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/8.4/sys/vm/vm_map.c 267019 2014-06-03 19:03:23Z delphij $");
67
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
71 #include <sys/ktr.h>
72 #include <sys/lock.h>
73 #include <sys/mutex.h>
74 #include <sys/proc.h>
75 #include <sys/vmmeter.h>
76 #include <sys/mman.h>
77 #include <sys/vnode.h>
78 #include <sys/resourcevar.h>
79 #include <sys/file.h>
80 #include <sys/sysctl.h>
81 #include <sys/sysent.h>
82 #include <sys/shm.h>
83
84 #include <vm/vm.h>
85 #include <vm/vm_param.h>
86 #include <vm/pmap.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_pager.h>
91 #include <vm/vm_kern.h>
92 #include <vm/vm_extern.h>
93 #include <vm/swap_pager.h>
94 #include <vm/uma.h>
95
96 /*
97 * Virtual memory maps provide for the mapping, protection,
98 * and sharing of virtual memory objects. In addition,
99 * this module provides for an efficient virtual copy of
100 * memory from one map to another.
101 *
102 * Synchronization is required prior to most operations.
103 *
104 * Maps consist of an ordered doubly-linked list of simple
105 * entries; a self-adjusting binary search tree of these
106 * entries is used to speed up lookups.
107 *
108 * Since portions of maps are specified by start/end addresses,
109 * which may not align with existing map entries, all
110 * routines merely "clip" entries to these start/end values.
111 * [That is, an entry is split into two, bordering at a
112 * start or end value.] Note that these clippings may not
113 * always be necessary (as the two resulting entries are then
114 * not changed); however, the clipping is done for convenience.
115 *
116 * As mentioned above, virtual copy operations are performed
117 * by copying VM object references from one map to
118 * another, and then marking both regions as copy-on-write.
119 */
120
121 static struct mtx map_sleep_mtx;
122 static uma_zone_t mapentzone;
123 static uma_zone_t kmapentzone;
124 static uma_zone_t mapzone;
125 static uma_zone_t vmspace_zone;
126 static struct vm_object kmapentobj;
127 static int vmspace_zinit(void *mem, int size, int flags);
128 static void vmspace_zfini(void *mem, int size);
129 static int vm_map_zinit(void *mem, int ize, int flags);
130 static void vm_map_zfini(void *mem, int size);
131 static void _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max);
132 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
133 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
134 #ifdef INVARIANTS
135 static void vm_map_zdtor(void *mem, int size, void *arg);
136 static void vmspace_zdtor(void *mem, int size, void *arg);
137 #endif
138
139 #define ENTRY_CHARGED(e) ((e)->uip != NULL || \
140 ((e)->object.vm_object != NULL && (e)->object.vm_object->uip != NULL && \
141 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
142
143 /*
144 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
145 * stable.
146 */
147 #define PROC_VMSPACE_LOCK(p) do { } while (0)
148 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
149
150 /*
151 * VM_MAP_RANGE_CHECK: [ internal use only ]
152 *
153 * Asserts that the starting and ending region
154 * addresses fall within the valid range of the map.
155 */
156 #define VM_MAP_RANGE_CHECK(map, start, end) \
157 { \
158 if (start < vm_map_min(map)) \
159 start = vm_map_min(map); \
160 if (end > vm_map_max(map)) \
161 end = vm_map_max(map); \
162 if (start > end) \
163 start = end; \
164 }
165
166 /*
167 * vm_map_startup:
168 *
169 * Initialize the vm_map module. Must be called before
170 * any other vm_map routines.
171 *
172 * Map and entry structures are allocated from the general
173 * purpose memory pool with some exceptions:
174 *
175 * - The kernel map and kmem submap are allocated statically.
176 * - Kernel map entries are allocated out of a static pool.
177 *
178 * These restrictions are necessary since malloc() uses the
179 * maps and requires map entries.
180 */
181
182 void
183 vm_map_startup(void)
184 {
185 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
186 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
187 #ifdef INVARIANTS
188 vm_map_zdtor,
189 #else
190 NULL,
191 #endif
192 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
193 uma_prealloc(mapzone, MAX_KMAP);
194 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
195 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
196 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
197 uma_prealloc(kmapentzone, MAX_KMAPENT);
198 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
199 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
200 }
201
202 static void
203 vmspace_zfini(void *mem, int size)
204 {
205 struct vmspace *vm;
206
207 vm = (struct vmspace *)mem;
208 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map));
209 }
210
211 static int
212 vmspace_zinit(void *mem, int size, int flags)
213 {
214 struct vmspace *vm;
215
216 vm = (struct vmspace *)mem;
217
218 vm->vm_map.pmap = NULL;
219 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
220 return (0);
221 }
222
223 static void
224 vm_map_zfini(void *mem, int size)
225 {
226 vm_map_t map;
227
228 map = (vm_map_t)mem;
229 mtx_destroy(&map->system_mtx);
230 sx_destroy(&map->lock);
231 }
232
233 static int
234 vm_map_zinit(void *mem, int size, int flags)
235 {
236 vm_map_t map;
237
238 map = (vm_map_t)mem;
239 map->nentries = 0;
240 map->size = 0;
241 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
242 sx_init(&map->lock, "user map");
243 return (0);
244 }
245
246 #ifdef INVARIANTS
247 static void
248 vmspace_zdtor(void *mem, int size, void *arg)
249 {
250 struct vmspace *vm;
251
252 vm = (struct vmspace *)mem;
253
254 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
255 }
256 static void
257 vm_map_zdtor(void *mem, int size, void *arg)
258 {
259 vm_map_t map;
260
261 map = (vm_map_t)mem;
262 KASSERT(map->nentries == 0,
263 ("map %p nentries == %d on free.",
264 map, map->nentries));
265 KASSERT(map->size == 0,
266 ("map %p size == %lu on free.",
267 map, (unsigned long)map->size));
268 }
269 #endif /* INVARIANTS */
270
271 /*
272 * Allocate a vmspace structure, including a vm_map and pmap,
273 * and initialize those structures. The refcnt is set to 1.
274 */
275 struct vmspace *
276 vmspace_alloc(min, max)
277 vm_offset_t min, max;
278 {
279 struct vmspace *vm;
280
281 vm = uma_zalloc(vmspace_zone, M_WAITOK);
282 if (vm->vm_map.pmap == NULL && !pmap_pinit(vmspace_pmap(vm))) {
283 uma_zfree(vmspace_zone, vm);
284 return (NULL);
285 }
286 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
287 _vm_map_init(&vm->vm_map, min, max);
288 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
289 vm->vm_refcnt = 1;
290 vm->vm_shm = NULL;
291 vm->vm_swrss = 0;
292 vm->vm_tsize = 0;
293 vm->vm_dsize = 0;
294 vm->vm_ssize = 0;
295 vm->vm_taddr = 0;
296 vm->vm_daddr = 0;
297 vm->vm_maxsaddr = 0;
298 return (vm);
299 }
300
301 void
302 vm_init2(void)
303 {
304 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count,
305 (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS) / PAGE_SIZE) / 8 +
306 maxproc * 2 + maxfiles);
307 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
308 #ifdef INVARIANTS
309 vmspace_zdtor,
310 #else
311 NULL,
312 #endif
313 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
314 }
315
316 static inline void
317 vmspace_dofree(struct vmspace *vm)
318 {
319
320 CTR1(KTR_VM, "vmspace_free: %p", vm);
321
322 /*
323 * Make sure any SysV shm is freed, it might not have been in
324 * exit1().
325 */
326 shmexit(vm);
327
328 /*
329 * Lock the map, to wait out all other references to it.
330 * Delete all of the mappings and pages they hold, then call
331 * the pmap module to reclaim anything left.
332 */
333 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
334 vm->vm_map.max_offset);
335
336 pmap_release(vmspace_pmap(vm));
337 vm->vm_map.pmap = NULL;
338 uma_zfree(vmspace_zone, vm);
339 }
340
341 void
342 vmspace_free(struct vmspace *vm)
343 {
344
345 if (vm->vm_refcnt == 0)
346 panic("vmspace_free: attempt to free already freed vmspace");
347
348 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
349 vmspace_dofree(vm);
350 }
351
352 void
353 vmspace_exitfree(struct proc *p)
354 {
355 struct vmspace *vm;
356
357 PROC_VMSPACE_LOCK(p);
358 vm = p->p_vmspace;
359 p->p_vmspace = NULL;
360 PROC_VMSPACE_UNLOCK(p);
361 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
362 vmspace_free(vm);
363 }
364
365 void
366 vmspace_exit(struct thread *td)
367 {
368 int refcnt;
369 struct vmspace *vm;
370 struct proc *p;
371
372 /*
373 * Release user portion of address space.
374 * This releases references to vnodes,
375 * which could cause I/O if the file has been unlinked.
376 * Need to do this early enough that we can still sleep.
377 *
378 * The last exiting process to reach this point releases as
379 * much of the environment as it can. vmspace_dofree() is the
380 * slower fallback in case another process had a temporary
381 * reference to the vmspace.
382 */
383
384 p = td->td_proc;
385 vm = p->p_vmspace;
386 atomic_add_int(&vmspace0.vm_refcnt, 1);
387 do {
388 refcnt = vm->vm_refcnt;
389 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
390 /* Switch now since other proc might free vmspace */
391 PROC_VMSPACE_LOCK(p);
392 p->p_vmspace = &vmspace0;
393 PROC_VMSPACE_UNLOCK(p);
394 pmap_activate(td);
395 }
396 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
397 if (refcnt == 1) {
398 if (p->p_vmspace != vm) {
399 /* vmspace not yet freed, switch back */
400 PROC_VMSPACE_LOCK(p);
401 p->p_vmspace = vm;
402 PROC_VMSPACE_UNLOCK(p);
403 pmap_activate(td);
404 }
405 pmap_remove_pages(vmspace_pmap(vm));
406 /* Switch now since this proc will free vmspace */
407 PROC_VMSPACE_LOCK(p);
408 p->p_vmspace = &vmspace0;
409 PROC_VMSPACE_UNLOCK(p);
410 pmap_activate(td);
411 vmspace_dofree(vm);
412 }
413 }
414
415 /* Acquire reference to vmspace owned by another process. */
416
417 struct vmspace *
418 vmspace_acquire_ref(struct proc *p)
419 {
420 struct vmspace *vm;
421 int refcnt;
422
423 PROC_VMSPACE_LOCK(p);
424 vm = p->p_vmspace;
425 if (vm == NULL) {
426 PROC_VMSPACE_UNLOCK(p);
427 return (NULL);
428 }
429 do {
430 refcnt = vm->vm_refcnt;
431 if (refcnt <= 0) { /* Avoid 0->1 transition */
432 PROC_VMSPACE_UNLOCK(p);
433 return (NULL);
434 }
435 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
436 if (vm != p->p_vmspace) {
437 PROC_VMSPACE_UNLOCK(p);
438 vmspace_free(vm);
439 return (NULL);
440 }
441 PROC_VMSPACE_UNLOCK(p);
442 return (vm);
443 }
444
445 void
446 _vm_map_lock(vm_map_t map, const char *file, int line)
447 {
448
449 if (map->system_map)
450 _mtx_lock_flags(&map->system_mtx, 0, file, line);
451 else
452 (void)_sx_xlock(&map->lock, 0, file, line);
453 map->timestamp++;
454 }
455
456 static void
457 vm_map_process_deferred(void)
458 {
459 struct thread *td;
460 vm_map_entry_t entry, next;
461
462 td = curthread;
463 entry = td->td_map_def_user;
464 td->td_map_def_user = NULL;
465 while (entry != NULL) {
466 next = entry->next;
467 vm_map_entry_deallocate(entry, FALSE);
468 entry = next;
469 }
470 }
471
472 void
473 _vm_map_unlock(vm_map_t map, const char *file, int line)
474 {
475
476 if (map->system_map)
477 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
478 else {
479 _sx_xunlock(&map->lock, file, line);
480 vm_map_process_deferred();
481 }
482 }
483
484 void
485 _vm_map_lock_read(vm_map_t map, const char *file, int line)
486 {
487
488 if (map->system_map)
489 _mtx_lock_flags(&map->system_mtx, 0, file, line);
490 else
491 (void)_sx_slock(&map->lock, 0, file, line);
492 }
493
494 void
495 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
496 {
497
498 if (map->system_map)
499 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
500 else {
501 _sx_sunlock(&map->lock, file, line);
502 vm_map_process_deferred();
503 }
504 }
505
506 int
507 _vm_map_trylock(vm_map_t map, const char *file, int line)
508 {
509 int error;
510
511 error = map->system_map ?
512 !_mtx_trylock(&map->system_mtx, 0, file, line) :
513 !_sx_try_xlock(&map->lock, file, line);
514 if (error == 0)
515 map->timestamp++;
516 return (error == 0);
517 }
518
519 int
520 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
521 {
522 int error;
523
524 error = map->system_map ?
525 !_mtx_trylock(&map->system_mtx, 0, file, line) :
526 !_sx_try_slock(&map->lock, file, line);
527 return (error == 0);
528 }
529
530 /*
531 * _vm_map_lock_upgrade: [ internal use only ]
532 *
533 * Tries to upgrade a read (shared) lock on the specified map to a write
534 * (exclusive) lock. Returns the value "" if the upgrade succeeds and a
535 * non-zero value if the upgrade fails. If the upgrade fails, the map is
536 * returned without a read or write lock held.
537 *
538 * Requires that the map be read locked.
539 */
540 int
541 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
542 {
543 unsigned int last_timestamp;
544
545 if (map->system_map) {
546 #ifdef INVARIANTS
547 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
548 #endif
549 } else {
550 if (!_sx_try_upgrade(&map->lock, file, line)) {
551 last_timestamp = map->timestamp;
552 _sx_sunlock(&map->lock, file, line);
553 vm_map_process_deferred();
554 /*
555 * If the map's timestamp does not change while the
556 * map is unlocked, then the upgrade succeeds.
557 */
558 (void)_sx_xlock(&map->lock, 0, file, line);
559 if (last_timestamp != map->timestamp) {
560 _sx_xunlock(&map->lock, file, line);
561 return (1);
562 }
563 }
564 }
565 map->timestamp++;
566 return (0);
567 }
568
569 void
570 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
571 {
572
573 if (map->system_map) {
574 #ifdef INVARIANTS
575 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
576 #endif
577 } else
578 _sx_downgrade(&map->lock, file, line);
579 }
580
581 /*
582 * vm_map_locked:
583 *
584 * Returns a non-zero value if the caller holds a write (exclusive) lock
585 * on the specified map and the value "" otherwise.
586 */
587 int
588 vm_map_locked(vm_map_t map)
589 {
590
591 if (map->system_map)
592 return (mtx_owned(&map->system_mtx));
593 else
594 return (sx_xlocked(&map->lock));
595 }
596
597 #ifdef INVARIANTS
598 static void
599 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
600 {
601
602 if (map->system_map)
603 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
604 else
605 _sx_assert(&map->lock, SA_XLOCKED, file, line);
606 }
607
608 #if 0
609 static void
610 _vm_map_assert_locked_read(vm_map_t map, const char *file, int line)
611 {
612
613 if (map->system_map)
614 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
615 else
616 _sx_assert(&map->lock, SA_SLOCKED, file, line);
617 }
618 #endif
619
620 #define VM_MAP_ASSERT_LOCKED(map) \
621 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
622 #define VM_MAP_ASSERT_LOCKED_READ(map) \
623 _vm_map_assert_locked_read(map, LOCK_FILE, LOCK_LINE)
624 #else
625 #define VM_MAP_ASSERT_LOCKED(map)
626 #define VM_MAP_ASSERT_LOCKED_READ(map)
627 #endif
628
629 /*
630 * _vm_map_unlock_and_wait:
631 *
632 * Atomically releases the lock on the specified map and puts the calling
633 * thread to sleep. The calling thread will remain asleep until either
634 * vm_map_wakeup() is performed on the map or the specified timeout is
635 * exceeded.
636 *
637 * WARNING! This function does not perform deferred deallocations of
638 * objects and map entries. Therefore, the calling thread is expected to
639 * reacquire the map lock after reawakening and later perform an ordinary
640 * unlock operation, such as vm_map_unlock(), before completing its
641 * operation on the map.
642 */
643 int
644 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
645 {
646
647 mtx_lock(&map_sleep_mtx);
648 if (map->system_map)
649 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
650 else
651 _sx_xunlock(&map->lock, file, line);
652 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
653 timo));
654 }
655
656 /*
657 * vm_map_wakeup:
658 *
659 * Awaken any threads that have slept on the map using
660 * vm_map_unlock_and_wait().
661 */
662 void
663 vm_map_wakeup(vm_map_t map)
664 {
665
666 /*
667 * Acquire and release map_sleep_mtx to prevent a wakeup()
668 * from being performed (and lost) between the map unlock
669 * and the msleep() in _vm_map_unlock_and_wait().
670 */
671 mtx_lock(&map_sleep_mtx);
672 mtx_unlock(&map_sleep_mtx);
673 wakeup(&map->root);
674 }
675
676 void
677 vm_map_busy(vm_map_t map)
678 {
679
680 VM_MAP_ASSERT_LOCKED(map);
681 map->busy++;
682 }
683
684 void
685 vm_map_unbusy(vm_map_t map)
686 {
687
688 VM_MAP_ASSERT_LOCKED(map);
689 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
690 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
691 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
692 wakeup(&map->busy);
693 }
694 }
695
696 void
697 vm_map_wait_busy(vm_map_t map)
698 {
699
700 VM_MAP_ASSERT_LOCKED(map);
701 while (map->busy) {
702 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
703 if (map->system_map)
704 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
705 else
706 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
707 }
708 map->timestamp++;
709 }
710
711 long
712 vmspace_resident_count(struct vmspace *vmspace)
713 {
714 return pmap_resident_count(vmspace_pmap(vmspace));
715 }
716
717 long
718 vmspace_wired_count(struct vmspace *vmspace)
719 {
720 return pmap_wired_count(vmspace_pmap(vmspace));
721 }
722
723 /*
724 * vm_map_create:
725 *
726 * Creates and returns a new empty VM map with
727 * the given physical map structure, and having
728 * the given lower and upper address bounds.
729 */
730 vm_map_t
731 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
732 {
733 vm_map_t result;
734
735 result = uma_zalloc(mapzone, M_WAITOK);
736 CTR1(KTR_VM, "vm_map_create: %p", result);
737 _vm_map_init(result, min, max);
738 result->pmap = pmap;
739 return (result);
740 }
741
742 /*
743 * Initialize an existing vm_map structure
744 * such as that in the vmspace structure.
745 * The pmap is set elsewhere.
746 */
747 static void
748 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
749 {
750
751 map->header.next = map->header.prev = &map->header;
752 map->needs_wakeup = FALSE;
753 map->system_map = 0;
754 map->min_offset = min;
755 map->max_offset = max;
756 map->flags = 0;
757 map->root = NULL;
758 map->timestamp = 0;
759 map->busy = 0;
760 }
761
762 void
763 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
764 {
765 _vm_map_init(map, min, max);
766 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
767 sx_init(&map->lock, "user map");
768 }
769
770 /*
771 * vm_map_entry_dispose: [ internal use only ]
772 *
773 * Inverse of vm_map_entry_create.
774 */
775 static void
776 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
777 {
778 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
779 }
780
781 /*
782 * vm_map_entry_create: [ internal use only ]
783 *
784 * Allocates a VM map entry for insertion.
785 * No entry fields are filled in.
786 */
787 static vm_map_entry_t
788 vm_map_entry_create(vm_map_t map)
789 {
790 vm_map_entry_t new_entry;
791
792 if (map->system_map)
793 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
794 else
795 new_entry = uma_zalloc(mapentzone, M_WAITOK);
796 if (new_entry == NULL)
797 panic("vm_map_entry_create: kernel resources exhausted");
798 return (new_entry);
799 }
800
801 /*
802 * vm_map_entry_set_behavior:
803 *
804 * Set the expected access behavior, either normal, random, or
805 * sequential.
806 */
807 static inline void
808 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
809 {
810 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
811 (behavior & MAP_ENTRY_BEHAV_MASK);
812 }
813
814 /*
815 * vm_map_entry_set_max_free:
816 *
817 * Set the max_free field in a vm_map_entry.
818 */
819 static inline void
820 vm_map_entry_set_max_free(vm_map_entry_t entry)
821 {
822
823 entry->max_free = entry->adj_free;
824 if (entry->left != NULL && entry->left->max_free > entry->max_free)
825 entry->max_free = entry->left->max_free;
826 if (entry->right != NULL && entry->right->max_free > entry->max_free)
827 entry->max_free = entry->right->max_free;
828 }
829
830 /*
831 * vm_map_entry_splay:
832 *
833 * The Sleator and Tarjan top-down splay algorithm with the
834 * following variation. Max_free must be computed bottom-up, so
835 * on the downward pass, maintain the left and right spines in
836 * reverse order. Then, make a second pass up each side to fix
837 * the pointers and compute max_free. The time bound is O(log n)
838 * amortized.
839 *
840 * The new root is the vm_map_entry containing "addr", or else an
841 * adjacent entry (lower or higher) if addr is not in the tree.
842 *
843 * The map must be locked, and leaves it so.
844 *
845 * Returns: the new root.
846 */
847 static vm_map_entry_t
848 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
849 {
850 vm_map_entry_t llist, rlist;
851 vm_map_entry_t ltree, rtree;
852 vm_map_entry_t y;
853
854 /* Special case of empty tree. */
855 if (root == NULL)
856 return (root);
857
858 /*
859 * Pass One: Splay down the tree until we find addr or a NULL
860 * pointer where addr would go. llist and rlist are the two
861 * sides in reverse order (bottom-up), with llist linked by
862 * the right pointer and rlist linked by the left pointer in
863 * the vm_map_entry. Wait until Pass Two to set max_free on
864 * the two spines.
865 */
866 llist = NULL;
867 rlist = NULL;
868 for (;;) {
869 /* root is never NULL in here. */
870 if (addr < root->start) {
871 y = root->left;
872 if (y == NULL)
873 break;
874 if (addr < y->start && y->left != NULL) {
875 /* Rotate right and put y on rlist. */
876 root->left = y->right;
877 y->right = root;
878 vm_map_entry_set_max_free(root);
879 root = y->left;
880 y->left = rlist;
881 rlist = y;
882 } else {
883 /* Put root on rlist. */
884 root->left = rlist;
885 rlist = root;
886 root = y;
887 }
888 } else if (addr >= root->end) {
889 y = root->right;
890 if (y == NULL)
891 break;
892 if (addr >= y->end && y->right != NULL) {
893 /* Rotate left and put y on llist. */
894 root->right = y->left;
895 y->left = root;
896 vm_map_entry_set_max_free(root);
897 root = y->right;
898 y->right = llist;
899 llist = y;
900 } else {
901 /* Put root on llist. */
902 root->right = llist;
903 llist = root;
904 root = y;
905 }
906 } else
907 break;
908 }
909
910 /*
911 * Pass Two: Walk back up the two spines, flip the pointers
912 * and set max_free. The subtrees of the root go at the
913 * bottom of llist and rlist.
914 */
915 ltree = root->left;
916 while (llist != NULL) {
917 y = llist->right;
918 llist->right = ltree;
919 vm_map_entry_set_max_free(llist);
920 ltree = llist;
921 llist = y;
922 }
923 rtree = root->right;
924 while (rlist != NULL) {
925 y = rlist->left;
926 rlist->left = rtree;
927 vm_map_entry_set_max_free(rlist);
928 rtree = rlist;
929 rlist = y;
930 }
931
932 /*
933 * Final assembly: add ltree and rtree as subtrees of root.
934 */
935 root->left = ltree;
936 root->right = rtree;
937 vm_map_entry_set_max_free(root);
938
939 return (root);
940 }
941
942 /*
943 * vm_map_entry_{un,}link:
944 *
945 * Insert/remove entries from maps.
946 */
947 static void
948 vm_map_entry_link(vm_map_t map,
949 vm_map_entry_t after_where,
950 vm_map_entry_t entry)
951 {
952
953 CTR4(KTR_VM,
954 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
955 map->nentries, entry, after_where);
956 VM_MAP_ASSERT_LOCKED(map);
957 map->nentries++;
958 entry->prev = after_where;
959 entry->next = after_where->next;
960 entry->next->prev = entry;
961 after_where->next = entry;
962
963 if (after_where != &map->header) {
964 if (after_where != map->root)
965 vm_map_entry_splay(after_where->start, map->root);
966 entry->right = after_where->right;
967 entry->left = after_where;
968 after_where->right = NULL;
969 after_where->adj_free = entry->start - after_where->end;
970 vm_map_entry_set_max_free(after_where);
971 } else {
972 entry->right = map->root;
973 entry->left = NULL;
974 }
975 entry->adj_free = (entry->next == &map->header ? map->max_offset :
976 entry->next->start) - entry->end;
977 vm_map_entry_set_max_free(entry);
978 map->root = entry;
979 }
980
981 static void
982 vm_map_entry_unlink(vm_map_t map,
983 vm_map_entry_t entry)
984 {
985 vm_map_entry_t next, prev, root;
986
987 VM_MAP_ASSERT_LOCKED(map);
988 if (entry != map->root)
989 vm_map_entry_splay(entry->start, map->root);
990 if (entry->left == NULL)
991 root = entry->right;
992 else {
993 root = vm_map_entry_splay(entry->start, entry->left);
994 root->right = entry->right;
995 root->adj_free = (entry->next == &map->header ? map->max_offset :
996 entry->next->start) - root->end;
997 vm_map_entry_set_max_free(root);
998 }
999 map->root = root;
1000
1001 prev = entry->prev;
1002 next = entry->next;
1003 next->prev = prev;
1004 prev->next = next;
1005 map->nentries--;
1006 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1007 map->nentries, entry);
1008 }
1009
1010 /*
1011 * vm_map_entry_resize_free:
1012 *
1013 * Recompute the amount of free space following a vm_map_entry
1014 * and propagate that value up the tree. Call this function after
1015 * resizing a map entry in-place, that is, without a call to
1016 * vm_map_entry_link() or _unlink().
1017 *
1018 * The map must be locked, and leaves it so.
1019 */
1020 static void
1021 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1022 {
1023
1024 /*
1025 * Using splay trees without parent pointers, propagating
1026 * max_free up the tree is done by moving the entry to the
1027 * root and making the change there.
1028 */
1029 if (entry != map->root)
1030 map->root = vm_map_entry_splay(entry->start, map->root);
1031
1032 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1033 entry->next->start) - entry->end;
1034 vm_map_entry_set_max_free(entry);
1035 }
1036
1037 /*
1038 * vm_map_lookup_entry: [ internal use only ]
1039 *
1040 * Finds the map entry containing (or
1041 * immediately preceding) the specified address
1042 * in the given map; the entry is returned
1043 * in the "entry" parameter. The boolean
1044 * result indicates whether the address is
1045 * actually contained in the map.
1046 */
1047 boolean_t
1048 vm_map_lookup_entry(
1049 vm_map_t map,
1050 vm_offset_t address,
1051 vm_map_entry_t *entry) /* OUT */
1052 {
1053 vm_map_entry_t cur;
1054 boolean_t locked;
1055
1056 /*
1057 * If the map is empty, then the map entry immediately preceding
1058 * "address" is the map's header.
1059 */
1060 cur = map->root;
1061 if (cur == NULL)
1062 *entry = &map->header;
1063 else if (address >= cur->start && cur->end > address) {
1064 *entry = cur;
1065 return (TRUE);
1066 } else if ((locked = vm_map_locked(map)) ||
1067 sx_try_upgrade(&map->lock)) {
1068 /*
1069 * Splay requires a write lock on the map. However, it only
1070 * restructures the binary search tree; it does not otherwise
1071 * change the map. Thus, the map's timestamp need not change
1072 * on a temporary upgrade.
1073 */
1074 map->root = cur = vm_map_entry_splay(address, cur);
1075 if (!locked)
1076 sx_downgrade(&map->lock);
1077
1078 /*
1079 * If "address" is contained within a map entry, the new root
1080 * is that map entry. Otherwise, the new root is a map entry
1081 * immediately before or after "address".
1082 */
1083 if (address >= cur->start) {
1084 *entry = cur;
1085 if (cur->end > address)
1086 return (TRUE);
1087 } else
1088 *entry = cur->prev;
1089 } else
1090 /*
1091 * Since the map is only locked for read access, perform a
1092 * standard binary search tree lookup for "address".
1093 */
1094 for (;;) {
1095 if (address < cur->start) {
1096 if (cur->left == NULL) {
1097 *entry = cur->prev;
1098 break;
1099 }
1100 cur = cur->left;
1101 } else if (cur->end > address) {
1102 *entry = cur;
1103 return (TRUE);
1104 } else {
1105 if (cur->right == NULL) {
1106 *entry = cur;
1107 break;
1108 }
1109 cur = cur->right;
1110 }
1111 }
1112 return (FALSE);
1113 }
1114
1115 /*
1116 * vm_map_insert:
1117 *
1118 * Inserts the given whole VM object into the target
1119 * map at the specified address range. The object's
1120 * size should match that of the address range.
1121 *
1122 * Requires that the map be locked, and leaves it so.
1123 *
1124 * If object is non-NULL, ref count must be bumped by caller
1125 * prior to making call to account for the new entry.
1126 */
1127 int
1128 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1129 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1130 int cow)
1131 {
1132 vm_map_entry_t new_entry;
1133 vm_map_entry_t prev_entry;
1134 vm_map_entry_t temp_entry;
1135 vm_eflags_t protoeflags;
1136 struct uidinfo *uip;
1137 boolean_t charge_prev_obj;
1138
1139 VM_MAP_ASSERT_LOCKED(map);
1140
1141 /*
1142 * Check that the start and end points are not bogus.
1143 */
1144 if ((start < map->min_offset) || (end > map->max_offset) ||
1145 (start >= end))
1146 return (KERN_INVALID_ADDRESS);
1147
1148 /*
1149 * Find the entry prior to the proposed starting address; if it's part
1150 * of an existing entry, this range is bogus.
1151 */
1152 if (vm_map_lookup_entry(map, start, &temp_entry))
1153 return (KERN_NO_SPACE);
1154
1155 prev_entry = temp_entry;
1156
1157 /*
1158 * Assert that the next entry doesn't overlap the end point.
1159 */
1160 if ((prev_entry->next != &map->header) &&
1161 (prev_entry->next->start < end))
1162 return (KERN_NO_SPACE);
1163
1164 protoeflags = 0;
1165 charge_prev_obj = FALSE;
1166
1167 if (cow & MAP_COPY_ON_WRITE)
1168 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1169
1170 if (cow & MAP_NOFAULT) {
1171 protoeflags |= MAP_ENTRY_NOFAULT;
1172
1173 KASSERT(object == NULL,
1174 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1175 }
1176 if (cow & MAP_DISABLE_SYNCER)
1177 protoeflags |= MAP_ENTRY_NOSYNC;
1178 if (cow & MAP_DISABLE_COREDUMP)
1179 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1180
1181 uip = NULL;
1182 KASSERT((object != kmem_object && object != kernel_object) ||
1183 ((object == kmem_object || object == kernel_object) &&
1184 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1185 ("kmem or kernel object and cow"));
1186 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1187 goto charged;
1188 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1189 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1190 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1191 return (KERN_RESOURCE_SHORTAGE);
1192 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1193 object->uip == NULL,
1194 ("OVERCOMMIT: vm_map_insert o %p", object));
1195 uip = curthread->td_ucred->cr_ruidinfo;
1196 uihold(uip);
1197 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1198 charge_prev_obj = TRUE;
1199 }
1200
1201 charged:
1202 /* Expand the kernel pmap, if necessary. */
1203 if (map == kernel_map && end > kernel_vm_end)
1204 pmap_growkernel(end);
1205 if (object != NULL) {
1206 /*
1207 * OBJ_ONEMAPPING must be cleared unless this mapping
1208 * is trivially proven to be the only mapping for any
1209 * of the object's pages. (Object granularity
1210 * reference counting is insufficient to recognize
1211 * aliases with precision.)
1212 */
1213 VM_OBJECT_LOCK(object);
1214 if (object->ref_count > 1 || object->shadow_count != 0)
1215 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1216 VM_OBJECT_UNLOCK(object);
1217 }
1218 else if ((prev_entry != &map->header) &&
1219 (prev_entry->eflags == protoeflags) &&
1220 (cow & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) == 0 &&
1221 (prev_entry->end == start) &&
1222 (prev_entry->wired_count == 0) &&
1223 (prev_entry->uip == uip ||
1224 (prev_entry->object.vm_object != NULL &&
1225 (prev_entry->object.vm_object->uip == uip))) &&
1226 vm_object_coalesce(prev_entry->object.vm_object,
1227 prev_entry->offset,
1228 (vm_size_t)(prev_entry->end - prev_entry->start),
1229 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1230 /*
1231 * We were able to extend the object. Determine if we
1232 * can extend the previous map entry to include the
1233 * new range as well.
1234 */
1235 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
1236 (prev_entry->protection == prot) &&
1237 (prev_entry->max_protection == max)) {
1238 map->size += (end - prev_entry->end);
1239 prev_entry->end = end;
1240 vm_map_entry_resize_free(map, prev_entry);
1241 vm_map_simplify_entry(map, prev_entry);
1242 if (uip != NULL)
1243 uifree(uip);
1244 return (KERN_SUCCESS);
1245 }
1246
1247 /*
1248 * If we can extend the object but cannot extend the
1249 * map entry, we have to create a new map entry. We
1250 * must bump the ref count on the extended object to
1251 * account for it. object may be NULL.
1252 */
1253 object = prev_entry->object.vm_object;
1254 offset = prev_entry->offset +
1255 (prev_entry->end - prev_entry->start);
1256 vm_object_reference(object);
1257 if (uip != NULL && object != NULL && object->uip != NULL &&
1258 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1259 /* Object already accounts for this uid. */
1260 uifree(uip);
1261 uip = NULL;
1262 }
1263 }
1264
1265 /*
1266 * NOTE: if conditionals fail, object can be NULL here. This occurs
1267 * in things like the buffer map where we manage kva but do not manage
1268 * backing objects.
1269 */
1270
1271 /*
1272 * Create a new entry
1273 */
1274 new_entry = vm_map_entry_create(map);
1275 new_entry->start = start;
1276 new_entry->end = end;
1277 new_entry->uip = NULL;
1278
1279 new_entry->eflags = protoeflags;
1280 new_entry->object.vm_object = object;
1281 new_entry->offset = offset;
1282 new_entry->avail_ssize = 0;
1283
1284 new_entry->inheritance = VM_INHERIT_DEFAULT;
1285 new_entry->protection = prot;
1286 new_entry->max_protection = max;
1287 new_entry->wired_count = 0;
1288
1289 KASSERT(uip == NULL || !ENTRY_CHARGED(new_entry),
1290 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1291 new_entry->uip = uip;
1292
1293 /*
1294 * Insert the new entry into the list
1295 */
1296 vm_map_entry_link(map, prev_entry, new_entry);
1297 map->size += new_entry->end - new_entry->start;
1298
1299 #if 0
1300 /*
1301 * Temporarily removed to avoid MAP_STACK panic, due to
1302 * MAP_STACK being a huge hack. Will be added back in
1303 * when MAP_STACK (and the user stack mapping) is fixed.
1304 */
1305 /*
1306 * It may be possible to simplify the entry
1307 */
1308 vm_map_simplify_entry(map, new_entry);
1309 #endif
1310
1311 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1312 vm_map_pmap_enter(map, start, prot,
1313 object, OFF_TO_IDX(offset), end - start,
1314 cow & MAP_PREFAULT_PARTIAL);
1315 }
1316
1317 return (KERN_SUCCESS);
1318 }
1319
1320 /*
1321 * vm_map_findspace:
1322 *
1323 * Find the first fit (lowest VM address) for "length" free bytes
1324 * beginning at address >= start in the given map.
1325 *
1326 * In a vm_map_entry, "adj_free" is the amount of free space
1327 * adjacent (higher address) to this entry, and "max_free" is the
1328 * maximum amount of contiguous free space in its subtree. This
1329 * allows finding a free region in one path down the tree, so
1330 * O(log n) amortized with splay trees.
1331 *
1332 * The map must be locked, and leaves it so.
1333 *
1334 * Returns: 0 on success, and starting address in *addr,
1335 * 1 if insufficient space.
1336 */
1337 int
1338 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1339 vm_offset_t *addr) /* OUT */
1340 {
1341 vm_map_entry_t entry;
1342 vm_offset_t st;
1343
1344 /*
1345 * Request must fit within min/max VM address and must avoid
1346 * address wrap.
1347 */
1348 if (start < map->min_offset)
1349 start = map->min_offset;
1350 if (start + length > map->max_offset || start + length < start)
1351 return (1);
1352
1353 /* Empty tree means wide open address space. */
1354 if (map->root == NULL) {
1355 *addr = start;
1356 return (0);
1357 }
1358
1359 /*
1360 * After splay, if start comes before root node, then there
1361 * must be a gap from start to the root.
1362 */
1363 map->root = vm_map_entry_splay(start, map->root);
1364 if (start + length <= map->root->start) {
1365 *addr = start;
1366 return (0);
1367 }
1368
1369 /*
1370 * Root is the last node that might begin its gap before
1371 * start, and this is the last comparison where address
1372 * wrap might be a problem.
1373 */
1374 st = (start > map->root->end) ? start : map->root->end;
1375 if (length <= map->root->end + map->root->adj_free - st) {
1376 *addr = st;
1377 return (0);
1378 }
1379
1380 /* With max_free, can immediately tell if no solution. */
1381 entry = map->root->right;
1382 if (entry == NULL || length > entry->max_free)
1383 return (1);
1384
1385 /*
1386 * Search the right subtree in the order: left subtree, root,
1387 * right subtree (first fit). The previous splay implies that
1388 * all regions in the right subtree have addresses > start.
1389 */
1390 while (entry != NULL) {
1391 if (entry->left != NULL && entry->left->max_free >= length)
1392 entry = entry->left;
1393 else if (entry->adj_free >= length) {
1394 *addr = entry->end;
1395 return (0);
1396 } else
1397 entry = entry->right;
1398 }
1399
1400 /* Can't get here, so panic if we do. */
1401 panic("vm_map_findspace: max_free corrupt");
1402 }
1403
1404 int
1405 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1406 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1407 vm_prot_t max, int cow)
1408 {
1409 vm_offset_t end;
1410 int result;
1411
1412 end = start + length;
1413 vm_map_lock(map);
1414 VM_MAP_RANGE_CHECK(map, start, end);
1415 (void) vm_map_delete(map, start, end);
1416 result = vm_map_insert(map, object, offset, start, end, prot,
1417 max, cow);
1418 vm_map_unlock(map);
1419 return (result);
1420 }
1421
1422 /*
1423 * vm_map_find finds an unallocated region in the target address
1424 * map with the given length. The search is defined to be
1425 * first-fit from the specified address; the region found is
1426 * returned in the same parameter.
1427 *
1428 * If object is non-NULL, ref count must be bumped by caller
1429 * prior to making call to account for the new entry.
1430 */
1431 int
1432 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1433 vm_offset_t *addr, /* IN/OUT */
1434 vm_size_t length, int find_space, vm_prot_t prot,
1435 vm_prot_t max, int cow)
1436 {
1437 vm_offset_t start;
1438 int result;
1439
1440 start = *addr;
1441 vm_map_lock(map);
1442 do {
1443 if (find_space != VMFS_NO_SPACE) {
1444 if (vm_map_findspace(map, start, length, addr)) {
1445 vm_map_unlock(map);
1446 return (KERN_NO_SPACE);
1447 }
1448 switch (find_space) {
1449 case VMFS_ALIGNED_SPACE:
1450 pmap_align_superpage(object, offset, addr,
1451 length);
1452 break;
1453 #ifdef VMFS_TLB_ALIGNED_SPACE
1454 case VMFS_TLB_ALIGNED_SPACE:
1455 pmap_align_tlb(addr);
1456 break;
1457 #endif
1458 default:
1459 break;
1460 }
1461
1462 start = *addr;
1463 }
1464 result = vm_map_insert(map, object, offset, start, start +
1465 length, prot, max, cow);
1466 } while (result == KERN_NO_SPACE && find_space == VMFS_ALIGNED_SPACE);
1467 vm_map_unlock(map);
1468 return (result);
1469 }
1470
1471 /*
1472 * vm_map_simplify_entry:
1473 *
1474 * Simplify the given map entry by merging with either neighbor. This
1475 * routine also has the ability to merge with both neighbors.
1476 *
1477 * The map must be locked.
1478 *
1479 * This routine guarentees that the passed entry remains valid (though
1480 * possibly extended). When merging, this routine may delete one or
1481 * both neighbors.
1482 */
1483 void
1484 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1485 {
1486 vm_map_entry_t next, prev;
1487 vm_size_t prevsize, esize;
1488
1489 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1490 return;
1491
1492 prev = entry->prev;
1493 if (prev != &map->header) {
1494 prevsize = prev->end - prev->start;
1495 if ( (prev->end == entry->start) &&
1496 (prev->object.vm_object == entry->object.vm_object) &&
1497 (!prev->object.vm_object ||
1498 (prev->offset + prevsize == entry->offset)) &&
1499 (prev->eflags == entry->eflags) &&
1500 (prev->protection == entry->protection) &&
1501 (prev->max_protection == entry->max_protection) &&
1502 (prev->inheritance == entry->inheritance) &&
1503 (prev->wired_count == entry->wired_count) &&
1504 (prev->uip == entry->uip)) {
1505 vm_map_entry_unlink(map, prev);
1506 entry->start = prev->start;
1507 entry->offset = prev->offset;
1508 if (entry->prev != &map->header)
1509 vm_map_entry_resize_free(map, entry->prev);
1510
1511 /*
1512 * If the backing object is a vnode object,
1513 * vm_object_deallocate() calls vrele().
1514 * However, vrele() does not lock the vnode
1515 * because the vnode has additional
1516 * references. Thus, the map lock can be kept
1517 * without causing a lock-order reversal with
1518 * the vnode lock.
1519 */
1520 if (prev->object.vm_object)
1521 vm_object_deallocate(prev->object.vm_object);
1522 if (prev->uip != NULL)
1523 uifree(prev->uip);
1524 vm_map_entry_dispose(map, prev);
1525 }
1526 }
1527
1528 next = entry->next;
1529 if (next != &map->header) {
1530 esize = entry->end - entry->start;
1531 if ((entry->end == next->start) &&
1532 (next->object.vm_object == entry->object.vm_object) &&
1533 (!entry->object.vm_object ||
1534 (entry->offset + esize == next->offset)) &&
1535 (next->eflags == entry->eflags) &&
1536 (next->protection == entry->protection) &&
1537 (next->max_protection == entry->max_protection) &&
1538 (next->inheritance == entry->inheritance) &&
1539 (next->wired_count == entry->wired_count) &&
1540 (next->uip == entry->uip)) {
1541 vm_map_entry_unlink(map, next);
1542 entry->end = next->end;
1543 vm_map_entry_resize_free(map, entry);
1544
1545 /*
1546 * See comment above.
1547 */
1548 if (next->object.vm_object)
1549 vm_object_deallocate(next->object.vm_object);
1550 if (next->uip != NULL)
1551 uifree(next->uip);
1552 vm_map_entry_dispose(map, next);
1553 }
1554 }
1555 }
1556 /*
1557 * vm_map_clip_start: [ internal use only ]
1558 *
1559 * Asserts that the given entry begins at or after
1560 * the specified address; if necessary,
1561 * it splits the entry into two.
1562 */
1563 #define vm_map_clip_start(map, entry, startaddr) \
1564 { \
1565 if (startaddr > entry->start) \
1566 _vm_map_clip_start(map, entry, startaddr); \
1567 }
1568
1569 /*
1570 * This routine is called only when it is known that
1571 * the entry must be split.
1572 */
1573 static void
1574 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1575 {
1576 vm_map_entry_t new_entry;
1577
1578 VM_MAP_ASSERT_LOCKED(map);
1579
1580 /*
1581 * Split off the front portion -- note that we must insert the new
1582 * entry BEFORE this one, so that this entry has the specified
1583 * starting address.
1584 */
1585 vm_map_simplify_entry(map, entry);
1586
1587 /*
1588 * If there is no object backing this entry, we might as well create
1589 * one now. If we defer it, an object can get created after the map
1590 * is clipped, and individual objects will be created for the split-up
1591 * map. This is a bit of a hack, but is also about the best place to
1592 * put this improvement.
1593 */
1594 if (entry->object.vm_object == NULL && !map->system_map) {
1595 vm_object_t object;
1596 object = vm_object_allocate(OBJT_DEFAULT,
1597 atop(entry->end - entry->start));
1598 entry->object.vm_object = object;
1599 entry->offset = 0;
1600 if (entry->uip != NULL) {
1601 object->uip = entry->uip;
1602 object->charge = entry->end - entry->start;
1603 entry->uip = NULL;
1604 }
1605 } else if (entry->object.vm_object != NULL &&
1606 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1607 entry->uip != NULL) {
1608 VM_OBJECT_LOCK(entry->object.vm_object);
1609 KASSERT(entry->object.vm_object->uip == NULL,
1610 ("OVERCOMMIT: vm_entry_clip_start: both uip e %p", entry));
1611 entry->object.vm_object->uip = entry->uip;
1612 entry->object.vm_object->charge = entry->end - entry->start;
1613 VM_OBJECT_UNLOCK(entry->object.vm_object);
1614 entry->uip = NULL;
1615 }
1616
1617 new_entry = vm_map_entry_create(map);
1618 *new_entry = *entry;
1619
1620 new_entry->end = start;
1621 entry->offset += (start - entry->start);
1622 entry->start = start;
1623 if (new_entry->uip != NULL)
1624 uihold(entry->uip);
1625
1626 vm_map_entry_link(map, entry->prev, new_entry);
1627
1628 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1629 vm_object_reference(new_entry->object.vm_object);
1630 }
1631 }
1632
1633 /*
1634 * vm_map_clip_end: [ internal use only ]
1635 *
1636 * Asserts that the given entry ends at or before
1637 * the specified address; if necessary,
1638 * it splits the entry into two.
1639 */
1640 #define vm_map_clip_end(map, entry, endaddr) \
1641 { \
1642 if ((endaddr) < (entry->end)) \
1643 _vm_map_clip_end((map), (entry), (endaddr)); \
1644 }
1645
1646 /*
1647 * This routine is called only when it is known that
1648 * the entry must be split.
1649 */
1650 static void
1651 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1652 {
1653 vm_map_entry_t new_entry;
1654
1655 VM_MAP_ASSERT_LOCKED(map);
1656
1657 /*
1658 * If there is no object backing this entry, we might as well create
1659 * one now. If we defer it, an object can get created after the map
1660 * is clipped, and individual objects will be created for the split-up
1661 * map. This is a bit of a hack, but is also about the best place to
1662 * put this improvement.
1663 */
1664 if (entry->object.vm_object == NULL && !map->system_map) {
1665 vm_object_t object;
1666 object = vm_object_allocate(OBJT_DEFAULT,
1667 atop(entry->end - entry->start));
1668 entry->object.vm_object = object;
1669 entry->offset = 0;
1670 if (entry->uip != NULL) {
1671 object->uip = entry->uip;
1672 object->charge = entry->end - entry->start;
1673 entry->uip = NULL;
1674 }
1675 } else if (entry->object.vm_object != NULL &&
1676 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1677 entry->uip != NULL) {
1678 VM_OBJECT_LOCK(entry->object.vm_object);
1679 KASSERT(entry->object.vm_object->uip == NULL,
1680 ("OVERCOMMIT: vm_entry_clip_end: both uip e %p", entry));
1681 entry->object.vm_object->uip = entry->uip;
1682 entry->object.vm_object->charge = entry->end - entry->start;
1683 VM_OBJECT_UNLOCK(entry->object.vm_object);
1684 entry->uip = NULL;
1685 }
1686
1687 /*
1688 * Create a new entry and insert it AFTER the specified entry
1689 */
1690 new_entry = vm_map_entry_create(map);
1691 *new_entry = *entry;
1692
1693 new_entry->start = entry->end = end;
1694 new_entry->offset += (end - entry->start);
1695 if (new_entry->uip != NULL)
1696 uihold(entry->uip);
1697
1698 vm_map_entry_link(map, entry, new_entry);
1699
1700 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1701 vm_object_reference(new_entry->object.vm_object);
1702 }
1703 }
1704
1705 /*
1706 * vm_map_submap: [ kernel use only ]
1707 *
1708 * Mark the given range as handled by a subordinate map.
1709 *
1710 * This range must have been created with vm_map_find,
1711 * and no other operations may have been performed on this
1712 * range prior to calling vm_map_submap.
1713 *
1714 * Only a limited number of operations can be performed
1715 * within this rage after calling vm_map_submap:
1716 * vm_fault
1717 * [Don't try vm_map_copy!]
1718 *
1719 * To remove a submapping, one must first remove the
1720 * range from the superior map, and then destroy the
1721 * submap (if desired). [Better yet, don't try it.]
1722 */
1723 int
1724 vm_map_submap(
1725 vm_map_t map,
1726 vm_offset_t start,
1727 vm_offset_t end,
1728 vm_map_t submap)
1729 {
1730 vm_map_entry_t entry;
1731 int result = KERN_INVALID_ARGUMENT;
1732
1733 vm_map_lock(map);
1734
1735 VM_MAP_RANGE_CHECK(map, start, end);
1736
1737 if (vm_map_lookup_entry(map, start, &entry)) {
1738 vm_map_clip_start(map, entry, start);
1739 } else
1740 entry = entry->next;
1741
1742 vm_map_clip_end(map, entry, end);
1743
1744 if ((entry->start == start) && (entry->end == end) &&
1745 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1746 (entry->object.vm_object == NULL)) {
1747 entry->object.sub_map = submap;
1748 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1749 result = KERN_SUCCESS;
1750 }
1751 vm_map_unlock(map);
1752
1753 return (result);
1754 }
1755
1756 /*
1757 * The maximum number of pages to map
1758 */
1759 #define MAX_INIT_PT 96
1760
1761 /*
1762 * vm_map_pmap_enter:
1763 *
1764 * Preload read-only mappings for the given object's resident pages into
1765 * the given map. This eliminates the soft faults on process startup and
1766 * immediately after an mmap(2). Because these are speculative mappings,
1767 * cached pages are not reactivated and mapped.
1768 */
1769 void
1770 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1771 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1772 {
1773 vm_offset_t start;
1774 vm_page_t p, p_start;
1775 vm_pindex_t psize, tmpidx;
1776 boolean_t are_queues_locked;
1777
1778 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1779 return;
1780 VM_OBJECT_LOCK(object);
1781 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1782 pmap_object_init_pt(map->pmap, addr, object, pindex, size);
1783 goto unlock_return;
1784 }
1785
1786 psize = atop(size);
1787
1788 if ((flags & MAP_PREFAULT_PARTIAL) && psize > MAX_INIT_PT &&
1789 object->resident_page_count > MAX_INIT_PT)
1790 goto unlock_return;
1791
1792 if (psize + pindex > object->size) {
1793 if (object->size < pindex)
1794 goto unlock_return;
1795 psize = object->size - pindex;
1796 }
1797
1798 are_queues_locked = FALSE;
1799 start = 0;
1800 p_start = NULL;
1801
1802 p = vm_page_find_least(object, pindex);
1803 /*
1804 * Assert: the variable p is either (1) the page with the
1805 * least pindex greater than or equal to the parameter pindex
1806 * or (2) NULL.
1807 */
1808 for (;
1809 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1810 p = TAILQ_NEXT(p, listq)) {
1811 /*
1812 * don't allow an madvise to blow away our really
1813 * free pages allocating pv entries.
1814 */
1815 if ((flags & MAP_PREFAULT_MADVISE) &&
1816 cnt.v_free_count < cnt.v_free_reserved) {
1817 psize = tmpidx;
1818 break;
1819 }
1820 if (p->valid == VM_PAGE_BITS_ALL) {
1821 if (p_start == NULL) {
1822 start = addr + ptoa(tmpidx);
1823 p_start = p;
1824 }
1825 } else if (p_start != NULL) {
1826 if (!are_queues_locked) {
1827 are_queues_locked = TRUE;
1828 vm_page_lock_queues();
1829 }
1830 pmap_enter_object(map->pmap, start, addr +
1831 ptoa(tmpidx), p_start, prot);
1832 p_start = NULL;
1833 }
1834 }
1835 if (p_start != NULL) {
1836 if (!are_queues_locked) {
1837 are_queues_locked = TRUE;
1838 vm_page_lock_queues();
1839 }
1840 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1841 p_start, prot);
1842 }
1843 if (are_queues_locked)
1844 vm_page_unlock_queues();
1845 unlock_return:
1846 VM_OBJECT_UNLOCK(object);
1847 }
1848
1849 /*
1850 * vm_map_protect:
1851 *
1852 * Sets the protection of the specified address
1853 * region in the target map. If "set_max" is
1854 * specified, the maximum protection is to be set;
1855 * otherwise, only the current protection is affected.
1856 */
1857 int
1858 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1859 vm_prot_t new_prot, boolean_t set_max)
1860 {
1861 vm_map_entry_t current, entry;
1862 vm_object_t obj;
1863 struct uidinfo *uip;
1864 vm_prot_t old_prot;
1865
1866 vm_map_lock(map);
1867
1868 VM_MAP_RANGE_CHECK(map, start, end);
1869
1870 if (vm_map_lookup_entry(map, start, &entry)) {
1871 vm_map_clip_start(map, entry, start);
1872 } else {
1873 entry = entry->next;
1874 }
1875
1876 /*
1877 * Make a first pass to check for protection violations.
1878 */
1879 current = entry;
1880 while ((current != &map->header) && (current->start < end)) {
1881 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1882 vm_map_unlock(map);
1883 return (KERN_INVALID_ARGUMENT);
1884 }
1885 if ((new_prot & current->max_protection) != new_prot) {
1886 vm_map_unlock(map);
1887 return (KERN_PROTECTION_FAILURE);
1888 }
1889 current = current->next;
1890 }
1891
1892
1893 /*
1894 * Do an accounting pass for private read-only mappings that
1895 * now will do cow due to allowed write (e.g. debugger sets
1896 * breakpoint on text segment)
1897 */
1898 for (current = entry; (current != &map->header) &&
1899 (current->start < end); current = current->next) {
1900
1901 vm_map_clip_end(map, current, end);
1902
1903 if (set_max ||
1904 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1905 ENTRY_CHARGED(current)) {
1906 continue;
1907 }
1908
1909 uip = curthread->td_ucred->cr_ruidinfo;
1910 obj = current->object.vm_object;
1911
1912 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1913 if (!swap_reserve(current->end - current->start)) {
1914 vm_map_unlock(map);
1915 return (KERN_RESOURCE_SHORTAGE);
1916 }
1917 uihold(uip);
1918 current->uip = uip;
1919 continue;
1920 }
1921
1922 VM_OBJECT_LOCK(obj);
1923 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1924 VM_OBJECT_UNLOCK(obj);
1925 continue;
1926 }
1927
1928 /*
1929 * Charge for the whole object allocation now, since
1930 * we cannot distinguish between non-charged and
1931 * charged clipped mapping of the same object later.
1932 */
1933 KASSERT(obj->charge == 0,
1934 ("vm_map_protect: object %p overcharged\n", obj));
1935 if (!swap_reserve(ptoa(obj->size))) {
1936 VM_OBJECT_UNLOCK(obj);
1937 vm_map_unlock(map);
1938 return (KERN_RESOURCE_SHORTAGE);
1939 }
1940
1941 uihold(uip);
1942 obj->uip = uip;
1943 obj->charge = ptoa(obj->size);
1944 VM_OBJECT_UNLOCK(obj);
1945 }
1946
1947 /*
1948 * Go back and fix up protections. [Note that clipping is not
1949 * necessary the second time.]
1950 */
1951 current = entry;
1952 while ((current != &map->header) && (current->start < end)) {
1953 old_prot = current->protection;
1954
1955 if (set_max)
1956 current->protection =
1957 (current->max_protection = new_prot) &
1958 old_prot;
1959 else
1960 current->protection = new_prot;
1961
1962 if ((current->eflags & (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED))
1963 == (MAP_ENTRY_COW | MAP_ENTRY_USER_WIRED) &&
1964 (current->protection & VM_PROT_WRITE) != 0 &&
1965 (old_prot & VM_PROT_WRITE) == 0) {
1966 vm_fault_copy_entry(map, map, current, current, NULL);
1967 }
1968
1969 /*
1970 * Update physical map if necessary. Worry about copy-on-write
1971 * here.
1972 */
1973 if (current->protection != old_prot) {
1974 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1975 VM_PROT_ALL)
1976 pmap_protect(map->pmap, current->start,
1977 current->end,
1978 current->protection & MASK(current));
1979 #undef MASK
1980 }
1981 vm_map_simplify_entry(map, current);
1982 current = current->next;
1983 }
1984 vm_map_unlock(map);
1985 return (KERN_SUCCESS);
1986 }
1987
1988 /*
1989 * vm_map_madvise:
1990 *
1991 * This routine traverses a processes map handling the madvise
1992 * system call. Advisories are classified as either those effecting
1993 * the vm_map_entry structure, or those effecting the underlying
1994 * objects.
1995 */
1996 int
1997 vm_map_madvise(
1998 vm_map_t map,
1999 vm_offset_t start,
2000 vm_offset_t end,
2001 int behav)
2002 {
2003 vm_map_entry_t current, entry;
2004 int modify_map = 0;
2005
2006 /*
2007 * Some madvise calls directly modify the vm_map_entry, in which case
2008 * we need to use an exclusive lock on the map and we need to perform
2009 * various clipping operations. Otherwise we only need a read-lock
2010 * on the map.
2011 */
2012 switch(behav) {
2013 case MADV_NORMAL:
2014 case MADV_SEQUENTIAL:
2015 case MADV_RANDOM:
2016 case MADV_NOSYNC:
2017 case MADV_AUTOSYNC:
2018 case MADV_NOCORE:
2019 case MADV_CORE:
2020 modify_map = 1;
2021 vm_map_lock(map);
2022 break;
2023 case MADV_WILLNEED:
2024 case MADV_DONTNEED:
2025 case MADV_FREE:
2026 vm_map_lock_read(map);
2027 break;
2028 default:
2029 return (KERN_INVALID_ARGUMENT);
2030 }
2031
2032 /*
2033 * Locate starting entry and clip if necessary.
2034 */
2035 VM_MAP_RANGE_CHECK(map, start, end);
2036
2037 if (vm_map_lookup_entry(map, start, &entry)) {
2038 if (modify_map)
2039 vm_map_clip_start(map, entry, start);
2040 } else {
2041 entry = entry->next;
2042 }
2043
2044 if (modify_map) {
2045 /*
2046 * madvise behaviors that are implemented in the vm_map_entry.
2047 *
2048 * We clip the vm_map_entry so that behavioral changes are
2049 * limited to the specified address range.
2050 */
2051 for (current = entry;
2052 (current != &map->header) && (current->start < end);
2053 current = current->next
2054 ) {
2055 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2056 continue;
2057
2058 vm_map_clip_end(map, current, end);
2059
2060 switch (behav) {
2061 case MADV_NORMAL:
2062 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2063 break;
2064 case MADV_SEQUENTIAL:
2065 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2066 break;
2067 case MADV_RANDOM:
2068 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2069 break;
2070 case MADV_NOSYNC:
2071 current->eflags |= MAP_ENTRY_NOSYNC;
2072 break;
2073 case MADV_AUTOSYNC:
2074 current->eflags &= ~MAP_ENTRY_NOSYNC;
2075 break;
2076 case MADV_NOCORE:
2077 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2078 break;
2079 case MADV_CORE:
2080 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2081 break;
2082 default:
2083 break;
2084 }
2085 vm_map_simplify_entry(map, current);
2086 }
2087 vm_map_unlock(map);
2088 } else {
2089 vm_pindex_t pstart, pend;
2090
2091 /*
2092 * madvise behaviors that are implemented in the underlying
2093 * vm_object.
2094 *
2095 * Since we don't clip the vm_map_entry, we have to clip
2096 * the vm_object pindex and count.
2097 */
2098 for (current = entry;
2099 (current != &map->header) && (current->start < end);
2100 current = current->next
2101 ) {
2102 vm_offset_t useStart;
2103
2104 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2105 continue;
2106
2107 pstart = OFF_TO_IDX(current->offset);
2108 pend = pstart + atop(current->end - current->start);
2109 useStart = current->start;
2110
2111 if (current->start < start) {
2112 pstart += atop(start - current->start);
2113 useStart = start;
2114 }
2115 if (current->end > end)
2116 pend -= atop(current->end - end);
2117
2118 if (pstart >= pend)
2119 continue;
2120
2121 vm_object_madvise(current->object.vm_object, pstart,
2122 pend, behav);
2123 if (behav == MADV_WILLNEED) {
2124 vm_map_pmap_enter(map,
2125 useStart,
2126 current->protection,
2127 current->object.vm_object,
2128 pstart,
2129 ptoa(pend - pstart),
2130 MAP_PREFAULT_MADVISE
2131 );
2132 }
2133 }
2134 vm_map_unlock_read(map);
2135 }
2136 return (0);
2137 }
2138
2139
2140 /*
2141 * vm_map_inherit:
2142 *
2143 * Sets the inheritance of the specified address
2144 * range in the target map. Inheritance
2145 * affects how the map will be shared with
2146 * child maps at the time of vmspace_fork.
2147 */
2148 int
2149 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2150 vm_inherit_t new_inheritance)
2151 {
2152 vm_map_entry_t entry;
2153 vm_map_entry_t temp_entry;
2154
2155 switch (new_inheritance) {
2156 case VM_INHERIT_NONE:
2157 case VM_INHERIT_COPY:
2158 case VM_INHERIT_SHARE:
2159 break;
2160 default:
2161 return (KERN_INVALID_ARGUMENT);
2162 }
2163 vm_map_lock(map);
2164 VM_MAP_RANGE_CHECK(map, start, end);
2165 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2166 entry = temp_entry;
2167 vm_map_clip_start(map, entry, start);
2168 } else
2169 entry = temp_entry->next;
2170 while ((entry != &map->header) && (entry->start < end)) {
2171 vm_map_clip_end(map, entry, end);
2172 entry->inheritance = new_inheritance;
2173 vm_map_simplify_entry(map, entry);
2174 entry = entry->next;
2175 }
2176 vm_map_unlock(map);
2177 return (KERN_SUCCESS);
2178 }
2179
2180 /*
2181 * vm_map_unwire:
2182 *
2183 * Implements both kernel and user unwiring.
2184 */
2185 int
2186 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2187 int flags)
2188 {
2189 vm_map_entry_t entry, first_entry, tmp_entry;
2190 vm_offset_t saved_start;
2191 unsigned int last_timestamp;
2192 int rv;
2193 boolean_t need_wakeup, result, user_unwire;
2194
2195 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2196 vm_map_lock(map);
2197 VM_MAP_RANGE_CHECK(map, start, end);
2198 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2199 if (flags & VM_MAP_WIRE_HOLESOK)
2200 first_entry = first_entry->next;
2201 else {
2202 vm_map_unlock(map);
2203 return (KERN_INVALID_ADDRESS);
2204 }
2205 }
2206 last_timestamp = map->timestamp;
2207 entry = first_entry;
2208 while (entry != &map->header && entry->start < end) {
2209 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2210 /*
2211 * We have not yet clipped the entry.
2212 */
2213 saved_start = (start >= entry->start) ? start :
2214 entry->start;
2215 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2216 if (vm_map_unlock_and_wait(map, 0)) {
2217 /*
2218 * Allow interruption of user unwiring?
2219 */
2220 }
2221 vm_map_lock(map);
2222 if (last_timestamp+1 != map->timestamp) {
2223 /*
2224 * Look again for the entry because the map was
2225 * modified while it was unlocked.
2226 * Specifically, the entry may have been
2227 * clipped, merged, or deleted.
2228 */
2229 if (!vm_map_lookup_entry(map, saved_start,
2230 &tmp_entry)) {
2231 if (flags & VM_MAP_WIRE_HOLESOK)
2232 tmp_entry = tmp_entry->next;
2233 else {
2234 if (saved_start == start) {
2235 /*
2236 * First_entry has been deleted.
2237 */
2238 vm_map_unlock(map);
2239 return (KERN_INVALID_ADDRESS);
2240 }
2241 end = saved_start;
2242 rv = KERN_INVALID_ADDRESS;
2243 goto done;
2244 }
2245 }
2246 if (entry == first_entry)
2247 first_entry = tmp_entry;
2248 else
2249 first_entry = NULL;
2250 entry = tmp_entry;
2251 }
2252 last_timestamp = map->timestamp;
2253 continue;
2254 }
2255 vm_map_clip_start(map, entry, start);
2256 vm_map_clip_end(map, entry, end);
2257 /*
2258 * Mark the entry in case the map lock is released. (See
2259 * above.)
2260 */
2261 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2262 /*
2263 * Check the map for holes in the specified region.
2264 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2265 */
2266 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2267 (entry->end < end && (entry->next == &map->header ||
2268 entry->next->start > entry->end))) {
2269 end = entry->end;
2270 rv = KERN_INVALID_ADDRESS;
2271 goto done;
2272 }
2273 /*
2274 * If system unwiring, require that the entry is system wired.
2275 */
2276 if (!user_unwire &&
2277 vm_map_entry_system_wired_count(entry) == 0) {
2278 end = entry->end;
2279 rv = KERN_INVALID_ARGUMENT;
2280 goto done;
2281 }
2282 entry = entry->next;
2283 }
2284 rv = KERN_SUCCESS;
2285 done:
2286 need_wakeup = FALSE;
2287 if (first_entry == NULL) {
2288 result = vm_map_lookup_entry(map, start, &first_entry);
2289 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2290 first_entry = first_entry->next;
2291 else
2292 KASSERT(result, ("vm_map_unwire: lookup failed"));
2293 }
2294 entry = first_entry;
2295 while (entry != &map->header && entry->start < end) {
2296 if (rv == KERN_SUCCESS && (!user_unwire ||
2297 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2298 if (user_unwire)
2299 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2300 entry->wired_count--;
2301 if (entry->wired_count == 0) {
2302 /*
2303 * Retain the map lock.
2304 */
2305 vm_fault_unwire(map, entry->start, entry->end,
2306 entry->object.vm_object != NULL &&
2307 (entry->object.vm_object->type == OBJT_DEVICE ||
2308 entry->object.vm_object->type == OBJT_SG));
2309 }
2310 }
2311 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2312 ("vm_map_unwire: in-transition flag missing"));
2313 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2314 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2315 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2316 need_wakeup = TRUE;
2317 }
2318 vm_map_simplify_entry(map, entry);
2319 entry = entry->next;
2320 }
2321 vm_map_unlock(map);
2322 if (need_wakeup)
2323 vm_map_wakeup(map);
2324 return (rv);
2325 }
2326
2327 /*
2328 * vm_map_wire:
2329 *
2330 * Implements both kernel and user wiring.
2331 */
2332 int
2333 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2334 int flags)
2335 {
2336 vm_map_entry_t entry, first_entry, tmp_entry;
2337 vm_offset_t saved_end, saved_start;
2338 unsigned int last_timestamp;
2339 int rv;
2340 boolean_t fictitious, need_wakeup, result, user_wire;
2341
2342 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2343 vm_map_lock(map);
2344 VM_MAP_RANGE_CHECK(map, start, end);
2345 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2346 if (flags & VM_MAP_WIRE_HOLESOK)
2347 first_entry = first_entry->next;
2348 else {
2349 vm_map_unlock(map);
2350 return (KERN_INVALID_ADDRESS);
2351 }
2352 }
2353 last_timestamp = map->timestamp;
2354 entry = first_entry;
2355 while (entry != &map->header && entry->start < end) {
2356 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2357 /*
2358 * We have not yet clipped the entry.
2359 */
2360 saved_start = (start >= entry->start) ? start :
2361 entry->start;
2362 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2363 if (vm_map_unlock_and_wait(map, 0)) {
2364 /*
2365 * Allow interruption of user wiring?
2366 */
2367 }
2368 vm_map_lock(map);
2369 if (last_timestamp + 1 != map->timestamp) {
2370 /*
2371 * Look again for the entry because the map was
2372 * modified while it was unlocked.
2373 * Specifically, the entry may have been
2374 * clipped, merged, or deleted.
2375 */
2376 if (!vm_map_lookup_entry(map, saved_start,
2377 &tmp_entry)) {
2378 if (flags & VM_MAP_WIRE_HOLESOK)
2379 tmp_entry = tmp_entry->next;
2380 else {
2381 if (saved_start == start) {
2382 /*
2383 * first_entry has been deleted.
2384 */
2385 vm_map_unlock(map);
2386 return (KERN_INVALID_ADDRESS);
2387 }
2388 end = saved_start;
2389 rv = KERN_INVALID_ADDRESS;
2390 goto done;
2391 }
2392 }
2393 if (entry == first_entry)
2394 first_entry = tmp_entry;
2395 else
2396 first_entry = NULL;
2397 entry = tmp_entry;
2398 }
2399 last_timestamp = map->timestamp;
2400 continue;
2401 }
2402 vm_map_clip_start(map, entry, start);
2403 vm_map_clip_end(map, entry, end);
2404 /*
2405 * Mark the entry in case the map lock is released. (See
2406 * above.)
2407 */
2408 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2409 /*
2410 *
2411 */
2412 if (entry->wired_count == 0) {
2413 if ((entry->protection & (VM_PROT_READ|VM_PROT_EXECUTE))
2414 == 0) {
2415 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2416 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2417 end = entry->end;
2418 rv = KERN_INVALID_ADDRESS;
2419 goto done;
2420 }
2421 goto next_entry;
2422 }
2423 entry->wired_count++;
2424 saved_start = entry->start;
2425 saved_end = entry->end;
2426 fictitious = entry->object.vm_object != NULL &&
2427 (entry->object.vm_object->type == OBJT_DEVICE ||
2428 entry->object.vm_object->type == OBJT_SG);
2429 /*
2430 * Release the map lock, relying on the in-transition
2431 * mark. Mark the map busy for fork.
2432 */
2433 vm_map_busy(map);
2434 vm_map_unlock(map);
2435 rv = vm_fault_wire(map, saved_start, saved_end,
2436 user_wire, fictitious);
2437 vm_map_lock(map);
2438 vm_map_unbusy(map);
2439 if (last_timestamp + 1 != map->timestamp) {
2440 /*
2441 * Look again for the entry because the map was
2442 * modified while it was unlocked. The entry
2443 * may have been clipped, but NOT merged or
2444 * deleted.
2445 */
2446 result = vm_map_lookup_entry(map, saved_start,
2447 &tmp_entry);
2448 KASSERT(result, ("vm_map_wire: lookup failed"));
2449 if (entry == first_entry)
2450 first_entry = tmp_entry;
2451 else
2452 first_entry = NULL;
2453 entry = tmp_entry;
2454 while (entry->end < saved_end) {
2455 if (rv != KERN_SUCCESS) {
2456 KASSERT(entry->wired_count == 1,
2457 ("vm_map_wire: bad count"));
2458 entry->wired_count = -1;
2459 }
2460 entry = entry->next;
2461 }
2462 }
2463 last_timestamp = map->timestamp;
2464 if (rv != KERN_SUCCESS) {
2465 KASSERT(entry->wired_count == 1,
2466 ("vm_map_wire: bad count"));
2467 /*
2468 * Assign an out-of-range value to represent
2469 * the failure to wire this entry.
2470 */
2471 entry->wired_count = -1;
2472 end = entry->end;
2473 goto done;
2474 }
2475 } else if (!user_wire ||
2476 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2477 entry->wired_count++;
2478 }
2479 /*
2480 * Check the map for holes in the specified region.
2481 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2482 */
2483 next_entry:
2484 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2485 (entry->end < end && (entry->next == &map->header ||
2486 entry->next->start > entry->end))) {
2487 end = entry->end;
2488 rv = KERN_INVALID_ADDRESS;
2489 goto done;
2490 }
2491 entry = entry->next;
2492 }
2493 rv = KERN_SUCCESS;
2494 done:
2495 need_wakeup = FALSE;
2496 if (first_entry == NULL) {
2497 result = vm_map_lookup_entry(map, start, &first_entry);
2498 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2499 first_entry = first_entry->next;
2500 else
2501 KASSERT(result, ("vm_map_wire: lookup failed"));
2502 }
2503 entry = first_entry;
2504 while (entry != &map->header && entry->start < end) {
2505 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2506 goto next_entry_done;
2507 if (rv == KERN_SUCCESS) {
2508 if (user_wire)
2509 entry->eflags |= MAP_ENTRY_USER_WIRED;
2510 } else if (entry->wired_count == -1) {
2511 /*
2512 * Wiring failed on this entry. Thus, unwiring is
2513 * unnecessary.
2514 */
2515 entry->wired_count = 0;
2516 } else {
2517 if (!user_wire ||
2518 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
2519 entry->wired_count--;
2520 if (entry->wired_count == 0) {
2521 /*
2522 * Retain the map lock.
2523 */
2524 vm_fault_unwire(map, entry->start, entry->end,
2525 entry->object.vm_object != NULL &&
2526 (entry->object.vm_object->type == OBJT_DEVICE ||
2527 entry->object.vm_object->type == OBJT_SG));
2528 }
2529 }
2530 next_entry_done:
2531 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
2532 ("vm_map_wire: in-transition flag missing"));
2533 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION|MAP_ENTRY_WIRE_SKIPPED);
2534 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2535 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2536 need_wakeup = TRUE;
2537 }
2538 vm_map_simplify_entry(map, entry);
2539 entry = entry->next;
2540 }
2541 vm_map_unlock(map);
2542 if (need_wakeup)
2543 vm_map_wakeup(map);
2544 return (rv);
2545 }
2546
2547 /*
2548 * vm_map_sync
2549 *
2550 * Push any dirty cached pages in the address range to their pager.
2551 * If syncio is TRUE, dirty pages are written synchronously.
2552 * If invalidate is TRUE, any cached pages are freed as well.
2553 *
2554 * If the size of the region from start to end is zero, we are
2555 * supposed to flush all modified pages within the region containing
2556 * start. Unfortunately, a region can be split or coalesced with
2557 * neighboring regions, making it difficult to determine what the
2558 * original region was. Therefore, we approximate this requirement by
2559 * flushing the current region containing start.
2560 *
2561 * Returns an error if any part of the specified range is not mapped.
2562 */
2563 int
2564 vm_map_sync(
2565 vm_map_t map,
2566 vm_offset_t start,
2567 vm_offset_t end,
2568 boolean_t syncio,
2569 boolean_t invalidate)
2570 {
2571 vm_map_entry_t current;
2572 vm_map_entry_t entry;
2573 vm_size_t size;
2574 vm_object_t object;
2575 vm_ooffset_t offset;
2576 unsigned int last_timestamp;
2577 boolean_t failed;
2578
2579 vm_map_lock_read(map);
2580 VM_MAP_RANGE_CHECK(map, start, end);
2581 if (!vm_map_lookup_entry(map, start, &entry)) {
2582 vm_map_unlock_read(map);
2583 return (KERN_INVALID_ADDRESS);
2584 } else if (start == end) {
2585 start = entry->start;
2586 end = entry->end;
2587 }
2588 /*
2589 * Make a first pass to check for user-wired memory and holes.
2590 */
2591 for (current = entry; current != &map->header && current->start < end;
2592 current = current->next) {
2593 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2594 vm_map_unlock_read(map);
2595 return (KERN_INVALID_ARGUMENT);
2596 }
2597 if (end > current->end &&
2598 (current->next == &map->header ||
2599 current->end != current->next->start)) {
2600 vm_map_unlock_read(map);
2601 return (KERN_INVALID_ADDRESS);
2602 }
2603 }
2604
2605 if (invalidate)
2606 pmap_remove(map->pmap, start, end);
2607 failed = FALSE;
2608
2609 /*
2610 * Make a second pass, cleaning/uncaching pages from the indicated
2611 * objects as we go.
2612 */
2613 for (current = entry; current != &map->header && current->start < end;) {
2614 offset = current->offset + (start - current->start);
2615 size = (end <= current->end ? end : current->end) - start;
2616 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2617 vm_map_t smap;
2618 vm_map_entry_t tentry;
2619 vm_size_t tsize;
2620
2621 smap = current->object.sub_map;
2622 vm_map_lock_read(smap);
2623 (void) vm_map_lookup_entry(smap, offset, &tentry);
2624 tsize = tentry->end - offset;
2625 if (tsize < size)
2626 size = tsize;
2627 object = tentry->object.vm_object;
2628 offset = tentry->offset + (offset - tentry->start);
2629 vm_map_unlock_read(smap);
2630 } else {
2631 object = current->object.vm_object;
2632 }
2633 vm_object_reference(object);
2634 last_timestamp = map->timestamp;
2635 vm_map_unlock_read(map);
2636 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2637 failed = TRUE;
2638 start += size;
2639 vm_object_deallocate(object);
2640 vm_map_lock_read(map);
2641 if (last_timestamp == map->timestamp ||
2642 !vm_map_lookup_entry(map, start, ¤t))
2643 current = current->next;
2644 }
2645
2646 vm_map_unlock_read(map);
2647 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2648 }
2649
2650 /*
2651 * vm_map_entry_unwire: [ internal use only ]
2652 *
2653 * Make the region specified by this entry pageable.
2654 *
2655 * The map in question should be locked.
2656 * [This is the reason for this routine's existence.]
2657 */
2658 static void
2659 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2660 {
2661 vm_fault_unwire(map, entry->start, entry->end,
2662 entry->object.vm_object != NULL &&
2663 (entry->object.vm_object->type == OBJT_DEVICE ||
2664 entry->object.vm_object->type == OBJT_SG));
2665 entry->wired_count = 0;
2666 }
2667
2668 static void
2669 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2670 {
2671
2672 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2673 vm_object_deallocate(entry->object.vm_object);
2674 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2675 }
2676
2677 /*
2678 * vm_map_entry_delete: [ internal use only ]
2679 *
2680 * Deallocate the given entry from the target map.
2681 */
2682 static void
2683 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2684 {
2685 vm_object_t object;
2686 vm_pindex_t offidxstart, offidxend, count, size1;
2687 vm_ooffset_t size;
2688
2689 vm_map_entry_unlink(map, entry);
2690 object = entry->object.vm_object;
2691 size = entry->end - entry->start;
2692 map->size -= size;
2693
2694 if (entry->uip != NULL) {
2695 swap_release_by_uid(size, entry->uip);
2696 uifree(entry->uip);
2697 }
2698
2699 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2700 (object != NULL)) {
2701 KASSERT(entry->uip == NULL || object->uip == NULL ||
2702 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2703 ("OVERCOMMIT vm_map_entry_delete: both uip %p", entry));
2704 count = OFF_TO_IDX(size);
2705 offidxstart = OFF_TO_IDX(entry->offset);
2706 offidxend = offidxstart + count;
2707 VM_OBJECT_LOCK(object);
2708 if (object->ref_count != 1 &&
2709 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2710 object == kernel_object || object == kmem_object)) {
2711 vm_object_collapse(object);
2712 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2713 if (object->type == OBJT_SWAP)
2714 swap_pager_freespace(object, offidxstart, count);
2715 if (offidxend >= object->size &&
2716 offidxstart < object->size) {
2717 size1 = object->size;
2718 object->size = offidxstart;
2719 if (object->uip != NULL) {
2720 size1 -= object->size;
2721 KASSERT(object->charge >= ptoa(size1),
2722 ("vm_map_entry_delete: object->charge < 0"));
2723 swap_release_by_uid(ptoa(size1), object->uip);
2724 object->charge -= ptoa(size1);
2725 }
2726 }
2727 }
2728 VM_OBJECT_UNLOCK(object);
2729 } else
2730 entry->object.vm_object = NULL;
2731 if (map->system_map)
2732 vm_map_entry_deallocate(entry, TRUE);
2733 else {
2734 entry->next = curthread->td_map_def_user;
2735 curthread->td_map_def_user = entry;
2736 }
2737 }
2738
2739 /*
2740 * vm_map_delete: [ internal use only ]
2741 *
2742 * Deallocates the given address range from the target
2743 * map.
2744 */
2745 int
2746 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2747 {
2748 vm_map_entry_t entry;
2749 vm_map_entry_t first_entry;
2750
2751 VM_MAP_ASSERT_LOCKED(map);
2752
2753 /*
2754 * Find the start of the region, and clip it
2755 */
2756 if (!vm_map_lookup_entry(map, start, &first_entry))
2757 entry = first_entry->next;
2758 else {
2759 entry = first_entry;
2760 vm_map_clip_start(map, entry, start);
2761 }
2762
2763 /*
2764 * Step through all entries in this region
2765 */
2766 while ((entry != &map->header) && (entry->start < end)) {
2767 vm_map_entry_t next;
2768
2769 /*
2770 * Wait for wiring or unwiring of an entry to complete.
2771 * Also wait for any system wirings to disappear on
2772 * user maps.
2773 */
2774 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2775 (vm_map_pmap(map) != kernel_pmap &&
2776 vm_map_entry_system_wired_count(entry) != 0)) {
2777 unsigned int last_timestamp;
2778 vm_offset_t saved_start;
2779 vm_map_entry_t tmp_entry;
2780
2781 saved_start = entry->start;
2782 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2783 last_timestamp = map->timestamp;
2784 (void) vm_map_unlock_and_wait(map, 0);
2785 vm_map_lock(map);
2786 if (last_timestamp + 1 != map->timestamp) {
2787 /*
2788 * Look again for the entry because the map was
2789 * modified while it was unlocked.
2790 * Specifically, the entry may have been
2791 * clipped, merged, or deleted.
2792 */
2793 if (!vm_map_lookup_entry(map, saved_start,
2794 &tmp_entry))
2795 entry = tmp_entry->next;
2796 else {
2797 entry = tmp_entry;
2798 vm_map_clip_start(map, entry,
2799 saved_start);
2800 }
2801 }
2802 continue;
2803 }
2804 vm_map_clip_end(map, entry, end);
2805
2806 next = entry->next;
2807
2808 /*
2809 * Unwire before removing addresses from the pmap; otherwise,
2810 * unwiring will put the entries back in the pmap.
2811 */
2812 if (entry->wired_count != 0) {
2813 vm_map_entry_unwire(map, entry);
2814 }
2815
2816 pmap_remove(map->pmap, entry->start, entry->end);
2817
2818 /*
2819 * Delete the entry only after removing all pmap
2820 * entries pointing to its pages. (Otherwise, its
2821 * page frames may be reallocated, and any modify bits
2822 * will be set in the wrong object!)
2823 */
2824 vm_map_entry_delete(map, entry);
2825 entry = next;
2826 }
2827 return (KERN_SUCCESS);
2828 }
2829
2830 /*
2831 * vm_map_remove:
2832 *
2833 * Remove the given address range from the target map.
2834 * This is the exported form of vm_map_delete.
2835 */
2836 int
2837 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2838 {
2839 int result;
2840
2841 vm_map_lock(map);
2842 VM_MAP_RANGE_CHECK(map, start, end);
2843 result = vm_map_delete(map, start, end);
2844 vm_map_unlock(map);
2845 return (result);
2846 }
2847
2848 /*
2849 * vm_map_check_protection:
2850 *
2851 * Assert that the target map allows the specified privilege on the
2852 * entire address region given. The entire region must be allocated.
2853 *
2854 * WARNING! This code does not and should not check whether the
2855 * contents of the region is accessible. For example a smaller file
2856 * might be mapped into a larger address space.
2857 *
2858 * NOTE! This code is also called by munmap().
2859 *
2860 * The map must be locked. A read lock is sufficient.
2861 */
2862 boolean_t
2863 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2864 vm_prot_t protection)
2865 {
2866 vm_map_entry_t entry;
2867 vm_map_entry_t tmp_entry;
2868
2869 if (!vm_map_lookup_entry(map, start, &tmp_entry))
2870 return (FALSE);
2871 entry = tmp_entry;
2872
2873 while (start < end) {
2874 if (entry == &map->header)
2875 return (FALSE);
2876 /*
2877 * No holes allowed!
2878 */
2879 if (start < entry->start)
2880 return (FALSE);
2881 /*
2882 * Check protection associated with entry.
2883 */
2884 if ((entry->protection & protection) != protection)
2885 return (FALSE);
2886 /* go to next entry */
2887 start = entry->end;
2888 entry = entry->next;
2889 }
2890 return (TRUE);
2891 }
2892
2893 /*
2894 * vm_map_copy_entry:
2895 *
2896 * Copies the contents of the source entry to the destination
2897 * entry. The entries *must* be aligned properly.
2898 */
2899 static void
2900 vm_map_copy_entry(
2901 vm_map_t src_map,
2902 vm_map_t dst_map,
2903 vm_map_entry_t src_entry,
2904 vm_map_entry_t dst_entry,
2905 vm_ooffset_t *fork_charge)
2906 {
2907 vm_object_t src_object;
2908 vm_offset_t size;
2909 struct uidinfo *uip;
2910 int charged;
2911
2912 VM_MAP_ASSERT_LOCKED(dst_map);
2913
2914 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2915 return;
2916
2917 if (src_entry->wired_count == 0) {
2918
2919 /*
2920 * If the source entry is marked needs_copy, it is already
2921 * write-protected.
2922 */
2923 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2924 pmap_protect(src_map->pmap,
2925 src_entry->start,
2926 src_entry->end,
2927 src_entry->protection & ~VM_PROT_WRITE);
2928 }
2929
2930 /*
2931 * Make a copy of the object.
2932 */
2933 size = src_entry->end - src_entry->start;
2934 if ((src_object = src_entry->object.vm_object) != NULL) {
2935 VM_OBJECT_LOCK(src_object);
2936 charged = ENTRY_CHARGED(src_entry);
2937 if ((src_object->handle == NULL) &&
2938 (src_object->type == OBJT_DEFAULT ||
2939 src_object->type == OBJT_SWAP)) {
2940 vm_object_collapse(src_object);
2941 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2942 vm_object_split(src_entry);
2943 src_object = src_entry->object.vm_object;
2944 }
2945 }
2946 vm_object_reference_locked(src_object);
2947 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2948 if (src_entry->uip != NULL &&
2949 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
2950 KASSERT(src_object->uip == NULL,
2951 ("OVERCOMMIT: vm_map_copy_entry: uip %p",
2952 src_object));
2953 src_object->uip = src_entry->uip;
2954 src_object->charge = size;
2955 }
2956 VM_OBJECT_UNLOCK(src_object);
2957 dst_entry->object.vm_object = src_object;
2958 if (charged) {
2959 uip = curthread->td_ucred->cr_ruidinfo;
2960 uihold(uip);
2961 dst_entry->uip = uip;
2962 *fork_charge += size;
2963 if (!(src_entry->eflags &
2964 MAP_ENTRY_NEEDS_COPY)) {
2965 uihold(uip);
2966 src_entry->uip = uip;
2967 *fork_charge += size;
2968 }
2969 }
2970 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2971 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2972 dst_entry->offset = src_entry->offset;
2973 } else {
2974 dst_entry->object.vm_object = NULL;
2975 dst_entry->offset = 0;
2976 if (src_entry->uip != NULL) {
2977 dst_entry->uip = curthread->td_ucred->cr_ruidinfo;
2978 uihold(dst_entry->uip);
2979 *fork_charge += size;
2980 }
2981 }
2982
2983 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2984 dst_entry->end - dst_entry->start, src_entry->start);
2985 } else {
2986 /*
2987 * Of course, wired down pages can't be set copy-on-write.
2988 * Cause wired pages to be copied into the new map by
2989 * simulating faults (the new pages are pageable)
2990 */
2991 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
2992 fork_charge);
2993 }
2994 }
2995
2996 /*
2997 * vmspace_map_entry_forked:
2998 * Update the newly-forked vmspace each time a map entry is inherited
2999 * or copied. The values for vm_dsize and vm_tsize are approximate
3000 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3001 */
3002 static void
3003 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3004 vm_map_entry_t entry)
3005 {
3006 vm_size_t entrysize;
3007 vm_offset_t newend;
3008
3009 entrysize = entry->end - entry->start;
3010 vm2->vm_map.size += entrysize;
3011 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3012 vm2->vm_ssize += btoc(entrysize);
3013 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3014 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3015 newend = MIN(entry->end,
3016 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3017 vm2->vm_dsize += btoc(newend - entry->start);
3018 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3019 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3020 newend = MIN(entry->end,
3021 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3022 vm2->vm_tsize += btoc(newend - entry->start);
3023 }
3024 }
3025
3026 /*
3027 * vmspace_fork:
3028 * Create a new process vmspace structure and vm_map
3029 * based on those of an existing process. The new map
3030 * is based on the old map, according to the inheritance
3031 * values on the regions in that map.
3032 *
3033 * XXX It might be worth coalescing the entries added to the new vmspace.
3034 *
3035 * The source map must not be locked.
3036 */
3037 struct vmspace *
3038 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3039 {
3040 struct vmspace *vm2;
3041 vm_map_t old_map = &vm1->vm_map;
3042 vm_map_t new_map;
3043 vm_map_entry_t old_entry;
3044 vm_map_entry_t new_entry;
3045 vm_object_t object;
3046 int locked;
3047
3048 vm_map_lock(old_map);
3049 if (old_map->busy)
3050 vm_map_wait_busy(old_map);
3051 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3052 if (vm2 == NULL)
3053 goto unlock_and_return;
3054 vm2->vm_taddr = vm1->vm_taddr;
3055 vm2->vm_daddr = vm1->vm_daddr;
3056 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3057 new_map = &vm2->vm_map; /* XXX */
3058 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3059 KASSERT(locked, ("vmspace_fork: lock failed"));
3060 new_map->timestamp = 1;
3061
3062 old_entry = old_map->header.next;
3063
3064 while (old_entry != &old_map->header) {
3065 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3066 panic("vm_map_fork: encountered a submap");
3067
3068 switch (old_entry->inheritance) {
3069 case VM_INHERIT_NONE:
3070 break;
3071
3072 case VM_INHERIT_SHARE:
3073 /*
3074 * Clone the entry, creating the shared object if necessary.
3075 */
3076 object = old_entry->object.vm_object;
3077 if (object == NULL) {
3078 object = vm_object_allocate(OBJT_DEFAULT,
3079 atop(old_entry->end - old_entry->start));
3080 old_entry->object.vm_object = object;
3081 old_entry->offset = 0;
3082 if (old_entry->uip != NULL) {
3083 object->uip = old_entry->uip;
3084 object->charge = old_entry->end -
3085 old_entry->start;
3086 old_entry->uip = NULL;
3087 }
3088 }
3089
3090 /*
3091 * Add the reference before calling vm_object_shadow
3092 * to insure that a shadow object is created.
3093 */
3094 vm_object_reference(object);
3095 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3096 vm_object_shadow(&old_entry->object.vm_object,
3097 &old_entry->offset,
3098 atop(old_entry->end - old_entry->start));
3099 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3100 /* Transfer the second reference too. */
3101 vm_object_reference(
3102 old_entry->object.vm_object);
3103
3104 /*
3105 * As in vm_map_simplify_entry(), the
3106 * vnode lock will not be acquired in
3107 * this call to vm_object_deallocate().
3108 */
3109 vm_object_deallocate(object);
3110 object = old_entry->object.vm_object;
3111 }
3112 VM_OBJECT_LOCK(object);
3113 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3114 if (old_entry->uip != NULL) {
3115 KASSERT(object->uip == NULL, ("vmspace_fork both uip"));
3116 object->uip = old_entry->uip;
3117 object->charge = old_entry->end - old_entry->start;
3118 old_entry->uip = NULL;
3119 }
3120 VM_OBJECT_UNLOCK(object);
3121
3122 /*
3123 * Clone the entry, referencing the shared object.
3124 */
3125 new_entry = vm_map_entry_create(new_map);
3126 *new_entry = *old_entry;
3127 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3128 MAP_ENTRY_IN_TRANSITION);
3129 new_entry->wired_count = 0;
3130
3131 /*
3132 * Insert the entry into the new map -- we know we're
3133 * inserting at the end of the new map.
3134 */
3135 vm_map_entry_link(new_map, new_map->header.prev,
3136 new_entry);
3137 vmspace_map_entry_forked(vm1, vm2, new_entry);
3138
3139 /*
3140 * Update the physical map
3141 */
3142 pmap_copy(new_map->pmap, old_map->pmap,
3143 new_entry->start,
3144 (old_entry->end - old_entry->start),
3145 old_entry->start);
3146 break;
3147
3148 case VM_INHERIT_COPY:
3149 /*
3150 * Clone the entry and link into the map.
3151 */
3152 new_entry = vm_map_entry_create(new_map);
3153 *new_entry = *old_entry;
3154 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3155 MAP_ENTRY_IN_TRANSITION);
3156 new_entry->wired_count = 0;
3157 new_entry->object.vm_object = NULL;
3158 new_entry->uip = NULL;
3159 vm_map_entry_link(new_map, new_map->header.prev,
3160 new_entry);
3161 vmspace_map_entry_forked(vm1, vm2, new_entry);
3162 vm_map_copy_entry(old_map, new_map, old_entry,
3163 new_entry, fork_charge);
3164 break;
3165 }
3166 old_entry = old_entry->next;
3167 }
3168 unlock_and_return:
3169 vm_map_unlock(old_map);
3170 if (vm2 != NULL)
3171 vm_map_unlock(new_map);
3172
3173 return (vm2);
3174 }
3175
3176 int
3177 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3178 vm_prot_t prot, vm_prot_t max, int cow)
3179 {
3180 vm_map_entry_t new_entry, prev_entry;
3181 vm_offset_t bot, top;
3182 vm_size_t growsize, init_ssize;
3183 int orient, rv;
3184 rlim_t vmemlim;
3185
3186 /*
3187 * The stack orientation is piggybacked with the cow argument.
3188 * Extract it into orient and mask the cow argument so that we
3189 * don't pass it around further.
3190 * NOTE: We explicitly allow bi-directional stacks.
3191 */
3192 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3193 KASSERT(orient != 0, ("No stack grow direction"));
3194
3195 if (addrbos < vm_map_min(map) ||
3196 addrbos > vm_map_max(map) ||
3197 addrbos + max_ssize < addrbos)
3198 return (KERN_NO_SPACE);
3199
3200 growsize = sgrowsiz;
3201 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3202
3203 PROC_LOCK(curthread->td_proc);
3204 vmemlim = lim_cur(curthread->td_proc, RLIMIT_VMEM);
3205 PROC_UNLOCK(curthread->td_proc);
3206
3207 vm_map_lock(map);
3208
3209 /* If addr is already mapped, no go */
3210 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3211 vm_map_unlock(map);
3212 return (KERN_NO_SPACE);
3213 }
3214
3215 /* If we would blow our VMEM resource limit, no go */
3216 if (map->size + init_ssize > vmemlim) {
3217 vm_map_unlock(map);
3218 return (KERN_NO_SPACE);
3219 }
3220
3221 /*
3222 * If we can't accomodate max_ssize in the current mapping, no go.
3223 * However, we need to be aware that subsequent user mappings might
3224 * map into the space we have reserved for stack, and currently this
3225 * space is not protected.
3226 *
3227 * Hopefully we will at least detect this condition when we try to
3228 * grow the stack.
3229 */
3230 if ((prev_entry->next != &map->header) &&
3231 (prev_entry->next->start < addrbos + max_ssize)) {
3232 vm_map_unlock(map);
3233 return (KERN_NO_SPACE);
3234 }
3235
3236 /*
3237 * We initially map a stack of only init_ssize. We will grow as
3238 * needed later. Depending on the orientation of the stack (i.e.
3239 * the grow direction) we either map at the top of the range, the
3240 * bottom of the range or in the middle.
3241 *
3242 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3243 * and cow to be 0. Possibly we should eliminate these as input
3244 * parameters, and just pass these values here in the insert call.
3245 */
3246 if (orient == MAP_STACK_GROWS_DOWN)
3247 bot = addrbos + max_ssize - init_ssize;
3248 else if (orient == MAP_STACK_GROWS_UP)
3249 bot = addrbos;
3250 else
3251 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3252 top = bot + init_ssize;
3253 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3254
3255 /* Now set the avail_ssize amount. */
3256 if (rv == KERN_SUCCESS) {
3257 if (prev_entry != &map->header)
3258 vm_map_clip_end(map, prev_entry, bot);
3259 new_entry = prev_entry->next;
3260 if (new_entry->end != top || new_entry->start != bot)
3261 panic("Bad entry start/end for new stack entry");
3262
3263 new_entry->avail_ssize = max_ssize - init_ssize;
3264 if (orient & MAP_STACK_GROWS_DOWN)
3265 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3266 if (orient & MAP_STACK_GROWS_UP)
3267 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3268 }
3269
3270 vm_map_unlock(map);
3271 return (rv);
3272 }
3273
3274 static int stack_guard_page = 0;
3275 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3276 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3277 &stack_guard_page, 0,
3278 "Insert stack guard page ahead of the growable segments.");
3279
3280 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3281 * desired address is already mapped, or if we successfully grow
3282 * the stack. Also returns KERN_SUCCESS if addr is outside the
3283 * stack range (this is strange, but preserves compatibility with
3284 * the grow function in vm_machdep.c).
3285 */
3286 int
3287 vm_map_growstack(struct proc *p, vm_offset_t addr)
3288 {
3289 vm_map_entry_t next_entry, prev_entry;
3290 vm_map_entry_t new_entry, stack_entry;
3291 struct vmspace *vm = p->p_vmspace;
3292 vm_map_t map = &vm->vm_map;
3293 vm_offset_t end;
3294 vm_size_t growsize;
3295 size_t grow_amount, max_grow;
3296 rlim_t stacklim, vmemlim;
3297 int is_procstack, rv;
3298 struct uidinfo *uip;
3299
3300 Retry:
3301 PROC_LOCK(p);
3302 stacklim = lim_cur(p, RLIMIT_STACK);
3303 vmemlim = lim_cur(p, RLIMIT_VMEM);
3304 PROC_UNLOCK(p);
3305
3306 vm_map_lock_read(map);
3307
3308 /* If addr is already in the entry range, no need to grow.*/
3309 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3310 vm_map_unlock_read(map);
3311 return (KERN_SUCCESS);
3312 }
3313
3314 next_entry = prev_entry->next;
3315 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3316 /*
3317 * This entry does not grow upwards. Since the address lies
3318 * beyond this entry, the next entry (if one exists) has to
3319 * be a downward growable entry. The entry list header is
3320 * never a growable entry, so it suffices to check the flags.
3321 */
3322 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3323 vm_map_unlock_read(map);
3324 return (KERN_SUCCESS);
3325 }
3326 stack_entry = next_entry;
3327 } else {
3328 /*
3329 * This entry grows upward. If the next entry does not at
3330 * least grow downwards, this is the entry we need to grow.
3331 * otherwise we have two possible choices and we have to
3332 * select one.
3333 */
3334 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3335 /*
3336 * We have two choices; grow the entry closest to
3337 * the address to minimize the amount of growth.
3338 */
3339 if (addr - prev_entry->end <= next_entry->start - addr)
3340 stack_entry = prev_entry;
3341 else
3342 stack_entry = next_entry;
3343 } else
3344 stack_entry = prev_entry;
3345 }
3346
3347 if (stack_entry == next_entry) {
3348 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3349 KASSERT(addr < stack_entry->start, ("foo"));
3350 end = (prev_entry != &map->header) ? prev_entry->end :
3351 stack_entry->start - stack_entry->avail_ssize;
3352 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3353 max_grow = stack_entry->start - end;
3354 } else {
3355 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3356 KASSERT(addr >= stack_entry->end, ("foo"));
3357 end = (next_entry != &map->header) ? next_entry->start :
3358 stack_entry->end + stack_entry->avail_ssize;
3359 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3360 max_grow = end - stack_entry->end;
3361 }
3362
3363 if (grow_amount > stack_entry->avail_ssize) {
3364 vm_map_unlock_read(map);
3365 return (KERN_NO_SPACE);
3366 }
3367
3368 /*
3369 * If there is no longer enough space between the entries nogo, and
3370 * adjust the available space. Note: this should only happen if the
3371 * user has mapped into the stack area after the stack was created,
3372 * and is probably an error.
3373 *
3374 * This also effectively destroys any guard page the user might have
3375 * intended by limiting the stack size.
3376 */
3377 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3378 if (vm_map_lock_upgrade(map))
3379 goto Retry;
3380
3381 stack_entry->avail_ssize = max_grow;
3382
3383 vm_map_unlock(map);
3384 return (KERN_NO_SPACE);
3385 }
3386
3387 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3388
3389 /*
3390 * If this is the main process stack, see if we're over the stack
3391 * limit.
3392 */
3393 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3394 vm_map_unlock_read(map);
3395 return (KERN_NO_SPACE);
3396 }
3397
3398 /* Round up the grow amount modulo sgrowsiz */
3399 growsize = sgrowsiz;
3400 grow_amount = roundup(grow_amount, growsize);
3401 if (grow_amount > stack_entry->avail_ssize)
3402 grow_amount = stack_entry->avail_ssize;
3403 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3404 grow_amount = trunc_page((vm_size_t)stacklim) -
3405 ctob(vm->vm_ssize);
3406 }
3407
3408 /* If we would blow our VMEM resource limit, no go */
3409 if (map->size + grow_amount > vmemlim) {
3410 vm_map_unlock_read(map);
3411 return (KERN_NO_SPACE);
3412 }
3413
3414 if (vm_map_lock_upgrade(map))
3415 goto Retry;
3416
3417 if (stack_entry == next_entry) {
3418 /*
3419 * Growing downward.
3420 */
3421 /* Get the preliminary new entry start value */
3422 addr = stack_entry->start - grow_amount;
3423
3424 /*
3425 * If this puts us into the previous entry, cut back our
3426 * growth to the available space. Also, see the note above.
3427 */
3428 if (addr < end) {
3429 stack_entry->avail_ssize = max_grow;
3430 addr = end;
3431 if (stack_guard_page)
3432 addr += PAGE_SIZE;
3433 }
3434
3435 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3436 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0);
3437
3438 /* Adjust the available stack space by the amount we grew. */
3439 if (rv == KERN_SUCCESS) {
3440 if (prev_entry != &map->header)
3441 vm_map_clip_end(map, prev_entry, addr);
3442 new_entry = prev_entry->next;
3443 KASSERT(new_entry == stack_entry->prev, ("foo"));
3444 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3445 KASSERT(new_entry->start == addr, ("foo"));
3446 grow_amount = new_entry->end - new_entry->start;
3447 new_entry->avail_ssize = stack_entry->avail_ssize -
3448 grow_amount;
3449 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3450 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3451 }
3452 } else {
3453 /*
3454 * Growing upward.
3455 */
3456 addr = stack_entry->end + grow_amount;
3457
3458 /*
3459 * If this puts us into the next entry, cut back our growth
3460 * to the available space. Also, see the note above.
3461 */
3462 if (addr > end) {
3463 stack_entry->avail_ssize = end - stack_entry->end;
3464 addr = end;
3465 if (stack_guard_page)
3466 addr -= PAGE_SIZE;
3467 }
3468
3469 grow_amount = addr - stack_entry->end;
3470 uip = stack_entry->uip;
3471 if (uip == NULL && stack_entry->object.vm_object != NULL)
3472 uip = stack_entry->object.vm_object->uip;
3473 if (uip != NULL && !swap_reserve_by_uid(grow_amount, uip))
3474 rv = KERN_NO_SPACE;
3475 /* Grow the underlying object if applicable. */
3476 else if (stack_entry->object.vm_object == NULL ||
3477 vm_object_coalesce(stack_entry->object.vm_object,
3478 stack_entry->offset,
3479 (vm_size_t)(stack_entry->end - stack_entry->start),
3480 (vm_size_t)grow_amount, uip != NULL)) {
3481 map->size += (addr - stack_entry->end);
3482 /* Update the current entry. */
3483 stack_entry->end = addr;
3484 stack_entry->avail_ssize -= grow_amount;
3485 vm_map_entry_resize_free(map, stack_entry);
3486 rv = KERN_SUCCESS;
3487
3488 if (next_entry != &map->header)
3489 vm_map_clip_start(map, next_entry, addr);
3490 } else
3491 rv = KERN_FAILURE;
3492 }
3493
3494 if (rv == KERN_SUCCESS && is_procstack)
3495 vm->vm_ssize += btoc(grow_amount);
3496
3497 vm_map_unlock(map);
3498
3499 /*
3500 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3501 */
3502 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3503 vm_map_wire(map,
3504 (stack_entry == next_entry) ? addr : addr - grow_amount,
3505 (stack_entry == next_entry) ? stack_entry->start : addr,
3506 (p->p_flag & P_SYSTEM)
3507 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3508 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3509 }
3510
3511 return (rv);
3512 }
3513
3514 /*
3515 * Unshare the specified VM space for exec. If other processes are
3516 * mapped to it, then create a new one. The new vmspace is null.
3517 */
3518 int
3519 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3520 {
3521 struct vmspace *oldvmspace = p->p_vmspace;
3522 struct vmspace *newvmspace;
3523
3524 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3525 ("vmspace_exec recursed"));
3526 newvmspace = vmspace_alloc(minuser, maxuser);
3527 if (newvmspace == NULL)
3528 return (ENOMEM);
3529 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3530 /*
3531 * This code is written like this for prototype purposes. The
3532 * goal is to avoid running down the vmspace here, but let the
3533 * other process's that are still using the vmspace to finally
3534 * run it down. Even though there is little or no chance of blocking
3535 * here, it is a good idea to keep this form for future mods.
3536 */
3537 PROC_VMSPACE_LOCK(p);
3538 p->p_vmspace = newvmspace;
3539 PROC_VMSPACE_UNLOCK(p);
3540 if (p == curthread->td_proc)
3541 pmap_activate(curthread);
3542 curthread->td_pflags |= TDP_EXECVMSPC;
3543 return (0);
3544 }
3545
3546 /*
3547 * Unshare the specified VM space for forcing COW. This
3548 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3549 */
3550 int
3551 vmspace_unshare(struct proc *p)
3552 {
3553 struct vmspace *oldvmspace = p->p_vmspace;
3554 struct vmspace *newvmspace;
3555 vm_ooffset_t fork_charge;
3556
3557 if (oldvmspace->vm_refcnt == 1)
3558 return (0);
3559 fork_charge = 0;
3560 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3561 if (newvmspace == NULL)
3562 return (ENOMEM);
3563 if (!swap_reserve_by_uid(fork_charge, p->p_ucred->cr_ruidinfo)) {
3564 vmspace_free(newvmspace);
3565 return (ENOMEM);
3566 }
3567 PROC_VMSPACE_LOCK(p);
3568 p->p_vmspace = newvmspace;
3569 PROC_VMSPACE_UNLOCK(p);
3570 if (p == curthread->td_proc)
3571 pmap_activate(curthread);
3572 vmspace_free(oldvmspace);
3573 return (0);
3574 }
3575
3576 /*
3577 * vm_map_lookup:
3578 *
3579 * Finds the VM object, offset, and
3580 * protection for a given virtual address in the
3581 * specified map, assuming a page fault of the
3582 * type specified.
3583 *
3584 * Leaves the map in question locked for read; return
3585 * values are guaranteed until a vm_map_lookup_done
3586 * call is performed. Note that the map argument
3587 * is in/out; the returned map must be used in
3588 * the call to vm_map_lookup_done.
3589 *
3590 * A handle (out_entry) is returned for use in
3591 * vm_map_lookup_done, to make that fast.
3592 *
3593 * If a lookup is requested with "write protection"
3594 * specified, the map may be changed to perform virtual
3595 * copying operations, although the data referenced will
3596 * remain the same.
3597 */
3598 int
3599 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3600 vm_offset_t vaddr,
3601 vm_prot_t fault_typea,
3602 vm_map_entry_t *out_entry, /* OUT */
3603 vm_object_t *object, /* OUT */
3604 vm_pindex_t *pindex, /* OUT */
3605 vm_prot_t *out_prot, /* OUT */
3606 boolean_t *wired) /* OUT */
3607 {
3608 vm_map_entry_t entry;
3609 vm_map_t map = *var_map;
3610 vm_prot_t prot;
3611 vm_prot_t fault_type = fault_typea;
3612 vm_object_t eobject;
3613 struct uidinfo *uip;
3614 vm_ooffset_t size;
3615
3616 RetryLookup:;
3617
3618 vm_map_lock_read(map);
3619
3620 /*
3621 * Lookup the faulting address.
3622 */
3623 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3624 vm_map_unlock_read(map);
3625 return (KERN_INVALID_ADDRESS);
3626 }
3627
3628 entry = *out_entry;
3629
3630 /*
3631 * Handle submaps.
3632 */
3633 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3634 vm_map_t old_map = map;
3635
3636 *var_map = map = entry->object.sub_map;
3637 vm_map_unlock_read(old_map);
3638 goto RetryLookup;
3639 }
3640
3641 /*
3642 * Check whether this task is allowed to have this page.
3643 * Note the special case for MAP_ENTRY_COW
3644 * pages with an override. This is to implement a forced
3645 * COW for debuggers.
3646 */
3647 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3648 prot = entry->max_protection;
3649 else
3650 prot = entry->protection;
3651 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3652 if ((fault_type & prot) != fault_type) {
3653 vm_map_unlock_read(map);
3654 return (KERN_PROTECTION_FAILURE);
3655 }
3656 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3657 (entry->eflags & MAP_ENTRY_COW) &&
3658 (fault_type & VM_PROT_WRITE) &&
3659 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3660 vm_map_unlock_read(map);
3661 return (KERN_PROTECTION_FAILURE);
3662 }
3663
3664 /*
3665 * If this page is not pageable, we have to get it for all possible
3666 * accesses.
3667 */
3668 *wired = (entry->wired_count != 0);
3669 if (*wired)
3670 prot = fault_type = entry->protection;
3671 size = entry->end - entry->start;
3672 /*
3673 * If the entry was copy-on-write, we either ...
3674 */
3675 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3676 /*
3677 * If we want to write the page, we may as well handle that
3678 * now since we've got the map locked.
3679 *
3680 * If we don't need to write the page, we just demote the
3681 * permissions allowed.
3682 */
3683 if (fault_type & VM_PROT_WRITE) {
3684 /*
3685 * Make a new object, and place it in the object
3686 * chain. Note that no new references have appeared
3687 * -- one just moved from the map to the new
3688 * object.
3689 */
3690 if (vm_map_lock_upgrade(map))
3691 goto RetryLookup;
3692
3693 if (entry->uip == NULL) {
3694 /*
3695 * The debugger owner is charged for
3696 * the memory.
3697 */
3698 uip = curthread->td_ucred->cr_ruidinfo;
3699 uihold(uip);
3700 if (!swap_reserve_by_uid(size, uip)) {
3701 uifree(uip);
3702 vm_map_unlock(map);
3703 return (KERN_RESOURCE_SHORTAGE);
3704 }
3705 entry->uip = uip;
3706 }
3707 vm_object_shadow(
3708 &entry->object.vm_object,
3709 &entry->offset,
3710 atop(size));
3711 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3712 eobject = entry->object.vm_object;
3713 if (eobject->uip != NULL) {
3714 /*
3715 * The object was not shadowed.
3716 */
3717 swap_release_by_uid(size, entry->uip);
3718 uifree(entry->uip);
3719 entry->uip = NULL;
3720 } else if (entry->uip != NULL) {
3721 VM_OBJECT_LOCK(eobject);
3722 eobject->uip = entry->uip;
3723 eobject->charge = size;
3724 VM_OBJECT_UNLOCK(eobject);
3725 entry->uip = NULL;
3726 }
3727
3728 vm_map_lock_downgrade(map);
3729 } else {
3730 /*
3731 * We're attempting to read a copy-on-write page --
3732 * don't allow writes.
3733 */
3734 prot &= ~VM_PROT_WRITE;
3735 }
3736 }
3737
3738 /*
3739 * Create an object if necessary.
3740 */
3741 if (entry->object.vm_object == NULL &&
3742 !map->system_map) {
3743 if (vm_map_lock_upgrade(map))
3744 goto RetryLookup;
3745 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3746 atop(size));
3747 entry->offset = 0;
3748 if (entry->uip != NULL) {
3749 VM_OBJECT_LOCK(entry->object.vm_object);
3750 entry->object.vm_object->uip = entry->uip;
3751 entry->object.vm_object->charge = size;
3752 VM_OBJECT_UNLOCK(entry->object.vm_object);
3753 entry->uip = NULL;
3754 }
3755 vm_map_lock_downgrade(map);
3756 }
3757
3758 /*
3759 * Return the object/offset from this entry. If the entry was
3760 * copy-on-write or empty, it has been fixed up.
3761 */
3762 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3763 *object = entry->object.vm_object;
3764
3765 *out_prot = prot;
3766 return (KERN_SUCCESS);
3767 }
3768
3769 /*
3770 * vm_map_lookup_locked:
3771 *
3772 * Lookup the faulting address. A version of vm_map_lookup that returns
3773 * KERN_FAILURE instead of blocking on map lock or memory allocation.
3774 */
3775 int
3776 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
3777 vm_offset_t vaddr,
3778 vm_prot_t fault_typea,
3779 vm_map_entry_t *out_entry, /* OUT */
3780 vm_object_t *object, /* OUT */
3781 vm_pindex_t *pindex, /* OUT */
3782 vm_prot_t *out_prot, /* OUT */
3783 boolean_t *wired) /* OUT */
3784 {
3785 vm_map_entry_t entry;
3786 vm_map_t map = *var_map;
3787 vm_prot_t prot;
3788 vm_prot_t fault_type = fault_typea;
3789
3790 /*
3791 * Lookup the faulting address.
3792 */
3793 if (!vm_map_lookup_entry(map, vaddr, out_entry))
3794 return (KERN_INVALID_ADDRESS);
3795
3796 entry = *out_entry;
3797
3798 /*
3799 * Fail if the entry refers to a submap.
3800 */
3801 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3802 return (KERN_FAILURE);
3803
3804 /*
3805 * Check whether this task is allowed to have this page.
3806 * Note the special case for MAP_ENTRY_COW
3807 * pages with an override. This is to implement a forced
3808 * COW for debuggers.
3809 */
3810 if (fault_type & VM_PROT_OVERRIDE_WRITE)
3811 prot = entry->max_protection;
3812 else
3813 prot = entry->protection;
3814 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
3815 if ((fault_type & prot) != fault_type)
3816 return (KERN_PROTECTION_FAILURE);
3817 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3818 (entry->eflags & MAP_ENTRY_COW) &&
3819 (fault_type & VM_PROT_WRITE) &&
3820 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0)
3821 return (KERN_PROTECTION_FAILURE);
3822
3823 /*
3824 * If this page is not pageable, we have to get it for all possible
3825 * accesses.
3826 */
3827 *wired = (entry->wired_count != 0);
3828 if (*wired)
3829 prot = fault_type = entry->protection;
3830
3831 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3832 /*
3833 * Fail if the entry was copy-on-write for a write fault.
3834 */
3835 if (fault_type & VM_PROT_WRITE)
3836 return (KERN_FAILURE);
3837 /*
3838 * We're attempting to read a copy-on-write page --
3839 * don't allow writes.
3840 */
3841 prot &= ~VM_PROT_WRITE;
3842 }
3843
3844 /*
3845 * Fail if an object should be created.
3846 */
3847 if (entry->object.vm_object == NULL && !map->system_map)
3848 return (KERN_FAILURE);
3849
3850 /*
3851 * Return the object/offset from this entry. If the entry was
3852 * copy-on-write or empty, it has been fixed up.
3853 */
3854 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3855 *object = entry->object.vm_object;
3856
3857 *out_prot = prot;
3858 return (KERN_SUCCESS);
3859 }
3860
3861 /*
3862 * vm_map_lookup_done:
3863 *
3864 * Releases locks acquired by a vm_map_lookup
3865 * (according to the handle returned by that lookup).
3866 */
3867 void
3868 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
3869 {
3870 /*
3871 * Unlock the main-level map
3872 */
3873 vm_map_unlock_read(map);
3874 }
3875
3876 #include "opt_ddb.h"
3877 #ifdef DDB
3878 #include <sys/kernel.h>
3879
3880 #include <ddb/ddb.h>
3881
3882 /*
3883 * vm_map_print: [ debug ]
3884 */
3885 DB_SHOW_COMMAND(map, vm_map_print)
3886 {
3887 static int nlines;
3888 /* XXX convert args. */
3889 vm_map_t map = (vm_map_t)addr;
3890 boolean_t full = have_addr;
3891
3892 vm_map_entry_t entry;
3893
3894 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3895 (void *)map,
3896 (void *)map->pmap, map->nentries, map->timestamp);
3897 nlines++;
3898
3899 if (!full && db_indent)
3900 return;
3901
3902 db_indent += 2;
3903 for (entry = map->header.next; entry != &map->header;
3904 entry = entry->next) {
3905 db_iprintf("map entry %p: start=%p, end=%p\n",
3906 (void *)entry, (void *)entry->start, (void *)entry->end);
3907 nlines++;
3908 {
3909 static char *inheritance_name[4] =
3910 {"share", "copy", "none", "donate_copy"};
3911
3912 db_iprintf(" prot=%x/%x/%s",
3913 entry->protection,
3914 entry->max_protection,
3915 inheritance_name[(int)(unsigned char)entry->inheritance]);
3916 if (entry->wired_count != 0)
3917 db_printf(", wired");
3918 }
3919 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3920 db_printf(", share=%p, offset=0x%jx\n",
3921 (void *)entry->object.sub_map,
3922 (uintmax_t)entry->offset);
3923 nlines++;
3924 if ((entry->prev == &map->header) ||
3925 (entry->prev->object.sub_map !=
3926 entry->object.sub_map)) {
3927 db_indent += 2;
3928 vm_map_print((db_expr_t)(intptr_t)
3929 entry->object.sub_map,
3930 full, 0, (char *)0);
3931 db_indent -= 2;
3932 }
3933 } else {
3934 if (entry->uip != NULL)
3935 db_printf(", uip %d", entry->uip->ui_uid);
3936 db_printf(", object=%p, offset=0x%jx",
3937 (void *)entry->object.vm_object,
3938 (uintmax_t)entry->offset);
3939 if (entry->object.vm_object && entry->object.vm_object->uip)
3940 db_printf(", obj uip %d charge %jx",
3941 entry->object.vm_object->uip->ui_uid,
3942 (uintmax_t)entry->object.vm_object->charge);
3943 if (entry->eflags & MAP_ENTRY_COW)
3944 db_printf(", copy (%s)",
3945 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3946 db_printf("\n");
3947 nlines++;
3948
3949 if ((entry->prev == &map->header) ||
3950 (entry->prev->object.vm_object !=
3951 entry->object.vm_object)) {
3952 db_indent += 2;
3953 vm_object_print((db_expr_t)(intptr_t)
3954 entry->object.vm_object,
3955 full, 0, (char *)0);
3956 nlines += 4;
3957 db_indent -= 2;
3958 }
3959 }
3960 }
3961 db_indent -= 2;
3962 if (db_indent == 0)
3963 nlines = 0;
3964 }
3965
3966
3967 DB_SHOW_COMMAND(procvm, procvm)
3968 {
3969 struct proc *p;
3970
3971 if (have_addr) {
3972 p = (struct proc *) addr;
3973 } else {
3974 p = curproc;
3975 }
3976
3977 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3978 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3979 (void *)vmspace_pmap(p->p_vmspace));
3980
3981 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3982 }
3983
3984 #endif /* DDB */
Cache object: 5fe110f3abfdb82eb29842303c153d58
|