FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_map.c
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 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
33 *
34 *
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
37 *
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39 *
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
45 *
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49 *
50 * Carnegie Mellon requests users of this software to return to
51 *
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
56 *
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
59 */
60
61 /*
62 * Virtual memory mapping module.
63 */
64
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD: releng/6.2/sys/vm/vm_map.c 159155 2006-06-02 00:26:27Z tegge $");
67
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/ktr.h>
71 #include <sys/lock.h>
72 #include <sys/mutex.h>
73 #include <sys/proc.h>
74 #include <sys/vmmeter.h>
75 #include <sys/mman.h>
76 #include <sys/vnode.h>
77 #include <sys/resourcevar.h>
78 #include <sys/file.h>
79 #include <sys/sysent.h>
80 #include <sys/shm.h>
81
82 #include <vm/vm.h>
83 #include <vm/vm_param.h>
84 #include <vm/pmap.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_page.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_pager.h>
89 #include <vm/vm_kern.h>
90 #include <vm/vm_extern.h>
91 #include <vm/swap_pager.h>
92 #include <vm/uma.h>
93
94 /*
95 * Virtual memory maps provide for the mapping, protection,
96 * and sharing of virtual memory objects. In addition,
97 * this module provides for an efficient virtual copy of
98 * memory from one map to another.
99 *
100 * Synchronization is required prior to most operations.
101 *
102 * Maps consist of an ordered doubly-linked list of simple
103 * entries; a single hint is used to speed up lookups.
104 *
105 * Since portions of maps are specified by start/end addresses,
106 * which may not align with existing map entries, all
107 * routines merely "clip" entries to these start/end values.
108 * [That is, an entry is split into two, bordering at a
109 * start or end value.] Note that these clippings may not
110 * always be necessary (as the two resulting entries are then
111 * not changed); however, the clipping is done for convenience.
112 *
113 * As mentioned above, virtual copy operations are performed
114 * by copying VM object references from one map to
115 * another, and then marking both regions as copy-on-write.
116 */
117
118 /*
119 * vm_map_startup:
120 *
121 * Initialize the vm_map module. Must be called before
122 * any other vm_map routines.
123 *
124 * Map and entry structures are allocated from the general
125 * purpose memory pool with some exceptions:
126 *
127 * - The kernel map and kmem submap are allocated statically.
128 * - Kernel map entries are allocated out of a static pool.
129 *
130 * These restrictions are necessary since malloc() uses the
131 * maps and requires map entries.
132 */
133
134 static struct mtx map_sleep_mtx;
135 static uma_zone_t mapentzone;
136 static uma_zone_t kmapentzone;
137 static uma_zone_t mapzone;
138 static uma_zone_t vmspace_zone;
139 static struct vm_object kmapentobj;
140 static int vmspace_zinit(void *mem, int size, int flags);
141 static void vmspace_zfini(void *mem, int size);
142 static int vm_map_zinit(void *mem, int ize, int flags);
143 static void vm_map_zfini(void *mem, int size);
144 static void _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max);
145
146 #ifdef INVARIANTS
147 static void vm_map_zdtor(void *mem, int size, void *arg);
148 static void vmspace_zdtor(void *mem, int size, void *arg);
149 #endif
150
151 /*
152 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
153 * stable.
154 */
155 #define PROC_VMSPACE_LOCK(p) do { } while (0)
156 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
157
158 void
159 vm_map_startup(void)
160 {
161 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
162 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
163 #ifdef INVARIANTS
164 vm_map_zdtor,
165 #else
166 NULL,
167 #endif
168 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
169 uma_prealloc(mapzone, MAX_KMAP);
170 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
171 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
172 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
173 uma_prealloc(kmapentzone, MAX_KMAPENT);
174 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
175 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
176 uma_prealloc(mapentzone, MAX_MAPENT);
177 }
178
179 static void
180 vmspace_zfini(void *mem, int size)
181 {
182 struct vmspace *vm;
183
184 vm = (struct vmspace *)mem;
185 pmap_release(vmspace_pmap(vm));
186 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map));
187 }
188
189 static int
190 vmspace_zinit(void *mem, int size, int flags)
191 {
192 struct vmspace *vm;
193
194 vm = (struct vmspace *)mem;
195
196 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
197 pmap_pinit(vmspace_pmap(vm));
198 return (0);
199 }
200
201 static void
202 vm_map_zfini(void *mem, int size)
203 {
204 vm_map_t map;
205
206 map = (vm_map_t)mem;
207 mtx_destroy(&map->system_mtx);
208 sx_destroy(&map->lock);
209 }
210
211 static int
212 vm_map_zinit(void *mem, int size, int flags)
213 {
214 vm_map_t map;
215
216 map = (vm_map_t)mem;
217 map->nentries = 0;
218 map->size = 0;
219 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
220 sx_init(&map->lock, "user map");
221 return (0);
222 }
223
224 #ifdef INVARIANTS
225 static void
226 vmspace_zdtor(void *mem, int size, void *arg)
227 {
228 struct vmspace *vm;
229
230 vm = (struct vmspace *)mem;
231
232 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
233 }
234 static void
235 vm_map_zdtor(void *mem, int size, void *arg)
236 {
237 vm_map_t map;
238
239 map = (vm_map_t)mem;
240 KASSERT(map->nentries == 0,
241 ("map %p nentries == %d on free.",
242 map, map->nentries));
243 KASSERT(map->size == 0,
244 ("map %p size == %lu on free.",
245 map, (unsigned long)map->size));
246 }
247 #endif /* INVARIANTS */
248
249 /*
250 * Allocate a vmspace structure, including a vm_map and pmap,
251 * and initialize those structures. The refcnt is set to 1.
252 */
253 struct vmspace *
254 vmspace_alloc(min, max)
255 vm_offset_t min, max;
256 {
257 struct vmspace *vm;
258
259 vm = uma_zalloc(vmspace_zone, M_WAITOK);
260 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
261 _vm_map_init(&vm->vm_map, min, max);
262 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
263 vm->vm_refcnt = 1;
264 vm->vm_shm = NULL;
265 vm->vm_swrss = 0;
266 vm->vm_tsize = 0;
267 vm->vm_dsize = 0;
268 vm->vm_ssize = 0;
269 vm->vm_taddr = 0;
270 vm->vm_daddr = 0;
271 vm->vm_maxsaddr = 0;
272 return (vm);
273 }
274
275 void
276 vm_init2(void)
277 {
278 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count,
279 (VM_MAX_KERNEL_ADDRESS - KERNBASE) / PAGE_SIZE) / 8 +
280 maxproc * 2 + maxfiles);
281 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
282 #ifdef INVARIANTS
283 vmspace_zdtor,
284 #else
285 NULL,
286 #endif
287 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
288 pmap_init2();
289 }
290
291 static __inline void
292 vmspace_dofree(struct vmspace *vm)
293 {
294 CTR1(KTR_VM, "vmspace_free: %p", vm);
295
296 /*
297 * Make sure any SysV shm is freed, it might not have been in
298 * exit1().
299 */
300 shmexit(vm);
301
302 /*
303 * Lock the map, to wait out all other references to it.
304 * Delete all of the mappings and pages they hold, then call
305 * the pmap module to reclaim anything left.
306 */
307 vm_map_lock(&vm->vm_map);
308 (void) vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
309 vm->vm_map.max_offset);
310 vm_map_unlock(&vm->vm_map);
311
312 uma_zfree(vmspace_zone, vm);
313 }
314
315 void
316 vmspace_free(struct vmspace *vm)
317 {
318 int refcnt;
319
320 if (vm->vm_refcnt == 0)
321 panic("vmspace_free: attempt to free already freed vmspace");
322
323 do
324 refcnt = vm->vm_refcnt;
325 while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
326 if (refcnt == 1)
327 vmspace_dofree(vm);
328 }
329
330 void
331 vmspace_exitfree(struct proc *p)
332 {
333 struct vmspace *vm;
334
335 PROC_VMSPACE_LOCK(p);
336 vm = p->p_vmspace;
337 p->p_vmspace = NULL;
338 PROC_VMSPACE_UNLOCK(p);
339 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
340 vmspace_free(vm);
341 }
342
343 void
344 vmspace_exit(struct thread *td)
345 {
346 int refcnt;
347 struct vmspace *vm;
348 struct proc *p;
349
350 /*
351 * Release user portion of address space.
352 * This releases references to vnodes,
353 * which could cause I/O if the file has been unlinked.
354 * Need to do this early enough that we can still sleep.
355 *
356 * The last exiting process to reach this point releases as
357 * much of the environment as it can. vmspace_dofree() is the
358 * slower fallback in case another process had a temporary
359 * reference to the vmspace.
360 */
361
362 p = td->td_proc;
363 vm = p->p_vmspace;
364 atomic_add_int(&vmspace0.vm_refcnt, 1);
365 do {
366 refcnt = vm->vm_refcnt;
367 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
368 /* Switch now since other proc might free vmspace */
369 PROC_VMSPACE_LOCK(p);
370 p->p_vmspace = &vmspace0;
371 PROC_VMSPACE_UNLOCK(p);
372 pmap_activate(td);
373 }
374 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
375 if (refcnt == 1) {
376 if (p->p_vmspace != vm) {
377 /* vmspace not yet freed, switch back */
378 PROC_VMSPACE_LOCK(p);
379 p->p_vmspace = vm;
380 PROC_VMSPACE_UNLOCK(p);
381 pmap_activate(td);
382 }
383 pmap_remove_pages(vmspace_pmap(vm),
384 vm_map_min(&vm->vm_map),
385 vm_map_max(&vm->vm_map));
386 /* Switch now since this proc will free vmspace */
387 PROC_VMSPACE_LOCK(p);
388 p->p_vmspace = &vmspace0;
389 PROC_VMSPACE_UNLOCK(p);
390 pmap_activate(td);
391 vmspace_dofree(vm);
392 }
393 }
394
395 /* Acquire reference to vmspace owned by another process. */
396
397 struct vmspace *
398 vmspace_acquire_ref(struct proc *p)
399 {
400 struct vmspace *vm;
401 int refcnt;
402
403 PROC_VMSPACE_LOCK(p);
404 vm = p->p_vmspace;
405 if (vm == NULL) {
406 PROC_VMSPACE_UNLOCK(p);
407 return (NULL);
408 }
409 do {
410 refcnt = vm->vm_refcnt;
411 if (refcnt <= 0) { /* Avoid 0->1 transition */
412 PROC_VMSPACE_UNLOCK(p);
413 return (NULL);
414 }
415 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
416 if (vm != p->p_vmspace) {
417 PROC_VMSPACE_UNLOCK(p);
418 vmspace_free(vm);
419 return (NULL);
420 }
421 PROC_VMSPACE_UNLOCK(p);
422 return (vm);
423 }
424
425 void
426 _vm_map_lock(vm_map_t map, const char *file, int line)
427 {
428
429 if (map->system_map)
430 _mtx_lock_flags(&map->system_mtx, 0, file, line);
431 else
432 _sx_xlock(&map->lock, file, line);
433 map->timestamp++;
434 }
435
436 void
437 _vm_map_unlock(vm_map_t map, const char *file, int line)
438 {
439
440 if (map->system_map)
441 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
442 else
443 _sx_xunlock(&map->lock, file, line);
444 }
445
446 void
447 _vm_map_lock_read(vm_map_t map, const char *file, int line)
448 {
449
450 if (map->system_map)
451 _mtx_lock_flags(&map->system_mtx, 0, file, line);
452 else
453 _sx_xlock(&map->lock, file, line);
454 }
455
456 void
457 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
458 {
459
460 if (map->system_map)
461 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
462 else
463 _sx_xunlock(&map->lock, file, line);
464 }
465
466 int
467 _vm_map_trylock(vm_map_t map, const char *file, int line)
468 {
469 int error;
470
471 error = map->system_map ?
472 !_mtx_trylock(&map->system_mtx, 0, file, line) :
473 !_sx_try_xlock(&map->lock, file, line);
474 if (error == 0)
475 map->timestamp++;
476 return (error == 0);
477 }
478
479 int
480 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
481 {
482 int error;
483
484 error = map->system_map ?
485 !_mtx_trylock(&map->system_mtx, 0, file, line) :
486 !_sx_try_xlock(&map->lock, file, line);
487 return (error == 0);
488 }
489
490 int
491 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
492 {
493
494 #ifdef INVARIANTS
495 if (map->system_map) {
496 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
497 } else
498 _sx_assert(&map->lock, SX_XLOCKED, file, line);
499 #endif
500 map->timestamp++;
501 return (0);
502 }
503
504 void
505 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
506 {
507
508 #ifdef INVARIANTS
509 if (map->system_map) {
510 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
511 } else
512 _sx_assert(&map->lock, SX_XLOCKED, file, line);
513 #endif
514 }
515
516 /*
517 * vm_map_unlock_and_wait:
518 */
519 int
520 vm_map_unlock_and_wait(vm_map_t map, boolean_t user_wait)
521 {
522
523 mtx_lock(&map_sleep_mtx);
524 vm_map_unlock(map);
525 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 0));
526 }
527
528 /*
529 * vm_map_wakeup:
530 */
531 void
532 vm_map_wakeup(vm_map_t map)
533 {
534
535 /*
536 * Acquire and release map_sleep_mtx to prevent a wakeup()
537 * from being performed (and lost) between the vm_map_unlock()
538 * and the msleep() in vm_map_unlock_and_wait().
539 */
540 mtx_lock(&map_sleep_mtx);
541 mtx_unlock(&map_sleep_mtx);
542 wakeup(&map->root);
543 }
544
545 long
546 vmspace_resident_count(struct vmspace *vmspace)
547 {
548 return pmap_resident_count(vmspace_pmap(vmspace));
549 }
550
551 long
552 vmspace_wired_count(struct vmspace *vmspace)
553 {
554 return pmap_wired_count(vmspace_pmap(vmspace));
555 }
556
557 /*
558 * vm_map_create:
559 *
560 * Creates and returns a new empty VM map with
561 * the given physical map structure, and having
562 * the given lower and upper address bounds.
563 */
564 vm_map_t
565 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
566 {
567 vm_map_t result;
568
569 result = uma_zalloc(mapzone, M_WAITOK);
570 CTR1(KTR_VM, "vm_map_create: %p", result);
571 _vm_map_init(result, min, max);
572 result->pmap = pmap;
573 return (result);
574 }
575
576 /*
577 * Initialize an existing vm_map structure
578 * such as that in the vmspace structure.
579 * The pmap is set elsewhere.
580 */
581 static void
582 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
583 {
584
585 map->header.next = map->header.prev = &map->header;
586 map->needs_wakeup = FALSE;
587 map->system_map = 0;
588 map->min_offset = min;
589 map->max_offset = max;
590 map->flags = 0;
591 map->root = NULL;
592 map->timestamp = 0;
593 }
594
595 void
596 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
597 {
598 _vm_map_init(map, min, max);
599 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
600 sx_init(&map->lock, "user map");
601 }
602
603 /*
604 * vm_map_entry_dispose: [ internal use only ]
605 *
606 * Inverse of vm_map_entry_create.
607 */
608 static void
609 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
610 {
611 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
612 }
613
614 /*
615 * vm_map_entry_create: [ internal use only ]
616 *
617 * Allocates a VM map entry for insertion.
618 * No entry fields are filled in.
619 */
620 static vm_map_entry_t
621 vm_map_entry_create(vm_map_t map)
622 {
623 vm_map_entry_t new_entry;
624
625 if (map->system_map)
626 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
627 else
628 new_entry = uma_zalloc(mapentzone, M_WAITOK);
629 if (new_entry == NULL)
630 panic("vm_map_entry_create: kernel resources exhausted");
631 return (new_entry);
632 }
633
634 /*
635 * vm_map_entry_set_behavior:
636 *
637 * Set the expected access behavior, either normal, random, or
638 * sequential.
639 */
640 static __inline void
641 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
642 {
643 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
644 (behavior & MAP_ENTRY_BEHAV_MASK);
645 }
646
647 /*
648 * vm_map_entry_set_max_free:
649 *
650 * Set the max_free field in a vm_map_entry.
651 */
652 static __inline void
653 vm_map_entry_set_max_free(vm_map_entry_t entry)
654 {
655
656 entry->max_free = entry->adj_free;
657 if (entry->left != NULL && entry->left->max_free > entry->max_free)
658 entry->max_free = entry->left->max_free;
659 if (entry->right != NULL && entry->right->max_free > entry->max_free)
660 entry->max_free = entry->right->max_free;
661 }
662
663 /*
664 * vm_map_entry_splay:
665 *
666 * The Sleator and Tarjan top-down splay algorithm with the
667 * following variation. Max_free must be computed bottom-up, so
668 * on the downward pass, maintain the left and right spines in
669 * reverse order. Then, make a second pass up each side to fix
670 * the pointers and compute max_free. The time bound is O(log n)
671 * amortized.
672 *
673 * The new root is the vm_map_entry containing "addr", or else an
674 * adjacent entry (lower or higher) if addr is not in the tree.
675 *
676 * The map must be locked, and leaves it so.
677 *
678 * Returns: the new root.
679 */
680 static vm_map_entry_t
681 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
682 {
683 vm_map_entry_t llist, rlist;
684 vm_map_entry_t ltree, rtree;
685 vm_map_entry_t y;
686
687 /* Special case of empty tree. */
688 if (root == NULL)
689 return (root);
690
691 /*
692 * Pass One: Splay down the tree until we find addr or a NULL
693 * pointer where addr would go. llist and rlist are the two
694 * sides in reverse order (bottom-up), with llist linked by
695 * the right pointer and rlist linked by the left pointer in
696 * the vm_map_entry. Wait until Pass Two to set max_free on
697 * the two spines.
698 */
699 llist = NULL;
700 rlist = NULL;
701 for (;;) {
702 /* root is never NULL in here. */
703 if (addr < root->start) {
704 y = root->left;
705 if (y == NULL)
706 break;
707 if (addr < y->start && y->left != NULL) {
708 /* Rotate right and put y on rlist. */
709 root->left = y->right;
710 y->right = root;
711 vm_map_entry_set_max_free(root);
712 root = y->left;
713 y->left = rlist;
714 rlist = y;
715 } else {
716 /* Put root on rlist. */
717 root->left = rlist;
718 rlist = root;
719 root = y;
720 }
721 } else {
722 y = root->right;
723 if (addr < root->end || y == NULL)
724 break;
725 if (addr >= y->end && y->right != NULL) {
726 /* Rotate left and put y on llist. */
727 root->right = y->left;
728 y->left = root;
729 vm_map_entry_set_max_free(root);
730 root = y->right;
731 y->right = llist;
732 llist = y;
733 } else {
734 /* Put root on llist. */
735 root->right = llist;
736 llist = root;
737 root = y;
738 }
739 }
740 }
741
742 /*
743 * Pass Two: Walk back up the two spines, flip the pointers
744 * and set max_free. The subtrees of the root go at the
745 * bottom of llist and rlist.
746 */
747 ltree = root->left;
748 while (llist != NULL) {
749 y = llist->right;
750 llist->right = ltree;
751 vm_map_entry_set_max_free(llist);
752 ltree = llist;
753 llist = y;
754 }
755 rtree = root->right;
756 while (rlist != NULL) {
757 y = rlist->left;
758 rlist->left = rtree;
759 vm_map_entry_set_max_free(rlist);
760 rtree = rlist;
761 rlist = y;
762 }
763
764 /*
765 * Final assembly: add ltree and rtree as subtrees of root.
766 */
767 root->left = ltree;
768 root->right = rtree;
769 vm_map_entry_set_max_free(root);
770
771 return (root);
772 }
773
774 /*
775 * vm_map_entry_{un,}link:
776 *
777 * Insert/remove entries from maps.
778 */
779 static void
780 vm_map_entry_link(vm_map_t map,
781 vm_map_entry_t after_where,
782 vm_map_entry_t entry)
783 {
784
785 CTR4(KTR_VM,
786 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
787 map->nentries, entry, after_where);
788 map->nentries++;
789 entry->prev = after_where;
790 entry->next = after_where->next;
791 entry->next->prev = entry;
792 after_where->next = entry;
793
794 if (after_where != &map->header) {
795 if (after_where != map->root)
796 vm_map_entry_splay(after_where->start, map->root);
797 entry->right = after_where->right;
798 entry->left = after_where;
799 after_where->right = NULL;
800 after_where->adj_free = entry->start - after_where->end;
801 vm_map_entry_set_max_free(after_where);
802 } else {
803 entry->right = map->root;
804 entry->left = NULL;
805 }
806 entry->adj_free = (entry->next == &map->header ? map->max_offset :
807 entry->next->start) - entry->end;
808 vm_map_entry_set_max_free(entry);
809 map->root = entry;
810 }
811
812 static void
813 vm_map_entry_unlink(vm_map_t map,
814 vm_map_entry_t entry)
815 {
816 vm_map_entry_t next, prev, root;
817
818 if (entry != map->root)
819 vm_map_entry_splay(entry->start, map->root);
820 if (entry->left == NULL)
821 root = entry->right;
822 else {
823 root = vm_map_entry_splay(entry->start, entry->left);
824 root->right = entry->right;
825 root->adj_free = (entry->next == &map->header ? map->max_offset :
826 entry->next->start) - root->end;
827 vm_map_entry_set_max_free(root);
828 }
829 map->root = root;
830
831 prev = entry->prev;
832 next = entry->next;
833 next->prev = prev;
834 prev->next = next;
835 map->nentries--;
836 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
837 map->nentries, entry);
838 }
839
840 /*
841 * vm_map_entry_resize_free:
842 *
843 * Recompute the amount of free space following a vm_map_entry
844 * and propagate that value up the tree. Call this function after
845 * resizing a map entry in-place, that is, without a call to
846 * vm_map_entry_link() or _unlink().
847 *
848 * The map must be locked, and leaves it so.
849 */
850 static void
851 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
852 {
853
854 /*
855 * Using splay trees without parent pointers, propagating
856 * max_free up the tree is done by moving the entry to the
857 * root and making the change there.
858 */
859 if (entry != map->root)
860 map->root = vm_map_entry_splay(entry->start, map->root);
861
862 entry->adj_free = (entry->next == &map->header ? map->max_offset :
863 entry->next->start) - entry->end;
864 vm_map_entry_set_max_free(entry);
865 }
866
867 /*
868 * vm_map_lookup_entry: [ internal use only ]
869 *
870 * Finds the map entry containing (or
871 * immediately preceding) the specified address
872 * in the given map; the entry is returned
873 * in the "entry" parameter. The boolean
874 * result indicates whether the address is
875 * actually contained in the map.
876 */
877 boolean_t
878 vm_map_lookup_entry(
879 vm_map_t map,
880 vm_offset_t address,
881 vm_map_entry_t *entry) /* OUT */
882 {
883 vm_map_entry_t cur;
884
885 cur = vm_map_entry_splay(address, map->root);
886 if (cur == NULL)
887 *entry = &map->header;
888 else {
889 map->root = cur;
890
891 if (address >= cur->start) {
892 *entry = cur;
893 if (cur->end > address)
894 return (TRUE);
895 } else
896 *entry = cur->prev;
897 }
898 return (FALSE);
899 }
900
901 /*
902 * vm_map_insert:
903 *
904 * Inserts the given whole VM object into the target
905 * map at the specified address range. The object's
906 * size should match that of the address range.
907 *
908 * Requires that the map be locked, and leaves it so.
909 *
910 * If object is non-NULL, ref count must be bumped by caller
911 * prior to making call to account for the new entry.
912 */
913 int
914 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
915 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
916 int cow)
917 {
918 vm_map_entry_t new_entry;
919 vm_map_entry_t prev_entry;
920 vm_map_entry_t temp_entry;
921 vm_eflags_t protoeflags;
922
923 /*
924 * Check that the start and end points are not bogus.
925 */
926 if ((start < map->min_offset) || (end > map->max_offset) ||
927 (start >= end))
928 return (KERN_INVALID_ADDRESS);
929
930 /*
931 * Find the entry prior to the proposed starting address; if it's part
932 * of an existing entry, this range is bogus.
933 */
934 if (vm_map_lookup_entry(map, start, &temp_entry))
935 return (KERN_NO_SPACE);
936
937 prev_entry = temp_entry;
938
939 /*
940 * Assert that the next entry doesn't overlap the end point.
941 */
942 if ((prev_entry->next != &map->header) &&
943 (prev_entry->next->start < end))
944 return (KERN_NO_SPACE);
945
946 protoeflags = 0;
947
948 if (cow & MAP_COPY_ON_WRITE)
949 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
950
951 if (cow & MAP_NOFAULT) {
952 protoeflags |= MAP_ENTRY_NOFAULT;
953
954 KASSERT(object == NULL,
955 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
956 }
957 if (cow & MAP_DISABLE_SYNCER)
958 protoeflags |= MAP_ENTRY_NOSYNC;
959 if (cow & MAP_DISABLE_COREDUMP)
960 protoeflags |= MAP_ENTRY_NOCOREDUMP;
961
962 if (object != NULL) {
963 /*
964 * OBJ_ONEMAPPING must be cleared unless this mapping
965 * is trivially proven to be the only mapping for any
966 * of the object's pages. (Object granularity
967 * reference counting is insufficient to recognize
968 * aliases with precision.)
969 */
970 VM_OBJECT_LOCK(object);
971 if (object->ref_count > 1 || object->shadow_count != 0)
972 vm_object_clear_flag(object, OBJ_ONEMAPPING);
973 VM_OBJECT_UNLOCK(object);
974 }
975 else if ((prev_entry != &map->header) &&
976 (prev_entry->eflags == protoeflags) &&
977 (prev_entry->end == start) &&
978 (prev_entry->wired_count == 0) &&
979 ((prev_entry->object.vm_object == NULL) ||
980 vm_object_coalesce(prev_entry->object.vm_object,
981 prev_entry->offset,
982 (vm_size_t)(prev_entry->end - prev_entry->start),
983 (vm_size_t)(end - prev_entry->end)))) {
984 /*
985 * We were able to extend the object. Determine if we
986 * can extend the previous map entry to include the
987 * new range as well.
988 */
989 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
990 (prev_entry->protection == prot) &&
991 (prev_entry->max_protection == max)) {
992 map->size += (end - prev_entry->end);
993 prev_entry->end = end;
994 vm_map_entry_resize_free(map, prev_entry);
995 vm_map_simplify_entry(map, prev_entry);
996 return (KERN_SUCCESS);
997 }
998
999 /*
1000 * If we can extend the object but cannot extend the
1001 * map entry, we have to create a new map entry. We
1002 * must bump the ref count on the extended object to
1003 * account for it. object may be NULL.
1004 */
1005 object = prev_entry->object.vm_object;
1006 offset = prev_entry->offset +
1007 (prev_entry->end - prev_entry->start);
1008 vm_object_reference(object);
1009 }
1010
1011 /*
1012 * NOTE: if conditionals fail, object can be NULL here. This occurs
1013 * in things like the buffer map where we manage kva but do not manage
1014 * backing objects.
1015 */
1016
1017 /*
1018 * Create a new entry
1019 */
1020 new_entry = vm_map_entry_create(map);
1021 new_entry->start = start;
1022 new_entry->end = end;
1023
1024 new_entry->eflags = protoeflags;
1025 new_entry->object.vm_object = object;
1026 new_entry->offset = offset;
1027 new_entry->avail_ssize = 0;
1028
1029 new_entry->inheritance = VM_INHERIT_DEFAULT;
1030 new_entry->protection = prot;
1031 new_entry->max_protection = max;
1032 new_entry->wired_count = 0;
1033
1034 /*
1035 * Insert the new entry into the list
1036 */
1037 vm_map_entry_link(map, prev_entry, new_entry);
1038 map->size += new_entry->end - new_entry->start;
1039
1040 #if 0
1041 /*
1042 * Temporarily removed to avoid MAP_STACK panic, due to
1043 * MAP_STACK being a huge hack. Will be added back in
1044 * when MAP_STACK (and the user stack mapping) is fixed.
1045 */
1046 /*
1047 * It may be possible to simplify the entry
1048 */
1049 vm_map_simplify_entry(map, new_entry);
1050 #endif
1051
1052 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1053 vm_map_pmap_enter(map, start, prot,
1054 object, OFF_TO_IDX(offset), end - start,
1055 cow & MAP_PREFAULT_PARTIAL);
1056 }
1057
1058 return (KERN_SUCCESS);
1059 }
1060
1061 /*
1062 * vm_map_findspace:
1063 *
1064 * Find the first fit (lowest VM address) for "length" free bytes
1065 * beginning at address >= start in the given map.
1066 *
1067 * In a vm_map_entry, "adj_free" is the amount of free space
1068 * adjacent (higher address) to this entry, and "max_free" is the
1069 * maximum amount of contiguous free space in its subtree. This
1070 * allows finding a free region in one path down the tree, so
1071 * O(log n) amortized with splay trees.
1072 *
1073 * The map must be locked, and leaves it so.
1074 *
1075 * Returns: 0 on success, and starting address in *addr,
1076 * 1 if insufficient space.
1077 */
1078 int
1079 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1080 vm_offset_t *addr) /* OUT */
1081 {
1082 vm_map_entry_t entry;
1083 vm_offset_t end, st;
1084
1085 /*
1086 * Request must fit within min/max VM address and must avoid
1087 * address wrap.
1088 */
1089 if (start < map->min_offset)
1090 start = map->min_offset;
1091 if (start + length > map->max_offset || start + length < start)
1092 return (1);
1093
1094 /* Empty tree means wide open address space. */
1095 if (map->root == NULL) {
1096 *addr = start;
1097 goto found;
1098 }
1099
1100 /*
1101 * After splay, if start comes before root node, then there
1102 * must be a gap from start to the root.
1103 */
1104 map->root = vm_map_entry_splay(start, map->root);
1105 if (start + length <= map->root->start) {
1106 *addr = start;
1107 goto found;
1108 }
1109
1110 /*
1111 * Root is the last node that might begin its gap before
1112 * start, and this is the last comparison where address
1113 * wrap might be a problem.
1114 */
1115 st = (start > map->root->end) ? start : map->root->end;
1116 if (length <= map->root->end + map->root->adj_free - st) {
1117 *addr = st;
1118 goto found;
1119 }
1120
1121 /* With max_free, can immediately tell if no solution. */
1122 entry = map->root->right;
1123 if (entry == NULL || length > entry->max_free)
1124 return (1);
1125
1126 /*
1127 * Search the right subtree in the order: left subtree, root,
1128 * right subtree (first fit). The previous splay implies that
1129 * all regions in the right subtree have addresses > start.
1130 */
1131 while (entry != NULL) {
1132 if (entry->left != NULL && entry->left->max_free >= length)
1133 entry = entry->left;
1134 else if (entry->adj_free >= length) {
1135 *addr = entry->end;
1136 goto found;
1137 } else
1138 entry = entry->right;
1139 }
1140
1141 /* Can't get here, so panic if we do. */
1142 panic("vm_map_findspace: max_free corrupt");
1143
1144 found:
1145 /* Expand the kernel pmap, if necessary. */
1146 if (map == kernel_map) {
1147 end = round_page(*addr + length);
1148 if (end > kernel_vm_end)
1149 pmap_growkernel(end);
1150 }
1151 return (0);
1152 }
1153
1154 /*
1155 * vm_map_find finds an unallocated region in the target address
1156 * map with the given length. The search is defined to be
1157 * first-fit from the specified address; the region found is
1158 * returned in the same parameter.
1159 *
1160 * If object is non-NULL, ref count must be bumped by caller
1161 * prior to making call to account for the new entry.
1162 */
1163 int
1164 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1165 vm_offset_t *addr, /* IN/OUT */
1166 vm_size_t length, boolean_t find_space, vm_prot_t prot,
1167 vm_prot_t max, int cow)
1168 {
1169 vm_offset_t start;
1170 int result;
1171
1172 start = *addr;
1173 vm_map_lock(map);
1174 if (find_space) {
1175 if (vm_map_findspace(map, start, length, addr)) {
1176 vm_map_unlock(map);
1177 return (KERN_NO_SPACE);
1178 }
1179 start = *addr;
1180 }
1181 result = vm_map_insert(map, object, offset,
1182 start, start + length, prot, max, cow);
1183 vm_map_unlock(map);
1184 return (result);
1185 }
1186
1187 /*
1188 * vm_map_simplify_entry:
1189 *
1190 * Simplify the given map entry by merging with either neighbor. This
1191 * routine also has the ability to merge with both neighbors.
1192 *
1193 * The map must be locked.
1194 *
1195 * This routine guarentees that the passed entry remains valid (though
1196 * possibly extended). When merging, this routine may delete one or
1197 * both neighbors.
1198 */
1199 void
1200 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1201 {
1202 vm_map_entry_t next, prev;
1203 vm_size_t prevsize, esize;
1204
1205 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1206 return;
1207
1208 prev = entry->prev;
1209 if (prev != &map->header) {
1210 prevsize = prev->end - prev->start;
1211 if ( (prev->end == entry->start) &&
1212 (prev->object.vm_object == entry->object.vm_object) &&
1213 (!prev->object.vm_object ||
1214 (prev->offset + prevsize == entry->offset)) &&
1215 (prev->eflags == entry->eflags) &&
1216 (prev->protection == entry->protection) &&
1217 (prev->max_protection == entry->max_protection) &&
1218 (prev->inheritance == entry->inheritance) &&
1219 (prev->wired_count == entry->wired_count)) {
1220 vm_map_entry_unlink(map, prev);
1221 entry->start = prev->start;
1222 entry->offset = prev->offset;
1223 if (entry->prev != &map->header)
1224 vm_map_entry_resize_free(map, entry->prev);
1225 if (prev->object.vm_object)
1226 vm_object_deallocate(prev->object.vm_object);
1227 vm_map_entry_dispose(map, prev);
1228 }
1229 }
1230
1231 next = entry->next;
1232 if (next != &map->header) {
1233 esize = entry->end - entry->start;
1234 if ((entry->end == next->start) &&
1235 (next->object.vm_object == entry->object.vm_object) &&
1236 (!entry->object.vm_object ||
1237 (entry->offset + esize == next->offset)) &&
1238 (next->eflags == entry->eflags) &&
1239 (next->protection == entry->protection) &&
1240 (next->max_protection == entry->max_protection) &&
1241 (next->inheritance == entry->inheritance) &&
1242 (next->wired_count == entry->wired_count)) {
1243 vm_map_entry_unlink(map, next);
1244 entry->end = next->end;
1245 vm_map_entry_resize_free(map, entry);
1246 if (next->object.vm_object)
1247 vm_object_deallocate(next->object.vm_object);
1248 vm_map_entry_dispose(map, next);
1249 }
1250 }
1251 }
1252 /*
1253 * vm_map_clip_start: [ internal use only ]
1254 *
1255 * Asserts that the given entry begins at or after
1256 * the specified address; if necessary,
1257 * it splits the entry into two.
1258 */
1259 #define vm_map_clip_start(map, entry, startaddr) \
1260 { \
1261 if (startaddr > entry->start) \
1262 _vm_map_clip_start(map, entry, startaddr); \
1263 }
1264
1265 /*
1266 * This routine is called only when it is known that
1267 * the entry must be split.
1268 */
1269 static void
1270 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1271 {
1272 vm_map_entry_t new_entry;
1273
1274 /*
1275 * Split off the front portion -- note that we must insert the new
1276 * entry BEFORE this one, so that this entry has the specified
1277 * starting address.
1278 */
1279 vm_map_simplify_entry(map, entry);
1280
1281 /*
1282 * If there is no object backing this entry, we might as well create
1283 * one now. If we defer it, an object can get created after the map
1284 * is clipped, and individual objects will be created for the split-up
1285 * map. This is a bit of a hack, but is also about the best place to
1286 * put this improvement.
1287 */
1288 if (entry->object.vm_object == NULL && !map->system_map) {
1289 vm_object_t object;
1290 object = vm_object_allocate(OBJT_DEFAULT,
1291 atop(entry->end - entry->start));
1292 entry->object.vm_object = object;
1293 entry->offset = 0;
1294 }
1295
1296 new_entry = vm_map_entry_create(map);
1297 *new_entry = *entry;
1298
1299 new_entry->end = start;
1300 entry->offset += (start - entry->start);
1301 entry->start = start;
1302
1303 vm_map_entry_link(map, entry->prev, new_entry);
1304
1305 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1306 vm_object_reference(new_entry->object.vm_object);
1307 }
1308 }
1309
1310 /*
1311 * vm_map_clip_end: [ internal use only ]
1312 *
1313 * Asserts that the given entry ends at or before
1314 * the specified address; if necessary,
1315 * it splits the entry into two.
1316 */
1317 #define vm_map_clip_end(map, entry, endaddr) \
1318 { \
1319 if ((endaddr) < (entry->end)) \
1320 _vm_map_clip_end((map), (entry), (endaddr)); \
1321 }
1322
1323 /*
1324 * This routine is called only when it is known that
1325 * the entry must be split.
1326 */
1327 static void
1328 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1329 {
1330 vm_map_entry_t new_entry;
1331
1332 /*
1333 * If there is no object backing this entry, we might as well create
1334 * one now. If we defer it, an object can get created after the map
1335 * is clipped, and individual objects will be created for the split-up
1336 * map. This is a bit of a hack, but is also about the best place to
1337 * put this improvement.
1338 */
1339 if (entry->object.vm_object == NULL && !map->system_map) {
1340 vm_object_t object;
1341 object = vm_object_allocate(OBJT_DEFAULT,
1342 atop(entry->end - entry->start));
1343 entry->object.vm_object = object;
1344 entry->offset = 0;
1345 }
1346
1347 /*
1348 * Create a new entry and insert it AFTER the specified entry
1349 */
1350 new_entry = vm_map_entry_create(map);
1351 *new_entry = *entry;
1352
1353 new_entry->start = entry->end = end;
1354 new_entry->offset += (end - entry->start);
1355
1356 vm_map_entry_link(map, entry, new_entry);
1357
1358 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1359 vm_object_reference(new_entry->object.vm_object);
1360 }
1361 }
1362
1363 /*
1364 * VM_MAP_RANGE_CHECK: [ internal use only ]
1365 *
1366 * Asserts that the starting and ending region
1367 * addresses fall within the valid range of the map.
1368 */
1369 #define VM_MAP_RANGE_CHECK(map, start, end) \
1370 { \
1371 if (start < vm_map_min(map)) \
1372 start = vm_map_min(map); \
1373 if (end > vm_map_max(map)) \
1374 end = vm_map_max(map); \
1375 if (start > end) \
1376 start = end; \
1377 }
1378
1379 /*
1380 * vm_map_submap: [ kernel use only ]
1381 *
1382 * Mark the given range as handled by a subordinate map.
1383 *
1384 * This range must have been created with vm_map_find,
1385 * and no other operations may have been performed on this
1386 * range prior to calling vm_map_submap.
1387 *
1388 * Only a limited number of operations can be performed
1389 * within this rage after calling vm_map_submap:
1390 * vm_fault
1391 * [Don't try vm_map_copy!]
1392 *
1393 * To remove a submapping, one must first remove the
1394 * range from the superior map, and then destroy the
1395 * submap (if desired). [Better yet, don't try it.]
1396 */
1397 int
1398 vm_map_submap(
1399 vm_map_t map,
1400 vm_offset_t start,
1401 vm_offset_t end,
1402 vm_map_t submap)
1403 {
1404 vm_map_entry_t entry;
1405 int result = KERN_INVALID_ARGUMENT;
1406
1407 vm_map_lock(map);
1408
1409 VM_MAP_RANGE_CHECK(map, start, end);
1410
1411 if (vm_map_lookup_entry(map, start, &entry)) {
1412 vm_map_clip_start(map, entry, start);
1413 } else
1414 entry = entry->next;
1415
1416 vm_map_clip_end(map, entry, end);
1417
1418 if ((entry->start == start) && (entry->end == end) &&
1419 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1420 (entry->object.vm_object == NULL)) {
1421 entry->object.sub_map = submap;
1422 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1423 result = KERN_SUCCESS;
1424 }
1425 vm_map_unlock(map);
1426
1427 return (result);
1428 }
1429
1430 /*
1431 * The maximum number of pages to map
1432 */
1433 #define MAX_INIT_PT 96
1434
1435 /*
1436 * vm_map_pmap_enter:
1437 *
1438 * Preload read-only mappings for the given object into the specified
1439 * map. This eliminates the soft faults on process startup and
1440 * immediately after an mmap(2).
1441 */
1442 void
1443 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1444 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1445 {
1446 vm_offset_t tmpidx;
1447 int psize;
1448 vm_page_t p, mpte;
1449 boolean_t are_queues_locked;
1450
1451 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1452 return;
1453 VM_OBJECT_LOCK(object);
1454 if (object->type == OBJT_DEVICE) {
1455 pmap_object_init_pt(map->pmap, addr, object, pindex, size);
1456 goto unlock_return;
1457 }
1458
1459 psize = atop(size);
1460
1461 if (object->type != OBJT_VNODE ||
1462 ((flags & MAP_PREFAULT_PARTIAL) && (psize > MAX_INIT_PT) &&
1463 (object->resident_page_count > MAX_INIT_PT))) {
1464 goto unlock_return;
1465 }
1466
1467 if (psize + pindex > object->size) {
1468 if (object->size < pindex)
1469 goto unlock_return;
1470 psize = object->size - pindex;
1471 }
1472
1473 are_queues_locked = FALSE;
1474 mpte = NULL;
1475
1476 if ((p = TAILQ_FIRST(&object->memq)) != NULL) {
1477 if (p->pindex < pindex) {
1478 p = vm_page_splay(pindex, object->root);
1479 if ((object->root = p)->pindex < pindex)
1480 p = TAILQ_NEXT(p, listq);
1481 }
1482 }
1483 /*
1484 * Assert: the variable p is either (1) the page with the
1485 * least pindex greater than or equal to the parameter pindex
1486 * or (2) NULL.
1487 */
1488 for (;
1489 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1490 p = TAILQ_NEXT(p, listq)) {
1491 /*
1492 * don't allow an madvise to blow away our really
1493 * free pages allocating pv entries.
1494 */
1495 if ((flags & MAP_PREFAULT_MADVISE) &&
1496 cnt.v_free_count < cnt.v_free_reserved) {
1497 break;
1498 }
1499 if ((p->valid & VM_PAGE_BITS_ALL) == VM_PAGE_BITS_ALL &&
1500 (p->busy == 0) &&
1501 (p->flags & (PG_BUSY | PG_FICTITIOUS)) == 0) {
1502 if (!are_queues_locked) {
1503 are_queues_locked = TRUE;
1504 vm_page_lock_queues();
1505 }
1506 if ((p->queue - p->pc) == PQ_CACHE)
1507 vm_page_deactivate(p);
1508 mpte = pmap_enter_quick(map->pmap,
1509 addr + ptoa(tmpidx), p, prot, mpte);
1510 }
1511 }
1512 if (are_queues_locked)
1513 vm_page_unlock_queues();
1514 unlock_return:
1515 VM_OBJECT_UNLOCK(object);
1516 }
1517
1518 /*
1519 * vm_map_protect:
1520 *
1521 * Sets the protection of the specified address
1522 * region in the target map. If "set_max" is
1523 * specified, the maximum protection is to be set;
1524 * otherwise, only the current protection is affected.
1525 */
1526 int
1527 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1528 vm_prot_t new_prot, boolean_t set_max)
1529 {
1530 vm_map_entry_t current;
1531 vm_map_entry_t entry;
1532
1533 vm_map_lock(map);
1534
1535 VM_MAP_RANGE_CHECK(map, start, end);
1536
1537 if (vm_map_lookup_entry(map, start, &entry)) {
1538 vm_map_clip_start(map, entry, start);
1539 } else {
1540 entry = entry->next;
1541 }
1542
1543 /*
1544 * Make a first pass to check for protection violations.
1545 */
1546 current = entry;
1547 while ((current != &map->header) && (current->start < end)) {
1548 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1549 vm_map_unlock(map);
1550 return (KERN_INVALID_ARGUMENT);
1551 }
1552 if ((new_prot & current->max_protection) != new_prot) {
1553 vm_map_unlock(map);
1554 return (KERN_PROTECTION_FAILURE);
1555 }
1556 current = current->next;
1557 }
1558
1559 /*
1560 * Go back and fix up protections. [Note that clipping is not
1561 * necessary the second time.]
1562 */
1563 current = entry;
1564 while ((current != &map->header) && (current->start < end)) {
1565 vm_prot_t old_prot;
1566
1567 vm_map_clip_end(map, current, end);
1568
1569 old_prot = current->protection;
1570 if (set_max)
1571 current->protection =
1572 (current->max_protection = new_prot) &
1573 old_prot;
1574 else
1575 current->protection = new_prot;
1576
1577 /*
1578 * Update physical map if necessary. Worry about copy-on-write
1579 * here -- CHECK THIS XXX
1580 */
1581 if (current->protection != old_prot) {
1582 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1583 VM_PROT_ALL)
1584 pmap_protect(map->pmap, current->start,
1585 current->end,
1586 current->protection & MASK(current));
1587 #undef MASK
1588 }
1589 vm_map_simplify_entry(map, current);
1590 current = current->next;
1591 }
1592 vm_map_unlock(map);
1593 return (KERN_SUCCESS);
1594 }
1595
1596 /*
1597 * vm_map_madvise:
1598 *
1599 * This routine traverses a processes map handling the madvise
1600 * system call. Advisories are classified as either those effecting
1601 * the vm_map_entry structure, or those effecting the underlying
1602 * objects.
1603 */
1604 int
1605 vm_map_madvise(
1606 vm_map_t map,
1607 vm_offset_t start,
1608 vm_offset_t end,
1609 int behav)
1610 {
1611 vm_map_entry_t current, entry;
1612 int modify_map = 0;
1613
1614 /*
1615 * Some madvise calls directly modify the vm_map_entry, in which case
1616 * we need to use an exclusive lock on the map and we need to perform
1617 * various clipping operations. Otherwise we only need a read-lock
1618 * on the map.
1619 */
1620 switch(behav) {
1621 case MADV_NORMAL:
1622 case MADV_SEQUENTIAL:
1623 case MADV_RANDOM:
1624 case MADV_NOSYNC:
1625 case MADV_AUTOSYNC:
1626 case MADV_NOCORE:
1627 case MADV_CORE:
1628 modify_map = 1;
1629 vm_map_lock(map);
1630 break;
1631 case MADV_WILLNEED:
1632 case MADV_DONTNEED:
1633 case MADV_FREE:
1634 vm_map_lock_read(map);
1635 break;
1636 default:
1637 return (KERN_INVALID_ARGUMENT);
1638 }
1639
1640 /*
1641 * Locate starting entry and clip if necessary.
1642 */
1643 VM_MAP_RANGE_CHECK(map, start, end);
1644
1645 if (vm_map_lookup_entry(map, start, &entry)) {
1646 if (modify_map)
1647 vm_map_clip_start(map, entry, start);
1648 } else {
1649 entry = entry->next;
1650 }
1651
1652 if (modify_map) {
1653 /*
1654 * madvise behaviors that are implemented in the vm_map_entry.
1655 *
1656 * We clip the vm_map_entry so that behavioral changes are
1657 * limited to the specified address range.
1658 */
1659 for (current = entry;
1660 (current != &map->header) && (current->start < end);
1661 current = current->next
1662 ) {
1663 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1664 continue;
1665
1666 vm_map_clip_end(map, current, end);
1667
1668 switch (behav) {
1669 case MADV_NORMAL:
1670 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1671 break;
1672 case MADV_SEQUENTIAL:
1673 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1674 break;
1675 case MADV_RANDOM:
1676 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1677 break;
1678 case MADV_NOSYNC:
1679 current->eflags |= MAP_ENTRY_NOSYNC;
1680 break;
1681 case MADV_AUTOSYNC:
1682 current->eflags &= ~MAP_ENTRY_NOSYNC;
1683 break;
1684 case MADV_NOCORE:
1685 current->eflags |= MAP_ENTRY_NOCOREDUMP;
1686 break;
1687 case MADV_CORE:
1688 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1689 break;
1690 default:
1691 break;
1692 }
1693 vm_map_simplify_entry(map, current);
1694 }
1695 vm_map_unlock(map);
1696 } else {
1697 vm_pindex_t pindex;
1698 int count;
1699
1700 /*
1701 * madvise behaviors that are implemented in the underlying
1702 * vm_object.
1703 *
1704 * Since we don't clip the vm_map_entry, we have to clip
1705 * the vm_object pindex and count.
1706 */
1707 for (current = entry;
1708 (current != &map->header) && (current->start < end);
1709 current = current->next
1710 ) {
1711 vm_offset_t useStart;
1712
1713 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1714 continue;
1715
1716 pindex = OFF_TO_IDX(current->offset);
1717 count = atop(current->end - current->start);
1718 useStart = current->start;
1719
1720 if (current->start < start) {
1721 pindex += atop(start - current->start);
1722 count -= atop(start - current->start);
1723 useStart = start;
1724 }
1725 if (current->end > end)
1726 count -= atop(current->end - end);
1727
1728 if (count <= 0)
1729 continue;
1730
1731 vm_object_madvise(current->object.vm_object,
1732 pindex, count, behav);
1733 if (behav == MADV_WILLNEED) {
1734 vm_map_pmap_enter(map,
1735 useStart,
1736 current->protection,
1737 current->object.vm_object,
1738 pindex,
1739 (count << PAGE_SHIFT),
1740 MAP_PREFAULT_MADVISE
1741 );
1742 }
1743 }
1744 vm_map_unlock_read(map);
1745 }
1746 return (0);
1747 }
1748
1749
1750 /*
1751 * vm_map_inherit:
1752 *
1753 * Sets the inheritance of the specified address
1754 * range in the target map. Inheritance
1755 * affects how the map will be shared with
1756 * child maps at the time of vm_map_fork.
1757 */
1758 int
1759 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1760 vm_inherit_t new_inheritance)
1761 {
1762 vm_map_entry_t entry;
1763 vm_map_entry_t temp_entry;
1764
1765 switch (new_inheritance) {
1766 case VM_INHERIT_NONE:
1767 case VM_INHERIT_COPY:
1768 case VM_INHERIT_SHARE:
1769 break;
1770 default:
1771 return (KERN_INVALID_ARGUMENT);
1772 }
1773 vm_map_lock(map);
1774 VM_MAP_RANGE_CHECK(map, start, end);
1775 if (vm_map_lookup_entry(map, start, &temp_entry)) {
1776 entry = temp_entry;
1777 vm_map_clip_start(map, entry, start);
1778 } else
1779 entry = temp_entry->next;
1780 while ((entry != &map->header) && (entry->start < end)) {
1781 vm_map_clip_end(map, entry, end);
1782 entry->inheritance = new_inheritance;
1783 vm_map_simplify_entry(map, entry);
1784 entry = entry->next;
1785 }
1786 vm_map_unlock(map);
1787 return (KERN_SUCCESS);
1788 }
1789
1790 /*
1791 * vm_map_unwire:
1792 *
1793 * Implements both kernel and user unwiring.
1794 */
1795 int
1796 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
1797 int flags)
1798 {
1799 vm_map_entry_t entry, first_entry, tmp_entry;
1800 vm_offset_t saved_start;
1801 unsigned int last_timestamp;
1802 int rv;
1803 boolean_t need_wakeup, result, user_unwire;
1804
1805 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
1806 vm_map_lock(map);
1807 VM_MAP_RANGE_CHECK(map, start, end);
1808 if (!vm_map_lookup_entry(map, start, &first_entry)) {
1809 if (flags & VM_MAP_WIRE_HOLESOK)
1810 first_entry = first_entry->next;
1811 else {
1812 vm_map_unlock(map);
1813 return (KERN_INVALID_ADDRESS);
1814 }
1815 }
1816 last_timestamp = map->timestamp;
1817 entry = first_entry;
1818 while (entry != &map->header && entry->start < end) {
1819 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1820 /*
1821 * We have not yet clipped the entry.
1822 */
1823 saved_start = (start >= entry->start) ? start :
1824 entry->start;
1825 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1826 if (vm_map_unlock_and_wait(map, user_unwire)) {
1827 /*
1828 * Allow interruption of user unwiring?
1829 */
1830 }
1831 vm_map_lock(map);
1832 if (last_timestamp+1 != map->timestamp) {
1833 /*
1834 * Look again for the entry because the map was
1835 * modified while it was unlocked.
1836 * Specifically, the entry may have been
1837 * clipped, merged, or deleted.
1838 */
1839 if (!vm_map_lookup_entry(map, saved_start,
1840 &tmp_entry)) {
1841 if (flags & VM_MAP_WIRE_HOLESOK)
1842 tmp_entry = tmp_entry->next;
1843 else {
1844 if (saved_start == start) {
1845 /*
1846 * First_entry has been deleted.
1847 */
1848 vm_map_unlock(map);
1849 return (KERN_INVALID_ADDRESS);
1850 }
1851 end = saved_start;
1852 rv = KERN_INVALID_ADDRESS;
1853 goto done;
1854 }
1855 }
1856 if (entry == first_entry)
1857 first_entry = tmp_entry;
1858 else
1859 first_entry = NULL;
1860 entry = tmp_entry;
1861 }
1862 last_timestamp = map->timestamp;
1863 continue;
1864 }
1865 vm_map_clip_start(map, entry, start);
1866 vm_map_clip_end(map, entry, end);
1867 /*
1868 * Mark the entry in case the map lock is released. (See
1869 * above.)
1870 */
1871 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1872 /*
1873 * Check the map for holes in the specified region.
1874 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
1875 */
1876 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
1877 (entry->end < end && (entry->next == &map->header ||
1878 entry->next->start > entry->end))) {
1879 end = entry->end;
1880 rv = KERN_INVALID_ADDRESS;
1881 goto done;
1882 }
1883 /*
1884 * If system unwiring, require that the entry is system wired.
1885 */
1886 if (!user_unwire &&
1887 vm_map_entry_system_wired_count(entry) == 0) {
1888 end = entry->end;
1889 rv = KERN_INVALID_ARGUMENT;
1890 goto done;
1891 }
1892 entry = entry->next;
1893 }
1894 rv = KERN_SUCCESS;
1895 done:
1896 need_wakeup = FALSE;
1897 if (first_entry == NULL) {
1898 result = vm_map_lookup_entry(map, start, &first_entry);
1899 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
1900 first_entry = first_entry->next;
1901 else
1902 KASSERT(result, ("vm_map_unwire: lookup failed"));
1903 }
1904 entry = first_entry;
1905 while (entry != &map->header && entry->start < end) {
1906 if (rv == KERN_SUCCESS && (!user_unwire ||
1907 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
1908 if (user_unwire)
1909 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1910 entry->wired_count--;
1911 if (entry->wired_count == 0) {
1912 /*
1913 * Retain the map lock.
1914 */
1915 vm_fault_unwire(map, entry->start, entry->end,
1916 entry->object.vm_object != NULL &&
1917 entry->object.vm_object->type == OBJT_DEVICE);
1918 }
1919 }
1920 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
1921 ("vm_map_unwire: in-transition flag missing"));
1922 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1923 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1924 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1925 need_wakeup = TRUE;
1926 }
1927 vm_map_simplify_entry(map, entry);
1928 entry = entry->next;
1929 }
1930 vm_map_unlock(map);
1931 if (need_wakeup)
1932 vm_map_wakeup(map);
1933 return (rv);
1934 }
1935
1936 /*
1937 * vm_map_wire:
1938 *
1939 * Implements both kernel and user wiring.
1940 */
1941 int
1942 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
1943 int flags)
1944 {
1945 vm_map_entry_t entry, first_entry, tmp_entry;
1946 vm_offset_t saved_end, saved_start;
1947 unsigned int last_timestamp;
1948 int rv;
1949 boolean_t fictitious, need_wakeup, result, user_wire;
1950
1951 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
1952 vm_map_lock(map);
1953 VM_MAP_RANGE_CHECK(map, start, end);
1954 if (!vm_map_lookup_entry(map, start, &first_entry)) {
1955 if (flags & VM_MAP_WIRE_HOLESOK)
1956 first_entry = first_entry->next;
1957 else {
1958 vm_map_unlock(map);
1959 return (KERN_INVALID_ADDRESS);
1960 }
1961 }
1962 last_timestamp = map->timestamp;
1963 entry = first_entry;
1964 while (entry != &map->header && entry->start < end) {
1965 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1966 /*
1967 * We have not yet clipped the entry.
1968 */
1969 saved_start = (start >= entry->start) ? start :
1970 entry->start;
1971 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1972 if (vm_map_unlock_and_wait(map, user_wire)) {
1973 /*
1974 * Allow interruption of user wiring?
1975 */
1976 }
1977 vm_map_lock(map);
1978 if (last_timestamp + 1 != map->timestamp) {
1979 /*
1980 * Look again for the entry because the map was
1981 * modified while it was unlocked.
1982 * Specifically, the entry may have been
1983 * clipped, merged, or deleted.
1984 */
1985 if (!vm_map_lookup_entry(map, saved_start,
1986 &tmp_entry)) {
1987 if (flags & VM_MAP_WIRE_HOLESOK)
1988 tmp_entry = tmp_entry->next;
1989 else {
1990 if (saved_start == start) {
1991 /*
1992 * first_entry has been deleted.
1993 */
1994 vm_map_unlock(map);
1995 return (KERN_INVALID_ADDRESS);
1996 }
1997 end = saved_start;
1998 rv = KERN_INVALID_ADDRESS;
1999 goto done;
2000 }
2001 }
2002 if (entry == first_entry)
2003 first_entry = tmp_entry;
2004 else
2005 first_entry = NULL;
2006 entry = tmp_entry;
2007 }
2008 last_timestamp = map->timestamp;
2009 continue;
2010 }
2011 vm_map_clip_start(map, entry, start);
2012 vm_map_clip_end(map, entry, end);
2013 /*
2014 * Mark the entry in case the map lock is released. (See
2015 * above.)
2016 */
2017 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2018 /*
2019 *
2020 */
2021 if (entry->wired_count == 0) {
2022 entry->wired_count++;
2023 saved_start = entry->start;
2024 saved_end = entry->end;
2025 fictitious = entry->object.vm_object != NULL &&
2026 entry->object.vm_object->type == OBJT_DEVICE;
2027 /*
2028 * Release the map lock, relying on the in-transition
2029 * mark.
2030 */
2031 vm_map_unlock(map);
2032 rv = vm_fault_wire(map, saved_start, saved_end,
2033 user_wire, fictitious);
2034 vm_map_lock(map);
2035 if (last_timestamp + 1 != map->timestamp) {
2036 /*
2037 * Look again for the entry because the map was
2038 * modified while it was unlocked. The entry
2039 * may have been clipped, but NOT merged or
2040 * deleted.
2041 */
2042 result = vm_map_lookup_entry(map, saved_start,
2043 &tmp_entry);
2044 KASSERT(result, ("vm_map_wire: lookup failed"));
2045 if (entry == first_entry)
2046 first_entry = tmp_entry;
2047 else
2048 first_entry = NULL;
2049 entry = tmp_entry;
2050 while (entry->end < saved_end) {
2051 if (rv != KERN_SUCCESS) {
2052 KASSERT(entry->wired_count == 1,
2053 ("vm_map_wire: bad count"));
2054 entry->wired_count = -1;
2055 }
2056 entry = entry->next;
2057 }
2058 }
2059 last_timestamp = map->timestamp;
2060 if (rv != KERN_SUCCESS) {
2061 KASSERT(entry->wired_count == 1,
2062 ("vm_map_wire: bad count"));
2063 /*
2064 * Assign an out-of-range value to represent
2065 * the failure to wire this entry.
2066 */
2067 entry->wired_count = -1;
2068 end = entry->end;
2069 goto done;
2070 }
2071 } else if (!user_wire ||
2072 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2073 entry->wired_count++;
2074 }
2075 /*
2076 * Check the map for holes in the specified region.
2077 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2078 */
2079 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2080 (entry->end < end && (entry->next == &map->header ||
2081 entry->next->start > entry->end))) {
2082 end = entry->end;
2083 rv = KERN_INVALID_ADDRESS;
2084 goto done;
2085 }
2086 entry = entry->next;
2087 }
2088 rv = KERN_SUCCESS;
2089 done:
2090 need_wakeup = FALSE;
2091 if (first_entry == NULL) {
2092 result = vm_map_lookup_entry(map, start, &first_entry);
2093 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2094 first_entry = first_entry->next;
2095 else
2096 KASSERT(result, ("vm_map_wire: lookup failed"));
2097 }
2098 entry = first_entry;
2099 while (entry != &map->header && entry->start < end) {
2100 if (rv == KERN_SUCCESS) {
2101 if (user_wire)
2102 entry->eflags |= MAP_ENTRY_USER_WIRED;
2103 } else if (entry->wired_count == -1) {
2104 /*
2105 * Wiring failed on this entry. Thus, unwiring is
2106 * unnecessary.
2107 */
2108 entry->wired_count = 0;
2109 } else {
2110 if (!user_wire ||
2111 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2112 entry->wired_count--;
2113 if (entry->wired_count == 0) {
2114 /*
2115 * Retain the map lock.
2116 */
2117 vm_fault_unwire(map, entry->start, entry->end,
2118 entry->object.vm_object != NULL &&
2119 entry->object.vm_object->type == OBJT_DEVICE);
2120 }
2121 }
2122 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2123 ("vm_map_wire: in-transition flag missing"));
2124 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2125 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2126 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2127 need_wakeup = TRUE;
2128 }
2129 vm_map_simplify_entry(map, entry);
2130 entry = entry->next;
2131 }
2132 vm_map_unlock(map);
2133 if (need_wakeup)
2134 vm_map_wakeup(map);
2135 return (rv);
2136 }
2137
2138 /*
2139 * vm_map_sync
2140 *
2141 * Push any dirty cached pages in the address range to their pager.
2142 * If syncio is TRUE, dirty pages are written synchronously.
2143 * If invalidate is TRUE, any cached pages are freed as well.
2144 *
2145 * If the size of the region from start to end is zero, we are
2146 * supposed to flush all modified pages within the region containing
2147 * start. Unfortunately, a region can be split or coalesced with
2148 * neighboring regions, making it difficult to determine what the
2149 * original region was. Therefore, we approximate this requirement by
2150 * flushing the current region containing start.
2151 *
2152 * Returns an error if any part of the specified range is not mapped.
2153 */
2154 int
2155 vm_map_sync(
2156 vm_map_t map,
2157 vm_offset_t start,
2158 vm_offset_t end,
2159 boolean_t syncio,
2160 boolean_t invalidate)
2161 {
2162 vm_map_entry_t current;
2163 vm_map_entry_t entry;
2164 vm_size_t size;
2165 vm_object_t object;
2166 vm_ooffset_t offset;
2167
2168 vm_map_lock_read(map);
2169 VM_MAP_RANGE_CHECK(map, start, end);
2170 if (!vm_map_lookup_entry(map, start, &entry)) {
2171 vm_map_unlock_read(map);
2172 return (KERN_INVALID_ADDRESS);
2173 } else if (start == end) {
2174 start = entry->start;
2175 end = entry->end;
2176 }
2177 /*
2178 * Make a first pass to check for user-wired memory and holes.
2179 */
2180 for (current = entry; current->start < end; current = current->next) {
2181 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2182 vm_map_unlock_read(map);
2183 return (KERN_INVALID_ARGUMENT);
2184 }
2185 if (end > current->end &&
2186 (current->next == &map->header ||
2187 current->end != current->next->start)) {
2188 vm_map_unlock_read(map);
2189 return (KERN_INVALID_ADDRESS);
2190 }
2191 }
2192
2193 if (invalidate) {
2194 VM_LOCK_GIANT();
2195 pmap_remove(map->pmap, start, end);
2196 VM_UNLOCK_GIANT();
2197 }
2198 /*
2199 * Make a second pass, cleaning/uncaching pages from the indicated
2200 * objects as we go.
2201 */
2202 for (current = entry; current->start < end; current = current->next) {
2203 offset = current->offset + (start - current->start);
2204 size = (end <= current->end ? end : current->end) - start;
2205 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2206 vm_map_t smap;
2207 vm_map_entry_t tentry;
2208 vm_size_t tsize;
2209
2210 smap = current->object.sub_map;
2211 vm_map_lock_read(smap);
2212 (void) vm_map_lookup_entry(smap, offset, &tentry);
2213 tsize = tentry->end - offset;
2214 if (tsize < size)
2215 size = tsize;
2216 object = tentry->object.vm_object;
2217 offset = tentry->offset + (offset - tentry->start);
2218 vm_map_unlock_read(smap);
2219 } else {
2220 object = current->object.vm_object;
2221 }
2222 vm_object_sync(object, offset, size, syncio, invalidate);
2223 start += size;
2224 }
2225
2226 vm_map_unlock_read(map);
2227 return (KERN_SUCCESS);
2228 }
2229
2230 /*
2231 * vm_map_entry_unwire: [ internal use only ]
2232 *
2233 * Make the region specified by this entry pageable.
2234 *
2235 * The map in question should be locked.
2236 * [This is the reason for this routine's existence.]
2237 */
2238 static void
2239 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2240 {
2241 vm_fault_unwire(map, entry->start, entry->end,
2242 entry->object.vm_object != NULL &&
2243 entry->object.vm_object->type == OBJT_DEVICE);
2244 entry->wired_count = 0;
2245 }
2246
2247 /*
2248 * vm_map_entry_delete: [ internal use only ]
2249 *
2250 * Deallocate the given entry from the target map.
2251 */
2252 static void
2253 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2254 {
2255 vm_object_t object;
2256 vm_pindex_t offidxstart, offidxend, count;
2257
2258 vm_map_entry_unlink(map, entry);
2259 map->size -= entry->end - entry->start;
2260
2261 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2262 (object = entry->object.vm_object) != NULL) {
2263 count = OFF_TO_IDX(entry->end - entry->start);
2264 offidxstart = OFF_TO_IDX(entry->offset);
2265 offidxend = offidxstart + count;
2266 VM_OBJECT_LOCK(object);
2267 if (object->ref_count != 1 &&
2268 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2269 object == kernel_object || object == kmem_object) &&
2270 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2271 vm_object_collapse(object);
2272 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2273 if (object->type == OBJT_SWAP)
2274 swap_pager_freespace(object, offidxstart, count);
2275 if (offidxend >= object->size &&
2276 offidxstart < object->size)
2277 object->size = offidxstart;
2278 }
2279 VM_OBJECT_UNLOCK(object);
2280 vm_object_deallocate(object);
2281 }
2282
2283 vm_map_entry_dispose(map, entry);
2284 }
2285
2286 /*
2287 * vm_map_delete: [ internal use only ]
2288 *
2289 * Deallocates the given address range from the target
2290 * map.
2291 */
2292 int
2293 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2294 {
2295 vm_map_entry_t entry;
2296 vm_map_entry_t first_entry;
2297
2298 /*
2299 * Find the start of the region, and clip it
2300 */
2301 if (!vm_map_lookup_entry(map, start, &first_entry))
2302 entry = first_entry->next;
2303 else {
2304 entry = first_entry;
2305 vm_map_clip_start(map, entry, start);
2306 }
2307
2308 /*
2309 * Step through all entries in this region
2310 */
2311 while ((entry != &map->header) && (entry->start < end)) {
2312 vm_map_entry_t next;
2313
2314 /*
2315 * Wait for wiring or unwiring of an entry to complete.
2316 * Also wait for any system wirings to disappear on
2317 * user maps.
2318 */
2319 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2320 (vm_map_pmap(map) != kernel_pmap &&
2321 vm_map_entry_system_wired_count(entry) != 0)) {
2322 unsigned int last_timestamp;
2323 vm_offset_t saved_start;
2324 vm_map_entry_t tmp_entry;
2325
2326 saved_start = entry->start;
2327 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2328 last_timestamp = map->timestamp;
2329 (void) vm_map_unlock_and_wait(map, FALSE);
2330 vm_map_lock(map);
2331 if (last_timestamp + 1 != map->timestamp) {
2332 /*
2333 * Look again for the entry because the map was
2334 * modified while it was unlocked.
2335 * Specifically, the entry may have been
2336 * clipped, merged, or deleted.
2337 */
2338 if (!vm_map_lookup_entry(map, saved_start,
2339 &tmp_entry))
2340 entry = tmp_entry->next;
2341 else {
2342 entry = tmp_entry;
2343 vm_map_clip_start(map, entry,
2344 saved_start);
2345 }
2346 }
2347 continue;
2348 }
2349 vm_map_clip_end(map, entry, end);
2350
2351 next = entry->next;
2352
2353 /*
2354 * Unwire before removing addresses from the pmap; otherwise,
2355 * unwiring will put the entries back in the pmap.
2356 */
2357 if (entry->wired_count != 0) {
2358 vm_map_entry_unwire(map, entry);
2359 }
2360
2361 if (!map->system_map)
2362 VM_LOCK_GIANT();
2363 pmap_remove(map->pmap, entry->start, entry->end);
2364 if (!map->system_map)
2365 VM_UNLOCK_GIANT();
2366
2367 /*
2368 * Delete the entry (which may delete the object) only after
2369 * removing all pmap entries pointing to its pages.
2370 * (Otherwise, its page frames may be reallocated, and any
2371 * modify bits will be set in the wrong object!)
2372 */
2373 vm_map_entry_delete(map, entry);
2374 entry = next;
2375 }
2376 return (KERN_SUCCESS);
2377 }
2378
2379 /*
2380 * vm_map_remove:
2381 *
2382 * Remove the given address range from the target map.
2383 * This is the exported form of vm_map_delete.
2384 */
2385 int
2386 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2387 {
2388 int result;
2389
2390 vm_map_lock(map);
2391 VM_MAP_RANGE_CHECK(map, start, end);
2392 result = vm_map_delete(map, start, end);
2393 vm_map_unlock(map);
2394 return (result);
2395 }
2396
2397 /*
2398 * vm_map_check_protection:
2399 *
2400 * Assert that the target map allows the specified privilege on the
2401 * entire address region given. The entire region must be allocated.
2402 *
2403 * WARNING! This code does not and should not check whether the
2404 * contents of the region is accessible. For example a smaller file
2405 * might be mapped into a larger address space.
2406 *
2407 * NOTE! This code is also called by munmap().
2408 *
2409 * The map must be locked. A read lock is sufficient.
2410 */
2411 boolean_t
2412 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2413 vm_prot_t protection)
2414 {
2415 vm_map_entry_t entry;
2416 vm_map_entry_t tmp_entry;
2417
2418 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2419 return (FALSE);
2420 entry = tmp_entry;
2421
2422 while (start < end) {
2423 if (entry == &map->header)
2424 return (FALSE);
2425 /*
2426 * No holes allowed!
2427 */
2428 if (start < entry->start)
2429 return (FALSE);
2430 /*
2431 * Check protection associated with entry.
2432 */
2433 if ((entry->protection & protection) != protection)
2434 return (FALSE);
2435 /* go to next entry */
2436 start = entry->end;
2437 entry = entry->next;
2438 }
2439 return (TRUE);
2440 }
2441
2442 /*
2443 * vm_map_copy_entry:
2444 *
2445 * Copies the contents of the source entry to the destination
2446 * entry. The entries *must* be aligned properly.
2447 */
2448 static void
2449 vm_map_copy_entry(
2450 vm_map_t src_map,
2451 vm_map_t dst_map,
2452 vm_map_entry_t src_entry,
2453 vm_map_entry_t dst_entry)
2454 {
2455 vm_object_t src_object;
2456
2457 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2458 return;
2459
2460 if (src_entry->wired_count == 0) {
2461
2462 /*
2463 * If the source entry is marked needs_copy, it is already
2464 * write-protected.
2465 */
2466 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2467 pmap_protect(src_map->pmap,
2468 src_entry->start,
2469 src_entry->end,
2470 src_entry->protection & ~VM_PROT_WRITE);
2471 }
2472
2473 /*
2474 * Make a copy of the object.
2475 */
2476 if ((src_object = src_entry->object.vm_object) != NULL) {
2477 VM_OBJECT_LOCK(src_object);
2478 if ((src_object->handle == NULL) &&
2479 (src_object->type == OBJT_DEFAULT ||
2480 src_object->type == OBJT_SWAP)) {
2481 vm_object_collapse(src_object);
2482 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2483 vm_object_split(src_entry);
2484 src_object = src_entry->object.vm_object;
2485 }
2486 }
2487 vm_object_reference_locked(src_object);
2488 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2489 VM_OBJECT_UNLOCK(src_object);
2490 dst_entry->object.vm_object = src_object;
2491 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2492 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2493 dst_entry->offset = src_entry->offset;
2494 } else {
2495 dst_entry->object.vm_object = NULL;
2496 dst_entry->offset = 0;
2497 }
2498
2499 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2500 dst_entry->end - dst_entry->start, src_entry->start);
2501 } else {
2502 /*
2503 * Of course, wired down pages can't be set copy-on-write.
2504 * Cause wired pages to be copied into the new map by
2505 * simulating faults (the new pages are pageable)
2506 */
2507 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2508 }
2509 }
2510
2511 /*
2512 * vmspace_map_entry_forked:
2513 * Update the newly-forked vmspace each time a map entry is inherited
2514 * or copied. The values for vm_dsize and vm_tsize are approximate
2515 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
2516 */
2517 static void
2518 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
2519 vm_map_entry_t entry)
2520 {
2521 vm_size_t entrysize;
2522 vm_offset_t newend;
2523
2524 entrysize = entry->end - entry->start;
2525 vm2->vm_map.size += entrysize;
2526 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
2527 vm2->vm_ssize += btoc(entrysize);
2528 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
2529 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
2530 newend = MIN(entry->end,
2531 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
2532 vm2->vm_dsize += btoc(newend - entry->start);
2533 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
2534 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
2535 newend = MIN(entry->end,
2536 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
2537 vm2->vm_tsize += btoc(newend - entry->start);
2538 }
2539 }
2540
2541 /*
2542 * vmspace_fork:
2543 * Create a new process vmspace structure and vm_map
2544 * based on those of an existing process. The new map
2545 * is based on the old map, according to the inheritance
2546 * values on the regions in that map.
2547 *
2548 * XXX It might be worth coalescing the entries added to the new vmspace.
2549 *
2550 * The source map must not be locked.
2551 */
2552 struct vmspace *
2553 vmspace_fork(struct vmspace *vm1)
2554 {
2555 struct vmspace *vm2;
2556 vm_map_t old_map = &vm1->vm_map;
2557 vm_map_t new_map;
2558 vm_map_entry_t old_entry;
2559 vm_map_entry_t new_entry;
2560 vm_object_t object;
2561
2562 vm_map_lock(old_map);
2563
2564 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
2565 vm2->vm_taddr = vm1->vm_taddr;
2566 vm2->vm_daddr = vm1->vm_daddr;
2567 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
2568 new_map = &vm2->vm_map; /* XXX */
2569 new_map->timestamp = 1;
2570
2571 /* Do not inherit the MAP_WIREFUTURE property. */
2572 if ((new_map->flags & MAP_WIREFUTURE) == MAP_WIREFUTURE)
2573 new_map->flags &= ~MAP_WIREFUTURE;
2574
2575 old_entry = old_map->header.next;
2576
2577 while (old_entry != &old_map->header) {
2578 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
2579 panic("vm_map_fork: encountered a submap");
2580
2581 switch (old_entry->inheritance) {
2582 case VM_INHERIT_NONE:
2583 break;
2584
2585 case VM_INHERIT_SHARE:
2586 /*
2587 * Clone the entry, creating the shared object if necessary.
2588 */
2589 object = old_entry->object.vm_object;
2590 if (object == NULL) {
2591 object = vm_object_allocate(OBJT_DEFAULT,
2592 atop(old_entry->end - old_entry->start));
2593 old_entry->object.vm_object = object;
2594 old_entry->offset = 0;
2595 }
2596
2597 /*
2598 * Add the reference before calling vm_object_shadow
2599 * to insure that a shadow object is created.
2600 */
2601 vm_object_reference(object);
2602 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2603 vm_object_shadow(&old_entry->object.vm_object,
2604 &old_entry->offset,
2605 atop(old_entry->end - old_entry->start));
2606 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2607 /* Transfer the second reference too. */
2608 vm_object_reference(
2609 old_entry->object.vm_object);
2610 vm_object_deallocate(object);
2611 object = old_entry->object.vm_object;
2612 }
2613 VM_OBJECT_LOCK(object);
2614 vm_object_clear_flag(object, OBJ_ONEMAPPING);
2615 VM_OBJECT_UNLOCK(object);
2616
2617 /*
2618 * Clone the entry, referencing the shared object.
2619 */
2620 new_entry = vm_map_entry_create(new_map);
2621 *new_entry = *old_entry;
2622 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2623 new_entry->wired_count = 0;
2624
2625 /*
2626 * Insert the entry into the new map -- we know we're
2627 * inserting at the end of the new map.
2628 */
2629 vm_map_entry_link(new_map, new_map->header.prev,
2630 new_entry);
2631 vmspace_map_entry_forked(vm1, vm2, new_entry);
2632
2633 /*
2634 * Update the physical map
2635 */
2636 pmap_copy(new_map->pmap, old_map->pmap,
2637 new_entry->start,
2638 (old_entry->end - old_entry->start),
2639 old_entry->start);
2640 break;
2641
2642 case VM_INHERIT_COPY:
2643 /*
2644 * Clone the entry and link into the map.
2645 */
2646 new_entry = vm_map_entry_create(new_map);
2647 *new_entry = *old_entry;
2648 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2649 new_entry->wired_count = 0;
2650 new_entry->object.vm_object = NULL;
2651 vm_map_entry_link(new_map, new_map->header.prev,
2652 new_entry);
2653 vmspace_map_entry_forked(vm1, vm2, new_entry);
2654 vm_map_copy_entry(old_map, new_map, old_entry,
2655 new_entry);
2656 break;
2657 }
2658 old_entry = old_entry->next;
2659 }
2660
2661 vm_map_unlock(old_map);
2662
2663 return (vm2);
2664 }
2665
2666 int
2667 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
2668 vm_prot_t prot, vm_prot_t max, int cow)
2669 {
2670 vm_map_entry_t new_entry, prev_entry;
2671 vm_offset_t bot, top;
2672 vm_size_t init_ssize;
2673 int orient, rv;
2674 rlim_t vmemlim;
2675
2676 /*
2677 * The stack orientation is piggybacked with the cow argument.
2678 * Extract it into orient and mask the cow argument so that we
2679 * don't pass it around further.
2680 * NOTE: We explicitly allow bi-directional stacks.
2681 */
2682 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
2683 cow &= ~orient;
2684 KASSERT(orient != 0, ("No stack grow direction"));
2685
2686 if (addrbos < vm_map_min(map) || addrbos > map->max_offset)
2687 return (KERN_NO_SPACE);
2688
2689 init_ssize = (max_ssize < sgrowsiz) ? max_ssize : sgrowsiz;
2690
2691 PROC_LOCK(curthread->td_proc);
2692 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM);
2693 PROC_UNLOCK(curthread->td_proc);
2694
2695 vm_map_lock(map);
2696
2697 /* If addr is already mapped, no go */
2698 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
2699 vm_map_unlock(map);
2700 return (KERN_NO_SPACE);
2701 }
2702
2703 /* If we would blow our VMEM resource limit, no go */
2704 if (map->size + init_ssize > vmemlim) {
2705 vm_map_unlock(map);
2706 return (KERN_NO_SPACE);
2707 }
2708
2709 /*
2710 * If we can't accomodate max_ssize in the current mapping, no go.
2711 * However, we need to be aware that subsequent user mappings might
2712 * map into the space we have reserved for stack, and currently this
2713 * space is not protected.
2714 *
2715 * Hopefully we will at least detect this condition when we try to
2716 * grow the stack.
2717 */
2718 if ((prev_entry->next != &map->header) &&
2719 (prev_entry->next->start < addrbos + max_ssize)) {
2720 vm_map_unlock(map);
2721 return (KERN_NO_SPACE);
2722 }
2723
2724 /*
2725 * We initially map a stack of only init_ssize. We will grow as
2726 * needed later. Depending on the orientation of the stack (i.e.
2727 * the grow direction) we either map at the top of the range, the
2728 * bottom of the range or in the middle.
2729 *
2730 * Note: we would normally expect prot and max to be VM_PROT_ALL,
2731 * and cow to be 0. Possibly we should eliminate these as input
2732 * parameters, and just pass these values here in the insert call.
2733 */
2734 if (orient == MAP_STACK_GROWS_DOWN)
2735 bot = addrbos + max_ssize - init_ssize;
2736 else if (orient == MAP_STACK_GROWS_UP)
2737 bot = addrbos;
2738 else
2739 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
2740 top = bot + init_ssize;
2741 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
2742
2743 /* Now set the avail_ssize amount. */
2744 if (rv == KERN_SUCCESS) {
2745 if (prev_entry != &map->header)
2746 vm_map_clip_end(map, prev_entry, bot);
2747 new_entry = prev_entry->next;
2748 if (new_entry->end != top || new_entry->start != bot)
2749 panic("Bad entry start/end for new stack entry");
2750
2751 new_entry->avail_ssize = max_ssize - init_ssize;
2752 if (orient & MAP_STACK_GROWS_DOWN)
2753 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
2754 if (orient & MAP_STACK_GROWS_UP)
2755 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
2756 }
2757
2758 vm_map_unlock(map);
2759 return (rv);
2760 }
2761
2762 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
2763 * desired address is already mapped, or if we successfully grow
2764 * the stack. Also returns KERN_SUCCESS if addr is outside the
2765 * stack range (this is strange, but preserves compatibility with
2766 * the grow function in vm_machdep.c).
2767 */
2768 int
2769 vm_map_growstack(struct proc *p, vm_offset_t addr)
2770 {
2771 vm_map_entry_t next_entry, prev_entry;
2772 vm_map_entry_t new_entry, stack_entry;
2773 struct vmspace *vm = p->p_vmspace;
2774 vm_map_t map = &vm->vm_map;
2775 vm_offset_t end;
2776 size_t grow_amount, max_grow;
2777 rlim_t stacklim, vmemlim;
2778 int is_procstack, rv;
2779
2780 Retry:
2781 PROC_LOCK(p);
2782 stacklim = lim_cur(p, RLIMIT_STACK);
2783 vmemlim = lim_cur(p, RLIMIT_VMEM);
2784 PROC_UNLOCK(p);
2785
2786 vm_map_lock_read(map);
2787
2788 /* If addr is already in the entry range, no need to grow.*/
2789 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
2790 vm_map_unlock_read(map);
2791 return (KERN_SUCCESS);
2792 }
2793
2794 next_entry = prev_entry->next;
2795 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
2796 /*
2797 * This entry does not grow upwards. Since the address lies
2798 * beyond this entry, the next entry (if one exists) has to
2799 * be a downward growable entry. The entry list header is
2800 * never a growable entry, so it suffices to check the flags.
2801 */
2802 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
2803 vm_map_unlock_read(map);
2804 return (KERN_SUCCESS);
2805 }
2806 stack_entry = next_entry;
2807 } else {
2808 /*
2809 * This entry grows upward. If the next entry does not at
2810 * least grow downwards, this is the entry we need to grow.
2811 * otherwise we have two possible choices and we have to
2812 * select one.
2813 */
2814 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
2815 /*
2816 * We have two choices; grow the entry closest to
2817 * the address to minimize the amount of growth.
2818 */
2819 if (addr - prev_entry->end <= next_entry->start - addr)
2820 stack_entry = prev_entry;
2821 else
2822 stack_entry = next_entry;
2823 } else
2824 stack_entry = prev_entry;
2825 }
2826
2827 if (stack_entry == next_entry) {
2828 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
2829 KASSERT(addr < stack_entry->start, ("foo"));
2830 end = (prev_entry != &map->header) ? prev_entry->end :
2831 stack_entry->start - stack_entry->avail_ssize;
2832 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
2833 max_grow = stack_entry->start - end;
2834 } else {
2835 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
2836 KASSERT(addr >= stack_entry->end, ("foo"));
2837 end = (next_entry != &map->header) ? next_entry->start :
2838 stack_entry->end + stack_entry->avail_ssize;
2839 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
2840 max_grow = end - stack_entry->end;
2841 }
2842
2843 if (grow_amount > stack_entry->avail_ssize) {
2844 vm_map_unlock_read(map);
2845 return (KERN_NO_SPACE);
2846 }
2847
2848 /*
2849 * If there is no longer enough space between the entries nogo, and
2850 * adjust the available space. Note: this should only happen if the
2851 * user has mapped into the stack area after the stack was created,
2852 * and is probably an error.
2853 *
2854 * This also effectively destroys any guard page the user might have
2855 * intended by limiting the stack size.
2856 */
2857 if (grow_amount > max_grow) {
2858 if (vm_map_lock_upgrade(map))
2859 goto Retry;
2860
2861 stack_entry->avail_ssize = max_grow;
2862
2863 vm_map_unlock(map);
2864 return (KERN_NO_SPACE);
2865 }
2866
2867 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
2868
2869 /*
2870 * If this is the main process stack, see if we're over the stack
2871 * limit.
2872 */
2873 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
2874 vm_map_unlock_read(map);
2875 return (KERN_NO_SPACE);
2876 }
2877
2878 /* Round up the grow amount modulo SGROWSIZ */
2879 grow_amount = roundup (grow_amount, sgrowsiz);
2880 if (grow_amount > stack_entry->avail_ssize)
2881 grow_amount = stack_entry->avail_ssize;
2882 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
2883 grow_amount = stacklim - ctob(vm->vm_ssize);
2884 }
2885
2886 /* If we would blow our VMEM resource limit, no go */
2887 if (map->size + grow_amount > vmemlim) {
2888 vm_map_unlock_read(map);
2889 return (KERN_NO_SPACE);
2890 }
2891
2892 if (vm_map_lock_upgrade(map))
2893 goto Retry;
2894
2895 if (stack_entry == next_entry) {
2896 /*
2897 * Growing downward.
2898 */
2899 /* Get the preliminary new entry start value */
2900 addr = stack_entry->start - grow_amount;
2901
2902 /*
2903 * If this puts us into the previous entry, cut back our
2904 * growth to the available space. Also, see the note above.
2905 */
2906 if (addr < end) {
2907 stack_entry->avail_ssize = max_grow;
2908 addr = end;
2909 }
2910
2911 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
2912 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0);
2913
2914 /* Adjust the available stack space by the amount we grew. */
2915 if (rv == KERN_SUCCESS) {
2916 if (prev_entry != &map->header)
2917 vm_map_clip_end(map, prev_entry, addr);
2918 new_entry = prev_entry->next;
2919 KASSERT(new_entry == stack_entry->prev, ("foo"));
2920 KASSERT(new_entry->end == stack_entry->start, ("foo"));
2921 KASSERT(new_entry->start == addr, ("foo"));
2922 grow_amount = new_entry->end - new_entry->start;
2923 new_entry->avail_ssize = stack_entry->avail_ssize -
2924 grow_amount;
2925 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
2926 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
2927 }
2928 } else {
2929 /*
2930 * Growing upward.
2931 */
2932 addr = stack_entry->end + grow_amount;
2933
2934 /*
2935 * If this puts us into the next entry, cut back our growth
2936 * to the available space. Also, see the note above.
2937 */
2938 if (addr > end) {
2939 stack_entry->avail_ssize = end - stack_entry->end;
2940 addr = end;
2941 }
2942
2943 grow_amount = addr - stack_entry->end;
2944
2945 /* Grow the underlying object if applicable. */
2946 if (stack_entry->object.vm_object == NULL ||
2947 vm_object_coalesce(stack_entry->object.vm_object,
2948 stack_entry->offset,
2949 (vm_size_t)(stack_entry->end - stack_entry->start),
2950 (vm_size_t)grow_amount)) {
2951 map->size += (addr - stack_entry->end);
2952 /* Update the current entry. */
2953 stack_entry->end = addr;
2954 stack_entry->avail_ssize -= grow_amount;
2955 vm_map_entry_resize_free(map, stack_entry);
2956 rv = KERN_SUCCESS;
2957
2958 if (next_entry != &map->header)
2959 vm_map_clip_start(map, next_entry, addr);
2960 } else
2961 rv = KERN_FAILURE;
2962 }
2963
2964 if (rv == KERN_SUCCESS && is_procstack)
2965 vm->vm_ssize += btoc(grow_amount);
2966
2967 vm_map_unlock(map);
2968
2969 /*
2970 * Heed the MAP_WIREFUTURE flag if it was set for this process.
2971 */
2972 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
2973 vm_map_wire(map,
2974 (stack_entry == next_entry) ? addr : addr - grow_amount,
2975 (stack_entry == next_entry) ? stack_entry->start : addr,
2976 (p->p_flag & P_SYSTEM)
2977 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
2978 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
2979 }
2980
2981 return (rv);
2982 }
2983
2984 /*
2985 * Unshare the specified VM space for exec. If other processes are
2986 * mapped to it, then create a new one. The new vmspace is null.
2987 */
2988 void
2989 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
2990 {
2991 struct vmspace *oldvmspace = p->p_vmspace;
2992 struct vmspace *newvmspace;
2993
2994 newvmspace = vmspace_alloc(minuser, maxuser);
2995 newvmspace->vm_swrss = oldvmspace->vm_swrss;
2996 /*
2997 * This code is written like this for prototype purposes. The
2998 * goal is to avoid running down the vmspace here, but let the
2999 * other process's that are still using the vmspace to finally
3000 * run it down. Even though there is little or no chance of blocking
3001 * here, it is a good idea to keep this form for future mods.
3002 */
3003 PROC_VMSPACE_LOCK(p);
3004 p->p_vmspace = newvmspace;
3005 PROC_VMSPACE_UNLOCK(p);
3006 if (p == curthread->td_proc) /* XXXKSE ? */
3007 pmap_activate(curthread);
3008 vmspace_free(oldvmspace);
3009 }
3010
3011 /*
3012 * Unshare the specified VM space for forcing COW. This
3013 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3014 */
3015 void
3016 vmspace_unshare(struct proc *p)
3017 {
3018 struct vmspace *oldvmspace = p->p_vmspace;
3019 struct vmspace *newvmspace;
3020
3021 if (oldvmspace->vm_refcnt == 1)
3022 return;
3023 newvmspace = vmspace_fork(oldvmspace);
3024 PROC_VMSPACE_LOCK(p);
3025 p->p_vmspace = newvmspace;
3026 PROC_VMSPACE_UNLOCK(p);
3027 if (p == curthread->td_proc) /* XXXKSE ? */
3028 pmap_activate(curthread);
3029 vmspace_free(oldvmspace);
3030 }
3031
3032 /*
3033 * vm_map_lookup:
3034 *
3035 * Finds the VM object, offset, and
3036 * protection for a given virtual address in the
3037 * specified map, assuming a page fault of the
3038 * type specified.
3039 *
3040 * Leaves the map in question locked for read; return
3041 * values are guaranteed until a vm_map_lookup_done
3042 * call is performed. Note that the map argument
3043 * is in/out; the returned map must be used in
3044 * the call to vm_map_lookup_done.
3045 *
3046 * A handle (out_entry) is returned for use in
3047 * vm_map_lookup_done, to make that fast.
3048 *
3049 * If a lookup is requested with "write protection"
3050 * specified, the map may be changed to perform virtual
3051 * copying operations, although the data referenced will
3052 * remain the same.
3053 */
3054 int
3055 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3056 vm_offset_t vaddr,
3057 vm_prot_t fault_typea,
3058 vm_map_entry_t *out_entry, /* OUT */
3059 vm_object_t *object, /* OUT */
3060 vm_pindex_t *pindex, /* OUT */
3061 vm_prot_t *out_prot, /* OUT */
3062 boolean_t *wired) /* OUT */
3063 {
3064 vm_map_entry_t entry;
3065 vm_map_t map = *var_map;
3066 vm_prot_t prot;
3067 vm_prot_t fault_type = fault_typea;
3068
3069 RetryLookup:;
3070 /*
3071 * Lookup the faulting address.
3072 */
3073
3074 vm_map_lock_read(map);
3075 #define RETURN(why) \
3076 { \
3077 vm_map_unlock_read(map); \
3078 return (why); \
3079 }
3080
3081 /*
3082 * If the map has an interesting hint, try it before calling full
3083 * blown lookup routine.
3084 */
3085 entry = map->root;
3086 *out_entry = entry;
3087 if (entry == NULL ||
3088 (vaddr < entry->start) || (vaddr >= entry->end)) {
3089 /*
3090 * Entry was either not a valid hint, or the vaddr was not
3091 * contained in the entry, so do a full lookup.
3092 */
3093 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3094 RETURN(KERN_INVALID_ADDRESS);
3095
3096 entry = *out_entry;
3097 }
3098
3099 /*
3100 * Handle submaps.
3101 */
3102 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3103 vm_map_t old_map = map;
3104
3105 *var_map = map = entry->object.sub_map;
3106 vm_map_unlock_read(old_map);
3107 goto RetryLookup;
3108 }
3109
3110 /*
3111 * Check whether this task is allowed to have this page.
3112 * Note the special case for MAP_ENTRY_COW
3113 * pages with an override. This is to implement a forced
3114 * COW for debuggers.
3115 */
3116 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3117 prot = entry->max_protection;
3118 else
3119 prot = entry->protection;
3120 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3121 if ((fault_type & prot) != fault_type) {
3122 RETURN(KERN_PROTECTION_FAILURE);
3123 }
3124 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3125 (entry->eflags & MAP_ENTRY_COW) &&
3126 (fault_type & VM_PROT_WRITE) &&
3127 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3128 RETURN(KERN_PROTECTION_FAILURE);
3129 }
3130
3131 /*
3132 * If this page is not pageable, we have to get it for all possible
3133 * accesses.
3134 */
3135 *wired = (entry->wired_count != 0);
3136 if (*wired)
3137 prot = fault_type = entry->protection;
3138
3139 /*
3140 * If the entry was copy-on-write, we either ...
3141 */
3142 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3143 /*
3144 * If we want to write the page, we may as well handle that
3145 * now since we've got the map locked.
3146 *
3147 * If we don't need to write the page, we just demote the
3148 * permissions allowed.
3149 */
3150 if (fault_type & VM_PROT_WRITE) {
3151 /*
3152 * Make a new object, and place it in the object
3153 * chain. Note that no new references have appeared
3154 * -- one just moved from the map to the new
3155 * object.
3156 */
3157 if (vm_map_lock_upgrade(map))
3158 goto RetryLookup;
3159
3160 vm_object_shadow(
3161 &entry->object.vm_object,
3162 &entry->offset,
3163 atop(entry->end - entry->start));
3164 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3165
3166 vm_map_lock_downgrade(map);
3167 } else {
3168 /*
3169 * We're attempting to read a copy-on-write page --
3170 * don't allow writes.
3171 */
3172 prot &= ~VM_PROT_WRITE;
3173 }
3174 }
3175
3176 /*
3177 * Create an object if necessary.
3178 */
3179 if (entry->object.vm_object == NULL &&
3180 !map->system_map) {
3181 if (vm_map_lock_upgrade(map))
3182 goto RetryLookup;
3183 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3184 atop(entry->end - entry->start));
3185 entry->offset = 0;
3186 vm_map_lock_downgrade(map);
3187 }
3188
3189 /*
3190 * Return the object/offset from this entry. If the entry was
3191 * copy-on-write or empty, it has been fixed up.
3192 */
3193 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3194 *object = entry->object.vm_object;
3195
3196 *out_prot = prot;
3197 return (KERN_SUCCESS);
3198
3199 #undef RETURN
3200 }
3201
3202 /*
3203 * vm_map_lookup_locked:
3204 *
3205 * Lookup the faulting address. A version of vm_map_lookup that returns
3206 * KERN_FAILURE instead of blocking on map lock or memory allocation.
3207 */
3208 int
3209 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
3210 vm_offset_t vaddr,
3211 vm_prot_t fault_typea,
3212 vm_map_entry_t *out_entry, /* OUT */
3213 vm_object_t *object, /* OUT */
3214 vm_pindex_t *pindex, /* OUT */
3215 vm_prot_t *out_prot, /* OUT */
3216 boolean_t *wired) /* OUT */
3217 {
3218 vm_map_entry_t entry;
3219 vm_map_t map = *var_map;
3220 vm_prot_t prot;
3221 vm_prot_t fault_type = fault_typea;
3222
3223 /*
3224 * If the map has an interesting hint, try it before calling full
3225 * blown lookup routine.
3226 */
3227 entry = map->root;
3228 *out_entry = entry;
3229 if (entry == NULL ||
3230 (vaddr < entry->start) || (vaddr >= entry->end)) {
3231 /*
3232 * Entry was either not a valid hint, or the vaddr was not
3233 * contained in the entry, so do a full lookup.
3234 */
3235 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3236 return (KERN_INVALID_ADDRESS);
3237
3238 entry = *out_entry;
3239 }
3240
3241 /*
3242 * Fail if the entry refers to a submap.
3243 */
3244 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3245 return (KERN_FAILURE);
3246
3247 /*
3248 * Check whether this task is allowed to have this page.
3249 * Note the special case for MAP_ENTRY_COW
3250 * pages with an override. This is to implement a forced
3251 * COW for debuggers.
3252 */
3253 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3254 prot = entry->max_protection;
3255 else
3256 prot = entry->protection;
3257 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
3258 if ((fault_type & prot) != fault_type)
3259 return (KERN_PROTECTION_FAILURE);
3260 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3261 (entry->eflags & MAP_ENTRY_COW) &&
3262 (fault_type & VM_PROT_WRITE) &&
3263 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0)
3264 return (KERN_PROTECTION_FAILURE);
3265
3266 /*
3267 * If this page is not pageable, we have to get it for all possible
3268 * accesses.
3269 */
3270 *wired = (entry->wired_count != 0);
3271 if (*wired)
3272 prot = fault_type = entry->protection;
3273
3274 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3275 /*
3276 * Fail if the entry was copy-on-write for a write fault.
3277 */
3278 if (fault_type & VM_PROT_WRITE)
3279 return (KERN_FAILURE);
3280 /*
3281 * We're attempting to read a copy-on-write page --
3282 * don't allow writes.
3283 */
3284 prot &= ~VM_PROT_WRITE;
3285 }
3286
3287 /*
3288 * Fail if an object should be created.
3289 */
3290 if (entry->object.vm_object == NULL && !map->system_map)
3291 return (KERN_FAILURE);
3292
3293 /*
3294 * Return the object/offset from this entry. If the entry was
3295 * copy-on-write or empty, it has been fixed up.
3296 */
3297 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3298 *object = entry->object.vm_object;
3299
3300 *out_prot = prot;
3301 return (KERN_SUCCESS);
3302 }
3303
3304 /*
3305 * vm_map_lookup_done:
3306 *
3307 * Releases locks acquired by a vm_map_lookup
3308 * (according to the handle returned by that lookup).
3309 */
3310 void
3311 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
3312 {
3313 /*
3314 * Unlock the main-level map
3315 */
3316 vm_map_unlock_read(map);
3317 }
3318
3319 #include "opt_ddb.h"
3320 #ifdef DDB
3321 #include <sys/kernel.h>
3322
3323 #include <ddb/ddb.h>
3324
3325 /*
3326 * vm_map_print: [ debug ]
3327 */
3328 DB_SHOW_COMMAND(map, vm_map_print)
3329 {
3330 static int nlines;
3331 /* XXX convert args. */
3332 vm_map_t map = (vm_map_t)addr;
3333 boolean_t full = have_addr;
3334
3335 vm_map_entry_t entry;
3336
3337 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3338 (void *)map,
3339 (void *)map->pmap, map->nentries, map->timestamp);
3340 nlines++;
3341
3342 if (!full && db_indent)
3343 return;
3344
3345 db_indent += 2;
3346 for (entry = map->header.next; entry != &map->header;
3347 entry = entry->next) {
3348 db_iprintf("map entry %p: start=%p, end=%p\n",
3349 (void *)entry, (void *)entry->start, (void *)entry->end);
3350 nlines++;
3351 {
3352 static char *inheritance_name[4] =
3353 {"share", "copy", "none", "donate_copy"};
3354
3355 db_iprintf(" prot=%x/%x/%s",
3356 entry->protection,
3357 entry->max_protection,
3358 inheritance_name[(int)(unsigned char)entry->inheritance]);
3359 if (entry->wired_count != 0)
3360 db_printf(", wired");
3361 }
3362 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3363 db_printf(", share=%p, offset=0x%jx\n",
3364 (void *)entry->object.sub_map,
3365 (uintmax_t)entry->offset);
3366 nlines++;
3367 if ((entry->prev == &map->header) ||
3368 (entry->prev->object.sub_map !=
3369 entry->object.sub_map)) {
3370 db_indent += 2;
3371 vm_map_print((db_expr_t)(intptr_t)
3372 entry->object.sub_map,
3373 full, 0, (char *)0);
3374 db_indent -= 2;
3375 }
3376 } else {
3377 db_printf(", object=%p, offset=0x%jx",
3378 (void *)entry->object.vm_object,
3379 (uintmax_t)entry->offset);
3380 if (entry->eflags & MAP_ENTRY_COW)
3381 db_printf(", copy (%s)",
3382 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3383 db_printf("\n");
3384 nlines++;
3385
3386 if ((entry->prev == &map->header) ||
3387 (entry->prev->object.vm_object !=
3388 entry->object.vm_object)) {
3389 db_indent += 2;
3390 vm_object_print((db_expr_t)(intptr_t)
3391 entry->object.vm_object,
3392 full, 0, (char *)0);
3393 nlines += 4;
3394 db_indent -= 2;
3395 }
3396 }
3397 }
3398 db_indent -= 2;
3399 if (db_indent == 0)
3400 nlines = 0;
3401 }
3402
3403
3404 DB_SHOW_COMMAND(procvm, procvm)
3405 {
3406 struct proc *p;
3407
3408 if (have_addr) {
3409 p = (struct proc *) addr;
3410 } else {
3411 p = curproc;
3412 }
3413
3414 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3415 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3416 (void *)vmspace_pmap(p->p_vmspace));
3417
3418 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3419 }
3420
3421 #endif /* DDB */
Cache object: a19356918bceb67b54199bbf930d08ee
|