Add the "wordexp" shell built-in command.
[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 $
e4846942 65 * $DragonFly: src/sys/vm/vm_map.c,v 1.54 2006/12/28 21:24:02 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);
54a764e8
MD
198 bzero(&vm->vm_startcopy,
199 (char *)&vm->vm_endcopy - (char *)&vm->vm_startcopy);
e4846942 200 vm_map_init(&vm->vm_map, min, max, NULL);
984263bc
MD
201 pmap_pinit(vmspace_pmap(vm));
202 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
203 vm->vm_refcnt = 1;
204 vm->vm_shm = NULL;
205 vm->vm_exitingcnt = 0;
206 return (vm);
207}
208
209void
a108bf71
MD
210vm_init2(void)
211{
c4ae567f
MD
212 zinitna(mapentzone, &mapentobj, NULL, 0, 0,
213 ZONE_USE_RESERVE | ZONE_SPECIAL, 1);
a108bf71 214 zinitna(mapzone, &mapobj, NULL, 0, 0, 0, 1);
984263bc
MD
215 vmspace_zone = zinit("VMSPACE", sizeof (struct vmspace), 0, 0, 3);
216 pmap_init2();
217 vm_object_init2();
218}
219
220static __inline void
221vmspace_dofree(struct vmspace *vm)
222{
a108bf71
MD
223 int count;
224
fef0fdf2
MD
225 /*
226 * Make sure any SysV shm is freed, it might not have in
227 * exit1()
228 */
229 shmexit(vm);
230
a722be49
MD
231 KKASSERT(vm->vm_upcalls == NULL);
232
984263bc
MD
233 /*
234 * Lock the map, to wait out all other references to it.
235 * Delete all of the mappings and pages they hold, then call
236 * the pmap module to reclaim anything left.
237 */
a108bf71 238 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc 239 vm_map_lock(&vm->vm_map);
a108bf71
MD
240 vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
241 vm->vm_map.max_offset, &count);
984263bc 242 vm_map_unlock(&vm->vm_map);
a108bf71 243 vm_map_entry_release(count);
984263bc
MD
244
245 pmap_release(vmspace_pmap(vm));
246 zfree(vmspace_zone, vm);
247}
248
249void
250vmspace_free(struct vmspace *vm)
251{
252 if (vm->vm_refcnt == 0)
253 panic("vmspace_free: attempt to free already freed vmspace");
254
255 if (--vm->vm_refcnt == 0 && vm->vm_exitingcnt == 0)
256 vmspace_dofree(vm);
257}
258
259void
260vmspace_exitfree(struct proc *p)
261{
262 struct vmspace *vm;
263
264 vm = p->p_vmspace;
265 p->p_vmspace = NULL;
266
267 /*
268 * cleanup by parent process wait()ing on exiting child. vm_refcnt
269 * may not be 0 (e.g. fork() and child exits without exec()ing).
270 * exitingcnt may increment above 0 and drop back down to zero
271 * several times while vm_refcnt is held non-zero. vm_refcnt
272 * may also increment above 0 and drop back down to zero several
273 * times while vm_exitingcnt is held non-zero.
274 *
275 * The last wait on the exiting child's vmspace will clean up
276 * the remainder of the vmspace.
277 */
278 if (--vm->vm_exitingcnt == 0 && vm->vm_refcnt == 0)
279 vmspace_dofree(vm);
280}
281
282/*
283 * vmspace_swap_count() - count the approximate swap useage in pages for a
284 * vmspace.
285 *
286 * Swap useage is determined by taking the proportional swap used by
287 * VM objects backing the VM map. To make up for fractional losses,
288 * if the VM object has any swap use at all the associated map entries
289 * count for at least 1 swap page.
290 */
291int
292vmspace_swap_count(struct vmspace *vmspace)
293{
294 vm_map_t map = &vmspace->vm_map;
295 vm_map_entry_t cur;
1b874851 296 vm_object_t object;
984263bc 297 int count = 0;
1b874851 298 int n;
984263bc
MD
299
300 for (cur = map->header.next; cur != &map->header; cur = cur->next) {
1b874851
MD
301 switch(cur->maptype) {
302 case VM_MAPTYPE_NORMAL:
303 case VM_MAPTYPE_VPAGETABLE:
304 if ((object = cur->object.vm_object) == NULL)
305 break;
306 if (object->type != OBJT_SWAP)
307 break;
308 n = (cur->end - cur->start) / PAGE_SIZE;
984263bc
MD
309 if (object->un_pager.swp.swp_bcount) {
310 count += object->un_pager.swp.swp_bcount *
311 SWAP_META_PAGES * n / object->size + 1;
312 }
1b874851
MD
313 break;
314 default:
315 break;
984263bc
MD
316 }
317 }
318 return(count);
319}
320
321
322/*
323 * vm_map_create:
324 *
325 * Creates and returns a new empty VM map with
326 * the given physical map structure, and having
327 * the given lower and upper address bounds.
328 */
329vm_map_t
e4846942 330vm_map_create(vm_map_t result, pmap_t pmap, vm_offset_t min, vm_offset_t max)
984263bc 331{
e4846942
MD
332 if (result == NULL)
333 result = zalloc(mapzone);
334 vm_map_init(result, min, max, pmap);
984263bc
MD
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
e4846942 344vm_map_init(struct vm_map *map, vm_offset_t min, vm_offset_t max, pmap_t pmap)
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;
e4846942 354 map->pmap = pmap;
984263bc
MD
355 map->first_free = &map->header;
356 map->hint = &map->header;
357 map->timestamp = 0;
ab6f251b 358 lockinit(&map->lock, "thrd_sleep", 0, 0);
984263bc
MD
359}
360
53025830
MD
361/*
362 * Shadow the vm_map_entry's object. This typically needs to be done when
363 * a write fault is taken on an entry which had previously been cloned by
364 * fork(). The shared object (which might be NULL) must become private so
365 * we add a shadow layer above it.
366 *
367 * Object allocation for anonymous mappings is defered as long as possible.
368 * When creating a shadow, however, the underlying object must be instantiated
369 * so it can be shared.
370 *
371 * If the map segment is governed by a virtual page table then it is
372 * possible to address offsets beyond the mapped area. Just allocate
373 * a maximally sized object for this case.
374 */
375static
376void
377vm_map_entry_shadow(vm_map_entry_t entry)
378{
379 if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
380 vm_object_shadow(&entry->object.vm_object, &entry->offset,
381 0x7FFFFFFF); /* XXX */
382 } else {
383 vm_object_shadow(&entry->object.vm_object, &entry->offset,
384 atop(entry->end - entry->start));
385 }
386 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
387}
388
389/*
390 * Allocate an object for a vm_map_entry.
391 *
392 * Object allocation for anonymous mappings is defered as long as possible.
393 * This function is called when we can defer no longer, generally when a map
394 * entry might be split or forked or takes a page fault.
395 *
396 * If the map segment is governed by a virtual page table then it is
397 * possible to address offsets beyond the mapped area. Just allocate
398 * a maximally sized object for this case.
399 */
400void
401vm_map_entry_allocate_object(vm_map_entry_t entry)
402{
403 vm_object_t obj;
404
405 if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
406 obj = vm_object_allocate(OBJT_DEFAULT, 0x7FFFFFFF); /* XXX */
407 } else {
408 obj = vm_object_allocate(OBJT_DEFAULT,
409 atop(entry->end - entry->start));
410 }
411 entry->object.vm_object = obj;
412 entry->offset = 0;
413}
414
41a01a4d 415/*
c4ae567f 416 * vm_map_entry_reserve_cpu_init:
41a01a4d
MD
417 *
418 * Set an initial negative count so the first attempt to reserve
c4ae567f
MD
419 * space preloads a bunch of vm_map_entry's for this cpu. Also
420 * pre-allocate 2 vm_map_entries which will be needed by zalloc() to
421 * map a new page for vm_map_entry structures. SMP systems are
422 * particularly sensitive.
423 *
424 * This routine is called in early boot so we cannot just call
41a01a4d
MD
425 * vm_map_entry_reserve().
426 *
c4ae567f
MD
427 * May be called for a gd other then mycpu, but may only be called
428 * during early boot.
41a01a4d
MD
429 */
430void
431vm_map_entry_reserve_cpu_init(globaldata_t gd)
432{
c4ae567f
MD
433 vm_map_entry_t entry;
434 int i;
435
41a01a4d 436 gd->gd_vme_avail -= MAP_RESERVE_COUNT * 2;
c4ae567f
MD
437 entry = &cpu_map_entry_init[gd->gd_cpuid][0];
438 for (i = 0; i < VMEPERCPU; ++i, ++entry) {
439 entry->next = gd->gd_vme_base;
440 gd->gd_vme_base = entry;
441 }
41a01a4d
MD
442}
443
a108bf71
MD
444/*
445 * vm_map_entry_reserve:
446 *
e1359933
MD
447 * Reserves vm_map_entry structures so code later on can manipulate
448 * map_entry structures within a locked map without blocking trying
449 * to allocate a new vm_map_entry.
a108bf71
MD
450 */
451int
452vm_map_entry_reserve(int count)
453{
454 struct globaldata *gd = mycpu;
455 vm_map_entry_t entry;
456
457 crit_enter();
a108bf71
MD
458
459 /*
460 * Make sure we have enough structures in gd_vme_base to handle
461 * the reservation request.
462 */
ac13eccd 463 while (gd->gd_vme_avail < count) {
a108bf71
MD
464 entry = zalloc(mapentzone);
465 entry->next = gd->gd_vme_base;
466 gd->gd_vme_base = entry;
467 ++gd->gd_vme_avail;
468 }
ac13eccd 469 gd->gd_vme_avail -= count;
a108bf71
MD
470 crit_exit();
471 return(count);
472}
473
474/*
475 * vm_map_entry_release:
476 *
477 * Releases previously reserved vm_map_entry structures that were not
478 * used. If we have too much junk in our per-cpu cache clean some of
479 * it out.
480 */
481void
482vm_map_entry_release(int count)
483{
484 struct globaldata *gd = mycpu;
485 vm_map_entry_t entry;
486
487 crit_enter();
488 gd->gd_vme_avail += count;
489 while (gd->gd_vme_avail > MAP_RESERVE_SLOP) {
490 entry = gd->gd_vme_base;
491 KKASSERT(entry != NULL);
492 gd->gd_vme_base = entry->next;
493 --gd->gd_vme_avail;
494 crit_exit();
495 zfree(mapentzone, entry);
496 crit_enter();
497 }
498 crit_exit();
499}
500
501/*
502 * vm_map_entry_kreserve:
503 *
c4ae567f
MD
504 * Reserve map entry structures for use in kernel_map itself. These
505 * entries have *ALREADY* been reserved on a per-cpu basis when the map
506 * was inited. This function is used by zalloc() to avoid a recursion
507 * when zalloc() itself needs to allocate additional kernel memory.
508 *
509 * This function works like the normal reserve but does not load the
510 * vm_map_entry cache (because that would result in an infinite
511 * recursion). Note that gd_vme_avail may go negative. This is expected.
a108bf71 512 *
c4ae567f
MD
513 * Any caller of this function must be sure to renormalize after
514 * potentially eating entries to ensure that the reserve supply
515 * remains intact.
a108bf71
MD
516 */
517int
518vm_map_entry_kreserve(int count)
519{
520 struct globaldata *gd = mycpu;
521
522 crit_enter();
c4ae567f 523 gd->gd_vme_avail -= count;
a108bf71 524 crit_exit();
c4ae567f 525 KASSERT(gd->gd_vme_base != NULL, ("no reserved entries left, gd_vme_avail = %d\n", gd->gd_vme_avail));
a108bf71
MD
526 return(count);
527}
528
529/*
530 * vm_map_entry_krelease:
531 *
c4ae567f
MD
532 * Release previously reserved map entries for kernel_map. We do not
533 * attempt to clean up like the normal release function as this would
534 * cause an unnecessary (but probably not fatal) deep procedure call.
a108bf71
MD
535 */
536void
537vm_map_entry_krelease(int count)
538{
539 struct globaldata *gd = mycpu;
540
541 crit_enter();
c4ae567f 542 gd->gd_vme_avail += count;
a108bf71
MD
543 crit_exit();
544}
545
984263bc 546/*
8a8d5d85 547 * vm_map_entry_create: [ internal use only ]
984263bc 548 *
8a8d5d85 549 * Allocates a VM map entry for insertion. No entry fields are filled
a108bf71
MD
550 * in.
551 *
552 * This routine may be called from an interrupt thread but not a FAST
553 * interrupt. This routine may recurse the map lock.
984263bc 554 */
8a8d5d85 555static vm_map_entry_t
a108bf71 556vm_map_entry_create(vm_map_t map, int *countp)
984263bc 557{
a108bf71
MD
558 struct globaldata *gd = mycpu;
559 vm_map_entry_t entry;
8a8d5d85 560
a108bf71
MD
561 KKASSERT(*countp > 0);
562 --*countp;
563 crit_enter();
564 entry = gd->gd_vme_base;
565 KASSERT(entry != NULL, ("gd_vme_base NULL! count %d", *countp));
566 gd->gd_vme_base = entry->next;
567 crit_exit();
568 return(entry);
984263bc
MD
569}
570
571/*
8a8d5d85 572 * vm_map_entry_dispose: [ internal use only ]
984263bc 573 *
8a8d5d85
MD
574 * Dispose of a vm_map_entry that is no longer being referenced. This
575 * function may be called from an interrupt.
984263bc 576 */
8a8d5d85 577static void
a108bf71 578vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc 579{
a108bf71
MD
580 struct globaldata *gd = mycpu;
581
686dbf64
MD
582 KKASSERT(map->hint != entry);
583 KKASSERT(map->first_free != entry);
584
a108bf71
MD
585 ++*countp;
586 crit_enter();
587 entry->next = gd->gd_vme_base;
588 gd->gd_vme_base = entry;
589 crit_exit();
984263bc
MD
590}
591
8a8d5d85 592
984263bc
MD
593/*
594 * vm_map_entry_{un,}link:
595 *
596 * Insert/remove entries from maps.
597 */
598static __inline void
599vm_map_entry_link(vm_map_t map,
600 vm_map_entry_t after_where,
601 vm_map_entry_t entry)
602{
603 map->nentries++;
604 entry->prev = after_where;
605 entry->next = after_where->next;
606 entry->next->prev = entry;
607 after_where->next = entry;
0cd275af
MD
608 if (vm_map_rb_tree_RB_INSERT(&map->rb_root, entry))
609 panic("vm_map_entry_link: dup addr map %p ent %p", map, entry);
984263bc
MD
610}
611
612static __inline void
613vm_map_entry_unlink(vm_map_t map,
614 vm_map_entry_t entry)
615{
616 vm_map_entry_t prev;
617 vm_map_entry_t next;
618
619 if (entry->eflags & MAP_ENTRY_IN_TRANSITION)
620 panic("vm_map_entry_unlink: attempt to mess with locked entry! %p", entry);
621 prev = entry->prev;
622 next = entry->next;
623 next->prev = prev;
624 prev->next = next;
686dbf64 625 vm_map_rb_tree_RB_REMOVE(&map->rb_root, entry);
984263bc
MD
626 map->nentries--;
627}
628
984263bc
MD
629/*
630 * vm_map_lookup_entry: [ internal use only ]
631 *
632 * Finds the map entry containing (or
633 * immediately preceding) the specified address
634 * in the given map; the entry is returned
635 * in the "entry" parameter. The boolean
636 * result indicates whether the address is
637 * actually contained in the map.
638 */
639boolean_t
57e43348
MD
640vm_map_lookup_entry(vm_map_t map, vm_offset_t address,
641 vm_map_entry_t *entry /* OUT */)
984263bc 642{
686dbf64 643 vm_map_entry_t tmp;
984263bc
MD
644 vm_map_entry_t last;
645
686dbf64 646#if 0
984263bc 647 /*
686dbf64
MD
648 * XXX TEMPORARILY DISABLED. For some reason our attempt to revive
649 * the hint code with the red-black lookup meets with system crashes
650 * and lockups. We do not yet know why.
651 *
652 * It is possible that the problem is related to the setting
653 * of the hint during map_entry deletion, in the code specified
654 * at the GGG comment later on in this file.
984263bc 655 */
686dbf64
MD
656 /*
657 * Quickly check the cached hint, there's a good chance of a match.
658 */
659 if (map->hint != &map->header) {
660 tmp = map->hint;
661 if (address >= tmp->start && address < tmp->end) {
662 *entry = tmp;
663 return(TRUE);
984263bc 664 }
984263bc 665 }
686dbf64 666#endif
984263bc
MD
667
668 /*
686dbf64
MD
669 * Locate the record from the top of the tree. 'last' tracks the
670 * closest prior record and is returned if no match is found, which
671 * in binary tree terms means tracking the most recent right-branch
672 * taken. If there is no prior record, &map->header is returned.
984263bc 673 */
686dbf64
MD
674 last = &map->header;
675 tmp = RB_ROOT(&map->rb_root);
676
677 while (tmp) {
678 if (address >= tmp->start) {
679 if (address < tmp->end) {
680 *entry = tmp;
681 map->hint = tmp;
682 return(TRUE);
984263bc 683 }
686dbf64
MD
684 last = tmp;
685 tmp = RB_RIGHT(tmp, rb_entry);
686 } else {
687 tmp = RB_LEFT(tmp, rb_entry);
984263bc 688 }
984263bc 689 }
686dbf64 690 *entry = last;
984263bc
MD
691 return (FALSE);
692}
693
694/*
695 * vm_map_insert:
696 *
697 * Inserts the given whole VM object into the target
698 * map at the specified address range. The object's
699 * size should match that of the address range.
700 *
a108bf71
MD
701 * Requires that the map be locked, and leaves it so. Requires that
702 * sufficient vm_map_entry structures have been reserved and tracks
703 * the use via countp.
984263bc
MD
704 *
705 * If object is non-NULL, ref count must be bumped by caller
706 * prior to making call to account for the new entry.
707 */
708int
a108bf71
MD
709vm_map_insert(vm_map_t map, int *countp,
710 vm_object_t object, vm_ooffset_t offset,
1b874851
MD
711 vm_offset_t start, vm_offset_t end,
712 vm_maptype_t maptype,
713 vm_prot_t prot, vm_prot_t max,
984263bc
MD
714 int cow)
715{
716 vm_map_entry_t new_entry;
717 vm_map_entry_t prev_entry;
718 vm_map_entry_t temp_entry;
719 vm_eflags_t protoeflags;
720
721 /*
722 * Check that the start and end points are not bogus.
723 */
724
725 if ((start < map->min_offset) || (end > map->max_offset) ||
726 (start >= end))
727 return (KERN_INVALID_ADDRESS);
728
729 /*
730 * Find the entry prior to the proposed starting address; if it's part
731 * of an existing entry, this range is bogus.
732 */
733
734 if (vm_map_lookup_entry(map, start, &temp_entry))
735 return (KERN_NO_SPACE);
736
737 prev_entry = temp_entry;
738
739 /*
740 * Assert that the next entry doesn't overlap the end point.
741 */
742
743 if ((prev_entry->next != &map->header) &&
744 (prev_entry->next->start < end))
745 return (KERN_NO_SPACE);
746
747 protoeflags = 0;
748
749 if (cow & MAP_COPY_ON_WRITE)
750 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
751
752 if (cow & MAP_NOFAULT) {
753 protoeflags |= MAP_ENTRY_NOFAULT;
754
755 KASSERT(object == NULL,
756 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
757 }
758 if (cow & MAP_DISABLE_SYNCER)
759 protoeflags |= MAP_ENTRY_NOSYNC;
760 if (cow & MAP_DISABLE_COREDUMP)
761 protoeflags |= MAP_ENTRY_NOCOREDUMP;
762
763 if (object) {
764 /*
765 * When object is non-NULL, it could be shared with another
766 * process. We have to set or clear OBJ_ONEMAPPING
767 * appropriately.
768 */
769 if ((object->ref_count > 1) || (object->shadow_count != 0)) {
770 vm_object_clear_flag(object, OBJ_ONEMAPPING);
771 }
772 }
773 else if ((prev_entry != &map->header) &&
774 (prev_entry->eflags == protoeflags) &&
775 (prev_entry->end == start) &&
776 (prev_entry->wired_count == 0) &&
1b874851 777 prev_entry->maptype == maptype &&
984263bc
MD
778 ((prev_entry->object.vm_object == NULL) ||
779 vm_object_coalesce(prev_entry->object.vm_object,
780 OFF_TO_IDX(prev_entry->offset),
781 (vm_size_t)(prev_entry->end - prev_entry->start),
782 (vm_size_t)(end - prev_entry->end)))) {
783 /*
784 * We were able to extend the object. Determine if we
785 * can extend the previous map entry to include the
786 * new range as well.
787 */
788 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
789 (prev_entry->protection == prot) &&
790 (prev_entry->max_protection == max)) {
791 map->size += (end - prev_entry->end);
792 prev_entry->end = end;
a108bf71 793 vm_map_simplify_entry(map, prev_entry, countp);
984263bc
MD
794 return (KERN_SUCCESS);
795 }
796
797 /*
798 * If we can extend the object but cannot extend the
799 * map entry, we have to create a new map entry. We
800 * must bump the ref count on the extended object to
801 * account for it. object may be NULL.
802 */
803 object = prev_entry->object.vm_object;
804 offset = prev_entry->offset +
805 (prev_entry->end - prev_entry->start);
806 vm_object_reference(object);
807 }
808
809 /*
810 * NOTE: if conditionals fail, object can be NULL here. This occurs
811 * in things like the buffer map where we manage kva but do not manage
812 * backing objects.
813 */
814
815 /*
816 * Create a new entry
817 */
818
a108bf71 819 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
820 new_entry->start = start;
821 new_entry->end = end;
822
1b874851 823 new_entry->maptype = maptype;
984263bc
MD
824 new_entry->eflags = protoeflags;
825 new_entry->object.vm_object = object;
826 new_entry->offset = offset;
afeabdca 827 new_entry->aux.master_pde = 0;
984263bc
MD
828
829 new_entry->inheritance = VM_INHERIT_DEFAULT;
830 new_entry->protection = prot;
831 new_entry->max_protection = max;
832 new_entry->wired_count = 0;
833
834 /*
835 * Insert the new entry into the list
836 */
837
838 vm_map_entry_link(map, prev_entry, new_entry);
839 map->size += new_entry->end - new_entry->start;
840
841 /*
842 * Update the free space hint
843 */
844 if ((map->first_free == prev_entry) &&
845 (prev_entry->end >= new_entry->start)) {
846 map->first_free = new_entry;
847 }
848
849#if 0
850 /*
851 * Temporarily removed to avoid MAP_STACK panic, due to
852 * MAP_STACK being a huge hack. Will be added back in
853 * when MAP_STACK (and the user stack mapping) is fixed.
854 */
855 /*
856 * It may be possible to simplify the entry
857 */
a108bf71 858 vm_map_simplify_entry(map, new_entry, countp);
984263bc
MD
859#endif
860
afeabdca
MD
861 /*
862 * Try to pre-populate the page table. Mappings governed by virtual
863 * page tables cannot be prepopulated without a lot of work, so
864 * don't try.
865 */
866 if ((cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) &&
867 maptype != VM_MAPTYPE_VPAGETABLE) {
083a7402 868 pmap_object_init_pt(map->pmap, start, prot,
984263bc
MD
869 object, OFF_TO_IDX(offset), end - start,
870 cow & MAP_PREFAULT_PARTIAL);
871 }
872
873 return (KERN_SUCCESS);
874}
875
876/*
877 * Find sufficient space for `length' bytes in the given map, starting at
878 * `start'. The map must be locked. Returns 0 on success, 1 on no space.
e9bb90e8
MD
879 *
880 * This function will returned an arbitrarily aligned pointer. If no
881 * particular alignment is required you should pass align as 1. Note that
882 * the map may return PAGE_SIZE aligned pointers if all the lengths used in
883 * the map are a multiple of PAGE_SIZE, even if you pass a smaller align
884 * argument.
885 *
886 * 'align' should be a power of 2 but is not required to be.
984263bc
MD
887 */
888int
e9bb90e8
MD
889vm_map_findspace(
890 vm_map_t map,
891 vm_offset_t start,
892 vm_size_t length,
893 vm_offset_t align,
894 vm_offset_t *addr)
984263bc
MD
895{
896 vm_map_entry_t entry, next;
897 vm_offset_t end;
e9bb90e8 898 vm_offset_t align_mask;
984263bc
MD
899
900 if (start < map->min_offset)
901 start = map->min_offset;
902 if (start > map->max_offset)
903 return (1);
904
e9bb90e8
MD
905 /*
906 * If the alignment is not a power of 2 we will have to use
907 * a mod/division, set align_mask to a special value.
908 */
909 if ((align | (align - 1)) + 1 != (align << 1))
910 align_mask = (vm_offset_t)-1;
911 else
912 align_mask = align - 1;
913
a108bf71 914retry:
984263bc
MD
915 /*
916 * Look for the first possible address; if there's already something
917 * at this address, we have to start after it.
918 */
919 if (start == map->min_offset) {
920 if ((entry = map->first_free) != &map->header)
921 start = entry->end;
922 } else {
923 vm_map_entry_t tmp;
924
925 if (vm_map_lookup_entry(map, start, &tmp))
926 start = tmp->end;
927 entry = tmp;
928 }
929
930 /*
931 * Look through the rest of the map, trying to fit a new region in the
932 * gap between existing regions, or after the very last region.
933 */
934 for (;; start = (entry = next)->end) {
e9bb90e8
MD
935 /*
936 * Adjust the proposed start by the requested alignment,
937 * be sure that we didn't wrap the address.
938 */
939 if (align_mask == (vm_offset_t)-1)
940 end = ((start + align - 1) / align) * align;
941 else
942 end = (start + align_mask) & ~align_mask;
943 if (end < start)
944 return (1);
945 start = end;
984263bc
MD
946 /*
947 * Find the end of the proposed new region. Be sure we didn't
e9bb90e8
MD
948 * go beyond the end of the map, or wrap around the address.
949 * Then check to see if this is the last entry or if the
950 * proposed end fits in the gap between this and the next
951 * entry.
984263bc
MD
952 */
953 end = start + length;
954 if (end > map->max_offset || end < start)
955 return (1);
956 next = entry->next;
957 if (next == &map->header || next->start >= end)
958 break;
959 }
686dbf64 960 map->hint = entry;
e4846942 961 if (map == &kernel_map) {
984263bc
MD
962 vm_offset_t ksize;
963 if ((ksize = round_page(start + length)) > kernel_vm_end) {
964 pmap_growkernel(ksize);
a108bf71 965 goto retry;
984263bc
MD
966 }
967 }
a108bf71 968 *addr = start;
984263bc
MD
969 return (0);
970}
971
972/*
973 * vm_map_find finds an unallocated region in the target address
974 * map with the given length. The search is defined to be
975 * first-fit from the specified address; the region found is
976 * returned in the same parameter.
977 *
978 * If object is non-NULL, ref count must be bumped by caller
979 * prior to making call to account for the new entry.
980 */
981int
982vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1b874851
MD
983 vm_offset_t *addr, vm_size_t length,
984 boolean_t find_space,
985 vm_maptype_t maptype,
986 vm_prot_t prot, vm_prot_t max,
987 int cow)
984263bc
MD
988{
989 vm_offset_t start;
03aa8d99 990 int result;
a108bf71 991 int count;
984263bc
MD
992
993 start = *addr;
994
a108bf71 995 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
996 vm_map_lock(map);
997 if (find_space) {
e9bb90e8 998 if (vm_map_findspace(map, start, length, 1, addr)) {
984263bc 999 vm_map_unlock(map);
a108bf71 1000 vm_map_entry_release(count);
984263bc
MD
1001 return (KERN_NO_SPACE);
1002 }
1003 start = *addr;
1004 }
a108bf71 1005 result = vm_map_insert(map, &count, object, offset,
1b874851
MD
1006 start, start + length,
1007 maptype,
1008 prot, max,
1009 cow);
984263bc 1010 vm_map_unlock(map);
a108bf71 1011 vm_map_entry_release(count);
984263bc 1012
984263bc
MD
1013 return (result);
1014}
1015
1016/*
1017 * vm_map_simplify_entry:
1018 *
1019 * Simplify the given map entry by merging with either neighbor. This
1020 * routine also has the ability to merge with both neighbors.
1021 *
1022 * The map must be locked.
1023 *
1024 * This routine guarentees that the passed entry remains valid (though
1025 * possibly extended). When merging, this routine may delete one or
1026 * both neighbors. No action is taken on entries which have their
1027 * in-transition flag set.
1028 */
1029void
a108bf71 1030vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc
MD
1031{
1032 vm_map_entry_t next, prev;
1033 vm_size_t prevsize, esize;
1034
1b874851 1035 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
12e4aaff 1036 ++mycpu->gd_cnt.v_intrans_coll;
984263bc
MD
1037 return;
1038 }
1039
1b874851
MD
1040 if (entry->maptype == VM_MAPTYPE_SUBMAP)
1041 return;
1042
984263bc
MD
1043 prev = entry->prev;
1044 if (prev != &map->header) {
1045 prevsize = prev->end - prev->start;
1046 if ( (prev->end == entry->start) &&
1b874851 1047 (prev->maptype == entry->maptype) &&
984263bc
MD
1048 (prev->object.vm_object == entry->object.vm_object) &&
1049 (!prev->object.vm_object ||
1050 (prev->offset + prevsize == entry->offset)) &&
1051 (prev->eflags == entry->eflags) &&
1052 (prev->protection == entry->protection) &&
1053 (prev->max_protection == entry->max_protection) &&
1054 (prev->inheritance == entry->inheritance) &&
1055 (prev->wired_count == entry->wired_count)) {
1056 if (map->first_free == prev)
1057 map->first_free = entry;
1058 if (map->hint == prev)
1059 map->hint = entry;
1060 vm_map_entry_unlink(map, prev);
1061 entry->start = prev->start;
1062 entry->offset = prev->offset;
1063 if (prev->object.vm_object)
1064 vm_object_deallocate(prev->object.vm_object);
a108bf71 1065 vm_map_entry_dispose(map, prev, countp);
984263bc
MD
1066 }
1067 }
1068
1069 next = entry->next;
1070 if (next != &map->header) {
1071 esize = entry->end - entry->start;
1072 if ((entry->end == next->start) &&
1b874851 1073 (next->maptype == entry->maptype) &&
984263bc
MD
1074 (next->object.vm_object == entry->object.vm_object) &&
1075 (!entry->object.vm_object ||
1076 (entry->offset + esize == next->offset)) &&
1077 (next->eflags == entry->eflags) &&
1078 (next->protection == entry->protection) &&
1079 (next->max_protection == entry->max_protection) &&
1080 (next->inheritance == entry->inheritance) &&
1081 (next->wired_count == entry->wired_count)) {
1082 if (map->first_free == next)
1083 map->first_free = entry;
1084 if (map->hint == next)
1085 map->hint = entry;
1086 vm_map_entry_unlink(map, next);
1087 entry->end = next->end;
1088 if (next->object.vm_object)
1089 vm_object_deallocate(next->object.vm_object);
a108bf71 1090 vm_map_entry_dispose(map, next, countp);
984263bc
MD
1091 }
1092 }
1093}
1094/*
1095 * vm_map_clip_start: [ internal use only ]
1096 *
1097 * Asserts that the given entry begins at or after
1098 * the specified address; if necessary,
1099 * it splits the entry into two.
1100 */
a108bf71 1101#define vm_map_clip_start(map, entry, startaddr, countp) \
984263bc
MD
1102{ \
1103 if (startaddr > entry->start) \
a108bf71 1104 _vm_map_clip_start(map, entry, startaddr, countp); \
984263bc
MD
1105}
1106
1107/*
1108 * This routine is called only when it is known that
1109 * the entry must be split.
1110 */
1111static void
a108bf71 1112_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start, int *countp)
984263bc
MD
1113{
1114 vm_map_entry_t new_entry;
1115
1116 /*
1117 * Split off the front portion -- note that we must insert the new
1118 * entry BEFORE this one, so that this entry has the specified
1119 * starting address.
1120 */
1121
a108bf71 1122 vm_map_simplify_entry(map, entry, countp);
984263bc
MD
1123
1124 /*
1125 * If there is no object backing this entry, we might as well create
1126 * one now. If we defer it, an object can get created after the map
1127 * is clipped, and individual objects will be created for the split-up
1128 * map. This is a bit of a hack, but is also about the best place to
1129 * put this improvement.
1130 */
984263bc 1131 if (entry->object.vm_object == NULL && !map->system_map) {
53025830 1132 vm_map_entry_allocate_object(entry);
984263bc
MD
1133 }
1134
a108bf71 1135 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
1136 *new_entry = *entry;
1137
1138 new_entry->end = start;
1139 entry->offset += (start - entry->start);
1140 entry->start = start;
1141
1142 vm_map_entry_link(map, entry->prev, new_entry);
1143
1b874851
MD
1144 switch(entry->maptype) {
1145 case VM_MAPTYPE_NORMAL:
1146 case VM_MAPTYPE_VPAGETABLE:
984263bc 1147 vm_object_reference(new_entry->object.vm_object);
1b874851
MD
1148 break;
1149 default:
1150 break;
984263bc
MD
1151 }
1152}
1153
1154/*
1155 * vm_map_clip_end: [ internal use only ]
1156 *
1157 * Asserts that the given entry ends at or before
1158 * the specified address; if necessary,
1159 * it splits the entry into two.
1160 */
1161
a108bf71 1162#define vm_map_clip_end(map, entry, endaddr, countp) \
984263bc
MD
1163{ \
1164 if (endaddr < entry->end) \
a108bf71 1165 _vm_map_clip_end(map, entry, endaddr, countp); \
984263bc
MD
1166}
1167
1168/*
1169 * This routine is called only when it is known that
1170 * the entry must be split.
1171 */
1172static void
a108bf71 1173_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end, int *countp)
984263bc
MD
1174{
1175 vm_map_entry_t new_entry;
1176
1177 /*
1178 * If there is no object backing this entry, we might as well create
1179 * one now. If we defer it, an object can get created after the map
1180 * is clipped, and individual objects will be created for the split-up
1181 * map. This is a bit of a hack, but is also about the best place to
1182 * put this improvement.
1183 */
1184
1185 if (entry->object.vm_object == NULL && !map->system_map) {
53025830 1186 vm_map_entry_allocate_object(entry);
984263bc
MD
1187 }
1188
1189 /*
1190 * Create a new entry and insert it AFTER the specified entry
1191 */
1192
a108bf71 1193 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
1194 *new_entry = *entry;
1195
1196 new_entry->start = entry->end = end;
1197 new_entry->offset += (end - entry->start);
1198
1199 vm_map_entry_link(map, entry, new_entry);
1200
1b874851
MD
1201 switch(entry->maptype) {
1202 case VM_MAPTYPE_NORMAL:
1203 case VM_MAPTYPE_VPAGETABLE:
984263bc 1204 vm_object_reference(new_entry->object.vm_object);
1b874851
MD
1205 break;
1206 default:
1207 break;
984263bc
MD
1208 }
1209}
1210
1211/*
1212 * VM_MAP_RANGE_CHECK: [ internal use only ]
1213 *
1214 * Asserts that the starting and ending region
1215 * addresses fall within the valid range of the map.
1216 */
1217#define VM_MAP_RANGE_CHECK(map, start, end) \
1218 { \
1219 if (start < vm_map_min(map)) \
1220 start = vm_map_min(map); \
1221 if (end > vm_map_max(map)) \
1222 end = vm_map_max(map); \
1223 if (start > end) \
1224 start = end; \
1225 }
1226
1227/*
1228 * vm_map_transition_wait: [ kernel use only ]
1229 *
1230 * Used to block when an in-transition collison occurs. The map
1231 * is unlocked for the sleep and relocked before the return.
1232 */
1233static
1234void
1235vm_map_transition_wait(vm_map_t map)
1236{
1237 vm_map_unlock(map);
377d4740 1238 tsleep(map, 0, "vment", 0);
984263bc
MD
1239 vm_map_lock(map);
1240}
1241
1242/*
1243 * CLIP_CHECK_BACK
1244 * CLIP_CHECK_FWD
1245 *
1246 * When we do blocking operations with the map lock held it is
1247 * possible that a clip might have occured on our in-transit entry,
1248 * requiring an adjustment to the entry in our loop. These macros
1249 * help the pageable and clip_range code deal with the case. The
1250 * conditional costs virtually nothing if no clipping has occured.
1251 */
1252
1253#define CLIP_CHECK_BACK(entry, save_start) \
1254 do { \
1255 while (entry->start != save_start) { \
1256 entry = entry->prev; \
1257 KASSERT(entry != &map->header, ("bad entry clip")); \
1258 } \
1259 } while(0)
1260
1261#define CLIP_CHECK_FWD(entry, save_end) \
1262 do { \
1263 while (entry->end != save_end) { \
1264 entry = entry->next; \
1265 KASSERT(entry != &map->header, ("bad entry clip")); \
1266 } \
1267 } while(0)
1268
1269
1270/*
1271 * vm_map_clip_range: [ kernel use only ]
1272 *
1273 * Clip the specified range and return the base entry. The
1274 * range may cover several entries starting at the returned base
1275 * and the first and last entry in the covering sequence will be
1276 * properly clipped to the requested start and end address.
1277 *
1278 * If no holes are allowed you should pass the MAP_CLIP_NO_HOLES
1279 * flag.
1280 *
1281 * The MAP_ENTRY_IN_TRANSITION flag will be set for the entries
1282 * covered by the requested range.
1283 *
1284 * The map must be exclusively locked on entry and will remain locked
1285 * on return. If no range exists or the range contains holes and you
1286 * specified that no holes were allowed, NULL will be returned. This
1287 * routine may temporarily unlock the map in order avoid a deadlock when
1288 * sleeping.
1289 */
1290static
1291vm_map_entry_t
a108bf71
MD
1292vm_map_clip_range(vm_map_t map, vm_offset_t start, vm_offset_t end,
1293 int *countp, int flags)
984263bc
MD
1294{
1295 vm_map_entry_t start_entry;
1296 vm_map_entry_t entry;
1297
1298 /*
1299 * Locate the entry and effect initial clipping. The in-transition
1300 * case does not occur very often so do not try to optimize it.
1301 */
1302again:
1303 if (vm_map_lookup_entry(map, start, &start_entry) == FALSE)
1304 return (NULL);
1305 entry = start_entry;
1306 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1307 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
12e4aaff
MD
1308 ++mycpu->gd_cnt.v_intrans_coll;
1309 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
1310 vm_map_transition_wait(map);
1311 /*
1312 * entry and/or start_entry may have been clipped while
1313 * we slept, or may have gone away entirely. We have
1314 * to restart from the lookup.
1315 */
1316 goto again;
1317 }
1318 /*
1319 * Since we hold an exclusive map lock we do not have to restart
1320 * after clipping, even though clipping may block in zalloc.
1321 */
a108bf71
MD
1322 vm_map_clip_start(map, entry, start, countp);
1323 vm_map_clip_end(map, entry, end, countp);
984263bc
MD
1324 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1325
1326 /*
1327 * Scan entries covered by the range. When working on the next
1328 * entry a restart need only re-loop on the current entry which
1329 * we have already locked, since 'next' may have changed. Also,
1330 * even though entry is safe, it may have been clipped so we
1331 * have to iterate forwards through the clip after sleeping.
1332 */
1333 while (entry->next != &map->header && entry->next->start < end) {
1334 vm_map_entry_t next = entry->next;
1335
1336 if (flags & MAP_CLIP_NO_HOLES) {
1337 if (next->start > entry->end) {
1338 vm_map_unclip_range(map, start_entry,
a108bf71 1339 start, entry->end, countp, flags);
984263bc
MD
1340 return(NULL);
1341 }
1342 }
1343
1344 if (next->eflags & MAP_ENTRY_IN_TRANSITION) {
1345 vm_offset_t save_end = entry->end;
1346 next->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
12e4aaff
MD
1347 ++mycpu->gd_cnt.v_intrans_coll;
1348 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
1349 vm_map_transition_wait(map);
1350
1351 /*
1352 * clips might have occured while we blocked.
1353 */
1354 CLIP_CHECK_FWD(entry, save_end);
1355 CLIP_CHECK_BACK(start_entry, start);
1356 continue;
1357 }
1358 /*
1359 * No restart necessary even though clip_end may block, we
1360 * are holding the map lock.
1361 */
a108bf71 1362 vm_map_clip_end(map, next, end, countp);
984263bc
MD
1363 next->eflags |= MAP_ENTRY_IN_TRANSITION;
1364 entry = next;
1365 }
1366 if (flags & MAP_CLIP_NO_HOLES) {
1367 if (entry->end != end) {
1368 vm_map_unclip_range(map, start_entry,
a108bf71 1369 start, entry->end, countp, flags);
984263bc
MD
1370 return(NULL);
1371 }
1372 }
1373 return(start_entry);
1374}
1375
1376/*
1377 * vm_map_unclip_range: [ kernel use only ]
1378 *
1379 * Undo the effect of vm_map_clip_range(). You should pass the same
1380 * flags and the same range that you passed to vm_map_clip_range().
1381 * This code will clear the in-transition flag on the entries and
1382 * wake up anyone waiting. This code will also simplify the sequence
1383 * and attempt to merge it with entries before and after the sequence.
1384 *
1385 * The map must be locked on entry and will remain locked on return.
1386 *
1387 * Note that you should also pass the start_entry returned by
1388 * vm_map_clip_range(). However, if you block between the two calls
1389 * with the map unlocked please be aware that the start_entry may
1390 * have been clipped and you may need to scan it backwards to find
1391 * the entry corresponding with the original start address. You are
1392 * responsible for this, vm_map_unclip_range() expects the correct
1393 * start_entry to be passed to it and will KASSERT otherwise.
1394 */
1395static
1396void
1397vm_map_unclip_range(
1398 vm_map_t map,
1399 vm_map_entry_t start_entry,
1400 vm_offset_t start,
1401 vm_offset_t end,
a108bf71 1402 int *countp,
984263bc
MD
1403 int flags)
1404{
1405 vm_map_entry_t entry;
1406
1407 entry = start_entry;
1408
1409 KASSERT(entry->start == start, ("unclip_range: illegal base entry"));
1410 while (entry != &map->header && entry->start < end) {
1411 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, ("in-transition flag not set during unclip on: %p", entry));
1412 KASSERT(entry->end <= end, ("unclip_range: tail wasn't clipped"));
1413 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1414 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1415 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1416 wakeup(map);
1417 }
1418 entry = entry->next;
1419 }
1420
1421 /*
1422 * Simplification does not block so there is no restart case.
1423 */
1424 entry = start_entry;
1425 while (entry != &map->header && entry->start < end) {
a108bf71 1426 vm_map_simplify_entry(map, entry, countp);
984263bc
MD
1427 entry = entry->next;
1428 }
1429}
1430
1431/*
1432 * vm_map_submap: [ kernel use only ]
1433 *
1434 * Mark the given range as handled by a subordinate map.
1435 *
1436 * This range must have been created with vm_map_find,
1437 * and no other operations may have been performed on this
1438 * range prior to calling vm_map_submap.
1439 *
1440 * Only a limited number of operations can be performed
1441 * within this rage after calling vm_map_submap:
1442 * vm_fault
1443 * [Don't try vm_map_copy!]
1444 *
1445 * To remove a submapping, one must first remove the
1446 * range from the superior map, and then destroy the
1447 * submap (if desired). [Better yet, don't try it.]
1448 */
1449int
a108bf71 1450vm_map_submap(vm_map_t map, vm_offset_t start, vm_offset_t end, vm_map_t submap)
984263bc
MD
1451{
1452 vm_map_entry_t entry;
1453 int result = KERN_INVALID_ARGUMENT;
a108bf71 1454 int count;
984263bc 1455
a108bf71 1456 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1457 vm_map_lock(map);
1458
1459 VM_MAP_RANGE_CHECK(map, start, end);
1460
1461 if (vm_map_lookup_entry(map, start, &entry)) {
a108bf71 1462 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1463 } else {
1464 entry = entry->next;
1465 }
1466
a108bf71 1467 vm_map_clip_end(map, entry, end, &count);
984263bc
MD
1468
1469 if ((entry->start == start) && (entry->end == end) &&
1470 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1471 (entry->object.vm_object == NULL)) {
1472 entry->object.sub_map = submap;
1b874851 1473 entry->maptype = VM_MAPTYPE_SUBMAP;
984263bc
MD
1474 result = KERN_SUCCESS;
1475 }
1476 vm_map_unlock(map);
a108bf71 1477 vm_map_entry_release(count);
984263bc
MD
1478
1479 return (result);
1480}
1481
1482/*
1b874851 1483 * vm_map_protect:
984263bc 1484 *
1b874851
MD
1485 * Sets the protection of the specified address region in the target map.
1486 * If "set_max" is specified, the maximum protection is to be set;
1487 * otherwise, only the current protection is affected.
1488 *
1489 * The protection is not applicable to submaps, but is applicable to normal
1490 * maps and maps governed by virtual page tables. For example, when operating
1491 * on a virtual page table our protection basically controls how COW occurs
1492 * on the backing object, whereas the virtual page table abstraction itself
1493 * is an abstraction for userland.
984263bc
MD
1494 */
1495int
1496vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1497 vm_prot_t new_prot, boolean_t set_max)
1498{
1499 vm_map_entry_t current;
1500 vm_map_entry_t entry;
a108bf71 1501 int count;
984263bc 1502
a108bf71 1503 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1504 vm_map_lock(map);
1505
1506 VM_MAP_RANGE_CHECK(map, start, end);
1507
1508 if (vm_map_lookup_entry(map, start, &entry)) {
a108bf71 1509 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1510 } else {
1511 entry = entry->next;
1512 }
1513
1514 /*
1515 * Make a first pass to check for protection violations.
1516 */
984263bc
MD
1517 current = entry;
1518 while ((current != &map->header) && (current->start < end)) {
1b874851 1519 if (current->maptype == VM_MAPTYPE_SUBMAP) {
984263bc 1520 vm_map_unlock(map);
a108bf71 1521 vm_map_entry_release(count);
984263bc
MD
1522 return (KERN_INVALID_ARGUMENT);
1523 }
1524 if ((new_prot & current->max_protection) != new_prot) {
1525 vm_map_unlock(map);
a108bf71 1526 vm_map_entry_release(count);
984263bc
MD
1527 return (KERN_PROTECTION_FAILURE);
1528 }
1529 current = current->next;
1530 }
1531
1532 /*
1533 * Go back and fix up protections. [Note that clipping is not
1534 * necessary the second time.]
1535 */
984263bc
MD
1536 current = entry;
1537
1538 while ((current != &map->header) && (current->start < end)) {
1539 vm_prot_t old_prot;
1540
a108bf71 1541 vm_map_clip_end(map, current, end, &count);
984263bc
MD
1542
1543 old_prot = current->protection;
1b874851 1544 if (set_max) {
984263bc
MD
1545 current->protection =
1546 (current->max_protection = new_prot) &
1547 old_prot;
1b874851 1548 } else {
984263bc 1549 current->protection = new_prot;
1b874851 1550 }
984263bc
MD
1551
1552 /*
1553 * Update physical map if necessary. Worry about copy-on-write
1554 * here -- CHECK THIS XXX
1555 */
1556
1557 if (current->protection != old_prot) {
1558#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1559 VM_PROT_ALL)
1560
1561 pmap_protect(map->pmap, current->start,
1562 current->end,
1563 current->protection & MASK(current));
1564#undef MASK
1565 }
1566
a108bf71 1567 vm_map_simplify_entry(map, current, &count);
984263bc
MD
1568
1569 current = current->next;
1570 }
1571
1572 vm_map_unlock(map);
a108bf71 1573 vm_map_entry_release(count);
984263bc
MD
1574 return (KERN_SUCCESS);
1575}
1576
1577/*
1578 * vm_map_madvise:
1579 *
1580 * This routine traverses a processes map handling the madvise
1581 * system call. Advisories are classified as either those effecting
1582 * the vm_map_entry structure, or those effecting the underlying
1583 * objects.
afeabdca
MD
1584 *
1585 * The <value> argument is used for extended madvise calls.
984263bc 1586 */
984263bc 1587int
afeabdca
MD
1588vm_map_madvise(vm_map_t map, vm_offset_t start, vm_offset_t end,
1589 int behav, off_t value)
984263bc
MD
1590{
1591 vm_map_entry_t current, entry;
1592 int modify_map = 0;
afeabdca 1593 int error = 0;
a108bf71 1594 int count;
984263bc
MD
1595
1596 /*
1597 * Some madvise calls directly modify the vm_map_entry, in which case
1598 * we need to use an exclusive lock on the map and we need to perform
1599 * various clipping operations. Otherwise we only need a read-lock
1600 * on the map.
1601 */
1602
a108bf71
MD
1603 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1604
984263bc
MD
1605 switch(behav) {
1606 case MADV_NORMAL:
1607 case MADV_SEQUENTIAL:
1608 case MADV_RANDOM:
1609 case MADV_NOSYNC:
1610 case MADV_AUTOSYNC:
1611 case MADV_NOCORE:
1612 case MADV_CORE:
afeabdca
MD
1613 case MADV_SETMAP:
1614 case MADV_INVAL:
984263bc
MD
1615 modify_map = 1;
1616 vm_map_lock(map);
1617 break;
1618 case MADV_WILLNEED:
1619 case MADV_DONTNEED:
1620 case MADV_FREE:
1621 vm_map_lock_read(map);
1622 break;
1623 default:
a108bf71 1624 vm_map_entry_release(count);
afeabdca 1625 return (EINVAL);
984263bc
MD
1626 }
1627
1628 /*
1629 * Locate starting entry and clip if necessary.
1630 */
1631
1632 VM_MAP_RANGE_CHECK(map, start, end);
1633
1634 if (vm_map_lookup_entry(map, start, &entry)) {
1635 if (modify_map)
a108bf71 1636 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1637 } else {
1638 entry = entry->next;
1639 }
1640
1641 if (modify_map) {
1642 /*
1643 * madvise behaviors that are implemented in the vm_map_entry.
1644 *
1645 * We clip the vm_map_entry so that behavioral changes are
1646 * limited to the specified address range.
1647 */
1648 for (current = entry;
1649 (current != &map->header) && (current->start < end);
1650 current = current->next
1651 ) {
1b874851 1652 if (current->maptype == VM_MAPTYPE_SUBMAP)
984263bc
MD
1653 continue;
1654
a108bf71 1655 vm_map_clip_end(map, current, end, &count);
984263bc
MD
1656
1657 switch (behav) {
1658 case MADV_NORMAL:
1659 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1660 break;
1661 case MADV_SEQUENTIAL:
1662 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1663 break;
1664 case MADV_RANDOM:
1665 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1666 break;
1667 case MADV_NOSYNC:
1668 current->eflags |= MAP_ENTRY_NOSYNC;
1669 break;
1670 case MADV_AUTOSYNC:
1671 current->eflags &= ~MAP_ENTRY_NOSYNC;
1672 break;
1673 case MADV_NOCORE:
1674 current->eflags |= MAP_ENTRY_NOCOREDUMP;
1675 break;
1676 case MADV_CORE:
1677 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1678 break;
afeabdca
MD
1679 case MADV_INVAL:
1680 /*
1681 * Invalidate the related pmap entries, used
1682 * to flush portions of the real kernel's
1683 * pmap when the caller has removed or
1684 * modified existing mappings in a virtual
1685 * page table.
1686 */
1687 pmap_remove(map->pmap,
1688 current->start, current->end);
1689 break;
1690 case MADV_SETMAP:
1691 /*
1692 * Set the page directory page for a map
1693 * governed by a virtual page table. Mark
1694 * the entry as being governed by a virtual
1695 * page table if it is not.
1696 *
1697 * XXX the page directory page is stored
1698 * in the avail_ssize field if the map_entry.
1699 *
1700 * XXX the map simplification code does not
1701 * compare this field so weird things may
1702 * happen if you do not apply this function
1703 * to the entire mapping governed by the
1704 * virtual page table.
1705 */
1706 if (current->maptype != VM_MAPTYPE_VPAGETABLE) {
1707 error = EINVAL;
1708 break;
1709 }
1710 current->aux.master_pde = value;
1711 pmap_remove(map->pmap,
1712 current->start, current->end);
1713 break;
984263bc 1714 default:
afeabdca 1715 error = EINVAL;
984263bc
MD
1716 break;
1717 }
a108bf71 1718 vm_map_simplify_entry(map, current, &count);
984263bc
MD
1719 }
1720 vm_map_unlock(map);
1721 } else {
1722 vm_pindex_t pindex;
1723 int count;
1724
1725 /*
1726 * madvise behaviors that are implemented in the underlying
1727 * vm_object.
1728 *
1729 * Since we don't clip the vm_map_entry, we have to clip
1730 * the vm_object pindex and count.
1b874851
MD
1731 *
1732 * NOTE! We currently do not support these functions on
1733 * virtual page tables.
984263bc
MD
1734 */
1735 for (current = entry;
1736 (current != &map->header) && (current->start < end);
1737 current = current->next
1738 ) {
1739 vm_offset_t useStart;
1740
1b874851 1741 if (current->maptype != VM_MAPTYPE_NORMAL)
984263bc
MD
1742 continue;
1743
1744 pindex = OFF_TO_IDX(current->offset);
1745 count = atop(current->end - current->start);
1746 useStart = current->start;
1747
1748 if (current->start < start) {
1749 pindex += atop(start - current->start);
1750 count -= atop(start - current->start);
1751 useStart = start;
1752 }
1753 if (current->end > end)
1754 count -= atop(current->end - end);
1755
1756 if (count <= 0)
1757 continue;
1758
1759 vm_object_madvise(current->object.vm_object,
1760 pindex, count, behav);
afeabdca
MD
1761
1762 /*
1763 * Try to populate the page table. Mappings governed
1764 * by virtual page tables cannot be pre-populated
1765 * without a lot of work so don't try.
1766 */
1767 if (behav == MADV_WILLNEED &&
1768 current->maptype != VM_MAPTYPE_VPAGETABLE) {
984263bc
MD
1769 pmap_object_init_pt(
1770 map->pmap,
1771 useStart,
083a7402 1772 current->protection,
984263bc
MD
1773 current->object.vm_object,
1774 pindex,
1775 (count << PAGE_SHIFT),
1776 MAP_PREFAULT_MADVISE
1777 );
1778 }
1779 }
1780 vm_map_unlock_read(map);
1781 }
a108bf71 1782 vm_map_entry_release(count);
afeabdca 1783 return(error);
984263bc
MD
1784}
1785
1786
1787/*
1788 * vm_map_inherit:
1789 *
1790 * Sets the inheritance of the specified address
1791 * range in the target map. Inheritance
1792 * affects how the map will be shared with
1793 * child maps at the time of vm_map_fork.
1794 */
1795int
1796vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1797 vm_inherit_t new_inheritance)
1798{
1799 vm_map_entry_t entry;
1800 vm_map_entry_t temp_entry;
a108bf71 1801 int count;
984263bc
MD
1802
1803 switch (new_inheritance) {
1804 case VM_INHERIT_NONE:
1805 case VM_INHERIT_COPY:
1806 case VM_INHERIT_SHARE:
1807 break;
1808 default:
1809 return (KERN_INVALID_ARGUMENT);
1810 }
1811
a108bf71 1812 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1813 vm_map_lock(map);
1814
1815 VM_MAP_RANGE_CHECK(map, start, end);
1816
1817 if (vm_map_lookup_entry(map, start, &temp_entry)) {
1818 entry = temp_entry;
a108bf71 1819 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1820 } else
1821 entry = temp_entry->next;
1822
1823 while ((entry != &map->header) && (entry->start < end)) {
a108bf71 1824 vm_map_clip_end(map, entry, end, &count);
984263bc
MD
1825
1826 entry->inheritance = new_inheritance;
1827
a108bf71 1828 vm_map_simplify_entry(map, entry, &count);
984263bc
MD
1829
1830 entry = entry->next;
1831 }
984263bc 1832 vm_map_unlock(map);
a108bf71 1833 vm_map_entry_release(count);
984263bc
MD
1834 return (KERN_SUCCESS);
1835}
1836
1837/*
1838 * Implement the semantics of mlock
1839 */
1840int
57e43348
MD
1841vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t real_end,
1842 boolean_t new_pageable)
984263bc
MD
1843{
1844 vm_map_entry_t entry;
1845 vm_map_entry_t start_entry;
1846 vm_offset_t end;
1847 int rv = KERN_SUCCESS;
a108bf71 1848 int count;
984263bc 1849
a108bf71 1850 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1851 vm_map_lock(map);
1852 VM_MAP_RANGE_CHECK(map, start, real_end);
1853 end = real_end;
1854
a108bf71 1855 start_entry = vm_map_clip_range(map, start, end, &count, MAP_CLIP_NO_HOLES);
984263bc
MD
1856 if (start_entry == NULL) {
1857 vm_map_unlock(map);
a108bf71 1858 vm_map_entry_release(count);
984263bc
MD
1859 return (KERN_INVALID_ADDRESS);
1860 }
1861
1862 if (new_pageable == 0) {
1863 entry = start_entry;
1864 while ((entry != &map->header) && (entry->start < end)) {
1865 vm_offset_t save_start;
1866 vm_offset_t save_end;
1867
1868 /*
1869 * Already user wired or hard wired (trivial cases)
1870 */
1871 if (entry->eflags & MAP_ENTRY_USER_WIRED) {
1872 entry = entry->next;
1873 continue;
1874 }
1875 if (entry->wired_count != 0) {
1876 entry->wired_count++;
1877 entry->eflags |= MAP_ENTRY_USER_WIRED;
1878 entry = entry->next;
1879 continue;
1880 }
1881
1882 /*
1883 * A new wiring requires instantiation of appropriate
1884 * management structures and the faulting in of the
1885 * page.
1886 */
1b874851 1887 if (entry->maptype != VM_MAPTYPE_SUBMAP) {
984263bc
MD
1888 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1889 if (copyflag && ((entry->protection & VM_PROT_WRITE) != 0)) {
53025830 1890 vm_map_entry_shadow(entry);
984263bc
MD
1891 } else if (entry->object.vm_object == NULL &&
1892 !map->system_map) {
53025830 1893 vm_map_entry_allocate_object(entry);
984263bc
MD
1894 }
1895 }
1896 entry->wired_count++;
1897 entry->eflags |= MAP_ENTRY_USER_WIRED;
1898
1899 /*
f2d22ebf
MD
1900 * Now fault in the area. Note that vm_fault_wire()
1901 * may release the map lock temporarily, it will be
1902 * relocked on return. The in-transition
984263bc
MD
1903 * flag protects the entries.
1904 */
1905 save_start = entry->start;
1906 save_end = entry->end;
f2d22ebf 1907 rv = vm_fault_wire(map, entry, TRUE);
984263bc
MD
1908 if (rv) {
1909 CLIP_CHECK_BACK(entry, save_start);
1910 for (;;) {
1911 KASSERT(entry->wired_count == 1, ("bad wired_count on entry"));
1912 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1913 entry->wired_count = 0;
1914 if (entry->end == save_end)
1915 break;
1916 entry = entry->next;
1917 KASSERT(entry != &map->header, ("bad entry clip during backout"));
1918 }
1919 end = save_start; /* unwire the rest */
1920 break;
1921 }
1922 /*
1923 * note that even though the entry might have been
1924 * clipped, the USER_WIRED flag we set prevents
1925 * duplication so we do not have to do a
1926 * clip check.
1927 */
1928 entry = entry->next;
1929 }
1930
1931 /*
1932 * If we failed fall through to the unwiring section to
1933 * unwire what we had wired so far. 'end' has already
1934 * been adjusted.
1935 */
1936 if (rv)
1937 new_pageable = 1;
1938
1939 /*
1940 * start_entry might have been clipped if we unlocked the
1941 * map and blocked. No matter how clipped it has gotten
1942 * there should be a fragment that is on our start boundary.
1943 */
1944 CLIP_CHECK_BACK(start_entry, start);
1945 }
1946
1947 /*
1948 * Deal with the unwiring case.
1949 */
1950 if (new_pageable) {
1951 /*
1952 * This is the unwiring case. We must first ensure that the
1953 * range to be unwired is really wired down. We know there
1954 * are no holes.
1955 */
1956 entry = start_entry;
1957 while ((entry != &map->header) && (entry->start < end)) {
1958 if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
1959 rv = KERN_INVALID_ARGUMENT;
1960 goto done;
1961 }
1962 KASSERT(entry->wired_count != 0, ("wired count was 0 with USER_WIRED set! %p", entry));
1963 entry = entry->next;
1964 }
1965
1966 /*
1967 * Now decrement the wiring count for each region. If a region
1968 * becomes completely unwired, unwire its physical pages and
1969 * mappings.
1970 */
b4eddbac
DR
1971 /*
1972 * The map entries are processed in a loop, checking to
1973 * make sure the entry is wired and asserting it has a wired
1974 * count. However, another loop was inserted more-or-less in
1975 * the middle of the unwiring path. This loop picks up the
1976 * "entry" loop variable from the first loop without first
1977 * setting it to start_entry. Naturally, the secound loop
1978 * is never entered and the pages backing the entries are
1979 * never unwired. This can lead to a leak of wired pages.
1980 */
1981 entry = start_entry;
984263bc 1982 while ((entry != &map->header) && (entry->start < end)) {
f2d22ebf
MD
1983 KASSERT(entry->eflags & MAP_ENTRY_USER_WIRED,
1984 ("expected USER_WIRED on entry %p", entry));
984263bc
MD
1985 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1986 entry->wired_count--;
1987 if (entry->wired_count == 0)
f2d22ebf 1988 vm_fault_unwire(map, entry);
984263bc
MD
1989 entry = entry->next;
1990 }
1991 }
1992done:
a108bf71 1993 vm_map_unclip_range(map, start_entry, start, real_end, &count,
984263bc
MD
1994 MAP_CLIP_NO_HOLES);
1995 map->timestamp++;
1996 vm_map_unlock(map);
a108bf71 1997 vm_map_entry_release(count);
984263bc
MD
1998 return (rv);
1999}
2000
2001/*
cde87949 2002 * vm_map_wire:
984263bc
MD
2003 *
2004 * Sets the pageability of the specified address
2005 * range in the target map. Regions specified
2006 * as not pageable require locked-down physical
2007 * memory and physical page maps.
2008 *
2009 * The map must not be locked, but a reference
2010 * must remain to the map throughout the call.
a108bf71
MD
2011 *
2012 * This function may be called via the zalloc path and must properly
2013 * reserve map entries for kernel_map.
984263bc
MD
2014 */
2015int
e1359933 2016vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t real_end, int kmflags)
984263bc
MD
2017{
2018 vm_map_entry_t entry;
2019 vm_map_entry_t start_entry;
2020 vm_offset_t end;
2021 int rv = KERN_SUCCESS;
a108bf71 2022 int count;
984263bc 2023
e1359933 2024 if (kmflags & KM_KRESERVE)
a108bf71 2025 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
a108bf71
MD
2026 else
2027 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2028 vm_map_lock(map);
2029 VM_MAP_RANGE_CHECK(map, start, real_end);
2030 end = real_end;
2031
a108bf71 2032 start_entry = vm_map_clip_range(map, start, end, &count, MAP_CLIP_NO_HOLES);
984263bc
MD
2033 if (start_entry == NULL) {
2034 vm_map_unlock(map);
a108bf71
MD
2035 rv = KERN_INVALID_ADDRESS;
2036 goto failure;
984263bc 2037 }
e1359933 2038 if ((kmflags & KM_PAGEABLE) == 0) {
984263bc
MD
2039 /*
2040 * Wiring.
2041 *
2042 * 1. Holding the write lock, we create any shadow or zero-fill
2043 * objects that need to be created. Then we clip each map
2044 * entry to the region to be wired and increment its wiring
2045 * count. We create objects before clipping the map entries
2046 * to avoid object proliferation.
2047 *
2048 * 2. We downgrade to a read lock, and call vm_fault_wire to
2049 * fault in the pages for any newly wired area (wired_count is
2050 * 1).
2051 *
2052 * Downgrading to a read lock for vm_fault_wire avoids a
2053 * possible deadlock with another process that may have faulted
2054 * on one of the pages to be wired (it would mark the page busy,
2055 * blocking us, then in turn block on the map lock that we
2056 * hold). Because of problems in the recursive lock package,
2057 * we cannot upgrade to a write lock in vm_map_lookup. Thus,
2058 * any actions that require the write lock must be done
2059 * beforehand. Because we keep the read lock on the map, the
2060 * copy-on-write status of the entries we modify here cannot
2061 * change.
2062 */
2063
2064 entry = start_entry;
2065 while ((entry != &map->header) && (entry->start < end)) {
2066 /*
2067 * Trivial case if the entry is already wired
2068 */
2069 if (entry->wired_count) {
2070 entry->wired_count++;
2071 entry = entry->next;
2072 continue;
2073 }
2074
2075 /*
2076 * The entry is being newly wired, we have to setup
2077 * appropriate management structures. A shadow
2078 * object is required for a copy-on-write region,
2079 * or a normal object for a zero-fill region. We
2080 * do not have to do this for entries that point to sub
2081 * maps because we won't hold the lock on the sub map.
2082 */
1b874851 2083 if (entry->maptype != VM_MAPTYPE_SUBMAP) {
984263bc
MD
2084 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
2085 if (copyflag &&
2086 ((entry->protection & VM_PROT_WRITE) != 0)) {
53025830 2087 vm_map_entry_shadow(entry);
984263bc
MD
2088 } else if (entry->object.vm_object == NULL &&
2089 !map->system_map) {
53025830 2090 vm_map_entry_allocate_object(entry);
984263bc
MD
2091 }
2092 }
2093
2094 entry->wired_count++;
2095 entry = entry->next;
2096 }
2097
2098 /*
2099 * Pass 2.
2100 */
2101
2102 /*
2103 * HACK HACK HACK HACK
2104 *
2105 * Unlock the map to avoid deadlocks. The in-transit flag
2106 * protects us from most changes but note that
2107 * clipping may still occur. To prevent clipping from
2108 * occuring after the unlock, except for when we are
5fd012e0
MD
2109 * blocking in vm_fault_wire, we must run in a critical
2110 * section, otherwise our accesses to entry->start and
2111 * entry->end could be corrupted. We have to enter the
2112 * critical section prior to unlocking so start_entry does
2113 * not change out from under us at the very beginning of the
2114 * loop.
984263bc
MD
2115 *
2116 * HACK HACK HACK HACK
2117 */
2118
5fd012e0 2119 crit_enter();
984263bc
MD
2120
2121 entry = start_entry;
2122 while (entry != &map->header && entry->start < end) {
2123 /*
2124 * If vm_fault_wire fails for any page we need to undo
2125 * what has been done. We decrement the wiring count
2126 * for those pages which have not yet been wired (now)
2127 * and unwire those that have (later).
2128 */
2129 vm_offset_t save_start = entry->start;
2130 vm_offset_t save_end = entry->end;
2131
2132 if (entry->wired_count == 1)
f2d22ebf 2133 rv = vm_fault_wire(map, entry, FALSE);
984263bc
MD
2134 if (rv) {
2135 CLIP_CHECK_BACK(entry, save_start);
2136 for (;;) {
2137 KASSERT(entry->wired_count == 1, ("wired_count changed unexpectedly"));
2138 entry->wired_count = 0;
2139 if (entry->end == save_end)
2140 break;
2141 entry = entry->next;
2142 KASSERT(entry != &map->header, ("bad entry clip during backout"));
2143 }
2144 end = save_start;
2145 break;
2146 }
2147 CLIP_CHECK_FWD(entry, save_end);
2148 entry = entry->next;
2149 }
5fd012e0 2150 crit_exit();
984263bc 2151
984263bc
MD
2152 /*
2153 * If a failure occured undo everything by falling through
2154 * to the unwiring code. 'end' has already been adjusted
2155 * appropriately.
2156 */
2157 if (rv)
e1359933 2158 kmflags |= KM_PAGEABLE;
984263bc
MD
2159
2160 /*
f2d22ebf
MD
2161 * start_entry is still IN_TRANSITION but may have been
2162 * clipped since vm_fault_wire() unlocks and relocks the
2163 * map. No matter how clipped it has gotten there should
2164 * be a fragment that is on our start boundary.
984263bc
MD
2165 */
2166 CLIP_CHECK_BACK(start_entry, start);
2167 }
2168
e1359933 2169 if (kmflags & KM_PAGEABLE) {
984263bc
MD
2170 /*
2171 * This is the unwiring case. We must first ensure that the
2172 * range to be unwired is really wired down. We know there
2173 * are no holes.
2174 */
2175 entry = start_entry;
2176 while ((entry != &map->header) && (entry->start < end)) {
2177 if (entry->wired_count == 0) {
2178 rv = KERN_INVALID_ARGUMENT;
2179 goto done;
2180 }
2181 entry = entry->next;
2182 }
2183
2184 /*
2185 * Now decrement the wiring count for each region. If a region
2186 * becomes completely unwired, unwire its physical pages and
2187 * mappings.
2188 */
2189 entry = start_entry;
2190 while ((entry != &map->header) && (entry->start < end)) {
2191 entry->wired_count--;
2192 if (entry->wired_count == 0)
f2d22ebf 2193 vm_fault_unwire(map, entry);
984263bc
MD
2194 entry = entry->next;
2195 }
2196 }
2197done:
a108bf71 2198 vm_map_unclip_range(map, start_entry, start, real_end, &count,
984263bc
MD
2199 MAP_CLIP_NO_HOLES);
2200 map->timestamp++;
2201 vm_map_unlock(map);
a108bf71 2202failure:
e1359933 2203 if (kmflags & KM_KRESERVE)
a108bf71 2204 vm_map_entry_krelease(count);
a108bf71
MD
2205 else
2206 vm_map_entry_release(count);
984263bc
MD
2207 return (rv);
2208}
2209
a108bf71
MD
2210/*
2211 * vm_map_set_wired_quick()
2212 *
2213 * Mark a newly allocated address range as wired but do not fault in
2214 * the pages. The caller is expected to load the pages into the object.
2215 *
2216 * The map must be locked on entry and will remain locked on return.
2217 */
2218void
2219vm_map_set_wired_quick(vm_map_t map, vm_offset_t addr, vm_size_t size, int *countp)
2220{
2221 vm_map_entry_t scan;
2222 vm_map_entry_t entry;
2223
2224 entry = vm_map_clip_range(map, addr, addr + size, countp, MAP_CLIP_NO_HOLES);
2225 for (scan = entry; scan != &map->header && scan->start < addr + size; scan = scan->next) {
2226 KKASSERT(entry->wired_count == 0);
2227 entry->wired_count = 1;
2228 }
2229 vm_map_unclip_range(map, entry, addr, addr + size, countp, MAP_CLIP_NO_HOLES);
2230}
2231
984263bc
MD
2232/*
2233 * vm_map_clean
2234 *
2235 * Push any dirty cached pages in the address range to their pager.
2236 * If syncio is TRUE, dirty pages are written synchronously.
2237 * If invalidate is TRUE, any cached pages are freed as well.
2238 *
2239 * Returns an error if any part of the specified range is not mapped.
2240 */
2241int
57e43348
MD
2242vm_map_clean(vm_map_t map, vm_offset_t start, vm_offset_t end, boolean_t syncio,
2243 boolean_t invalidate)
984263bc
MD
2244{
2245 vm_map_entry_t current;
2246 vm_map_entry_t entry;
2247 vm_size_t size;
2248 vm_object_t object;
2249 vm_ooffset_t offset;
2250
2251 vm_map_lock_read(map);
2252 VM_MAP_RANGE_CHECK(map, start, end);
2253 if (!vm_map_lookup_entry(map, start, &entry)) {
2254 vm_map_unlock_read(map);
2255 return (KERN_INVALID_ADDRESS);
2256 }
2257 /*
2258 * Make a first pass to check for holes.
2259 */
2260 for (current = entry; current->start < end; current = current->next) {
1b874851 2261 if (current->maptype == VM_MAPTYPE_SUBMAP) {
984263bc
MD
2262 vm_map_unlock_read(map);
2263 return (KERN_INVALID_ARGUMENT);
2264 }
2265 if (end > current->end &&
2266 (current->next == &map->header ||
2267 current->end != current->next->start)) {
2268 vm_map_unlock_read(map);
2269 return (KERN_INVALID_ADDRESS);
2270 }
2271 }
2272
2273 if (invalidate)
2274 pmap_remove(vm_map_pmap(map), start, end);
2275 /*
2276 * Make a second pass, cleaning/uncaching pages from the indicated
2277 * objects as we go.
2278 */
2279 for (current = entry; current->start < end; current = current->next) {
2280 offset = current->offset + (start - current->start);
2281 size = (end <= current->end ? end : current->end) - start;
1b874851 2282 if (current->maptype == VM_MAPTYPE_SUBMAP) {
984263bc
MD
2283 vm_map_t smap;
2284 vm_map_entry_t tentry;
2285 vm_size_t tsize;
2286
2287 smap = current->object.sub_map;
2288 vm_map_lock_read(smap);
418ff780 2289 vm_map_lookup_entry(smap, offset, &tentry);
984263bc
MD
2290 tsize = tentry->end - offset;
2291 if (tsize < size)
2292 size = tsize;
2293 object = tentry->object.vm_object;
2294 offset = tentry->offset + (offset - tentry->start);
2295 vm_map_unlock_read(smap);
2296 } else {
2297 object = current->object.vm_object;
2298 }
2299 /*
2300 * Note that there is absolutely no sense in writing out
2301 * anonymous objects, so we track down the vnode object
2302 * to write out.
2303 * We invalidate (remove) all pages from the address space
2304 * anyway, for semantic correctness.
2305 *
2306 * note: certain anonymous maps, such as MAP_NOSYNC maps,
2307 * may start out with a NULL object.
2308 */
2309 while (object && object->backing_object) {
984263bc 2310 offset += object->backing_object_offset;
7cad6903 2311 object = object->backing_object;
984263bc
MD
2312 if (object->size < OFF_TO_IDX( offset + size))
2313 size = IDX_TO_OFF(object->size) - offset;
2314 }
2315 if (object && (object->type == OBJT_VNODE) &&
2316 (current->protection & VM_PROT_WRITE)) {
2317 /*
2318 * Flush pages if writing is allowed, invalidate them
2319 * if invalidation requested. Pages undergoing I/O
2320 * will be ignored by vm_object_page_remove().
2321 *
2322 * We cannot lock the vnode and then wait for paging
2323 * to complete without deadlocking against vm_fault.
2324 * Instead we simply call vm_object_page_remove() and
2325 * allow it to block internally on a page-by-page
2326 * basis when it encounters pages undergoing async
2327 * I/O.
2328 */
2329 int flags;
2330
2331 vm_object_reference(object);
ca466bae 2332 vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY);
984263bc
MD
2333 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
2334 flags |= invalidate ? OBJPC_INVAL : 0;
1b874851
MD
2335
2336 /*
2337 * When operating on a virtual page table just
2338 * flush the whole object. XXX we probably ought
2339 * to
2340 */
2341 switch(current->maptype) {
2342 case VM_MAPTYPE_NORMAL:
2343 vm_object_page_clean(object,
2344 OFF_TO_IDX(offset),
2345 OFF_TO_IDX(offset + size + PAGE_MASK),
2346 flags);
2347 break;
2348 case VM_MAPTYPE_VPAGETABLE:
2349 vm_object_page_clean(object, 0, 0, flags);
2350 break;
2351 }
a11aaa81 2352 vn_unlock(((struct vnode *)object->handle));
984263bc
MD
2353 vm_object_deallocate(object);
2354 }
2355 if (object && invalidate &&
2356 ((object->type == OBJT_VNODE) ||
2357 (object->type == OBJT_DEVICE))) {
2f1821ca
MD
2358 int clean_only =
2359 (object->type == OBJT_DEVICE) ? FALSE : TRUE;
984263bc 2360 vm_object_reference(object);
1b874851
MD
2361 switch(current->maptype) {
2362 case VM_MAPTYPE_NORMAL:
2363 vm_object_page_remove(object,
2364 OFF_TO_IDX(offset),
2365 OFF_TO_IDX(offset + size + PAGE_MASK),
2366 clean_only);
2367 break;
2368 case VM_MAPTYPE_VPAGETABLE:
2369 vm_object_page_remove(object, 0, 0, clean_only);
2370 break;
2371 }
984263bc
MD
2372 vm_object_deallocate(object);
2373 }
2374 start += size;
2375 }
2376
2377 vm_map_unlock_read(map);
2378 return (KERN_SUCCESS);
2379}
2380
2381/*
2382 * vm_map_entry_unwire: [ internal use only ]
2383 *
2384 * Make the region specified by this entry pageable.
2385 *
2386 * The map in question should be locked.
2387 * [This is the reason for this routine's existence.]
2388 */
2389static void
a108bf71 2390vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
984263bc 2391{
f2d22ebf 2392 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
984263bc 2393 entry->wired_count = 0;
f2d22ebf 2394 vm_fault_unwire(map, entry);
984263bc
MD
2395}
2396
2397/*
2398 * vm_map_entry_delete: [ internal use only ]
2399 *
2400 * Deallocate the given entry from the target map.
2401 */
2402static void
a108bf71 2403vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc
MD
2404{
2405 vm_map_entry_unlink(map, entry);
2406 map->size -= entry->end - entry->start;
2407
1b874851
MD
2408 switch(entry->maptype) {
2409 case VM_MAPTYPE_NORMAL:
2410 case VM_MAPTYPE_VPAGETABLE:
984263bc 2411 vm_object_deallocate(entry->object.vm_object);
1b874851
MD
2412 break;
2413 default:
2414 break;
984263bc
MD
2415 }
2416
a108bf71 2417 vm_map_entry_dispose(map, entry, countp);
984263bc
MD
2418}
2419
2420/*
2421 * vm_map_delete: [ internal use only ]
2422 *
2423 * Deallocates the given address range from the target
2424 * map.
2425 */
2426int
a108bf71 2427vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end, int *countp)
984263bc
MD
2428{
2429 vm_object_t object;
2430 vm_map_entry_t entry;
2431 vm_map_entry_t first_entry;
2432
686dbf64 2433again:
984263bc 2434 /*
686dbf64
MD
2435 * Find the start of the region, and clip it. Set entry to point
2436 * at the first record containing the requested address or, if no
2437 * such record exists, the next record with a greater address. The
2438 * loop will run from this point until a record beyond the termination
2439 * address is encountered.
2440 *
2441 * map->hint must be adjusted to not point to anything we delete,
2442 * so set it to the entry prior to the one being deleted.
2443 *
2444 * GGG see other GGG comment.
984263bc 2445 */
686dbf64 2446 if (vm_map_lookup_entry(map, start, &first_entry)) {
984263bc 2447 entry = first_entry;
a108bf71 2448 vm_map_clip_start(map, entry, start, countp);
686dbf64
MD
2449 map->hint = entry->prev; /* possible problem XXX */
2450 } else {
2451 map->hint = first_entry; /* possible problem XXX */
2452 entry = first_entry->next;
984263bc
MD
2453 }
2454
2455 /*
686dbf64
MD
2456 * If a hole opens up prior to the current first_free then
2457 * adjust first_free. As with map->hint, map->first_free
2458 * cannot be left set to anything we might delete.
984263bc 2459 */
984263bc
MD
2460 if (entry == &map->header) {
2461 map->first_free = &map->header;
2462 } else if (map->first_free->start >= start) {
2463 map->first_free = entry->prev;
2464 }
2465
2466 /*
2467 * Step through all entries in this region
2468 */
2469
2470 while ((entry != &map->header) && (entry->start < end)) {
2471 vm_map_entry_t next;
2472 vm_offset_t s, e;
2473 vm_pindex_t offidxstart, offidxend, count;
2474
2475 /*
2476 * If we hit an in-transition entry we have to sleep and
2477 * retry. It's easier (and not really slower) to just retry
2478 * since this case occurs so rarely and the hint is already
2479 * pointing at the right place. We have to reset the
2480 * start offset so as not to accidently delete an entry
2481 * another process just created in vacated space.
2482 */
2483 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2484 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2485 start = entry->start;
12e4aaff
MD
2486 ++mycpu->gd_cnt.v_intrans_coll;
2487 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
2488 vm_map_transition_wait(map);
2489 goto again;
2490 }
a108bf71 2491 vm_map_clip_end(map, entry, end, countp);
984263bc
MD
2492
2493 s = entry->start;
2494 e = entry->end;
2495 next = entry->next;
2496
2497 offidxstart = OFF_TO_IDX(entry->offset);
2498 count = OFF_TO_IDX(e - s);
2499 object = entry->object.vm_object;
2500
2501 /*
2502 * Unwire before removing addresses from the pmap; otherwise,
2503 * unwiring will put the entries back in the pmap.
2504 */
f2d22ebf 2505 if (entry->wired_count != 0)
984263bc 2506 vm_map_entry_unwire(map, entry);
984263bc
MD
2507
2508 offidxend = offidxstart + count;
2509
c439ad8f 2510 if (object == &kernel_object) {
984263bc
MD
2511 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2512 } else {
2513 pmap_remove(map->pmap, s, e);
2514 if (object != NULL &&
2515 object->ref_count != 1 &&
2516 (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING &&
2517 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2518 vm_object_collapse(object);
2519 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2520 if (object->type == OBJT_SWAP) {
2521 swap_pager_freespace(object, offidxstart, count);
2522 }
2523 if (offidxend >= object->size &&
2524 offidxstart < object->size) {
2525 object->size = offidxstart;
2526 }
2527 }
2528 }
2529
2530 /*
2531 * Delete the entry (which may delete the object) only after
2532 * removing all pmap entries pointing to its pages.
2533 * (Otherwise, its page frames may be reallocated, and any
2534 * modify bits will be set in the wrong object!)
2535 */
a108bf71 2536 vm_map_entry_delete(map, entry, countp);
984263bc
MD
2537 entry = next;
2538 }
2539 return (KERN_SUCCESS);
2540}
2541
2542/*
2543 * vm_map_remove:
2544 *
2545 * Remove the given address range from the target map.
2546 * This is the exported form of vm_map_delete.
2547 */
2548int
a108bf71 2549vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
984263bc 2550{
03aa8d99 2551 int result;
a108bf71 2552 int count;
984263bc 2553
a108bf71 2554 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2555 vm_map_lock(map);
2556 VM_MAP_RANGE_CHECK(map, start, end);
a108bf71 2557 result = vm_map_delete(map, start, end, &count);
984263bc 2558 vm_map_unlock(map);
a108bf71 2559 vm_map_entry_release(count);
984263bc 2560
984263bc
MD
2561 return (result);
2562}
2563
2564/*
2565 * vm_map_check_protection:
2566 *
2567 * Assert that the target map allows the specified
2568 * privilege on the entire address region given.
2569 * The entire region must be allocated.
2570 */
2571boolean_t
2572vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2573 vm_prot_t protection)
2574{
2575 vm_map_entry_t entry;
2576 vm_map_entry_t tmp_entry;
2577
2578 if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2579 return (FALSE);
2580 }
2581 entry = tmp_entry;
2582
2583 while (start < end) {
2584 if (entry == &map->header) {
2585 return (FALSE);
2586 }
2587 /*
2588 * No holes allowed!
2589 */
2590
2591 if (start < entry->start) {
2592 return (FALSE);
2593 }
2594 /*
2595 * Check protection associated with entry.
2596 */
2597
2598 if ((entry->protection & protection) != protection) {
2599 return (FALSE);
2600 }
2601 /* go to next entry */
2602
2603 start = entry->end;
2604 entry = entry->next;
2605 }
2606 return (TRUE);
2607}
2608
2609/*
2610 * Split the pages in a map entry into a new object. This affords
2611 * easier removal of unused pages, and keeps object inheritance from
2612 * being a negative impact on memory usage.
2613 */
2614static void
a108bf71 2615vm_map_split(vm_map_entry_t entry)
984263bc
MD
2616{
2617 vm_page_t m;
2618 vm_object_t orig_object, new_object, source;
2619 vm_offset_t s, e;
2620 vm_pindex_t offidxstart, offidxend, idx;
2621 vm_size_t size;
2622 vm_ooffset_t offset;
2623
2624 orig_object = entry->object.vm_object;
2625 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
2626 return;
2627 if (orig_object->ref_count <= 1)
2628 return;
2629
2630 offset = entry->offset;
2631 s = entry->start;
2632 e = entry->end;
2633
2634 offidxstart = OFF_TO_IDX(offset);
2635 offidxend = offidxstart + OFF_TO_IDX(e - s);
2636 size = offidxend - offidxstart;
2637
57f7b636
MD
2638 new_object = vm_pager_allocate(orig_object->type, NULL,
2639 IDX_TO_OFF(size), VM_PROT_ALL, 0);
984263bc
MD
2640 if (new_object == NULL)
2641 return;
2642
2643 source = orig_object->backing_object;
2644 if (source != NULL) {
2645 vm_object_reference(source); /* Referenced by new_object */
2646 LIST_INSERT_HEAD(&source->shadow_head,
2647 new_object, shadow_list);
2648 vm_object_clear_flag(source, OBJ_ONEMAPPING);
2649 new_object->backing_object_offset =
2650 orig_object->backing_object_offset + IDX_TO_OFF(offidxstart);
2651 new_object->backing_object = source;
2652 source->shadow_count++;
2653 source->generation++;
2654 }
2655
2656 for (idx = 0; idx < size; idx++) {
2657 vm_page_t m;
2658
06ecca5a 2659 /*
5fd012e0 2660 * A critical section is required to avoid a race between
06ecca5a
MD
2661 * the lookup and an interrupt/unbusy/free and our busy
2662 * check.
2663 */
5fd012e0 2664 crit_enter();
984263bc
MD
2665 retry:
2666 m = vm_page_lookup(orig_object, offidxstart + idx);
06ecca5a 2667 if (m == NULL) {
5fd012e0 2668 crit_exit();
984263bc 2669 continue;
06ecca5a 2670 }
984263bc
MD
2671
2672 /*
2673 * We must wait for pending I/O to complete before we can
2674 * rename the page.
2675 *
2676 * We do not have to VM_PROT_NONE the page as mappings should
2677 * not be changed by this operation.
2678 */
2679 if (vm_page_sleep_busy(m, TRUE, "spltwt"))
2680 goto retry;
984263bc
MD
2681 vm_page_busy(m);
2682 vm_page_rename(m, new_object, idx);
2683 /* page automatically made dirty by rename and cache handled */
2684 vm_page_busy(m);
5fd012e0 2685 crit_exit();
984263bc
MD
2686 }
2687
2688 if (orig_object->type == OBJT_SWAP) {
2689 vm_object_pip_add(orig_object, 1);
2690 /*
2691 * copy orig_object pages into new_object
2692 * and destroy unneeded pages in
2693 * shadow object.
2694 */
2695 swap_pager_copy(orig_object, new_object, offidxstart, 0);
2696 vm_object_pip_wakeup(orig_object);
2697 }
2698
06ecca5a
MD
2699 /*
2700 * Wakeup the pages we played with. No spl protection is needed
2701 * for a simple wakeup.
2702 */
984263bc
MD
2703 for (idx = 0; idx < size; idx++) {
2704 m = vm_page_lookup(new_object, idx);
06ecca5a 2705 if (m)
984263bc 2706 vm_page_wakeup(m);
984263bc
MD
2707 }
2708
2709 entry->object.vm_object = new_object;
2710 entry->offset = 0LL;
2711 vm_object_deallocate(orig_object);
2712}
2713
2714/*
2715 * vm_map_copy_entry:
2716 *
2717 * Copies the contents of the source entry to the destination
2718 * entry. The entries *must* be aligned properly.
2719 */
2720static void
a108bf71
MD
2721vm_map_copy_entry(vm_map_t src_map, vm_map_t dst_map,
2722 vm_map_entry_t src_entry, vm_map_entry_t dst_entry)
984263bc
MD
2723{
2724 vm_object_t src_object;
2725
1b874851
MD
2726 if (dst_entry->maptype == VM_MAPTYPE_SUBMAP)
2727 return;
2728 if (src_entry->maptype == VM_MAPTYPE_SUBMAP)
984263bc
MD
2729 return;
2730
2731 if (src_entry->wired_count == 0) {
984263bc
MD
2732 /*
2733 * If the source entry is marked needs_copy, it is already
2734 * write-protected.
2735 */
2736 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2737 pmap_protect(src_map->pmap,
2738 src_entry->start,
2739 src_entry->end,
2740 src_entry->protection & ~VM_PROT_WRITE);
2741 }
2742
2743 /*
2744 * Make a copy of the object.
2745 */
2746 if ((src_object = src_entry->object.vm_object) != NULL) {
984263bc
MD
2747 if ((src_object->handle == NULL) &&
2748 (src_object->type == OBJT_DEFAULT ||
2749 src_object->type == OBJT_SWAP)) {
2750 vm_object_collapse(src_object);
2751 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2752 vm_map_split(src_entry);
2753 src_object = src_entry->object.vm_object;
2754 }
2755 }
2756
2757 vm_object_reference(src_object);
2758 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2759 dst_entry->object.vm_object = src_object;
2760 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2761 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2762 dst_entry->offset = src_entry->offset;
2763 } else {
2764 dst_entry->object.vm_object = NULL;
2765 dst_entry->offset = 0;
2766 }
2767
2768 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2769 dst_entry->end - dst_entry->start, src_entry->start);
2770 } else {
2771 /*
2772 * Of course, wired down pages can't be set copy-on-write.
2773 * Cause wired pages to be copied into the new map by
2774 * simulating faults (the new pages are pageable)
2775 */
2776 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2777 }
2778}
2779
2780/*
2781 * vmspace_fork:
2782 * Create a new process vmspace structure and vm_map
2783 * based on those of an existing process. The new map
2784 * is based on the old map, according to the inheritance
2785 * values on the regions in that map.
2786 *
2787 * The source map must not be locked.
2788 */
2789struct vmspace *
a108bf71 2790vmspace_fork(struct vmspace *vm1)
984263bc
MD
2791{
2792 struct vmspace *vm2;
2793 vm_map_t old_map = &vm1->vm_map;
2794 vm_map_t new_map;
2795 vm_map_entry_t old_entry;
2796 vm_map_entry_t new_entry;
2797 vm_object_t object;
a108bf71 2798 int count;
984263bc
MD
2799
2800 vm_map_lock(old_map);
2801 old_map->infork = 1;
2802
239b4df9
MD
2803 /*
2804 * XXX Note: upcalls are not copied.
2805 */
984263bc
MD
2806 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
2807 bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
239b4df9 2808 (caddr_t)&vm1->vm_endcopy - (caddr_t)&vm1->vm_startcopy);
984263bc
MD
2809 new_map = &vm2->vm_map; /* XXX */
2810 new_map->timestamp = 1;
2811
a108bf71 2812 count = 0;
984263bc 2813 old_entry = old_map->header.next;
a108bf71
MD
2814 while (old_entry != &old_map->header) {
2815 ++count;
2816 old_entry = old_entry->next;
2817 }
984263bc 2818
a108bf71
MD
2819 count = vm_map_entry_reserve(count + MAP_RESERVE_COUNT);
2820
2821 old_entry = old_map->header.next;
984263bc 2822 while (old_entry != &old_map->header) {
1b874851 2823 if (old_entry->maptype == VM_MAPTYPE_SUBMAP)
984263bc
MD
2824 panic("vm_map_fork: encountered a submap");
2825
2826 switch (old_entry->inheritance) {
2827 case VM_INHERIT_NONE:
2828 break;
2829
2830 case VM_INHERIT_SHARE:
2831 /*
1b874851
MD
2832 * Clone the entry, creating the shared object if
2833 * necessary.
984263bc
MD
2834 */
2835 object = old_entry->object.vm_object;
2836 if (object == NULL) {
53025830
MD
2837 vm_map_entry_allocate_object(old_entry);
2838 object = old_entry->object.vm_object;
984263bc
MD
2839 }
2840
2841 /*
53025830 2842 * Add the reference before calling vm_map_entry_shadow
984263bc
MD
2843 * to insure that a shadow object is created.
2844 */
2845 vm_object_reference(object);
2846 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
53025830 2847 vm_map_entry_shadow(old_entry);
984263bc
MD
2848 /* Transfer the second reference too. */
2849 vm_object_reference(
2850 old_entry->object.vm_object);
2851 vm_object_deallocate(object);
2852 object = old_entry->object.vm_object;
2853 }
2854 vm_object_clear_flag(object, OBJ_ONEMAPPING);
2855
2856 /*
2857 * Clone the entry, referencing the shared object.
2858 */
a108bf71 2859 new_entry = vm_map_entry_create(new_map, &count);
984263bc
MD
2860 *new_entry = *old_entry;
2861 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2862 new_entry->wired_count = 0;
2863
2864 /*
2865 * Insert the entry into the new map -- we know we're
2866 * inserting at the end of the new map.
2867 */
2868
2869 vm_map_entry_link(new_map, new_map->header.prev,
2870 new_entry);
2871
2872 /*
2873 * Update the physical map
2874 */
2875
2876 pmap_copy(new_map->pmap, old_map->pmap,
2877 new_entry->start,
2878 (old_entry->end - old_entry->start),
2879 old_entry->start);
2880 break;
2881
2882 case VM_INHERIT_COPY:
2883 /*
2884 * Clone the entry and link into the map.
2885 */
a108bf71 2886 new_entry = vm_map_entry_create(new_map, &count);
984263bc
MD
2887 *new_entry = *old_entry;
2888 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2889 new_entry->wired_count = 0;
2890 new_entry->object.vm_object = NULL;
2891 vm_map_entry_link(new_map, new_map->header.prev,
2892 new_entry);
2893 vm_map_copy_entry(old_map, new_map, old_entry,
2894 new_entry);
2895 break;
2896 }
2897 old_entry = old_entry->next;
2898 }
2899
2900 new_map->size = old_map->size;
2901 old_map->infork = 0;
2902 vm_map_unlock(old_map);
a108bf71 2903 vm_map_entry_release(count);
984263bc
MD
2904
2905 return (vm2);
2906}
2907
2908int
2909vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
2910 vm_prot_t prot, vm_prot_t max, int cow)
2911{
2912 vm_map_entry_t prev_entry;
2913 vm_map_entry_t new_stack_entry;
2914 vm_size_t init_ssize;
2915 int rv;
a108bf71 2916 int count;
984263bc 2917
88181b08 2918 if (VM_MIN_USER_ADDRESS > 0 && addrbos < VM_MIN_USER_ADDRESS)
984263bc
MD
2919 return (KERN_NO_SPACE);
2920
2921 if (max_ssize < sgrowsiz)
2922 init_ssize = max_ssize;
2923 else
2924 init_ssize = sgrowsiz;
2925
a108bf71 2926 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2927 vm_map_lock(map);
2928
2929 /* If addr is already mapped, no go */
2930 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
2931 vm_map_unlock(map);
a108bf71 2932 vm_map_entry_release(count);
984263bc
MD
2933 return (KERN_NO_SPACE);
2934 }
2935
2936 /* If we would blow our VMEM resource limit, no go */
2937 if (map->size + init_ssize >
2938 curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
2939 vm_map_unlock(map);
a108bf71 2940 vm_map_entry_release(count);
984263bc
MD
2941 return (KERN_NO_SPACE);
2942 }
2943
2944 /* If we can't accomodate max_ssize in the current mapping,
2945 * no go. However, we need to be aware that subsequent user
2946 * mappings might map into the space we have reserved for
2947 * stack, and currently this space is not protected.
2948 *
2949 * Hopefully we will at least detect this condition
2950 * when we try to grow the stack.
2951 */
2952 if ((prev_entry->next != &map->header) &&
2953 (prev_entry->next->start < addrbos + max_ssize)) {
2954 vm_map_unlock(map);
a108bf71 2955 vm_map_entry_release(count);
984263bc
MD
2956 return (KERN_NO_SPACE);
2957 }
2958
2959 /* We initially map a stack of only init_ssize. We will
2960 * grow as needed later. Since this is to be a grow
2961 * down stack, we map at the top of the range.
2962 *
2963 * Note: we would normally expect prot and max to be
2964 * VM_PROT_ALL, and cow to be 0. Possibly we should
2965 * eliminate these as input parameters, and just
2966 * pass these values here in the insert call.
2967 */
a108bf71
MD
2968 rv = vm_map_insert(map, &count,
2969 NULL, 0, addrbos + max_ssize - init_ssize,
1b874851
MD
2970 addrbos + max_ssize,
2971 VM_MAPTYPE_NORMAL,
2972 prot, max,
2973 cow);
984263bc
MD
2974
2975 /* Now set the avail_ssize amount */
517e1666 2976 if (rv == KERN_SUCCESS) {
984263bc 2977 if (prev_entry != &map->header)
a108bf71 2978 vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize, &count);
984263bc
MD
2979 new_stack_entry = prev_entry->next;
2980 if (new_stack_entry->end != addrbos + max_ssize ||
2981 new_stack_entry->start != addrbos + max_ssize - init_ssize)
2982 panic ("Bad entry start/end for new stack entry");
2983 else
afeabdca 2984 new_stack_entry->aux.avail_ssize = max_ssize - init_ssize;
984263bc
MD
2985 }
2986
2987 vm_map_unlock(map);
a108bf71 2988 vm_map_entry_release(count);
984263bc
MD
2989 return (rv);
2990}
2991
2992/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
2993 * desired address is already mapped, or if we successfully grow
2994 * the stack. Also returns KERN_SUCCESS if addr is outside the
2995 * stack range (this is strange, but preserves compatibility with
2996 * the grow function in vm_machdep.c).
2997 */
2998int
2999vm_map_growstack (struct proc *p, vm_offset_t addr)
3000{
3001 vm_map_entry_t prev_entry;
3002 vm_map_entry_t stack_entry;
3003 vm_map_entry_t new_stack_entry;
3004 struct vmspace *vm = p->p_vmspace;
3005 vm_map_t map = &vm->vm_map;
3006 vm_offset_t end;
a108bf71
MD
3007 int grow_amount;
3008 int rv = KERN_SUCCESS;
3009 int is_procstack;
3010 int use_read_lock = 1;
3011 int count;
984263bc 3012
a108bf71 3013 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
3014Retry:
3015 if (use_read_lock)
3016 vm_map_lock_read(map);
3017 else
3018 vm_map_lock(map);
3019
3020 /* If addr is already in the entry range, no need to grow.*/
3021 if (vm_map_lookup_entry(map, addr, &prev_entry))
3022 goto done;
3023
3024 if ((stack_entry = prev_entry->next) == &map->header)
3025 goto done;
3026 if (prev_entry == &map->header)
afeabdca 3027 end = stack_entry->start - stack_entry->aux.avail_ssize;
984263bc
MD
3028 else
3029 end = prev_entry->end;
3030
3031 /* This next test mimics the old grow function in vm_machdep.c.
3032 * It really doesn't quite make sense, but we do it anyway
3033 * for compatibility.
3034 *
3035 * If not growable stack, return success. This signals the
3036 * caller to proceed as he would normally with normal vm.
3037 */
afeabdca 3038 if (stack_entry->aux.avail_ssize < 1 ||
984263bc 3039 addr >= stack_entry->start ||
afeabdca 3040 addr < stack_entry->start - stack_entry->aux.avail_ssize) {
984263bc
MD
3041 goto done;
3042 }
3043
3044 /* Find the minimum grow amount */
3045 grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
afeabdca 3046 if (grow_amount > stack_entry->aux.avail_ssize) {
984263bc
MD
3047 rv = KERN_NO_SPACE;
3048 goto done;
3049 }
3050
3051 /* If there is no longer enough space between the entries
3052 * nogo, and adjust the available space. Note: this
3053 * should only happen if the user has mapped into the
3054 * stack area after the stack was created, and is
3055 * probably an error.
3056 *
3057 * This also effectively destroys any guard page the user
3058 * might have intended by limiting the stack size.
3059 */
3060 if (grow_amount > stack_entry->start - end) {
3061 if (use_read_lock && vm_map_lock_upgrade(map)) {
3062 use_read_lock = 0;
3063 goto Retry;
3064 }
3065 use_read_lock = 0;
afeabdca 3066 stack_entry->aux.avail_ssize = stack_entry->start - end;
984263bc
MD
3067 rv = KERN_NO_SPACE;
3068 goto done;
3069 }
3070
3071 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
3072
3073 /* If this is the main process stack, see if we're over the
3074 * stack limit.
3075 */
3076 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
3077 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
3078 rv = KERN_NO_SPACE;
3079 goto done;
3080 }
3081
3082 /* Round up the grow amount modulo SGROWSIZ */
3083 grow_amount = roundup (grow_amount, sgrowsiz);
afeabdca
MD
3084 if (grow_amount > stack_entry->aux.avail_ssize) {
3085 grow_amount = stack_entry->aux.avail_ssize;
984263bc
MD
3086 }
3087 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
3088 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
3089 grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
3090 ctob(vm->vm_ssize);
3091 }
3092
3093 /* If we would blow our VMEM resource limit, no go */
be77b5f9 3094 if (map->size + grow_amount > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
984263bc
MD
3095 rv = KERN_NO_SPACE;
3096 goto done;
3097 }
3098
3099 if (use_read_lock && vm_map_lock_upgrade(map)) {
3100 use_read_lock = 0;
3101 goto Retry;
3102 }
3103 use_read_lock = 0;
3104
3105 /* Get the preliminary new entry start value */
3106 addr = stack_entry->start - grow_amount;
3107
3108 /* If this puts us into the previous entry, cut back our growth
3109 * to the available space. Also, see the note above.
3110 */
3111 if (addr < end) {
afeabdca 3112 stack_entry->aux.avail_ssize = stack_entry->start - end;
984263bc
MD
3113 addr = end;
3114 }
3115
a108bf71
MD
3116 rv = vm_map_insert(map, &count,
3117 NULL, 0, addr, stack_entry->start,
1b874851
MD
3118 VM_MAPTYPE_NORMAL,
3119 VM_PROT_ALL, VM_PROT_ALL,
984263bc
MD
3120 0);
3121
3122 /* Adjust the available stack space by the amount we grew. */
3123 if (rv == KERN_SUCCESS) {
3124 if (prev_entry != &map->header)
a108bf71 3125 vm_map_clip_end(map, prev_entry, addr, &count);
984263bc
MD
3126 new_stack_entry = prev_entry->next;
3127 if (new_stack_entry->end != stack_entry->start ||
3128 new_stack_entry->start != addr)
3129 panic ("Bad stack grow start/end in new stack entry");
3130 else {
afeabdca
MD
3131 new_stack_entry->aux.avail_ssize =
3132 stack_entry->aux.avail_ssize -
3133 (new_stack_entry->end - new_stack_entry->start);
984263bc
MD
3134 if (is_procstack)
3135 vm->vm_ssize += btoc(new_stack_entry->end -
3136 new_stack_entry->start);
3137 }
3138 }
3139
3140done:
3141 if (use_read_lock)
3142 vm_map_unlock_read(map);
3143 else
3144 vm_map_unlock(map);
a108bf71 3145 vm_map_entry_release(count);
984263bc
MD
3146 return (rv);
3147}
3148
3149/*
3150 * Unshare the specified VM space for exec. If other processes are
3151 * mapped to it, then create a new one. The new vmspace is null.
3152 */
3153
3154void
29802dbb 3155vmspace_exec(struct proc *p, struct vmspace *vmcopy)
a108bf71 3156{
984263bc
MD
3157 struct vmspace *oldvmspace = p->p_vmspace;
3158 struct vmspace *newvmspace;
3159 vm_map_t map = &p->p_vmspace->vm_map;
3160
29802dbb
MD
3161 /*
3162 * If we are execing a resident vmspace we fork it, otherwise
239b4df9
MD
3163 * we create a new vmspace. Note that exitingcnt and upcalls
3164 * are not copied to the new vmspace.
29802dbb
MD
3165 */
3166 if (vmcopy) {
3167 newvmspace = vmspace_fork(vmcopy);
3168 } else {
3169 newvmspace = vmspace_alloc(map->min_offset, map->max_offset);
3170 bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
239b4df9
MD
3171 (caddr_t)&oldvmspace->vm_endcopy -
3172 (caddr_t)&oldvmspace->vm_startcopy);
29802dbb
MD
3173 }
3174
984263bc
MD
3175 /*
3176 * This code is written like this for prototype purposes. The
3177 * goal is to avoid running down the vmspace here, but let the
3178 * other process's that are still using the vmspace to finally
3179 * run it down. Even though there is little or no chance of blocking
3180 * here, it is a good idea to keep this form for future mods.
3181 */
984263bc
MD
3182 p->p_vmspace = newvmspace;
3183 pmap_pinit2(vmspace_pmap(newvmspace));
3184 if (p == curproc)
3185 pmap_activate(p);
29802dbb 3186 vmspace_free(oldvmspace);
984263bc
MD
3187}
3188
3189/*
3190 * Unshare the specified VM space for forcing COW. This
3191 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
239b4df9
MD
3192 *
3193 * The exitingcnt test is not strictly necessary but has been
3194 * included for code sanity (to make the code a bit more deterministic).
984263bc
MD
3195 */
3196
3197void
a108bf71
MD
3198vmspace_unshare(struct proc *p)
3199{
984263bc
MD
3200 struct vmspace *oldvmspace = p->p_vmspace;
3201 struct vmspace *newvmspace;
3202
239b4df9 3203 if (oldvmspace->vm_refcnt == 1 && oldvmspace->vm_exitingcnt == 0)
984263bc
MD
3204 return;
3205 newvmspace = vmspace_fork(oldvmspace);
984263bc
MD
3206 p->p_vmspace = newvmspace;
3207 pmap_pinit2(vmspace_pmap(newvmspace));
3208 if (p == curproc)
3209 pmap_activate(p);
29802dbb 3210 vmspace_free(oldvmspace);
984263bc 3211}
984263bc
MD
3212
3213/*
3214 * vm_map_lookup:
3215 *
3216 * Finds the VM object, offset, and
3217 * protection for a given virtual address in the
3218 * specified map, assuming a page fault of the
3219 * type specified.
3220 *
3221 * Leaves the map in question locked for read; return
3222 * values are guaranteed until a vm_map_lookup_done
3223 * call is performed. Note that the map argument
3224 * is in/out; the returned map must be used in
3225 * the call to vm_map_lookup_done.
3226 *
3227 * A handle (out_entry) is returned for use in
3228 * vm_map_lookup_done, to make that fast.
3229 *
3230 * If a lookup is requested with "write protection"
3231 * specified, the map may be changed to perform virtual
3232 * copying operations, although the data referenced will
3233 * remain the same.
3234 */
3235int
3236vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3237 vm_offset_t vaddr,
3238 vm_prot_t fault_typea,
3239 vm_map_entry_t *out_entry, /* OUT */
3240 vm_object_t *object, /* OUT */
3241 vm_pindex_t *pindex, /* OUT */
3242 vm_prot_t *out_prot, /* OUT */
3243 boolean_t *wired) /* OUT */
3244{
3245 vm_map_entry_t entry;
3246 vm_map_t map = *var_map;
3247 vm_prot_t prot;
3248 vm_prot_t fault_type = fault_typea;
3249 int use_read_lock = 1;
3250 int rv = KERN_SUCCESS;
3251
3252RetryLookup:
3253 if (use_read_lock)
3254 vm_map_lock_read(map);
3255 else
3256 vm_map_lock(map);
3257
3258 /*
3259 * If the map has an interesting hint, try it before calling full
3260 * blown lookup routine.
3261 */
3262 entry = map->hint;
3263 *out_entry = entry;
3264
3265 if ((entry == &map->header) ||
3266 (vaddr < entry->start) || (vaddr >= entry->end)) {
3267 vm_map_entry_t tmp_entry;
3268
3269 /*
3270 * Entry was either not a valid hint, or the vaddr was not
3271 * contained in the entry, so do a full lookup.
3272 */
3273 if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) {
3274 rv = KERN_INVALID_ADDRESS;
3275 goto done;
3276 }
3277
3278 entry = tmp_entry;
3279 *out_entry = entry;
3280 }
3281
3282 /*
3283 * Handle submaps.
3284 */
1b874851 3285 if (entry->maptype == VM_MAPTYPE_SUBMAP) {
984263bc
MD
3286 vm_map_t old_map = map;
3287
3288 *var_map = map = entry->object.sub_map;
3289 if (use_read_lock)
3290 vm_map_unlock_read(old_map);
3291 else
3292 vm_map_unlock(old_map);
3293 use_read_lock = 1;
3294 goto RetryLookup;
3295 }
3296
3297 /*
3298 * Check whether this task is allowed to have this page.
3299 * Note the special case for MAP_ENTRY_COW
3300 * pages with an override. This is to implement a forced
3301 * COW for debuggers.
3302 */
3303
3304 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3305 prot = entry->max_protection;
3306 else
3307 prot = entry->protection;
3308
3309 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3310 if ((fault_type & prot) != fault_type) {
3311 rv = KERN_PROTECTION_FAILURE;
3312 goto done;
3313 }
3314
3315 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3316 (entry->eflags & MAP_ENTRY_COW) &&
3317 (fault_type & VM_PROT_WRITE) &&
3318 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3319 rv = KERN_PROTECTION_FAILURE;
3320 goto done;
3321 }
3322
3323 /*
3324 * If this page is not pageable, we have to get it for all possible
3325 * accesses.
3326 */
984263bc
MD
3327 *wired = (entry->wired_count != 0);
3328 if (*wired)
3329 prot = fault_type = entry->protection;
3330
3331 /*
568e6804
MD
3332 * Virtual page tables may need to update the accessed (A) bit
3333 * in a page table entry. Upgrade the fault to a write fault for
3334 * that case if the map will support it. If the map does not support
3335 * it the page table entry simply will not be updated.
984263bc 3336 */
568e6804
MD
3337 if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
3338 if (prot & VM_PROT_WRITE)
3339 fault_type |= VM_PROT_WRITE;
3340 }
984263bc 3341
568e6804
MD
3342 /*
3343 * If the entry was copy-on-write, we either ...
3344 */
984263bc
MD
3345 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3346 /*
3347 * If we want to write the page, we may as well handle that
3348 * now since we've got the map locked.
3349 *
3350 * If we don't need to write the page, we just demote the
3351 * permissions allowed.
3352 */
3353
3354 if (fault_type & VM_PROT_WRITE) {
3355 /*
3356 * Make a new object, and place it in the object
3357 * chain. Note that no new references have appeared
3358 * -- one just moved from the map to the new
3359 * object.
3360 */
3361
3362 if (use_read_lock && vm_map_lock_upgrade(map)) {
3363 use_read_lock = 0;
3364 goto RetryLookup;
3365 }
3366 use_read_lock = 0;
3367
53025830 3368 vm_map_entry_shadow(entry);
984263bc
MD
3369 } else {
3370 /*
3371 * We're attempting to read a copy-on-write page --
3372 * don't allow writes.
3373 */
3374
3375 prot &= ~VM_PROT_WRITE;
3376 }
3377 }
3378
3379 /*
3380 * Create an object if necessary.
3381 */
3382 if (entry->object.vm_object == NULL &&
3383 !map->system_map) {
3384 if (use_read_lock && vm_map_lock_upgrade(map)) {
3385 use_read_lock = 0;
3386 goto RetryLookup;
3387 }
3388 use_read_lock = 0;
53025830 3389 vm_map_entry_allocate_object(entry);
984263bc
MD
3390 }
3391
3392 /*
3393 * Return the object/offset from this entry. If the entry was
3394 * copy-on-write or empty, it has been fixed up.
3395 */
3396
3397 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3398 *object = entry->object.vm_object;
3399
3400 /*
3401 * Return whether this is the only map sharing this data. On
3402 * success we return with a read lock held on the map. On failure
3403 * we return with the map unlocked.
3404 */
3405 *out_prot = prot;
3406done:
3407 if (rv == KERN_SUCCESS) {
3408 if (use_read_lock == 0)
3409 vm_map_lock_downgrade(map);
3410 } else if (use_read_lock) {
3411 vm_map_unlock_read(map);
3412 } else {
3413 vm_map_unlock(map);
3414 }
3415 return (rv);
3416}
3417
3418/*
3419 * vm_map_lookup_done:
3420 *
3421 * Releases locks acquired by a vm_map_lookup
3422 * (according to the handle returned by that lookup).
3423 */
3424
3425void
a108bf71 3426vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry, int count)
984263bc
MD
3427{
3428 /*
3429 * Unlock the main-level map
3430 */
984263bc 3431 vm_map_unlock_read(map);
a108bf71
MD
3432 if (count)
3433 vm_map_entry_release(count);
984263bc
MD
3434}
3435
984263bc
MD
3436#include "opt_ddb.h"
3437#ifdef DDB
3438#include <sys/kernel.h>
3439
3440#include <ddb/ddb.h>
3441
3442/*
3443 * vm_map_print: [ debug ]
3444 */
3445DB_SHOW_COMMAND(map, vm_map_print)
3446{
3447 static int nlines;
3448 /* XXX convert args. */
3449 vm_map_t map = (vm_map_t)addr;
3450 boolean_t full = have_addr;
3451
3452 vm_map_entry_t entry;
3453
3454 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3455 (void *)map,
3456 (void *)map->pmap, map->nentries, map->timestamp);
3457 nlines++;
3458
3459 if (!full && db_indent)
3460 return;
3461
3462 db_indent += 2;
3463 for (entry = map->header.next; entry != &map->header;
3464 entry = entry->next) {
3465 db_iprintf("map entry %p: start=%p, end=%p\n",
3466 (void *)entry, (void *)entry->start, (void *)entry->end);
3467 nlines++;
3468 {
3469 static char *inheritance_name[4] =
3470 {"share", "copy", "none", "donate_copy"};
3471
3472 db_iprintf(" prot=%x/%x/%s",
3473 entry->protection,
3474 entry->max_protection,
3475 inheritance_name[(int)(unsigned char)entry->inheritance]);
3476 if (entry->wired_count != 0)
3477 db_printf(", wired");
3478 }
1b874851 3479 if (entry->maptype == VM_MAPTYPE_SUBMAP) {
984263bc
MD
3480 /* XXX no %qd in kernel. Truncate entry->offset. */
3481 db_printf(", share=%p, offset=0x%lx\n",
3482 (void *)entry->object.sub_map,
3483 (long)entry->offset);
3484 nlines++;
3485 if ((entry->prev == &map->header) ||
3486 (entry->prev->object.sub_map !=
3487 entry->object.sub_map)) {
3488 db_indent += 2;
3489 vm_map_print((db_expr_t)(intptr_t)
3490 entry->object.sub_map,
3491 full, 0, (char *)0);
3492 db_indent -= 2;
3493 }
3494 } else {
3495 /* XXX no %qd in kernel. Truncate entry->offset. */
3496 db_printf(", object=%p, offset=0x%lx",
3497 (void *)entry->object.vm_object,
3498 (long)entry->offset);
3499 if (entry->eflags & MAP_ENTRY_COW)
3500 db_printf(", copy (%s)",
3501 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3502 db_printf("\n");
3503 nlines++;
3504
3505 if ((entry->prev == &map->header) ||
3506 (entry->prev->object.vm_object !=
3507 entry->object.vm_object)) {
3508 db_indent += 2;
3509 vm_object_print((db_expr_t)(intptr_t)
3510 entry->object.vm_object,
3511 full, 0, (char *)0);
3512 nlines += 4;
3513 db_indent -= 2;
3514 }
3515 }
3516 }
3517 db_indent -= 2;
3518 if (db_indent == 0)
3519 nlines = 0;
3520}
3521
3522
3523DB_SHOW_COMMAND(procvm, procvm)
3524{
3525 struct proc *p;
3526
3527 if (have_addr) {
3528 p = (struct proc *) addr;
3529 } else {
3530 p = curproc;
3531 }
3532
3533 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3534 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3535 (void *)vmspace_pmap(p->p_vmspace));
3536
3537 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3538}
3539
3540#endif /* DDB */