Do all the length checks before returning even if "ip_mthread_enable"
[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 $
239b4df9 65 * $DragonFly: src/sys/vm/vm_map.c,v 1.23 2004/03/12 23:09:37 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
a108bf71
MD
94#include <sys/thread2.h>
95
984263bc
MD
96/*
97 * Virtual memory maps provide for the mapping, protection,
98 * and sharing of virtual memory objects. In addition,
99 * this module provides for an efficient virtual copy of
100 * memory from one map to another.
101 *
102 * Synchronization is required prior to most operations.
103 *
104 * Maps consist of an ordered doubly-linked list of simple
105 * entries; a single hint is used to speed up lookups.
106 *
107 * Since portions of maps are specified by start/end addresses,
108 * which may not align with existing map entries, all
109 * routines merely "clip" entries to these start/end values.
110 * [That is, an entry is split into two, bordering at a
111 * start or end value.] Note that these clippings may not
112 * always be necessary (as the two resulting entries are then
113 * not changed); however, the clipping is done for convenience.
114 *
115 * As mentioned above, virtual copy operations are performed
116 * by copying VM object references from one map to
117 * another, and then marking both regions as copy-on-write.
118 */
119
120/*
121 * vm_map_startup:
122 *
123 * Initialize the vm_map module. Must be called before
124 * any other vm_map routines.
125 *
126 * Map and entry structures are allocated from the general
127 * purpose memory pool with some exceptions:
128 *
129 * - The kernel map and kmem submap are allocated statically.
130 * - Kernel map entries are allocated out of a static pool.
131 *
132 * These restrictions are necessary since malloc() uses the
133 * maps and requires map entries.
134 */
135
a108bf71
MD
136static struct vm_zone mapentzone_store, mapzone_store;
137static vm_zone_t mapentzone, mapzone, vmspace_zone;
138static struct vm_object mapentobj, mapobj;
984263bc
MD
139
140static struct vm_map_entry map_entry_init[MAX_MAPENT];
984263bc
MD
141static struct vm_map map_init[MAX_KMAP];
142
a108bf71
MD
143static vm_map_entry_t vm_map_entry_create(vm_map_t map, int *);
144static void vm_map_entry_dispose (vm_map_t map, vm_map_entry_t entry, int *);
145static void _vm_map_clip_end (vm_map_t, vm_map_entry_t, vm_offset_t, int *);
146static void _vm_map_clip_start (vm_map_t, vm_map_entry_t, vm_offset_t, int *);
147static void vm_map_entry_delete (vm_map_t, vm_map_entry_t, int *);
1388df65
RG
148static void vm_map_entry_unwire (vm_map_t, vm_map_entry_t);
149static void vm_map_copy_entry (vm_map_t, vm_map_t, vm_map_entry_t,
150 vm_map_entry_t);
151static void vm_map_split (vm_map_entry_t);
a108bf71 152static void vm_map_unclip_range (vm_map_t map, vm_map_entry_t start_entry, vm_offset_t start, vm_offset_t end, int *count, int flags);
984263bc
MD
153
154void
155vm_map_startup()
156{
157 mapzone = &mapzone_store;
158 zbootinit(mapzone, "MAP", sizeof (struct vm_map),
159 map_init, MAX_KMAP);
984263bc
MD
160 mapentzone = &mapentzone_store;
161 zbootinit(mapentzone, "MAP ENTRY", sizeof (struct vm_map_entry),
162 map_entry_init, MAX_MAPENT);
163}
164
165/*
166 * Allocate a vmspace structure, including a vm_map and pmap,
167 * and initialize those structures. The refcnt is set to 1.
168 * The remaining fields must be initialized by the caller.
169 */
170struct vmspace *
171vmspace_alloc(min, max)
172 vm_offset_t min, max;
173{
174 struct vmspace *vm;
175
176 vm = zalloc(vmspace_zone);
177 vm_map_init(&vm->vm_map, min, max);
178 pmap_pinit(vmspace_pmap(vm));
179 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
180 vm->vm_refcnt = 1;
181 vm->vm_shm = NULL;
182 vm->vm_exitingcnt = 0;
183 return (vm);
184}
185
186void
a108bf71
MD
187vm_init2(void)
188{
189 zinitna(mapentzone, &mapentobj, NULL, 0, 0, ZONE_USE_RESERVE, 1);
190 zinitna(mapzone, &mapobj, NULL, 0, 0, 0, 1);
984263bc
MD
191 vmspace_zone = zinit("VMSPACE", sizeof (struct vmspace), 0, 0, 3);
192 pmap_init2();
193 vm_object_init2();
194}
195
196static __inline void
197vmspace_dofree(struct vmspace *vm)
198{
a108bf71
MD
199 int count;
200
fef0fdf2
MD
201 /*
202 * Make sure any SysV shm is freed, it might not have in
203 * exit1()
204 */
205 shmexit(vm);
206
a722be49
MD
207 KKASSERT(vm->vm_upcalls == NULL);
208
984263bc
MD
209 /*
210 * Lock the map, to wait out all other references to it.
211 * Delete all of the mappings and pages they hold, then call
212 * the pmap module to reclaim anything left.
213 */
a108bf71 214 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc 215 vm_map_lock(&vm->vm_map);
a108bf71
MD
216 vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
217 vm->vm_map.max_offset, &count);
984263bc 218 vm_map_unlock(&vm->vm_map);
a108bf71 219 vm_map_entry_release(count);
984263bc
MD
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
a108bf71 301vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
984263bc
MD
302{
303 vm_map_t result;
304
305 result = zalloc(mapzone);
306 vm_map_init(result, min, max);
307 result->pmap = pmap;
308 return (result);
309}
310
311/*
312 * Initialize an existing vm_map structure
313 * such as that in the vmspace structure.
314 * The pmap is set elsewhere.
315 */
316void
a108bf71 317vm_map_init(struct vm_map *map, vm_offset_t min, vm_offset_t max)
984263bc
MD
318{
319 map->header.next = map->header.prev = &map->header;
320 map->nentries = 0;
321 map->size = 0;
322 map->system_map = 0;
323 map->infork = 0;
324 map->min_offset = min;
325 map->max_offset = max;
326 map->first_free = &map->header;
327 map->hint = &map->header;
328 map->timestamp = 0;
377d4740 329 lockinit(&map->lock, 0, "thrd_sleep", 0, LK_NOPAUSE);
984263bc
MD
330}
331
41a01a4d
MD
332/*
333 * vm_map_entry_cpu_init:
334 *
335 * Set an initial negative count so the first attempt to reserve
336 * space preloads a bunch of vm_map_entry's for this cpu. This
337 * routine is called in early boot so we cannot just call
338 * vm_map_entry_reserve().
339 *
340 * May be called for a gd other then mycpu.
341 */
342void
343vm_map_entry_reserve_cpu_init(globaldata_t gd)
344{
345 gd->gd_vme_avail -= MAP_RESERVE_COUNT * 2;
346}
347
a108bf71
MD
348/*
349 * vm_map_entry_reserve:
350 *
e1359933
MD
351 * Reserves vm_map_entry structures so code later on can manipulate
352 * map_entry structures within a locked map without blocking trying
353 * to allocate a new vm_map_entry.
a108bf71
MD
354 */
355int
356vm_map_entry_reserve(int count)
357{
358 struct globaldata *gd = mycpu;
359 vm_map_entry_t entry;
360
361 crit_enter();
362 gd->gd_vme_avail -= count;
363
364 /*
365 * Make sure we have enough structures in gd_vme_base to handle
366 * the reservation request.
367 */
368 while (gd->gd_vme_avail < 0) {
369 entry = zalloc(mapentzone);
370 entry->next = gd->gd_vme_base;
371 gd->gd_vme_base = entry;
372 ++gd->gd_vme_avail;
373 }
374 crit_exit();
375 return(count);
376}
377
378/*
379 * vm_map_entry_release:
380 *
381 * Releases previously reserved vm_map_entry structures that were not
382 * used. If we have too much junk in our per-cpu cache clean some of
383 * it out.
384 */
385void
386vm_map_entry_release(int count)
387{
388 struct globaldata *gd = mycpu;
389 vm_map_entry_t entry;
390
391 crit_enter();
392 gd->gd_vme_avail += count;
393 while (gd->gd_vme_avail > MAP_RESERVE_SLOP) {
394 entry = gd->gd_vme_base;
395 KKASSERT(entry != NULL);
396 gd->gd_vme_base = entry->next;
397 --gd->gd_vme_avail;
398 crit_exit();
399 zfree(mapentzone, entry);
400 crit_enter();
401 }
402 crit_exit();
403}
404
405/*
406 * vm_map_entry_kreserve:
407 *
408 * Reserve map entry structures for use in kernel_map or (if it exists)
409 * kmem_map. These entries have *ALREADY* been reserved on a per-cpu
e1359933
MD
410 * basis when the map was inited. This function is used by zalloc()
411 * to avoid a recursion when zalloc() itself needs to allocate additional
412 * kernel memory.
a108bf71 413 *
e1359933
MD
414 * This function should only be used when the caller intends to later
415 * call vm_map_entry_reserve() to 'normalize' the reserve cache.
a108bf71
MD
416 */
417int
418vm_map_entry_kreserve(int count)
419{
420 struct globaldata *gd = mycpu;
421
422 crit_enter();
423 gd->gd_vme_kdeficit += count;
424 crit_exit();
425 KKASSERT(gd->gd_vme_base != NULL);
426 return(count);
427}
428
429/*
430 * vm_map_entry_krelease:
431 *
432 * Release previously reserved map entries for kernel_map or kmem_map
433 * use. This routine determines how many entries were actually used and
434 * replentishes the kernel reserve supply from vme_avail.
435 *
436 * If there is insufficient supply vme_avail will go negative, which is
437 * ok. We cannot safely call zalloc in this function without getting
438 * into a recursion deadlock. zalloc() will call vm_map_entry_reserve()
439 * to regenerate the lost entries.
440 */
441void
442vm_map_entry_krelease(int count)
443{
444 struct globaldata *gd = mycpu;
445
446 crit_enter();
447 gd->gd_vme_kdeficit -= count;
448 gd->gd_vme_avail -= gd->gd_vme_kdeficit; /* can go negative */
449 gd->gd_vme_kdeficit = 0;
450 crit_exit();
451}
452
984263bc 453/*
8a8d5d85 454 * vm_map_entry_create: [ internal use only ]
984263bc 455 *
8a8d5d85 456 * Allocates a VM map entry for insertion. No entry fields are filled
a108bf71
MD
457 * in.
458 *
459 * This routine may be called from an interrupt thread but not a FAST
460 * interrupt. This routine may recurse the map lock.
984263bc 461 */
8a8d5d85 462static vm_map_entry_t
a108bf71 463vm_map_entry_create(vm_map_t map, int *countp)
984263bc 464{
a108bf71
MD
465 struct globaldata *gd = mycpu;
466 vm_map_entry_t entry;
8a8d5d85 467
a108bf71
MD
468 KKASSERT(*countp > 0);
469 --*countp;
470 crit_enter();
471 entry = gd->gd_vme_base;
472 KASSERT(entry != NULL, ("gd_vme_base NULL! count %d", *countp));
473 gd->gd_vme_base = entry->next;
474 crit_exit();
475 return(entry);
984263bc
MD
476}
477
478/*
8a8d5d85 479 * vm_map_entry_dispose: [ internal use only ]
984263bc 480 *
8a8d5d85
MD
481 * Dispose of a vm_map_entry that is no longer being referenced. This
482 * function may be called from an interrupt.
984263bc 483 */
8a8d5d85 484static void
a108bf71 485vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc 486{
a108bf71
MD
487 struct globaldata *gd = mycpu;
488
489 ++*countp;
490 crit_enter();
491 entry->next = gd->gd_vme_base;
492 gd->gd_vme_base = entry;
493 crit_exit();
984263bc
MD
494}
495
8a8d5d85 496
984263bc
MD
497/*
498 * vm_map_entry_{un,}link:
499 *
500 * Insert/remove entries from maps.
501 */
502static __inline void
503vm_map_entry_link(vm_map_t map,
504 vm_map_entry_t after_where,
505 vm_map_entry_t entry)
506{
507 map->nentries++;
508 entry->prev = after_where;
509 entry->next = after_where->next;
510 entry->next->prev = entry;
511 after_where->next = entry;
512}
513
514static __inline void
515vm_map_entry_unlink(vm_map_t map,
516 vm_map_entry_t entry)
517{
518 vm_map_entry_t prev;
519 vm_map_entry_t next;
520
521 if (entry->eflags & MAP_ENTRY_IN_TRANSITION)
522 panic("vm_map_entry_unlink: attempt to mess with locked entry! %p", entry);
523 prev = entry->prev;
524 next = entry->next;
525 next->prev = prev;
526 prev->next = next;
527 map->nentries--;
528}
529
530/*
531 * SAVE_HINT:
532 *
533 * Saves the specified entry as the hint for
534 * future lookups.
535 */
536#define SAVE_HINT(map,value) \
537 (map)->hint = (value);
538
539/*
540 * vm_map_lookup_entry: [ internal use only ]
541 *
542 * Finds the map entry containing (or
543 * immediately preceding) the specified address
544 * in the given map; the entry is returned
545 * in the "entry" parameter. The boolean
546 * result indicates whether the address is
547 * actually contained in the map.
548 */
549boolean_t
550vm_map_lookup_entry(map, address, entry)
551 vm_map_t map;
552 vm_offset_t address;
553 vm_map_entry_t *entry; /* OUT */
554{
555 vm_map_entry_t cur;
556 vm_map_entry_t last;
557
558 /*
559 * Start looking either from the head of the list, or from the hint.
560 */
561
562 cur = map->hint;
563
564 if (cur == &map->header)
565 cur = cur->next;
566
567 if (address >= cur->start) {
568 /*
569 * Go from hint to end of list.
570 *
571 * But first, make a quick check to see if we are already looking
572 * at the entry we want (which is usually the case). Note also
573 * that we don't need to save the hint here... it is the same
574 * hint (unless we are at the header, in which case the hint
575 * didn't buy us anything anyway).
576 */
577 last = &map->header;
578 if ((cur != last) && (cur->end > address)) {
579 *entry = cur;
580 return (TRUE);
581 }
582 } else {
583 /*
584 * Go from start to hint, *inclusively*
585 */
586 last = cur->next;
587 cur = map->header.next;
588 }
589
590 /*
591 * Search linearly
592 */
593
594 while (cur != last) {
595 if (cur->end > address) {
596 if (address >= cur->start) {
597 /*
598 * Save this lookup for future hints, and
599 * return
600 */
601
602 *entry = cur;
603 SAVE_HINT(map, cur);
604 return (TRUE);
605 }
606 break;
607 }
608 cur = cur->next;
609 }
610 *entry = cur->prev;
611 SAVE_HINT(map, *entry);
612 return (FALSE);
613}
614
615/*
616 * vm_map_insert:
617 *
618 * Inserts the given whole VM object into the target
619 * map at the specified address range. The object's
620 * size should match that of the address range.
621 *
a108bf71
MD
622 * Requires that the map be locked, and leaves it so. Requires that
623 * sufficient vm_map_entry structures have been reserved and tracks
624 * the use via countp.
984263bc
MD
625 *
626 * If object is non-NULL, ref count must be bumped by caller
627 * prior to making call to account for the new entry.
628 */
629int
a108bf71
MD
630vm_map_insert(vm_map_t map, int *countp,
631 vm_object_t object, vm_ooffset_t offset,
984263bc
MD
632 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
633 int cow)
634{
635 vm_map_entry_t new_entry;
636 vm_map_entry_t prev_entry;
637 vm_map_entry_t temp_entry;
638 vm_eflags_t protoeflags;
639
640 /*
641 * Check that the start and end points are not bogus.
642 */
643
644 if ((start < map->min_offset) || (end > map->max_offset) ||
645 (start >= end))
646 return (KERN_INVALID_ADDRESS);
647
648 /*
649 * Find the entry prior to the proposed starting address; if it's part
650 * of an existing entry, this range is bogus.
651 */
652
653 if (vm_map_lookup_entry(map, start, &temp_entry))
654 return (KERN_NO_SPACE);
655
656 prev_entry = temp_entry;
657
658 /*
659 * Assert that the next entry doesn't overlap the end point.
660 */
661
662 if ((prev_entry->next != &map->header) &&
663 (prev_entry->next->start < end))
664 return (KERN_NO_SPACE);
665
666 protoeflags = 0;
667
668 if (cow & MAP_COPY_ON_WRITE)
669 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
670
671 if (cow & MAP_NOFAULT) {
672 protoeflags |= MAP_ENTRY_NOFAULT;
673
674 KASSERT(object == NULL,
675 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
676 }
677 if (cow & MAP_DISABLE_SYNCER)
678 protoeflags |= MAP_ENTRY_NOSYNC;
679 if (cow & MAP_DISABLE_COREDUMP)
680 protoeflags |= MAP_ENTRY_NOCOREDUMP;
681
682 if (object) {
683 /*
684 * When object is non-NULL, it could be shared with another
685 * process. We have to set or clear OBJ_ONEMAPPING
686 * appropriately.
687 */
688 if ((object->ref_count > 1) || (object->shadow_count != 0)) {
689 vm_object_clear_flag(object, OBJ_ONEMAPPING);
690 }
691 }
692 else if ((prev_entry != &map->header) &&
693 (prev_entry->eflags == protoeflags) &&
694 (prev_entry->end == start) &&
695 (prev_entry->wired_count == 0) &&
696 ((prev_entry->object.vm_object == NULL) ||
697 vm_object_coalesce(prev_entry->object.vm_object,
698 OFF_TO_IDX(prev_entry->offset),
699 (vm_size_t)(prev_entry->end - prev_entry->start),
700 (vm_size_t)(end - prev_entry->end)))) {
701 /*
702 * We were able to extend the object. Determine if we
703 * can extend the previous map entry to include the
704 * new range as well.
705 */
706 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
707 (prev_entry->protection == prot) &&
708 (prev_entry->max_protection == max)) {
709 map->size += (end - prev_entry->end);
710 prev_entry->end = end;
a108bf71 711 vm_map_simplify_entry(map, prev_entry, countp);
984263bc
MD
712 return (KERN_SUCCESS);
713 }
714
715 /*
716 * If we can extend the object but cannot extend the
717 * map entry, we have to create a new map entry. We
718 * must bump the ref count on the extended object to
719 * account for it. object may be NULL.
720 */
721 object = prev_entry->object.vm_object;
722 offset = prev_entry->offset +
723 (prev_entry->end - prev_entry->start);
724 vm_object_reference(object);
725 }
726
727 /*
728 * NOTE: if conditionals fail, object can be NULL here. This occurs
729 * in things like the buffer map where we manage kva but do not manage
730 * backing objects.
731 */
732
733 /*
734 * Create a new entry
735 */
736
a108bf71 737 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
738 new_entry->start = start;
739 new_entry->end = end;
740
741 new_entry->eflags = protoeflags;
742 new_entry->object.vm_object = object;
743 new_entry->offset = offset;
744 new_entry->avail_ssize = 0;
745
746 new_entry->inheritance = VM_INHERIT_DEFAULT;
747 new_entry->protection = prot;
748 new_entry->max_protection = max;
749 new_entry->wired_count = 0;
750
751 /*
752 * Insert the new entry into the list
753 */
754
755 vm_map_entry_link(map, prev_entry, new_entry);
756 map->size += new_entry->end - new_entry->start;
757
758 /*
759 * Update the free space hint
760 */
761 if ((map->first_free == prev_entry) &&
762 (prev_entry->end >= new_entry->start)) {
763 map->first_free = new_entry;
764 }
765
766#if 0
767 /*
768 * Temporarily removed to avoid MAP_STACK panic, due to
769 * MAP_STACK being a huge hack. Will be added back in
770 * when MAP_STACK (and the user stack mapping) is fixed.
771 */
772 /*
773 * It may be possible to simplify the entry
774 */
a108bf71 775 vm_map_simplify_entry(map, new_entry, countp);
984263bc
MD
776#endif
777
778 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
779 pmap_object_init_pt(map->pmap, start,
780 object, OFF_TO_IDX(offset), end - start,
781 cow & MAP_PREFAULT_PARTIAL);
782 }
783
784 return (KERN_SUCCESS);
785}
786
787/*
788 * Find sufficient space for `length' bytes in the given map, starting at
789 * `start'. The map must be locked. Returns 0 on success, 1 on no space.
e9bb90e8
MD
790 *
791 * This function will returned an arbitrarily aligned pointer. If no
792 * particular alignment is required you should pass align as 1. Note that
793 * the map may return PAGE_SIZE aligned pointers if all the lengths used in
794 * the map are a multiple of PAGE_SIZE, even if you pass a smaller align
795 * argument.
796 *
797 * 'align' should be a power of 2 but is not required to be.
984263bc
MD
798 */
799int
e9bb90e8
MD
800vm_map_findspace(
801 vm_map_t map,
802 vm_offset_t start,
803 vm_size_t length,
804 vm_offset_t align,
805 vm_offset_t *addr)
984263bc
MD
806{
807 vm_map_entry_t entry, next;
808 vm_offset_t end;
e9bb90e8 809 vm_offset_t align_mask;
984263bc
MD
810
811 if (start < map->min_offset)
812 start = map->min_offset;
813 if (start > map->max_offset)
814 return (1);
815
e9bb90e8
MD
816 /*
817 * If the alignment is not a power of 2 we will have to use
818 * a mod/division, set align_mask to a special value.
819 */
820 if ((align | (align - 1)) + 1 != (align << 1))
821 align_mask = (vm_offset_t)-1;
822 else
823 align_mask = align - 1;
824
a108bf71 825retry:
984263bc
MD
826 /*
827 * Look for the first possible address; if there's already something
828 * at this address, we have to start after it.
829 */
830 if (start == map->min_offset) {
831 if ((entry = map->first_free) != &map->header)
832 start = entry->end;
833 } else {
834 vm_map_entry_t tmp;
835
836 if (vm_map_lookup_entry(map, start, &tmp))
837 start = tmp->end;
838 entry = tmp;
839 }
840
841 /*
842 * Look through the rest of the map, trying to fit a new region in the
843 * gap between existing regions, or after the very last region.
844 */
845 for (;; start = (entry = next)->end) {
e9bb90e8
MD
846 /*
847 * Adjust the proposed start by the requested alignment,
848 * be sure that we didn't wrap the address.
849 */
850 if (align_mask == (vm_offset_t)-1)
851 end = ((start + align - 1) / align) * align;
852 else
853 end = (start + align_mask) & ~align_mask;
854 if (end < start)
855 return (1);
856 start = end;
984263bc
MD
857 /*
858 * Find the end of the proposed new region. Be sure we didn't
e9bb90e8
MD
859 * go beyond the end of the map, or wrap around the address.
860 * Then check to see if this is the last entry or if the
861 * proposed end fits in the gap between this and the next
862 * entry.
984263bc
MD
863 */
864 end = start + length;
865 if (end > map->max_offset || end < start)
866 return (1);
867 next = entry->next;
868 if (next == &map->header || next->start >= end)
869 break;
870 }
871 SAVE_HINT(map, entry);
984263bc
MD
872 if (map == kernel_map) {
873 vm_offset_t ksize;
874 if ((ksize = round_page(start + length)) > kernel_vm_end) {
875 pmap_growkernel(ksize);
a108bf71 876 goto retry;
984263bc
MD
877 }
878 }
a108bf71 879 *addr = start;
984263bc
MD
880 return (0);
881}
882
883/*
884 * vm_map_find finds an unallocated region in the target address
885 * map with the given length. The search is defined to be
886 * first-fit from the specified address; the region found is
887 * returned in the same parameter.
888 *
889 * If object is non-NULL, ref count must be bumped by caller
890 * prior to making call to account for the new entry.
891 */
892int
893vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
894 vm_offset_t *addr, /* IN/OUT */
895 vm_size_t length, boolean_t find_space, vm_prot_t prot,
896 vm_prot_t max, int cow)
897{
898 vm_offset_t start;
03aa8d99 899 int result;
a108bf71 900 int count;
984263bc
MD
901
902 start = *addr;
903
a108bf71 904 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
905 vm_map_lock(map);
906 if (find_space) {
e9bb90e8 907 if (vm_map_findspace(map, start, length, 1, addr)) {
984263bc 908 vm_map_unlock(map);
a108bf71 909 vm_map_entry_release(count);
984263bc
MD
910 return (KERN_NO_SPACE);
911 }
912 start = *addr;
913 }
a108bf71 914 result = vm_map_insert(map, &count, object, offset,
984263bc
MD
915 start, start + length, prot, max, cow);
916 vm_map_unlock(map);
a108bf71 917 vm_map_entry_release(count);
984263bc 918
984263bc
MD
919 return (result);
920}
921
922/*
923 * vm_map_simplify_entry:
924 *
925 * Simplify the given map entry by merging with either neighbor. This
926 * routine also has the ability to merge with both neighbors.
927 *
928 * The map must be locked.
929 *
930 * This routine guarentees that the passed entry remains valid (though
931 * possibly extended). When merging, this routine may delete one or
932 * both neighbors. No action is taken on entries which have their
933 * in-transition flag set.
934 */
935void
a108bf71 936vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc
MD
937{
938 vm_map_entry_t next, prev;
939 vm_size_t prevsize, esize;
940
941 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) {
12e4aaff 942 ++mycpu->gd_cnt.v_intrans_coll;
984263bc
MD
943 return;
944 }
945
946 prev = entry->prev;
947 if (prev != &map->header) {
948 prevsize = prev->end - prev->start;
949 if ( (prev->end == entry->start) &&
950 (prev->object.vm_object == entry->object.vm_object) &&
951 (!prev->object.vm_object ||
952 (prev->offset + prevsize == entry->offset)) &&
953 (prev->eflags == entry->eflags) &&
954 (prev->protection == entry->protection) &&
955 (prev->max_protection == entry->max_protection) &&
956 (prev->inheritance == entry->inheritance) &&
957 (prev->wired_count == entry->wired_count)) {
958 if (map->first_free == prev)
959 map->first_free = entry;
960 if (map->hint == prev)
961 map->hint = entry;
962 vm_map_entry_unlink(map, prev);
963 entry->start = prev->start;
964 entry->offset = prev->offset;
965 if (prev->object.vm_object)
966 vm_object_deallocate(prev->object.vm_object);
a108bf71 967 vm_map_entry_dispose(map, prev, countp);
984263bc
MD
968 }
969 }
970
971 next = entry->next;
972 if (next != &map->header) {
973 esize = entry->end - entry->start;
974 if ((entry->end == next->start) &&
975 (next->object.vm_object == entry->object.vm_object) &&
976 (!entry->object.vm_object ||
977 (entry->offset + esize == next->offset)) &&
978 (next->eflags == entry->eflags) &&
979 (next->protection == entry->protection) &&
980 (next->max_protection == entry->max_protection) &&
981 (next->inheritance == entry->inheritance) &&
982 (next->wired_count == entry->wired_count)) {
983 if (map->first_free == next)
984 map->first_free = entry;
985 if (map->hint == next)
986 map->hint = entry;
987 vm_map_entry_unlink(map, next);
988 entry->end = next->end;
989 if (next->object.vm_object)
990 vm_object_deallocate(next->object.vm_object);
a108bf71 991 vm_map_entry_dispose(map, next, countp);
984263bc
MD
992 }
993 }
994}
995/*
996 * vm_map_clip_start: [ internal use only ]
997 *
998 * Asserts that the given entry begins at or after
999 * the specified address; if necessary,
1000 * it splits the entry into two.
1001 */
a108bf71 1002#define vm_map_clip_start(map, entry, startaddr, countp) \
984263bc
MD
1003{ \
1004 if (startaddr > entry->start) \
a108bf71 1005 _vm_map_clip_start(map, entry, startaddr, countp); \
984263bc
MD
1006}
1007
1008/*
1009 * This routine is called only when it is known that
1010 * the entry must be split.
1011 */
1012static void
a108bf71 1013_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start, int *countp)
984263bc
MD
1014{
1015 vm_map_entry_t new_entry;
1016
1017 /*
1018 * Split off the front portion -- note that we must insert the new
1019 * entry BEFORE this one, so that this entry has the specified
1020 * starting address.
1021 */
1022
a108bf71 1023 vm_map_simplify_entry(map, entry, countp);
984263bc
MD
1024
1025 /*
1026 * If there is no object backing this entry, we might as well create
1027 * one now. If we defer it, an object can get created after the map
1028 * is clipped, and individual objects will be created for the split-up
1029 * map. This is a bit of a hack, but is also about the best place to
1030 * put this improvement.
1031 */
1032
1033 if (entry->object.vm_object == NULL && !map->system_map) {
1034 vm_object_t object;
1035 object = vm_object_allocate(OBJT_DEFAULT,
1036 atop(entry->end - entry->start));
1037 entry->object.vm_object = object;
1038 entry->offset = 0;
1039 }
1040
a108bf71 1041 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
1042 *new_entry = *entry;
1043
1044 new_entry->end = start;
1045 entry->offset += (start - entry->start);
1046 entry->start = start;
1047
1048 vm_map_entry_link(map, entry->prev, new_entry);
1049
1050 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1051 vm_object_reference(new_entry->object.vm_object);
1052 }
1053}
1054
1055/*
1056 * vm_map_clip_end: [ internal use only ]
1057 *
1058 * Asserts that the given entry ends at or before
1059 * the specified address; if necessary,
1060 * it splits the entry into two.
1061 */
1062
a108bf71 1063#define vm_map_clip_end(map, entry, endaddr, countp) \
984263bc
MD
1064{ \
1065 if (endaddr < entry->end) \
a108bf71 1066 _vm_map_clip_end(map, entry, endaddr, countp); \
984263bc
MD
1067}
1068
1069/*
1070 * This routine is called only when it is known that
1071 * the entry must be split.
1072 */
1073static void
a108bf71 1074_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end, int *countp)
984263bc
MD
1075{
1076 vm_map_entry_t new_entry;
1077
1078 /*
1079 * If there is no object backing this entry, we might as well create
1080 * one now. If we defer it, an object can get created after the map
1081 * is clipped, and individual objects will be created for the split-up
1082 * map. This is a bit of a hack, but is also about the best place to
1083 * put this improvement.
1084 */
1085
1086 if (entry->object.vm_object == NULL && !map->system_map) {
1087 vm_object_t object;
1088 object = vm_object_allocate(OBJT_DEFAULT,
1089 atop(entry->end - entry->start));
1090 entry->object.vm_object = object;
1091 entry->offset = 0;
1092 }
1093
1094 /*
1095 * Create a new entry and insert it AFTER the specified entry
1096 */
1097
a108bf71 1098 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
1099 *new_entry = *entry;
1100
1101 new_entry->start = entry->end = end;
1102 new_entry->offset += (end - entry->start);
1103
1104 vm_map_entry_link(map, entry, new_entry);
1105
1106 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1107 vm_object_reference(new_entry->object.vm_object);
1108 }
1109}
1110
1111/*
1112 * VM_MAP_RANGE_CHECK: [ internal use only ]
1113 *
1114 * Asserts that the starting and ending region
1115 * addresses fall within the valid range of the map.
1116 */
1117#define VM_MAP_RANGE_CHECK(map, start, end) \
1118 { \
1119 if (start < vm_map_min(map)) \
1120 start = vm_map_min(map); \
1121 if (end > vm_map_max(map)) \
1122 end = vm_map_max(map); \
1123 if (start > end) \
1124 start = end; \
1125 }
1126
1127/*
1128 * vm_map_transition_wait: [ kernel use only ]
1129 *
1130 * Used to block when an in-transition collison occurs. The map
1131 * is unlocked for the sleep and relocked before the return.
1132 */
1133static
1134void
1135vm_map_transition_wait(vm_map_t map)
1136{
1137 vm_map_unlock(map);
377d4740 1138 tsleep(map, 0, "vment", 0);
984263bc
MD
1139 vm_map_lock(map);
1140}
1141
1142/*
1143 * CLIP_CHECK_BACK
1144 * CLIP_CHECK_FWD
1145 *
1146 * When we do blocking operations with the map lock held it is
1147 * possible that a clip might have occured on our in-transit entry,
1148 * requiring an adjustment to the entry in our loop. These macros
1149 * help the pageable and clip_range code deal with the case. The
1150 * conditional costs virtually nothing if no clipping has occured.
1151 */
1152
1153#define CLIP_CHECK_BACK(entry, save_start) \
1154 do { \
1155 while (entry->start != save_start) { \
1156 entry = entry->prev; \
1157 KASSERT(entry != &map->header, ("bad entry clip")); \
1158 } \
1159 } while(0)
1160
1161#define CLIP_CHECK_FWD(entry, save_end) \
1162 do { \
1163 while (entry->end != save_end) { \
1164 entry = entry->next; \
1165 KASSERT(entry != &map->header, ("bad entry clip")); \
1166 } \
1167 } while(0)
1168
1169
1170/*
1171 * vm_map_clip_range: [ kernel use only ]
1172 *
1173 * Clip the specified range and return the base entry. The
1174 * range may cover several entries starting at the returned base
1175 * and the first and last entry in the covering sequence will be
1176 * properly clipped to the requested start and end address.
1177 *
1178 * If no holes are allowed you should pass the MAP_CLIP_NO_HOLES
1179 * flag.
1180 *
1181 * The MAP_ENTRY_IN_TRANSITION flag will be set for the entries
1182 * covered by the requested range.
1183 *
1184 * The map must be exclusively locked on entry and will remain locked
1185 * on return. If no range exists or the range contains holes and you
1186 * specified that no holes were allowed, NULL will be returned. This
1187 * routine may temporarily unlock the map in order avoid a deadlock when
1188 * sleeping.
1189 */
1190static
1191vm_map_entry_t
a108bf71
MD
1192vm_map_clip_range(vm_map_t map, vm_offset_t start, vm_offset_t end,
1193 int *countp, int flags)
984263bc
MD
1194{
1195 vm_map_entry_t start_entry;
1196 vm_map_entry_t entry;
1197
1198 /*
1199 * Locate the entry and effect initial clipping. The in-transition
1200 * case does not occur very often so do not try to optimize it.
1201 */
1202again:
1203 if (vm_map_lookup_entry(map, start, &start_entry) == FALSE)
1204 return (NULL);
1205 entry = start_entry;
1206 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1207 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
12e4aaff
MD
1208 ++mycpu->gd_cnt.v_intrans_coll;
1209 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
1210 vm_map_transition_wait(map);
1211 /*
1212 * entry and/or start_entry may have been clipped while
1213 * we slept, or may have gone away entirely. We have
1214 * to restart from the lookup.
1215 */
1216 goto again;
1217 }
1218 /*
1219 * Since we hold an exclusive map lock we do not have to restart
1220 * after clipping, even though clipping may block in zalloc.
1221 */
a108bf71
MD
1222 vm_map_clip_start(map, entry, start, countp);
1223 vm_map_clip_end(map, entry, end, countp);
984263bc
MD
1224 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1225
1226 /*
1227 * Scan entries covered by the range. When working on the next
1228 * entry a restart need only re-loop on the current entry which
1229 * we have already locked, since 'next' may have changed. Also,
1230 * even though entry is safe, it may have been clipped so we
1231 * have to iterate forwards through the clip after sleeping.
1232 */
1233 while (entry->next != &map->header && entry->next->start < end) {
1234 vm_map_entry_t next = entry->next;
1235
1236 if (flags & MAP_CLIP_NO_HOLES) {
1237 if (next->start > entry->end) {
1238 vm_map_unclip_range(map, start_entry,
a108bf71 1239 start, entry->end, countp, flags);
984263bc
MD
1240 return(NULL);
1241 }
1242 }
1243
1244 if (next->eflags & MAP_ENTRY_IN_TRANSITION) {
1245 vm_offset_t save_end = entry->end;
1246 next->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
12e4aaff
MD
1247 ++mycpu->gd_cnt.v_intrans_coll;
1248 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
1249 vm_map_transition_wait(map);
1250
1251 /*
1252 * clips might have occured while we blocked.
1253 */
1254 CLIP_CHECK_FWD(entry, save_end);
1255 CLIP_CHECK_BACK(start_entry, start);
1256 continue;
1257 }
1258 /*
1259 * No restart necessary even though clip_end may block, we
1260 * are holding the map lock.
1261 */
a108bf71 1262 vm_map_clip_end(map, next, end, countp);
984263bc
MD
1263 next->eflags |= MAP_ENTRY_IN_TRANSITION;
1264 entry = next;
1265 }
1266 if (flags & MAP_CLIP_NO_HOLES) {
1267 if (entry->end != end) {
1268 vm_map_unclip_range(map, start_entry,
a108bf71 1269 start, entry->end, countp, flags);
984263bc
MD
1270 return(NULL);
1271 }
1272 }
1273 return(start_entry);
1274}
1275
1276/*
1277 * vm_map_unclip_range: [ kernel use only ]
1278 *
1279 * Undo the effect of vm_map_clip_range(). You should pass the same
1280 * flags and the same range that you passed to vm_map_clip_range().
1281 * This code will clear the in-transition flag on the entries and
1282 * wake up anyone waiting. This code will also simplify the sequence
1283 * and attempt to merge it with entries before and after the sequence.
1284 *
1285 * The map must be locked on entry and will remain locked on return.
1286 *
1287 * Note that you should also pass the start_entry returned by
1288 * vm_map_clip_range(). However, if you block between the two calls
1289 * with the map unlocked please be aware that the start_entry may
1290 * have been clipped and you may need to scan it backwards to find
1291 * the entry corresponding with the original start address. You are
1292 * responsible for this, vm_map_unclip_range() expects the correct
1293 * start_entry to be passed to it and will KASSERT otherwise.
1294 */
1295static
1296void
1297vm_map_unclip_range(
1298 vm_map_t map,
1299 vm_map_entry_t start_entry,
1300 vm_offset_t start,
1301 vm_offset_t end,
a108bf71 1302 int *countp,
984263bc
MD
1303 int flags)
1304{
1305 vm_map_entry_t entry;
1306
1307 entry = start_entry;
1308
1309 KASSERT(entry->start == start, ("unclip_range: illegal base entry"));
1310 while (entry != &map->header && entry->start < end) {
1311 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, ("in-transition flag not set during unclip on: %p", entry));
1312 KASSERT(entry->end <= end, ("unclip_range: tail wasn't clipped"));
1313 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1314 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1315 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1316 wakeup(map);
1317 }
1318 entry = entry->next;
1319 }
1320
1321 /*
1322 * Simplification does not block so there is no restart case.
1323 */
1324 entry = start_entry;
1325 while (entry != &map->header && entry->start < end) {
a108bf71 1326 vm_map_simplify_entry(map, entry, countp);
984263bc
MD
1327 entry = entry->next;
1328 }
1329}
1330
1331/*
1332 * vm_map_submap: [ kernel use only ]
1333 *
1334 * Mark the given range as handled by a subordinate map.
1335 *
1336 * This range must have been created with vm_map_find,
1337 * and no other operations may have been performed on this
1338 * range prior to calling vm_map_submap.
1339 *
1340 * Only a limited number of operations can be performed
1341 * within this rage after calling vm_map_submap:
1342 * vm_fault
1343 * [Don't try vm_map_copy!]
1344 *
1345 * To remove a submapping, one must first remove the
1346 * range from the superior map, and then destroy the
1347 * submap (if desired). [Better yet, don't try it.]
1348 */
1349int
a108bf71 1350vm_map_submap(vm_map_t map, vm_offset_t start, vm_offset_t end, vm_map_t submap)
984263bc
MD
1351{
1352 vm_map_entry_t entry;
1353 int result = KERN_INVALID_ARGUMENT;
a108bf71 1354 int count;
984263bc 1355
a108bf71 1356 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1357 vm_map_lock(map);
1358
1359 VM_MAP_RANGE_CHECK(map, start, end);
1360
1361 if (vm_map_lookup_entry(map, start, &entry)) {
a108bf71 1362 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1363 } else {
1364 entry = entry->next;
1365 }
1366
a108bf71 1367 vm_map_clip_end(map, entry, end, &count);
984263bc
MD
1368
1369 if ((entry->start == start) && (entry->end == end) &&
1370 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1371 (entry->object.vm_object == NULL)) {
1372 entry->object.sub_map = submap;
1373 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1374 result = KERN_SUCCESS;
1375 }
1376 vm_map_unlock(map);
a108bf71 1377 vm_map_entry_release(count);
984263bc
MD
1378
1379 return (result);
1380}
1381
1382/*
1383 * vm_map_protect:
1384 *
1385 * Sets the protection of the specified address
1386 * region in the target map. If "set_max" is
1387 * specified, the maximum protection is to be set;
1388 * otherwise, only the current protection is affected.
1389 */
1390int
1391vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1392 vm_prot_t new_prot, boolean_t set_max)
1393{
1394 vm_map_entry_t current;
1395 vm_map_entry_t entry;
a108bf71 1396 int count;
984263bc 1397
a108bf71 1398 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1399 vm_map_lock(map);
1400
1401 VM_MAP_RANGE_CHECK(map, start, end);
1402
1403 if (vm_map_lookup_entry(map, start, &entry)) {
a108bf71 1404 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1405 } else {
1406 entry = entry->next;
1407 }
1408
1409 /*
1410 * Make a first pass to check for protection violations.
1411 */
1412
1413 current = entry;
1414 while ((current != &map->header) && (current->start < end)) {
1415 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1416 vm_map_unlock(map);
a108bf71 1417 vm_map_entry_release(count);
984263bc
MD
1418 return (KERN_INVALID_ARGUMENT);
1419 }
1420 if ((new_prot & current->max_protection) != new_prot) {
1421 vm_map_unlock(map);
a108bf71 1422 vm_map_entry_release(count);
984263bc
MD
1423 return (KERN_PROTECTION_FAILURE);
1424 }
1425 current = current->next;
1426 }
1427
1428 /*
1429 * Go back and fix up protections. [Note that clipping is not
1430 * necessary the second time.]
1431 */
984263bc
MD
1432 current = entry;
1433
1434 while ((current != &map->header) && (current->start < end)) {
1435 vm_prot_t old_prot;
1436
a108bf71 1437 vm_map_clip_end(map, current, end, &count);
984263bc
MD
1438
1439 old_prot = current->protection;
1440 if (set_max)
1441 current->protection =
1442 (current->max_protection = new_prot) &
1443 old_prot;
1444 else
1445 current->protection = new_prot;
1446
1447 /*
1448 * Update physical map if necessary. Worry about copy-on-write
1449 * here -- CHECK THIS XXX
1450 */
1451
1452 if (current->protection != old_prot) {
1453#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1454 VM_PROT_ALL)
1455
1456 pmap_protect(map->pmap, current->start,
1457 current->end,
1458 current->protection & MASK(current));
1459#undef MASK
1460 }
1461
a108bf71 1462 vm_map_simplify_entry(map, current, &count);
984263bc
MD
1463
1464 current = current->next;
1465 }
1466
1467 vm_map_unlock(map);
a108bf71 1468 vm_map_entry_release(count);
984263bc
MD
1469 return (KERN_SUCCESS);
1470}
1471
1472/*
1473 * vm_map_madvise:
1474 *
1475 * This routine traverses a processes map handling the madvise
1476 * system call. Advisories are classified as either those effecting
1477 * the vm_map_entry structure, or those effecting the underlying
1478 * objects.
1479 */
1480
1481int
a108bf71 1482vm_map_madvise(vm_map_t map, vm_offset_t start, vm_offset_t end, int behav)
984263bc
MD
1483{
1484 vm_map_entry_t current, entry;
1485 int modify_map = 0;
a108bf71 1486 int count;
984263bc
MD
1487
1488 /*
1489 * Some madvise calls directly modify the vm_map_entry, in which case
1490 * we need to use an exclusive lock on the map and we need to perform
1491 * various clipping operations. Otherwise we only need a read-lock
1492 * on the map.
1493 */
1494
a108bf71
MD
1495 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1496
984263bc
MD
1497 switch(behav) {
1498 case MADV_NORMAL:
1499 case MADV_SEQUENTIAL:
1500 case MADV_RANDOM:
1501 case MADV_NOSYNC:
1502 case MADV_AUTOSYNC:
1503 case MADV_NOCORE:
1504 case MADV_CORE:
1505 modify_map = 1;
1506 vm_map_lock(map);
1507 break;
1508 case MADV_WILLNEED:
1509 case MADV_DONTNEED:
1510 case MADV_FREE:
1511 vm_map_lock_read(map);
1512 break;
1513 default:
a108bf71 1514 vm_map_entry_release(count);
984263bc
MD
1515 return (KERN_INVALID_ARGUMENT);
1516 }
1517
1518 /*
1519 * Locate starting entry and clip if necessary.
1520 */
1521
1522 VM_MAP_RANGE_CHECK(map, start, end);
1523
1524 if (vm_map_lookup_entry(map, start, &entry)) {
1525 if (modify_map)
a108bf71 1526 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1527 } else {
1528 entry = entry->next;
1529 }
1530
1531 if (modify_map) {
1532 /*
1533 * madvise behaviors that are implemented in the vm_map_entry.
1534 *
1535 * We clip the vm_map_entry so that behavioral changes are
1536 * limited to the specified address range.
1537 */
1538 for (current = entry;
1539 (current != &map->header) && (current->start < end);
1540 current = current->next
1541 ) {
1542 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1543 continue;
1544
a108bf71 1545 vm_map_clip_end(map, current, end, &count);
984263bc
MD
1546
1547 switch (behav) {
1548 case MADV_NORMAL:
1549 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1550 break;
1551 case MADV_SEQUENTIAL:
1552 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1553 break;
1554 case MADV_RANDOM:
1555 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1556 break;
1557 case MADV_NOSYNC:
1558 current->eflags |= MAP_ENTRY_NOSYNC;
1559 break;
1560 case MADV_AUTOSYNC:
1561 current->eflags &= ~MAP_ENTRY_NOSYNC;
1562 break;
1563 case MADV_NOCORE:
1564 current->eflags |= MAP_ENTRY_NOCOREDUMP;
1565 break;
1566 case MADV_CORE:
1567 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1568 break;
1569 default:
1570 break;
1571 }
a108bf71 1572 vm_map_simplify_entry(map, current, &count);
984263bc
MD
1573 }
1574 vm_map_unlock(map);
1575 } else {
1576 vm_pindex_t pindex;
1577 int count;
1578
1579 /*
1580 * madvise behaviors that are implemented in the underlying
1581 * vm_object.
1582 *
1583 * Since we don't clip the vm_map_entry, we have to clip
1584 * the vm_object pindex and count.
1585 */
1586 for (current = entry;
1587 (current != &map->header) && (current->start < end);
1588 current = current->next
1589 ) {
1590 vm_offset_t useStart;
1591
1592 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1593 continue;
1594
1595 pindex = OFF_TO_IDX(current->offset);
1596 count = atop(current->end - current->start);
1597 useStart = current->start;
1598
1599 if (current->start < start) {
1600 pindex += atop(start - current->start);
1601 count -= atop(start - current->start);
1602 useStart = start;
1603 }
1604 if (current->end > end)
1605 count -= atop(current->end - end);
1606
1607 if (count <= 0)
1608 continue;
1609
1610 vm_object_madvise(current->object.vm_object,
1611 pindex, count, behav);
1612 if (behav == MADV_WILLNEED) {
1613 pmap_object_init_pt(
1614 map->pmap,
1615 useStart,
1616 current->object.vm_object,
1617 pindex,
1618 (count << PAGE_SHIFT),
1619 MAP_PREFAULT_MADVISE
1620 );
1621 }
1622 }
1623 vm_map_unlock_read(map);
1624 }
a108bf71 1625 vm_map_entry_release(count);
984263bc
MD
1626 return(0);
1627}
1628
1629
1630/*
1631 * vm_map_inherit:
1632 *
1633 * Sets the inheritance of the specified address
1634 * range in the target map. Inheritance
1635 * affects how the map will be shared with
1636 * child maps at the time of vm_map_fork.
1637 */
1638int
1639vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1640 vm_inherit_t new_inheritance)
1641{
1642 vm_map_entry_t entry;
1643 vm_map_entry_t temp_entry;
a108bf71 1644 int count;
984263bc
MD
1645
1646 switch (new_inheritance) {
1647 case VM_INHERIT_NONE:
1648 case VM_INHERIT_COPY:
1649 case VM_INHERIT_SHARE:
1650 break;
1651 default:
1652 return (KERN_INVALID_ARGUMENT);
1653 }
1654
a108bf71 1655 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1656 vm_map_lock(map);
1657
1658 VM_MAP_RANGE_CHECK(map, start, end);
1659
1660 if (vm_map_lookup_entry(map, start, &temp_entry)) {
1661 entry = temp_entry;
a108bf71 1662 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1663 } else
1664 entry = temp_entry->next;
1665
1666 while ((entry != &map->header) && (entry->start < end)) {
a108bf71 1667 vm_map_clip_end(map, entry, end, &count);
984263bc
MD
1668
1669 entry->inheritance = new_inheritance;
1670
a108bf71 1671 vm_map_simplify_entry(map, entry, &count);
984263bc
MD
1672
1673 entry = entry->next;
1674 }
984263bc 1675 vm_map_unlock(map);
a108bf71 1676 vm_map_entry_release(count);
984263bc
MD
1677 return (KERN_SUCCESS);
1678}
1679
1680/*
1681 * Implement the semantics of mlock
1682 */
1683int
cde87949 1684vm_map_unwire(map, start, real_end, new_pageable)
984263bc
MD
1685 vm_map_t map;
1686 vm_offset_t start;
1687 vm_offset_t real_end;
1688 boolean_t new_pageable;
1689{
1690 vm_map_entry_t entry;
1691 vm_map_entry_t start_entry;
1692 vm_offset_t end;
1693 int rv = KERN_SUCCESS;
a108bf71 1694 int count;
984263bc 1695
a108bf71 1696 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1697 vm_map_lock(map);
1698 VM_MAP_RANGE_CHECK(map, start, real_end);
1699 end = real_end;
1700
a108bf71 1701 start_entry = vm_map_clip_range(map, start, end, &count, MAP_CLIP_NO_HOLES);
984263bc
MD
1702 if (start_entry == NULL) {
1703 vm_map_unlock(map);
a108bf71 1704 vm_map_entry_release(count);
984263bc
MD
1705 return (KERN_INVALID_ADDRESS);
1706 }
1707
1708 if (new_pageable == 0) {
1709 entry = start_entry;
1710 while ((entry != &map->header) && (entry->start < end)) {
1711 vm_offset_t save_start;
1712 vm_offset_t save_end;
1713
1714 /*
1715 * Already user wired or hard wired (trivial cases)
1716 */
1717 if (entry->eflags & MAP_ENTRY_USER_WIRED) {
1718 entry = entry->next;
1719 continue;
1720 }
1721 if (entry->wired_count != 0) {
1722 entry->wired_count++;
1723 entry->eflags |= MAP_ENTRY_USER_WIRED;
1724 entry = entry->next;
1725 continue;
1726 }
1727
1728 /*
1729 * A new wiring requires instantiation of appropriate
1730 * management structures and the faulting in of the
1731 * page.
1732 */
1733 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1734 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1735 if (copyflag && ((entry->protection & VM_PROT_WRITE) != 0)) {
1736
1737 vm_object_shadow(&entry->object.vm_object,
1738 &entry->offset,
1739 atop(entry->end - entry->start));
1740 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1741
1742 } else if (entry->object.vm_object == NULL &&
1743 !map->system_map) {
1744
1745 entry->object.vm_object =
1746 vm_object_allocate(OBJT_DEFAULT,
1747 atop(entry->end - entry->start));
1748 entry->offset = (vm_offset_t) 0;
1749
1750 }
1751 }
1752 entry->wired_count++;
1753 entry->eflags |= MAP_ENTRY_USER_WIRED;
1754
1755 /*
1756 * Now fault in the area. The map lock needs to be
1757 * manipulated to avoid deadlocks. The in-transition
1758 * flag protects the entries.
1759 */
1760 save_start = entry->start;
1761 save_end = entry->end;
1762 vm_map_unlock(map);
1763 map->timestamp++;
1764 rv = vm_fault_user_wire(map, save_start, save_end);
1765 vm_map_lock(map);
1766 if (rv) {
1767 CLIP_CHECK_BACK(entry, save_start);
1768 for (;;) {
1769 KASSERT(entry->wired_count == 1, ("bad wired_count on entry"));
1770 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1771 entry->wired_count = 0;
1772 if (entry->end == save_end)
1773 break;
1774 entry = entry->next;
1775 KASSERT(entry != &map->header, ("bad entry clip during backout"));
1776 }
1777 end = save_start; /* unwire the rest */
1778 break;
1779 }
1780 /*
1781 * note that even though the entry might have been
1782 * clipped, the USER_WIRED flag we set prevents
1783 * duplication so we do not have to do a
1784 * clip check.
1785 */
1786 entry = entry->next;
1787 }
1788
1789 /*
1790 * If we failed fall through to the unwiring section to
1791 * unwire what we had wired so far. 'end' has already
1792 * been adjusted.
1793 */
1794 if (rv)
1795 new_pageable = 1;
1796
1797 /*
1798 * start_entry might have been clipped if we unlocked the
1799 * map and blocked. No matter how clipped it has gotten
1800 * there should be a fragment that is on our start boundary.
1801 */
1802 CLIP_CHECK_BACK(start_entry, start);
1803 }
1804
1805 /*
1806 * Deal with the unwiring case.
1807 */
1808 if (new_pageable) {
1809 /*
1810 * This is the unwiring case. We must first ensure that the
1811 * range to be unwired is really wired down. We know there
1812 * are no holes.
1813 */
1814 entry = start_entry;
1815 while ((entry != &map->header) && (entry->start < end)) {
1816 if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
1817 rv = KERN_INVALID_ARGUMENT;
1818 goto done;
1819 }
1820 KASSERT(entry->wired_count != 0, ("wired count was 0 with USER_WIRED set! %p", entry));
1821 entry = entry->next;
1822 }
1823
1824 /*
1825 * Now decrement the wiring count for each region. If a region
1826 * becomes completely unwired, unwire its physical pages and
1827 * mappings.
1828 */
b4eddbac
DR
1829 /*
1830 * The map entries are processed in a loop, checking to
1831 * make sure the entry is wired and asserting it has a wired
1832 * count. However, another loop was inserted more-or-less in
1833 * the middle of the unwiring path. This loop picks up the
1834 * "entry" loop variable from the first loop without first
1835 * setting it to start_entry. Naturally, the secound loop
1836 * is never entered and the pages backing the entries are
1837 * never unwired. This can lead to a leak of wired pages.
1838 */
1839 entry = start_entry;
984263bc
MD
1840 while ((entry != &map->header) && (entry->start < end)) {
1841 KASSERT(entry->eflags & MAP_ENTRY_USER_WIRED, ("expected USER_WIRED on entry %p", entry));
1842 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1843 entry->wired_count--;
1844 if (entry->wired_count == 0)
1845 vm_fault_unwire(map, entry->start, entry->end);
1846 entry = entry->next;
1847 }
1848 }
1849done:
a108bf71 1850 vm_map_unclip_range(map, start_entry, start, real_end, &count,
984263bc
MD
1851 MAP_CLIP_NO_HOLES);
1852 map->timestamp++;
1853 vm_map_unlock(map);
a108bf71 1854 vm_map_entry_release(count);
984263bc
MD
1855 return (rv);
1856}
1857
1858/*
cde87949 1859 * vm_map_wire:
984263bc
MD
1860 *
1861 * Sets the pageability of the specified address
1862 * range in the target map. Regions specified
1863 * as not pageable require locked-down physical
1864 * memory and physical page maps.
1865 *
1866 * The map must not be locked, but a reference
1867 * must remain to the map throughout the call.
a108bf71
MD
1868 *
1869 * This function may be called via the zalloc path and must properly
1870 * reserve map entries for kernel_map.
984263bc
MD
1871 */
1872int
e1359933 1873vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t real_end, int kmflags)
984263bc
MD
1874{
1875 vm_map_entry_t entry;
1876 vm_map_entry_t start_entry;
1877 vm_offset_t end;
1878 int rv = KERN_SUCCESS;
a108bf71 1879 int count;
984263bc
MD
1880 int s;
1881
e1359933 1882 if (kmflags & KM_KRESERVE)
a108bf71 1883 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
a108bf71
MD
1884 else
1885 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1886 vm_map_lock(map);
1887 VM_MAP_RANGE_CHECK(map, start, real_end);
1888 end = real_end;
1889
a108bf71 1890 start_entry = vm_map_clip_range(map, start, end, &count, MAP_CLIP_NO_HOLES);
984263bc
MD
1891 if (start_entry == NULL) {
1892 vm_map_unlock(map);
a108bf71
MD
1893 rv = KERN_INVALID_ADDRESS;
1894 goto failure;
984263bc 1895 }
e1359933 1896 if ((kmflags & KM_PAGEABLE) == 0) {
984263bc
MD
1897 /*
1898 * Wiring.
1899 *
1900 * 1. Holding the write lock, we create any shadow or zero-fill
1901 * objects that need to be created. Then we clip each map
1902 * entry to the region to be wired and increment its wiring
1903 * count. We create objects before clipping the map entries
1904 * to avoid object proliferation.
1905 *
1906 * 2. We downgrade to a read lock, and call vm_fault_wire to
1907 * fault in the pages for any newly wired area (wired_count is
1908 * 1).
1909 *
1910 * Downgrading to a read lock for vm_fault_wire avoids a
1911 * possible deadlock with another process that may have faulted
1912 * on one of the pages to be wired (it would mark the page busy,
1913 * blocking us, then in turn block on the map lock that we
1914 * hold). Because of problems in the recursive lock package,
1915 * we cannot upgrade to a write lock in vm_map_lookup. Thus,
1916 * any actions that require the write lock must be done
1917 * beforehand. Because we keep the read lock on the map, the
1918 * copy-on-write status of the entries we modify here cannot
1919 * change.
1920 */
1921
1922 entry = start_entry;
1923 while ((entry != &map->header) && (entry->start < end)) {
1924 /*
1925 * Trivial case if the entry is already wired
1926 */
1927 if (entry->wired_count) {
1928 entry->wired_count++;
1929 entry = entry->next;
1930 continue;
1931 }
1932
1933 /*
1934 * The entry is being newly wired, we have to setup
1935 * appropriate management structures. A shadow
1936 * object is required for a copy-on-write region,
1937 * or a normal object for a zero-fill region. We
1938 * do not have to do this for entries that point to sub
1939 * maps because we won't hold the lock on the sub map.
1940 */
1941 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1942 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1943 if (copyflag &&
1944 ((entry->protection & VM_PROT_WRITE) != 0)) {
1945
1946 vm_object_shadow(&entry->object.vm_object,
1947 &entry->offset,
1948 atop(entry->end - entry->start));
1949 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1950 } else if (entry->object.vm_object == NULL &&
1951 !map->system_map) {
1952 entry->object.vm_object =
1953 vm_object_allocate(OBJT_DEFAULT,
1954 atop(entry->end - entry->start));
1955 entry->offset = (vm_offset_t) 0;
1956 }
1957 }
1958
1959 entry->wired_count++;
1960 entry = entry->next;
1961 }
1962
1963 /*
1964 * Pass 2.
1965 */
1966
1967 /*
1968 * HACK HACK HACK HACK
1969 *
1970 * Unlock the map to avoid deadlocks. The in-transit flag
1971 * protects us from most changes but note that
1972 * clipping may still occur. To prevent clipping from
1973 * occuring after the unlock, except for when we are
1974 * blocking in vm_fault_wire, we must run at splvm().
1975 * Otherwise our accesses to entry->start and entry->end
1976 * could be corrupted. We have to set splvm() prior to
1977 * unlocking so start_entry does not change out from
1978 * under us at the very beginning of the loop.
1979 *
1980 * HACK HACK HACK HACK
1981 */
1982
1983 s = splvm();
1984 vm_map_unlock(map);
1985
1986 entry = start_entry;
1987 while (entry != &map->header && entry->start < end) {
1988 /*
1989 * If vm_fault_wire fails for any page we need to undo
1990 * what has been done. We decrement the wiring count
1991 * for those pages which have not yet been wired (now)
1992 * and unwire those that have (later).
1993 */
1994 vm_offset_t save_start = entry->start;
1995 vm_offset_t save_end = entry->end;
1996
1997 if (entry->wired_count == 1)
1998 rv = vm_fault_wire(map, entry->start, entry->end);
1999 if (rv) {
2000 CLIP_CHECK_BACK(entry, save_start);
2001 for (;;) {
2002 KASSERT(entry->wired_count == 1, ("wired_count changed unexpectedly"));
2003 entry->wired_count = 0;
2004 if (entry->end == save_end)
2005 break;
2006 entry = entry->next;
2007 KASSERT(entry != &map->header, ("bad entry clip during backout"));
2008 }
2009 end = save_start;
2010 break;
2011 }
2012 CLIP_CHECK_FWD(entry, save_end);
2013 entry = entry->next;
2014 }
2015 splx(s);
2016
2017 /*
2018 * relock. start_entry is still IN_TRANSITION and must
2019 * still exist, but may have been clipped (handled just
2020 * below).
2021 */
2022 vm_map_lock(map);
2023
2024 /*
2025 * If a failure occured undo everything by falling through
2026 * to the unwiring code. 'end' has already been adjusted
2027 * appropriately.
2028 */
2029 if (rv)
e1359933 2030 kmflags |= KM_PAGEABLE;
984263bc
MD
2031
2032 /*
2033 * start_entry might have been clipped if we unlocked the
2034 * map and blocked. No matter how clipped it has gotten
2035 * there should be a fragment that is on our start boundary.
2036 */
2037 CLIP_CHECK_BACK(start_entry, start);
2038 }
2039
e1359933 2040 if (kmflags & KM_PAGEABLE) {
984263bc
MD
2041 /*
2042 * This is the unwiring case. We must first ensure that the
2043 * range to be unwired is really wired down. We know there
2044 * are no holes.
2045 */
2046 entry = start_entry;
2047 while ((entry != &map->header) && (entry->start < end)) {
2048 if (entry->wired_count == 0) {
2049 rv = KERN_INVALID_ARGUMENT;
2050 goto done;
2051 }
2052 entry = entry->next;
2053 }
2054
2055 /*
2056 * Now decrement the wiring count for each region. If a region
2057 * becomes completely unwired, unwire its physical pages and
2058 * mappings.
2059 */
2060 entry = start_entry;
2061 while ((entry != &map->header) && (entry->start < end)) {
2062 entry->wired_count--;
2063 if (entry->wired_count == 0)
2064 vm_fault_unwire(map, entry->start, entry->end);
2065 entry = entry->next;
2066 }
2067 }
2068done:
a108bf71 2069 vm_map_unclip_range(map, start_entry, start, real_end, &count,
984263bc
MD
2070 MAP_CLIP_NO_HOLES);
2071 map->timestamp++;
2072 vm_map_unlock(map);
a108bf71 2073failure:
e1359933 2074 if (kmflags & KM_KRESERVE)
a108bf71 2075 vm_map_entry_krelease(count);
a108bf71
MD
2076 else
2077 vm_map_entry_release(count);
984263bc
MD
2078 return (rv);
2079}
2080
a108bf71
MD
2081/*
2082 * vm_map_set_wired_quick()
2083 *
2084 * Mark a newly allocated address range as wired but do not fault in
2085 * the pages. The caller is expected to load the pages into the object.
2086 *
2087 * The map must be locked on entry and will remain locked on return.
2088 */
2089void
2090vm_map_set_wired_quick(vm_map_t map, vm_offset_t addr, vm_size_t size, int *countp)
2091{
2092 vm_map_entry_t scan;
2093 vm_map_entry_t entry;
2094
2095 entry = vm_map_clip_range(map, addr, addr + size, countp, MAP_CLIP_NO_HOLES);
2096 for (scan = entry; scan != &map->header && scan->start < addr + size; scan = scan->next) {
2097 KKASSERT(entry->wired_count == 0);
2098 entry->wired_count = 1;
2099 }
2100 vm_map_unclip_range(map, entry, addr, addr + size, countp, MAP_CLIP_NO_HOLES);
2101}
2102
984263bc
MD
2103/*
2104 * vm_map_clean
2105 *
2106 * Push any dirty cached pages in the address range to their pager.
2107 * If syncio is TRUE, dirty pages are written synchronously.
2108 * If invalidate is TRUE, any cached pages are freed as well.
2109 *
2110 * Returns an error if any part of the specified range is not mapped.
2111 */
2112int
2113vm_map_clean(map, start, end, syncio, invalidate)
2114 vm_map_t map;
2115 vm_offset_t start;
2116 vm_offset_t end;
2117 boolean_t syncio;
2118 boolean_t invalidate;
2119{
2120 vm_map_entry_t current;
2121 vm_map_entry_t entry;
2122 vm_size_t size;
2123 vm_object_t object;
2124 vm_ooffset_t offset;
2125
2126 vm_map_lock_read(map);
2127 VM_MAP_RANGE_CHECK(map, start, end);
2128 if (!vm_map_lookup_entry(map, start, &entry)) {
2129 vm_map_unlock_read(map);
2130 return (KERN_INVALID_ADDRESS);
2131 }
2132 /*
2133 * Make a first pass to check for holes.
2134 */
2135 for (current = entry; current->start < end; current = current->next) {
2136 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2137 vm_map_unlock_read(map);
2138 return (KERN_INVALID_ARGUMENT);
2139 }
2140 if (end > current->end &&
2141 (current->next == &map->header ||
2142 current->end != current->next->start)) {
2143 vm_map_unlock_read(map);
2144 return (KERN_INVALID_ADDRESS);
2145 }
2146 }
2147
2148 if (invalidate)
2149 pmap_remove(vm_map_pmap(map), start, end);
2150 /*
2151 * Make a second pass, cleaning/uncaching pages from the indicated
2152 * objects as we go.
2153 */
2154 for (current = entry; current->start < end; current = current->next) {
2155 offset = current->offset + (start - current->start);
2156 size = (end <= current->end ? end : current->end) - start;
2157 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2158 vm_map_t smap;
2159 vm_map_entry_t tentry;
2160 vm_size_t tsize;
2161
2162 smap = current->object.sub_map;
2163 vm_map_lock_read(smap);
2164 (void) vm_map_lookup_entry(smap, offset, &tentry);
2165 tsize = tentry->end - offset;
2166 if (tsize < size)
2167 size = tsize;
2168 object = tentry->object.vm_object;
2169 offset = tentry->offset + (offset - tentry->start);
2170 vm_map_unlock_read(smap);
2171 } else {
2172 object = current->object.vm_object;
2173 }
2174 /*
2175 * Note that there is absolutely no sense in writing out
2176 * anonymous objects, so we track down the vnode object
2177 * to write out.
2178 * We invalidate (remove) all pages from the address space
2179 * anyway, for semantic correctness.
2180 *
2181 * note: certain anonymous maps, such as MAP_NOSYNC maps,
2182 * may start out with a NULL object.
2183 */
2184 while (object && object->backing_object) {
2185 object = object->backing_object;
2186 offset += object->backing_object_offset;
2187 if (object->size < OFF_TO_IDX( offset + size))
2188 size = IDX_TO_OFF(object->size) - offset;
2189 }
2190 if (object && (object->type == OBJT_VNODE) &&
2191 (current->protection & VM_PROT_WRITE)) {
2192 /*
2193 * Flush pages if writing is allowed, invalidate them
2194 * if invalidation requested. Pages undergoing I/O
2195 * will be ignored by vm_object_page_remove().
2196 *
2197 * We cannot lock the vnode and then wait for paging
2198 * to complete without deadlocking against vm_fault.
2199 * Instead we simply call vm_object_page_remove() and
2200 * allow it to block internally on a page-by-page
2201 * basis when it encounters pages undergoing async
2202 * I/O.
2203 */
2204 int flags;
2205
2206 vm_object_reference(object);
41a01a4d
MD
2207 vn_lock(object->handle, NULL,
2208 LK_EXCLUSIVE | LK_RETRY, curthread);
984263bc
MD
2209 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
2210 flags |= invalidate ? OBJPC_INVAL : 0;
2211 vm_object_page_clean(object,
2212 OFF_TO_IDX(offset),
2213 OFF_TO_IDX(offset + size + PAGE_MASK),
2214 flags);
41a01a4d 2215 VOP_UNLOCK(object->handle, NULL, 0, curthread);
984263bc
MD
2216 vm_object_deallocate(object);
2217 }
2218 if (object && invalidate &&
2219 ((object->type == OBJT_VNODE) ||
2220 (object->type == OBJT_DEVICE))) {
2221 vm_object_reference(object);
2222 vm_object_page_remove(object,
2223 OFF_TO_IDX(offset),
2224 OFF_TO_IDX(offset + size + PAGE_MASK),
2225 FALSE);
2226 vm_object_deallocate(object);
2227 }
2228 start += size;
2229 }
2230
2231 vm_map_unlock_read(map);
2232 return (KERN_SUCCESS);
2233}
2234
2235/*
2236 * vm_map_entry_unwire: [ internal use only ]
2237 *
2238 * Make the region specified by this entry pageable.
2239 *
2240 * The map in question should be locked.
2241 * [This is the reason for this routine's existence.]
2242 */
2243static void
a108bf71 2244vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
984263bc
MD
2245{
2246 vm_fault_unwire(map, entry->start, entry->end);
2247 entry->wired_count = 0;
2248}
2249
2250/*
2251 * vm_map_entry_delete: [ internal use only ]
2252 *
2253 * Deallocate the given entry from the target map.
2254 */
2255static void
a108bf71 2256vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc
MD
2257{
2258 vm_map_entry_unlink(map, entry);
2259 map->size -= entry->end - entry->start;
2260
2261 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2262 vm_object_deallocate(entry->object.vm_object);
2263 }
2264
a108bf71 2265 vm_map_entry_dispose(map, entry, countp);
984263bc
MD
2266}
2267
2268/*
2269 * vm_map_delete: [ internal use only ]
2270 *
2271 * Deallocates the given address range from the target
2272 * map.
2273 */
2274int
a108bf71 2275vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end, int *countp)
984263bc
MD
2276{
2277 vm_object_t object;
2278 vm_map_entry_t entry;
2279 vm_map_entry_t first_entry;
2280
2281 /*
2282 * Find the start of the region, and clip it
2283 */
2284
2285again:
2286 if (!vm_map_lookup_entry(map, start, &first_entry))
2287 entry = first_entry->next;
2288 else {
2289 entry = first_entry;
a108bf71 2290 vm_map_clip_start(map, entry, start, countp);
984263bc
MD
2291 /*
2292 * Fix the lookup hint now, rather than each time though the
2293 * loop.
2294 */
2295 SAVE_HINT(map, entry->prev);
2296 }
2297
2298 /*
2299 * Save the free space hint
2300 */
2301
2302 if (entry == &map->header) {
2303 map->first_free = &map->header;
2304 } else if (map->first_free->start >= start) {
2305 map->first_free = entry->prev;
2306 }
2307
2308 /*
2309 * Step through all entries in this region
2310 */
2311
2312 while ((entry != &map->header) && (entry->start < end)) {
2313 vm_map_entry_t next;
2314 vm_offset_t s, e;
2315 vm_pindex_t offidxstart, offidxend, count;
2316
2317 /*
2318 * If we hit an in-transition entry we have to sleep and
2319 * retry. It's easier (and not really slower) to just retry
2320 * since this case occurs so rarely and the hint is already
2321 * pointing at the right place. We have to reset the
2322 * start offset so as not to accidently delete an entry
2323 * another process just created in vacated space.
2324 */
2325 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2326 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2327 start = entry->start;
12e4aaff
MD
2328 ++mycpu->gd_cnt.v_intrans_coll;
2329 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
2330 vm_map_transition_wait(map);
2331 goto again;
2332 }
a108bf71 2333 vm_map_clip_end(map, entry, end, countp);
984263bc
MD
2334
2335 s = entry->start;
2336 e = entry->end;
2337 next = entry->next;
2338
2339 offidxstart = OFF_TO_IDX(entry->offset);
2340 count = OFF_TO_IDX(e - s);
2341 object = entry->object.vm_object;
2342
2343 /*
2344 * Unwire before removing addresses from the pmap; otherwise,
2345 * unwiring will put the entries back in the pmap.
2346 */
2347 if (entry->wired_count != 0) {
2348 vm_map_entry_unwire(map, entry);
2349 }
2350
2351 offidxend = offidxstart + count;
2352
2353 if ((object == kernel_object) || (object == kmem_object)) {
2354 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2355 } else {
2356 pmap_remove(map->pmap, s, e);
2357 if (object != NULL &&
2358 object->ref_count != 1 &&
2359 (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING &&
2360 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2361 vm_object_collapse(object);
2362 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2363 if (object->type == OBJT_SWAP) {
2364 swap_pager_freespace(object, offidxstart, count);
2365 }
2366 if (offidxend >= object->size &&
2367 offidxstart < object->size) {
2368 object->size = offidxstart;
2369 }
2370 }
2371 }
2372
2373 /*
2374 * Delete the entry (which may delete the object) only after
2375 * removing all pmap entries pointing to its pages.
2376 * (Otherwise, its page frames may be reallocated, and any
2377 * modify bits will be set in the wrong object!)
2378 */
a108bf71 2379 vm_map_entry_delete(map, entry, countp);
984263bc
MD
2380 entry = next;
2381 }
2382 return (KERN_SUCCESS);
2383}
2384
2385/*
2386 * vm_map_remove:
2387 *
2388 * Remove the given address range from the target map.
2389 * This is the exported form of vm_map_delete.
2390 */
2391int
a108bf71 2392vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
984263bc 2393{
03aa8d99 2394 int result;
a108bf71 2395 int count;
984263bc 2396
a108bf71 2397 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2398 vm_map_lock(map);
2399 VM_MAP_RANGE_CHECK(map, start, end);
a108bf71 2400 result = vm_map_delete(map, start, end, &count);
984263bc 2401 vm_map_unlock(map);
a108bf71 2402 vm_map_entry_release(count);
984263bc 2403
984263bc
MD
2404 return (result);
2405}
2406
2407/*
2408 * vm_map_check_protection:
2409 *
2410 * Assert that the target map allows the specified
2411 * privilege on the entire address region given.
2412 * The entire region must be allocated.
2413 */
2414boolean_t
2415vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2416 vm_prot_t protection)
2417{
2418 vm_map_entry_t entry;
2419 vm_map_entry_t tmp_entry;
2420
2421 if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2422 return (FALSE);
2423 }
2424 entry = tmp_entry;
2425
2426 while (start < end) {
2427 if (entry == &map->header) {
2428 return (FALSE);
2429 }
2430 /*
2431 * No holes allowed!
2432 */
2433
2434 if (start < entry->start) {
2435 return (FALSE);
2436 }
2437 /*
2438 * Check protection associated with entry.
2439 */
2440
2441 if ((entry->protection & protection) != protection) {
2442 return (FALSE);
2443 }
2444 /* go to next entry */
2445
2446 start = entry->end;
2447 entry = entry->next;
2448 }
2449 return (TRUE);
2450}
2451
2452/*
2453 * Split the pages in a map entry into a new object. This affords
2454 * easier removal of unused pages, and keeps object inheritance from
2455 * being a negative impact on memory usage.
2456 */
2457static void
a108bf71 2458vm_map_split(vm_map_entry_t entry)
984263bc
MD
2459{
2460 vm_page_t m;
2461 vm_object_t orig_object, new_object, source;
2462 vm_offset_t s, e;
2463 vm_pindex_t offidxstart, offidxend, idx;
2464 vm_size_t size;
2465 vm_ooffset_t offset;
2466
2467 orig_object = entry->object.vm_object;
2468 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
2469 return;
2470 if (orig_object->ref_count <= 1)
2471 return;
2472
2473 offset = entry->offset;
2474 s = entry->start;
2475 e = entry->end;
2476
2477 offidxstart = OFF_TO_IDX(offset);
2478 offidxend = offidxstart + OFF_TO_IDX(e - s);
2479 size = offidxend - offidxstart;
2480
2481 new_object = vm_pager_allocate(orig_object->type,
2482 NULL, IDX_TO_OFF(size), VM_PROT_ALL, 0LL);
2483 if (new_object == NULL)
2484 return;
2485
2486 source = orig_object->backing_object;
2487 if (source != NULL) {
2488 vm_object_reference(source); /* Referenced by new_object */
2489 LIST_INSERT_HEAD(&source->shadow_head,
2490 new_object, shadow_list);
2491 vm_object_clear_flag(source, OBJ_ONEMAPPING);
2492 new_object->backing_object_offset =
2493 orig_object->backing_object_offset + IDX_TO_OFF(offidxstart);
2494 new_object->backing_object = source;
2495 source->shadow_count++;
2496 source->generation++;
2497 }
2498
2499 for (idx = 0; idx < size; idx++) {
2500 vm_page_t m;
2501
2502 retry:
2503 m = vm_page_lookup(orig_object, offidxstart + idx);
2504 if (m == NULL)
2505 continue;
2506
2507 /*
2508 * We must wait for pending I/O to complete before we can
2509 * rename the page.
2510 *
2511 * We do not have to VM_PROT_NONE the page as mappings should
2512 * not be changed by this operation.
2513 */
2514 if (vm_page_sleep_busy(m, TRUE, "spltwt"))
2515 goto retry;
2516
2517 vm_page_busy(m);
2518 vm_page_rename(m, new_object, idx);
2519 /* page automatically made dirty by rename and cache handled */
2520 vm_page_busy(m);
2521 }
2522
2523 if (orig_object->type == OBJT_SWAP) {
2524 vm_object_pip_add(orig_object, 1);
2525 /*
2526 * copy orig_object pages into new_object
2527 * and destroy unneeded pages in
2528 * shadow object.
2529 */
2530 swap_pager_copy(orig_object, new_object, offidxstart, 0);
2531 vm_object_pip_wakeup(orig_object);
2532 }
2533
2534 for (idx = 0; idx < size; idx++) {
2535 m = vm_page_lookup(new_object, idx);
2536 if (m) {
2537 vm_page_wakeup(m);
2538 }
2539 }
2540
2541 entry->object.vm_object = new_object;
2542 entry->offset = 0LL;
2543 vm_object_deallocate(orig_object);
2544}
2545
2546/*
2547 * vm_map_copy_entry:
2548 *
2549 * Copies the contents of the source entry to the destination
2550 * entry. The entries *must* be aligned properly.
2551 */
2552static void
a108bf71
MD
2553vm_map_copy_entry(vm_map_t src_map, vm_map_t dst_map,
2554 vm_map_entry_t src_entry, vm_map_entry_t dst_entry)
984263bc
MD
2555{
2556 vm_object_t src_object;
2557
2558 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2559 return;
2560
2561 if (src_entry->wired_count == 0) {
2562
2563 /*
2564 * If the source entry is marked needs_copy, it is already
2565 * write-protected.
2566 */
2567 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2568 pmap_protect(src_map->pmap,
2569 src_entry->start,
2570 src_entry->end,
2571 src_entry->protection & ~VM_PROT_WRITE);
2572 }
2573
2574 /*
2575 * Make a copy of the object.
2576 */
2577 if ((src_object = src_entry->object.vm_object) != NULL) {
2578
2579 if ((src_object->handle == NULL) &&
2580 (src_object->type == OBJT_DEFAULT ||
2581 src_object->type == OBJT_SWAP)) {
2582 vm_object_collapse(src_object);
2583 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2584 vm_map_split(src_entry);
2585 src_object = src_entry->object.vm_object;
2586 }
2587 }
2588
2589 vm_object_reference(src_object);
2590 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2591 dst_entry->object.vm_object = src_object;
2592 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2593 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2594 dst_entry->offset = src_entry->offset;
2595 } else {
2596 dst_entry->object.vm_object = NULL;
2597 dst_entry->offset = 0;
2598 }
2599
2600 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2601 dst_entry->end - dst_entry->start, src_entry->start);
2602 } else {
2603 /*
2604 * Of course, wired down pages can't be set copy-on-write.
2605 * Cause wired pages to be copied into the new map by
2606 * simulating faults (the new pages are pageable)
2607 */
2608 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2609 }
2610}
2611
2612/*
2613 * vmspace_fork:
2614 * Create a new process vmspace structure and vm_map
2615 * based on those of an existing process. The new map
2616 * is based on the old map, according to the inheritance
2617 * values on the regions in that map.
2618 *
2619 * The source map must not be locked.
2620 */
2621struct vmspace *
a108bf71 2622vmspace_fork(struct vmspace *vm1)
984263bc
MD
2623{
2624 struct vmspace *vm2;
2625 vm_map_t old_map = &vm1->vm_map;
2626 vm_map_t new_map;
2627 vm_map_entry_t old_entry;
2628 vm_map_entry_t new_entry;
2629 vm_object_t object;
a108bf71 2630 int count;
984263bc
MD
2631
2632 vm_map_lock(old_map);
2633 old_map->infork = 1;
2634
239b4df9
MD
2635 /*
2636 * XXX Note: upcalls are not copied.
2637 */
984263bc
MD
2638 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
2639 bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
239b4df9 2640 (caddr_t)&vm1->vm_endcopy - (caddr_t)&vm1->vm_startcopy);
984263bc
MD
2641 new_map = &vm2->vm_map; /* XXX */
2642 new_map->timestamp = 1;
2643
a108bf71 2644 count = 0;
984263bc 2645 old_entry = old_map->header.next;
a108bf71
MD
2646 while (old_entry != &old_map->header) {
2647 ++count;
2648 old_entry = old_entry->next;
2649 }
984263bc 2650
a108bf71
MD
2651 count = vm_map_entry_reserve(count + MAP_RESERVE_COUNT);
2652
2653 old_entry = old_map->header.next;
984263bc
MD
2654 while (old_entry != &old_map->header) {
2655 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
2656 panic("vm_map_fork: encountered a submap");
2657
2658 switch (old_entry->inheritance) {
2659 case VM_INHERIT_NONE:
2660 break;
2661
2662 case VM_INHERIT_SHARE:
2663 /*
2664 * Clone the entry, creating the shared object if necessary.
2665 */
2666 object = old_entry->object.vm_object;
2667 if (object == NULL) {
2668 object = vm_object_allocate(OBJT_DEFAULT,
2669 atop(old_entry->end - old_entry->start));
2670 old_entry->object.vm_object = object;
2671 old_entry->offset = (vm_offset_t) 0;
2672 }
2673
2674 /*
2675 * Add the reference before calling vm_object_shadow
2676 * to insure that a shadow object is created.
2677 */
2678 vm_object_reference(object);
2679 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2680 vm_object_shadow(&old_entry->object.vm_object,
2681 &old_entry->offset,
2682 atop(old_entry->end - old_entry->start));
2683 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2684 /* Transfer the second reference too. */
2685 vm_object_reference(
2686 old_entry->object.vm_object);
2687 vm_object_deallocate(object);
2688 object = old_entry->object.vm_object;
2689 }
2690 vm_object_clear_flag(object, OBJ_ONEMAPPING);
2691
2692 /*
2693 * Clone the entry, referencing the shared object.
2694 */
a108bf71 2695 new_entry = vm_map_entry_create(new_map, &count);
984263bc
MD
2696 *new_entry = *old_entry;
2697 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2698 new_entry->wired_count = 0;
2699
2700 /*
2701 * Insert the entry into the new map -- we know we're
2702 * inserting at the end of the new map.
2703 */
2704
2705 vm_map_entry_link(new_map, new_map->header.prev,
2706 new_entry);
2707
2708 /*
2709 * Update the physical map
2710 */
2711
2712 pmap_copy(new_map->pmap, old_map->pmap,
2713 new_entry->start,
2714 (old_entry->end - old_entry->start),
2715 old_entry->start);
2716 break;
2717
2718 case VM_INHERIT_COPY:
2719 /*
2720 * Clone the entry and link into the map.
2721 */
a108bf71 2722 new_entry = vm_map_entry_create(new_map, &count);
984263bc
MD
2723 *new_entry = *old_entry;
2724 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2725 new_entry->wired_count = 0;
2726 new_entry->object.vm_object = NULL;
2727 vm_map_entry_link(new_map, new_map->header.prev,
2728 new_entry);
2729 vm_map_copy_entry(old_map, new_map, old_entry,
2730 new_entry);
2731 break;
2732 }
2733 old_entry = old_entry->next;
2734 }
2735
2736 new_map->size = old_map->size;
2737 old_map->infork = 0;
2738 vm_map_unlock(old_map);
a108bf71 2739 vm_map_entry_release(count);
984263bc
MD
2740
2741 return (vm2);
2742}
2743
2744int
2745vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
2746 vm_prot_t prot, vm_prot_t max, int cow)
2747{
2748 vm_map_entry_t prev_entry;
2749 vm_map_entry_t new_stack_entry;
2750 vm_size_t init_ssize;
2751 int rv;
a108bf71 2752 int count;
984263bc
MD
2753
2754 if (VM_MIN_ADDRESS > 0 && addrbos < VM_MIN_ADDRESS)
2755 return (KERN_NO_SPACE);
2756
2757 if (max_ssize < sgrowsiz)
2758 init_ssize = max_ssize;
2759 else
2760 init_ssize = sgrowsiz;
2761
a108bf71 2762 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2763 vm_map_lock(map);
2764
2765 /* If addr is already mapped, no go */
2766 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
2767 vm_map_unlock(map);
a108bf71 2768 vm_map_entry_release(count);
984263bc
MD
2769 return (KERN_NO_SPACE);
2770 }
2771
2772 /* If we would blow our VMEM resource limit, no go */
2773 if (map->size + init_ssize >
2774 curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
2775 vm_map_unlock(map);
a108bf71 2776 vm_map_entry_release(count);
984263bc
MD
2777 return (KERN_NO_SPACE);
2778 }
2779
2780 /* If we can't accomodate max_ssize in the current mapping,
2781 * no go. However, we need to be aware that subsequent user
2782 * mappings might map into the space we have reserved for
2783 * stack, and currently this space is not protected.
2784 *
2785 * Hopefully we will at least detect this condition
2786 * when we try to grow the stack.
2787 */
2788 if ((prev_entry->next != &map->header) &&
2789 (prev_entry->next->start < addrbos + max_ssize)) {
2790 vm_map_unlock(map);
a108bf71 2791 vm_map_entry_release(count);
984263bc
MD
2792 return (KERN_NO_SPACE);
2793 }
2794
2795 /* We initially map a stack of only init_ssize. We will
2796 * grow as needed later. Since this is to be a grow
2797 * down stack, we map at the top of the range.
2798 *
2799 * Note: we would normally expect prot and max to be
2800 * VM_PROT_ALL, and cow to be 0. Possibly we should
2801 * eliminate these as input parameters, and just
2802 * pass these values here in the insert call.
2803 */
a108bf71
MD
2804 rv = vm_map_insert(map, &count,
2805 NULL, 0, addrbos + max_ssize - init_ssize,
984263bc
MD
2806 addrbos + max_ssize, prot, max, cow);
2807
2808 /* Now set the avail_ssize amount */
2809 if (rv == KERN_SUCCESS){
2810 if (prev_entry != &map->header)
a108bf71 2811 vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize, &count);
984263bc
MD
2812 new_stack_entry = prev_entry->next;
2813 if (new_stack_entry->end != addrbos + max_ssize ||
2814 new_stack_entry->start != addrbos + max_ssize - init_ssize)
2815 panic ("Bad entry start/end for new stack entry");
2816 else
2817 new_stack_entry->avail_ssize = max_ssize - init_ssize;
2818 }
2819
2820 vm_map_unlock(map);
a108bf71 2821 vm_map_entry_release(count);
984263bc
MD
2822 return (rv);
2823}
2824
2825/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
2826 * desired address is already mapped, or if we successfully grow
2827 * the stack. Also returns KERN_SUCCESS if addr is outside the
2828 * stack range (this is strange, but preserves compatibility with
2829 * the grow function in vm_machdep.c).
2830 */
2831int
2832vm_map_growstack (struct proc *p, vm_offset_t addr)
2833{
2834 vm_map_entry_t prev_entry;
2835 vm_map_entry_t stack_entry;
2836 vm_map_entry_t new_stack_entry;
2837 struct vmspace *vm = p->p_vmspace;
2838 vm_map_t map = &vm->vm_map;
2839 vm_offset_t end;
a108bf71
MD
2840 int grow_amount;
2841 int rv = KERN_SUCCESS;
2842 int is_procstack;
2843 int use_read_lock = 1;
2844 int count;
984263bc 2845
a108bf71 2846 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2847Retry:
2848 if (use_read_lock)
2849 vm_map_lock_read(map);
2850 else
2851 vm_map_lock(map);
2852
2853 /* If addr is already in the entry range, no need to grow.*/
2854 if (vm_map_lookup_entry(map, addr, &prev_entry))
2855 goto done;
2856
2857 if ((stack_entry = prev_entry->next) == &map->header)
2858 goto done;
2859 if (prev_entry == &map->header)
2860 end = stack_entry->start - stack_entry->avail_ssize;
2861 else
2862 end = prev_entry->end;
2863
2864 /* This next test mimics the old grow function in vm_machdep.c.
2865 * It really doesn't quite make sense, but we do it anyway
2866 * for compatibility.
2867 *
2868 * If not growable stack, return success. This signals the
2869 * caller to proceed as he would normally with normal vm.
2870 */
2871 if (stack_entry->avail_ssize < 1 ||
2872 addr >= stack_entry->start ||
2873 addr < stack_entry->start - stack_entry->avail_ssize) {
2874 goto done;
2875 }
2876
2877 /* Find the minimum grow amount */
2878 grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
2879 if (grow_amount > stack_entry->avail_ssize) {
2880 rv = KERN_NO_SPACE;
2881 goto done;
2882 }
2883
2884 /* If there is no longer enough space between the entries
2885 * nogo, and adjust the available space. Note: this
2886 * should only happen if the user has mapped into the
2887 * stack area after the stack was created, and is
2888 * probably an error.
2889 *
2890 * This also effectively destroys any guard page the user
2891 * might have intended by limiting the stack size.
2892 */
2893 if (grow_amount > stack_entry->start - end) {
2894 if (use_read_lock && vm_map_lock_upgrade(map)) {
2895 use_read_lock = 0;
2896 goto Retry;
2897 }
2898 use_read_lock = 0;
2899 stack_entry->avail_ssize = stack_entry->start - end;
2900 rv = KERN_NO_SPACE;
2901 goto done;
2902 }
2903
2904 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
2905
2906 /* If this is the main process stack, see if we're over the
2907 * stack limit.
2908 */
2909 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2910 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2911 rv = KERN_NO_SPACE;
2912 goto done;
2913 }
2914
2915 /* Round up the grow amount modulo SGROWSIZ */
2916 grow_amount = roundup (grow_amount, sgrowsiz);
2917 if (grow_amount > stack_entry->avail_ssize) {
2918 grow_amount = stack_entry->avail_ssize;
2919 }
2920 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2921 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2922 grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
2923 ctob(vm->vm_ssize);
2924 }
2925
2926 /* If we would blow our VMEM resource limit, no go */
be77b5f9 2927 if (map->size + grow_amount > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
984263bc
MD
2928 rv = KERN_NO_SPACE;
2929 goto done;
2930 }
2931
2932 if (use_read_lock && vm_map_lock_upgrade(map)) {
2933 use_read_lock = 0;
2934 goto Retry;
2935 }
2936 use_read_lock = 0;
2937
2938 /* Get the preliminary new entry start value */
2939 addr = stack_entry->start - grow_amount;
2940
2941 /* If this puts us into the previous entry, cut back our growth
2942 * to the available space. Also, see the note above.
2943 */
2944 if (addr < end) {
2945 stack_entry->avail_ssize = stack_entry->start - end;
2946 addr = end;
2947 }
2948
a108bf71
MD
2949 rv = vm_map_insert(map, &count,
2950 NULL, 0, addr, stack_entry->start,
984263bc
MD
2951 VM_PROT_ALL,
2952 VM_PROT_ALL,
2953 0);
2954
2955 /* Adjust the available stack space by the amount we grew. */
2956 if (rv == KERN_SUCCESS) {
2957 if (prev_entry != &map->header)
a108bf71 2958 vm_map_clip_end(map, prev_entry, addr, &count);
984263bc
MD
2959 new_stack_entry = prev_entry->next;
2960 if (new_stack_entry->end != stack_entry->start ||
2961 new_stack_entry->start != addr)
2962 panic ("Bad stack grow start/end in new stack entry");
2963 else {
2964 new_stack_entry->avail_ssize = stack_entry->avail_ssize -
2965 (new_stack_entry->end -
2966 new_stack_entry->start);
2967 if (is_procstack)
2968 vm->vm_ssize += btoc(new_stack_entry->end -
2969 new_stack_entry->start);
2970 }
2971 }
2972
2973done:
2974 if (use_read_lock)
2975 vm_map_unlock_read(map);
2976 else
2977 vm_map_unlock(map);
a108bf71 2978 vm_map_entry_release(count);
984263bc
MD
2979 return (rv);
2980}
2981
2982/*
2983 * Unshare the specified VM space for exec. If other processes are
2984 * mapped to it, then create a new one. The new vmspace is null.
2985 */
2986
2987void
29802dbb 2988vmspace_exec(struct proc *p, struct vmspace *vmcopy)
a108bf71 2989{
984263bc
MD
2990 struct vmspace *oldvmspace = p->p_vmspace;
2991 struct vmspace *newvmspace;
2992 vm_map_t map = &p->p_vmspace->vm_map;
2993
29802dbb
MD
2994 /*
2995 * If we are execing a resident vmspace we fork it, otherwise
239b4df9
MD
2996 * we create a new vmspace. Note that exitingcnt and upcalls
2997 * are not copied to the new vmspace.
29802dbb
MD
2998 */
2999 if (vmcopy) {
3000 newvmspace = vmspace_fork(vmcopy);
3001 } else {
3002 newvmspace = vmspace_alloc(map->min_offset, map->max_offset);
3003 bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
239b4df9
MD
3004 (caddr_t)&oldvmspace->vm_endcopy -
3005 (caddr_t)&oldvmspace->vm_startcopy);
29802dbb
MD
3006 }
3007
984263bc
MD
3008 /*
3009 * This code is written like this for prototype purposes. The
3010 * goal is to avoid running down the vmspace here, but let the
3011 * other process's that are still using the vmspace to finally
3012 * run it down. Even though there is little or no chance of blocking
3013 * here, it is a good idea to keep this form for future mods.
3014 */
984263bc
MD
3015 p->p_vmspace = newvmspace;
3016 pmap_pinit2(vmspace_pmap(newvmspace));
3017 if (p == curproc)
3018 pmap_activate(p);
29802dbb 3019 vmspace_free(oldvmspace);
984263bc
MD
3020}
3021
3022/*
3023 * Unshare the specified VM space for forcing COW. This
3024 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
239b4df9
MD
3025 *
3026 * The exitingcnt test is not strictly necessary but has been
3027 * included for code sanity (to make the code a bit more deterministic).
984263bc
MD
3028 */
3029
3030void
a108bf71
MD
3031vmspace_unshare(struct proc *p)
3032{
984263bc
MD
3033 struct vmspace *oldvmspace = p->p_vmspace;
3034 struct vmspace *newvmspace;
3035
239b4df9 3036 if (oldvmspace->vm_refcnt == 1 && oldvmspace->vm_exitingcnt == 0)
984263bc
MD
3037 return;
3038 newvmspace = vmspace_fork(oldvmspace);
984263bc
MD
3039 p->p_vmspace = newvmspace;
3040 pmap_pinit2(vmspace_pmap(newvmspace));
3041 if (p == curproc)
3042 pmap_activate(p);
29802dbb 3043 vmspace_free(oldvmspace);
984263bc 3044}
984263bc
MD
3045
3046/*
3047 * vm_map_lookup:
3048 *
3049 * Finds the VM object, offset, and
3050 * protection for a given virtual address in the
3051 * specified map, assuming a page fault of the
3052 * type specified.
3053 *
3054 * Leaves the map in question locked for read; return
3055 * values are guaranteed until a vm_map_lookup_done
3056 * call is performed. Note that the map argument
3057 * is in/out; the returned map must be used in
3058 * the call to vm_map_lookup_done.
3059 *
3060 * A handle (out_entry) is returned for use in
3061 * vm_map_lookup_done, to make that fast.
3062 *
3063 * If a lookup is requested with "write protection"
3064 * specified, the map may be changed to perform virtual
3065 * copying operations, although the data referenced will
3066 * remain the same.
3067 */
3068int
3069vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3070 vm_offset_t vaddr,
3071 vm_prot_t fault_typea,
3072 vm_map_entry_t *out_entry, /* OUT */
3073 vm_object_t *object, /* OUT */
3074 vm_pindex_t *pindex, /* OUT */
3075 vm_prot_t *out_prot, /* OUT */
3076 boolean_t *wired) /* OUT */
3077{
3078 vm_map_entry_t entry;
3079 vm_map_t map = *var_map;
3080 vm_prot_t prot;
3081 vm_prot_t fault_type = fault_typea;
3082 int use_read_lock = 1;
3083 int rv = KERN_SUCCESS;
3084
3085RetryLookup:
3086 if (use_read_lock)
3087 vm_map_lock_read(map);
3088 else
3089 vm_map_lock(map);
3090
3091 /*
3092 * If the map has an interesting hint, try it before calling full
3093 * blown lookup routine.
3094 */
3095 entry = map->hint;
3096 *out_entry = entry;
3097
3098 if ((entry == &map->header) ||
3099 (vaddr < entry->start) || (vaddr >= entry->end)) {
3100 vm_map_entry_t tmp_entry;
3101
3102 /*
3103 * Entry was either not a valid hint, or the vaddr was not
3104 * contained in the entry, so do a full lookup.
3105 */
3106 if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) {
3107 rv = KERN_INVALID_ADDRESS;
3108 goto done;
3109 }
3110
3111 entry = tmp_entry;
3112 *out_entry = entry;
3113 }
3114
3115 /*
3116 * Handle submaps.
3117 */
3118
3119 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3120 vm_map_t old_map = map;
3121
3122 *var_map = map = entry->object.sub_map;
3123 if (use_read_lock)
3124 vm_map_unlock_read(old_map);
3125 else
3126 vm_map_unlock(old_map);
3127 use_read_lock = 1;
3128 goto RetryLookup;
3129 }
3130
3131 /*
3132 * Check whether this task is allowed to have this page.
3133 * Note the special case for MAP_ENTRY_COW
3134 * pages with an override. This is to implement a forced
3135 * COW for debuggers.
3136 */
3137
3138 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3139 prot = entry->max_protection;
3140 else
3141 prot = entry->protection;
3142
3143 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3144 if ((fault_type & prot) != fault_type) {
3145 rv = KERN_PROTECTION_FAILURE;
3146 goto done;
3147 }
3148
3149 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3150 (entry->eflags & MAP_ENTRY_COW) &&
3151 (fault_type & VM_PROT_WRITE) &&
3152 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3153 rv = KERN_PROTECTION_FAILURE;
3154 goto done;
3155 }
3156
3157 /*
3158 * If this page is not pageable, we have to get it for all possible
3159 * accesses.
3160 */
3161
3162 *wired = (entry->wired_count != 0);
3163 if (*wired)
3164 prot = fault_type = entry->protection;
3165
3166 /*
3167 * If the entry was copy-on-write, we either ...
3168 */
3169
3170 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3171 /*
3172 * If we want to write the page, we may as well handle that
3173 * now since we've got the map locked.
3174 *
3175 * If we don't need to write the page, we just demote the
3176 * permissions allowed.
3177 */
3178
3179 if (fault_type & VM_PROT_WRITE) {
3180 /*
3181 * Make a new object, and place it in the object
3182 * chain. Note that no new references have appeared
3183 * -- one just moved from the map to the new
3184 * object.
3185 */
3186
3187 if (use_read_lock && vm_map_lock_upgrade(map)) {
3188 use_read_lock = 0;
3189 goto RetryLookup;
3190 }
3191 use_read_lock = 0;
3192
3193 vm_object_shadow(
3194 &entry->object.vm_object,
3195 &entry->offset,
3196 atop(entry->end - entry->start));
3197
3198 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3199 } else {
3200 /*
3201 * We're attempting to read a copy-on-write page --
3202 * don't allow writes.
3203 */
3204
3205 prot &= ~VM_PROT_WRITE;
3206 }
3207 }
3208
3209 /*
3210 * Create an object if necessary.
3211 */
3212 if (entry->object.vm_object == NULL &&
3213 !map->system_map) {
3214 if (use_read_lock && vm_map_lock_upgrade(map)) {
3215 use_read_lock = 0;
3216 goto RetryLookup;
3217 }
3218 use_read_lock = 0;
3219 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3220 atop(entry->end - entry->start));
3221 entry->offset = 0;
3222 }
3223
3224 /*
3225 * Return the object/offset from this entry. If the entry was
3226 * copy-on-write or empty, it has been fixed up.
3227 */
3228
3229 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3230 *object = entry->object.vm_object;
3231
3232 /*
3233 * Return whether this is the only map sharing this data. On
3234 * success we return with a read lock held on the map. On failure
3235 * we return with the map unlocked.
3236 */
3237 *out_prot = prot;
3238done:
3239 if (rv == KERN_SUCCESS) {
3240 if (use_read_lock == 0)
3241 vm_map_lock_downgrade(map);
3242 } else if (use_read_lock) {
3243 vm_map_unlock_read(map);
3244 } else {
3245 vm_map_unlock(map);
3246 }
3247 return (rv);
3248}
3249
3250/*
3251 * vm_map_lookup_done:
3252 *
3253 * Releases locks acquired by a vm_map_lookup
3254 * (according to the handle returned by that lookup).
3255 */
3256
3257void
a108bf71 3258vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry, int count)
984263bc
MD
3259{
3260 /*
3261 * Unlock the main-level map
3262 */
984263bc 3263 vm_map_unlock_read(map);
a108bf71
MD
3264 if (count)
3265 vm_map_entry_release(count);
984263bc
MD
3266}
3267
cfbd9dd0
MD
3268#ifdef ENABLE_VFS_IOOPT
3269
984263bc
MD
3270/*
3271 * Implement uiomove with VM operations. This handles (and collateral changes)
3272 * support every combination of source object modification, and COW type
3273 * operations.
3274 */
3275int
3276vm_uiomove(mapa, srcobject, cp, cnta, uaddra, npages)
3277 vm_map_t mapa;
3278 vm_object_t srcobject;
3279 off_t cp;
3280 int cnta;
3281 vm_offset_t uaddra;
3282 int *npages;
3283{
3284 vm_map_t map;
3285 vm_object_t first_object, oldobject, object;
3286 vm_map_entry_t entry;
3287 vm_prot_t prot;
3288 boolean_t wired;
3289 int tcnt, rv;
3290 vm_offset_t uaddr, start, end, tend;
3291 vm_pindex_t first_pindex, osize, oindex;
3292 off_t ooffset;
3293 int cnt;
a108bf71 3294 int count;
984263bc
MD
3295
3296 if (npages)
3297 *npages = 0;
3298
3299 cnt = cnta;
3300 uaddr = uaddra;
3301
3302 while (cnt > 0) {
3303 map = mapa;
3304
a108bf71
MD
3305 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
3306
984263bc
MD
3307 if ((vm_map_lookup(&map, uaddr,
3308 VM_PROT_READ, &entry, &first_object,
3309 &first_pindex, &prot, &wired)) != KERN_SUCCESS) {
3310 return EFAULT;
3311 }
3312
a108bf71 3313 vm_map_clip_start(map, entry, uaddr, &count);
984263bc
MD
3314
3315 tcnt = cnt;
3316 tend = uaddr + tcnt;
3317 if (tend > entry->end) {
3318 tcnt = entry->end - uaddr;
3319 tend = entry->end;
3320 }
3321
a108bf71 3322 vm_map_clip_end(map, entry, tend, &count);
984263bc
MD
3323
3324 start = entry->start;
3325 end = entry->end;
3326
3327 osize = atop(tcnt);
3328
3329 oindex = OFF_TO_IDX(cp);
3330 if (npages) {
3331 vm_pindex_t idx;
3332 for (idx = 0; idx < osize; idx++) {
3333 vm_page_t m;
3334 if ((m = vm_page_lookup(srcobject, oindex + idx)) == NULL) {
a108bf71 3335 vm_map_lookup_done(map, entry, count);
984263bc
MD
3336 return 0;
3337 }
3338 /*
3339 * disallow busy or invalid pages, but allow
3340 * m->busy pages if they are entirely valid.
3341 */
3342 if ((m->flags & PG_BUSY) ||
3343 ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL)) {
a108bf71 3344 vm_map_lookup_done(map, entry, count);
984263bc
MD
3345 return 0;
3346 }
3347 }
3348 }
3349
3350/*
3351 * If we are changing an existing map entry, just redirect
3352 * the object, and change mappings.
3353 */
3354 if ((first_object->type == OBJT_VNODE) &&
3355 ((oldobject = entry->object.vm_object) == first_object)) {
3356
3357 if ((entry->offset != cp) || (oldobject != srcobject)) {
3358 /*
3359 * Remove old window into the file
3360 */
3361 pmap_remove (map->pmap, uaddr, tend);
3362
3363 /*
3364 * Force copy on write for mmaped regions
3365 */
3366 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3367
3368 /*
3369 * Point the object appropriately
3370 */
3371 if (oldobject != srcobject) {
3372
3373 /*
3374 * Set the object optimization hint flag
3375 */
3376 vm_object_set_flag(srcobject, OBJ_OPT);
3377 vm_object_reference(srcobject);
3378 entry->object.vm_object = srcobject;
3379
3380 if (oldobject) {
3381 vm_object_deallocate(oldobject);
3382 }
3383 }
3384
3385 entry->offset = cp;
3386 map->timestamp++;
3387 } else {
3388 pmap_remove (map->pmap, uaddr, tend);
3389 }
3390
3391 } else if ((first_object->ref_count == 1) &&
3392 (first_object->size == osize) &&
3393 ((first_object->type == OBJT_DEFAULT) ||
3394 (first_object->type == OBJT_SWAP)) ) {
3395
3396 oldobject = first_object->backing_object;
3397
3398 if ((first_object->backing_object_offset != cp) ||
3399 (oldobject != srcobject)) {
3400 /*
3401 * Remove old window into the file
3402 */
3403 pmap_remove (map->pmap, uaddr, tend);
3404
3405 /*
3406 * Remove unneeded old pages
3407 */
3408 vm_object_page_remove(first_object, 0, 0, 0);
3409
3410 /*
3411 * Invalidate swap space
3412 */
3413 if (first_object->type == OBJT_SWAP) {
3414 swap_pager_freespace(first_object,
3415 0,
3416 first_object->size);
3417 }
3418
3419 /*
3420 * Force copy on write for mmaped regions
3421 */
3422 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3423
3424 /*
3425 * Point the object appropriately
3426 */
3427 if (oldobject != srcobject) {
3428
3429 /*
3430 * Set the object optimization hint flag
3431 */
3432 vm_object_set_flag(srcobject, OBJ_OPT);
3433 vm_object_reference(srcobject);
3434
3435 if (oldobject) {
3436 LIST_REMOVE(
3437 first_object, shadow_list);
3438 oldobject->shadow_count--;
3439 /* XXX bump generation? */
3440 vm_object_deallocate(oldobject);
3441 }
3442
3443 LIST_INSERT_HEAD(&srcobject->shadow_head,
3444 first_object, shadow_list);
3445 srcobject->shadow_count++;
3446 /* XXX bump generation? */
3447
3448 first_object->backing_object = srcobject;
3449 }
3450 first_object->backing_object_offset = cp;
3451 map->timestamp++;
3452 } else {
3453 pmap_remove (map->pmap, uaddr, tend);
3454 }
3455/*
3456 * Otherwise, we have to do a logical mmap.
3457 */
3458 } else {
3459
3460 vm_object_set_flag(srcobject, OBJ_OPT);
3461 vm_object_reference(srcobject);
3462
3463 pmap_remove (map->pmap, uaddr, tend);
3464
3465 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3466 vm_map_lock_upgrade(map);
3467
3468 if (entry == &map->header) {
3469 map->first_free = &map->header;
3470 } else if (map->first_free->start >= start) {
3471 map->first_free = entry->prev;
3472 }
3473
3474 SAVE_HINT(map, entry->prev);
a108bf71 3475 vm_map_entry_delete(map, entry, &count);
984263bc
MD
3476
3477 object = srcobject;
3478 ooffset = cp;
3479
a108bf71
MD
3480 rv = vm_map_insert(map, &count,
3481 object, ooffset, start, tend,
984263bc
MD
3482 VM_PROT_ALL, VM_PROT_ALL, MAP_COPY_ON_WRITE);
3483
3484 if (rv != KERN_SUCCESS)
3485 panic("vm_uiomove: could not insert new entry: %d", rv);
3486 }
3487
3488/*
3489 * Map the window directly, if it is already in memory
3490 */
3491 pmap_object_init_pt(map->pmap, uaddr,
3492 srcobject, oindex, tcnt, 0);
3493
3494 map->timestamp++;
3495 vm_map_unlock(map);
a108bf71 3496 vm_map_entry_release(count);
984263bc
MD
3497
3498 cnt -= tcnt;
3499 uaddr += tcnt;
3500 cp += tcnt;
3501 if (npages)
3502 *npages += osize;
3503 }
3504 return 0;
3505}
3506
cfbd9dd0
MD
3507#endif
3508
984263bc
MD
3509/*
3510 * Performs the copy_on_write operations necessary to allow the virtual copies
3511 * into user space to work. This has to be called for write(2) system calls
3512 * from other processes, file unlinking, and file size shrinkage.
3513 */
3514void
3515vm_freeze_copyopts(object, froma, toa)
3516 vm_object_t object;
3517 vm_pindex_t froma, toa;
3518{
3519 int rv;
3520 vm_object_t robject;
3521 vm_pindex_t idx;
3522
3523 if ((object == NULL) ||
3524 ((object->flags & OBJ_OPT) == 0))
3525 return;
3526
3527 if (object->shadow_count > object->ref_count)
3528 panic("vm_freeze_copyopts: sc > rc");
3529
3530 while((robject = LIST_FIRST(&object->shadow_head)) != NULL) {
3531 vm_pindex_t bo_pindex;
3532 vm_page_t m_in, m_out;
3533
3534 bo_pindex = OFF_TO_IDX(robject->backing_object_offset);
3535
3536 vm_object_reference(robject);
3537
3538 vm_object_pip_wait(robject, "objfrz");
3539
3540 if (robject->ref_count == 1) {
3541 vm_object_deallocate(robject);
3542 continue;
3543 }
3544
3545 vm_object_pip_add(robject, 1);
3546
3547 for (idx = 0; idx < robject->size; idx++) {
3548
3549 m_out = vm_page_grab(robject, idx,
dc1fd4b3 3550 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
984263bc
MD
3551
3552 if (m_out->valid == 0) {
3553 m_in = vm_page_grab(object, bo_pindex + idx,
dc1fd4b3 3554 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
984263bc
MD
3555 if (m_in->valid == 0) {
3556 rv = vm_pager_get_pages(object, &m_in, 1, 0);
3557 if (rv != VM_PAGER_OK) {
3558 printf("vm_freeze_copyopts: cannot read page from file: %lx\n", (long)m_in->pindex);
3559 continue;
3560 }
3561 vm_page_deactivate(m_in);
3562 }
3563
3564 vm_page_protect(m_in, VM_PROT_NONE);
3565 pmap_copy_page(VM_PAGE_TO_PHYS(m_in), VM_PAGE_TO_PHYS(m_out));
3566 m_out->valid = m_in->valid;
3567 vm_page_dirty(m_out);
3568 vm_page_activate(m_out);
3569 vm_page_wakeup(m_in);
3570 }
3571 vm_page_wakeup(m_out);
3572 }
3573
3574 object->shadow_count--;
3575 object->ref_count--;
3576 LIST_REMOVE(robject, shadow_list);
3577 robject->backing_object = NULL;
3578 robject->backing_object_offset = 0;
3579
3580 vm_object_pip_wakeup(robject);
3581 vm_object_deallocate(robject);
3582 }
3583
3584 vm_object_clear_flag(object, OBJ_OPT);
3585}
3586
3587#include "opt_ddb.h"
3588#ifdef DDB
3589#include <sys/kernel.h>
3590
3591#include <ddb/ddb.h>
3592
3593/*
3594 * vm_map_print: [ debug ]
3595 */
3596DB_SHOW_COMMAND(map, vm_map_print)
3597{
3598 static int nlines;
3599 /* XXX convert args. */
3600 vm_map_t map = (vm_map_t)addr;
3601 boolean_t full = have_addr;
3602
3603 vm_map_entry_t entry;
3604
3605 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3606 (void *)map,
3607 (void *)map->pmap, map->nentries, map->timestamp);
3608 nlines++;
3609
3610 if (!full && db_indent)
3611 return;
3612
3613 db_indent += 2;
3614 for (entry = map->header.next; entry != &map->header;
3615 entry = entry->next) {
3616 db_iprintf("map entry %p: start=%p, end=%p\n",
3617 (void *)entry, (void *)entry->start, (void *)entry->end);
3618 nlines++;
3619 {
3620 static char *inheritance_name[4] =
3621 {"share", "copy", "none", "donate_copy"};
3622
3623 db_iprintf(" prot=%x/%x/%s",
3624 entry->protection,
3625 entry->max_protection,
3626 inheritance_name[(int)(unsigned char)entry->inheritance]);
3627 if (entry->wired_count != 0)
3628 db_printf(", wired");
3629 }
3630 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3631 /* XXX no %qd in kernel. Truncate entry->offset. */
3632 db_printf(", share=%p, offset=0x%lx\n",
3633 (void *)entry->object.sub_map,
3634 (long)entry->offset);
3635 nlines++;
3636 if ((entry->prev == &map->header) ||
3637 (entry->prev->object.sub_map !=
3638 entry->object.sub_map)) {
3639 db_indent += 2;
3640 vm_map_print((db_expr_t)(intptr_t)
3641 entry->object.sub_map,
3642 full, 0, (char *)0);
3643 db_indent -= 2;
3644 }
3645 } else {
3646 /* XXX no %qd in kernel. Truncate entry->offset. */
3647 db_printf(", object=%p, offset=0x%lx",
3648 (void *)entry->object.vm_object,
3649 (long)entry->offset);
3650 if (entry->eflags & MAP_ENTRY_COW)
3651 db_printf(", copy (%s)",
3652 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3653 db_printf("\n");
3654 nlines++;
3655
3656 if ((entry->prev == &map->header) ||
3657 (entry->prev->object.vm_object !=
3658 entry->object.vm_object)) {
3659 db_indent += 2;
3660 vm_object_print((db_expr_t)(intptr_t)
3661 entry->object.vm_object,
3662 full, 0, (char *)0);
3663 nlines += 4;
3664 db_indent -= 2;
3665 }
3666 }
3667 }
3668 db_indent -= 2;
3669 if (db_indent == 0)
3670 nlines = 0;
3671}
3672
3673
3674DB_SHOW_COMMAND(procvm, procvm)
3675{
3676 struct proc *p;
3677
3678 if (have_addr) {
3679 p = (struct proc *) addr;
3680 } else {
3681 p = curproc;
3682 }
3683
3684 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3685 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3686 (void *)vmspace_pmap(p->p_vmspace));
3687
3688 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3689}
3690
3691#endif /* DDB */