Mask bits properly for pte_prot() in case it is called with additional
[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 $
06ecca5a 65 * $DragonFly: src/sys/vm/vm_map.c,v 1.27 2004/05/13 17:40:19 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
57e43348 155vm_map_startup(void)
984263bc
MD
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 *
57e43348 171vmspace_alloc(vm_offset_t min, vm_offset_t max)
984263bc
MD
172{
173 struct vmspace *vm;
174
175 vm = zalloc(vmspace_zone);
176 vm_map_init(&vm->vm_map, min, max);
177 pmap_pinit(vmspace_pmap(vm));
178 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
179 vm->vm_refcnt = 1;
180 vm->vm_shm = NULL;
181 vm->vm_exitingcnt = 0;
182 return (vm);
183}
184
185void
a108bf71
MD
186vm_init2(void)
187{
188 zinitna(mapentzone, &mapentobj, NULL, 0, 0, ZONE_USE_RESERVE, 1);
189 zinitna(mapzone, &mapobj, NULL, 0, 0, 0, 1);
984263bc
MD
190 vmspace_zone = zinit("VMSPACE", sizeof (struct vmspace), 0, 0, 3);
191 pmap_init2();
192 vm_object_init2();
193}
194
195static __inline void
196vmspace_dofree(struct vmspace *vm)
197{
a108bf71
MD
198 int count;
199
fef0fdf2
MD
200 /*
201 * Make sure any SysV shm is freed, it might not have in
202 * exit1()
203 */
204 shmexit(vm);
205
a722be49
MD
206 KKASSERT(vm->vm_upcalls == NULL);
207
984263bc
MD
208 /*
209 * Lock the map, to wait out all other references to it.
210 * Delete all of the mappings and pages they hold, then call
211 * the pmap module to reclaim anything left.
212 */
a108bf71 213 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc 214 vm_map_lock(&vm->vm_map);
a108bf71
MD
215 vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
216 vm->vm_map.max_offset, &count);
984263bc 217 vm_map_unlock(&vm->vm_map);
a108bf71 218 vm_map_entry_release(count);
984263bc
MD
219
220 pmap_release(vmspace_pmap(vm));
221 zfree(vmspace_zone, vm);
222}
223
224void
225vmspace_free(struct vmspace *vm)
226{
227 if (vm->vm_refcnt == 0)
228 panic("vmspace_free: attempt to free already freed vmspace");
229
230 if (--vm->vm_refcnt == 0 && vm->vm_exitingcnt == 0)
231 vmspace_dofree(vm);
232}
233
234void
235vmspace_exitfree(struct proc *p)
236{
237 struct vmspace *vm;
238
239 vm = p->p_vmspace;
240 p->p_vmspace = NULL;
241
242 /*
243 * cleanup by parent process wait()ing on exiting child. vm_refcnt
244 * may not be 0 (e.g. fork() and child exits without exec()ing).
245 * exitingcnt may increment above 0 and drop back down to zero
246 * several times while vm_refcnt is held non-zero. vm_refcnt
247 * may also increment above 0 and drop back down to zero several
248 * times while vm_exitingcnt is held non-zero.
249 *
250 * The last wait on the exiting child's vmspace will clean up
251 * the remainder of the vmspace.
252 */
253 if (--vm->vm_exitingcnt == 0 && vm->vm_refcnt == 0)
254 vmspace_dofree(vm);
255}
256
257/*
258 * vmspace_swap_count() - count the approximate swap useage in pages for a
259 * vmspace.
260 *
261 * Swap useage is determined by taking the proportional swap used by
262 * VM objects backing the VM map. To make up for fractional losses,
263 * if the VM object has any swap use at all the associated map entries
264 * count for at least 1 swap page.
265 */
266int
267vmspace_swap_count(struct vmspace *vmspace)
268{
269 vm_map_t map = &vmspace->vm_map;
270 vm_map_entry_t cur;
271 int count = 0;
272
273 for (cur = map->header.next; cur != &map->header; cur = cur->next) {
274 vm_object_t object;
275
276 if ((cur->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
277 (object = cur->object.vm_object) != NULL &&
278 object->type == OBJT_SWAP
279 ) {
280 int n = (cur->end - cur->start) / PAGE_SIZE;
281
282 if (object->un_pager.swp.swp_bcount) {
283 count += object->un_pager.swp.swp_bcount *
284 SWAP_META_PAGES * n / object->size + 1;
285 }
286 }
287 }
288 return(count);
289}
290
291
292/*
293 * vm_map_create:
294 *
295 * Creates and returns a new empty VM map with
296 * the given physical map structure, and having
297 * the given lower and upper address bounds.
298 */
299vm_map_t
a108bf71 300vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
984263bc
MD
301{
302 vm_map_t result;
303
304 result = zalloc(mapzone);
305 vm_map_init(result, min, max);
306 result->pmap = pmap;
307 return (result);
308}
309
310/*
311 * Initialize an existing vm_map structure
312 * such as that in the vmspace structure.
313 * The pmap is set elsewhere.
314 */
315void
a108bf71 316vm_map_init(struct vm_map *map, vm_offset_t min, vm_offset_t max)
984263bc
MD
317{
318 map->header.next = map->header.prev = &map->header;
319 map->nentries = 0;
320 map->size = 0;
321 map->system_map = 0;
322 map->infork = 0;
323 map->min_offset = min;
324 map->max_offset = max;
325 map->first_free = &map->header;
326 map->hint = &map->header;
327 map->timestamp = 0;
377d4740 328 lockinit(&map->lock, 0, "thrd_sleep", 0, LK_NOPAUSE);
984263bc
MD
329}
330
41a01a4d
MD
331/*
332 * vm_map_entry_cpu_init:
333 *
334 * Set an initial negative count so the first attempt to reserve
335 * space preloads a bunch of vm_map_entry's for this cpu. This
336 * routine is called in early boot so we cannot just call
337 * vm_map_entry_reserve().
338 *
339 * May be called for a gd other then mycpu.
340 */
341void
342vm_map_entry_reserve_cpu_init(globaldata_t gd)
343{
344 gd->gd_vme_avail -= MAP_RESERVE_COUNT * 2;
345}
346
a108bf71
MD
347/*
348 * vm_map_entry_reserve:
349 *
e1359933
MD
350 * Reserves vm_map_entry structures so code later on can manipulate
351 * map_entry structures within a locked map without blocking trying
352 * to allocate a new vm_map_entry.
a108bf71
MD
353 */
354int
355vm_map_entry_reserve(int count)
356{
357 struct globaldata *gd = mycpu;
358 vm_map_entry_t entry;
359
360 crit_enter();
361 gd->gd_vme_avail -= count;
362
363 /*
364 * Make sure we have enough structures in gd_vme_base to handle
365 * the reservation request.
366 */
367 while (gd->gd_vme_avail < 0) {
368 entry = zalloc(mapentzone);
369 entry->next = gd->gd_vme_base;
370 gd->gd_vme_base = entry;
371 ++gd->gd_vme_avail;
372 }
373 crit_exit();
374 return(count);
375}
376
377/*
378 * vm_map_entry_release:
379 *
380 * Releases previously reserved vm_map_entry structures that were not
381 * used. If we have too much junk in our per-cpu cache clean some of
382 * it out.
383 */
384void
385vm_map_entry_release(int count)
386{
387 struct globaldata *gd = mycpu;
388 vm_map_entry_t entry;
389
390 crit_enter();
391 gd->gd_vme_avail += count;
392 while (gd->gd_vme_avail > MAP_RESERVE_SLOP) {
393 entry = gd->gd_vme_base;
394 KKASSERT(entry != NULL);
395 gd->gd_vme_base = entry->next;
396 --gd->gd_vme_avail;
397 crit_exit();
398 zfree(mapentzone, entry);
399 crit_enter();
400 }
401 crit_exit();
402}
403
404/*
405 * vm_map_entry_kreserve:
406 *
407 * Reserve map entry structures for use in kernel_map or (if it exists)
408 * kmem_map. These entries have *ALREADY* been reserved on a per-cpu
e1359933
MD
409 * basis when the map was inited. This function is used by zalloc()
410 * to avoid a recursion when zalloc() itself needs to allocate additional
411 * kernel memory.
a108bf71 412 *
e1359933
MD
413 * This function should only be used when the caller intends to later
414 * call vm_map_entry_reserve() to 'normalize' the reserve cache.
a108bf71
MD
415 */
416int
417vm_map_entry_kreserve(int count)
418{
419 struct globaldata *gd = mycpu;
420
421 crit_enter();
422 gd->gd_vme_kdeficit += count;
423 crit_exit();
424 KKASSERT(gd->gd_vme_base != NULL);
425 return(count);
426}
427
428/*
429 * vm_map_entry_krelease:
430 *
431 * Release previously reserved map entries for kernel_map or kmem_map
432 * use. This routine determines how many entries were actually used and
433 * replentishes the kernel reserve supply from vme_avail.
434 *
435 * If there is insufficient supply vme_avail will go negative, which is
436 * ok. We cannot safely call zalloc in this function without getting
437 * into a recursion deadlock. zalloc() will call vm_map_entry_reserve()
438 * to regenerate the lost entries.
439 */
440void
441vm_map_entry_krelease(int count)
442{
443 struct globaldata *gd = mycpu;
444
445 crit_enter();
446 gd->gd_vme_kdeficit -= count;
447 gd->gd_vme_avail -= gd->gd_vme_kdeficit; /* can go negative */
448 gd->gd_vme_kdeficit = 0;
449 crit_exit();
450}
451
984263bc 452/*
8a8d5d85 453 * vm_map_entry_create: [ internal use only ]
984263bc 454 *
8a8d5d85 455 * Allocates a VM map entry for insertion. No entry fields are filled
a108bf71
MD
456 * in.
457 *
458 * This routine may be called from an interrupt thread but not a FAST
459 * interrupt. This routine may recurse the map lock.
984263bc 460 */
8a8d5d85 461static vm_map_entry_t
a108bf71 462vm_map_entry_create(vm_map_t map, int *countp)
984263bc 463{
a108bf71
MD
464 struct globaldata *gd = mycpu;
465 vm_map_entry_t entry;
8a8d5d85 466
a108bf71
MD
467 KKASSERT(*countp > 0);
468 --*countp;
469 crit_enter();
470 entry = gd->gd_vme_base;
471 KASSERT(entry != NULL, ("gd_vme_base NULL! count %d", *countp));
472 gd->gd_vme_base = entry->next;
473 crit_exit();
474 return(entry);
984263bc
MD
475}
476
477/*
8a8d5d85 478 * vm_map_entry_dispose: [ internal use only ]
984263bc 479 *
8a8d5d85
MD
480 * Dispose of a vm_map_entry that is no longer being referenced. This
481 * function may be called from an interrupt.
984263bc 482 */
8a8d5d85 483static void
a108bf71 484vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc 485{
a108bf71
MD
486 struct globaldata *gd = mycpu;
487
488 ++*countp;
489 crit_enter();
490 entry->next = gd->gd_vme_base;
491 gd->gd_vme_base = entry;
492 crit_exit();
984263bc
MD
493}
494
8a8d5d85 495
984263bc
MD
496/*
497 * vm_map_entry_{un,}link:
498 *
499 * Insert/remove entries from maps.
500 */
501static __inline void
502vm_map_entry_link(vm_map_t map,
503 vm_map_entry_t after_where,
504 vm_map_entry_t entry)
505{
506 map->nentries++;
507 entry->prev = after_where;
508 entry->next = after_where->next;
509 entry->next->prev = entry;
510 after_where->next = entry;
511}
512
513static __inline void
514vm_map_entry_unlink(vm_map_t map,
515 vm_map_entry_t entry)
516{
517 vm_map_entry_t prev;
518 vm_map_entry_t next;
519
520 if (entry->eflags & MAP_ENTRY_IN_TRANSITION)
521 panic("vm_map_entry_unlink: attempt to mess with locked entry! %p", entry);
522 prev = entry->prev;
523 next = entry->next;
524 next->prev = prev;
525 prev->next = next;
526 map->nentries--;
527}
528
529/*
530 * SAVE_HINT:
531 *
532 * Saves the specified entry as the hint for
533 * future lookups.
534 */
535#define SAVE_HINT(map,value) \
536 (map)->hint = (value);
537
538/*
539 * vm_map_lookup_entry: [ internal use only ]
540 *
541 * Finds the map entry containing (or
542 * immediately preceding) the specified address
543 * in the given map; the entry is returned
544 * in the "entry" parameter. The boolean
545 * result indicates whether the address is
546 * actually contained in the map.
547 */
548boolean_t
57e43348
MD
549vm_map_lookup_entry(vm_map_t map, vm_offset_t address,
550 vm_map_entry_t *entry /* OUT */)
984263bc
MD
551{
552 vm_map_entry_t cur;
553 vm_map_entry_t last;
554
555 /*
556 * Start looking either from the head of the list, or from the hint.
557 */
558
559 cur = map->hint;
560
561 if (cur == &map->header)
562 cur = cur->next;
563
564 if (address >= cur->start) {
565 /*
566 * Go from hint to end of list.
567 *
568 * But first, make a quick check to see if we are already looking
569 * at the entry we want (which is usually the case). Note also
570 * that we don't need to save the hint here... it is the same
571 * hint (unless we are at the header, in which case the hint
572 * didn't buy us anything anyway).
573 */
574 last = &map->header;
575 if ((cur != last) && (cur->end > address)) {
576 *entry = cur;
577 return (TRUE);
578 }
579 } else {
580 /*
581 * Go from start to hint, *inclusively*
582 */
583 last = cur->next;
584 cur = map->header.next;
585 }
586
587 /*
588 * Search linearly
589 */
590
591 while (cur != last) {
592 if (cur->end > address) {
593 if (address >= cur->start) {
594 /*
595 * Save this lookup for future hints, and
596 * return
597 */
598
599 *entry = cur;
600 SAVE_HINT(map, cur);
601 return (TRUE);
602 }
603 break;
604 }
605 cur = cur->next;
606 }
607 *entry = cur->prev;
608 SAVE_HINT(map, *entry);
609 return (FALSE);
610}
611
612/*
613 * vm_map_insert:
614 *
615 * Inserts the given whole VM object into the target
616 * map at the specified address range. The object's
617 * size should match that of the address range.
618 *
a108bf71
MD
619 * Requires that the map be locked, and leaves it so. Requires that
620 * sufficient vm_map_entry structures have been reserved and tracks
621 * the use via countp.
984263bc
MD
622 *
623 * If object is non-NULL, ref count must be bumped by caller
624 * prior to making call to account for the new entry.
625 */
626int
a108bf71
MD
627vm_map_insert(vm_map_t map, int *countp,
628 vm_object_t object, vm_ooffset_t offset,
984263bc
MD
629 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
630 int cow)
631{
632 vm_map_entry_t new_entry;
633 vm_map_entry_t prev_entry;
634 vm_map_entry_t temp_entry;
635 vm_eflags_t protoeflags;
636
637 /*
638 * Check that the start and end points are not bogus.
639 */
640
641 if ((start < map->min_offset) || (end > map->max_offset) ||
642 (start >= end))
643 return (KERN_INVALID_ADDRESS);
644
645 /*
646 * Find the entry prior to the proposed starting address; if it's part
647 * of an existing entry, this range is bogus.
648 */
649
650 if (vm_map_lookup_entry(map, start, &temp_entry))
651 return (KERN_NO_SPACE);
652
653 prev_entry = temp_entry;
654
655 /*
656 * Assert that the next entry doesn't overlap the end point.
657 */
658
659 if ((prev_entry->next != &map->header) &&
660 (prev_entry->next->start < end))
661 return (KERN_NO_SPACE);
662
663 protoeflags = 0;
664
665 if (cow & MAP_COPY_ON_WRITE)
666 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
667
668 if (cow & MAP_NOFAULT) {
669 protoeflags |= MAP_ENTRY_NOFAULT;
670
671 KASSERT(object == NULL,
672 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
673 }
674 if (cow & MAP_DISABLE_SYNCER)
675 protoeflags |= MAP_ENTRY_NOSYNC;
676 if (cow & MAP_DISABLE_COREDUMP)
677 protoeflags |= MAP_ENTRY_NOCOREDUMP;
678
679 if (object) {
680 /*
681 * When object is non-NULL, it could be shared with another
682 * process. We have to set or clear OBJ_ONEMAPPING
683 * appropriately.
684 */
685 if ((object->ref_count > 1) || (object->shadow_count != 0)) {
686 vm_object_clear_flag(object, OBJ_ONEMAPPING);
687 }
688 }
689 else if ((prev_entry != &map->header) &&
690 (prev_entry->eflags == protoeflags) &&
691 (prev_entry->end == start) &&
692 (prev_entry->wired_count == 0) &&
693 ((prev_entry->object.vm_object == NULL) ||
694 vm_object_coalesce(prev_entry->object.vm_object,
695 OFF_TO_IDX(prev_entry->offset),
696 (vm_size_t)(prev_entry->end - prev_entry->start),
697 (vm_size_t)(end - prev_entry->end)))) {
698 /*
699 * We were able to extend the object. Determine if we
700 * can extend the previous map entry to include the
701 * new range as well.
702 */
703 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
704 (prev_entry->protection == prot) &&
705 (prev_entry->max_protection == max)) {
706 map->size += (end - prev_entry->end);
707 prev_entry->end = end;
a108bf71 708 vm_map_simplify_entry(map, prev_entry, countp);
984263bc
MD
709 return (KERN_SUCCESS);
710 }
711
712 /*
713 * If we can extend the object but cannot extend the
714 * map entry, we have to create a new map entry. We
715 * must bump the ref count on the extended object to
716 * account for it. object may be NULL.
717 */
718 object = prev_entry->object.vm_object;
719 offset = prev_entry->offset +
720 (prev_entry->end - prev_entry->start);
721 vm_object_reference(object);
722 }
723
724 /*
725 * NOTE: if conditionals fail, object can be NULL here. This occurs
726 * in things like the buffer map where we manage kva but do not manage
727 * backing objects.
728 */
729
730 /*
731 * Create a new entry
732 */
733
a108bf71 734 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
735 new_entry->start = start;
736 new_entry->end = end;
737
738 new_entry->eflags = protoeflags;
739 new_entry->object.vm_object = object;
740 new_entry->offset = offset;
741 new_entry->avail_ssize = 0;
742
743 new_entry->inheritance = VM_INHERIT_DEFAULT;
744 new_entry->protection = prot;
745 new_entry->max_protection = max;
746 new_entry->wired_count = 0;
747
748 /*
749 * Insert the new entry into the list
750 */
751
752 vm_map_entry_link(map, prev_entry, new_entry);
753 map->size += new_entry->end - new_entry->start;
754
755 /*
756 * Update the free space hint
757 */
758 if ((map->first_free == prev_entry) &&
759 (prev_entry->end >= new_entry->start)) {
760 map->first_free = new_entry;
761 }
762
763#if 0
764 /*
765 * Temporarily removed to avoid MAP_STACK panic, due to
766 * MAP_STACK being a huge hack. Will be added back in
767 * when MAP_STACK (and the user stack mapping) is fixed.
768 */
769 /*
770 * It may be possible to simplify the entry
771 */
a108bf71 772 vm_map_simplify_entry(map, new_entry, countp);
984263bc
MD
773#endif
774
775 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
083a7402 776 pmap_object_init_pt(map->pmap, start, prot,
984263bc
MD
777 object, OFF_TO_IDX(offset), end - start,
778 cow & MAP_PREFAULT_PARTIAL);
779 }
780
781 return (KERN_SUCCESS);
782}
783
784/*
785 * Find sufficient space for `length' bytes in the given map, starting at
786 * `start'. The map must be locked. Returns 0 on success, 1 on no space.
e9bb90e8
MD
787 *
788 * This function will returned an arbitrarily aligned pointer. If no
789 * particular alignment is required you should pass align as 1. Note that
790 * the map may return PAGE_SIZE aligned pointers if all the lengths used in
791 * the map are a multiple of PAGE_SIZE, even if you pass a smaller align
792 * argument.
793 *
794 * 'align' should be a power of 2 but is not required to be.
984263bc
MD
795 */
796int
e9bb90e8
MD
797vm_map_findspace(
798 vm_map_t map,
799 vm_offset_t start,
800 vm_size_t length,
801 vm_offset_t align,
802 vm_offset_t *addr)
984263bc
MD
803{
804 vm_map_entry_t entry, next;
805 vm_offset_t end;
e9bb90e8 806 vm_offset_t align_mask;
984263bc
MD
807
808 if (start < map->min_offset)
809 start = map->min_offset;
810 if (start > map->max_offset)
811 return (1);
812
e9bb90e8
MD
813 /*
814 * If the alignment is not a power of 2 we will have to use
815 * a mod/division, set align_mask to a special value.
816 */
817 if ((align | (align - 1)) + 1 != (align << 1))
818 align_mask = (vm_offset_t)-1;
819 else
820 align_mask = align - 1;
821
a108bf71 822retry:
984263bc
MD
823 /*
824 * Look for the first possible address; if there's already something
825 * at this address, we have to start after it.
826 */
827 if (start == map->min_offset) {
828 if ((entry = map->first_free) != &map->header)
829 start = entry->end;
830 } else {
831 vm_map_entry_t tmp;
832
833 if (vm_map_lookup_entry(map, start, &tmp))
834 start = tmp->end;
835 entry = tmp;
836 }
837
838 /*
839 * Look through the rest of the map, trying to fit a new region in the
840 * gap between existing regions, or after the very last region.
841 */
842 for (;; start = (entry = next)->end) {
e9bb90e8
MD
843 /*
844 * Adjust the proposed start by the requested alignment,
845 * be sure that we didn't wrap the address.
846 */
847 if (align_mask == (vm_offset_t)-1)
848 end = ((start + align - 1) / align) * align;
849 else
850 end = (start + align_mask) & ~align_mask;
851 if (end < start)
852 return (1);
853 start = end;
984263bc
MD
854 /*
855 * Find the end of the proposed new region. Be sure we didn't
e9bb90e8
MD
856 * go beyond the end of the map, or wrap around the address.
857 * Then check to see if this is the last entry or if the
858 * proposed end fits in the gap between this and the next
859 * entry.
984263bc
MD
860 */
861 end = start + length;
862 if (end > map->max_offset || end < start)
863 return (1);
864 next = entry->next;
865 if (next == &map->header || next->start >= end)
866 break;
867 }
868 SAVE_HINT(map, entry);
984263bc
MD
869 if (map == kernel_map) {
870 vm_offset_t ksize;
871 if ((ksize = round_page(start + length)) > kernel_vm_end) {
872 pmap_growkernel(ksize);
a108bf71 873 goto retry;
984263bc
MD
874 }
875 }
a108bf71 876 *addr = start;
984263bc
MD
877 return (0);
878}
879
880/*
881 * vm_map_find finds an unallocated region in the target address
882 * map with the given length. The search is defined to be
883 * first-fit from the specified address; the region found is
884 * returned in the same parameter.
885 *
886 * If object is non-NULL, ref count must be bumped by caller
887 * prior to making call to account for the new entry.
888 */
889int
890vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
891 vm_offset_t *addr, /* IN/OUT */
892 vm_size_t length, boolean_t find_space, vm_prot_t prot,
893 vm_prot_t max, int cow)
894{
895 vm_offset_t start;
03aa8d99 896 int result;
a108bf71 897 int count;
984263bc
MD
898
899 start = *addr;
900
a108bf71 901 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
902 vm_map_lock(map);
903 if (find_space) {
e9bb90e8 904 if (vm_map_findspace(map, start, length, 1, addr)) {
984263bc 905 vm_map_unlock(map);
a108bf71 906 vm_map_entry_release(count);
984263bc
MD
907 return (KERN_NO_SPACE);
908 }
909 start = *addr;
910 }
a108bf71 911 result = vm_map_insert(map, &count, object, offset,
984263bc
MD
912 start, start + length, prot, max, cow);
913 vm_map_unlock(map);
a108bf71 914 vm_map_entry_release(count);
984263bc 915
984263bc
MD
916 return (result);
917}
918
919/*
920 * vm_map_simplify_entry:
921 *
922 * Simplify the given map entry by merging with either neighbor. This
923 * routine also has the ability to merge with both neighbors.
924 *
925 * The map must be locked.
926 *
927 * This routine guarentees that the passed entry remains valid (though
928 * possibly extended). When merging, this routine may delete one or
929 * both neighbors. No action is taken on entries which have their
930 * in-transition flag set.
931 */
932void
a108bf71 933vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc
MD
934{
935 vm_map_entry_t next, prev;
936 vm_size_t prevsize, esize;
937
938 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) {
12e4aaff 939 ++mycpu->gd_cnt.v_intrans_coll;
984263bc
MD
940 return;
941 }
942
943 prev = entry->prev;
944 if (prev != &map->header) {
945 prevsize = prev->end - prev->start;
946 if ( (prev->end == entry->start) &&
947 (prev->object.vm_object == entry->object.vm_object) &&
948 (!prev->object.vm_object ||
949 (prev->offset + prevsize == entry->offset)) &&
950 (prev->eflags == entry->eflags) &&
951 (prev->protection == entry->protection) &&
952 (prev->max_protection == entry->max_protection) &&
953 (prev->inheritance == entry->inheritance) &&
954 (prev->wired_count == entry->wired_count)) {
955 if (map->first_free == prev)
956 map->first_free = entry;
957 if (map->hint == prev)
958 map->hint = entry;
959 vm_map_entry_unlink(map, prev);
960 entry->start = prev->start;
961 entry->offset = prev->offset;
962 if (prev->object.vm_object)
963 vm_object_deallocate(prev->object.vm_object);
a108bf71 964 vm_map_entry_dispose(map, prev, countp);
984263bc
MD
965 }
966 }
967
968 next = entry->next;
969 if (next != &map->header) {
970 esize = entry->end - entry->start;
971 if ((entry->end == next->start) &&
972 (next->object.vm_object == entry->object.vm_object) &&
973 (!entry->object.vm_object ||
974 (entry->offset + esize == next->offset)) &&
975 (next->eflags == entry->eflags) &&
976 (next->protection == entry->protection) &&
977 (next->max_protection == entry->max_protection) &&
978 (next->inheritance == entry->inheritance) &&
979 (next->wired_count == entry->wired_count)) {
980 if (map->first_free == next)
981 map->first_free = entry;
982 if (map->hint == next)
983 map->hint = entry;
984 vm_map_entry_unlink(map, next);
985 entry->end = next->end;
986 if (next->object.vm_object)
987 vm_object_deallocate(next->object.vm_object);
a108bf71 988 vm_map_entry_dispose(map, next, countp);
984263bc
MD
989 }
990 }
991}
992/*
993 * vm_map_clip_start: [ internal use only ]
994 *
995 * Asserts that the given entry begins at or after
996 * the specified address; if necessary,
997 * it splits the entry into two.
998 */
a108bf71 999#define vm_map_clip_start(map, entry, startaddr, countp) \
984263bc
MD
1000{ \
1001 if (startaddr > entry->start) \
a108bf71 1002 _vm_map_clip_start(map, entry, startaddr, countp); \
984263bc
MD
1003}
1004
1005/*
1006 * This routine is called only when it is known that
1007 * the entry must be split.
1008 */
1009static void
a108bf71 1010_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start, int *countp)
984263bc
MD
1011{
1012 vm_map_entry_t new_entry;
1013
1014 /*
1015 * Split off the front portion -- note that we must insert the new
1016 * entry BEFORE this one, so that this entry has the specified
1017 * starting address.
1018 */
1019
a108bf71 1020 vm_map_simplify_entry(map, entry, countp);
984263bc
MD
1021
1022 /*
1023 * If there is no object backing this entry, we might as well create
1024 * one now. If we defer it, an object can get created after the map
1025 * is clipped, and individual objects will be created for the split-up
1026 * map. This is a bit of a hack, but is also about the best place to
1027 * put this improvement.
1028 */
1029
1030 if (entry->object.vm_object == NULL && !map->system_map) {
1031 vm_object_t object;
1032 object = vm_object_allocate(OBJT_DEFAULT,
1033 atop(entry->end - entry->start));
1034 entry->object.vm_object = object;
1035 entry->offset = 0;
1036 }
1037
a108bf71 1038 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
1039 *new_entry = *entry;
1040
1041 new_entry->end = start;
1042 entry->offset += (start - entry->start);
1043 entry->start = start;
1044
1045 vm_map_entry_link(map, entry->prev, new_entry);
1046
1047 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1048 vm_object_reference(new_entry->object.vm_object);
1049 }
1050}
1051
1052/*
1053 * vm_map_clip_end: [ internal use only ]
1054 *
1055 * Asserts that the given entry ends at or before
1056 * the specified address; if necessary,
1057 * it splits the entry into two.
1058 */
1059
a108bf71 1060#define vm_map_clip_end(map, entry, endaddr, countp) \
984263bc
MD
1061{ \
1062 if (endaddr < entry->end) \
a108bf71 1063 _vm_map_clip_end(map, entry, endaddr, countp); \
984263bc
MD
1064}
1065
1066/*
1067 * This routine is called only when it is known that
1068 * the entry must be split.
1069 */
1070static void
a108bf71 1071_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end, int *countp)
984263bc
MD
1072{
1073 vm_map_entry_t new_entry;
1074
1075 /*
1076 * If there is no object backing this entry, we might as well create
1077 * one now. If we defer it, an object can get created after the map
1078 * is clipped, and individual objects will be created for the split-up
1079 * map. This is a bit of a hack, but is also about the best place to
1080 * put this improvement.
1081 */
1082
1083 if (entry->object.vm_object == NULL && !map->system_map) {
1084 vm_object_t object;
1085 object = vm_object_allocate(OBJT_DEFAULT,
1086 atop(entry->end - entry->start));
1087 entry->object.vm_object = object;
1088 entry->offset = 0;
1089 }
1090
1091 /*
1092 * Create a new entry and insert it AFTER the specified entry
1093 */
1094
a108bf71 1095 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
1096 *new_entry = *entry;
1097
1098 new_entry->start = entry->end = end;
1099 new_entry->offset += (end - entry->start);
1100
1101 vm_map_entry_link(map, entry, new_entry);
1102
1103 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1104 vm_object_reference(new_entry->object.vm_object);
1105 }
1106}
1107
1108/*
1109 * VM_MAP_RANGE_CHECK: [ internal use only ]
1110 *
1111 * Asserts that the starting and ending region
1112 * addresses fall within the valid range of the map.
1113 */
1114#define VM_MAP_RANGE_CHECK(map, start, end) \
1115 { \
1116 if (start < vm_map_min(map)) \
1117 start = vm_map_min(map); \
1118 if (end > vm_map_max(map)) \
1119 end = vm_map_max(map); \
1120 if (start > end) \
1121 start = end; \
1122 }
1123
1124/*
1125 * vm_map_transition_wait: [ kernel use only ]
1126 *
1127 * Used to block when an in-transition collison occurs. The map
1128 * is unlocked for the sleep and relocked before the return.
1129 */
1130static
1131void
1132vm_map_transition_wait(vm_map_t map)
1133{
1134 vm_map_unlock(map);
377d4740 1135 tsleep(map, 0, "vment", 0);
984263bc
MD
1136 vm_map_lock(map);
1137}
1138
1139/*
1140 * CLIP_CHECK_BACK
1141 * CLIP_CHECK_FWD
1142 *
1143 * When we do blocking operations with the map lock held it is
1144 * possible that a clip might have occured on our in-transit entry,
1145 * requiring an adjustment to the entry in our loop. These macros
1146 * help the pageable and clip_range code deal with the case. The
1147 * conditional costs virtually nothing if no clipping has occured.
1148 */
1149
1150#define CLIP_CHECK_BACK(entry, save_start) \
1151 do { \
1152 while (entry->start != save_start) { \
1153 entry = entry->prev; \
1154 KASSERT(entry != &map->header, ("bad entry clip")); \
1155 } \
1156 } while(0)
1157
1158#define CLIP_CHECK_FWD(entry, save_end) \
1159 do { \
1160 while (entry->end != save_end) { \
1161 entry = entry->next; \
1162 KASSERT(entry != &map->header, ("bad entry clip")); \
1163 } \
1164 } while(0)
1165
1166
1167/*
1168 * vm_map_clip_range: [ kernel use only ]
1169 *
1170 * Clip the specified range and return the base entry. The
1171 * range may cover several entries starting at the returned base
1172 * and the first and last entry in the covering sequence will be
1173 * properly clipped to the requested start and end address.
1174 *
1175 * If no holes are allowed you should pass the MAP_CLIP_NO_HOLES
1176 * flag.
1177 *
1178 * The MAP_ENTRY_IN_TRANSITION flag will be set for the entries
1179 * covered by the requested range.
1180 *
1181 * The map must be exclusively locked on entry and will remain locked
1182 * on return. If no range exists or the range contains holes and you
1183 * specified that no holes were allowed, NULL will be returned. This
1184 * routine may temporarily unlock the map in order avoid a deadlock when
1185 * sleeping.
1186 */
1187static
1188vm_map_entry_t
a108bf71
MD
1189vm_map_clip_range(vm_map_t map, vm_offset_t start, vm_offset_t end,
1190 int *countp, int flags)
984263bc
MD
1191{
1192 vm_map_entry_t start_entry;
1193 vm_map_entry_t entry;
1194
1195 /*
1196 * Locate the entry and effect initial clipping. The in-transition
1197 * case does not occur very often so do not try to optimize it.
1198 */
1199again:
1200 if (vm_map_lookup_entry(map, start, &start_entry) == FALSE)
1201 return (NULL);
1202 entry = start_entry;
1203 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1204 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
12e4aaff
MD
1205 ++mycpu->gd_cnt.v_intrans_coll;
1206 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
1207 vm_map_transition_wait(map);
1208 /*
1209 * entry and/or start_entry may have been clipped while
1210 * we slept, or may have gone away entirely. We have
1211 * to restart from the lookup.
1212 */
1213 goto again;
1214 }
1215 /*
1216 * Since we hold an exclusive map lock we do not have to restart
1217 * after clipping, even though clipping may block in zalloc.
1218 */
a108bf71
MD
1219 vm_map_clip_start(map, entry, start, countp);
1220 vm_map_clip_end(map, entry, end, countp);
984263bc
MD
1221 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1222
1223 /*
1224 * Scan entries covered by the range. When working on the next
1225 * entry a restart need only re-loop on the current entry which
1226 * we have already locked, since 'next' may have changed. Also,
1227 * even though entry is safe, it may have been clipped so we
1228 * have to iterate forwards through the clip after sleeping.
1229 */
1230 while (entry->next != &map->header && entry->next->start < end) {
1231 vm_map_entry_t next = entry->next;
1232
1233 if (flags & MAP_CLIP_NO_HOLES) {
1234 if (next->start > entry->end) {
1235 vm_map_unclip_range(map, start_entry,
a108bf71 1236 start, entry->end, countp, flags);
984263bc
MD
1237 return(NULL);
1238 }
1239 }
1240
1241 if (next->eflags & MAP_ENTRY_IN_TRANSITION) {
1242 vm_offset_t save_end = entry->end;
1243 next->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
12e4aaff
MD
1244 ++mycpu->gd_cnt.v_intrans_coll;
1245 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
1246 vm_map_transition_wait(map);
1247
1248 /*
1249 * clips might have occured while we blocked.
1250 */
1251 CLIP_CHECK_FWD(entry, save_end);
1252 CLIP_CHECK_BACK(start_entry, start);
1253 continue;
1254 }
1255 /*
1256 * No restart necessary even though clip_end may block, we
1257 * are holding the map lock.
1258 */
a108bf71 1259 vm_map_clip_end(map, next, end, countp);
984263bc
MD
1260 next->eflags |= MAP_ENTRY_IN_TRANSITION;
1261 entry = next;
1262 }
1263 if (flags & MAP_CLIP_NO_HOLES) {
1264 if (entry->end != end) {
1265 vm_map_unclip_range(map, start_entry,
a108bf71 1266 start, entry->end, countp, flags);
984263bc
MD
1267 return(NULL);
1268 }
1269 }
1270 return(start_entry);
1271}
1272
1273/*
1274 * vm_map_unclip_range: [ kernel use only ]
1275 *
1276 * Undo the effect of vm_map_clip_range(). You should pass the same
1277 * flags and the same range that you passed to vm_map_clip_range().
1278 * This code will clear the in-transition flag on the entries and
1279 * wake up anyone waiting. This code will also simplify the sequence
1280 * and attempt to merge it with entries before and after the sequence.
1281 *
1282 * The map must be locked on entry and will remain locked on return.
1283 *
1284 * Note that you should also pass the start_entry returned by
1285 * vm_map_clip_range(). However, if you block between the two calls
1286 * with the map unlocked please be aware that the start_entry may
1287 * have been clipped and you may need to scan it backwards to find
1288 * the entry corresponding with the original start address. You are
1289 * responsible for this, vm_map_unclip_range() expects the correct
1290 * start_entry to be passed to it and will KASSERT otherwise.
1291 */
1292static
1293void
1294vm_map_unclip_range(
1295 vm_map_t map,
1296 vm_map_entry_t start_entry,
1297 vm_offset_t start,
1298 vm_offset_t end,
a108bf71 1299 int *countp,
984263bc
MD
1300 int flags)
1301{
1302 vm_map_entry_t entry;
1303
1304 entry = start_entry;
1305
1306 KASSERT(entry->start == start, ("unclip_range: illegal base entry"));
1307 while (entry != &map->header && entry->start < end) {
1308 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, ("in-transition flag not set during unclip on: %p", entry));
1309 KASSERT(entry->end <= end, ("unclip_range: tail wasn't clipped"));
1310 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1311 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1312 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1313 wakeup(map);
1314 }
1315 entry = entry->next;
1316 }
1317
1318 /*
1319 * Simplification does not block so there is no restart case.
1320 */
1321 entry = start_entry;
1322 while (entry != &map->header && entry->start < end) {
a108bf71 1323 vm_map_simplify_entry(map, entry, countp);
984263bc
MD
1324 entry = entry->next;
1325 }
1326}
1327
1328/*
1329 * vm_map_submap: [ kernel use only ]
1330 *
1331 * Mark the given range as handled by a subordinate map.
1332 *
1333 * This range must have been created with vm_map_find,
1334 * and no other operations may have been performed on this
1335 * range prior to calling vm_map_submap.
1336 *
1337 * Only a limited number of operations can be performed
1338 * within this rage after calling vm_map_submap:
1339 * vm_fault
1340 * [Don't try vm_map_copy!]
1341 *
1342 * To remove a submapping, one must first remove the
1343 * range from the superior map, and then destroy the
1344 * submap (if desired). [Better yet, don't try it.]
1345 */
1346int
a108bf71 1347vm_map_submap(vm_map_t map, vm_offset_t start, vm_offset_t end, vm_map_t submap)
984263bc
MD
1348{
1349 vm_map_entry_t entry;
1350 int result = KERN_INVALID_ARGUMENT;
a108bf71 1351 int count;
984263bc 1352
a108bf71 1353 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1354 vm_map_lock(map);
1355
1356 VM_MAP_RANGE_CHECK(map, start, end);
1357
1358 if (vm_map_lookup_entry(map, start, &entry)) {
a108bf71 1359 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1360 } else {
1361 entry = entry->next;
1362 }
1363
a108bf71 1364 vm_map_clip_end(map, entry, end, &count);
984263bc
MD
1365
1366 if ((entry->start == start) && (entry->end == end) &&
1367 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1368 (entry->object.vm_object == NULL)) {
1369 entry->object.sub_map = submap;
1370 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1371 result = KERN_SUCCESS;
1372 }
1373 vm_map_unlock(map);
a108bf71 1374 vm_map_entry_release(count);
984263bc
MD
1375
1376 return (result);
1377}
1378
1379/*
1380 * vm_map_protect:
1381 *
1382 * Sets the protection of the specified address
1383 * region in the target map. If "set_max" is
1384 * specified, the maximum protection is to be set;
1385 * otherwise, only the current protection is affected.
1386 */
1387int
1388vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1389 vm_prot_t new_prot, boolean_t set_max)
1390{
1391 vm_map_entry_t current;
1392 vm_map_entry_t entry;
a108bf71 1393 int count;
984263bc 1394
a108bf71 1395 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1396 vm_map_lock(map);
1397
1398 VM_MAP_RANGE_CHECK(map, start, end);
1399
1400 if (vm_map_lookup_entry(map, start, &entry)) {
a108bf71 1401 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1402 } else {
1403 entry = entry->next;
1404 }
1405
1406 /*
1407 * Make a first pass to check for protection violations.
1408 */
1409
1410 current = entry;
1411 while ((current != &map->header) && (current->start < end)) {
1412 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1413 vm_map_unlock(map);
a108bf71 1414 vm_map_entry_release(count);
984263bc
MD
1415 return (KERN_INVALID_ARGUMENT);
1416 }
1417 if ((new_prot & current->max_protection) != new_prot) {
1418 vm_map_unlock(map);
a108bf71 1419 vm_map_entry_release(count);
984263bc
MD
1420 return (KERN_PROTECTION_FAILURE);
1421 }
1422 current = current->next;
1423 }
1424
1425 /*
1426 * Go back and fix up protections. [Note that clipping is not
1427 * necessary the second time.]
1428 */
984263bc
MD
1429 current = entry;
1430
1431 while ((current != &map->header) && (current->start < end)) {
1432 vm_prot_t old_prot;
1433
a108bf71 1434 vm_map_clip_end(map, current, end, &count);
984263bc
MD
1435
1436 old_prot = current->protection;
1437 if (set_max)
1438 current->protection =
1439 (current->max_protection = new_prot) &
1440 old_prot;
1441 else
1442 current->protection = new_prot;
1443
1444 /*
1445 * Update physical map if necessary. Worry about copy-on-write
1446 * here -- CHECK THIS XXX
1447 */
1448
1449 if (current->protection != old_prot) {
1450#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1451 VM_PROT_ALL)
1452
1453 pmap_protect(map->pmap, current->start,
1454 current->end,
1455 current->protection & MASK(current));
1456#undef MASK
1457 }
1458
a108bf71 1459 vm_map_simplify_entry(map, current, &count);
984263bc
MD
1460
1461 current = current->next;
1462 }
1463
1464 vm_map_unlock(map);
a108bf71 1465 vm_map_entry_release(count);
984263bc
MD
1466 return (KERN_SUCCESS);
1467}
1468
1469/*
1470 * vm_map_madvise:
1471 *
1472 * This routine traverses a processes map handling the madvise
1473 * system call. Advisories are classified as either those effecting
1474 * the vm_map_entry structure, or those effecting the underlying
1475 * objects.
1476 */
1477
1478int
a108bf71 1479vm_map_madvise(vm_map_t map, vm_offset_t start, vm_offset_t end, int behav)
984263bc
MD
1480{
1481 vm_map_entry_t current, entry;
1482 int modify_map = 0;
a108bf71 1483 int count;
984263bc
MD
1484
1485 /*
1486 * Some madvise calls directly modify the vm_map_entry, in which case
1487 * we need to use an exclusive lock on the map and we need to perform
1488 * various clipping operations. Otherwise we only need a read-lock
1489 * on the map.
1490 */
1491
a108bf71
MD
1492 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1493
984263bc
MD
1494 switch(behav) {
1495 case MADV_NORMAL:
1496 case MADV_SEQUENTIAL:
1497 case MADV_RANDOM:
1498 case MADV_NOSYNC:
1499 case MADV_AUTOSYNC:
1500 case MADV_NOCORE:
1501 case MADV_CORE:
1502 modify_map = 1;
1503 vm_map_lock(map);
1504 break;
1505 case MADV_WILLNEED:
1506 case MADV_DONTNEED:
1507 case MADV_FREE:
1508 vm_map_lock_read(map);
1509 break;
1510 default:
a108bf71 1511 vm_map_entry_release(count);
984263bc
MD
1512 return (KERN_INVALID_ARGUMENT);
1513 }
1514
1515 /*
1516 * Locate starting entry and clip if necessary.
1517 */
1518
1519 VM_MAP_RANGE_CHECK(map, start, end);
1520
1521 if (vm_map_lookup_entry(map, start, &entry)) {
1522 if (modify_map)
a108bf71 1523 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1524 } else {
1525 entry = entry->next;
1526 }
1527
1528 if (modify_map) {
1529 /*
1530 * madvise behaviors that are implemented in the vm_map_entry.
1531 *
1532 * We clip the vm_map_entry so that behavioral changes are
1533 * limited to the specified address range.
1534 */
1535 for (current = entry;
1536 (current != &map->header) && (current->start < end);
1537 current = current->next
1538 ) {
1539 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1540 continue;
1541
a108bf71 1542 vm_map_clip_end(map, current, end, &count);
984263bc
MD
1543
1544 switch (behav) {
1545 case MADV_NORMAL:
1546 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1547 break;
1548 case MADV_SEQUENTIAL:
1549 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1550 break;
1551 case MADV_RANDOM:
1552 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1553 break;
1554 case MADV_NOSYNC:
1555 current->eflags |= MAP_ENTRY_NOSYNC;
1556 break;
1557 case MADV_AUTOSYNC:
1558 current->eflags &= ~MAP_ENTRY_NOSYNC;
1559 break;
1560 case MADV_NOCORE:
1561 current->eflags |= MAP_ENTRY_NOCOREDUMP;
1562 break;
1563 case MADV_CORE:
1564 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1565 break;
1566 default:
1567 break;
1568 }
a108bf71 1569 vm_map_simplify_entry(map, current, &count);
984263bc
MD
1570 }
1571 vm_map_unlock(map);
1572 } else {
1573 vm_pindex_t pindex;
1574 int count;
1575
1576 /*
1577 * madvise behaviors that are implemented in the underlying
1578 * vm_object.
1579 *
1580 * Since we don't clip the vm_map_entry, we have to clip
1581 * the vm_object pindex and count.
1582 */
1583 for (current = entry;
1584 (current != &map->header) && (current->start < end);
1585 current = current->next
1586 ) {
1587 vm_offset_t useStart;
1588
1589 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1590 continue;
1591
1592 pindex = OFF_TO_IDX(current->offset);
1593 count = atop(current->end - current->start);
1594 useStart = current->start;
1595
1596 if (current->start < start) {
1597 pindex += atop(start - current->start);
1598 count -= atop(start - current->start);
1599 useStart = start;
1600 }
1601 if (current->end > end)
1602 count -= atop(current->end - end);
1603
1604 if (count <= 0)
1605 continue;
1606
1607 vm_object_madvise(current->object.vm_object,
1608 pindex, count, behav);
1609 if (behav == MADV_WILLNEED) {
1610 pmap_object_init_pt(
1611 map->pmap,
1612 useStart,
083a7402 1613 current->protection,
984263bc
MD
1614 current->object.vm_object,
1615 pindex,
1616 (count << PAGE_SHIFT),
1617 MAP_PREFAULT_MADVISE
1618 );
1619 }
1620 }
1621 vm_map_unlock_read(map);
1622 }
a108bf71 1623 vm_map_entry_release(count);
984263bc
MD
1624 return(0);
1625}
1626
1627
1628/*
1629 * vm_map_inherit:
1630 *
1631 * Sets the inheritance of the specified address
1632 * range in the target map. Inheritance
1633 * affects how the map will be shared with
1634 * child maps at the time of vm_map_fork.
1635 */
1636int
1637vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1638 vm_inherit_t new_inheritance)
1639{
1640 vm_map_entry_t entry;
1641 vm_map_entry_t temp_entry;
a108bf71 1642 int count;
984263bc
MD
1643
1644 switch (new_inheritance) {
1645 case VM_INHERIT_NONE:
1646 case VM_INHERIT_COPY:
1647 case VM_INHERIT_SHARE:
1648 break;
1649 default:
1650 return (KERN_INVALID_ARGUMENT);
1651 }
1652
a108bf71 1653 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1654 vm_map_lock(map);
1655
1656 VM_MAP_RANGE_CHECK(map, start, end);
1657
1658 if (vm_map_lookup_entry(map, start, &temp_entry)) {
1659 entry = temp_entry;
a108bf71 1660 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1661 } else
1662 entry = temp_entry->next;
1663
1664 while ((entry != &map->header) && (entry->start < end)) {
a108bf71 1665 vm_map_clip_end(map, entry, end, &count);
984263bc
MD
1666
1667 entry->inheritance = new_inheritance;
1668
a108bf71 1669 vm_map_simplify_entry(map, entry, &count);
984263bc
MD
1670
1671 entry = entry->next;
1672 }
984263bc 1673 vm_map_unlock(map);
a108bf71 1674 vm_map_entry_release(count);
984263bc
MD
1675 return (KERN_SUCCESS);
1676}
1677
1678/*
1679 * Implement the semantics of mlock
1680 */
1681int
57e43348
MD
1682vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t real_end,
1683 boolean_t new_pageable)
984263bc
MD
1684{
1685 vm_map_entry_t entry;
1686 vm_map_entry_t start_entry;
1687 vm_offset_t end;
1688 int rv = KERN_SUCCESS;
a108bf71 1689 int count;
984263bc 1690
a108bf71 1691 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1692 vm_map_lock(map);
1693 VM_MAP_RANGE_CHECK(map, start, real_end);
1694 end = real_end;
1695
a108bf71 1696 start_entry = vm_map_clip_range(map, start, end, &count, MAP_CLIP_NO_HOLES);
984263bc
MD
1697 if (start_entry == NULL) {
1698 vm_map_unlock(map);
a108bf71 1699 vm_map_entry_release(count);
984263bc
MD
1700 return (KERN_INVALID_ADDRESS);
1701 }
1702
1703 if (new_pageable == 0) {
1704 entry = start_entry;
1705 while ((entry != &map->header) && (entry->start < end)) {
1706 vm_offset_t save_start;
1707 vm_offset_t save_end;
1708
1709 /*
1710 * Already user wired or hard wired (trivial cases)
1711 */
1712 if (entry->eflags & MAP_ENTRY_USER_WIRED) {
1713 entry = entry->next;
1714 continue;
1715 }
1716 if (entry->wired_count != 0) {
1717 entry->wired_count++;
1718 entry->eflags |= MAP_ENTRY_USER_WIRED;
1719 entry = entry->next;
1720 continue;
1721 }
1722
1723 /*
1724 * A new wiring requires instantiation of appropriate
1725 * management structures and the faulting in of the
1726 * page.
1727 */
1728 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1729 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1730 if (copyflag && ((entry->protection & VM_PROT_WRITE) != 0)) {
1731
1732 vm_object_shadow(&entry->object.vm_object,
1733 &entry->offset,
1734 atop(entry->end - entry->start));
1735 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1736
1737 } else if (entry->object.vm_object == NULL &&
1738 !map->system_map) {
1739
1740 entry->object.vm_object =
1741 vm_object_allocate(OBJT_DEFAULT,
1742 atop(entry->end - entry->start));
1743 entry->offset = (vm_offset_t) 0;
1744
1745 }
1746 }
1747 entry->wired_count++;
1748 entry->eflags |= MAP_ENTRY_USER_WIRED;
1749
1750 /*
1751 * Now fault in the area. The map lock needs to be
1752 * manipulated to avoid deadlocks. The in-transition
1753 * flag protects the entries.
1754 */
1755 save_start = entry->start;
1756 save_end = entry->end;
1757 vm_map_unlock(map);
1758 map->timestamp++;
1759 rv = vm_fault_user_wire(map, save_start, save_end);
1760 vm_map_lock(map);
1761 if (rv) {
1762 CLIP_CHECK_BACK(entry, save_start);
1763 for (;;) {
1764 KASSERT(entry->wired_count == 1, ("bad wired_count on entry"));
1765 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1766 entry->wired_count = 0;
1767 if (entry->end == save_end)
1768 break;
1769 entry = entry->next;
1770 KASSERT(entry != &map->header, ("bad entry clip during backout"));
1771 }
1772 end = save_start; /* unwire the rest */
1773 break;
1774 }
1775 /*
1776 * note that even though the entry might have been
1777 * clipped, the USER_WIRED flag we set prevents
1778 * duplication so we do not have to do a
1779 * clip check.
1780 */
1781 entry = entry->next;
1782 }
1783
1784 /*
1785 * If we failed fall through to the unwiring section to
1786 * unwire what we had wired so far. 'end' has already
1787 * been adjusted.
1788 */
1789 if (rv)
1790 new_pageable = 1;
1791
1792 /*
1793 * start_entry might have been clipped if we unlocked the
1794 * map and blocked. No matter how clipped it has gotten
1795 * there should be a fragment that is on our start boundary.
1796 */
1797 CLIP_CHECK_BACK(start_entry, start);
1798 }
1799
1800 /*
1801 * Deal with the unwiring case.
1802 */
1803 if (new_pageable) {
1804 /*
1805 * This is the unwiring case. We must first ensure that the
1806 * range to be unwired is really wired down. We know there
1807 * are no holes.
1808 */
1809 entry = start_entry;
1810 while ((entry != &map->header) && (entry->start < end)) {
1811 if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
1812 rv = KERN_INVALID_ARGUMENT;
1813 goto done;
1814 }
1815 KASSERT(entry->wired_count != 0, ("wired count was 0 with USER_WIRED set! %p", entry));
1816 entry = entry->next;
1817 }
1818
1819 /*
1820 * Now decrement the wiring count for each region. If a region
1821 * becomes completely unwired, unwire its physical pages and
1822 * mappings.
1823 */
b4eddbac
DR
1824 /*
1825 * The map entries are processed in a loop, checking to
1826 * make sure the entry is wired and asserting it has a wired
1827 * count. However, another loop was inserted more-or-less in
1828 * the middle of the unwiring path. This loop picks up the
1829 * "entry" loop variable from the first loop without first
1830 * setting it to start_entry. Naturally, the secound loop
1831 * is never entered and the pages backing the entries are
1832 * never unwired. This can lead to a leak of wired pages.
1833 */
1834 entry = start_entry;
984263bc
MD
1835 while ((entry != &map->header) && (entry->start < end)) {
1836 KASSERT(entry->eflags & MAP_ENTRY_USER_WIRED, ("expected USER_WIRED on entry %p", entry));
1837 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1838 entry->wired_count--;
1839 if (entry->wired_count == 0)
1840 vm_fault_unwire(map, entry->start, entry->end);
1841 entry = entry->next;
1842 }
1843 }
1844done:
a108bf71 1845 vm_map_unclip_range(map, start_entry, start, real_end, &count,
984263bc
MD
1846 MAP_CLIP_NO_HOLES);
1847 map->timestamp++;
1848 vm_map_unlock(map);
a108bf71 1849 vm_map_entry_release(count);
984263bc
MD
1850 return (rv);
1851}
1852
1853/*
cde87949 1854 * vm_map_wire:
984263bc
MD
1855 *
1856 * Sets the pageability of the specified address
1857 * range in the target map. Regions specified
1858 * as not pageable require locked-down physical
1859 * memory and physical page maps.
1860 *
1861 * The map must not be locked, but a reference
1862 * must remain to the map throughout the call.
a108bf71
MD
1863 *
1864 * This function may be called via the zalloc path and must properly
1865 * reserve map entries for kernel_map.
984263bc
MD
1866 */
1867int
e1359933 1868vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t real_end, int kmflags)
984263bc
MD
1869{
1870 vm_map_entry_t entry;
1871 vm_map_entry_t start_entry;
1872 vm_offset_t end;
1873 int rv = KERN_SUCCESS;
a108bf71 1874 int count;
984263bc
MD
1875 int s;
1876
e1359933 1877 if (kmflags & KM_KRESERVE)
a108bf71 1878 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
a108bf71
MD
1879 else
1880 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1881 vm_map_lock(map);
1882 VM_MAP_RANGE_CHECK(map, start, real_end);
1883 end = real_end;
1884
a108bf71 1885 start_entry = vm_map_clip_range(map, start, end, &count, MAP_CLIP_NO_HOLES);
984263bc
MD
1886 if (start_entry == NULL) {
1887 vm_map_unlock(map);
a108bf71
MD
1888 rv = KERN_INVALID_ADDRESS;
1889 goto failure;
984263bc 1890 }
e1359933 1891 if ((kmflags & KM_PAGEABLE) == 0) {
984263bc
MD
1892 /*
1893 * Wiring.
1894 *
1895 * 1. Holding the write lock, we create any shadow or zero-fill
1896 * objects that need to be created. Then we clip each map
1897 * entry to the region to be wired and increment its wiring
1898 * count. We create objects before clipping the map entries
1899 * to avoid object proliferation.
1900 *
1901 * 2. We downgrade to a read lock, and call vm_fault_wire to
1902 * fault in the pages for any newly wired area (wired_count is
1903 * 1).
1904 *
1905 * Downgrading to a read lock for vm_fault_wire avoids a
1906 * possible deadlock with another process that may have faulted
1907 * on one of the pages to be wired (it would mark the page busy,
1908 * blocking us, then in turn block on the map lock that we
1909 * hold). Because of problems in the recursive lock package,
1910 * we cannot upgrade to a write lock in vm_map_lookup. Thus,
1911 * any actions that require the write lock must be done
1912 * beforehand. Because we keep the read lock on the map, the
1913 * copy-on-write status of the entries we modify here cannot
1914 * change.
1915 */
1916
1917 entry = start_entry;
1918 while ((entry != &map->header) && (entry->start < end)) {
1919 /*
1920 * Trivial case if the entry is already wired
1921 */
1922 if (entry->wired_count) {
1923 entry->wired_count++;
1924 entry = entry->next;
1925 continue;
1926 }
1927
1928 /*
1929 * The entry is being newly wired, we have to setup
1930 * appropriate management structures. A shadow
1931 * object is required for a copy-on-write region,
1932 * or a normal object for a zero-fill region. We
1933 * do not have to do this for entries that point to sub
1934 * maps because we won't hold the lock on the sub map.
1935 */
1936 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1937 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1938 if (copyflag &&
1939 ((entry->protection & VM_PROT_WRITE) != 0)) {
1940
1941 vm_object_shadow(&entry->object.vm_object,
1942 &entry->offset,
1943 atop(entry->end - entry->start));
1944 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1945 } else if (entry->object.vm_object == NULL &&
1946 !map->system_map) {
1947 entry->object.vm_object =
1948 vm_object_allocate(OBJT_DEFAULT,
1949 atop(entry->end - entry->start));
1950 entry->offset = (vm_offset_t) 0;
1951 }
1952 }
1953
1954 entry->wired_count++;
1955 entry = entry->next;
1956 }
1957
1958 /*
1959 * Pass 2.
1960 */
1961
1962 /*
1963 * HACK HACK HACK HACK
1964 *
1965 * Unlock the map to avoid deadlocks. The in-transit flag
1966 * protects us from most changes but note that
1967 * clipping may still occur. To prevent clipping from
1968 * occuring after the unlock, except for when we are
1969 * blocking in vm_fault_wire, we must run at splvm().
1970 * Otherwise our accesses to entry->start and entry->end
1971 * could be corrupted. We have to set splvm() prior to
1972 * unlocking so start_entry does not change out from
1973 * under us at the very beginning of the loop.
1974 *
1975 * HACK HACK HACK HACK
1976 */
1977
1978 s = splvm();
1979 vm_map_unlock(map);
1980
1981 entry = start_entry;
1982 while (entry != &map->header && entry->start < end) {
1983 /*
1984 * If vm_fault_wire fails for any page we need to undo
1985 * what has been done. We decrement the wiring count
1986 * for those pages which have not yet been wired (now)
1987 * and unwire those that have (later).
1988 */
1989 vm_offset_t save_start = entry->start;
1990 vm_offset_t save_end = entry->end;
1991
1992 if (entry->wired_count == 1)
1993 rv = vm_fault_wire(map, entry->start, entry->end);
1994 if (rv) {
1995 CLIP_CHECK_BACK(entry, save_start);
1996 for (;;) {
1997 KASSERT(entry->wired_count == 1, ("wired_count changed unexpectedly"));
1998 entry->wired_count = 0;
1999 if (entry->end == save_end)
2000 break;
2001 entry = entry->next;
2002 KASSERT(entry != &map->header, ("bad entry clip during backout"));
2003 }
2004 end = save_start;
2005 break;
2006 }
2007 CLIP_CHECK_FWD(entry, save_end);
2008 entry = entry->next;
2009 }
2010 splx(s);
2011
2012 /*
2013 * relock. start_entry is still IN_TRANSITION and must
2014 * still exist, but may have been clipped (handled just
2015 * below).
2016 */
2017 vm_map_lock(map);
2018
2019 /*
2020 * If a failure occured undo everything by falling through
2021 * to the unwiring code. 'end' has already been adjusted
2022 * appropriately.
2023 */
2024 if (rv)
e1359933 2025 kmflags |= KM_PAGEABLE;
984263bc
MD
2026
2027 /*
2028 * start_entry might have been clipped if we unlocked the
2029 * map and blocked. No matter how clipped it has gotten
2030 * there should be a fragment that is on our start boundary.
2031 */
2032 CLIP_CHECK_BACK(start_entry, start);
2033 }
2034
e1359933 2035 if (kmflags & KM_PAGEABLE) {
984263bc
MD
2036 /*
2037 * This is the unwiring case. We must first ensure that the
2038 * range to be unwired is really wired down. We know there
2039 * are no holes.
2040 */
2041 entry = start_entry;
2042 while ((entry != &map->header) && (entry->start < end)) {
2043 if (entry->wired_count == 0) {
2044 rv = KERN_INVALID_ARGUMENT;
2045 goto done;
2046 }
2047 entry = entry->next;
2048 }
2049
2050 /*
2051 * Now decrement the wiring count for each region. If a region
2052 * becomes completely unwired, unwire its physical pages and
2053 * mappings.
2054 */
2055 entry = start_entry;
2056 while ((entry != &map->header) && (entry->start < end)) {
2057 entry->wired_count--;
2058 if (entry->wired_count == 0)
2059 vm_fault_unwire(map, entry->start, entry->end);
2060 entry = entry->next;
2061 }
2062 }
2063done:
a108bf71 2064 vm_map_unclip_range(map, start_entry, start, real_end, &count,
984263bc
MD
2065 MAP_CLIP_NO_HOLES);
2066 map->timestamp++;
2067 vm_map_unlock(map);
a108bf71 2068failure:
e1359933 2069 if (kmflags & KM_KRESERVE)
a108bf71 2070 vm_map_entry_krelease(count);
a108bf71
MD
2071 else
2072 vm_map_entry_release(count);
984263bc
MD
2073 return (rv);
2074}
2075
a108bf71
MD
2076/*
2077 * vm_map_set_wired_quick()
2078 *
2079 * Mark a newly allocated address range as wired but do not fault in
2080 * the pages. The caller is expected to load the pages into the object.
2081 *
2082 * The map must be locked on entry and will remain locked on return.
2083 */
2084void
2085vm_map_set_wired_quick(vm_map_t map, vm_offset_t addr, vm_size_t size, int *countp)
2086{
2087 vm_map_entry_t scan;
2088 vm_map_entry_t entry;
2089
2090 entry = vm_map_clip_range(map, addr, addr + size, countp, MAP_CLIP_NO_HOLES);
2091 for (scan = entry; scan != &map->header && scan->start < addr + size; scan = scan->next) {
2092 KKASSERT(entry->wired_count == 0);
2093 entry->wired_count = 1;
2094 }
2095 vm_map_unclip_range(map, entry, addr, addr + size, countp, MAP_CLIP_NO_HOLES);
2096}
2097
984263bc
MD
2098/*
2099 * vm_map_clean
2100 *
2101 * Push any dirty cached pages in the address range to their pager.
2102 * If syncio is TRUE, dirty pages are written synchronously.
2103 * If invalidate is TRUE, any cached pages are freed as well.
2104 *
2105 * Returns an error if any part of the specified range is not mapped.
2106 */
2107int
57e43348
MD
2108vm_map_clean(vm_map_t map, vm_offset_t start, vm_offset_t end, boolean_t syncio,
2109 boolean_t invalidate)
984263bc
MD
2110{
2111 vm_map_entry_t current;
2112 vm_map_entry_t entry;
2113 vm_size_t size;
2114 vm_object_t object;
2115 vm_ooffset_t offset;
2116
2117 vm_map_lock_read(map);
2118 VM_MAP_RANGE_CHECK(map, start, end);
2119 if (!vm_map_lookup_entry(map, start, &entry)) {
2120 vm_map_unlock_read(map);
2121 return (KERN_INVALID_ADDRESS);
2122 }
2123 /*
2124 * Make a first pass to check for holes.
2125 */
2126 for (current = entry; current->start < end; current = current->next) {
2127 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2128 vm_map_unlock_read(map);
2129 return (KERN_INVALID_ARGUMENT);
2130 }
2131 if (end > current->end &&
2132 (current->next == &map->header ||
2133 current->end != current->next->start)) {
2134 vm_map_unlock_read(map);
2135 return (KERN_INVALID_ADDRESS);
2136 }
2137 }
2138
2139 if (invalidate)
2140 pmap_remove(vm_map_pmap(map), start, end);
2141 /*
2142 * Make a second pass, cleaning/uncaching pages from the indicated
2143 * objects as we go.
2144 */
2145 for (current = entry; current->start < end; current = current->next) {
2146 offset = current->offset + (start - current->start);
2147 size = (end <= current->end ? end : current->end) - start;
2148 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2149 vm_map_t smap;
2150 vm_map_entry_t tentry;
2151 vm_size_t tsize;
2152
2153 smap = current->object.sub_map;
2154 vm_map_lock_read(smap);
2155 (void) vm_map_lookup_entry(smap, offset, &tentry);
2156 tsize = tentry->end - offset;
2157 if (tsize < size)
2158 size = tsize;
2159 object = tentry->object.vm_object;
2160 offset = tentry->offset + (offset - tentry->start);
2161 vm_map_unlock_read(smap);
2162 } else {
2163 object = current->object.vm_object;
2164 }
2165 /*
2166 * Note that there is absolutely no sense in writing out
2167 * anonymous objects, so we track down the vnode object
2168 * to write out.
2169 * We invalidate (remove) all pages from the address space
2170 * anyway, for semantic correctness.
2171 *
2172 * note: certain anonymous maps, such as MAP_NOSYNC maps,
2173 * may start out with a NULL object.
2174 */
2175 while (object && object->backing_object) {
2176 object = object->backing_object;
2177 offset += object->backing_object_offset;
2178 if (object->size < OFF_TO_IDX( offset + size))
2179 size = IDX_TO_OFF(object->size) - offset;
2180 }
2181 if (object && (object->type == OBJT_VNODE) &&
2182 (current->protection & VM_PROT_WRITE)) {
2183 /*
2184 * Flush pages if writing is allowed, invalidate them
2185 * if invalidation requested. Pages undergoing I/O
2186 * will be ignored by vm_object_page_remove().
2187 *
2188 * We cannot lock the vnode and then wait for paging
2189 * to complete without deadlocking against vm_fault.
2190 * Instead we simply call vm_object_page_remove() and
2191 * allow it to block internally on a page-by-page
2192 * basis when it encounters pages undergoing async
2193 * I/O.
2194 */
2195 int flags;
2196
2197 vm_object_reference(object);
41a01a4d
MD
2198 vn_lock(object->handle, NULL,
2199 LK_EXCLUSIVE | LK_RETRY, curthread);
984263bc
MD
2200 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
2201 flags |= invalidate ? OBJPC_INVAL : 0;
2202 vm_object_page_clean(object,
2203 OFF_TO_IDX(offset),
2204 OFF_TO_IDX(offset + size + PAGE_MASK),
2205 flags);
41a01a4d 2206 VOP_UNLOCK(object->handle, NULL, 0, curthread);
984263bc
MD
2207 vm_object_deallocate(object);
2208 }
2209 if (object && invalidate &&
2210 ((object->type == OBJT_VNODE) ||
2211 (object->type == OBJT_DEVICE))) {
2212 vm_object_reference(object);
2213 vm_object_page_remove(object,
2214 OFF_TO_IDX(offset),
2215 OFF_TO_IDX(offset + size + PAGE_MASK),
10e49b8b 2216 TRUE);
984263bc
MD
2217 vm_object_deallocate(object);
2218 }
2219 start += size;
2220 }
2221
2222 vm_map_unlock_read(map);
2223 return (KERN_SUCCESS);
2224}
2225
2226/*
2227 * vm_map_entry_unwire: [ internal use only ]
2228 *
2229 * Make the region specified by this entry pageable.
2230 *
2231 * The map in question should be locked.
2232 * [This is the reason for this routine's existence.]
2233 */
2234static void
a108bf71 2235vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
984263bc
MD
2236{
2237 vm_fault_unwire(map, entry->start, entry->end);
2238 entry->wired_count = 0;
2239}
2240
2241/*
2242 * vm_map_entry_delete: [ internal use only ]
2243 *
2244 * Deallocate the given entry from the target map.
2245 */
2246static void
a108bf71 2247vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc
MD
2248{
2249 vm_map_entry_unlink(map, entry);
2250 map->size -= entry->end - entry->start;
2251
2252 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2253 vm_object_deallocate(entry->object.vm_object);
2254 }
2255
a108bf71 2256 vm_map_entry_dispose(map, entry, countp);
984263bc
MD
2257}
2258
2259/*
2260 * vm_map_delete: [ internal use only ]
2261 *
2262 * Deallocates the given address range from the target
2263 * map.
2264 */
2265int
a108bf71 2266vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end, int *countp)
984263bc
MD
2267{
2268 vm_object_t object;
2269 vm_map_entry_t entry;
2270 vm_map_entry_t first_entry;
2271
2272 /*
2273 * Find the start of the region, and clip it
2274 */
2275
2276again:
2277 if (!vm_map_lookup_entry(map, start, &first_entry))
2278 entry = first_entry->next;
2279 else {
2280 entry = first_entry;
a108bf71 2281 vm_map_clip_start(map, entry, start, countp);
984263bc
MD
2282 /*
2283 * Fix the lookup hint now, rather than each time though the
2284 * loop.
2285 */
2286 SAVE_HINT(map, entry->prev);
2287 }
2288
2289 /*
2290 * Save the free space hint
2291 */
2292
2293 if (entry == &map->header) {
2294 map->first_free = &map->header;
2295 } else if (map->first_free->start >= start) {
2296 map->first_free = entry->prev;
2297 }
2298
2299 /*
2300 * Step through all entries in this region
2301 */
2302
2303 while ((entry != &map->header) && (entry->start < end)) {
2304 vm_map_entry_t next;
2305 vm_offset_t s, e;
2306 vm_pindex_t offidxstart, offidxend, count;
2307
2308 /*
2309 * If we hit an in-transition entry we have to sleep and
2310 * retry. It's easier (and not really slower) to just retry
2311 * since this case occurs so rarely and the hint is already
2312 * pointing at the right place. We have to reset the
2313 * start offset so as not to accidently delete an entry
2314 * another process just created in vacated space.
2315 */
2316 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2317 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2318 start = entry->start;
12e4aaff
MD
2319 ++mycpu->gd_cnt.v_intrans_coll;
2320 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
2321 vm_map_transition_wait(map);
2322 goto again;
2323 }
a108bf71 2324 vm_map_clip_end(map, entry, end, countp);
984263bc
MD
2325
2326 s = entry->start;
2327 e = entry->end;
2328 next = entry->next;
2329
2330 offidxstart = OFF_TO_IDX(entry->offset);
2331 count = OFF_TO_IDX(e - s);
2332 object = entry->object.vm_object;
2333
2334 /*
2335 * Unwire before removing addresses from the pmap; otherwise,
2336 * unwiring will put the entries back in the pmap.
2337 */
2338 if (entry->wired_count != 0) {
2339 vm_map_entry_unwire(map, entry);
2340 }
2341
2342 offidxend = offidxstart + count;
2343
2344 if ((object == kernel_object) || (object == kmem_object)) {
2345 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2346 } else {
2347 pmap_remove(map->pmap, s, e);
2348 if (object != NULL &&
2349 object->ref_count != 1 &&
2350 (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING &&
2351 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2352 vm_object_collapse(object);
2353 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2354 if (object->type == OBJT_SWAP) {
2355 swap_pager_freespace(object, offidxstart, count);
2356 }
2357 if (offidxend >= object->size &&
2358 offidxstart < object->size) {
2359 object->size = offidxstart;
2360 }
2361 }
2362 }
2363
2364 /*
2365 * Delete the entry (which may delete the object) only after
2366 * removing all pmap entries pointing to its pages.
2367 * (Otherwise, its page frames may be reallocated, and any
2368 * modify bits will be set in the wrong object!)
2369 */
a108bf71 2370 vm_map_entry_delete(map, entry, countp);
984263bc
MD
2371 entry = next;
2372 }
2373 return (KERN_SUCCESS);
2374}
2375
2376/*
2377 * vm_map_remove:
2378 *
2379 * Remove the given address range from the target map.
2380 * This is the exported form of vm_map_delete.
2381 */
2382int
a108bf71 2383vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
984263bc 2384{
03aa8d99 2385 int result;
a108bf71 2386 int count;
984263bc 2387
a108bf71 2388 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2389 vm_map_lock(map);
2390 VM_MAP_RANGE_CHECK(map, start, end);
a108bf71 2391 result = vm_map_delete(map, start, end, &count);
984263bc 2392 vm_map_unlock(map);
a108bf71 2393 vm_map_entry_release(count);
984263bc 2394
984263bc
MD
2395 return (result);
2396}
2397
2398/*
2399 * vm_map_check_protection:
2400 *
2401 * Assert that the target map allows the specified
2402 * privilege on the entire address region given.
2403 * The entire region must be allocated.
2404 */
2405boolean_t
2406vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2407 vm_prot_t protection)
2408{
2409 vm_map_entry_t entry;
2410 vm_map_entry_t tmp_entry;
2411
2412 if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2413 return (FALSE);
2414 }
2415 entry = tmp_entry;
2416
2417 while (start < end) {
2418 if (entry == &map->header) {
2419 return (FALSE);
2420 }
2421 /*
2422 * No holes allowed!
2423 */
2424
2425 if (start < entry->start) {
2426 return (FALSE);
2427 }
2428 /*
2429 * Check protection associated with entry.
2430 */
2431
2432 if ((entry->protection & protection) != protection) {
2433 return (FALSE);
2434 }
2435 /* go to next entry */
2436
2437 start = entry->end;
2438 entry = entry->next;
2439 }
2440 return (TRUE);
2441}
2442
2443/*
2444 * Split the pages in a map entry into a new object. This affords
2445 * easier removal of unused pages, and keeps object inheritance from
2446 * being a negative impact on memory usage.
2447 */
2448static void
a108bf71 2449vm_map_split(vm_map_entry_t entry)
984263bc
MD
2450{
2451 vm_page_t m;
2452 vm_object_t orig_object, new_object, source;
2453 vm_offset_t s, e;
2454 vm_pindex_t offidxstart, offidxend, idx;
2455 vm_size_t size;
2456 vm_ooffset_t offset;
2457
2458 orig_object = entry->object.vm_object;
2459 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
2460 return;
2461 if (orig_object->ref_count <= 1)
2462 return;
2463
2464 offset = entry->offset;
2465 s = entry->start;
2466 e = entry->end;
2467
2468 offidxstart = OFF_TO_IDX(offset);
2469 offidxend = offidxstart + OFF_TO_IDX(e - s);
2470 size = offidxend - offidxstart;
2471
2472 new_object = vm_pager_allocate(orig_object->type,
2473 NULL, IDX_TO_OFF(size), VM_PROT_ALL, 0LL);
2474 if (new_object == NULL)
2475 return;
2476
2477 source = orig_object->backing_object;
2478 if (source != NULL) {
2479 vm_object_reference(source); /* Referenced by new_object */
2480 LIST_INSERT_HEAD(&source->shadow_head,
2481 new_object, shadow_list);
2482 vm_object_clear_flag(source, OBJ_ONEMAPPING);
2483 new_object->backing_object_offset =
2484 orig_object->backing_object_offset + IDX_TO_OFF(offidxstart);
2485 new_object->backing_object = source;
2486 source->shadow_count++;
2487 source->generation++;
2488 }
2489
2490 for (idx = 0; idx < size; idx++) {
2491 vm_page_t m;
06ecca5a 2492 int ss; /* s used */
984263bc 2493
06ecca5a
MD
2494 /*
2495 * splvm protection is required to avoid a race between
2496 * the lookup and an interrupt/unbusy/free and our busy
2497 * check.
2498 */
2499 ss = splvm();
984263bc
MD
2500 retry:
2501 m = vm_page_lookup(orig_object, offidxstart + idx);
06ecca5a
MD
2502 if (m == NULL) {
2503 splx(ss);
984263bc 2504 continue;
06ecca5a 2505 }
984263bc
MD
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;
984263bc
MD
2516 vm_page_busy(m);
2517 vm_page_rename(m, new_object, idx);
2518 /* page automatically made dirty by rename and cache handled */
2519 vm_page_busy(m);
06ecca5a 2520 splx(ss);
984263bc
MD
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
06ecca5a
MD
2534 /*
2535 * Wakeup the pages we played with. No spl protection is needed
2536 * for a simple wakeup.
2537 */
984263bc
MD
2538 for (idx = 0; idx < size; idx++) {
2539 m = vm_page_lookup(new_object, idx);
06ecca5a 2540 if (m)
984263bc 2541 vm_page_wakeup(m);
984263bc
MD
2542 }
2543
2544 entry->object.vm_object = new_object;
2545 entry->offset = 0LL;
2546 vm_object_deallocate(orig_object);
2547}
2548
2549/*
2550 * vm_map_copy_entry:
2551 *
2552 * Copies the contents of the source entry to the destination
2553 * entry. The entries *must* be aligned properly.
2554 */
2555static void
a108bf71
MD
2556vm_map_copy_entry(vm_map_t src_map, vm_map_t dst_map,
2557 vm_map_entry_t src_entry, vm_map_entry_t dst_entry)
984263bc
MD
2558{
2559 vm_object_t src_object;
2560
2561 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2562 return;
2563
2564 if (src_entry->wired_count == 0) {
2565
2566 /*
2567 * If the source entry is marked needs_copy, it is already
2568 * write-protected.
2569 */
2570 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2571 pmap_protect(src_map->pmap,
2572 src_entry->start,
2573 src_entry->end,
2574 src_entry->protection & ~VM_PROT_WRITE);
2575 }
2576
2577 /*
2578 * Make a copy of the object.
2579 */
2580 if ((src_object = src_entry->object.vm_object) != NULL) {
2581
2582 if ((src_object->handle == NULL) &&
2583 (src_object->type == OBJT_DEFAULT ||
2584 src_object->type == OBJT_SWAP)) {
2585 vm_object_collapse(src_object);
2586 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2587 vm_map_split(src_entry);
2588 src_object = src_entry->object.vm_object;
2589 }
2590 }
2591
2592 vm_object_reference(src_object);
2593 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2594 dst_entry->object.vm_object = src_object;
2595 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2596 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2597 dst_entry->offset = src_entry->offset;
2598 } else {
2599 dst_entry->object.vm_object = NULL;
2600 dst_entry->offset = 0;
2601 }
2602
2603 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2604 dst_entry->end - dst_entry->start, src_entry->start);
2605 } else {
2606 /*
2607 * Of course, wired down pages can't be set copy-on-write.
2608 * Cause wired pages to be copied into the new map by
2609 * simulating faults (the new pages are pageable)
2610 */
2611 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2612 }
2613}
2614
2615/*
2616 * vmspace_fork:
2617 * Create a new process vmspace structure and vm_map
2618 * based on those of an existing process. The new map
2619 * is based on the old map, according to the inheritance
2620 * values on the regions in that map.
2621 *
2622 * The source map must not be locked.
2623 */
2624struct vmspace *
a108bf71 2625vmspace_fork(struct vmspace *vm1)
984263bc
MD
2626{
2627 struct vmspace *vm2;
2628 vm_map_t old_map = &vm1->vm_map;
2629 vm_map_t new_map;
2630 vm_map_entry_t old_entry;
2631 vm_map_entry_t new_entry;
2632 vm_object_t object;
a108bf71 2633 int count;
984263bc
MD
2634
2635 vm_map_lock(old_map);
2636 old_map->infork = 1;
2637
239b4df9
MD
2638 /*
2639 * XXX Note: upcalls are not copied.
2640 */
984263bc
MD
2641 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
2642 bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
239b4df9 2643 (caddr_t)&vm1->vm_endcopy - (caddr_t)&vm1->vm_startcopy);
984263bc
MD
2644 new_map = &vm2->vm_map; /* XXX */
2645 new_map->timestamp = 1;
2646
a108bf71 2647 count = 0;
984263bc 2648 old_entry = old_map->header.next;
a108bf71
MD
2649 while (old_entry != &old_map->header) {
2650 ++count;
2651 old_entry = old_entry->next;
2652 }
984263bc 2653
a108bf71
MD
2654 count = vm_map_entry_reserve(count + MAP_RESERVE_COUNT);
2655
2656 old_entry = old_map->header.next;
984263bc
MD
2657 while (old_entry != &old_map->header) {
2658 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
2659 panic("vm_map_fork: encountered a submap");
2660
2661 switch (old_entry->inheritance) {
2662 case VM_INHERIT_NONE:
2663 break;
2664
2665 case VM_INHERIT_SHARE:
2666 /*
2667 * Clone the entry, creating the shared object if necessary.
2668 */
2669 object = old_entry->object.vm_object;
2670 if (object == NULL) {
2671 object = vm_object_allocate(OBJT_DEFAULT,
2672 atop(old_entry->end - old_entry->start));
2673 old_entry->object.vm_object = object;
2674 old_entry->offset = (vm_offset_t) 0;
2675 }
2676
2677 /*
2678 * Add the reference before calling vm_object_shadow
2679 * to insure that a shadow object is created.
2680 */
2681 vm_object_reference(object);
2682 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2683 vm_object_shadow(&old_entry->object.vm_object,
2684 &old_entry->offset,
2685 atop(old_entry->end - old_entry->start));
2686 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2687 /* Transfer the second reference too. */
2688 vm_object_reference(
2689 old_entry->object.vm_object);
2690 vm_object_deallocate(object);
2691 object = old_entry->object.vm_object;
2692 }
2693 vm_object_clear_flag(object, OBJ_ONEMAPPING);
2694
2695 /*
2696 * Clone the entry, referencing the shared object.
2697 */
a108bf71 2698 new_entry = vm_map_entry_create(new_map, &count);
984263bc
MD
2699 *new_entry = *old_entry;
2700 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2701 new_entry->wired_count = 0;
2702
2703 /*
2704 * Insert the entry into the new map -- we know we're
2705 * inserting at the end of the new map.
2706 */
2707
2708 vm_map_entry_link(new_map, new_map->header.prev,
2709 new_entry);
2710
2711 /*
2712 * Update the physical map
2713 */
2714
2715 pmap_copy(new_map->pmap, old_map->pmap,
2716 new_entry->start,
2717 (old_entry->end - old_entry->start),
2718 old_entry->start);
2719 break;
2720
2721 case VM_INHERIT_COPY:
2722 /*
2723 * Clone the entry and link into the map.
2724 */
a108bf71 2725 new_entry = vm_map_entry_create(new_map, &count);
984263bc
MD
2726 *new_entry = *old_entry;
2727 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2728 new_entry->wired_count = 0;
2729 new_entry->object.vm_object = NULL;
2730 vm_map_entry_link(new_map, new_map->header.prev,
2731 new_entry);
2732 vm_map_copy_entry(old_map, new_map, old_entry,
2733 new_entry);
2734 break;
2735 }
2736 old_entry = old_entry->next;
2737 }
2738
2739 new_map->size = old_map->size;
2740 old_map->infork = 0;
2741 vm_map_unlock(old_map);
a108bf71 2742 vm_map_entry_release(count);
984263bc
MD
2743
2744 return (vm2);
2745}
2746
2747int
2748vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
2749 vm_prot_t prot, vm_prot_t max, int cow)
2750{
2751 vm_map_entry_t prev_entry;
2752 vm_map_entry_t new_stack_entry;
2753 vm_size_t init_ssize;
2754 int rv;
a108bf71 2755 int count;
984263bc
MD
2756
2757 if (VM_MIN_ADDRESS > 0 && addrbos < VM_MIN_ADDRESS)
2758 return (KERN_NO_SPACE);
2759
2760 if (max_ssize < sgrowsiz)
2761 init_ssize = max_ssize;
2762 else
2763 init_ssize = sgrowsiz;
2764
a108bf71 2765 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2766 vm_map_lock(map);
2767
2768 /* If addr is already mapped, no go */
2769 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
2770 vm_map_unlock(map);
a108bf71 2771 vm_map_entry_release(count);
984263bc
MD
2772 return (KERN_NO_SPACE);
2773 }
2774
2775 /* If we would blow our VMEM resource limit, no go */
2776 if (map->size + init_ssize >
2777 curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
2778 vm_map_unlock(map);
a108bf71 2779 vm_map_entry_release(count);
984263bc
MD
2780 return (KERN_NO_SPACE);
2781 }
2782
2783 /* If we can't accomodate max_ssize in the current mapping,
2784 * no go. However, we need to be aware that subsequent user
2785 * mappings might map into the space we have reserved for
2786 * stack, and currently this space is not protected.
2787 *
2788 * Hopefully we will at least detect this condition
2789 * when we try to grow the stack.
2790 */
2791 if ((prev_entry->next != &map->header) &&
2792 (prev_entry->next->start < addrbos + max_ssize)) {
2793 vm_map_unlock(map);
a108bf71 2794 vm_map_entry_release(count);
984263bc
MD
2795 return (KERN_NO_SPACE);
2796 }
2797
2798 /* We initially map a stack of only init_ssize. We will
2799 * grow as needed later. Since this is to be a grow
2800 * down stack, we map at the top of the range.
2801 *
2802 * Note: we would normally expect prot and max to be
2803 * VM_PROT_ALL, and cow to be 0. Possibly we should
2804 * eliminate these as input parameters, and just
2805 * pass these values here in the insert call.
2806 */
a108bf71
MD
2807 rv = vm_map_insert(map, &count,
2808 NULL, 0, addrbos + max_ssize - init_ssize,
984263bc
MD
2809 addrbos + max_ssize, prot, max, cow);
2810
2811 /* Now set the avail_ssize amount */
2812 if (rv == KERN_SUCCESS){
2813 if (prev_entry != &map->header)
a108bf71 2814 vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize, &count);
984263bc
MD
2815 new_stack_entry = prev_entry->next;
2816 if (new_stack_entry->end != addrbos + max_ssize ||
2817 new_stack_entry->start != addrbos + max_ssize - init_ssize)
2818 panic ("Bad entry start/end for new stack entry");
2819 else
2820 new_stack_entry->avail_ssize = max_ssize - init_ssize;
2821 }
2822
2823 vm_map_unlock(map);
a108bf71 2824 vm_map_entry_release(count);
984263bc
MD
2825 return (rv);
2826}
2827
2828/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
2829 * desired address is already mapped, or if we successfully grow
2830 * the stack. Also returns KERN_SUCCESS if addr is outside the
2831 * stack range (this is strange, but preserves compatibility with
2832 * the grow function in vm_machdep.c).
2833 */
2834int
2835vm_map_growstack (struct proc *p, vm_offset_t addr)
2836{
2837 vm_map_entry_t prev_entry;
2838 vm_map_entry_t stack_entry;
2839 vm_map_entry_t new_stack_entry;
2840 struct vmspace *vm = p->p_vmspace;
2841 vm_map_t map = &vm->vm_map;
2842 vm_offset_t end;
a108bf71
MD
2843 int grow_amount;
2844 int rv = KERN_SUCCESS;
2845 int is_procstack;
2846 int use_read_lock = 1;
2847 int count;
984263bc 2848
a108bf71 2849 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2850Retry:
2851 if (use_read_lock)
2852 vm_map_lock_read(map);
2853 else
2854 vm_map_lock(map);
2855
2856 /* If addr is already in the entry range, no need to grow.*/
2857 if (vm_map_lookup_entry(map, addr, &prev_entry))
2858 goto done;
2859
2860 if ((stack_entry = prev_entry->next) == &map->header)
2861 goto done;
2862 if (prev_entry == &map->header)
2863 end = stack_entry->start - stack_entry->avail_ssize;
2864 else
2865 end = prev_entry->end;
2866
2867 /* This next test mimics the old grow function in vm_machdep.c.
2868 * It really doesn't quite make sense, but we do it anyway
2869 * for compatibility.
2870 *
2871 * If not growable stack, return success. This signals the
2872 * caller to proceed as he would normally with normal vm.
2873 */
2874 if (stack_entry->avail_ssize < 1 ||
2875 addr >= stack_entry->start ||
2876 addr < stack_entry->start - stack_entry->avail_ssize) {
2877 goto done;
2878 }
2879
2880 /* Find the minimum grow amount */
2881 grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
2882 if (grow_amount > stack_entry->avail_ssize) {
2883 rv = KERN_NO_SPACE;
2884 goto done;
2885 }
2886
2887 /* If there is no longer enough space between the entries
2888 * nogo, and adjust the available space. Note: this
2889 * should only happen if the user has mapped into the
2890 * stack area after the stack was created, and is
2891 * probably an error.
2892 *
2893 * This also effectively destroys any guard page the user
2894 * might have intended by limiting the stack size.
2895 */
2896 if (grow_amount > stack_entry->start - end) {
2897 if (use_read_lock && vm_map_lock_upgrade(map)) {
2898 use_read_lock = 0;
2899 goto Retry;
2900 }
2901 use_read_lock = 0;
2902 stack_entry->avail_ssize = stack_entry->start - end;
2903 rv = KERN_NO_SPACE;
2904 goto done;
2905 }
2906
2907 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
2908
2909 /* If this is the main process stack, see if we're over the
2910 * stack limit.
2911 */
2912 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2913 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2914 rv = KERN_NO_SPACE;
2915 goto done;
2916 }
2917
2918 /* Round up the grow amount modulo SGROWSIZ */
2919 grow_amount = roundup (grow_amount, sgrowsiz);
2920 if (grow_amount > stack_entry->avail_ssize) {
2921 grow_amount = stack_entry->avail_ssize;
2922 }
2923 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2924 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2925 grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
2926 ctob(vm->vm_ssize);
2927 }
2928
2929 /* If we would blow our VMEM resource limit, no go */
be77b5f9 2930 if (map->size + grow_amount > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
984263bc
MD
2931 rv = KERN_NO_SPACE;
2932 goto done;
2933 }
2934
2935 if (use_read_lock && vm_map_lock_upgrade(map)) {
2936 use_read_lock = 0;
2937 goto Retry;
2938 }
2939 use_read_lock = 0;
2940
2941 /* Get the preliminary new entry start value */
2942 addr = stack_entry->start - grow_amount;
2943
2944 /* If this puts us into the previous entry, cut back our growth
2945 * to the available space. Also, see the note above.
2946 */
2947 if (addr < end) {
2948 stack_entry->avail_ssize = stack_entry->start - end;
2949 addr = end;
2950 }
2951
a108bf71
MD
2952 rv = vm_map_insert(map, &count,
2953 NULL, 0, addr, stack_entry->start,
984263bc
MD
2954 VM_PROT_ALL,
2955 VM_PROT_ALL,
2956 0);
2957
2958 /* Adjust the available stack space by the amount we grew. */
2959 if (rv == KERN_SUCCESS) {
2960 if (prev_entry != &map->header)
a108bf71 2961 vm_map_clip_end(map, prev_entry, addr, &count);
984263bc
MD
2962 new_stack_entry = prev_entry->next;
2963 if (new_stack_entry->end != stack_entry->start ||
2964 new_stack_entry->start != addr)
2965 panic ("Bad stack grow start/end in new stack entry");
2966 else {
2967 new_stack_entry->avail_ssize = stack_entry->avail_ssize -
2968 (new_stack_entry->end -
2969 new_stack_entry->start);
2970 if (is_procstack)
2971 vm->vm_ssize += btoc(new_stack_entry->end -
2972 new_stack_entry->start);
2973 }
2974 }
2975
2976done:
2977 if (use_read_lock)
2978 vm_map_unlock_read(map);
2979 else
2980 vm_map_unlock(map);
a108bf71 2981 vm_map_entry_release(count);
984263bc
MD
2982 return (rv);
2983}
2984
2985/*
2986 * Unshare the specified VM space for exec. If other processes are
2987 * mapped to it, then create a new one. The new vmspace is null.
2988 */
2989
2990void
29802dbb 2991vmspace_exec(struct proc *p, struct vmspace *vmcopy)
a108bf71 2992{
984263bc
MD
2993 struct vmspace *oldvmspace = p->p_vmspace;
2994 struct vmspace *newvmspace;
2995 vm_map_t map = &p->p_vmspace->vm_map;
2996
29802dbb
MD
2997 /*
2998 * If we are execing a resident vmspace we fork it, otherwise
239b4df9
MD
2999 * we create a new vmspace. Note that exitingcnt and upcalls
3000 * are not copied to the new vmspace.
29802dbb
MD
3001 */
3002 if (vmcopy) {
3003 newvmspace = vmspace_fork(vmcopy);
3004 } else {
3005 newvmspace = vmspace_alloc(map->min_offset, map->max_offset);
3006 bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
239b4df9
MD
3007 (caddr_t)&oldvmspace->vm_endcopy -
3008 (caddr_t)&oldvmspace->vm_startcopy);
29802dbb
MD
3009 }
3010
984263bc
MD
3011 /*
3012 * This code is written like this for prototype purposes. The
3013 * goal is to avoid running down the vmspace here, but let the
3014 * other process's that are still using the vmspace to finally
3015 * run it down. Even though there is little or no chance of blocking
3016 * here, it is a good idea to keep this form for future mods.
3017 */
984263bc
MD
3018 p->p_vmspace = newvmspace;
3019 pmap_pinit2(vmspace_pmap(newvmspace));
3020 if (p == curproc)
3021 pmap_activate(p);
29802dbb 3022 vmspace_free(oldvmspace);
984263bc
MD
3023}
3024
3025/*
3026 * Unshare the specified VM space for forcing COW. This
3027 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
239b4df9
MD
3028 *
3029 * The exitingcnt test is not strictly necessary but has been
3030 * included for code sanity (to make the code a bit more deterministic).
984263bc
MD
3031 */
3032
3033void
a108bf71
MD
3034vmspace_unshare(struct proc *p)
3035{
984263bc
MD
3036 struct vmspace *oldvmspace = p->p_vmspace;
3037 struct vmspace *newvmspace;
3038
239b4df9 3039 if (oldvmspace->vm_refcnt == 1 && oldvmspace->vm_exitingcnt == 0)
984263bc
MD
3040 return;
3041 newvmspace = vmspace_fork(oldvmspace);
984263bc
MD
3042 p->p_vmspace = newvmspace;
3043 pmap_pinit2(vmspace_pmap(newvmspace));
3044 if (p == curproc)
3045 pmap_activate(p);
29802dbb 3046 vmspace_free(oldvmspace);
984263bc 3047}
984263bc
MD
3048
3049/*
3050 * vm_map_lookup:
3051 *
3052 * Finds the VM object, offset, and
3053 * protection for a given virtual address in the
3054 * specified map, assuming a page fault of the
3055 * type specified.
3056 *
3057 * Leaves the map in question locked for read; return
3058 * values are guaranteed until a vm_map_lookup_done
3059 * call is performed. Note that the map argument
3060 * is in/out; the returned map must be used in
3061 * the call to vm_map_lookup_done.
3062 *
3063 * A handle (out_entry) is returned for use in
3064 * vm_map_lookup_done, to make that fast.
3065 *
3066 * If a lookup is requested with "write protection"
3067 * specified, the map may be changed to perform virtual
3068 * copying operations, although the data referenced will
3069 * remain the same.
3070 */
3071int
3072vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3073 vm_offset_t vaddr,
3074 vm_prot_t fault_typea,
3075 vm_map_entry_t *out_entry, /* OUT */
3076 vm_object_t *object, /* OUT */
3077 vm_pindex_t *pindex, /* OUT */
3078 vm_prot_t *out_prot, /* OUT */
3079 boolean_t *wired) /* OUT */
3080{
3081 vm_map_entry_t entry;
3082 vm_map_t map = *var_map;
3083 vm_prot_t prot;
3084 vm_prot_t fault_type = fault_typea;
3085 int use_read_lock = 1;
3086 int rv = KERN_SUCCESS;
3087
3088RetryLookup:
3089 if (use_read_lock)
3090 vm_map_lock_read(map);
3091 else
3092 vm_map_lock(map);
3093
3094 /*
3095 * If the map has an interesting hint, try it before calling full
3096 * blown lookup routine.
3097 */
3098 entry = map->hint;
3099 *out_entry = entry;
3100
3101 if ((entry == &map->header) ||
3102 (vaddr < entry->start) || (vaddr >= entry->end)) {
3103 vm_map_entry_t tmp_entry;
3104
3105 /*
3106 * Entry was either not a valid hint, or the vaddr was not
3107 * contained in the entry, so do a full lookup.
3108 */
3109 if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) {
3110 rv = KERN_INVALID_ADDRESS;
3111 goto done;
3112 }
3113
3114 entry = tmp_entry;
3115 *out_entry = entry;
3116 }
3117
3118 /*
3119 * Handle submaps.
3120 */
3121
3122 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3123 vm_map_t old_map = map;
3124
3125 *var_map = map = entry->object.sub_map;
3126 if (use_read_lock)
3127 vm_map_unlock_read(old_map);
3128 else
3129 vm_map_unlock(old_map);
3130 use_read_lock = 1;
3131 goto RetryLookup;
3132 }
3133
3134 /*
3135 * Check whether this task is allowed to have this page.
3136 * Note the special case for MAP_ENTRY_COW
3137 * pages with an override. This is to implement a forced
3138 * COW for debuggers.
3139 */
3140
3141 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3142 prot = entry->max_protection;
3143 else
3144 prot = entry->protection;
3145
3146 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3147 if ((fault_type & prot) != fault_type) {
3148 rv = KERN_PROTECTION_FAILURE;
3149 goto done;
3150 }
3151
3152 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3153 (entry->eflags & MAP_ENTRY_COW) &&
3154 (fault_type & VM_PROT_WRITE) &&
3155 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3156 rv = KERN_PROTECTION_FAILURE;
3157 goto done;
3158 }
3159
3160 /*
3161 * If this page is not pageable, we have to get it for all possible
3162 * accesses.
3163 */
3164
3165 *wired = (entry->wired_count != 0);
3166 if (*wired)
3167 prot = fault_type = entry->protection;
3168
3169 /*
3170 * If the entry was copy-on-write, we either ...
3171 */
3172
3173 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3174 /*
3175 * If we want to write the page, we may as well handle that
3176 * now since we've got the map locked.
3177 *
3178 * If we don't need to write the page, we just demote the
3179 * permissions allowed.
3180 */
3181
3182 if (fault_type & VM_PROT_WRITE) {
3183 /*
3184 * Make a new object, and place it in the object
3185 * chain. Note that no new references have appeared
3186 * -- one just moved from the map to the new
3187 * object.
3188 */
3189
3190 if (use_read_lock && vm_map_lock_upgrade(map)) {
3191 use_read_lock = 0;
3192 goto RetryLookup;
3193 }
3194 use_read_lock = 0;
3195
3196 vm_object_shadow(
3197 &entry->object.vm_object,
3198 &entry->offset,
3199 atop(entry->end - entry->start));
3200
3201 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3202 } else {
3203 /*
3204 * We're attempting to read a copy-on-write page --
3205 * don't allow writes.
3206 */
3207
3208 prot &= ~VM_PROT_WRITE;
3209 }
3210 }
3211
3212 /*
3213 * Create an object if necessary.
3214 */
3215 if (entry->object.vm_object == NULL &&
3216 !map->system_map) {
3217 if (use_read_lock && vm_map_lock_upgrade(map)) {
3218 use_read_lock = 0;
3219 goto RetryLookup;
3220 }
3221 use_read_lock = 0;
3222 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3223 atop(entry->end - entry->start));
3224 entry->offset = 0;
3225 }
3226
3227 /*
3228 * Return the object/offset from this entry. If the entry was
3229 * copy-on-write or empty, it has been fixed up.
3230 */
3231
3232 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3233 *object = entry->object.vm_object;
3234
3235 /*
3236 * Return whether this is the only map sharing this data. On
3237 * success we return with a read lock held on the map. On failure
3238 * we return with the map unlocked.
3239 */
3240 *out_prot = prot;
3241done:
3242 if (rv == KERN_SUCCESS) {
3243 if (use_read_lock == 0)
3244 vm_map_lock_downgrade(map);
3245 } else if (use_read_lock) {
3246 vm_map_unlock_read(map);
3247 } else {
3248 vm_map_unlock(map);
3249 }
3250 return (rv);
3251}
3252
3253/*
3254 * vm_map_lookup_done:
3255 *
3256 * Releases locks acquired by a vm_map_lookup
3257 * (according to the handle returned by that lookup).
3258 */
3259
3260void
a108bf71 3261vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry, int count)
984263bc
MD
3262{
3263 /*
3264 * Unlock the main-level map
3265 */
984263bc 3266 vm_map_unlock_read(map);
a108bf71
MD
3267 if (count)
3268 vm_map_entry_release(count);
984263bc
MD
3269}
3270
cfbd9dd0
MD
3271#ifdef ENABLE_VFS_IOOPT
3272
984263bc
MD
3273/*
3274 * Implement uiomove with VM operations. This handles (and collateral changes)
3275 * support every combination of source object modification, and COW type
3276 * operations.
06ecca5a
MD
3277 *
3278 * XXX this is extremely dangerous, enabling this option is NOT recommended.
984263bc
MD
3279 */
3280int
57e43348
MD
3281vm_uiomove(vm_map_t mapa, vm_object_t srcobject, off_t cp, int cnta,
3282 vm_offset_t uaddra, int *npages)
984263bc
MD
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;
06ecca5a 3295 int s;
984263bc
MD
3296
3297 if (npages)
3298 *npages = 0;
3299
3300 cnt = cnta;
3301 uaddr = uaddra;
3302
3303 while (cnt > 0) {
3304 map = mapa;
3305
a108bf71
MD
3306 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
3307
984263bc
MD
3308 if ((vm_map_lookup(&map, uaddr,
3309 VM_PROT_READ, &entry, &first_object,
3310 &first_pindex, &prot, &wired)) != KERN_SUCCESS) {
3311 return EFAULT;
3312 }
3313
a108bf71 3314 vm_map_clip_start(map, entry, uaddr, &count);
984263bc
MD
3315
3316 tcnt = cnt;
3317 tend = uaddr + tcnt;
3318 if (tend > entry->end) {
3319 tcnt = entry->end - uaddr;
3320 tend = entry->end;
3321 }
3322
a108bf71 3323 vm_map_clip_end(map, entry, tend, &count);
984263bc
MD
3324
3325 start = entry->start;
3326 end = entry->end;
3327
3328 osize = atop(tcnt);
3329
3330 oindex = OFF_TO_IDX(cp);
3331 if (npages) {
3332 vm_pindex_t idx;
06ecca5a
MD
3333
3334 /*
3335 * spl protection is needed to avoid a race between
3336 * the lookup and an interrupt/unbusy/free occuring
3337 * prior to our busy check.
3338 */
3339 s = splvm();
984263bc
MD
3340 for (idx = 0; idx < osize; idx++) {
3341 vm_page_t m;
3342 if ((m = vm_page_lookup(srcobject, oindex + idx)) == NULL) {
06ecca5a 3343 splx(s);
a108bf71 3344 vm_map_lookup_done(map, entry, count);
984263bc
MD
3345 return 0;
3346 }
3347 /*
3348 * disallow busy or invalid pages, but allow
3349 * m->busy pages if they are entirely valid.
3350 */
3351 if ((m->flags & PG_BUSY) ||
3352 ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL)) {
06ecca5a 3353 splx(s);
a108bf71 3354 vm_map_lookup_done(map, entry, count);
984263bc
MD
3355 return 0;
3356 }
3357 }
06ecca5a 3358 splx(s);
984263bc
MD
3359 }
3360
3361/*
3362 * If we are changing an existing map entry, just redirect
3363 * the object, and change mappings.
3364 */
3365 if ((first_object->type == OBJT_VNODE) &&
3366 ((oldobject = entry->object.vm_object) == first_object)) {
3367
3368 if ((entry->offset != cp) || (oldobject != srcobject)) {
3369 /*
3370 * Remove old window into the file
3371 */
3372 pmap_remove (map->pmap, uaddr, tend);
3373
3374 /*
3375 * Force copy on write for mmaped regions
3376 */
3377 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3378
3379 /*
3380 * Point the object appropriately
3381 */
3382 if (oldobject != srcobject) {
3383
3384 /*
3385 * Set the object optimization hint flag
3386 */
3387 vm_object_set_flag(srcobject, OBJ_OPT);
3388 vm_object_reference(srcobject);
3389 entry->object.vm_object = srcobject;
3390
3391 if (oldobject) {
3392 vm_object_deallocate(oldobject);
3393 }
3394 }
3395
3396 entry->offset = cp;
3397 map->timestamp++;
3398 } else {
3399 pmap_remove (map->pmap, uaddr, tend);
3400 }
3401
3402 } else if ((first_object->ref_count == 1) &&
3403 (first_object->size == osize) &&
3404 ((first_object->type == OBJT_DEFAULT) ||
3405 (first_object->type == OBJT_SWAP)) ) {
3406
3407 oldobject = first_object->backing_object;
3408
3409 if ((first_object->backing_object_offset != cp) ||
3410 (oldobject != srcobject)) {
3411 /*
3412 * Remove old window into the file
3413 */
3414 pmap_remove (map->pmap, uaddr, tend);
3415
3416 /*
3417 * Remove unneeded old pages
3418 */
3419 vm_object_page_remove(first_object, 0, 0, 0);
3420
3421 /*
3422 * Invalidate swap space
3423 */
3424 if (first_object->type == OBJT_SWAP) {
3425 swap_pager_freespace(first_object,
3426 0,
3427 first_object->size);
3428 }
3429
3430 /*
3431 * Force copy on write for mmaped regions
3432 */
3433 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3434
3435 /*
3436 * Point the object appropriately
3437 */
3438 if (oldobject != srcobject) {
3439
3440 /*
3441 * Set the object optimization hint flag
3442 */
3443 vm_object_set_flag(srcobject, OBJ_OPT);
3444 vm_object_reference(srcobject);
3445
3446 if (oldobject) {
3447 LIST_REMOVE(
3448 first_object, shadow_list);
3449 oldobject->shadow_count--;
3450 /* XXX bump generation? */
3451 vm_object_deallocate(oldobject);
3452 }
3453
3454 LIST_INSERT_HEAD(&srcobject->shadow_head,
3455 first_object, shadow_list);
3456 srcobject->shadow_count++;
3457 /* XXX bump generation? */
3458
3459 first_object->backing_object = srcobject;
3460 }
3461 first_object->backing_object_offset = cp;
3462 map->timestamp++;
3463 } else {
3464 pmap_remove (map->pmap, uaddr, tend);
3465 }
3466/*
3467 * Otherwise, we have to do a logical mmap.
3468 */
3469 } else {
3470
3471 vm_object_set_flag(srcobject, OBJ_OPT);
3472 vm_object_reference(srcobject);
3473
3474 pmap_remove (map->pmap, uaddr, tend);
3475
3476 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3477 vm_map_lock_upgrade(map);
3478
3479 if (entry == &map->header) {
3480 map->first_free = &map->header;
3481 } else if (map->first_free->start >= start) {
3482 map->first_free = entry->prev;
3483 }
3484
3485 SAVE_HINT(map, entry->prev);
a108bf71 3486 vm_map_entry_delete(map, entry, &count);
984263bc
MD
3487
3488 object = srcobject;
3489 ooffset = cp;
3490
a108bf71
MD
3491 rv = vm_map_insert(map, &count,
3492 object, ooffset, start, tend,
984263bc
MD
3493 VM_PROT_ALL, VM_PROT_ALL, MAP_COPY_ON_WRITE);
3494
3495 if (rv != KERN_SUCCESS)
3496 panic("vm_uiomove: could not insert new entry: %d", rv);
3497 }
3498
3499/*
3500 * Map the window directly, if it is already in memory
3501 */
083a7402 3502 pmap_object_init_pt(map->pmap, uaddr, entry->protection,
984263bc
MD
3503 srcobject, oindex, tcnt, 0);
3504
3505 map->timestamp++;
3506 vm_map_unlock(map);
a108bf71 3507 vm_map_entry_release(count);
984263bc
MD
3508
3509 cnt -= tcnt;
3510 uaddr += tcnt;
3511 cp += tcnt;
3512 if (npages)
3513 *npages += osize;
3514 }
3515 return 0;
3516}
3517
cfbd9dd0
MD
3518#endif
3519
984263bc
MD
3520/*
3521 * Performs the copy_on_write operations necessary to allow the virtual copies
3522 * into user space to work. This has to be called for write(2) system calls
3523 * from other processes, file unlinking, and file size shrinkage.
3524 */
3525void
57e43348 3526vm_freeze_copyopts(vm_object_t object, vm_pindex_t froma, vm_pindex_t toa)
984263bc
MD
3527{
3528 int rv;
3529 vm_object_t robject;
3530 vm_pindex_t idx;
3531
3532 if ((object == NULL) ||
3533 ((object->flags & OBJ_OPT) == 0))
3534 return;
3535
3536 if (object->shadow_count > object->ref_count)
3537 panic("vm_freeze_copyopts: sc > rc");
3538
06ecca5a 3539 while ((robject = LIST_FIRST(&object->shadow_head)) != NULL) {
984263bc
MD
3540 vm_pindex_t bo_pindex;
3541 vm_page_t m_in, m_out;
3542
3543 bo_pindex = OFF_TO_IDX(robject->backing_object_offset);
3544
3545 vm_object_reference(robject);
3546
3547 vm_object_pip_wait(robject, "objfrz");
3548
3549 if (robject->ref_count == 1) {
3550 vm_object_deallocate(robject);
3551 continue;
3552 }
3553
3554 vm_object_pip_add(robject, 1);
3555
3556 for (idx = 0; idx < robject->size; idx++) {
3557
3558 m_out = vm_page_grab(robject, idx,
dc1fd4b3 3559 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
984263bc
MD
3560
3561 if (m_out->valid == 0) {
3562 m_in = vm_page_grab(object, bo_pindex + idx,
dc1fd4b3 3563 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
984263bc
MD
3564 if (m_in->valid == 0) {
3565 rv = vm_pager_get_pages(object, &m_in, 1, 0);
3566 if (rv != VM_PAGER_OK) {
3567 printf("vm_freeze_copyopts: cannot read page from file: %lx\n", (long)m_in->pindex);
3568 continue;
3569 }
3570 vm_page_deactivate(m_in);
3571 }
3572
3573 vm_page_protect(m_in, VM_PROT_NONE);
3574 pmap_copy_page(VM_PAGE_TO_PHYS(m_in), VM_PAGE_TO_PHYS(m_out));
3575 m_out->valid = m_in->valid;
3576 vm_page_dirty(m_out);
3577 vm_page_activate(m_out);
3578 vm_page_wakeup(m_in);
3579 }
3580 vm_page_wakeup(m_out);
3581 }
3582
3583 object->shadow_count--;
3584 object->ref_count--;
3585 LIST_REMOVE(robject, shadow_list);
3586 robject->backing_object = NULL;
3587 robject->backing_object_offset = 0;
3588
3589 vm_object_pip_wakeup(robject);
3590 vm_object_deallocate(robject);
3591 }
3592
3593 vm_object_clear_flag(object, OBJ_OPT);
3594}
3595
3596#include "opt_ddb.h"
3597#ifdef DDB
3598#include <sys/kernel.h>
3599
3600#include <ddb/ddb.h>
3601
3602/*
3603 * vm_map_print: [ debug ]
3604 */
3605DB_SHOW_COMMAND(map, vm_map_print)
3606{
3607 static int nlines;
3608 /* XXX convert args. */
3609 vm_map_t map = (vm_map_t)addr;
3610 boolean_t full = have_addr;
3611
3612 vm_map_entry_t entry;
3613
3614 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3615 (void *)map,
3616 (void *)map->pmap, map->nentries, map->timestamp);
3617 nlines++;
3618
3619 if (!full && db_indent)
3620 return;
3621
3622 db_indent += 2;
3623 for (entry = map->header.next; entry != &map->header;
3624 entry = entry->next) {
3625 db_iprintf("map entry %p: start=%p, end=%p\n",
3626 (void *)entry, (void *)entry->start, (void *)entry->end);
3627 nlines++;
3628 {
3629 static char *inheritance_name[4] =
3630 {"share", "copy", "none", "donate_copy"};
3631
3632 db_iprintf(" prot=%x/%x/%s",
3633 entry->protection,
3634 entry->max_protection,
3635 inheritance_name[(int)(unsigned char)entry->inheritance]);
3636 if (entry->wired_count != 0)
3637 db_printf(", wired");
3638 }
3639 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3640 /* XXX no %qd in kernel. Truncate entry->offset. */
3641 db_printf(", share=%p, offset=0x%lx\n",
3642 (void *)entry->object.sub_map,
3643 (long)entry->offset);
3644 nlines++;
3645 if ((entry->prev == &map->header) ||
3646 (entry->prev->object.sub_map !=
3647 entry->object.sub_map)) {
3648 db_indent += 2;
3649 vm_map_print((db_expr_t)(intptr_t)
3650 entry->object.sub_map,
3651 full, 0, (char *)0);
3652 db_indent -= 2;
3653 }
3654 } else {
3655 /* XXX no %qd in kernel. Truncate entry->offset. */
3656 db_printf(", object=%p, offset=0x%lx",
3657 (void *)entry->object.vm_object,
3658 (long)entry->offset);
3659 if (entry->eflags & MAP_ENTRY_COW)
3660 db_printf(", copy (%s)",
3661 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3662 db_printf("\n");
3663 nlines++;
3664
3665 if ((entry->prev == &map->header) ||
3666 (entry->prev->object.vm_object !=
3667 entry->object.vm_object)) {
3668 db_indent += 2;
3669 vm_object_print((db_expr_t)(intptr_t)
3670 entry->object.vm_object,
3671 full, 0, (char *)0);
3672 nlines += 4;
3673 db_indent -= 2;
3674 }
3675 }
3676 }
3677 db_indent -= 2;
3678 if (db_indent == 0)
3679 nlines = 0;
3680}
3681
3682
3683DB_SHOW_COMMAND(procvm, procvm)
3684{
3685 struct proc *p;
3686
3687 if (have_addr) {
3688 p = (struct proc *) addr;
3689 } else {
3690 p = curproc;
3691 }
3692
3693 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3694 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3695 (void *)vmspace_pmap(p->p_vmspace));
3696
3697 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3698}
3699
3700#endif /* DDB */