Add some casts to bring us up to WARNS2. This is almost WARNS6 except for
[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 $
53025830 65 * $DragonFly: src/sys/vm/vm_map.c,v 1.50 2006/09/13 22:25:00 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>
686dbf64 81#include <sys/tree.h>
984263bc
MD
82
83#include <vm/vm.h>
84#include <vm/vm_param.h>
984263bc
MD
85#include <vm/pmap.h>
86#include <vm/vm_map.h>
87#include <vm/vm_page.h>
88#include <vm/vm_object.h>
89#include <vm/vm_pager.h>
90#include <vm/vm_kern.h>
91#include <vm/vm_extern.h>
92#include <vm/swap_pager.h>
93#include <vm/vm_zone.h>
94
a108bf71
MD
95#include <sys/thread2.h>
96
984263bc
MD
97/*
98 * Virtual memory maps provide for the mapping, protection,
99 * and sharing of virtual memory objects. In addition,
100 * this module provides for an efficient virtual copy of
101 * memory from one map to another.
102 *
103 * Synchronization is required prior to most operations.
104 *
105 * Maps consist of an ordered doubly-linked list of simple
106 * entries; a single hint is used to speed up lookups.
107 *
108 * Since portions of maps are specified by start/end addresses,
109 * which may not align with existing map entries, all
110 * routines merely "clip" entries to these start/end values.
111 * [That is, an entry is split into two, bordering at a
112 * start or end value.] Note that these clippings may not
113 * always be necessary (as the two resulting entries are then
114 * not changed); however, the clipping is done for convenience.
115 *
116 * As mentioned above, virtual copy operations are performed
117 * by copying VM object references from one map to
118 * another, and then marking both regions as copy-on-write.
119 */
120
121/*
122 * vm_map_startup:
123 *
124 * Initialize the vm_map module. Must be called before
125 * any other vm_map routines.
126 *
127 * Map and entry structures are allocated from the general
128 * purpose memory pool with some exceptions:
129 *
130 * - The kernel map and kmem submap are allocated statically.
131 * - Kernel map entries are allocated out of a static pool.
132 *
133 * These restrictions are necessary since malloc() uses the
134 * maps and requires map entries.
135 */
136
c4ae567f
MD
137#define VMEPERCPU 2
138
a108bf71
MD
139static struct vm_zone mapentzone_store, mapzone_store;
140static vm_zone_t mapentzone, mapzone, vmspace_zone;
141static struct vm_object mapentobj, mapobj;
984263bc
MD
142
143static struct vm_map_entry map_entry_init[MAX_MAPENT];
c4ae567f 144static struct vm_map_entry cpu_map_entry_init[MAXCPU][VMEPERCPU];
984263bc
MD
145static struct vm_map map_init[MAX_KMAP];
146
53025830 147static void vm_map_entry_shadow(vm_map_entry_t entry);
a108bf71
MD
148static vm_map_entry_t vm_map_entry_create(vm_map_t map, int *);
149static void vm_map_entry_dispose (vm_map_t map, vm_map_entry_t entry, int *);
150static void _vm_map_clip_end (vm_map_t, vm_map_entry_t, vm_offset_t, int *);
151static void _vm_map_clip_start (vm_map_t, vm_map_entry_t, vm_offset_t, int *);
152static void vm_map_entry_delete (vm_map_t, vm_map_entry_t, int *);
1388df65
RG
153static void vm_map_entry_unwire (vm_map_t, vm_map_entry_t);
154static void vm_map_copy_entry (vm_map_t, vm_map_t, vm_map_entry_t,
155 vm_map_entry_t);
156static void vm_map_split (vm_map_entry_t);
a108bf71 157static 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
158
159void
57e43348 160vm_map_startup(void)
984263bc
MD
161{
162 mapzone = &mapzone_store;
163 zbootinit(mapzone, "MAP", sizeof (struct vm_map),
164 map_init, MAX_KMAP);
984263bc
MD
165 mapentzone = &mapentzone_store;
166 zbootinit(mapentzone, "MAP ENTRY", sizeof (struct vm_map_entry),
167 map_entry_init, MAX_MAPENT);
168}
169
686dbf64
MD
170/*
171 * Red black tree functions
172 */
173static int rb_vm_map_compare(vm_map_entry_t a, vm_map_entry_t b);
174RB_GENERATE(vm_map_rb_tree, vm_map_entry, rb_entry, rb_vm_map_compare);
175
176/* a->start is address, and the only field has to be initialized */
177static int
178rb_vm_map_compare(vm_map_entry_t a, vm_map_entry_t b)
179{
180 if (a->start < b->start)
181 return(-1);
182 else if (a->start > b->start)
183 return(1);
184 return(0);
185}
186
984263bc
MD
187/*
188 * Allocate a vmspace structure, including a vm_map and pmap,
189 * and initialize those structures. The refcnt is set to 1.
190 * The remaining fields must be initialized by the caller.
191 */
192struct vmspace *
57e43348 193vmspace_alloc(vm_offset_t min, vm_offset_t max)
984263bc
MD
194{
195 struct vmspace *vm;
196
197 vm = zalloc(vmspace_zone);
198 vm_map_init(&vm->vm_map, min, max);
199 pmap_pinit(vmspace_pmap(vm));
200 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
201 vm->vm_refcnt = 1;
202 vm->vm_shm = NULL;
203 vm->vm_exitingcnt = 0;
204 return (vm);
205}
206
207void
a108bf71
MD
208vm_init2(void)
209{
c4ae567f
MD
210 zinitna(mapentzone, &mapentobj, NULL, 0, 0,
211 ZONE_USE_RESERVE | ZONE_SPECIAL, 1);
a108bf71 212 zinitna(mapzone, &mapobj, NULL, 0, 0, 0, 1);
984263bc
MD
213 vmspace_zone = zinit("VMSPACE", sizeof (struct vmspace), 0, 0, 3);
214 pmap_init2();
215 vm_object_init2();
216}
217
218static __inline void
219vmspace_dofree(struct vmspace *vm)
220{
a108bf71
MD
221 int count;
222
fef0fdf2
MD
223 /*
224 * Make sure any SysV shm is freed, it might not have in
225 * exit1()
226 */
227 shmexit(vm);
228
a722be49
MD
229 KKASSERT(vm->vm_upcalls == NULL);
230
984263bc
MD
231 /*
232 * Lock the map, to wait out all other references to it.
233 * Delete all of the mappings and pages they hold, then call
234 * the pmap module to reclaim anything left.
235 */
a108bf71 236 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc 237 vm_map_lock(&vm->vm_map);
a108bf71
MD
238 vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
239 vm->vm_map.max_offset, &count);
984263bc 240 vm_map_unlock(&vm->vm_map);
a108bf71 241 vm_map_entry_release(count);
984263bc
MD
242
243 pmap_release(vmspace_pmap(vm));
244 zfree(vmspace_zone, vm);
245}
246
247void
248vmspace_free(struct vmspace *vm)
249{
250 if (vm->vm_refcnt == 0)
251 panic("vmspace_free: attempt to free already freed vmspace");
252
253 if (--vm->vm_refcnt == 0 && vm->vm_exitingcnt == 0)
254 vmspace_dofree(vm);
255}
256
257void
258vmspace_exitfree(struct proc *p)
259{
260 struct vmspace *vm;
261
262 vm = p->p_vmspace;
263 p->p_vmspace = NULL;
264
265 /*
266 * cleanup by parent process wait()ing on exiting child. vm_refcnt
267 * may not be 0 (e.g. fork() and child exits without exec()ing).
268 * exitingcnt may increment above 0 and drop back down to zero
269 * several times while vm_refcnt is held non-zero. vm_refcnt
270 * may also increment above 0 and drop back down to zero several
271 * times while vm_exitingcnt is held non-zero.
272 *
273 * The last wait on the exiting child's vmspace will clean up
274 * the remainder of the vmspace.
275 */
276 if (--vm->vm_exitingcnt == 0 && vm->vm_refcnt == 0)
277 vmspace_dofree(vm);
278}
279
280/*
281 * vmspace_swap_count() - count the approximate swap useage in pages for a
282 * vmspace.
283 *
284 * Swap useage is determined by taking the proportional swap used by
285 * VM objects backing the VM map. To make up for fractional losses,
286 * if the VM object has any swap use at all the associated map entries
287 * count for at least 1 swap page.
288 */
289int
290vmspace_swap_count(struct vmspace *vmspace)
291{
292 vm_map_t map = &vmspace->vm_map;
293 vm_map_entry_t cur;
1b874851 294 vm_object_t object;
984263bc 295 int count = 0;
1b874851 296 int n;
984263bc
MD
297
298 for (cur = map->header.next; cur != &map->header; cur = cur->next) {
1b874851
MD
299 switch(cur->maptype) {
300 case VM_MAPTYPE_NORMAL:
301 case VM_MAPTYPE_VPAGETABLE:
302 if ((object = cur->object.vm_object) == NULL)
303 break;
304 if (object->type != OBJT_SWAP)
305 break;
306 n = (cur->end - cur->start) / PAGE_SIZE;
984263bc
MD
307 if (object->un_pager.swp.swp_bcount) {
308 count += object->un_pager.swp.swp_bcount *
309 SWAP_META_PAGES * n / object->size + 1;
310 }
1b874851
MD
311 break;
312 default:
313 break;
984263bc
MD
314 }
315 }
316 return(count);
317}
318
319
320/*
321 * vm_map_create:
322 *
323 * Creates and returns a new empty VM map with
324 * the given physical map structure, and having
325 * the given lower and upper address bounds.
326 */
327vm_map_t
a108bf71 328vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
984263bc
MD
329{
330 vm_map_t result;
331
332 result = zalloc(mapzone);
333 vm_map_init(result, min, max);
334 result->pmap = pmap;
335 return (result);
336}
337
338/*
339 * Initialize an existing vm_map structure
340 * such as that in the vmspace structure.
341 * The pmap is set elsewhere.
342 */
343void
a108bf71 344vm_map_init(struct vm_map *map, vm_offset_t min, vm_offset_t max)
984263bc
MD
345{
346 map->header.next = map->header.prev = &map->header;
686dbf64 347 RB_INIT(&map->rb_root);
984263bc
MD
348 map->nentries = 0;
349 map->size = 0;
350 map->system_map = 0;
351 map->infork = 0;
352 map->min_offset = min;
353 map->max_offset = max;
354 map->first_free = &map->header;
355 map->hint = &map->header;
356 map->timestamp = 0;
ab6f251b 357 lockinit(&map->lock, "thrd_sleep", 0, 0);
984263bc
MD
358}
359
53025830
MD
360/*
361 * Shadow the vm_map_entry's object. This typically needs to be done when
362 * a write fault is taken on an entry which had previously been cloned by
363 * fork(). The shared object (which might be NULL) must become private so
364 * we add a shadow layer above it.
365 *
366 * Object allocation for anonymous mappings is defered as long as possible.
367 * When creating a shadow, however, the underlying object must be instantiated
368 * so it can be shared.
369 *
370 * If the map segment is governed by a virtual page table then it is
371 * possible to address offsets beyond the mapped area. Just allocate
372 * a maximally sized object for this case.
373 */
374static
375void
376vm_map_entry_shadow(vm_map_entry_t entry)
377{
378 if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
379 vm_object_shadow(&entry->object.vm_object, &entry->offset,
380 0x7FFFFFFF); /* XXX */
381 } else {
382 vm_object_shadow(&entry->object.vm_object, &entry->offset,
383 atop(entry->end - entry->start));
384 }
385 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
386}
387
388/*
389 * Allocate an object for a vm_map_entry.
390 *
391 * Object allocation for anonymous mappings is defered as long as possible.
392 * This function is called when we can defer no longer, generally when a map
393 * entry might be split or forked or takes a page fault.
394 *
395 * If the map segment is governed by a virtual page table then it is
396 * possible to address offsets beyond the mapped area. Just allocate
397 * a maximally sized object for this case.
398 */
399void
400vm_map_entry_allocate_object(vm_map_entry_t entry)
401{
402 vm_object_t obj;
403
404 if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
405 obj = vm_object_allocate(OBJT_DEFAULT, 0x7FFFFFFF); /* XXX */
406 } else {
407 obj = vm_object_allocate(OBJT_DEFAULT,
408 atop(entry->end - entry->start));
409 }
410 entry->object.vm_object = obj;
411 entry->offset = 0;
412}
413
41a01a4d 414/*
c4ae567f 415 * vm_map_entry_reserve_cpu_init:
41a01a4d
MD
416 *
417 * Set an initial negative count so the first attempt to reserve
c4ae567f
MD
418 * space preloads a bunch of vm_map_entry's for this cpu. Also
419 * pre-allocate 2 vm_map_entries which will be needed by zalloc() to
420 * map a new page for vm_map_entry structures. SMP systems are
421 * particularly sensitive.
422 *
423 * This routine is called in early boot so we cannot just call
41a01a4d
MD
424 * vm_map_entry_reserve().
425 *
c4ae567f
MD
426 * May be called for a gd other then mycpu, but may only be called
427 * during early boot.
41a01a4d
MD
428 */
429void
430vm_map_entry_reserve_cpu_init(globaldata_t gd)
431{
c4ae567f
MD
432 vm_map_entry_t entry;
433 int i;
434
41a01a4d 435 gd->gd_vme_avail -= MAP_RESERVE_COUNT * 2;
c4ae567f
MD
436 entry = &cpu_map_entry_init[gd->gd_cpuid][0];
437 for (i = 0; i < VMEPERCPU; ++i, ++entry) {
438 entry->next = gd->gd_vme_base;
439 gd->gd_vme_base = entry;
440 }
41a01a4d
MD
441}
442
a108bf71
MD
443/*
444 * vm_map_entry_reserve:
445 *
e1359933
MD
446 * Reserves vm_map_entry structures so code later on can manipulate
447 * map_entry structures within a locked map without blocking trying
448 * to allocate a new vm_map_entry.
a108bf71
MD
449 */
450int
451vm_map_entry_reserve(int count)
452{
453 struct globaldata *gd = mycpu;
454 vm_map_entry_t entry;
455
456 crit_enter();
a108bf71
MD
457
458 /*
459 * Make sure we have enough structures in gd_vme_base to handle
460 * the reservation request.
461 */
ac13eccd 462 while (gd->gd_vme_avail < count) {
a108bf71
MD
463 entry = zalloc(mapentzone);
464 entry->next = gd->gd_vme_base;
465 gd->gd_vme_base = entry;
466 ++gd->gd_vme_avail;
467 }
ac13eccd 468 gd->gd_vme_avail -= count;
a108bf71
MD
469 crit_exit();
470 return(count);
471}
472
473/*
474 * vm_map_entry_release:
475 *
476 * Releases previously reserved vm_map_entry structures that were not
477 * used. If we have too much junk in our per-cpu cache clean some of
478 * it out.
479 */
480void
481vm_map_entry_release(int count)
482{
483 struct globaldata *gd = mycpu;
484 vm_map_entry_t entry;
485
486 crit_enter();
487 gd->gd_vme_avail += count;
488 while (gd->gd_vme_avail > MAP_RESERVE_SLOP) {
489 entry = gd->gd_vme_base;
490 KKASSERT(entry != NULL);
491 gd->gd_vme_base = entry->next;
492 --gd->gd_vme_avail;
493 crit_exit();
494 zfree(mapentzone, entry);
495 crit_enter();
496 }
497 crit_exit();
498}
499
500/*
501 * vm_map_entry_kreserve:
502 *
c4ae567f
MD
503 * Reserve map entry structures for use in kernel_map itself. These
504 * entries have *ALREADY* been reserved on a per-cpu basis when the map
505 * was inited. This function is used by zalloc() to avoid a recursion
506 * when zalloc() itself needs to allocate additional kernel memory.
507 *
508 * This function works like the normal reserve but does not load the
509 * vm_map_entry cache (because that would result in an infinite
510 * recursion). Note that gd_vme_avail may go negative. This is expected.
a108bf71 511 *
c4ae567f
MD
512 * Any caller of this function must be sure to renormalize after
513 * potentially eating entries to ensure that the reserve supply
514 * remains intact.
a108bf71
MD
515 */
516int
517vm_map_entry_kreserve(int count)
518{
519 struct globaldata *gd = mycpu;
520
521 crit_enter();
c4ae567f 522 gd->gd_vme_avail -= count;
a108bf71 523 crit_exit();
c4ae567f 524 KASSERT(gd->gd_vme_base != NULL, ("no reserved entries left, gd_vme_avail = %d\n", gd->gd_vme_avail));
a108bf71
MD
525 return(count);
526}
527
528/*
529 * vm_map_entry_krelease:
530 *
c4ae567f
MD
531 * Release previously reserved map entries for kernel_map. We do not
532 * attempt to clean up like the normal release function as this would
533 * cause an unnecessary (but probably not fatal) deep procedure call.
a108bf71
MD
534 */
535void
536vm_map_entry_krelease(int count)
537{
538 struct globaldata *gd = mycpu;
539
540 crit_enter();
c4ae567f 541 gd->gd_vme_avail += count;
a108bf71
MD
542 crit_exit();
543}
544
984263bc 545/*
8a8d5d85 546 * vm_map_entry_create: [ internal use only ]
984263bc 547 *
8a8d5d85 548 * Allocates a VM map entry for insertion. No entry fields are filled
a108bf71
MD
549 * in.
550 *
551 * This routine may be called from an interrupt thread but not a FAST
552 * interrupt. This routine may recurse the map lock.
984263bc 553 */
8a8d5d85 554static vm_map_entry_t
a108bf71 555vm_map_entry_create(vm_map_t map, int *countp)
984263bc 556{
a108bf71
MD
557 struct globaldata *gd = mycpu;
558 vm_map_entry_t entry;
8a8d5d85 559
a108bf71
MD
560 KKASSERT(*countp > 0);
561 --*countp;
562 crit_enter();
563 entry = gd->gd_vme_base;
564 KASSERT(entry != NULL, ("gd_vme_base NULL! count %d", *countp));
565 gd->gd_vme_base = entry->next;
566 crit_exit();
567 return(entry);
984263bc
MD
568}
569
570/*
8a8d5d85 571 * vm_map_entry_dispose: [ internal use only ]
984263bc 572 *
8a8d5d85
MD
573 * Dispose of a vm_map_entry that is no longer being referenced. This
574 * function may be called from an interrupt.
984263bc 575 */
8a8d5d85 576static void
a108bf71 577vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc 578{
a108bf71
MD
579 struct globaldata *gd = mycpu;
580
686dbf64
MD
581 KKASSERT(map->hint != entry);
582 KKASSERT(map->first_free != entry);
583
a108bf71
MD
584 ++*countp;
585 crit_enter();
586 entry->next = gd->gd_vme_base;
587 gd->gd_vme_base = entry;
588 crit_exit();
984263bc
MD
589}
590
8a8d5d85 591
984263bc
MD
592/*
593 * vm_map_entry_{un,}link:
594 *
595 * Insert/remove entries from maps.
596 */
597static __inline void
598vm_map_entry_link(vm_map_t map,
599 vm_map_entry_t after_where,
600 vm_map_entry_t entry)
601{
602 map->nentries++;
603 entry->prev = after_where;
604 entry->next = after_where->next;
605 entry->next->prev = entry;
606 after_where->next = entry;
0cd275af
MD
607 if (vm_map_rb_tree_RB_INSERT(&map->rb_root, entry))
608 panic("vm_map_entry_link: dup addr map %p ent %p", map, entry);
984263bc
MD
609}
610
611static __inline void
612vm_map_entry_unlink(vm_map_t map,
613 vm_map_entry_t entry)
614{
615 vm_map_entry_t prev;
616 vm_map_entry_t next;
617
618 if (entry->eflags & MAP_ENTRY_IN_TRANSITION)
619 panic("vm_map_entry_unlink: attempt to mess with locked entry! %p", entry);
620 prev = entry->prev;
621 next = entry->next;
622 next->prev = prev;
623 prev->next = next;
686dbf64 624 vm_map_rb_tree_RB_REMOVE(&map->rb_root, entry);
984263bc
MD
625 map->nentries--;
626}
627
984263bc
MD
628/*
629 * vm_map_lookup_entry: [ internal use only ]
630 *
631 * Finds the map entry containing (or
632 * immediately preceding) the specified address
633 * in the given map; the entry is returned
634 * in the "entry" parameter. The boolean
635 * result indicates whether the address is
636 * actually contained in the map.
637 */
638boolean_t
57e43348
MD
639vm_map_lookup_entry(vm_map_t map, vm_offset_t address,
640 vm_map_entry_t *entry /* OUT */)
984263bc 641{
686dbf64 642 vm_map_entry_t tmp;
984263bc
MD
643 vm_map_entry_t last;
644
686dbf64 645#if 0
984263bc 646 /*
686dbf64
MD
647 * XXX TEMPORARILY DISABLED. For some reason our attempt to revive
648 * the hint code with the red-black lookup meets with system crashes
649 * and lockups. We do not yet know why.
650 *
651 * It is possible that the problem is related to the setting
652 * of the hint during map_entry deletion, in the code specified
653 * at the GGG comment later on in this file.
984263bc 654 */
686dbf64
MD
655 /*
656 * Quickly check the cached hint, there's a good chance of a match.
657 */
658 if (map->hint != &map->header) {
659 tmp = map->hint;
660 if (address >= tmp->start && address < tmp->end) {
661 *entry = tmp;
662 return(TRUE);
984263bc 663 }
984263bc 664 }
686dbf64 665#endif
984263bc
MD
666
667 /*
686dbf64
MD
668 * Locate the record from the top of the tree. 'last' tracks the
669 * closest prior record and is returned if no match is found, which
670 * in binary tree terms means tracking the most recent right-branch
671 * taken. If there is no prior record, &map->header is returned.
984263bc 672 */
686dbf64
MD
673 last = &map->header;
674 tmp = RB_ROOT(&map->rb_root);
675
676 while (tmp) {
677 if (address >= tmp->start) {
678 if (address < tmp->end) {
679 *entry = tmp;
680 map->hint = tmp;
681 return(TRUE);
984263bc 682 }
686dbf64
MD
683 last = tmp;
684 tmp = RB_RIGHT(tmp, rb_entry);
685 } else {
686 tmp = RB_LEFT(tmp, rb_entry);
984263bc 687 }
984263bc 688 }
686dbf64 689 *entry = last;
984263bc
MD
690 return (FALSE);
691}
692
693/*
694 * vm_map_insert:
695 *
696 * Inserts the given whole VM object into the target
697 * map at the specified address range. The object's
698 * size should match that of the address range.
699 *
a108bf71
MD
700 * Requires that the map be locked, and leaves it so. Requires that
701 * sufficient vm_map_entry structures have been reserved and tracks
702 * the use via countp.
984263bc
MD
703 *
704 * If object is non-NULL, ref count must be bumped by caller
705 * prior to making call to account for the new entry.
706 */
707int
a108bf71
MD
708vm_map_insert(vm_map_t map, int *countp,
709 vm_object_t object, vm_ooffset_t offset,
1b874851
MD
710 vm_offset_t start, vm_offset_t end,
711 vm_maptype_t maptype,
712 vm_prot_t prot, vm_prot_t max,
984263bc
MD
713 int cow)
714{
715 vm_map_entry_t new_entry;
716 vm_map_entry_t prev_entry;
717 vm_map_entry_t temp_entry;
718 vm_eflags_t protoeflags;
719
720 /*
721 * Check that the start and end points are not bogus.
722 */
723
724 if ((start < map->min_offset) || (end > map->max_offset) ||
725 (start >= end))
726 return (KERN_INVALID_ADDRESS);
727
728 /*
729 * Find the entry prior to the proposed starting address; if it's part
730 * of an existing entry, this range is bogus.
731 */
732
733 if (vm_map_lookup_entry(map, start, &temp_entry))
734 return (KERN_NO_SPACE);
735
736 prev_entry = temp_entry;
737
738 /*
739 * Assert that the next entry doesn't overlap the end point.
740 */
741
742 if ((prev_entry->next != &map->header) &&
743 (prev_entry->next->start < end))
744 return (KERN_NO_SPACE);
745
746 protoeflags = 0;
747
748 if (cow & MAP_COPY_ON_WRITE)
749 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
750
751 if (cow & MAP_NOFAULT) {
752 protoeflags |= MAP_ENTRY_NOFAULT;
753
754 KASSERT(object == NULL,
755 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
756 }
757 if (cow & MAP_DISABLE_SYNCER)
758 protoeflags |= MAP_ENTRY_NOSYNC;
759 if (cow & MAP_DISABLE_COREDUMP)
760 protoeflags |= MAP_ENTRY_NOCOREDUMP;
761
762 if (object) {
763 /*
764 * When object is non-NULL, it could be shared with another
765 * process. We have to set or clear OBJ_ONEMAPPING
766 * appropriately.
767 */
768 if ((object->ref_count > 1) || (object->shadow_count != 0)) {
769 vm_object_clear_flag(object, OBJ_ONEMAPPING);
770 }
771 }
772 else if ((prev_entry != &map->header) &&
773 (prev_entry->eflags == protoeflags) &&
774 (prev_entry->end == start) &&
775 (prev_entry->wired_count == 0) &&
1b874851 776 prev_entry->maptype == maptype &&
984263bc
MD
777 ((prev_entry->object.vm_object == NULL) ||
778 vm_object_coalesce(prev_entry->object.vm_object,
779 OFF_TO_IDX(prev_entry->offset),
780 (vm_size_t)(prev_entry->end - prev_entry->start),
781 (vm_size_t)(end - prev_entry->end)))) {
782 /*
783 * We were able to extend the object. Determine if we
784 * can extend the previous map entry to include the
785 * new range as well.
786 */
787 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
788 (prev_entry->protection == prot) &&
789 (prev_entry->max_protection == max)) {
790 map->size += (end - prev_entry->end);
791 prev_entry->end = end;
a108bf71 792 vm_map_simplify_entry(map, prev_entry, countp);
984263bc
MD
793 return (KERN_SUCCESS);
794 }
795
796 /*
797 * If we can extend the object but cannot extend the
798 * map entry, we have to create a new map entry. We
799 * must bump the ref count on the extended object to
800 * account for it. object may be NULL.
801 */
802 object = prev_entry->object.vm_object;
803 offset = prev_entry->offset +
804 (prev_entry->end - prev_entry->start);
805 vm_object_reference(object);
806 }
807
808 /*
809 * NOTE: if conditionals fail, object can be NULL here. This occurs
810 * in things like the buffer map where we manage kva but do not manage
811 * backing objects.
812 */
813
814 /*
815 * Create a new entry
816 */
817
a108bf71 818 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
819 new_entry->start = start;
820 new_entry->end = end;
821
1b874851 822 new_entry->maptype = maptype;
984263bc
MD
823 new_entry->eflags = protoeflags;
824 new_entry->object.vm_object = object;
825 new_entry->offset = offset;
afeabdca 826 new_entry->aux.master_pde = 0;
984263bc
MD
827
828 new_entry->inheritance = VM_INHERIT_DEFAULT;
829 new_entry->protection = prot;
830 new_entry->max_protection = max;
831 new_entry->wired_count = 0;
832
833 /*
834 * Insert the new entry into the list
835 */
836
837 vm_map_entry_link(map, prev_entry, new_entry);
838 map->size += new_entry->end - new_entry->start;
839
840 /*
841 * Update the free space hint
842 */
843 if ((map->first_free == prev_entry) &&
844 (prev_entry->end >= new_entry->start)) {
845 map->first_free = new_entry;
846 }
847
848#if 0
849 /*
850 * Temporarily removed to avoid MAP_STACK panic, due to
851 * MAP_STACK being a huge hack. Will be added back in
852 * when MAP_STACK (and the user stack mapping) is fixed.
853 */
854 /*
855 * It may be possible to simplify the entry
856 */
a108bf71 857 vm_map_simplify_entry(map, new_entry, countp);
984263bc
MD
858#endif
859
afeabdca
MD
860 /*
861 * Try to pre-populate the page table. Mappings governed by virtual
862 * page tables cannot be prepopulated without a lot of work, so
863 * don't try.
864 */
865 if ((cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) &&
866 maptype != VM_MAPTYPE_VPAGETABLE) {
083a7402 867 pmap_object_init_pt(map->pmap, start, prot,
984263bc
MD
868 object, OFF_TO_IDX(offset), end - start,
869 cow & MAP_PREFAULT_PARTIAL);
870 }
871
872 return (KERN_SUCCESS);
873}
874
875/*
876 * Find sufficient space for `length' bytes in the given map, starting at
877 * `start'. The map must be locked. Returns 0 on success, 1 on no space.
e9bb90e8
MD
878 *
879 * This function will returned an arbitrarily aligned pointer. If no
880 * particular alignment is required you should pass align as 1. Note that
881 * the map may return PAGE_SIZE aligned pointers if all the lengths used in
882 * the map are a multiple of PAGE_SIZE, even if you pass a smaller align
883 * argument.
884 *
885 * 'align' should be a power of 2 but is not required to be.
984263bc
MD
886 */
887int
e9bb90e8
MD
888vm_map_findspace(
889 vm_map_t map,
890 vm_offset_t start,
891 vm_size_t length,
892 vm_offset_t align,
893 vm_offset_t *addr)
984263bc
MD
894{
895 vm_map_entry_t entry, next;
896 vm_offset_t end;
e9bb90e8 897 vm_offset_t align_mask;
984263bc
MD
898
899 if (start < map->min_offset)
900 start = map->min_offset;
901 if (start > map->max_offset)
902 return (1);
903
e9bb90e8
MD
904 /*
905 * If the alignment is not a power of 2 we will have to use
906 * a mod/division, set align_mask to a special value.
907 */
908 if ((align | (align - 1)) + 1 != (align << 1))
909 align_mask = (vm_offset_t)-1;
910 else
911 align_mask = align - 1;
912
a108bf71 913retry:
984263bc
MD
914 /*
915 * Look for the first possible address; if there's already something
916 * at this address, we have to start after it.
917 */
918 if (start == map->min_offset) {
919 if ((entry = map->first_free) != &map->header)
920 start = entry->end;
921 } else {
922 vm_map_entry_t tmp;
923
924 if (vm_map_lookup_entry(map, start, &tmp))
925 start = tmp->end;
926 entry = tmp;
927 }
928
929 /*
930 * Look through the rest of the map, trying to fit a new region in the
931 * gap between existing regions, or after the very last region.
932 */
933 for (;; start = (entry = next)->end) {
e9bb90e8
MD
934 /*
935 * Adjust the proposed start by the requested alignment,
936 * be sure that we didn't wrap the address.
937 */
938 if (align_mask == (vm_offset_t)-1)
939 end = ((start + align - 1) / align) * align;
940 else
941 end = (start + align_mask) & ~align_mask;
942 if (end < start)
943 return (1);
944 start = end;
984263bc
MD
945 /*
946 * Find the end of the proposed new region. Be sure we didn't
e9bb90e8
MD
947 * go beyond the end of the map, or wrap around the address.
948 * Then check to see if this is the last entry or if the
949 * proposed end fits in the gap between this and the next
950 * entry.
984263bc
MD
951 */
952 end = start + length;
953 if (end > map->max_offset || end < start)
954 return (1);
955 next = entry->next;
956 if (next == &map->header || next->start >= end)
957 break;
958 }
686dbf64 959 map->hint = entry;
984263bc
MD
960 if (map == kernel_map) {
961 vm_offset_t ksize;
962 if ((ksize = round_page(start + length)) > kernel_vm_end) {
963 pmap_growkernel(ksize);
a108bf71 964 goto retry;
984263bc
MD
965 }
966 }
a108bf71 967 *addr = start;
984263bc
MD
968 return (0);
969}
970
971/*
972 * vm_map_find finds an unallocated region in the target address
973 * map with the given length. The search is defined to be
974 * first-fit from the specified address; the region found is
975 * returned in the same parameter.
976 *
977 * If object is non-NULL, ref count must be bumped by caller
978 * prior to making call to account for the new entry.
979 */
980int
981vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1b874851
MD
982 vm_offset_t *addr, vm_size_t length,
983 boolean_t find_space,
984 vm_maptype_t maptype,
985 vm_prot_t prot, vm_prot_t max,
986 int cow)
984263bc
MD
987{
988 vm_offset_t start;
03aa8d99 989 int result;
a108bf71 990 int count;
984263bc
MD
991
992 start = *addr;
993
a108bf71 994 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
995 vm_map_lock(map);
996 if (find_space) {
e9bb90e8 997 if (vm_map_findspace(map, start, length, 1, addr)) {
984263bc 998 vm_map_unlock(map);
a108bf71 999 vm_map_entry_release(count);
984263bc
MD
1000 return (KERN_NO_SPACE);
1001 }
1002 start = *addr;
1003 }
a108bf71 1004 result = vm_map_insert(map, &count, object, offset,
1b874851
MD
1005 start, start + length,
1006 maptype,
1007 prot, max,
1008 cow);
984263bc 1009 vm_map_unlock(map);
a108bf71 1010 vm_map_entry_release(count);
984263bc 1011
984263bc
MD
1012 return (result);
1013}
1014
1015/*
1016 * vm_map_simplify_entry:
1017 *
1018 * Simplify the given map entry by merging with either neighbor. This
1019 * routine also has the ability to merge with both neighbors.
1020 *
1021 * The map must be locked.
1022 *
1023 * This routine guarentees that the passed entry remains valid (though
1024 * possibly extended). When merging, this routine may delete one or
1025 * both neighbors. No action is taken on entries which have their
1026 * in-transition flag set.
1027 */
1028void
a108bf71 1029vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc
MD
1030{
1031 vm_map_entry_t next, prev;
1032 vm_size_t prevsize, esize;
1033
1b874851 1034 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
12e4aaff 1035 ++mycpu->gd_cnt.v_intrans_coll;
984263bc
MD
1036 return;
1037 }
1038
1b874851
MD
1039 if (entry->maptype == VM_MAPTYPE_SUBMAP)
1040 return;
1041
984263bc
MD
1042 prev = entry->prev;
1043 if (prev != &map->header) {
1044 prevsize = prev->end - prev->start;
1045 if ( (prev->end == entry->start) &&
1b874851 1046 (prev->maptype == entry->maptype) &&
984263bc
MD
1047 (prev->object.vm_object == entry->object.vm_object) &&
1048 (!prev->object.vm_object ||
1049 (prev->offset + prevsize == entry->offset)) &&
1050 (prev->eflags == entry->eflags) &&
1051 (prev->protection == entry->protection) &&
1052 (prev->max_protection == entry->max_protection) &&
1053 (prev->inheritance == entry->inheritance) &&
1054 (prev->wired_count == entry->wired_count)) {
1055 if (map->first_free == prev)
1056 map->first_free = entry;
1057 if (map->hint == prev)
1058 map->hint = entry;
1059 vm_map_entry_unlink(map, prev);
1060 entry->start = prev->start;
1061 entry->offset = prev->offset;
1062 if (prev->object.vm_object)
1063 vm_object_deallocate(prev->object.vm_object);
a108bf71 1064 vm_map_entry_dispose(map, prev, countp);
984263bc
MD
1065 }
1066 }
1067
1068 next = entry->next;
1069 if (next != &map->header) {
1070 esize = entry->end - entry->start;
1071 if ((entry->end == next->start) &&
1b874851 1072 (next->maptype == entry->maptype) &&
984263bc
MD
1073 (next->object.vm_object == entry->object.vm_object) &&
1074 (!entry->object.vm_object ||
1075 (entry->offset + esize == next->offset)) &&
1076 (next->eflags == entry->eflags) &&
1077 (next->protection == entry->protection) &&
1078 (next->max_protection == entry->max_protection) &&
1079 (next->inheritance == entry->inheritance) &&
1080 (next->wired_count == entry->wired_count)) {
1081 if (map->first_free == next)
1082 map->first_free = entry;
1083 if (map->hint == next)
1084 map->hint = entry;
1085 vm_map_entry_unlink(map, next);
1086 entry->end = next->end;
1087 if (next->object.vm_object)
1088 vm_object_deallocate(next->object.vm_object);
a108bf71 1089 vm_map_entry_dispose(map, next, countp);
984263bc
MD
1090 }
1091 }
1092}
1093/*
1094 * vm_map_clip_start: [ internal use only ]
1095 *
1096 * Asserts that the given entry begins at or after
1097 * the specified address; if necessary,
1098 * it splits the entry into two.
1099 */
a108bf71 1100#define vm_map_clip_start(map, entry, startaddr, countp) \
984263bc
MD
1101{ \
1102 if (startaddr > entry->start) \
a108bf71 1103 _vm_map_clip_start(map, entry, startaddr, countp); \
984263bc
MD
1104}
1105
1106/*
1107 * This routine is called only when it is known that
1108 * the entry must be split.
1109 */
1110static void
a108bf71 1111_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start, int *countp)
984263bc
MD
1112{
1113 vm_map_entry_t new_entry;
1114
1115 /*
1116 * Split off the front portion -- note that we must insert the new
1117 * entry BEFORE this one, so that this entry has the specified
1118 * starting address.
1119 */
1120
a108bf71 1121 vm_map_simplify_entry(map, entry, countp);
984263bc
MD
1122
1123 /*
1124 * If there is no object backing this entry, we might as well create
1125 * one now. If we defer it, an object can get created after the map
1126 * is clipped, and individual objects will be created for the split-up
1127 * map. This is a bit of a hack, but is also about the best place to
1128 * put this improvement.
1129 */
984263bc 1130 if (entry->object.vm_object == NULL && !map->system_map) {
53025830 1131 vm_map_entry_allocate_object(entry);
984263bc
MD
1132 }
1133
a108bf71 1134 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
1135 *new_entry = *entry;
1136
1137 new_entry->end = start;
1138 entry->offset += (start - entry->start);
1139 entry->start = start;
1140
1141 vm_map_entry_link(map, entry->prev, new_entry);
1142
1b874851
MD
1143 switch(entry->maptype) {
1144 case VM_MAPTYPE_NORMAL:
1145 case VM_MAPTYPE_VPAGETABLE:
984263bc 1146 vm_object_reference(new_entry->object.vm_object);
1b874851
MD
1147 break;
1148 default:
1149 break;
984263bc
MD
1150 }
1151}
1152
1153/*
1154 * vm_map_clip_end: [ internal use only ]
1155 *
1156 * Asserts that the given entry ends at or before
1157 * the specified address; if necessary,
1158 * it splits the entry into two.
1159 */
1160
a108bf71 1161#define vm_map_clip_end(map, entry, endaddr, countp) \
984263bc
MD
1162{ \
1163 if (endaddr < entry->end) \
a108bf71 1164 _vm_map_clip_end(map, entry, endaddr, countp); \
984263bc
MD
1165}
1166
1167/*
1168 * This routine is called only when it is known that
1169 * the entry must be split.
1170 */
1171static void
a108bf71 1172_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end, int *countp)
984263bc
MD
1173{
1174 vm_map_entry_t new_entry;
1175
1176 /*
1177 * If there is no object backing this entry, we might as well create
1178 * one now. If we defer it, an object can get created after the map
1179 * is clipped, and individual objects will be created for the split-up
1180 * map. This is a bit of a hack, but is also about the best place to
1181 * put this improvement.
1182 */
1183
1184 if (entry->object.vm_object == NULL && !map->system_map) {
53025830 1185 vm_map_entry_allocate_object(entry);
984263bc
MD
1186 }
1187
1188 /*
1189 * Create a new entry and insert it AFTER the specified entry
1190 */
1191
a108bf71 1192 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
1193 *new_entry = *entry;
1194
1195 new_entry->start = entry->end = end;
1196 new_entry->offset += (end - entry->start);
1197
1198 vm_map_entry_link(map, entry, new_entry);
1199
1b874851
MD
1200 switch(entry->maptype) {
1201 case VM_MAPTYPE_NORMAL:
1202 case VM_MAPTYPE_VPAGETABLE:
984263bc 1203 vm_object_reference(new_entry->object.vm_object);
1b874851
MD
1204 break;
1205 default:
1206 break;
984263bc
MD
1207 }
1208}
1209
1210/*
1211 * VM_MAP_RANGE_CHECK: [ internal use only ]
1212 *
1213 * Asserts that the starting and ending region
1214 * addresses fall within the valid range of the map.
1215 */
1216#define VM_MAP_RANGE_CHECK(map, start, end) \
1217 { \
1218 if (start < vm_map_min(map)) \
1219 start = vm_map_min(map); \
1220 if (end > vm_map_max(map)) \
1221 end = vm_map_max(map); \
1222 if (start > end) \
1223 start = end; \
1224 }
1225
1226/*
1227 * vm_map_transition_wait: [ kernel use only ]
1228 *
1229 * Used to block when an in-transition collison occurs. The map
1230 * is unlocked for the sleep and relocked before the return.
1231 */
1232static
1233void
1234vm_map_transition_wait(vm_map_t map)
1235{
1236 vm_map_unlock(map);
377d4740 1237 tsleep(map, 0, "vment", 0);
984263bc
MD
1238 vm_map_lock(map);
1239}
1240
1241/*
1242 * CLIP_CHECK_BACK
1243 * CLIP_CHECK_FWD
1244 *
1245 * When we do blocking operations with the map lock held it is
1246 * possible that a clip might have occured on our in-transit entry,
1247 * requiring an adjustment to the entry in our loop. These macros
1248 * help the pageable and clip_range code deal with the case. The
1249 * conditional costs virtually nothing if no clipping has occured.
1250 */
1251
1252#define CLIP_CHECK_BACK(entry, save_start) \
1253 do { \
1254 while (entry->start != save_start) { \
1255 entry = entry->prev; \
1256 KASSERT(entry != &map->header, ("bad entry clip")); \
1257 } \
1258 } while(0)
1259
1260#define CLIP_CHECK_FWD(entry, save_end) \
1261 do { \
1262 while (entry->end != save_end) { \
1263 entry = entry->next; \
1264 KASSERT(entry != &map->header, ("bad entry clip")); \
1265 } \
1266 } while(0)
1267
1268
1269/*
1270 * vm_map_clip_range: [ kernel use only ]
1271 *
1272 * Clip the specified range and return the base entry. The
1273 * range may cover several entries starting at the returned base
1274 * and the first and last entry in the covering sequence will be
1275 * properly clipped to the requested start and end address.
1276 *
1277 * If no holes are allowed you should pass the MAP_CLIP_NO_HOLES
1278 * flag.
1279 *
1280 * The MAP_ENTRY_IN_TRANSITION flag will be set for the entries
1281 * covered by the requested range.
1282 *
1283 * The map must be exclusively locked on entry and will remain locked
1284 * on return. If no range exists or the range contains holes and you
1285 * specified that no holes were allowed, NULL will be returned. This
1286 * routine may temporarily unlock the map in order avoid a deadlock when
1287 * sleeping.
1288 */
1289static
1290vm_map_entry_t
a108bf71
MD
1291vm_map_clip_range(vm_map_t map, vm_offset_t start, vm_offset_t end,
1292 int *countp, int flags)
984263bc
MD
1293{
1294 vm_map_entry_t start_entry;
1295 vm_map_entry_t entry;
1296
1297 /*
1298 * Locate the entry and effect initial clipping. The in-transition
1299 * case does not occur very often so do not try to optimize it.
1300 */
1301again:
1302 if (vm_map_lookup_entry(map, start, &start_entry) == FALSE)
1303 return (NULL);
1304 entry = start_entry;
1305 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1306 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
12e4aaff
MD
1307 ++mycpu->gd_cnt.v_intrans_coll;
1308 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
1309 vm_map_transition_wait(map);
1310 /*
1311 * entry and/or start_entry may have been clipped while
1312 * we slept, or may have gone away entirely. We have
1313 * to restart from the lookup.
1314 */
1315 goto again;
1316 }
1317 /*
1318 * Since we hold an exclusive map lock we do not have to restart
1319 * after clipping, even though clipping may block in zalloc.
1320 */
a108bf71
MD
1321 vm_map_clip_start(map, entry, start, countp);
1322 vm_map_clip_end(map, entry, end, countp);
984263bc
MD
1323 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1324
1325 /*
1326 * Scan entries covered by the range. When working on the next
1327 * entry a restart need only re-loop on the current entry which
1328 * we have already locked, since 'next' may have changed. Also,
1329 * even though entry is safe, it may have been clipped so we
1330 * have to iterate forwards through the clip after sleeping.
1331 */
1332 while (entry->next != &map->header && entry->next->start < end) {
1333 vm_map_entry_t next = entry->next;
1334
1335 if (flags & MAP_CLIP_NO_HOLES) {
1336 if (next->start > entry->end) {
1337 vm_map_unclip_range(map, start_entry,
a108bf71 1338 start, entry->end, countp, flags);
984263bc
MD
1339 return(NULL);
1340 }
1341 }
1342
1343 if (next->eflags & MAP_ENTRY_IN_TRANSITION) {
1344 vm_offset_t save_end = entry->end;
1345 next->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
12e4aaff
MD
1346 ++mycpu->gd_cnt.v_intrans_coll;
1347 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
1348 vm_map_transition_wait(map);
1349
1350 /*
1351 * clips might have occured while we blocked.
1352 */
1353 CLIP_CHECK_FWD(entry, save_end);
1354 CLIP_CHECK_BACK(start_entry, start);
1355 continue;
1356 }
1357 /*
1358 * No restart necessary even though clip_end may block, we
1359 * are holding the map lock.
1360 */
a108bf71 1361 vm_map_clip_end(map, next, end, countp);
984263bc
MD
1362 next->eflags |= MAP_ENTRY_IN_TRANSITION;
1363 entry = next;
1364 }
1365 if (flags & MAP_CLIP_NO_HOLES) {
1366 if (entry->end != end) {
1367 vm_map_unclip_range(map, start_entry,
a108bf71 1368 start, entry->end, countp, flags);
984263bc
MD
1369 return(NULL);
1370 }
1371 }
1372 return(start_entry);
1373}
1374
1375/*
1376 * vm_map_unclip_range: [ kernel use only ]
1377 *
1378 * Undo the effect of vm_map_clip_range(). You should pass the same
1379 * flags and the same range that you passed to vm_map_clip_range().
1380 * This code will clear the in-transition flag on the entries and
1381 * wake up anyone waiting. This code will also simplify the sequence
1382 * and attempt to merge it with entries before and after the sequence.
1383 *
1384 * The map must be locked on entry and will remain locked on return.
1385 *
1386 * Note that you should also pass the start_entry returned by
1387 * vm_map_clip_range(). However, if you block between the two calls
1388 * with the map unlocked please be aware that the start_entry may
1389 * have been clipped and you may need to scan it backwards to find
1390 * the entry corresponding with the original start address. You are
1391 * responsible for this, vm_map_unclip_range() expects the correct
1392 * start_entry to be passed to it and will KASSERT otherwise.
1393 */
1394static
1395void
1396vm_map_unclip_range(
1397 vm_map_t map,
1398 vm_map_entry_t start_entry,
1399 vm_offset_t start,
1400 vm_offset_t end,
a108bf71 1401 int *countp,
984263bc
MD
1402 int flags)
1403{
1404 vm_map_entry_t entry;
1405
1406 entry = start_entry;
1407
1408 KASSERT(entry->start == start, ("unclip_range: illegal base entry"));
1409 while (entry != &map->header && entry->start < end) {
1410 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, ("in-transition flag not set during unclip on: %p", entry));
1411 KASSERT(entry->end <= end, ("unclip_range: tail wasn't clipped"));
1412 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1413 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1414 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1415 wakeup(map);
1416 }
1417 entry = entry->next;
1418 }
1419
1420 /*
1421 * Simplification does not block so there is no restart case.
1422 */
1423 entry = start_entry;
1424 while (entry != &map->header && entry->start < end) {
a108bf71 1425 vm_map_simplify_entry(map, entry, countp);
984263bc
MD
1426 entry = entry->next;
1427 }
1428}
1429
1430/*
1431 * vm_map_submap: [ kernel use only ]
1432 *
1433 * Mark the given range as handled by a subordinate map.
1434 *
1435 * This range must have been created with vm_map_find,
1436 * and no other operations may have been performed on this
1437 * range prior to calling vm_map_submap.
1438 *
1439 * Only a limited number of operations can be performed
1440 * within this rage after calling vm_map_submap:
1441 * vm_fault
1442 * [Don't try vm_map_copy!]
1443 *
1444 * To remove a submapping, one must first remove the
1445 * range from the superior map, and then destroy the
1446 * submap (if desired). [Better yet, don't try it.]
1447 */
1448int
a108bf71 1449vm_map_submap(vm_map_t map, vm_offset_t start, vm_offset_t end, vm_map_t submap)
984263bc
MD
1450{
1451 vm_map_entry_t entry;
1452 int result = KERN_INVALID_ARGUMENT;
a108bf71 1453 int count;
984263bc 1454
a108bf71 1455 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1456 vm_map_lock(map);
1457
1458 VM_MAP_RANGE_CHECK(map, start, end);
1459
1460 if (vm_map_lookup_entry(map, start, &entry)) {
a108bf71 1461 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1462 } else {
1463 entry = entry->next;
1464 }
1465
a108bf71 1466 vm_map_clip_end(map, entry, end, &count);
984263bc
MD
1467
1468 if ((entry->start == start) && (entry->end == end) &&
1469 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1470 (entry->object.vm_object == NULL)) {
1471 entry->object.sub_map = submap;
1b874851 1472 entry->maptype = VM_MAPTYPE_SUBMAP;
984263bc
MD
1473 result = KERN_SUCCESS;
1474 }
1475 vm_map_unlock(map);
a108bf71 1476 vm_map_entry_release(count);
984263bc
MD
1477
1478 return (result);
1479}
1480
1481/*
1b874851 1482 * vm_map_protect:
984263bc 1483 *
1b874851
MD
1484 * Sets the protection of the specified address region in the target map.
1485 * If "set_max" is specified, the maximum protection is to be set;
1486 * otherwise, only the current protection is affected.
1487 *
1488 * The protection is not applicable to submaps, but is applicable to normal
1489 * maps and maps governed by virtual page tables. For example, when operating
1490 * on a virtual page table our protection basically controls how COW occurs
1491 * on the backing object, whereas the virtual page table abstraction itself
1492 * is an abstraction for userland.
984263bc
MD
1493 */
1494int
1495vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1496 vm_prot_t new_prot, boolean_t set_max)
1497{
1498 vm_map_entry_t current;
1499 vm_map_entry_t entry;
a108bf71 1500 int count;
984263bc 1501
a108bf71 1502 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1503 vm_map_lock(map);
1504
1505 VM_MAP_RANGE_CHECK(map, start, end);
1506
1507 if (vm_map_lookup_entry(map, start, &entry)) {
a108bf71 1508 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1509 } else {
1510 entry = entry->next;
1511 }
1512
1513 /*
1514 * Make a first pass to check for protection violations.
1515 */
984263bc
MD
1516 current = entry;
1517 while ((current != &map->header) && (current->start < end)) {
1b874851 1518 if (current->maptype == VM_MAPTYPE_SUBMAP) {
984263bc 1519 vm_map_unlock(map);
a108bf71 1520 vm_map_entry_release(count);
984263bc
MD
1521 return (KERN_INVALID_ARGUMENT);
1522 }
1523 if ((new_prot & current->max_protection) != new_prot) {
1524 vm_map_unlock(map);
a108bf71 1525 vm_map_entry_release(count);
984263bc
MD
1526 return (KERN_PROTECTION_FAILURE);
1527 }
1528 current = current->next;
1529 }
1530
1531 /*
1532 * Go back and fix up protections. [Note that clipping is not
1533 * necessary the second time.]
1534 */
984263bc
MD
1535 current = entry;
1536
1537 while ((current != &map->header) && (current->start < end)) {
1538 vm_prot_t old_prot;
1539
a108bf71 1540 vm_map_clip_end(map, current, end, &count);
984263bc
MD
1541
1542 old_prot = current->protection;
1b874851 1543 if (set_max) {
984263bc
MD
1544 current->protection =
1545 (current->max_protection = new_prot) &
1546 old_prot;
1b874851 1547 } else {
984263bc 1548 current->protection = new_prot;
1b874851 1549 }
984263bc
MD
1550
1551 /*
1552 * Update physical map if necessary. Worry about copy-on-write
1553 * here -- CHECK THIS XXX
1554 */
1555
1556 if (current->protection != old_prot) {
1557#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1558 VM_PROT_ALL)
1559
1560 pmap_protect(map->pmap, current->start,
1561 current->end,
1562 current->protection & MASK(current));
1563#undef MASK
1564 }
1565
a108bf71 1566 vm_map_simplify_entry(map, current, &count);
984263bc
MD
1567
1568 current = current->next;
1569 }
1570
1571 vm_map_unlock(map);
a108bf71 1572 vm_map_entry_release(count);
984263bc
MD
1573 return (KERN_SUCCESS);
1574}
1575
1576/*
1577 * vm_map_madvise:
1578 *
1579 * This routine traverses a processes map handling the madvise
1580 * system call. Advisories are classified as either those effecting
1581 * the vm_map_entry structure, or those effecting the underlying
1582 * objects.
afeabdca
MD
1583 *
1584 * The <value> argument is used for extended madvise calls.
984263bc 1585 */
984263bc 1586int
afeabdca
MD
1587vm_map_madvise(vm_map_t map, vm_offset_t start, vm_offset_t end,
1588 int behav, off_t value)
984263bc
MD
1589{
1590 vm_map_entry_t current, entry;
1591 int modify_map = 0;
afeabdca 1592 int error = 0;
a108bf71 1593 int count;
984263bc
MD
1594
1595 /*
1596 * Some madvise calls directly modify the vm_map_entry, in which case
1597 * we need to use an exclusive lock on the map and we need to perform
1598 * various clipping operations. Otherwise we only need a read-lock
1599 * on the map.
1600 */
1601
a108bf71
MD
1602 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1603
984263bc
MD
1604 switch(behav) {
1605 case MADV_NORMAL:
1606 case MADV_SEQUENTIAL:
1607 case MADV_RANDOM:
1608 case MADV_NOSYNC:
1609 case MADV_AUTOSYNC:
1610 case MADV_NOCORE:
1611 case MADV_CORE:
afeabdca
MD
1612 case MADV_SETMAP:
1613 case MADV_INVAL:
984263bc
MD
1614 modify_map = 1;
1615 vm_map_lock(map);
1616 break;
1617 case MADV_WILLNEED:
1618 case MADV_DONTNEED:
1619 case MADV_FREE:
1620 vm_map_lock_read(map);
1621 break;
1622 default:
a108bf71 1623 vm_map_entry_release(count);
afeabdca 1624 return (EINVAL);
984263bc
MD
1625 }
1626
1627 /*
1628 * Locate starting entry and clip if necessary.
1629 */
1630
1631 VM_MAP_RANGE_CHECK(map, start, end);
1632
1633 if (vm_map_lookup_entry(map, start, &entry)) {
1634 if (modify_map)
a108bf71 1635 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1636 } else {
1637 entry = entry->next;
1638 }
1639
1640 if (modify_map) {
1641 /*
1642 * madvise behaviors that are implemented in the vm_map_entry.
1643 *
1644 * We clip the vm_map_entry so that behavioral changes are
1645 * limited to the specified address range.
1646 */
1647 for (current = entry;
1648 (current != &map->header) && (current->start < end);
1649 current = current->next
1650 ) {
1b874851 1651 if (current->maptype == VM_MAPTYPE_SUBMAP)
984263bc
MD
1652 continue;
1653
a108bf71 1654 vm_map_clip_end(map, current, end, &count);
984263bc
MD
1655
1656 switch (behav) {
1657 case MADV_NORMAL:
1658 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1659 break;
1660 case MADV_SEQUENTIAL:
1661 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1662 break;
1663 case MADV_RANDOM:
1664 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1665 break;
1666 case MADV_NOSYNC:
1667 current->eflags |= MAP_ENTRY_NOSYNC;
1668 break;
1669 case MADV_AUTOSYNC:
1670 current->eflags &= ~MAP_ENTRY_NOSYNC;
1671 break;
1672 case MADV_NOCORE:
1673 current->eflags |= MAP_ENTRY_NOCOREDUMP;
1674 break;
1675 case MADV_CORE:
1676 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1677 break;
afeabdca
MD
1678 case MADV_INVAL:
1679 /*
1680 * Invalidate the related pmap entries, used
1681 * to flush portions of the real kernel's
1682 * pmap when the caller has removed or
1683 * modified existing mappings in a virtual
1684 * page table.
1685 */
1686 pmap_remove(map->pmap,
1687 current->start, current->end);
1688 break;
1689 case MADV_SETMAP:
1690 /*
1691 * Set the page directory page for a map
1692 * governed by a virtual page table. Mark
1693 * the entry as being governed by a virtual
1694 * page table if it is not.
1695 *
1696 * XXX the page directory page is stored
1697 * in the avail_ssize field if the map_entry.
1698 *
1699 * XXX the map simplification code does not
1700 * compare this field so weird things may
1701 * happen if you do not apply this function
1702 * to the entire mapping governed by the
1703 * virtual page table.
1704 */
1705 if (current->maptype != VM_MAPTYPE_VPAGETABLE) {
1706 error = EINVAL;
1707 break;
1708 }
1709 current->aux.master_pde = value;
1710 pmap_remove(map->pmap,
1711 current->start, current->end);
1712 break;
984263bc 1713 default:
afeabdca 1714 error = EINVAL;
984263bc
MD
1715 break;
1716 }
a108bf71 1717 vm_map_simplify_entry(map, current, &count);
984263bc
MD
1718 }
1719 vm_map_unlock(map);
1720 } else {
1721 vm_pindex_t pindex;
1722 int count;
1723
1724 /*
1725 * madvise behaviors that are implemented in the underlying
1726 * vm_object.
1727 *
1728 * Since we don't clip the vm_map_entry, we have to clip
1729 * the vm_object pindex and count.
1b874851
MD
1730 *
1731 * NOTE! We currently do not support these functions on
1732 * virtual page tables.
984263bc
MD
1733 */
1734 for (current = entry;
1735 (current != &map->header) && (current->start < end);
1736 current = current->next
1737 ) {
1738 vm_offset_t useStart;
1739
1b874851 1740 if (current->maptype != VM_MAPTYPE_NORMAL)
984263bc
MD
1741 continue;
1742
1743 pindex = OFF_TO_IDX(current->offset);
1744 count = atop(current->end - current->start);
1745 useStart = current->start;
1746
1747 if (current->start < start) {
1748 pindex += atop(start - current->start);
1749 count -= atop(start - current->start);
1750 useStart = start;
1751 }
1752 if (current->end > end)
1753 count -= atop(current->end - end);
1754
1755 if (count <= 0)
1756 continue;
1757
1758 vm_object_madvise(current->object.vm_object,
1759 pindex, count, behav);
afeabdca
MD
1760
1761 /*
1762 * Try to populate the page table. Mappings governed
1763 * by virtual page tables cannot be pre-populated
1764 * without a lot of work so don't try.
1765 */
1766 if (behav == MADV_WILLNEED &&
1767 current->maptype != VM_MAPTYPE_VPAGETABLE) {
984263bc
MD
1768 pmap_object_init_pt(
1769 map->pmap,
1770 useStart,
083a7402 1771 current->protection,
984263bc
MD
1772 current->object.vm_object,
1773 pindex,
1774 (count << PAGE_SHIFT),
1775 MAP_PREFAULT_MADVISE
1776 );
1777 }
1778 }
1779 vm_map_unlock_read(map);
1780 }
a108bf71 1781 vm_map_entry_release(count);
afeabdca 1782 return(error);
984263bc
MD
1783}
1784
1785
1786/*
1787 * vm_map_inherit:
1788 *
1789 * Sets the inheritance of the specified address
1790 * range in the target map. Inheritance
1791 * affects how the map will be shared with
1792 * child maps at the time of vm_map_fork.
1793 */
1794int
1795vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1796 vm_inherit_t new_inheritance)
1797{
1798 vm_map_entry_t entry;
1799 vm_map_entry_t temp_entry;
a108bf71 1800 int count;
984263bc
MD
1801
1802 switch (new_inheritance) {
1803 case VM_INHERIT_NONE:
1804 case VM_INHERIT_COPY:
1805 case VM_INHERIT_SHARE:
1806 break;
1807 default:
1808 return (KERN_INVALID_ARGUMENT);
1809 }
1810
a108bf71 1811 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1812 vm_map_lock(map);
1813
1814 VM_MAP_RANGE_CHECK(map, start, end);
1815
1816 if (vm_map_lookup_entry(map, start, &temp_entry)) {
1817 entry = temp_entry;
a108bf71 1818 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1819 } else
1820 entry = temp_entry->next;
1821
1822 while ((entry != &map->header) && (entry->start < end)) {
a108bf71 1823 vm_map_clip_end(map, entry, end, &count);
984263bc
MD
1824
1825 entry->inheritance = new_inheritance;
1826
a108bf71 1827 vm_map_simplify_entry(map, entry, &count);
984263bc
MD
1828
1829 entry = entry->next;
1830 }
984263bc 1831 vm_map_unlock(map);
a108bf71 1832 vm_map_entry_release(count);
984263bc
MD
1833 return (KERN_SUCCESS);
1834}
1835
1836/*
1837 * Implement the semantics of mlock
1838 */
1839int
57e43348
MD
1840vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t real_end,
1841 boolean_t new_pageable)
984263bc
MD
1842{
1843 vm_map_entry_t entry;
1844 vm_map_entry_t start_entry;
1845 vm_offset_t end;
1846 int rv = KERN_SUCCESS;
a108bf71 1847 int count;
984263bc 1848
a108bf71 1849 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1850 vm_map_lock(map);
1851 VM_MAP_RANGE_CHECK(map, start, real_end);
1852 end = real_end;
1853
a108bf71 1854 start_entry = vm_map_clip_range(map, start, end, &count, MAP_CLIP_NO_HOLES);
984263bc
MD
1855 if (start_entry == NULL) {
1856 vm_map_unlock(map);
a108bf71 1857 vm_map_entry_release(count);
984263bc
MD
1858 return (KERN_INVALID_ADDRESS);
1859 }
1860
1861 if (new_pageable == 0) {
1862 entry = start_entry;
1863 while ((entry != &map->header) && (entry->start < end)) {
1864 vm_offset_t save_start;
1865 vm_offset_t save_end;
1866
1867 /*
1868 * Already user wired or hard wired (trivial cases)
1869 */
1870 if (entry->eflags & MAP_ENTRY_USER_WIRED) {
1871 entry = entry->next;
1872 continue;
1873 }
1874 if (entry->wired_count != 0) {
1875 entry->wired_count++;
1876 entry->eflags |= MAP_ENTRY_USER_WIRED;
1877 entry = entry->next;
1878 continue;
1879 }
1880
1881 /*
1882 * A new wiring requires instantiation of appropriate
1883 * management structures and the faulting in of the
1884 * page.
1885 */
1b874851 1886 if (entry->maptype != VM_MAPTYPE_SUBMAP) {
984263bc
MD
1887 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1888 if (copyflag && ((entry->protection & VM_PROT_WRITE) != 0)) {
53025830 1889 vm_map_entry_shadow(entry);
984263bc
MD
1890 } else if (entry->object.vm_object == NULL &&
1891 !map->system_map) {
53025830 1892 vm_map_entry_allocate_object(entry);
984263bc
MD
1893 }
1894 }
1895 entry->wired_count++;
1896 entry->eflags |= MAP_ENTRY_USER_WIRED;
1897
1898 /*
f2d22ebf
MD
1899 * Now fault in the area. Note that vm_fault_wire()
1900 * may release the map lock temporarily, it will be
1901 * relocked on return. The in-transition
984263bc
MD
1902 * flag protects the entries.
1903 */
1904 save_start = entry->start;
1905 save_end = entry->end;
f2d22ebf 1906 rv = vm_fault_wire(map, entry, TRUE);
984263bc
MD
1907 if (rv) {
1908 CLIP_CHECK_BACK(entry, save_start);
1909 for (;;) {
1910 KASSERT(entry->wired_count == 1, ("bad wired_count on entry"));
1911 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1912 entry->wired_count = 0;
1913 if (entry->end == save_end)
1914 break;
1915 entry = entry->next;
1916 KASSERT(entry != &map->header, ("bad entry clip during backout"));
1917 }
1918 end = save_start; /* unwire the rest */
1919 break;
1920 }
1921 /*
1922 * note that even though the entry might have been
1923 * clipped, the USER_WIRED flag we set prevents
1924 * duplication so we do not have to do a
1925 * clip check.
1926 */
1927 entry = entry->next;
1928 }
1929
1930 /*
1931 * If we failed fall through to the unwiring section to
1932 * unwire what we had wired so far. 'end' has already
1933 * been adjusted.
1934 */
1935 if (rv)
1936 new_pageable = 1;
1937
1938 /*
1939 * start_entry might have been clipped if we unlocked the
1940 * map and blocked. No matter how clipped it has gotten
1941 * there should be a fragment that is on our start boundary.
1942 */
1943 CLIP_CHECK_BACK(start_entry, start);
1944 }
1945
1946 /*
1947 * Deal with the unwiring case.
1948 */
1949 if (new_pageable) {
1950 /*
1951 * This is the unwiring case. We must first ensure that the
1952 * range to be unwired is really wired down. We know there
1953 * are no holes.
1954 */
1955 entry = start_entry;
1956 while ((entry != &map->header) && (entry->start < end)) {
1957 if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
1958 rv = KERN_INVALID_ARGUMENT;
1959 goto done;
1960 }
1961 KASSERT(entry->wired_count != 0, ("wired count was 0 with USER_WIRED set! %p", entry));
1962 entry = entry->next;
1963 }
1964
1965 /*
1966 * Now decrement the wiring count for each region. If a region
1967 * becomes completely unwired, unwire its physical pages and
1968 * mappings.
1969 */
b4eddbac
DR
1970 /*
1971 * The map entries are processed in a loop, checking to
1972 * make sure the entry is wired and asserting it has a wired
1973 * count. However, another loop was inserted more-or-less in
1974 * the middle of the unwiring path. This loop picks up the
1975 * "entry" loop variable from the first loop without first
1976 * setting it to start_entry. Naturally, the secound loop
1977 * is never entered and the pages backing the entries are
1978 * never unwired. This can lead to a leak of wired pages.
1979 */
1980 entry = start_entry;
984263bc 1981 while ((entry != &map->header) && (entry->start < end)) {
f2d22ebf
MD
1982 KASSERT(entry->eflags & MAP_ENTRY_USER_WIRED,
1983 ("expected USER_WIRED on entry %p", entry));
984263bc
MD
1984 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1985 entry->wired_count--;
1986 if (entry->wired_count == 0)
f2d22ebf 1987 vm_fault_unwire(map, entry);
984263bc
MD
1988 entry = entry->next;
1989 }
1990 }
1991done:
a108bf71 1992 vm_map_unclip_range(map, start_entry, start, real_end, &count,
984263bc
MD
1993 MAP_CLIP_NO_HOLES);
1994 map->timestamp++;
1995 vm_map_unlock(map);
a108bf71 1996 vm_map_entry_release(count);
984263bc
MD
1997 return (rv);
1998}
1999
2000/*
cde87949 2001 * vm_map_wire:
984263bc
MD
2002 *
2003 * Sets the pageability of the specified address
2004 * range in the target map. Regions specified
2005 * as not pageable require locked-down physical
2006 * memory and physical page maps.
2007 *
2008 * The map must not be locked, but a reference
2009 * must remain to the map throughout the call.
a108bf71
MD
2010 *
2011 * This function may be called via the zalloc path and must properly
2012 * reserve map entries for kernel_map.
984263bc
MD
2013 */
2014int
e1359933 2015vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t real_end, int kmflags)
984263bc
MD
2016{
2017 vm_map_entry_t entry;
2018 vm_map_entry_t start_entry;
2019 vm_offset_t end;
2020 int rv = KERN_SUCCESS;
a108bf71 2021 int count;
984263bc 2022
e1359933 2023 if (kmflags & KM_KRESERVE)
a108bf71 2024 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
a108bf71
MD
2025 else
2026 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2027 vm_map_lock(map);
2028 VM_MAP_RANGE_CHECK(map, start, real_end);
2029 end = real_end;
2030
a108bf71 2031 start_entry = vm_map_clip_range(map, start, end, &count, MAP_CLIP_NO_HOLES);
984263bc
MD
2032 if (start_entry == NULL) {
2033 vm_map_unlock(map);
a108bf71
MD
2034 rv = KERN_INVALID_ADDRESS;
2035 goto failure;
984263bc 2036 }
e1359933 2037 if ((kmflags & KM_PAGEABLE) == 0) {
984263bc
MD
2038 /*
2039 * Wiring.
2040 *
2041 * 1. Holding the write lock, we create any shadow or zero-fill
2042 * objects that need to be created. Then we clip each map
2043 * entry to the region to be wired and increment its wiring
2044 * count. We create objects before clipping the map entries
2045 * to avoid object proliferation.
2046 *
2047 * 2. We downgrade to a read lock, and call vm_fault_wire to
2048 * fault in the pages for any newly wired area (wired_count is
2049 * 1).
2050 *
2051 * Downgrading to a read lock for vm_fault_wire avoids a
2052 * possible deadlock with another process that may have faulted
2053 * on one of the pages to be wired (it would mark the page busy,
2054 * blocking us, then in turn block on the map lock that we
2055 * hold). Because of problems in the recursive lock package,
2056 * we cannot upgrade to a write lock in vm_map_lookup. Thus,
2057 * any actions that require the write lock must be done
2058 * beforehand. Because we keep the read lock on the map, the
2059 * copy-on-write status of the entries we modify here cannot
2060 * change.
2061 */
2062
2063 entry = start_entry;
2064 while ((entry != &map->header) && (entry->start < end)) {
2065 /*
2066 * Trivial case if the entry is already wired
2067 */
2068 if (entry->wired_count) {
2069 entry->wired_count++;
2070 entry = entry->next;
2071 continue;
2072 }
2073
2074 /*
2075 * The entry is being newly wired, we have to setup
2076 * appropriate management structures. A shadow
2077 * object is required for a copy-on-write region,
2078 * or a normal object for a zero-fill region. We
2079 * do not have to do this for entries that point to sub
2080 * maps because we won't hold the lock on the sub map.
2081 */
1b874851 2082 if (entry->maptype != VM_MAPTYPE_SUBMAP) {
984263bc
MD
2083 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
2084 if (copyflag &&
2085 ((entry->protection & VM_PROT_WRITE) != 0)) {
53025830 2086 vm_map_entry_shadow(entry);
984263bc
MD
2087 } else if (entry->object.vm_object == NULL &&
2088 !map->system_map) {
53025830 2089 vm_map_entry_allocate_object(entry);
984263bc
MD
2090 }
2091 }
2092
2093 entry->wired_count++;
2094 entry = entry->next;
2095 }
2096
2097 /*
2098 * Pass 2.
2099 */
2100
2101 /*
2102 * HACK HACK HACK HACK
2103 *
2104 * Unlock the map to avoid deadlocks. The in-transit flag
2105 * protects us from most changes but note that
2106 * clipping may still occur. To prevent clipping from
2107 * occuring after the unlock, except for when we are
5fd012e0
MD
2108 * blocking in vm_fault_wire, we must run in a critical
2109 * section, otherwise our accesses to entry->start and
2110 * entry->end could be corrupted. We have to enter the
2111 * critical section prior to unlocking so start_entry does
2112 * not change out from under us at the very beginning of the
2113 * loop.
984263bc
MD
2114 *
2115 * HACK HACK HACK HACK
2116 */
2117
5fd012e0 2118 crit_enter();
984263bc
MD
2119
2120 entry = start_entry;
2121 while (entry != &map->header && entry->start < end) {
2122 /*
2123 * If vm_fault_wire fails for any page we need to undo
2124 * what has been done. We decrement the wiring count
2125 * for those pages which have not yet been wired (now)
2126 * and unwire those that have (later).
2127 */
2128 vm_offset_t save_start = entry->start;
2129 vm_offset_t save_end = entry->end;
2130
2131 if (entry->wired_count == 1)
f2d22ebf 2132 rv = vm_fault_wire(map, entry, FALSE);
984263bc
MD
2133 if (rv) {
2134 CLIP_CHECK_BACK(entry, save_start);
2135 for (;;) {
2136 KASSERT(entry->wired_count == 1, ("wired_count changed unexpectedly"));
2137 entry->wired_count = 0;
2138 if (entry->end == save_end)
2139 break;
2140 entry = entry->next;
2141 KASSERT(entry != &map->header, ("bad entry clip during backout"));
2142 }
2143 end = save_start;
2144 break;
2145 }
2146 CLIP_CHECK_FWD(entry, save_end);
2147 entry = entry->next;
2148 }
5fd012e0 2149 crit_exit();
984263bc 2150
984263bc
MD
2151 /*
2152 * If a failure occured undo everything by falling through
2153 * to the unwiring code. 'end' has already been adjusted
2154 * appropriately.
2155 */
2156 if (rv)
e1359933 2157 kmflags |= KM_PAGEABLE;
984263bc
MD
2158
2159 /*
f2d22ebf
MD
2160 * start_entry is still IN_TRANSITION but may have been
2161 * clipped since vm_fault_wire() unlocks and relocks the
2162 * map. No matter how clipped it has gotten there should
2163 * be a fragment that is on our start boundary.
984263bc
MD
2164 */
2165 CLIP_CHECK_BACK(start_entry, start);
2166 }
2167
e1359933 2168 if (kmflags & KM_PAGEABLE) {
984263bc
MD
2169 /*
2170 * This is the unwiring case. We must first ensure that the
2171 * range to be unwired is really wired down. We know there
2172 * are no holes.
2173 */
2174 entry = start_entry;
2175 while ((entry != &map->header) && (entry->start < end)) {
2176 if (entry->wired_count == 0) {
2177 rv = KERN_INVALID_ARGUMENT;
2178 goto done;
2179 }
2180 entry = entry->next;
2181 }
2182
2183 /*
2184 * Now decrement the wiring count for each region. If a region
2185 * becomes completely unwired, unwire its physical pages and
2186 * mappings.
2187 */
2188 entry = start_entry;
2189 while ((entry != &map->header) && (entry->start < end)) {
2190 entry->wired_count--;
2191 if (entry->wired_count == 0)
f2d22ebf 2192 vm_fault_unwire(map, entry);
984263bc
MD
2193 entry = entry->next;
2194 }
2195 }
2196done:
a108bf71 2197 vm_map_unclip_range(map, start_entry, start, real_end, &count,
984263bc
MD
2198 MAP_CLIP_NO_HOLES);
2199 map->timestamp++;
2200 vm_map_unlock(map);
a108bf71 2201failure:
e1359933 2202 if (kmflags & KM_KRESERVE)
a108bf71 2203 vm_map_entry_krelease(count);
a108bf71
MD
2204 else
2205 vm_map_entry_release(count);
984263bc
MD
2206 return (rv);
2207}
2208
a108bf71
MD
2209/*
2210 * vm_map_set_wired_quick()
2211 *
2212 * Mark a newly allocated address range as wired but do not fault in
2213 * the pages. The caller is expected to load the pages into the object.
2214 *
2215 * The map must be locked on entry and will remain locked on return.
2216 */
2217void
2218vm_map_set_wired_quick(vm_map_t map, vm_offset_t addr, vm_size_t size, int *countp)
2219{
2220 vm_map_entry_t scan;
2221 vm_map_entry_t entry;
2222
2223 entry = vm_map_clip_range(map, addr, addr + size, countp, MAP_CLIP_NO_HOLES);
2224 for (scan = entry; scan != &map->header && scan->start < addr + size; scan = scan->next) {
2225 KKASSERT(entry->wired_count == 0);
2226 entry->wired_count = 1;
2227 }
2228 vm_map_unclip_range(map, entry, addr, addr + size, countp, MAP_CLIP_NO_HOLES);
2229}
2230
984263bc
MD
2231/*
2232 * vm_map_clean
2233 *
2234 * Push any dirty cached pages in the address range to their pager.
2235 * If syncio is TRUE, dirty pages are written synchronously.
2236 * If invalidate is TRUE, any cached pages are freed as well.
2237 *
2238 * Returns an error if any part of the specified range is not mapped.
2239 */
2240int
57e43348
MD
2241vm_map_clean(vm_map_t map, vm_offset_t start, vm_offset_t end, boolean_t syncio,
2242 boolean_t invalidate)
984263bc
MD
2243{
2244 vm_map_entry_t current;
2245 vm_map_entry_t entry;
2246 vm_size_t size;
2247 vm_object_t object;
2248 vm_ooffset_t offset;
2249
2250 vm_map_lock_read(map);
2251 VM_MAP_RANGE_CHECK(map, start, end);
2252 if (!vm_map_lookup_entry(map, start, &entry)) {
2253 vm_map_unlock_read(map);
2254 return (KERN_INVALID_ADDRESS);
2255 }
2256 /*
2257 * Make a first pass to check for holes.
2258 */
2259 for (current = entry; current->start < end; current = current->next) {
1b874851 2260 if (current->maptype == VM_MAPTYPE_SUBMAP) {
984263bc
MD
2261 vm_map_unlock_read(map);
2262 return (KERN_INVALID_ARGUMENT);
2263 }
2264 if (end > current->end &&
2265 (current->next == &map->header ||
2266 current->end != current->next->start)) {
2267 vm_map_unlock_read(map);
2268 return (KERN_INVALID_ADDRESS);
2269 }
2270 }
2271
2272 if (invalidate)
2273 pmap_remove(vm_map_pmap(map), start, end);
2274 /*
2275 * Make a second pass, cleaning/uncaching pages from the indicated
2276 * objects as we go.
2277 */
2278 for (current = entry; current->start < end; current = current->next) {
2279 offset = current->offset + (start - current->start);
2280 size = (end <= current->end ? end : current->end) - start;
1b874851 2281 if (current->maptype == VM_MAPTYPE_SUBMAP) {
984263bc
MD
2282 vm_map_t smap;
2283 vm_map_entry_t tentry;
2284 vm_size_t tsize;
2285
2286 smap = current->object.sub_map;
2287 vm_map_lock_read(smap);
418ff780 2288 vm_map_lookup_entry(smap, offset, &tentry);
984263bc
MD
2289 tsize = tentry->end - offset;
2290 if (tsize < size)
2291 size = tsize;
2292 object = tentry->object.vm_object;
2293 offset = tentry->offset + (offset - tentry->start);
2294 vm_map_unlock_read(smap);
2295 } else {
2296 object = current->object.vm_object;
2297 }
2298 /*
2299 * Note that there is absolutely no sense in writing out
2300 * anonymous objects, so we track down the vnode object
2301 * to write out.
2302 * We invalidate (remove) all pages from the address space
2303 * anyway, for semantic correctness.
2304 *
2305 * note: certain anonymous maps, such as MAP_NOSYNC maps,
2306 * may start out with a NULL object.
2307 */
2308 while (object && object->backing_object) {
984263bc 2309 offset += object->backing_object_offset;
7cad6903 2310 object = object->backing_object;
984263bc
MD
2311 if (object->size < OFF_TO_IDX( offset + size))
2312 size = IDX_TO_OFF(object->size) - offset;
2313 }
2314 if (object && (object->type == OBJT_VNODE) &&
2315 (current->protection & VM_PROT_WRITE)) {
2316 /*
2317 * Flush pages if writing is allowed, invalidate them
2318 * if invalidation requested. Pages undergoing I/O
2319 * will be ignored by vm_object_page_remove().
2320 *
2321 * We cannot lock the vnode and then wait for paging
2322 * to complete without deadlocking against vm_fault.
2323 * Instead we simply call vm_object_page_remove() and
2324 * allow it to block internally on a page-by-page
2325 * basis when it encounters pages undergoing async
2326 * I/O.
2327 */
2328 int flags;
2329
2330 vm_object_reference(object);
ca466bae 2331 vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY);
984263bc
MD
2332 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
2333 flags |= invalidate ? OBJPC_INVAL : 0;
1b874851
MD
2334
2335 /*
2336 * When operating on a virtual page table just
2337 * flush the whole object. XXX we probably ought
2338 * to
2339 */
2340 switch(current->maptype) {
2341 case VM_MAPTYPE_NORMAL:
2342 vm_object_page_clean(object,
2343 OFF_TO_IDX(offset),
2344 OFF_TO_IDX(offset + size + PAGE_MASK),
2345 flags);
2346 break;
2347 case VM_MAPTYPE_VPAGETABLE:
2348 vm_object_page_clean(object, 0, 0, flags);
2349 break;
2350 }
a11aaa81 2351 vn_unlock(((struct vnode *)object->handle));
984263bc
MD
2352 vm_object_deallocate(object);
2353 }
2354 if (object && invalidate &&
2355 ((object->type == OBJT_VNODE) ||
2356 (object->type == OBJT_DEVICE))) {
2f1821ca
MD
2357 int clean_only =
2358 (object->type == OBJT_DEVICE) ? FALSE : TRUE;
984263bc 2359 vm_object_reference(object);
1b874851
MD
2360 switch(current->maptype) {
2361 case VM_MAPTYPE_NORMAL:
2362 vm_object_page_remove(object,
2363 OFF_TO_IDX(offset),
2364 OFF_TO_IDX(offset + size + PAGE_MASK),
2365 clean_only);
2366 break;
2367 case VM_MAPTYPE_VPAGETABLE:
2368 vm_object_page_remove(object, 0, 0, clean_only);
2369 break;
2370 }
984263bc
MD
2371 vm_object_deallocate(object);
2372 }
2373 start += size;
2374 }
2375
2376 vm_map_unlock_read(map);
2377 return (KERN_SUCCESS);
2378}
2379
2380/*
2381 * vm_map_entry_unwire: [ internal use only ]
2382 *
2383 * Make the region specified by this entry pageable.
2384 *
2385 * The map in question should be locked.
2386 * [This is the reason for this routine's existence.]
2387 */
2388static void
a108bf71 2389vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
984263bc 2390{
f2d22ebf 2391 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
984263bc 2392 entry->wired_count = 0;
f2d22ebf 2393 vm_fault_unwire(map, entry);
984263bc
MD
2394}
2395
2396/*
2397 * vm_map_entry_delete: [ internal use only ]
2398 *
2399 * Deallocate the given entry from the target map.
2400 */
2401static void
a108bf71 2402vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc
MD
2403{
2404 vm_map_entry_unlink(map, entry);
2405 map->size -= entry->end - entry->start;
2406
1b874851
MD
2407 switch(entry->maptype) {
2408 case VM_MAPTYPE_NORMAL:
2409 case VM_MAPTYPE_VPAGETABLE:
984263bc 2410 vm_object_deallocate(entry->object.vm_object);
1b874851
MD
2411 break;
2412 default:
2413 break;
984263bc
MD
2414 }
2415
a108bf71 2416 vm_map_entry_dispose(map, entry, countp);
984263bc
MD
2417}
2418
2419/*
2420 * vm_map_delete: [ internal use only ]
2421 *
2422 * Deallocates the given address range from the target
2423 * map.
2424 */
2425int
a108bf71 2426vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end, int *countp)
984263bc
MD
2427{
2428 vm_object_t object;
2429 vm_map_entry_t entry;
2430 vm_map_entry_t first_entry;
2431
686dbf64 2432again:
984263bc 2433 /*
686dbf64
MD
2434 * Find the start of the region, and clip it. Set entry to point
2435 * at the first record containing the requested address or, if no
2436 * such record exists, the next record with a greater address. The
2437 * loop will run from this point until a record beyond the termination
2438 * address is encountered.
2439 *
2440 * map->hint must be adjusted to not point to anything we delete,
2441 * so set it to the entry prior to the one being deleted.
2442 *
2443 * GGG see other GGG comment.
984263bc 2444 */
686dbf64 2445 if (vm_map_lookup_entry(map, start, &first_entry)) {
984263bc 2446 entry = first_entry;
a108bf71 2447 vm_map_clip_start(map, entry, start, countp);
686dbf64
MD
2448 map->hint = entry->prev; /* possible problem XXX */
2449 } else {
2450 map->hint = first_entry; /* possible problem XXX */
2451 entry = first_entry->next;
984263bc
MD
2452 }
2453
2454 /*
686dbf64
MD
2455 * If a hole opens up prior to the current first_free then
2456 * adjust first_free. As with map->hint, map->first_free
2457 * cannot be left set to anything we might delete.
984263bc 2458 */
984263bc
MD
2459 if (entry == &map->header) {
2460 map->first_free = &map->header;
2461 } else if (map->first_free->start >= start) {
2462 map->first_free = entry->prev;
2463 }
2464
2465 /*
2466 * Step through all entries in this region
2467 */
2468
2469 while ((entry != &map->header) && (entry->start < end)) {
2470 vm_map_entry_t next;
2471 vm_offset_t s, e;
2472 vm_pindex_t offidxstart, offidxend, count;
2473
2474 /*
2475 * If we hit an in-transition entry we have to sleep and
2476 * retry. It's easier (and not really slower) to just retry
2477 * since this case occurs so rarely and the hint is already
2478 * pointing at the right place. We have to reset the
2479 * start offset so as not to accidently delete an entry
2480 * another process just created in vacated space.
2481 */
2482 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2483 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2484 start = entry->start;
12e4aaff
MD
2485 ++mycpu->gd_cnt.v_intrans_coll;
2486 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
2487 vm_map_transition_wait(map);
2488 goto again;
2489 }
a108bf71 2490 vm_map_clip_end(map, entry, end, countp);
984263bc
MD
2491
2492 s = entry->start;
2493 e = entry->end;
2494 next = entry->next;
2495
2496 offidxstart = OFF_TO_IDX(entry->offset);
2497 count = OFF_TO_IDX(e - s);
2498 object = entry->object.vm_object;
2499
2500 /*
2501 * Unwire before removing addresses from the pmap; otherwise,
2502 * unwiring will put the entries back in the pmap.
2503 */
f2d22ebf 2504 if (entry->wired_count != 0)
984263bc 2505 vm_map_entry_unwire(map, entry);
984263bc
MD
2506
2507 offidxend = offidxstart + count;
2508
2509 if ((object == kernel_object) || (object == kmem_object)) {
2510 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2511 } else {
2512 pmap_remove(map->pmap, s, e);
2513 if (object != NULL &&
2514 object->ref_count != 1 &&
2515 (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING &&
2516 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2517 vm_object_collapse(object);
2518 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2519 if (object->type == OBJT_SWAP) {
2520 swap_pager_freespace(object, offidxstart, count);
2521 }
2522 if (offidxend >= object->size &&
2523 offidxstart < object->size) {
2524 object->size = offidxstart;
2525 }
2526 }
2527 }
2528
2529 /*
2530 * Delete the entry (which may delete the object) only after
2531 * removing all pmap entries pointing to its pages.
2532 * (Otherwise, its page frames may be reallocated, and any
2533 * modify bits will be set in the wrong object!)
2534 */
a108bf71 2535 vm_map_entry_delete(map, entry, countp);
984263bc
MD
2536 entry = next;
2537 }
2538 return (KERN_SUCCESS);
2539}
2540
2541/*
2542 * vm_map_remove:
2543 *
2544 * Remove the given address range from the target map.
2545 * This is the exported form of vm_map_delete.
2546 */
2547int
a108bf71 2548vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
984263bc 2549{
03aa8d99 2550 int result;
a108bf71 2551 int count;
984263bc 2552
a108bf71 2553 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2554 vm_map_lock(map);
2555 VM_MAP_RANGE_CHECK(map, start, end);
a108bf71 2556 result = vm_map_delete(map, start, end, &count);
984263bc 2557 vm_map_unlock(map);
a108bf71 2558 vm_map_entry_release(count);
984263bc 2559
984263bc
MD
2560 return (result);
2561}
2562
2563/*
2564 * vm_map_check_protection:
2565 *
2566 * Assert that the target map allows the specified
2567 * privilege on the entire address region given.
2568 * The entire region must be allocated.
2569 */
2570boolean_t
2571vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2572 vm_prot_t protection)
2573{
2574 vm_map_entry_t entry;
2575 vm_map_entry_t tmp_entry;
2576
2577 if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2578 return (FALSE);
2579 }
2580 entry = tmp_entry;
2581
2582 while (start < end) {
2583 if (entry == &map->header) {
2584 return (FALSE);
2585 }
2586 /*
2587 * No holes allowed!
2588 */
2589
2590 if (start < entry->start) {
2591 return (FALSE);
2592 }
2593 /*
2594 * Check protection associated with entry.
2595 */
2596
2597 if ((entry->protection & protection) != protection) {
2598 return (FALSE);
2599 }
2600 /* go to next entry */
2601
2602 start = entry->end;
2603 entry = entry->next;
2604 }
2605 return (TRUE);
2606}
2607
2608/*
2609 * Split the pages in a map entry into a new object. This affords
2610 * easier removal of unused pages, and keeps object inheritance from
2611 * being a negative impact on memory usage.
2612 */
2613static void
a108bf71 2614vm_map_split(vm_map_entry_t entry)
984263bc
MD
2615{
2616 vm_page_t m;
2617 vm_object_t orig_object, new_object, source;
2618 vm_offset_t s, e;
2619 vm_pindex_t offidxstart, offidxend, idx;
2620 vm_size_t size;
2621 vm_ooffset_t offset;
2622
2623 orig_object = entry->object.vm_object;
2624 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
2625 return;
2626 if (orig_object->ref_count <= 1)
2627 return;
2628
2629 offset = entry->offset;
2630 s = entry->start;
2631 e = entry->end;
2632
2633 offidxstart = OFF_TO_IDX(offset);
2634 offidxend = offidxstart + OFF_TO_IDX(e - s);
2635 size = offidxend - offidxstart;
2636
57f7b636
MD
2637 new_object = vm_pager_allocate(orig_object->type, NULL,
2638 IDX_TO_OFF(size), VM_PROT_ALL, 0);
984263bc
MD
2639 if (new_object == NULL)
2640 return;
2641
2642 source = orig_object->backing_object;
2643 if (source != NULL) {
2644 vm_object_reference(source); /* Referenced by new_object */
2645 LIST_INSERT_HEAD(&source->shadow_head,
2646 new_object, shadow_list);
2647 vm_object_clear_flag(source, OBJ_ONEMAPPING);
2648 new_object->backing_object_offset =
2649 orig_object->backing_object_offset + IDX_TO_OFF(offidxstart);
2650 new_object->backing_object = source;
2651 source->shadow_count++;
2652 source->generation++;
2653 }
2654
2655 for (idx = 0; idx < size; idx++) {
2656 vm_page_t m;
2657
06ecca5a 2658 /*
5fd012e0 2659 * A critical section is required to avoid a race between
06ecca5a
MD
2660 * the lookup and an interrupt/unbusy/free and our busy
2661 * check.
2662 */
5fd012e0 2663 crit_enter();
984263bc
MD
2664 retry:
2665 m = vm_page_lookup(orig_object, offidxstart + idx);
06ecca5a 2666 if (m == NULL) {
5fd012e0 2667 crit_exit();
984263bc 2668 continue;
06ecca5a 2669 }
984263bc
MD
2670
2671 /*
2672 * We must wait for pending I/O to complete before we can
2673 * rename the page.
2674 *
2675 * We do not have to VM_PROT_NONE the page as mappings should
2676 * not be changed by this operation.
2677 */
2678 if (vm_page_sleep_busy(m, TRUE, "spltwt"))
2679 goto retry;
984263bc
MD
2680 vm_page_busy(m);
2681 vm_page_rename(m, new_object, idx);
2682 /* page automatically made dirty by rename and cache handled */
2683 vm_page_busy(m);
5fd012e0 2684 crit_exit();
984263bc
MD
2685 }
2686
2687 if (orig_object->type == OBJT_SWAP) {
2688 vm_object_pip_add(orig_object, 1);
2689 /*
2690 * copy orig_object pages into new_object
2691 * and destroy unneeded pages in
2692 * shadow object.
2693 */
2694 swap_pager_copy(orig_object, new_object, offidxstart, 0);
2695 vm_object_pip_wakeup(orig_object);
2696 }
2697
06ecca5a
MD
2698 /*
2699 * Wakeup the pages we played with. No spl protection is needed
2700 * for a simple wakeup.
2701 */
984263bc
MD
2702 for (idx = 0; idx < size; idx++) {
2703 m = vm_page_lookup(new_object, idx);
06ecca5a 2704 if (m)
984263bc 2705 vm_page_wakeup(m);
984263bc
MD
2706 }
2707
2708 entry->object.vm_object = new_object;
2709 entry->offset = 0LL;
2710 vm_object_deallocate(orig_object);
2711}
2712
2713/*
2714 * vm_map_copy_entry:
2715 *
2716 * Copies the contents of the source entry to the destination
2717 * entry. The entries *must* be aligned properly.
2718 */
2719static void
a108bf71
MD
2720vm_map_copy_entry(vm_map_t src_map, vm_map_t dst_map,
2721 vm_map_entry_t src_entry, vm_map_entry_t dst_entry)
984263bc
MD
2722{
2723 vm_object_t src_object;
2724
1b874851
MD
2725 if (dst_entry->maptype == VM_MAPTYPE_SUBMAP)
2726 return;
2727 if (src_entry->maptype == VM_MAPTYPE_SUBMAP)
984263bc
MD
2728 return;
2729
2730 if (src_entry->wired_count == 0) {
984263bc
MD
2731 /*
2732 * If the source entry is marked needs_copy, it is already
2733 * write-protected.
2734 */
2735 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2736 pmap_protect(src_map->pmap,
2737 src_entry->start,
2738 src_entry->end,
2739 src_entry->protection & ~VM_PROT_WRITE);
2740 }
2741
2742 /*
2743 * Make a copy of the object.
2744 */
2745 if ((src_object = src_entry->object.vm_object) != NULL) {
984263bc
MD
2746 if ((src_object->handle == NULL) &&
2747 (src_object->type == OBJT_DEFAULT ||
2748 src_object->type == OBJT_SWAP)) {
2749 vm_object_collapse(src_object);
2750 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2751 vm_map_split(src_entry);
2752 src_object = src_entry->object.vm_object;
2753 }
2754 }
2755
2756 vm_object_reference(src_object);
2757 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2758 dst_entry->object.vm_object = src_object;
2759 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2760 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2761 dst_entry->offset = src_entry->offset;
2762 } else {
2763 dst_entry->object.vm_object = NULL;
2764 dst_entry->offset = 0;
2765 }
2766
2767 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2768 dst_entry->end - dst_entry->start, src_entry->start);
2769 } else {
2770 /*
2771 * Of course, wired down pages can't be set copy-on-write.
2772 * Cause wired pages to be copied into the new map by
2773 * simulating faults (the new pages are pageable)
2774 */
2775 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2776 }
2777}
2778
2779/*
2780 * vmspace_fork:
2781 * Create a new process vmspace structure and vm_map
2782 * based on those of an existing process. The new map
2783 * is based on the old map, according to the inheritance
2784 * values on the regions in that map.
2785 *
2786 * The source map must not be locked.
2787 */
2788struct vmspace *
a108bf71 2789vmspace_fork(struct vmspace *vm1)
984263bc
MD
2790{
2791 struct vmspace *vm2;
2792 vm_map_t old_map = &vm1->vm_map;
2793 vm_map_t new_map;
2794 vm_map_entry_t old_entry;
2795 vm_map_entry_t new_entry;
2796 vm_object_t object;
a108bf71 2797 int count;
984263bc
MD
2798
2799 vm_map_lock(old_map);
2800 old_map->infork = 1;
2801
239b4df9
MD
2802 /*
2803 * XXX Note: upcalls are not copied.
2804 */
984263bc
MD
2805 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
2806 bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
239b4df9 2807 (caddr_t)&vm1->vm_endcopy - (caddr_t)&vm1->vm_startcopy);
984263bc
MD
2808 new_map = &vm2->vm_map; /* XXX */
2809 new_map->timestamp = 1;
2810
a108bf71 2811 count = 0;
984263bc 2812 old_entry = old_map->header.next;
a108bf71
MD
2813 while (old_entry != &old_map->header) {
2814 ++count;
2815 old_entry = old_entry->next;
2816 }
984263bc 2817
a108bf71
MD
2818 count = vm_map_entry_reserve(count + MAP_RESERVE_COUNT);
2819
2820 old_entry = old_map->header.next;
984263bc 2821 while (old_entry != &old_map->header) {
1b874851 2822 if (old_entry->maptype == VM_MAPTYPE_SUBMAP)
984263bc
MD
2823 panic("vm_map_fork: encountered a submap");
2824
2825 switch (old_entry->inheritance) {
2826 case VM_INHERIT_NONE:
2827 break;
2828
2829 case VM_INHERIT_SHARE:
2830 /*
1b874851
MD
2831 * Clone the entry, creating the shared object if
2832 * necessary.
984263bc
MD
2833 */
2834 object = old_entry->object.vm_object;
2835 if (object == NULL) {
53025830
MD
2836 vm_map_entry_allocate_object(old_entry);
2837 object = old_entry->object.vm_object;
984263bc
MD
2838 }
2839
2840 /*
53025830 2841 * Add the reference before calling vm_map_entry_shadow
984263bc
MD
2842 * to insure that a shadow object is created.
2843 */
2844 vm_object_reference(object);
2845 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
53025830 2846 vm_map_entry_shadow(old_entry);
984263bc
MD
2847 /* Transfer the second reference too. */
2848 vm_object_reference(
2849 old_entry->object.vm_object);
2850 vm_object_deallocate(object);
2851 object = old_entry->object.vm_object;
2852 }
2853 vm_object_clear_flag(object, OBJ_ONEMAPPING);
2854
2855 /*
2856 * Clone the entry, referencing the shared object.
2857 */
a108bf71 2858 new_entry = vm_map_entry_create(new_map, &count);
984263bc
MD
2859 *new_entry = *old_entry;
2860 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2861 new_entry->wired_count = 0;
2862
2863 /*
2864 * Insert the entry into the new map -- we know we're
2865 * inserting at the end of the new map.
2866 */
2867
2868 vm_map_entry_link(new_map, new_map->header.prev,
2869 new_entry);
2870
2871 /*
2872 * Update the physical map
2873 */
2874
2875 pmap_copy(new_map->pmap, old_map->pmap,
2876 new_entry->start,
2877 (old_entry->end - old_entry->start),
2878 old_entry->start);
2879 break;
2880
2881 case VM_INHERIT_COPY:
2882 /*
2883 * Clone the entry and link into the map.
2884 */
a108bf71 2885 new_entry = vm_map_entry_create(new_map, &count);
984263bc
MD
2886 *new_entry = *old_entry;
2887 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2888 new_entry->wired_count = 0;
2889 new_entry->object.vm_object = NULL;
2890 vm_map_entry_link(new_map, new_map->header.prev,
2891 new_entry);
2892 vm_map_copy_entry(old_map, new_map, old_entry,
2893 new_entry);
2894 break;
2895 }
2896 old_entry = old_entry->next;
2897 }
2898
2899 new_map->size = old_map->size;
2900 old_map->infork = 0;
2901 vm_map_unlock(old_map);
a108bf71 2902 vm_map_entry_release(count);
984263bc
MD
2903
2904 return (vm2);
2905}
2906
2907int
2908vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
2909 vm_prot_t prot, vm_prot_t max, int cow)
2910{
2911 vm_map_entry_t prev_entry;
2912 vm_map_entry_t new_stack_entry;
2913 vm_size_t init_ssize;
2914 int rv;
a108bf71 2915 int count;
984263bc
MD
2916
2917 if (VM_MIN_ADDRESS > 0 && addrbos < VM_MIN_ADDRESS)
2918 return (KERN_NO_SPACE);
2919
2920 if (max_ssize < sgrowsiz)
2921 init_ssize = max_ssize;
2922 else
2923 init_ssize = sgrowsiz;
2924
a108bf71 2925 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2926 vm_map_lock(map);
2927
2928 /* If addr is already mapped, no go */
2929 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
2930 vm_map_unlock(map);
a108bf71 2931 vm_map_entry_release(count);
984263bc
MD
2932 return (KERN_NO_SPACE);
2933 }
2934
2935 /* If we would blow our VMEM resource limit, no go */
2936 if (map->size + init_ssize >
2937 curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
2938 vm_map_unlock(map);
a108bf71 2939 vm_map_entry_release(count);
984263bc
MD
2940 return (KERN_NO_SPACE);
2941 }
2942
2943 /* If we can't accomodate max_ssize in the current mapping,
2944 * no go. However, we need to be aware that subsequent user
2945 * mappings might map into the space we have reserved for
2946 * stack, and currently this space is not protected.
2947 *
2948 * Hopefully we will at least detect this condition
2949 * when we try to grow the stack.
2950 */
2951 if ((prev_entry->next != &map->header) &&
2952 (prev_entry->next->start < addrbos + max_ssize)) {
2953 vm_map_unlock(map);
a108bf71 2954 vm_map_entry_release(count);
984263bc
MD
2955 return (KERN_NO_SPACE);
2956 }
2957
2958 /* We initially map a stack of only init_ssize. We will
2959 * grow as needed later. Since this is to be a grow
2960 * down stack, we map at the top of the range.
2961 *
2962 * Note: we would normally expect prot and max to be
2963 * VM_PROT_ALL, and cow to be 0. Possibly we should
2964 * eliminate these as input parameters, and just
2965 * pass these values here in the insert call.
2966 */
a108bf71
MD
2967 rv = vm_map_insert(map, &count,
2968 NULL, 0, addrbos + max_ssize - init_ssize,
1b874851
MD
2969 addrbos + max_ssize,
2970 VM_MAPTYPE_NORMAL,
2971 prot, max,
2972 cow);
984263bc
MD
2973
2974 /* Now set the avail_ssize amount */
517e1666 2975 if (rv == KERN_SUCCESS) {
984263bc 2976 if (prev_entry != &map->header)
a108bf71 2977 vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize, &count);
984263bc
MD
2978 new_stack_entry = prev_entry->next;
2979 if (new_stack_entry->end != addrbos + max_ssize ||
2980 new_stack_entry->start != addrbos + max_ssize - init_ssize)
2981 panic ("Bad entry start/end for new stack entry");
2982 else
afeabdca 2983 new_stack_entry->aux.avail_ssize = max_ssize - init_ssize;
984263bc
MD
2984 }
2985
2986 vm_map_unlock(map);
a108bf71 2987 vm_map_entry_release(count);
984263bc
MD
2988 return (rv);
2989}
2990
2991/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
2992 * desired address is already mapped, or if we successfully grow
2993 * the stack. Also returns KERN_SUCCESS if addr is outside the
2994 * stack range (this is strange, but preserves compatibility with
2995 * the grow function in vm_machdep.c).
2996 */
2997int
2998vm_map_growstack (struct proc *p, vm_offset_t addr)
2999{
3000 vm_map_entry_t prev_entry;
3001 vm_map_entry_t stack_entry;
3002 vm_map_entry_t new_stack_entry;
3003 struct vmspace *vm = p->p_vmspace;
3004 vm_map_t map = &vm->vm_map;
3005 vm_offset_t end;
a108bf71
MD
3006 int grow_amount;
3007 int rv = KERN_SUCCESS;
3008 int is_procstack;
3009 int use_read_lock = 1;
3010 int count;
984263bc 3011
a108bf71 3012 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
3013Retry:
3014 if (use_read_lock)
3015 vm_map_lock_read(map);
3016 else
3017 vm_map_lock(map);
3018
3019 /* If addr is already in the entry range, no need to grow.*/
3020 if (vm_map_lookup_entry(map, addr, &prev_entry))
3021 goto done;
3022
3023 if ((stack_entry = prev_entry->next) == &map->header)
3024 goto done;
3025 if (prev_entry == &map->header)
afeabdca 3026 end = stack_entry->start - stack_entry->aux.avail_ssize;
984263bc
MD
3027 else
3028 end = prev_entry->end;
3029
3030 /* This next test mimics the old grow function in vm_machdep.c.
3031 * It really doesn't quite make sense, but we do it anyway
3032 * for compatibility.
3033 *
3034 * If not growable stack, return success. This signals the
3035 * caller to proceed as he would normally with normal vm.
3036 */
afeabdca 3037 if (stack_entry->aux.avail_ssize < 1 ||
984263bc 3038 addr >= stack_entry->start ||
afeabdca 3039 addr < stack_entry->start - stack_entry->aux.avail_ssize) {
984263bc
MD
3040 goto done;
3041 }
3042
3043 /* Find the minimum grow amount */
3044 grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
afeabdca 3045 if (grow_amount > stack_entry->aux.avail_ssize) {
984263bc
MD
3046 rv = KERN_NO_SPACE;
3047 goto done;
3048 }
3049
3050 /* If there is no longer enough space between the entries
3051 * nogo, and adjust the available space. Note: this
3052 * should only happen if the user has mapped into the
3053 * stack area after the stack was created, and is
3054 * probably an error.
3055 *
3056 * This also effectively destroys any guard page the user
3057 * might have intended by limiting the stack size.
3058 */
3059 if (grow_amount > stack_entry->start - end) {
3060 if (use_read_lock && vm_map_lock_upgrade(map)) {
3061 use_read_lock = 0;
3062 goto Retry;
3063 }
3064 use_read_lock = 0;
afeabdca 3065 stack_entry->aux.avail_ssize = stack_entry->start - end;
984263bc
MD
3066 rv = KERN_NO_SPACE;
3067 goto done;
3068 }
3069
3070 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
3071
3072 /* If this is the main process stack, see if we're over the
3073 * stack limit.
3074 */
3075 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
3076 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
3077 rv = KERN_NO_SPACE;
3078 goto done;
3079 }
3080
3081 /* Round up the grow amount modulo SGROWSIZ */
3082 grow_amount = roundup (grow_amount, sgrowsiz);
afeabdca
MD
3083 if (grow_amount > stack_entry->aux.avail_ssize) {
3084 grow_amount = stack_entry->aux.avail_ssize;
984263bc
MD
3085 }
3086 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
3087 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
3088 grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
3089 ctob(vm->vm_ssize);
3090 }
3091
3092 /* If we would blow our VMEM resource limit, no go */
be77b5f9 3093 if (map->size + grow_amount > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
984263bc
MD
3094 rv = KERN_NO_SPACE;
3095 goto done;
3096 }
3097
3098 if (use_read_lock && vm_map_lock_upgrade(map)) {
3099 use_read_lock = 0;
3100 goto Retry;
3101 }
3102 use_read_lock = 0;
3103
3104 /* Get the preliminary new entry start value */
3105 addr = stack_entry->start - grow_amount;
3106
3107 /* If this puts us into the previous entry, cut back our growth
3108 * to the available space. Also, see the note above.
3109 */
3110 if (addr < end) {
afeabdca 3111 stack_entry->aux.avail_ssize = stack_entry->start - end;
984263bc
MD
3112 addr = end;
3113 }
3114
a108bf71
MD
3115 rv = vm_map_insert(map, &count,
3116 NULL, 0, addr, stack_entry->start,
1b874851
MD
3117 VM_MAPTYPE_NORMAL,
3118 VM_PROT_ALL, VM_PROT_ALL,
984263bc
MD
3119 0);
3120
3121 /* Adjust the available stack space by the amount we grew. */
3122 if (rv == KERN_SUCCESS) {
3123 if (prev_entry != &map->header)
a108bf71 3124 vm_map_clip_end(map, prev_entry, addr, &count);
984263bc
MD
3125 new_stack_entry = prev_entry->next;
3126 if (new_stack_entry->end != stack_entry->start ||
3127 new_stack_entry->start != addr)
3128 panic ("Bad stack grow start/end in new stack entry");
3129 else {
afeabdca
MD
3130 new_stack_entry->aux.avail_ssize =
3131 stack_entry->aux.avail_ssize -
3132 (new_stack_entry->end - new_stack_entry->start);
984263bc
MD
3133 if (is_procstack)
3134 vm->vm_ssize += btoc(new_stack_entry->end -
3135 new_stack_entry->start);
3136 }
3137 }
3138
3139done:
3140 if (use_read_lock)
3141 vm_map_unlock_read(map);
3142 else
3143 vm_map_unlock(map);
a108bf71 3144 vm_map_entry_release(count);
984263bc
MD
3145 return (rv);
3146}
3147
3148/*
3149 * Unshare the specified VM space for exec. If other processes are
3150 * mapped to it, then create a new one. The new vmspace is null.
3151 */
3152
3153void
29802dbb 3154vmspace_exec(struct proc *p, struct vmspace *vmcopy)
a108bf71 3155{
984263bc
MD
3156 struct vmspace *oldvmspace = p->p_vmspace;
3157 struct vmspace *newvmspace;
3158 vm_map_t map = &p->p_vmspace->vm_map;
3159
29802dbb
MD
3160 /*
3161 * If we are execing a resident vmspace we fork it, otherwise
239b4df9
MD
3162 * we create a new vmspace. Note that exitingcnt and upcalls
3163 * are not copied to the new vmspace.
29802dbb
MD
3164 */
3165 if (vmcopy) {
3166 newvmspace = vmspace_fork(vmcopy);
3167 } else {
3168 newvmspace = vmspace_alloc(map->min_offset, map->max_offset);
3169 bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
239b4df9
MD
3170 (caddr_t)&oldvmspace->vm_endcopy -
3171 (caddr_t)&oldvmspace->vm_startcopy);
29802dbb
MD
3172 }
3173
984263bc
MD
3174 /*
3175 * This code is written like this for prototype purposes. The
3176 * goal is to avoid running down the vmspace here, but let the
3177 * other process's that are still using the vmspace to finally
3178 * run it down. Even though there is little or no chance of blocking
3179 * here, it is a good idea to keep this form for future mods.
3180 */
984263bc
MD
3181 p->p_vmspace = newvmspace;
3182 pmap_pinit2(vmspace_pmap(newvmspace));
3183 if (p == curproc)
3184 pmap_activate(p);
29802dbb 3185 vmspace_free(oldvmspace);
984263bc
MD
3186}
3187
3188/*
3189 * Unshare the specified VM space for forcing COW. This
3190 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
239b4df9
MD
3191 *
3192 * The exitingcnt test is not strictly necessary but has been
3193 * included for code sanity (to make the code a bit more deterministic).
984263bc
MD
3194 */
3195
3196void
a108bf71
MD
3197vmspace_unshare(struct proc *p)
3198{
984263bc
MD
3199 struct vmspace *oldvmspace = p->p_vmspace;
3200 struct vmspace *newvmspace;
3201
239b4df9 3202 if (oldvmspace->vm_refcnt == 1 && oldvmspace->vm_exitingcnt == 0)
984263bc
MD
3203 return;
3204 newvmspace = vmspace_fork(oldvmspace);
984263bc
MD
3205 p->p_vmspace = newvmspace;
3206 pmap_pinit2(vmspace_pmap(newvmspace));
3207 if (p == curproc)
3208 pmap_activate(p);
29802dbb 3209 vmspace_free(oldvmspace);
984263bc 3210}
984263bc
MD
3211
3212/*
3213 * vm_map_lookup:
3214 *
3215 * Finds the VM object, offset, and
3216 * protection for a given virtual address in the
3217 * specified map, assuming a page fault of the
3218 * type specified.
3219 *
3220 * Leaves the map in question locked for read; return
3221 * values are guaranteed until a vm_map_lookup_done
3222 * call is performed. Note that the map argument
3223 * is in/out; the returned map must be used in
3224 * the call to vm_map_lookup_done.
3225 *
3226 * A handle (out_entry) is returned for use in
3227 * vm_map_lookup_done, to make that fast.
3228 *
3229 * If a lookup is requested with "write protection"
3230 * specified, the map may be changed to perform virtual
3231 * copying operations, although the data referenced will
3232 * remain the same.
3233 */
3234int
3235vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3236 vm_offset_t vaddr,
3237 vm_prot_t fault_typea,
3238 vm_map_entry_t *out_entry, /* OUT */
3239 vm_object_t *object, /* OUT */
3240 vm_pindex_t *pindex, /* OUT */
3241 vm_prot_t *out_prot, /* OUT */
3242 boolean_t *wired) /* OUT */
3243{
3244 vm_map_entry_t entry;
3245 vm_map_t map = *var_map;
3246 vm_prot_t prot;
3247 vm_prot_t fault_type = fault_typea;
3248 int use_read_lock = 1;
3249 int rv = KERN_SUCCESS;
3250
3251RetryLookup:
3252 if (use_read_lock)
3253 vm_map_lock_read(map);
3254 else
3255 vm_map_lock(map);
3256
3257 /*
3258 * If the map has an interesting hint, try it before calling full
3259 * blown lookup routine.
3260 */
3261 entry = map->hint;
3262 *out_entry = entry;
3263
3264 if ((entry == &map->header) ||
3265 (vaddr < entry->start) || (vaddr >= entry->end)) {
3266 vm_map_entry_t tmp_entry;
3267
3268 /*
3269 * Entry was either not a valid hint, or the vaddr was not
3270 * contained in the entry, so do a full lookup.
3271 */
3272 if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) {
3273 rv = KERN_INVALID_ADDRESS;
3274 goto done;
3275 }
3276
3277 entry = tmp_entry;
3278 *out_entry = entry;
3279 }
3280
3281 /*
3282 * Handle submaps.
3283 */
1b874851 3284 if (entry->maptype == VM_MAPTYPE_SUBMAP) {
984263bc
MD
3285 vm_map_t old_map = map;
3286
3287 *var_map = map = entry->object.sub_map;
3288 if (use_read_lock)
3289 vm_map_unlock_read(old_map);
3290 else
3291 vm_map_unlock(old_map);
3292 use_read_lock = 1;
3293 goto RetryLookup;
3294 }
3295
3296 /*
3297 * Check whether this task is allowed to have this page.
3298 * Note the special case for MAP_ENTRY_COW
3299 * pages with an override. This is to implement a forced
3300 * COW for debuggers.
3301 */
3302
3303 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3304 prot = entry->max_protection;
3305 else
3306 prot = entry->protection;
3307
3308 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3309 if ((fault_type & prot) != fault_type) {
3310 rv = KERN_PROTECTION_FAILURE;
3311 goto done;
3312 }
3313
3314 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3315 (entry->eflags & MAP_ENTRY_COW) &&
3316 (fault_type & VM_PROT_WRITE) &&
3317 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3318 rv = KERN_PROTECTION_FAILURE;
3319 goto done;
3320 }
3321
3322 /*
3323 * If this page is not pageable, we have to get it for all possible
3324 * accesses.
3325 */
984263bc
MD
3326 *wired = (entry->wired_count != 0);
3327 if (*wired)
3328 prot = fault_type = entry->protection;
3329
3330 /*
568e6804
MD
3331 * Virtual page tables may need to update the accessed (A) bit
3332 * in a page table entry. Upgrade the fault to a write fault for
3333 * that case if the map will support it. If the map does not support
3334 * it the page table entry simply will not be updated.
984263bc 3335 */
568e6804
MD
3336 if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
3337 if (prot & VM_PROT_WRITE)
3338 fault_type |= VM_PROT_WRITE;
3339 }
984263bc 3340
568e6804
MD
3341 /*
3342 * If the entry was copy-on-write, we either ...
3343 */
984263bc
MD
3344 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3345 /*
3346 * If we want to write the page, we may as well handle that
3347 * now since we've got the map locked.
3348 *
3349 * If we don't need to write the page, we just demote the
3350 * permissions allowed.
3351 */
3352
3353 if (fault_type & VM_PROT_WRITE) {
3354 /*
3355 * Make a new object, and place it in the object
3356 * chain. Note that no new references have appeared
3357 * -- one just moved from the map to the new
3358 * object.
3359 */
3360
3361 if (use_read_lock && vm_map_lock_upgrade(map)) {
3362 use_read_lock = 0;
3363 goto RetryLookup;
3364 }
3365 use_read_lock = 0;
3366
53025830 3367 vm_map_entry_shadow(entry);
984263bc
MD
3368 } else {
3369 /*
3370 * We're attempting to read a copy-on-write page --
3371 * don't allow writes.
3372 */
3373
3374 prot &= ~VM_PROT_WRITE;
3375 }
3376 }
3377
3378 /*
3379 * Create an object if necessary.
3380 */
3381 if (entry->object.vm_object == NULL &&
3382 !map->system_map) {
3383 if (use_read_lock && vm_map_lock_upgrade(map)) {
3384 use_read_lock = 0;
3385 goto RetryLookup;
3386 }
3387 use_read_lock = 0;
53025830 3388 vm_map_entry_allocate_object(entry);
984263bc
MD
3389 }
3390
3391 /*
3392 * Return the object/offset from this entry. If the entry was
3393 * copy-on-write or empty, it has been fixed up.
3394 */
3395
3396 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3397 *object = entry->object.vm_object;
3398
3399 /*
3400 * Return whether this is the only map sharing this data. On
3401 * success we return with a read lock held on the map. On failure
3402 * we return with the map unlocked.
3403 */
3404 *out_prot = prot;
3405done:
3406 if (rv == KERN_SUCCESS) {
3407 if (use_read_lock == 0)
3408 vm_map_lock_downgrade(map);
3409 } else if (use_read_lock) {
3410 vm_map_unlock_read(map);
3411 } else {
3412 vm_map_unlock(map);
3413 }
3414 return (rv);
3415}
3416
3417/*
3418 * vm_map_lookup_done:
3419 *
3420 * Releases locks acquired by a vm_map_lookup
3421 * (according to the handle returned by that lookup).
3422 */
3423
3424void
a108bf71 3425vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry, int count)
984263bc
MD
3426{
3427 /*
3428 * Unlock the main-level map
3429 */
984263bc 3430 vm_map_unlock_read(map);
a108bf71
MD
3431 if (count)
3432 vm_map_entry_release(count);
984263bc
MD
3433}
3434
984263bc
MD
3435#include "opt_ddb.h"
3436#ifdef DDB
3437#include <sys/kernel.h>
3438
3439#include <ddb/ddb.h>
3440
3441/*
3442 * vm_map_print: [ debug ]
3443 */
3444DB_SHOW_COMMAND(map, vm_map_print)
3445{
3446 static int nlines;
3447 /* XXX convert args. */
3448 vm_map_t map = (vm_map_t)addr;
3449 boolean_t full = have_addr;
3450
3451 vm_map_entry_t entry;
3452
3453 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3454 (void *)map,
3455 (void *)map->pmap, map->nentries, map->timestamp);
3456 nlines++;
3457
3458 if (!full && db_indent)
3459 return;
3460
3461 db_indent += 2;
3462 for (entry = map->header.next; entry != &map->header;
3463 entry = entry->next) {
3464 db_iprintf("map entry %p: start=%p, end=%p\n",
3465 (void *)entry, (void *)entry->start, (void *)entry->end);
3466 nlines++;
3467 {
3468 static char *inheritance_name[4] =
3469 {"share", "copy", "none", "donate_copy"};
3470
3471 db_iprintf(" prot=%x/%x/%s",
3472 entry->protection,
3473 entry->max_protection,
3474 inheritance_name[(int)(unsigned char)entry->inheritance]);
3475 if (entry->wired_count != 0)
3476 db_printf(", wired");
3477 }
1b874851 3478 if (entry->maptype == VM_MAPTYPE_SUBMAP) {
984263bc
MD
3479 /* XXX no %qd in kernel. Truncate entry->offset. */
3480 db_printf(", share=%p, offset=0x%lx\n",
3481 (void *)entry->object.sub_map,
3482 (long)entry->offset);
3483 nlines++;
3484 if ((entry->prev == &map->header) ||
3485 (entry->prev->object.sub_map !=
3486 entry->object.sub_map)) {
3487 db_indent += 2;
3488 vm_map_print((db_expr_t)(intptr_t)
3489 entry->object.sub_map,
3490 full, 0, (char *)0);
3491 db_indent -= 2;
3492 }
3493 } else {
3494 /* XXX no %qd in kernel. Truncate entry->offset. */
3495 db_printf(", object=%p, offset=0x%lx",
3496 (void *)entry->object.vm_object,
3497 (long)entry->offset);
3498 if (entry->eflags & MAP_ENTRY_COW)
3499 db_printf(", copy (%s)",
3500 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3501 db_printf("\n");
3502 nlines++;
3503
3504 if ((entry->prev == &map->header) ||
3505 (entry->prev->object.vm_object !=
3506 entry->object.vm_object)) {
3507 db_indent += 2;
3508 vm_object_print((db_expr_t)(intptr_t)
3509 entry->object.vm_object,
3510 full, 0, (char *)0);
3511 nlines += 4;
3512 db_indent -= 2;
3513 }
3514 }
3515 }
3516 db_indent -= 2;
3517 if (db_indent == 0)
3518 nlines = 0;
3519}
3520
3521
3522DB_SHOW_COMMAND(procvm, procvm)
3523{
3524 struct proc *p;
3525
3526 if (have_addr) {
3527 p = (struct proc *) addr;
3528 } else {
3529 p = curproc;
3530 }
3531
3532 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3533 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3534 (void *)vmspace_pmap(p->p_vmspace));
3535
3536 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3537}
3538
3539#endif /* DDB */