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/10.1/sys/vm/vm_map.c 290362 2015-11-04 11:27:21Z glebius $");
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/racct.h>
79 #include <sys/resourcevar.h>
80 #include <sys/rwlock.h>
81 #include <sys/file.h>
82 #include <sys/sysctl.h>
83 #include <sys/sysent.h>
84 #include <sys/shm.h>
85
86 #include <vm/vm.h>
87 #include <vm/vm_param.h>
88 #include <vm/pmap.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_pager.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_extern.h>
95 #include <vm/vnode_pager.h>
96 #include <vm/swap_pager.h>
97 #include <vm/uma.h>
98
99 /*
100 * Virtual memory maps provide for the mapping, protection,
101 * and sharing of virtual memory objects. In addition,
102 * this module provides for an efficient virtual copy of
103 * memory from one map to another.
104 *
105 * Synchronization is required prior to most operations.
106 *
107 * Maps consist of an ordered doubly-linked list of simple
108 * entries; a self-adjusting binary search tree of these
109 * entries is used to speed up lookups.
110 *
111 * Since portions of maps are specified by start/end addresses,
112 * which may not align with existing map entries, all
113 * routines merely "clip" entries to these start/end values.
114 * [That is, an entry is split into two, bordering at a
115 * start or end value.] Note that these clippings may not
116 * always be necessary (as the two resulting entries are then
117 * not changed); however, the clipping is done for convenience.
118 *
119 * As mentioned above, virtual copy operations are performed
120 * by copying VM object references from one map to
121 * another, and then marking both regions as copy-on-write.
122 */
123
124 static struct mtx map_sleep_mtx;
125 static uma_zone_t mapentzone;
126 static uma_zone_t kmapentzone;
127 static uma_zone_t mapzone;
128 static uma_zone_t vmspace_zone;
129 static int vmspace_zinit(void *mem, int size, int flags);
130 static int vm_map_zinit(void *mem, int ize, int flags);
131 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
132 vm_offset_t max);
133 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
134 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
135 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
136 #ifdef INVARIANTS
137 static void vm_map_zdtor(void *mem, int size, void *arg);
138 static void vmspace_zdtor(void *mem, int size, void *arg);
139 #endif
140 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
141 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
142 int cow);
143 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
144 vm_offset_t failed_addr);
145
146 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
147 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
148 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
149
150 /*
151 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
152 * stable.
153 */
154 #define PROC_VMSPACE_LOCK(p) do { } while (0)
155 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
156
157 /*
158 * VM_MAP_RANGE_CHECK: [ internal use only ]
159 *
160 * Asserts that the starting and ending region
161 * addresses fall within the valid range of the map.
162 */
163 #define VM_MAP_RANGE_CHECK(map, start, end) \
164 { \
165 if (start < vm_map_min(map)) \
166 start = vm_map_min(map); \
167 if (end > vm_map_max(map)) \
168 end = vm_map_max(map); \
169 if (start > end) \
170 start = end; \
171 }
172
173 /*
174 * vm_map_startup:
175 *
176 * Initialize the vm_map module. Must be called before
177 * any other vm_map routines.
178 *
179 * Map and entry structures are allocated from the general
180 * purpose memory pool with some exceptions:
181 *
182 * - The kernel map and kmem submap are allocated statically.
183 * - Kernel map entries are allocated out of a static pool.
184 *
185 * These restrictions are necessary since malloc() uses the
186 * maps and requires map entries.
187 */
188
189 void
190 vm_map_startup(void)
191 {
192 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
193 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
194 #ifdef INVARIANTS
195 vm_map_zdtor,
196 #else
197 NULL,
198 #endif
199 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
200 uma_prealloc(mapzone, MAX_KMAP);
201 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
202 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
203 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
204 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
205 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
206 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
207 #ifdef INVARIANTS
208 vmspace_zdtor,
209 #else
210 NULL,
211 #endif
212 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
213 }
214
215 static int
216 vmspace_zinit(void *mem, int size, int flags)
217 {
218 struct vmspace *vm;
219
220 vm = (struct vmspace *)mem;
221
222 vm->vm_map.pmap = NULL;
223 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
224 PMAP_LOCK_INIT(vmspace_pmap(vm));
225 return (0);
226 }
227
228 static int
229 vm_map_zinit(void *mem, int size, int flags)
230 {
231 vm_map_t map;
232
233 map = (vm_map_t)mem;
234 memset(map, 0, sizeof(*map));
235 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
236 sx_init(&map->lock, "vm map (user)");
237 return (0);
238 }
239
240 #ifdef INVARIANTS
241 static void
242 vmspace_zdtor(void *mem, int size, void *arg)
243 {
244 struct vmspace *vm;
245
246 vm = (struct vmspace *)mem;
247
248 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
249 }
250 static void
251 vm_map_zdtor(void *mem, int size, void *arg)
252 {
253 vm_map_t map;
254
255 map = (vm_map_t)mem;
256 KASSERT(map->nentries == 0,
257 ("map %p nentries == %d on free.",
258 map, map->nentries));
259 KASSERT(map->size == 0,
260 ("map %p size == %lu on free.",
261 map, (unsigned long)map->size));
262 }
263 #endif /* INVARIANTS */
264
265 /*
266 * Allocate a vmspace structure, including a vm_map and pmap,
267 * and initialize those structures. The refcnt is set to 1.
268 *
269 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
270 */
271 struct vmspace *
272 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
273 {
274 struct vmspace *vm;
275
276 vm = uma_zalloc(vmspace_zone, M_WAITOK);
277
278 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
279
280 if (pinit == NULL)
281 pinit = &pmap_pinit;
282
283 if (!pinit(vmspace_pmap(vm))) {
284 uma_zfree(vmspace_zone, vm);
285 return (NULL);
286 }
287 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
288 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
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 static void
302 vmspace_container_reset(struct proc *p)
303 {
304
305 #ifdef RACCT
306 PROC_LOCK(p);
307 racct_set(p, RACCT_DATA, 0);
308 racct_set(p, RACCT_STACK, 0);
309 racct_set(p, RACCT_RSS, 0);
310 racct_set(p, RACCT_MEMLOCK, 0);
311 racct_set(p, RACCT_VMEM, 0);
312 PROC_UNLOCK(p);
313 #endif
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 vmspace_container_reset(p);
414 }
415
416 /* Acquire reference to vmspace owned by another process. */
417
418 struct vmspace *
419 vmspace_acquire_ref(struct proc *p)
420 {
421 struct vmspace *vm;
422 int refcnt;
423
424 PROC_VMSPACE_LOCK(p);
425 vm = p->p_vmspace;
426 if (vm == NULL) {
427 PROC_VMSPACE_UNLOCK(p);
428 return (NULL);
429 }
430 do {
431 refcnt = vm->vm_refcnt;
432 if (refcnt <= 0) { /* Avoid 0->1 transition */
433 PROC_VMSPACE_UNLOCK(p);
434 return (NULL);
435 }
436 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
437 if (vm != p->p_vmspace) {
438 PROC_VMSPACE_UNLOCK(p);
439 vmspace_free(vm);
440 return (NULL);
441 }
442 PROC_VMSPACE_UNLOCK(p);
443 return (vm);
444 }
445
446 void
447 _vm_map_lock(vm_map_t map, const char *file, int line)
448 {
449
450 if (map->system_map)
451 mtx_lock_flags_(&map->system_mtx, 0, file, line);
452 else
453 sx_xlock_(&map->lock, file, line);
454 map->timestamp++;
455 }
456
457 static void
458 vm_map_process_deferred(void)
459 {
460 struct thread *td;
461 vm_map_entry_t entry, next;
462 vm_object_t object;
463
464 td = curthread;
465 entry = td->td_map_def_user;
466 td->td_map_def_user = NULL;
467 while (entry != NULL) {
468 next = entry->next;
469 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
470 /*
471 * Decrement the object's writemappings and
472 * possibly the vnode's v_writecount.
473 */
474 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
475 ("Submap with writecount"));
476 object = entry->object.vm_object;
477 KASSERT(object != NULL, ("No object for writecount"));
478 vnode_pager_release_writecount(object, entry->start,
479 entry->end);
480 }
481 vm_map_entry_deallocate(entry, FALSE);
482 entry = next;
483 }
484 }
485
486 void
487 _vm_map_unlock(vm_map_t map, const char *file, int line)
488 {
489
490 if (map->system_map)
491 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
492 else {
493 sx_xunlock_(&map->lock, file, line);
494 vm_map_process_deferred();
495 }
496 }
497
498 void
499 _vm_map_lock_read(vm_map_t map, const char *file, int line)
500 {
501
502 if (map->system_map)
503 mtx_lock_flags_(&map->system_mtx, 0, file, line);
504 else
505 sx_slock_(&map->lock, file, line);
506 }
507
508 void
509 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
510 {
511
512 if (map->system_map)
513 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
514 else {
515 sx_sunlock_(&map->lock, file, line);
516 vm_map_process_deferred();
517 }
518 }
519
520 int
521 _vm_map_trylock(vm_map_t map, const char *file, int line)
522 {
523 int error;
524
525 error = map->system_map ?
526 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
527 !sx_try_xlock_(&map->lock, file, line);
528 if (error == 0)
529 map->timestamp++;
530 return (error == 0);
531 }
532
533 int
534 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
535 {
536 int error;
537
538 error = map->system_map ?
539 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
540 !sx_try_slock_(&map->lock, file, line);
541 return (error == 0);
542 }
543
544 /*
545 * _vm_map_lock_upgrade: [ internal use only ]
546 *
547 * Tries to upgrade a read (shared) lock on the specified map to a write
548 * (exclusive) lock. Returns the value "" if the upgrade succeeds and a
549 * non-zero value if the upgrade fails. If the upgrade fails, the map is
550 * returned without a read or write lock held.
551 *
552 * Requires that the map be read locked.
553 */
554 int
555 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
556 {
557 unsigned int last_timestamp;
558
559 if (map->system_map) {
560 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
561 } else {
562 if (!sx_try_upgrade_(&map->lock, file, line)) {
563 last_timestamp = map->timestamp;
564 sx_sunlock_(&map->lock, file, line);
565 vm_map_process_deferred();
566 /*
567 * If the map's timestamp does not change while the
568 * map is unlocked, then the upgrade succeeds.
569 */
570 sx_xlock_(&map->lock, file, line);
571 if (last_timestamp != map->timestamp) {
572 sx_xunlock_(&map->lock, file, line);
573 return (1);
574 }
575 }
576 }
577 map->timestamp++;
578 return (0);
579 }
580
581 void
582 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
583 {
584
585 if (map->system_map) {
586 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
587 } else
588 sx_downgrade_(&map->lock, file, line);
589 }
590
591 /*
592 * vm_map_locked:
593 *
594 * Returns a non-zero value if the caller holds a write (exclusive) lock
595 * on the specified map and the value "" otherwise.
596 */
597 int
598 vm_map_locked(vm_map_t map)
599 {
600
601 if (map->system_map)
602 return (mtx_owned(&map->system_mtx));
603 else
604 return (sx_xlocked(&map->lock));
605 }
606
607 #ifdef INVARIANTS
608 static void
609 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
610 {
611
612 if (map->system_map)
613 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
614 else
615 sx_assert_(&map->lock, SA_XLOCKED, file, line);
616 }
617
618 #define VM_MAP_ASSERT_LOCKED(map) \
619 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
620 #else
621 #define VM_MAP_ASSERT_LOCKED(map)
622 #endif
623
624 /*
625 * _vm_map_unlock_and_wait:
626 *
627 * Atomically releases the lock on the specified map and puts the calling
628 * thread to sleep. The calling thread will remain asleep until either
629 * vm_map_wakeup() is performed on the map or the specified timeout is
630 * exceeded.
631 *
632 * WARNING! This function does not perform deferred deallocations of
633 * objects and map entries. Therefore, the calling thread is expected to
634 * reacquire the map lock after reawakening and later perform an ordinary
635 * unlock operation, such as vm_map_unlock(), before completing its
636 * operation on the map.
637 */
638 int
639 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
640 {
641
642 mtx_lock(&map_sleep_mtx);
643 if (map->system_map)
644 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
645 else
646 sx_xunlock_(&map->lock, file, line);
647 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
648 timo));
649 }
650
651 /*
652 * vm_map_wakeup:
653 *
654 * Awaken any threads that have slept on the map using
655 * vm_map_unlock_and_wait().
656 */
657 void
658 vm_map_wakeup(vm_map_t map)
659 {
660
661 /*
662 * Acquire and release map_sleep_mtx to prevent a wakeup()
663 * from being performed (and lost) between the map unlock
664 * and the msleep() in _vm_map_unlock_and_wait().
665 */
666 mtx_lock(&map_sleep_mtx);
667 mtx_unlock(&map_sleep_mtx);
668 wakeup(&map->root);
669 }
670
671 void
672 vm_map_busy(vm_map_t map)
673 {
674
675 VM_MAP_ASSERT_LOCKED(map);
676 map->busy++;
677 }
678
679 void
680 vm_map_unbusy(vm_map_t map)
681 {
682
683 VM_MAP_ASSERT_LOCKED(map);
684 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
685 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
686 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
687 wakeup(&map->busy);
688 }
689 }
690
691 void
692 vm_map_wait_busy(vm_map_t map)
693 {
694
695 VM_MAP_ASSERT_LOCKED(map);
696 while (map->busy) {
697 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
698 if (map->system_map)
699 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
700 else
701 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
702 }
703 map->timestamp++;
704 }
705
706 long
707 vmspace_resident_count(struct vmspace *vmspace)
708 {
709 return pmap_resident_count(vmspace_pmap(vmspace));
710 }
711
712 /*
713 * vm_map_create:
714 *
715 * Creates and returns a new empty VM map with
716 * the given physical map structure, and having
717 * the given lower and upper address bounds.
718 */
719 vm_map_t
720 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
721 {
722 vm_map_t result;
723
724 result = uma_zalloc(mapzone, M_WAITOK);
725 CTR1(KTR_VM, "vm_map_create: %p", result);
726 _vm_map_init(result, pmap, min, max);
727 return (result);
728 }
729
730 /*
731 * Initialize an existing vm_map structure
732 * such as that in the vmspace structure.
733 */
734 static void
735 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
736 {
737
738 map->header.next = map->header.prev = &map->header;
739 map->needs_wakeup = FALSE;
740 map->system_map = 0;
741 map->pmap = pmap;
742 map->min_offset = min;
743 map->max_offset = max;
744 map->flags = 0;
745 map->root = NULL;
746 map->timestamp = 0;
747 map->busy = 0;
748 }
749
750 void
751 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
752 {
753
754 _vm_map_init(map, pmap, min, max);
755 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
756 sx_init(&map->lock, "user map");
757 }
758
759 /*
760 * vm_map_entry_dispose: [ internal use only ]
761 *
762 * Inverse of vm_map_entry_create.
763 */
764 static void
765 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
766 {
767 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
768 }
769
770 /*
771 * vm_map_entry_create: [ internal use only ]
772 *
773 * Allocates a VM map entry for insertion.
774 * No entry fields are filled in.
775 */
776 static vm_map_entry_t
777 vm_map_entry_create(vm_map_t map)
778 {
779 vm_map_entry_t new_entry;
780
781 if (map->system_map)
782 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
783 else
784 new_entry = uma_zalloc(mapentzone, M_WAITOK);
785 if (new_entry == NULL)
786 panic("vm_map_entry_create: kernel resources exhausted");
787 return (new_entry);
788 }
789
790 /*
791 * vm_map_entry_set_behavior:
792 *
793 * Set the expected access behavior, either normal, random, or
794 * sequential.
795 */
796 static inline void
797 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
798 {
799 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
800 (behavior & MAP_ENTRY_BEHAV_MASK);
801 }
802
803 /*
804 * vm_map_entry_set_max_free:
805 *
806 * Set the max_free field in a vm_map_entry.
807 */
808 static inline void
809 vm_map_entry_set_max_free(vm_map_entry_t entry)
810 {
811
812 entry->max_free = entry->adj_free;
813 if (entry->left != NULL && entry->left->max_free > entry->max_free)
814 entry->max_free = entry->left->max_free;
815 if (entry->right != NULL && entry->right->max_free > entry->max_free)
816 entry->max_free = entry->right->max_free;
817 }
818
819 /*
820 * vm_map_entry_splay:
821 *
822 * The Sleator and Tarjan top-down splay algorithm with the
823 * following variation. Max_free must be computed bottom-up, so
824 * on the downward pass, maintain the left and right spines in
825 * reverse order. Then, make a second pass up each side to fix
826 * the pointers and compute max_free. The time bound is O(log n)
827 * amortized.
828 *
829 * The new root is the vm_map_entry containing "addr", or else an
830 * adjacent entry (lower or higher) if addr is not in the tree.
831 *
832 * The map must be locked, and leaves it so.
833 *
834 * Returns: the new root.
835 */
836 static vm_map_entry_t
837 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
838 {
839 vm_map_entry_t llist, rlist;
840 vm_map_entry_t ltree, rtree;
841 vm_map_entry_t y;
842
843 /* Special case of empty tree. */
844 if (root == NULL)
845 return (root);
846
847 /*
848 * Pass One: Splay down the tree until we find addr or a NULL
849 * pointer where addr would go. llist and rlist are the two
850 * sides in reverse order (bottom-up), with llist linked by
851 * the right pointer and rlist linked by the left pointer in
852 * the vm_map_entry. Wait until Pass Two to set max_free on
853 * the two spines.
854 */
855 llist = NULL;
856 rlist = NULL;
857 for (;;) {
858 /* root is never NULL in here. */
859 if (addr < root->start) {
860 y = root->left;
861 if (y == NULL)
862 break;
863 if (addr < y->start && y->left != NULL) {
864 /* Rotate right and put y on rlist. */
865 root->left = y->right;
866 y->right = root;
867 vm_map_entry_set_max_free(root);
868 root = y->left;
869 y->left = rlist;
870 rlist = y;
871 } else {
872 /* Put root on rlist. */
873 root->left = rlist;
874 rlist = root;
875 root = y;
876 }
877 } else if (addr >= root->end) {
878 y = root->right;
879 if (y == NULL)
880 break;
881 if (addr >= y->end && y->right != NULL) {
882 /* Rotate left and put y on llist. */
883 root->right = y->left;
884 y->left = root;
885 vm_map_entry_set_max_free(root);
886 root = y->right;
887 y->right = llist;
888 llist = y;
889 } else {
890 /* Put root on llist. */
891 root->right = llist;
892 llist = root;
893 root = y;
894 }
895 } else
896 break;
897 }
898
899 /*
900 * Pass Two: Walk back up the two spines, flip the pointers
901 * and set max_free. The subtrees of the root go at the
902 * bottom of llist and rlist.
903 */
904 ltree = root->left;
905 while (llist != NULL) {
906 y = llist->right;
907 llist->right = ltree;
908 vm_map_entry_set_max_free(llist);
909 ltree = llist;
910 llist = y;
911 }
912 rtree = root->right;
913 while (rlist != NULL) {
914 y = rlist->left;
915 rlist->left = rtree;
916 vm_map_entry_set_max_free(rlist);
917 rtree = rlist;
918 rlist = y;
919 }
920
921 /*
922 * Final assembly: add ltree and rtree as subtrees of root.
923 */
924 root->left = ltree;
925 root->right = rtree;
926 vm_map_entry_set_max_free(root);
927
928 return (root);
929 }
930
931 /*
932 * vm_map_entry_{un,}link:
933 *
934 * Insert/remove entries from maps.
935 */
936 static void
937 vm_map_entry_link(vm_map_t map,
938 vm_map_entry_t after_where,
939 vm_map_entry_t entry)
940 {
941
942 CTR4(KTR_VM,
943 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
944 map->nentries, entry, after_where);
945 VM_MAP_ASSERT_LOCKED(map);
946 KASSERT(after_where == &map->header ||
947 after_where->end <= entry->start,
948 ("vm_map_entry_link: prev end %jx new start %jx overlap",
949 (uintmax_t)after_where->end, (uintmax_t)entry->start));
950 KASSERT(after_where->next == &map->header ||
951 entry->end <= after_where->next->start,
952 ("vm_map_entry_link: new end %jx next start %jx overlap",
953 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
954
955 map->nentries++;
956 entry->prev = after_where;
957 entry->next = after_where->next;
958 entry->next->prev = entry;
959 after_where->next = entry;
960
961 if (after_where != &map->header) {
962 if (after_where != map->root)
963 vm_map_entry_splay(after_where->start, map->root);
964 entry->right = after_where->right;
965 entry->left = after_where;
966 after_where->right = NULL;
967 after_where->adj_free = entry->start - after_where->end;
968 vm_map_entry_set_max_free(after_where);
969 } else {
970 entry->right = map->root;
971 entry->left = NULL;
972 }
973 entry->adj_free = (entry->next == &map->header ? map->max_offset :
974 entry->next->start) - entry->end;
975 vm_map_entry_set_max_free(entry);
976 map->root = entry;
977 }
978
979 static void
980 vm_map_entry_unlink(vm_map_t map,
981 vm_map_entry_t entry)
982 {
983 vm_map_entry_t next, prev, root;
984
985 VM_MAP_ASSERT_LOCKED(map);
986 if (entry != map->root)
987 vm_map_entry_splay(entry->start, map->root);
988 if (entry->left == NULL)
989 root = entry->right;
990 else {
991 root = vm_map_entry_splay(entry->start, entry->left);
992 root->right = entry->right;
993 root->adj_free = (entry->next == &map->header ? map->max_offset :
994 entry->next->start) - root->end;
995 vm_map_entry_set_max_free(root);
996 }
997 map->root = root;
998
999 prev = entry->prev;
1000 next = entry->next;
1001 next->prev = prev;
1002 prev->next = next;
1003 map->nentries--;
1004 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1005 map->nentries, entry);
1006 }
1007
1008 /*
1009 * vm_map_entry_resize_free:
1010 *
1011 * Recompute the amount of free space following a vm_map_entry
1012 * and propagate that value up the tree. Call this function after
1013 * resizing a map entry in-place, that is, without a call to
1014 * vm_map_entry_link() or _unlink().
1015 *
1016 * The map must be locked, and leaves it so.
1017 */
1018 static void
1019 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1020 {
1021
1022 /*
1023 * Using splay trees without parent pointers, propagating
1024 * max_free up the tree is done by moving the entry to the
1025 * root and making the change there.
1026 */
1027 if (entry != map->root)
1028 map->root = vm_map_entry_splay(entry->start, map->root);
1029
1030 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1031 entry->next->start) - entry->end;
1032 vm_map_entry_set_max_free(entry);
1033 }
1034
1035 /*
1036 * vm_map_lookup_entry: [ internal use only ]
1037 *
1038 * Finds the map entry containing (or
1039 * immediately preceding) the specified address
1040 * in the given map; the entry is returned
1041 * in the "entry" parameter. The boolean
1042 * result indicates whether the address is
1043 * actually contained in the map.
1044 */
1045 boolean_t
1046 vm_map_lookup_entry(
1047 vm_map_t map,
1048 vm_offset_t address,
1049 vm_map_entry_t *entry) /* OUT */
1050 {
1051 vm_map_entry_t cur;
1052 boolean_t locked;
1053
1054 /*
1055 * If the map is empty, then the map entry immediately preceding
1056 * "address" is the map's header.
1057 */
1058 cur = map->root;
1059 if (cur == NULL)
1060 *entry = &map->header;
1061 else if (address >= cur->start && cur->end > address) {
1062 *entry = cur;
1063 return (TRUE);
1064 } else if ((locked = vm_map_locked(map)) ||
1065 sx_try_upgrade(&map->lock)) {
1066 /*
1067 * Splay requires a write lock on the map. However, it only
1068 * restructures the binary search tree; it does not otherwise
1069 * change the map. Thus, the map's timestamp need not change
1070 * on a temporary upgrade.
1071 */
1072 map->root = cur = vm_map_entry_splay(address, cur);
1073 if (!locked)
1074 sx_downgrade(&map->lock);
1075
1076 /*
1077 * If "address" is contained within a map entry, the new root
1078 * is that map entry. Otherwise, the new root is a map entry
1079 * immediately before or after "address".
1080 */
1081 if (address >= cur->start) {
1082 *entry = cur;
1083 if (cur->end > address)
1084 return (TRUE);
1085 } else
1086 *entry = cur->prev;
1087 } else
1088 /*
1089 * Since the map is only locked for read access, perform a
1090 * standard binary search tree lookup for "address".
1091 */
1092 for (;;) {
1093 if (address < cur->start) {
1094 if (cur->left == NULL) {
1095 *entry = cur->prev;
1096 break;
1097 }
1098 cur = cur->left;
1099 } else if (cur->end > address) {
1100 *entry = cur;
1101 return (TRUE);
1102 } else {
1103 if (cur->right == NULL) {
1104 *entry = cur;
1105 break;
1106 }
1107 cur = cur->right;
1108 }
1109 }
1110 return (FALSE);
1111 }
1112
1113 /*
1114 * vm_map_insert:
1115 *
1116 * Inserts the given whole VM object into the target
1117 * map at the specified address range. The object's
1118 * size should match that of the address range.
1119 *
1120 * Requires that the map be locked, and leaves it so.
1121 *
1122 * If object is non-NULL, ref count must be bumped by caller
1123 * prior to making call to account for the new entry.
1124 */
1125 int
1126 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1127 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1128 int cow)
1129 {
1130 vm_map_entry_t new_entry;
1131 vm_map_entry_t prev_entry;
1132 vm_map_entry_t temp_entry;
1133 vm_eflags_t protoeflags;
1134 struct ucred *cred;
1135 vm_inherit_t inheritance;
1136 boolean_t charge_prev_obj;
1137
1138 VM_MAP_ASSERT_LOCKED(map);
1139
1140 /*
1141 * Check that the start and end points are not bogus.
1142 */
1143 if ((start < map->min_offset) || (end > map->max_offset) ||
1144 (start >= end))
1145 return (KERN_INVALID_ADDRESS);
1146
1147 /*
1148 * Find the entry prior to the proposed starting address; if it's part
1149 * of an existing entry, this range is bogus.
1150 */
1151 if (vm_map_lookup_entry(map, start, &temp_entry))
1152 return (KERN_NO_SPACE);
1153
1154 prev_entry = temp_entry;
1155
1156 /*
1157 * Assert that the next entry doesn't overlap the end point.
1158 */
1159 if ((prev_entry->next != &map->header) &&
1160 (prev_entry->next->start < end))
1161 return (KERN_NO_SPACE);
1162
1163 protoeflags = 0;
1164 charge_prev_obj = FALSE;
1165
1166 if (cow & MAP_COPY_ON_WRITE)
1167 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1168
1169 if (cow & MAP_NOFAULT) {
1170 protoeflags |= MAP_ENTRY_NOFAULT;
1171
1172 KASSERT(object == NULL,
1173 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1174 }
1175 if (cow & MAP_DISABLE_SYNCER)
1176 protoeflags |= MAP_ENTRY_NOSYNC;
1177 if (cow & MAP_DISABLE_COREDUMP)
1178 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1179 if (cow & MAP_VN_WRITECOUNT)
1180 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1181 if (cow & MAP_INHERIT_SHARE)
1182 inheritance = VM_INHERIT_SHARE;
1183 else
1184 inheritance = VM_INHERIT_DEFAULT;
1185
1186 cred = NULL;
1187 KASSERT((object != kmem_object && object != kernel_object) ||
1188 ((object == kmem_object || object == kernel_object) &&
1189 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1190 ("kmem or kernel object and cow"));
1191 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1192 goto charged;
1193 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1194 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1195 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1196 return (KERN_RESOURCE_SHORTAGE);
1197 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1198 object->cred == NULL,
1199 ("OVERCOMMIT: vm_map_insert o %p", object));
1200 cred = curthread->td_ucred;
1201 crhold(cred);
1202 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1203 charge_prev_obj = TRUE;
1204 }
1205
1206 charged:
1207 /* Expand the kernel pmap, if necessary. */
1208 if (map == kernel_map && end > kernel_vm_end)
1209 pmap_growkernel(end);
1210 if (object != NULL) {
1211 /*
1212 * OBJ_ONEMAPPING must be cleared unless this mapping
1213 * is trivially proven to be the only mapping for any
1214 * of the object's pages. (Object granularity
1215 * reference counting is insufficient to recognize
1216 * aliases with precision.)
1217 */
1218 VM_OBJECT_WLOCK(object);
1219 if (object->ref_count > 1 || object->shadow_count != 0)
1220 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1221 VM_OBJECT_WUNLOCK(object);
1222 }
1223 else if ((prev_entry != &map->header) &&
1224 (prev_entry->eflags == protoeflags) &&
1225 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1226 (prev_entry->end == start) &&
1227 (prev_entry->wired_count == 0) &&
1228 (prev_entry->cred == cred ||
1229 (prev_entry->object.vm_object != NULL &&
1230 (prev_entry->object.vm_object->cred == cred))) &&
1231 vm_object_coalesce(prev_entry->object.vm_object,
1232 prev_entry->offset,
1233 (vm_size_t)(prev_entry->end - prev_entry->start),
1234 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1235 /*
1236 * We were able to extend the object. Determine if we
1237 * can extend the previous map entry to include the
1238 * new range as well.
1239 */
1240 if ((prev_entry->inheritance == inheritance) &&
1241 (prev_entry->protection == prot) &&
1242 (prev_entry->max_protection == max)) {
1243 map->size += (end - prev_entry->end);
1244 prev_entry->end = end;
1245 vm_map_entry_resize_free(map, prev_entry);
1246 vm_map_simplify_entry(map, prev_entry);
1247 if (cred != NULL)
1248 crfree(cred);
1249 return (KERN_SUCCESS);
1250 }
1251
1252 /*
1253 * If we can extend the object but cannot extend the
1254 * map entry, we have to create a new map entry. We
1255 * must bump the ref count on the extended object to
1256 * account for it. object may be NULL.
1257 */
1258 object = prev_entry->object.vm_object;
1259 offset = prev_entry->offset +
1260 (prev_entry->end - prev_entry->start);
1261 vm_object_reference(object);
1262 if (cred != NULL && object != NULL && object->cred != NULL &&
1263 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1264 /* Object already accounts for this uid. */
1265 crfree(cred);
1266 cred = NULL;
1267 }
1268 }
1269
1270 /*
1271 * NOTE: if conditionals fail, object can be NULL here. This occurs
1272 * in things like the buffer map where we manage kva but do not manage
1273 * backing objects.
1274 */
1275
1276 /*
1277 * Create a new entry
1278 */
1279 new_entry = vm_map_entry_create(map);
1280 new_entry->start = start;
1281 new_entry->end = end;
1282 new_entry->cred = NULL;
1283
1284 new_entry->eflags = protoeflags;
1285 new_entry->object.vm_object = object;
1286 new_entry->offset = offset;
1287 new_entry->avail_ssize = 0;
1288
1289 new_entry->inheritance = inheritance;
1290 new_entry->protection = prot;
1291 new_entry->max_protection = max;
1292 new_entry->wired_count = 0;
1293 new_entry->wiring_thread = NULL;
1294 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1295 new_entry->next_read = OFF_TO_IDX(offset);
1296
1297 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1298 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1299 new_entry->cred = cred;
1300
1301 /*
1302 * Insert the new entry into the list
1303 */
1304 vm_map_entry_link(map, prev_entry, new_entry);
1305 map->size += new_entry->end - new_entry->start;
1306
1307 /*
1308 * It may be possible to merge the new entry with the next and/or
1309 * previous entries. However, due to MAP_STACK_* being a hack, a
1310 * panic can result from merging such entries.
1311 */
1312 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)
1313 vm_map_simplify_entry(map, new_entry);
1314
1315 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1316 vm_map_pmap_enter(map, start, prot,
1317 object, OFF_TO_IDX(offset), end - start,
1318 cow & MAP_PREFAULT_PARTIAL);
1319 }
1320
1321 return (KERN_SUCCESS);
1322 }
1323
1324 /*
1325 * vm_map_findspace:
1326 *
1327 * Find the first fit (lowest VM address) for "length" free bytes
1328 * beginning at address >= start in the given map.
1329 *
1330 * In a vm_map_entry, "adj_free" is the amount of free space
1331 * adjacent (higher address) to this entry, and "max_free" is the
1332 * maximum amount of contiguous free space in its subtree. This
1333 * allows finding a free region in one path down the tree, so
1334 * O(log n) amortized with splay trees.
1335 *
1336 * The map must be locked, and leaves it so.
1337 *
1338 * Returns: 0 on success, and starting address in *addr,
1339 * 1 if insufficient space.
1340 */
1341 int
1342 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1343 vm_offset_t *addr) /* OUT */
1344 {
1345 vm_map_entry_t entry;
1346 vm_offset_t st;
1347
1348 /*
1349 * Request must fit within min/max VM address and must avoid
1350 * address wrap.
1351 */
1352 if (start < map->min_offset)
1353 start = map->min_offset;
1354 if (start + length > map->max_offset || start + length < start)
1355 return (1);
1356
1357 /* Empty tree means wide open address space. */
1358 if (map->root == NULL) {
1359 *addr = start;
1360 return (0);
1361 }
1362
1363 /*
1364 * After splay, if start comes before root node, then there
1365 * must be a gap from start to the root.
1366 */
1367 map->root = vm_map_entry_splay(start, map->root);
1368 if (start + length <= map->root->start) {
1369 *addr = start;
1370 return (0);
1371 }
1372
1373 /*
1374 * Root is the last node that might begin its gap before
1375 * start, and this is the last comparison where address
1376 * wrap might be a problem.
1377 */
1378 st = (start > map->root->end) ? start : map->root->end;
1379 if (length <= map->root->end + map->root->adj_free - st) {
1380 *addr = st;
1381 return (0);
1382 }
1383
1384 /* With max_free, can immediately tell if no solution. */
1385 entry = map->root->right;
1386 if (entry == NULL || length > entry->max_free)
1387 return (1);
1388
1389 /*
1390 * Search the right subtree in the order: left subtree, root,
1391 * right subtree (first fit). The previous splay implies that
1392 * all regions in the right subtree have addresses > start.
1393 */
1394 while (entry != NULL) {
1395 if (entry->left != NULL && entry->left->max_free >= length)
1396 entry = entry->left;
1397 else if (entry->adj_free >= length) {
1398 *addr = entry->end;
1399 return (0);
1400 } else
1401 entry = entry->right;
1402 }
1403
1404 /* Can't get here, so panic if we do. */
1405 panic("vm_map_findspace: max_free corrupt");
1406 }
1407
1408 int
1409 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1410 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1411 vm_prot_t max, int cow)
1412 {
1413 vm_offset_t end;
1414 int result;
1415
1416 end = start + length;
1417 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1418 object == NULL,
1419 ("vm_map_fixed: non-NULL backing object for stack"));
1420 vm_map_lock(map);
1421 VM_MAP_RANGE_CHECK(map, start, end);
1422 if ((cow & MAP_CHECK_EXCL) == 0)
1423 vm_map_delete(map, start, end);
1424 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1425 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1426 prot, max, cow);
1427 } else {
1428 result = vm_map_insert(map, object, offset, start, end,
1429 prot, max, cow);
1430 }
1431 vm_map_unlock(map);
1432 return (result);
1433 }
1434
1435 /*
1436 * vm_map_find finds an unallocated region in the target address
1437 * map with the given length. The search is defined to be
1438 * first-fit from the specified address; the region found is
1439 * returned in the same parameter.
1440 *
1441 * If object is non-NULL, ref count must be bumped by caller
1442 * prior to making call to account for the new entry.
1443 */
1444 int
1445 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1446 vm_offset_t *addr, /* IN/OUT */
1447 vm_size_t length, vm_offset_t max_addr, int find_space,
1448 vm_prot_t prot, vm_prot_t max, int cow)
1449 {
1450 vm_offset_t alignment, initial_addr, start;
1451 int result;
1452
1453 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1454 object == NULL,
1455 ("vm_map_find: non-NULL backing object for stack"));
1456 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1457 (object->flags & OBJ_COLORED) == 0))
1458 find_space = VMFS_ANY_SPACE;
1459 if (find_space >> 8 != 0) {
1460 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1461 alignment = (vm_offset_t)1 << (find_space >> 8);
1462 } else
1463 alignment = 0;
1464 initial_addr = *addr;
1465 again:
1466 start = initial_addr;
1467 vm_map_lock(map);
1468 do {
1469 if (find_space != VMFS_NO_SPACE) {
1470 if (vm_map_findspace(map, start, length, addr) ||
1471 (max_addr != 0 && *addr + length > max_addr)) {
1472 vm_map_unlock(map);
1473 if (find_space == VMFS_OPTIMAL_SPACE) {
1474 find_space = VMFS_ANY_SPACE;
1475 goto again;
1476 }
1477 return (KERN_NO_SPACE);
1478 }
1479 switch (find_space) {
1480 case VMFS_SUPER_SPACE:
1481 case VMFS_OPTIMAL_SPACE:
1482 pmap_align_superpage(object, offset, addr,
1483 length);
1484 break;
1485 case VMFS_ANY_SPACE:
1486 break;
1487 default:
1488 if ((*addr & (alignment - 1)) != 0) {
1489 *addr &= ~(alignment - 1);
1490 *addr += alignment;
1491 }
1492 break;
1493 }
1494
1495 start = *addr;
1496 }
1497 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1498 result = vm_map_stack_locked(map, start, length,
1499 sgrowsiz, prot, max, cow);
1500 } else {
1501 result = vm_map_insert(map, object, offset, start,
1502 start + length, prot, max, cow);
1503 }
1504 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1505 find_space != VMFS_ANY_SPACE);
1506 vm_map_unlock(map);
1507 return (result);
1508 }
1509
1510 /*
1511 * vm_map_simplify_entry:
1512 *
1513 * Simplify the given map entry by merging with either neighbor. This
1514 * routine also has the ability to merge with both neighbors.
1515 *
1516 * The map must be locked.
1517 *
1518 * This routine guarentees that the passed entry remains valid (though
1519 * possibly extended). When merging, this routine may delete one or
1520 * both neighbors.
1521 */
1522 void
1523 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1524 {
1525 vm_map_entry_t next, prev;
1526 vm_size_t prevsize, esize;
1527
1528 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1529 return;
1530
1531 prev = entry->prev;
1532 if (prev != &map->header) {
1533 prevsize = prev->end - prev->start;
1534 if ( (prev->end == entry->start) &&
1535 (prev->object.vm_object == entry->object.vm_object) &&
1536 (!prev->object.vm_object ||
1537 (prev->offset + prevsize == entry->offset)) &&
1538 (prev->eflags == entry->eflags) &&
1539 (prev->protection == entry->protection) &&
1540 (prev->max_protection == entry->max_protection) &&
1541 (prev->inheritance == entry->inheritance) &&
1542 (prev->wired_count == entry->wired_count) &&
1543 (prev->cred == entry->cred)) {
1544 vm_map_entry_unlink(map, prev);
1545 entry->start = prev->start;
1546 entry->offset = prev->offset;
1547 if (entry->prev != &map->header)
1548 vm_map_entry_resize_free(map, entry->prev);
1549
1550 /*
1551 * If the backing object is a vnode object,
1552 * vm_object_deallocate() calls vrele().
1553 * However, vrele() does not lock the vnode
1554 * because the vnode has additional
1555 * references. Thus, the map lock can be kept
1556 * without causing a lock-order reversal with
1557 * the vnode lock.
1558 *
1559 * Since we count the number of virtual page
1560 * mappings in object->un_pager.vnp.writemappings,
1561 * the writemappings value should not be adjusted
1562 * when the entry is disposed of.
1563 */
1564 if (prev->object.vm_object)
1565 vm_object_deallocate(prev->object.vm_object);
1566 if (prev->cred != NULL)
1567 crfree(prev->cred);
1568 vm_map_entry_dispose(map, prev);
1569 }
1570 }
1571
1572 next = entry->next;
1573 if (next != &map->header) {
1574 esize = entry->end - entry->start;
1575 if ((entry->end == next->start) &&
1576 (next->object.vm_object == entry->object.vm_object) &&
1577 (!entry->object.vm_object ||
1578 (entry->offset + esize == next->offset)) &&
1579 (next->eflags == entry->eflags) &&
1580 (next->protection == entry->protection) &&
1581 (next->max_protection == entry->max_protection) &&
1582 (next->inheritance == entry->inheritance) &&
1583 (next->wired_count == entry->wired_count) &&
1584 (next->cred == entry->cred)) {
1585 vm_map_entry_unlink(map, next);
1586 entry->end = next->end;
1587 vm_map_entry_resize_free(map, entry);
1588
1589 /*
1590 * See comment above.
1591 */
1592 if (next->object.vm_object)
1593 vm_object_deallocate(next->object.vm_object);
1594 if (next->cred != NULL)
1595 crfree(next->cred);
1596 vm_map_entry_dispose(map, next);
1597 }
1598 }
1599 }
1600 /*
1601 * vm_map_clip_start: [ internal use only ]
1602 *
1603 * Asserts that the given entry begins at or after
1604 * the specified address; if necessary,
1605 * it splits the entry into two.
1606 */
1607 #define vm_map_clip_start(map, entry, startaddr) \
1608 { \
1609 if (startaddr > entry->start) \
1610 _vm_map_clip_start(map, entry, startaddr); \
1611 }
1612
1613 /*
1614 * This routine is called only when it is known that
1615 * the entry must be split.
1616 */
1617 static void
1618 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1619 {
1620 vm_map_entry_t new_entry;
1621
1622 VM_MAP_ASSERT_LOCKED(map);
1623
1624 /*
1625 * Split off the front portion -- note that we must insert the new
1626 * entry BEFORE this one, so that this entry has the specified
1627 * starting address.
1628 */
1629 vm_map_simplify_entry(map, entry);
1630
1631 /*
1632 * If there is no object backing this entry, we might as well create
1633 * one now. If we defer it, an object can get created after the map
1634 * is clipped, and individual objects will be created for the split-up
1635 * map. This is a bit of a hack, but is also about the best place to
1636 * put this improvement.
1637 */
1638 if (entry->object.vm_object == NULL && !map->system_map) {
1639 vm_object_t object;
1640 object = vm_object_allocate(OBJT_DEFAULT,
1641 atop(entry->end - entry->start));
1642 entry->object.vm_object = object;
1643 entry->offset = 0;
1644 if (entry->cred != NULL) {
1645 object->cred = entry->cred;
1646 object->charge = entry->end - entry->start;
1647 entry->cred = NULL;
1648 }
1649 } else if (entry->object.vm_object != NULL &&
1650 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1651 entry->cred != NULL) {
1652 VM_OBJECT_WLOCK(entry->object.vm_object);
1653 KASSERT(entry->object.vm_object->cred == NULL,
1654 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1655 entry->object.vm_object->cred = entry->cred;
1656 entry->object.vm_object->charge = entry->end - entry->start;
1657 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1658 entry->cred = NULL;
1659 }
1660
1661 new_entry = vm_map_entry_create(map);
1662 *new_entry = *entry;
1663
1664 new_entry->end = start;
1665 entry->offset += (start - entry->start);
1666 entry->start = start;
1667 if (new_entry->cred != NULL)
1668 crhold(entry->cred);
1669
1670 vm_map_entry_link(map, entry->prev, new_entry);
1671
1672 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1673 vm_object_reference(new_entry->object.vm_object);
1674 /*
1675 * The object->un_pager.vnp.writemappings for the
1676 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1677 * kept as is here. The virtual pages are
1678 * re-distributed among the clipped entries, so the sum is
1679 * left the same.
1680 */
1681 }
1682 }
1683
1684 /*
1685 * vm_map_clip_end: [ internal use only ]
1686 *
1687 * Asserts that the given entry ends at or before
1688 * the specified address; if necessary,
1689 * it splits the entry into two.
1690 */
1691 #define vm_map_clip_end(map, entry, endaddr) \
1692 { \
1693 if ((endaddr) < (entry->end)) \
1694 _vm_map_clip_end((map), (entry), (endaddr)); \
1695 }
1696
1697 /*
1698 * This routine is called only when it is known that
1699 * the entry must be split.
1700 */
1701 static void
1702 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1703 {
1704 vm_map_entry_t new_entry;
1705
1706 VM_MAP_ASSERT_LOCKED(map);
1707
1708 /*
1709 * If there is no object backing this entry, we might as well create
1710 * one now. If we defer it, an object can get created after the map
1711 * is clipped, and individual objects will be created for the split-up
1712 * map. This is a bit of a hack, but is also about the best place to
1713 * put this improvement.
1714 */
1715 if (entry->object.vm_object == NULL && !map->system_map) {
1716 vm_object_t object;
1717 object = vm_object_allocate(OBJT_DEFAULT,
1718 atop(entry->end - entry->start));
1719 entry->object.vm_object = object;
1720 entry->offset = 0;
1721 if (entry->cred != NULL) {
1722 object->cred = entry->cred;
1723 object->charge = entry->end - entry->start;
1724 entry->cred = NULL;
1725 }
1726 } else if (entry->object.vm_object != NULL &&
1727 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1728 entry->cred != NULL) {
1729 VM_OBJECT_WLOCK(entry->object.vm_object);
1730 KASSERT(entry->object.vm_object->cred == NULL,
1731 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1732 entry->object.vm_object->cred = entry->cred;
1733 entry->object.vm_object->charge = entry->end - entry->start;
1734 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1735 entry->cred = NULL;
1736 }
1737
1738 /*
1739 * Create a new entry and insert it AFTER the specified entry
1740 */
1741 new_entry = vm_map_entry_create(map);
1742 *new_entry = *entry;
1743
1744 new_entry->start = entry->end = end;
1745 new_entry->offset += (end - entry->start);
1746 if (new_entry->cred != NULL)
1747 crhold(entry->cred);
1748
1749 vm_map_entry_link(map, entry, new_entry);
1750
1751 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1752 vm_object_reference(new_entry->object.vm_object);
1753 }
1754 }
1755
1756 /*
1757 * vm_map_submap: [ kernel use only ]
1758 *
1759 * Mark the given range as handled by a subordinate map.
1760 *
1761 * This range must have been created with vm_map_find,
1762 * and no other operations may have been performed on this
1763 * range prior to calling vm_map_submap.
1764 *
1765 * Only a limited number of operations can be performed
1766 * within this rage after calling vm_map_submap:
1767 * vm_fault
1768 * [Don't try vm_map_copy!]
1769 *
1770 * To remove a submapping, one must first remove the
1771 * range from the superior map, and then destroy the
1772 * submap (if desired). [Better yet, don't try it.]
1773 */
1774 int
1775 vm_map_submap(
1776 vm_map_t map,
1777 vm_offset_t start,
1778 vm_offset_t end,
1779 vm_map_t submap)
1780 {
1781 vm_map_entry_t entry;
1782 int result = KERN_INVALID_ARGUMENT;
1783
1784 vm_map_lock(map);
1785
1786 VM_MAP_RANGE_CHECK(map, start, end);
1787
1788 if (vm_map_lookup_entry(map, start, &entry)) {
1789 vm_map_clip_start(map, entry, start);
1790 } else
1791 entry = entry->next;
1792
1793 vm_map_clip_end(map, entry, end);
1794
1795 if ((entry->start == start) && (entry->end == end) &&
1796 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1797 (entry->object.vm_object == NULL)) {
1798 entry->object.sub_map = submap;
1799 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1800 result = KERN_SUCCESS;
1801 }
1802 vm_map_unlock(map);
1803
1804 return (result);
1805 }
1806
1807 /*
1808 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1809 */
1810 #define MAX_INIT_PT 96
1811
1812 /*
1813 * vm_map_pmap_enter:
1814 *
1815 * Preload the specified map's pmap with mappings to the specified
1816 * object's memory-resident pages. No further physical pages are
1817 * allocated, and no further virtual pages are retrieved from secondary
1818 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1819 * limited number of page mappings are created at the low-end of the
1820 * specified address range. (For this purpose, a superpage mapping
1821 * counts as one page mapping.) Otherwise, all resident pages within
1822 * the specified address range are mapped. Because these mappings are
1823 * being created speculatively, cached pages are not reactivated and
1824 * mapped.
1825 */
1826 void
1827 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1828 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1829 {
1830 vm_offset_t start;
1831 vm_page_t p, p_start;
1832 vm_pindex_t mask, psize, threshold, tmpidx;
1833
1834 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1835 return;
1836 VM_OBJECT_RLOCK(object);
1837 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1838 VM_OBJECT_RUNLOCK(object);
1839 VM_OBJECT_WLOCK(object);
1840 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1841 pmap_object_init_pt(map->pmap, addr, object, pindex,
1842 size);
1843 VM_OBJECT_WUNLOCK(object);
1844 return;
1845 }
1846 VM_OBJECT_LOCK_DOWNGRADE(object);
1847 }
1848
1849 psize = atop(size);
1850 if (psize + pindex > object->size) {
1851 if (object->size < pindex) {
1852 VM_OBJECT_RUNLOCK(object);
1853 return;
1854 }
1855 psize = object->size - pindex;
1856 }
1857
1858 start = 0;
1859 p_start = NULL;
1860 threshold = MAX_INIT_PT;
1861
1862 p = vm_page_find_least(object, pindex);
1863 /*
1864 * Assert: the variable p is either (1) the page with the
1865 * least pindex greater than or equal to the parameter pindex
1866 * or (2) NULL.
1867 */
1868 for (;
1869 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1870 p = TAILQ_NEXT(p, listq)) {
1871 /*
1872 * don't allow an madvise to blow away our really
1873 * free pages allocating pv entries.
1874 */
1875 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1876 cnt.v_free_count < cnt.v_free_reserved) ||
1877 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1878 tmpidx >= threshold)) {
1879 psize = tmpidx;
1880 break;
1881 }
1882 if (p->valid == VM_PAGE_BITS_ALL) {
1883 if (p_start == NULL) {
1884 start = addr + ptoa(tmpidx);
1885 p_start = p;
1886 }
1887 /* Jump ahead if a superpage mapping is possible. */
1888 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1889 (pagesizes[p->psind] - 1)) == 0) {
1890 mask = atop(pagesizes[p->psind]) - 1;
1891 if (tmpidx + mask < psize &&
1892 vm_page_ps_is_valid(p)) {
1893 p += mask;
1894 threshold += mask;
1895 }
1896 }
1897 } else if (p_start != NULL) {
1898 pmap_enter_object(map->pmap, start, addr +
1899 ptoa(tmpidx), p_start, prot);
1900 p_start = NULL;
1901 }
1902 }
1903 if (p_start != NULL)
1904 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1905 p_start, prot);
1906 VM_OBJECT_RUNLOCK(object);
1907 }
1908
1909 /*
1910 * vm_map_protect:
1911 *
1912 * Sets the protection of the specified address
1913 * region in the target map. If "set_max" is
1914 * specified, the maximum protection is to be set;
1915 * otherwise, only the current protection is affected.
1916 */
1917 int
1918 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1919 vm_prot_t new_prot, boolean_t set_max)
1920 {
1921 vm_map_entry_t current, entry;
1922 vm_object_t obj;
1923 struct ucred *cred;
1924 vm_prot_t old_prot;
1925
1926 if (start == end)
1927 return (KERN_SUCCESS);
1928
1929 vm_map_lock(map);
1930
1931 VM_MAP_RANGE_CHECK(map, start, end);
1932
1933 if (vm_map_lookup_entry(map, start, &entry)) {
1934 vm_map_clip_start(map, entry, start);
1935 } else {
1936 entry = entry->next;
1937 }
1938
1939 /*
1940 * Make a first pass to check for protection violations.
1941 */
1942 current = entry;
1943 while ((current != &map->header) && (current->start < end)) {
1944 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1945 vm_map_unlock(map);
1946 return (KERN_INVALID_ARGUMENT);
1947 }
1948 if ((new_prot & current->max_protection) != new_prot) {
1949 vm_map_unlock(map);
1950 return (KERN_PROTECTION_FAILURE);
1951 }
1952 current = current->next;
1953 }
1954
1955
1956 /*
1957 * Do an accounting pass for private read-only mappings that
1958 * now will do cow due to allowed write (e.g. debugger sets
1959 * breakpoint on text segment)
1960 */
1961 for (current = entry; (current != &map->header) &&
1962 (current->start < end); current = current->next) {
1963
1964 vm_map_clip_end(map, current, end);
1965
1966 if (set_max ||
1967 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1968 ENTRY_CHARGED(current)) {
1969 continue;
1970 }
1971
1972 cred = curthread->td_ucred;
1973 obj = current->object.vm_object;
1974
1975 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1976 if (!swap_reserve(current->end - current->start)) {
1977 vm_map_unlock(map);
1978 return (KERN_RESOURCE_SHORTAGE);
1979 }
1980 crhold(cred);
1981 current->cred = cred;
1982 continue;
1983 }
1984
1985 VM_OBJECT_WLOCK(obj);
1986 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1987 VM_OBJECT_WUNLOCK(obj);
1988 continue;
1989 }
1990
1991 /*
1992 * Charge for the whole object allocation now, since
1993 * we cannot distinguish between non-charged and
1994 * charged clipped mapping of the same object later.
1995 */
1996 KASSERT(obj->charge == 0,
1997 ("vm_map_protect: object %p overcharged (entry %p)",
1998 obj, current));
1999 if (!swap_reserve(ptoa(obj->size))) {
2000 VM_OBJECT_WUNLOCK(obj);
2001 vm_map_unlock(map);
2002 return (KERN_RESOURCE_SHORTAGE);
2003 }
2004
2005 crhold(cred);
2006 obj->cred = cred;
2007 obj->charge = ptoa(obj->size);
2008 VM_OBJECT_WUNLOCK(obj);
2009 }
2010
2011 /*
2012 * Go back and fix up protections. [Note that clipping is not
2013 * necessary the second time.]
2014 */
2015 current = entry;
2016 while ((current != &map->header) && (current->start < end)) {
2017 old_prot = current->protection;
2018
2019 if (set_max)
2020 current->protection =
2021 (current->max_protection = new_prot) &
2022 old_prot;
2023 else
2024 current->protection = new_prot;
2025
2026 /*
2027 * For user wired map entries, the normal lazy evaluation of
2028 * write access upgrades through soft page faults is
2029 * undesirable. Instead, immediately copy any pages that are
2030 * copy-on-write and enable write access in the physical map.
2031 */
2032 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2033 (current->protection & VM_PROT_WRITE) != 0 &&
2034 (old_prot & VM_PROT_WRITE) == 0)
2035 vm_fault_copy_entry(map, map, current, current, NULL);
2036
2037 /*
2038 * When restricting access, update the physical map. Worry
2039 * about copy-on-write here.
2040 */
2041 if ((old_prot & ~current->protection) != 0) {
2042 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2043 VM_PROT_ALL)
2044 pmap_protect(map->pmap, current->start,
2045 current->end,
2046 current->protection & MASK(current));
2047 #undef MASK
2048 }
2049 vm_map_simplify_entry(map, current);
2050 current = current->next;
2051 }
2052 vm_map_unlock(map);
2053 return (KERN_SUCCESS);
2054 }
2055
2056 /*
2057 * vm_map_madvise:
2058 *
2059 * This routine traverses a processes map handling the madvise
2060 * system call. Advisories are classified as either those effecting
2061 * the vm_map_entry structure, or those effecting the underlying
2062 * objects.
2063 */
2064 int
2065 vm_map_madvise(
2066 vm_map_t map,
2067 vm_offset_t start,
2068 vm_offset_t end,
2069 int behav)
2070 {
2071 vm_map_entry_t current, entry;
2072 int modify_map = 0;
2073
2074 /*
2075 * Some madvise calls directly modify the vm_map_entry, in which case
2076 * we need to use an exclusive lock on the map and we need to perform
2077 * various clipping operations. Otherwise we only need a read-lock
2078 * on the map.
2079 */
2080 switch(behav) {
2081 case MADV_NORMAL:
2082 case MADV_SEQUENTIAL:
2083 case MADV_RANDOM:
2084 case MADV_NOSYNC:
2085 case MADV_AUTOSYNC:
2086 case MADV_NOCORE:
2087 case MADV_CORE:
2088 if (start == end)
2089 return (KERN_SUCCESS);
2090 modify_map = 1;
2091 vm_map_lock(map);
2092 break;
2093 case MADV_WILLNEED:
2094 case MADV_DONTNEED:
2095 case MADV_FREE:
2096 if (start == end)
2097 return (KERN_SUCCESS);
2098 vm_map_lock_read(map);
2099 break;
2100 default:
2101 return (KERN_INVALID_ARGUMENT);
2102 }
2103
2104 /*
2105 * Locate starting entry and clip if necessary.
2106 */
2107 VM_MAP_RANGE_CHECK(map, start, end);
2108
2109 if (vm_map_lookup_entry(map, start, &entry)) {
2110 if (modify_map)
2111 vm_map_clip_start(map, entry, start);
2112 } else {
2113 entry = entry->next;
2114 }
2115
2116 if (modify_map) {
2117 /*
2118 * madvise behaviors that are implemented in the vm_map_entry.
2119 *
2120 * We clip the vm_map_entry so that behavioral changes are
2121 * limited to the specified address range.
2122 */
2123 for (current = entry;
2124 (current != &map->header) && (current->start < end);
2125 current = current->next
2126 ) {
2127 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2128 continue;
2129
2130 vm_map_clip_end(map, current, end);
2131
2132 switch (behav) {
2133 case MADV_NORMAL:
2134 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2135 break;
2136 case MADV_SEQUENTIAL:
2137 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2138 break;
2139 case MADV_RANDOM:
2140 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2141 break;
2142 case MADV_NOSYNC:
2143 current->eflags |= MAP_ENTRY_NOSYNC;
2144 break;
2145 case MADV_AUTOSYNC:
2146 current->eflags &= ~MAP_ENTRY_NOSYNC;
2147 break;
2148 case MADV_NOCORE:
2149 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2150 break;
2151 case MADV_CORE:
2152 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2153 break;
2154 default:
2155 break;
2156 }
2157 vm_map_simplify_entry(map, current);
2158 }
2159 vm_map_unlock(map);
2160 } else {
2161 vm_pindex_t pstart, pend;
2162
2163 /*
2164 * madvise behaviors that are implemented in the underlying
2165 * vm_object.
2166 *
2167 * Since we don't clip the vm_map_entry, we have to clip
2168 * the vm_object pindex and count.
2169 */
2170 for (current = entry;
2171 (current != &map->header) && (current->start < end);
2172 current = current->next
2173 ) {
2174 vm_offset_t useEnd, useStart;
2175
2176 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2177 continue;
2178
2179 pstart = OFF_TO_IDX(current->offset);
2180 pend = pstart + atop(current->end - current->start);
2181 useStart = current->start;
2182 useEnd = current->end;
2183
2184 if (current->start < start) {
2185 pstart += atop(start - current->start);
2186 useStart = start;
2187 }
2188 if (current->end > end) {
2189 pend -= atop(current->end - end);
2190 useEnd = end;
2191 }
2192
2193 if (pstart >= pend)
2194 continue;
2195
2196 /*
2197 * Perform the pmap_advise() before clearing
2198 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2199 * concurrent pmap operation, such as pmap_remove(),
2200 * could clear a reference in the pmap and set
2201 * PGA_REFERENCED on the page before the pmap_advise()
2202 * had completed. Consequently, the page would appear
2203 * referenced based upon an old reference that
2204 * occurred before this pmap_advise() ran.
2205 */
2206 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2207 pmap_advise(map->pmap, useStart, useEnd,
2208 behav);
2209
2210 vm_object_madvise(current->object.vm_object, pstart,
2211 pend, behav);
2212
2213 /*
2214 * Pre-populate paging structures in the
2215 * WILLNEED case. For wired entries, the
2216 * paging structures are already populated.
2217 */
2218 if (behav == MADV_WILLNEED &&
2219 current->wired_count == 0) {
2220 vm_map_pmap_enter(map,
2221 useStart,
2222 current->protection,
2223 current->object.vm_object,
2224 pstart,
2225 ptoa(pend - pstart),
2226 MAP_PREFAULT_MADVISE
2227 );
2228 }
2229 }
2230 vm_map_unlock_read(map);
2231 }
2232 return (0);
2233 }
2234
2235
2236 /*
2237 * vm_map_inherit:
2238 *
2239 * Sets the inheritance of the specified address
2240 * range in the target map. Inheritance
2241 * affects how the map will be shared with
2242 * child maps at the time of vmspace_fork.
2243 */
2244 int
2245 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2246 vm_inherit_t new_inheritance)
2247 {
2248 vm_map_entry_t entry;
2249 vm_map_entry_t temp_entry;
2250
2251 switch (new_inheritance) {
2252 case VM_INHERIT_NONE:
2253 case VM_INHERIT_COPY:
2254 case VM_INHERIT_SHARE:
2255 break;
2256 default:
2257 return (KERN_INVALID_ARGUMENT);
2258 }
2259 if (start == end)
2260 return (KERN_SUCCESS);
2261 vm_map_lock(map);
2262 VM_MAP_RANGE_CHECK(map, start, end);
2263 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2264 entry = temp_entry;
2265 vm_map_clip_start(map, entry, start);
2266 } else
2267 entry = temp_entry->next;
2268 while ((entry != &map->header) && (entry->start < end)) {
2269 vm_map_clip_end(map, entry, end);
2270 entry->inheritance = new_inheritance;
2271 vm_map_simplify_entry(map, entry);
2272 entry = entry->next;
2273 }
2274 vm_map_unlock(map);
2275 return (KERN_SUCCESS);
2276 }
2277
2278 /*
2279 * vm_map_unwire:
2280 *
2281 * Implements both kernel and user unwiring.
2282 */
2283 int
2284 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2285 int flags)
2286 {
2287 vm_map_entry_t entry, first_entry, tmp_entry;
2288 vm_offset_t saved_start;
2289 unsigned int last_timestamp;
2290 int rv;
2291 boolean_t need_wakeup, result, user_unwire;
2292
2293 if (start == end)
2294 return (KERN_SUCCESS);
2295 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2296 vm_map_lock(map);
2297 VM_MAP_RANGE_CHECK(map, start, end);
2298 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2299 if (flags & VM_MAP_WIRE_HOLESOK)
2300 first_entry = first_entry->next;
2301 else {
2302 vm_map_unlock(map);
2303 return (KERN_INVALID_ADDRESS);
2304 }
2305 }
2306 last_timestamp = map->timestamp;
2307 entry = first_entry;
2308 while (entry != &map->header && entry->start < end) {
2309 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2310 /*
2311 * We have not yet clipped the entry.
2312 */
2313 saved_start = (start >= entry->start) ? start :
2314 entry->start;
2315 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2316 if (vm_map_unlock_and_wait(map, 0)) {
2317 /*
2318 * Allow interruption of user unwiring?
2319 */
2320 }
2321 vm_map_lock(map);
2322 if (last_timestamp+1 != map->timestamp) {
2323 /*
2324 * Look again for the entry because the map was
2325 * modified while it was unlocked.
2326 * Specifically, the entry may have been
2327 * clipped, merged, or deleted.
2328 */
2329 if (!vm_map_lookup_entry(map, saved_start,
2330 &tmp_entry)) {
2331 if (flags & VM_MAP_WIRE_HOLESOK)
2332 tmp_entry = tmp_entry->next;
2333 else {
2334 if (saved_start == start) {
2335 /*
2336 * First_entry has been deleted.
2337 */
2338 vm_map_unlock(map);
2339 return (KERN_INVALID_ADDRESS);
2340 }
2341 end = saved_start;
2342 rv = KERN_INVALID_ADDRESS;
2343 goto done;
2344 }
2345 }
2346 if (entry == first_entry)
2347 first_entry = tmp_entry;
2348 else
2349 first_entry = NULL;
2350 entry = tmp_entry;
2351 }
2352 last_timestamp = map->timestamp;
2353 continue;
2354 }
2355 vm_map_clip_start(map, entry, start);
2356 vm_map_clip_end(map, entry, end);
2357 /*
2358 * Mark the entry in case the map lock is released. (See
2359 * above.)
2360 */
2361 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2362 entry->wiring_thread == NULL,
2363 ("owned map entry %p", entry));
2364 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2365 entry->wiring_thread = curthread;
2366 /*
2367 * Check the map for holes in the specified region.
2368 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2369 */
2370 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2371 (entry->end < end && (entry->next == &map->header ||
2372 entry->next->start > entry->end))) {
2373 end = entry->end;
2374 rv = KERN_INVALID_ADDRESS;
2375 goto done;
2376 }
2377 /*
2378 * If system unwiring, require that the entry is system wired.
2379 */
2380 if (!user_unwire &&
2381 vm_map_entry_system_wired_count(entry) == 0) {
2382 end = entry->end;
2383 rv = KERN_INVALID_ARGUMENT;
2384 goto done;
2385 }
2386 entry = entry->next;
2387 }
2388 rv = KERN_SUCCESS;
2389 done:
2390 need_wakeup = FALSE;
2391 if (first_entry == NULL) {
2392 result = vm_map_lookup_entry(map, start, &first_entry);
2393 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2394 first_entry = first_entry->next;
2395 else
2396 KASSERT(result, ("vm_map_unwire: lookup failed"));
2397 }
2398 for (entry = first_entry; entry != &map->header && entry->start < end;
2399 entry = entry->next) {
2400 /*
2401 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2402 * space in the unwired region could have been mapped
2403 * while the map lock was dropped for draining
2404 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2405 * could be simultaneously wiring this new mapping
2406 * entry. Detect these cases and skip any entries
2407 * marked as in transition by us.
2408 */
2409 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2410 entry->wiring_thread != curthread) {
2411 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2412 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2413 continue;
2414 }
2415
2416 if (rv == KERN_SUCCESS && (!user_unwire ||
2417 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2418 if (user_unwire)
2419 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2420 if (entry->wired_count == 1)
2421 vm_map_entry_unwire(map, entry);
2422 else
2423 entry->wired_count--;
2424 }
2425 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2426 ("vm_map_unwire: in-transition flag missing %p", entry));
2427 KASSERT(entry->wiring_thread == curthread,
2428 ("vm_map_unwire: alien wire %p", entry));
2429 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2430 entry->wiring_thread = NULL;
2431 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2432 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2433 need_wakeup = TRUE;
2434 }
2435 vm_map_simplify_entry(map, entry);
2436 }
2437 vm_map_unlock(map);
2438 if (need_wakeup)
2439 vm_map_wakeup(map);
2440 return (rv);
2441 }
2442
2443 /*
2444 * vm_map_wire_entry_failure:
2445 *
2446 * Handle a wiring failure on the given entry.
2447 *
2448 * The map should be locked.
2449 */
2450 static void
2451 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2452 vm_offset_t failed_addr)
2453 {
2454
2455 VM_MAP_ASSERT_LOCKED(map);
2456 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2457 entry->wired_count == 1,
2458 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2459 KASSERT(failed_addr < entry->end,
2460 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2461
2462 /*
2463 * If any pages at the start of this entry were successfully wired,
2464 * then unwire them.
2465 */
2466 if (failed_addr > entry->start) {
2467 pmap_unwire(map->pmap, entry->start, failed_addr);
2468 vm_object_unwire(entry->object.vm_object, entry->offset,
2469 failed_addr - entry->start, PQ_ACTIVE);
2470 }
2471
2472 /*
2473 * Assign an out-of-range value to represent the failure to wire this
2474 * entry.
2475 */
2476 entry->wired_count = -1;
2477 }
2478
2479 /*
2480 * vm_map_wire:
2481 *
2482 * Implements both kernel and user wiring.
2483 */
2484 int
2485 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2486 int flags)
2487 {
2488 vm_map_entry_t entry, first_entry, tmp_entry;
2489 vm_offset_t faddr, saved_end, saved_start;
2490 unsigned int last_timestamp;
2491 int rv;
2492 boolean_t need_wakeup, result, user_wire;
2493 vm_prot_t prot;
2494
2495 if (start == end)
2496 return (KERN_SUCCESS);
2497 prot = 0;
2498 if (flags & VM_MAP_WIRE_WRITE)
2499 prot |= VM_PROT_WRITE;
2500 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2501 vm_map_lock(map);
2502 VM_MAP_RANGE_CHECK(map, start, end);
2503 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2504 if (flags & VM_MAP_WIRE_HOLESOK)
2505 first_entry = first_entry->next;
2506 else {
2507 vm_map_unlock(map);
2508 return (KERN_INVALID_ADDRESS);
2509 }
2510 }
2511 last_timestamp = map->timestamp;
2512 entry = first_entry;
2513 while (entry != &map->header && entry->start < end) {
2514 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2515 /*
2516 * We have not yet clipped the entry.
2517 */
2518 saved_start = (start >= entry->start) ? start :
2519 entry->start;
2520 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2521 if (vm_map_unlock_and_wait(map, 0)) {
2522 /*
2523 * Allow interruption of user wiring?
2524 */
2525 }
2526 vm_map_lock(map);
2527 if (last_timestamp + 1 != map->timestamp) {
2528 /*
2529 * Look again for the entry because the map was
2530 * modified while it was unlocked.
2531 * Specifically, the entry may have been
2532 * clipped, merged, or deleted.
2533 */
2534 if (!vm_map_lookup_entry(map, saved_start,
2535 &tmp_entry)) {
2536 if (flags & VM_MAP_WIRE_HOLESOK)
2537 tmp_entry = tmp_entry->next;
2538 else {
2539 if (saved_start == start) {
2540 /*
2541 * first_entry has been deleted.
2542 */
2543 vm_map_unlock(map);
2544 return (KERN_INVALID_ADDRESS);
2545 }
2546 end = saved_start;
2547 rv = KERN_INVALID_ADDRESS;
2548 goto done;
2549 }
2550 }
2551 if (entry == first_entry)
2552 first_entry = tmp_entry;
2553 else
2554 first_entry = NULL;
2555 entry = tmp_entry;
2556 }
2557 last_timestamp = map->timestamp;
2558 continue;
2559 }
2560 vm_map_clip_start(map, entry, start);
2561 vm_map_clip_end(map, entry, end);
2562 /*
2563 * Mark the entry in case the map lock is released. (See
2564 * above.)
2565 */
2566 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2567 entry->wiring_thread == NULL,
2568 ("owned map entry %p", entry));
2569 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2570 entry->wiring_thread = curthread;
2571 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2572 || (entry->protection & prot) != prot) {
2573 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2574 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2575 end = entry->end;
2576 rv = KERN_INVALID_ADDRESS;
2577 goto done;
2578 }
2579 goto next_entry;
2580 }
2581 if (entry->wired_count == 0) {
2582 entry->wired_count++;
2583 saved_start = entry->start;
2584 saved_end = entry->end;
2585
2586 /*
2587 * Release the map lock, relying on the in-transition
2588 * mark. Mark the map busy for fork.
2589 */
2590 vm_map_busy(map);
2591 vm_map_unlock(map);
2592
2593 faddr = saved_start;
2594 do {
2595 /*
2596 * Simulate a fault to get the page and enter
2597 * it into the physical map.
2598 */
2599 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2600 VM_FAULT_CHANGE_WIRING)) != KERN_SUCCESS)
2601 break;
2602 } while ((faddr += PAGE_SIZE) < saved_end);
2603 vm_map_lock(map);
2604 vm_map_unbusy(map);
2605 if (last_timestamp + 1 != map->timestamp) {
2606 /*
2607 * Look again for the entry because the map was
2608 * modified while it was unlocked. The entry
2609 * may have been clipped, but NOT merged or
2610 * deleted.
2611 */
2612 result = vm_map_lookup_entry(map, saved_start,
2613 &tmp_entry);
2614 KASSERT(result, ("vm_map_wire: lookup failed"));
2615 if (entry == first_entry)
2616 first_entry = tmp_entry;
2617 else
2618 first_entry = NULL;
2619 entry = tmp_entry;
2620 while (entry->end < saved_end) {
2621 /*
2622 * In case of failure, handle entries
2623 * that were not fully wired here;
2624 * fully wired entries are handled
2625 * later.
2626 */
2627 if (rv != KERN_SUCCESS &&
2628 faddr < entry->end)
2629 vm_map_wire_entry_failure(map,
2630 entry, faddr);
2631 entry = entry->next;
2632 }
2633 }
2634 last_timestamp = map->timestamp;
2635 if (rv != KERN_SUCCESS) {
2636 vm_map_wire_entry_failure(map, entry, faddr);
2637 end = entry->end;
2638 goto done;
2639 }
2640 } else if (!user_wire ||
2641 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2642 entry->wired_count++;
2643 }
2644 /*
2645 * Check the map for holes in the specified region.
2646 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2647 */
2648 next_entry:
2649 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2650 (entry->end < end && (entry->next == &map->header ||
2651 entry->next->start > entry->end))) {
2652 end = entry->end;
2653 rv = KERN_INVALID_ADDRESS;
2654 goto done;
2655 }
2656 entry = entry->next;
2657 }
2658 rv = KERN_SUCCESS;
2659 done:
2660 need_wakeup = FALSE;
2661 if (first_entry == NULL) {
2662 result = vm_map_lookup_entry(map, start, &first_entry);
2663 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2664 first_entry = first_entry->next;
2665 else
2666 KASSERT(result, ("vm_map_wire: lookup failed"));
2667 }
2668 for (entry = first_entry; entry != &map->header && entry->start < end;
2669 entry = entry->next) {
2670 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2671 goto next_entry_done;
2672
2673 /*
2674 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2675 * space in the unwired region could have been mapped
2676 * while the map lock was dropped for faulting in the
2677 * pages or draining MAP_ENTRY_IN_TRANSITION.
2678 * Moreover, another thread could be simultaneously
2679 * wiring this new mapping entry. Detect these cases
2680 * and skip any entries marked as in transition by us.
2681 */
2682 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2683 entry->wiring_thread != curthread) {
2684 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2685 ("vm_map_wire: !HOLESOK and new/changed entry"));
2686 continue;
2687 }
2688
2689 if (rv == KERN_SUCCESS) {
2690 if (user_wire)
2691 entry->eflags |= MAP_ENTRY_USER_WIRED;
2692 } else if (entry->wired_count == -1) {
2693 /*
2694 * Wiring failed on this entry. Thus, unwiring is
2695 * unnecessary.
2696 */
2697 entry->wired_count = 0;
2698 } else if (!user_wire ||
2699 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2700 /*
2701 * Undo the wiring. Wiring succeeded on this entry
2702 * but failed on a later entry.
2703 */
2704 if (entry->wired_count == 1)
2705 vm_map_entry_unwire(map, entry);
2706 else
2707 entry->wired_count--;
2708 }
2709 next_entry_done:
2710 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2711 ("vm_map_wire: in-transition flag missing %p", entry));
2712 KASSERT(entry->wiring_thread == curthread,
2713 ("vm_map_wire: alien wire %p", entry));
2714 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2715 MAP_ENTRY_WIRE_SKIPPED);
2716 entry->wiring_thread = NULL;
2717 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2718 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2719 need_wakeup = TRUE;
2720 }
2721 vm_map_simplify_entry(map, entry);
2722 }
2723 vm_map_unlock(map);
2724 if (need_wakeup)
2725 vm_map_wakeup(map);
2726 return (rv);
2727 }
2728
2729 /*
2730 * vm_map_sync
2731 *
2732 * Push any dirty cached pages in the address range to their pager.
2733 * If syncio is TRUE, dirty pages are written synchronously.
2734 * If invalidate is TRUE, any cached pages are freed as well.
2735 *
2736 * If the size of the region from start to end is zero, we are
2737 * supposed to flush all modified pages within the region containing
2738 * start. Unfortunately, a region can be split or coalesced with
2739 * neighboring regions, making it difficult to determine what the
2740 * original region was. Therefore, we approximate this requirement by
2741 * flushing the current region containing start.
2742 *
2743 * Returns an error if any part of the specified range is not mapped.
2744 */
2745 int
2746 vm_map_sync(
2747 vm_map_t map,
2748 vm_offset_t start,
2749 vm_offset_t end,
2750 boolean_t syncio,
2751 boolean_t invalidate)
2752 {
2753 vm_map_entry_t current;
2754 vm_map_entry_t entry;
2755 vm_size_t size;
2756 vm_object_t object;
2757 vm_ooffset_t offset;
2758 unsigned int last_timestamp;
2759 boolean_t failed;
2760
2761 vm_map_lock_read(map);
2762 VM_MAP_RANGE_CHECK(map, start, end);
2763 if (!vm_map_lookup_entry(map, start, &entry)) {
2764 vm_map_unlock_read(map);
2765 return (KERN_INVALID_ADDRESS);
2766 } else if (start == end) {
2767 start = entry->start;
2768 end = entry->end;
2769 }
2770 /*
2771 * Make a first pass to check for user-wired memory and holes.
2772 */
2773 for (current = entry; current != &map->header && current->start < end;
2774 current = current->next) {
2775 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2776 vm_map_unlock_read(map);
2777 return (KERN_INVALID_ARGUMENT);
2778 }
2779 if (end > current->end &&
2780 (current->next == &map->header ||
2781 current->end != current->next->start)) {
2782 vm_map_unlock_read(map);
2783 return (KERN_INVALID_ADDRESS);
2784 }
2785 }
2786
2787 if (invalidate)
2788 pmap_remove(map->pmap, start, end);
2789 failed = FALSE;
2790
2791 /*
2792 * Make a second pass, cleaning/uncaching pages from the indicated
2793 * objects as we go.
2794 */
2795 for (current = entry; current != &map->header && current->start < end;) {
2796 offset = current->offset + (start - current->start);
2797 size = (end <= current->end ? end : current->end) - start;
2798 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2799 vm_map_t smap;
2800 vm_map_entry_t tentry;
2801 vm_size_t tsize;
2802
2803 smap = current->object.sub_map;
2804 vm_map_lock_read(smap);
2805 (void) vm_map_lookup_entry(smap, offset, &tentry);
2806 tsize = tentry->end - offset;
2807 if (tsize < size)
2808 size = tsize;
2809 object = tentry->object.vm_object;
2810 offset = tentry->offset + (offset - tentry->start);
2811 vm_map_unlock_read(smap);
2812 } else {
2813 object = current->object.vm_object;
2814 }
2815 vm_object_reference(object);
2816 last_timestamp = map->timestamp;
2817 vm_map_unlock_read(map);
2818 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2819 failed = TRUE;
2820 start += size;
2821 vm_object_deallocate(object);
2822 vm_map_lock_read(map);
2823 if (last_timestamp == map->timestamp ||
2824 !vm_map_lookup_entry(map, start, ¤t))
2825 current = current->next;
2826 }
2827
2828 vm_map_unlock_read(map);
2829 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2830 }
2831
2832 /*
2833 * vm_map_entry_unwire: [ internal use only ]
2834 *
2835 * Make the region specified by this entry pageable.
2836 *
2837 * The map in question should be locked.
2838 * [This is the reason for this routine's existence.]
2839 */
2840 static void
2841 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2842 {
2843
2844 VM_MAP_ASSERT_LOCKED(map);
2845 KASSERT(entry->wired_count > 0,
2846 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2847 pmap_unwire(map->pmap, entry->start, entry->end);
2848 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2849 entry->start, PQ_ACTIVE);
2850 entry->wired_count = 0;
2851 }
2852
2853 static void
2854 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2855 {
2856
2857 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2858 vm_object_deallocate(entry->object.vm_object);
2859 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2860 }
2861
2862 /*
2863 * vm_map_entry_delete: [ internal use only ]
2864 *
2865 * Deallocate the given entry from the target map.
2866 */
2867 static void
2868 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2869 {
2870 vm_object_t object;
2871 vm_pindex_t offidxstart, offidxend, count, size1;
2872 vm_ooffset_t size;
2873
2874 vm_map_entry_unlink(map, entry);
2875 object = entry->object.vm_object;
2876 size = entry->end - entry->start;
2877 map->size -= size;
2878
2879 if (entry->cred != NULL) {
2880 swap_release_by_cred(size, entry->cred);
2881 crfree(entry->cred);
2882 }
2883
2884 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2885 (object != NULL)) {
2886 KASSERT(entry->cred == NULL || object->cred == NULL ||
2887 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2888 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2889 count = OFF_TO_IDX(size);
2890 offidxstart = OFF_TO_IDX(entry->offset);
2891 offidxend = offidxstart + count;
2892 VM_OBJECT_WLOCK(object);
2893 if (object->ref_count != 1 &&
2894 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2895 object == kernel_object || object == kmem_object)) {
2896 vm_object_collapse(object);
2897
2898 /*
2899 * The option OBJPR_NOTMAPPED can be passed here
2900 * because vm_map_delete() already performed
2901 * pmap_remove() on the only mapping to this range
2902 * of pages.
2903 */
2904 vm_object_page_remove(object, offidxstart, offidxend,
2905 OBJPR_NOTMAPPED);
2906 if (object->type == OBJT_SWAP)
2907 swap_pager_freespace(object, offidxstart, count);
2908 if (offidxend >= object->size &&
2909 offidxstart < object->size) {
2910 size1 = object->size;
2911 object->size = offidxstart;
2912 if (object->cred != NULL) {
2913 size1 -= object->size;
2914 KASSERT(object->charge >= ptoa(size1),
2915 ("vm_map_entry_delete: object->charge < 0"));
2916 swap_release_by_cred(ptoa(size1), object->cred);
2917 object->charge -= ptoa(size1);
2918 }
2919 }
2920 }
2921 VM_OBJECT_WUNLOCK(object);
2922 } else
2923 entry->object.vm_object = NULL;
2924 if (map->system_map)
2925 vm_map_entry_deallocate(entry, TRUE);
2926 else {
2927 entry->next = curthread->td_map_def_user;
2928 curthread->td_map_def_user = entry;
2929 }
2930 }
2931
2932 /*
2933 * vm_map_delete: [ internal use only ]
2934 *
2935 * Deallocates the given address range from the target
2936 * map.
2937 */
2938 int
2939 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2940 {
2941 vm_map_entry_t entry;
2942 vm_map_entry_t first_entry;
2943
2944 VM_MAP_ASSERT_LOCKED(map);
2945 if (start == end)
2946 return (KERN_SUCCESS);
2947
2948 /*
2949 * Find the start of the region, and clip it
2950 */
2951 if (!vm_map_lookup_entry(map, start, &first_entry))
2952 entry = first_entry->next;
2953 else {
2954 entry = first_entry;
2955 vm_map_clip_start(map, entry, start);
2956 }
2957
2958 /*
2959 * Step through all entries in this region
2960 */
2961 while ((entry != &map->header) && (entry->start < end)) {
2962 vm_map_entry_t next;
2963
2964 /*
2965 * Wait for wiring or unwiring of an entry to complete.
2966 * Also wait for any system wirings to disappear on
2967 * user maps.
2968 */
2969 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2970 (vm_map_pmap(map) != kernel_pmap &&
2971 vm_map_entry_system_wired_count(entry) != 0)) {
2972 unsigned int last_timestamp;
2973 vm_offset_t saved_start;
2974 vm_map_entry_t tmp_entry;
2975
2976 saved_start = entry->start;
2977 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2978 last_timestamp = map->timestamp;
2979 (void) vm_map_unlock_and_wait(map, 0);
2980 vm_map_lock(map);
2981 if (last_timestamp + 1 != map->timestamp) {
2982 /*
2983 * Look again for the entry because the map was
2984 * modified while it was unlocked.
2985 * Specifically, the entry may have been
2986 * clipped, merged, or deleted.
2987 */
2988 if (!vm_map_lookup_entry(map, saved_start,
2989 &tmp_entry))
2990 entry = tmp_entry->next;
2991 else {
2992 entry = tmp_entry;
2993 vm_map_clip_start(map, entry,
2994 saved_start);
2995 }
2996 }
2997 continue;
2998 }
2999 vm_map_clip_end(map, entry, end);
3000
3001 next = entry->next;
3002
3003 /*
3004 * Unwire before removing addresses from the pmap; otherwise,
3005 * unwiring will put the entries back in the pmap.
3006 */
3007 if (entry->wired_count != 0) {
3008 vm_map_entry_unwire(map, entry);
3009 }
3010
3011 pmap_remove(map->pmap, entry->start, entry->end);
3012
3013 /*
3014 * Delete the entry only after removing all pmap
3015 * entries pointing to its pages. (Otherwise, its
3016 * page frames may be reallocated, and any modify bits
3017 * will be set in the wrong object!)
3018 */
3019 vm_map_entry_delete(map, entry);
3020 entry = next;
3021 }
3022 return (KERN_SUCCESS);
3023 }
3024
3025 /*
3026 * vm_map_remove:
3027 *
3028 * Remove the given address range from the target map.
3029 * This is the exported form of vm_map_delete.
3030 */
3031 int
3032 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3033 {
3034 int result;
3035
3036 vm_map_lock(map);
3037 VM_MAP_RANGE_CHECK(map, start, end);
3038 result = vm_map_delete(map, start, end);
3039 vm_map_unlock(map);
3040 return (result);
3041 }
3042
3043 /*
3044 * vm_map_check_protection:
3045 *
3046 * Assert that the target map allows the specified privilege on the
3047 * entire address region given. The entire region must be allocated.
3048 *
3049 * WARNING! This code does not and should not check whether the
3050 * contents of the region is accessible. For example a smaller file
3051 * might be mapped into a larger address space.
3052 *
3053 * NOTE! This code is also called by munmap().
3054 *
3055 * The map must be locked. A read lock is sufficient.
3056 */
3057 boolean_t
3058 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3059 vm_prot_t protection)
3060 {
3061 vm_map_entry_t entry;
3062 vm_map_entry_t tmp_entry;
3063
3064 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3065 return (FALSE);
3066 entry = tmp_entry;
3067
3068 while (start < end) {
3069 if (entry == &map->header)
3070 return (FALSE);
3071 /*
3072 * No holes allowed!
3073 */
3074 if (start < entry->start)
3075 return (FALSE);
3076 /*
3077 * Check protection associated with entry.
3078 */
3079 if ((entry->protection & protection) != protection)
3080 return (FALSE);
3081 /* go to next entry */
3082 start = entry->end;
3083 entry = entry->next;
3084 }
3085 return (TRUE);
3086 }
3087
3088 /*
3089 * vm_map_copy_entry:
3090 *
3091 * Copies the contents of the source entry to the destination
3092 * entry. The entries *must* be aligned properly.
3093 */
3094 static void
3095 vm_map_copy_entry(
3096 vm_map_t src_map,
3097 vm_map_t dst_map,
3098 vm_map_entry_t src_entry,
3099 vm_map_entry_t dst_entry,
3100 vm_ooffset_t *fork_charge)
3101 {
3102 vm_object_t src_object;
3103 vm_map_entry_t fake_entry;
3104 vm_offset_t size;
3105 struct ucred *cred;
3106 int charged;
3107
3108 VM_MAP_ASSERT_LOCKED(dst_map);
3109
3110 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3111 return;
3112
3113 if (src_entry->wired_count == 0 ||
3114 (src_entry->protection & VM_PROT_WRITE) == 0) {
3115 /*
3116 * If the source entry is marked needs_copy, it is already
3117 * write-protected.
3118 */
3119 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3120 (src_entry->protection & VM_PROT_WRITE) != 0) {
3121 pmap_protect(src_map->pmap,
3122 src_entry->start,
3123 src_entry->end,
3124 src_entry->protection & ~VM_PROT_WRITE);
3125 }
3126
3127 /*
3128 * Make a copy of the object.
3129 */
3130 size = src_entry->end - src_entry->start;
3131 if ((src_object = src_entry->object.vm_object) != NULL) {
3132 VM_OBJECT_WLOCK(src_object);
3133 charged = ENTRY_CHARGED(src_entry);
3134 if ((src_object->handle == NULL) &&
3135 (src_object->type == OBJT_DEFAULT ||
3136 src_object->type == OBJT_SWAP)) {
3137 vm_object_collapse(src_object);
3138 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3139 vm_object_split(src_entry);
3140 src_object = src_entry->object.vm_object;
3141 }
3142 }
3143 vm_object_reference_locked(src_object);
3144 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3145 if (src_entry->cred != NULL &&
3146 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3147 KASSERT(src_object->cred == NULL,
3148 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3149 src_object));
3150 src_object->cred = src_entry->cred;
3151 src_object->charge = size;
3152 }
3153 VM_OBJECT_WUNLOCK(src_object);
3154 dst_entry->object.vm_object = src_object;
3155 if (charged) {
3156 cred = curthread->td_ucred;
3157 crhold(cred);
3158 dst_entry->cred = cred;
3159 *fork_charge += size;
3160 if (!(src_entry->eflags &
3161 MAP_ENTRY_NEEDS_COPY)) {
3162 crhold(cred);
3163 src_entry->cred = cred;
3164 *fork_charge += size;
3165 }
3166 }
3167 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3168 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3169 dst_entry->offset = src_entry->offset;
3170 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3171 /*
3172 * MAP_ENTRY_VN_WRITECNT cannot
3173 * indicate write reference from
3174 * src_entry, since the entry is
3175 * marked as needs copy. Allocate a
3176 * fake entry that is used to
3177 * decrement object->un_pager.vnp.writecount
3178 * at the appropriate time. Attach
3179 * fake_entry to the deferred list.
3180 */
3181 fake_entry = vm_map_entry_create(dst_map);
3182 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3183 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3184 vm_object_reference(src_object);
3185 fake_entry->object.vm_object = src_object;
3186 fake_entry->start = src_entry->start;
3187 fake_entry->end = src_entry->end;
3188 fake_entry->next = curthread->td_map_def_user;
3189 curthread->td_map_def_user = fake_entry;
3190 }
3191 } else {
3192 dst_entry->object.vm_object = NULL;
3193 dst_entry->offset = 0;
3194 if (src_entry->cred != NULL) {
3195 dst_entry->cred = curthread->td_ucred;
3196 crhold(dst_entry->cred);
3197 *fork_charge += size;
3198 }
3199 }
3200
3201 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3202 dst_entry->end - dst_entry->start, src_entry->start);
3203 } else {
3204 /*
3205 * We don't want to make writeable wired pages copy-on-write.
3206 * Immediately copy these pages into the new map by simulating
3207 * page faults. The new pages are pageable.
3208 */
3209 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3210 fork_charge);
3211 }
3212 }
3213
3214 /*
3215 * vmspace_map_entry_forked:
3216 * Update the newly-forked vmspace each time a map entry is inherited
3217 * or copied. The values for vm_dsize and vm_tsize are approximate
3218 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3219 */
3220 static void
3221 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3222 vm_map_entry_t entry)
3223 {
3224 vm_size_t entrysize;
3225 vm_offset_t newend;
3226
3227 entrysize = entry->end - entry->start;
3228 vm2->vm_map.size += entrysize;
3229 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3230 vm2->vm_ssize += btoc(entrysize);
3231 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3232 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3233 newend = MIN(entry->end,
3234 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3235 vm2->vm_dsize += btoc(newend - entry->start);
3236 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3237 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3238 newend = MIN(entry->end,
3239 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3240 vm2->vm_tsize += btoc(newend - entry->start);
3241 }
3242 }
3243
3244 /*
3245 * vmspace_fork:
3246 * Create a new process vmspace structure and vm_map
3247 * based on those of an existing process. The new map
3248 * is based on the old map, according to the inheritance
3249 * values on the regions in that map.
3250 *
3251 * XXX It might be worth coalescing the entries added to the new vmspace.
3252 *
3253 * The source map must not be locked.
3254 */
3255 struct vmspace *
3256 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3257 {
3258 struct vmspace *vm2;
3259 vm_map_t new_map, old_map;
3260 vm_map_entry_t new_entry, old_entry;
3261 vm_object_t object;
3262 int locked;
3263
3264 old_map = &vm1->vm_map;
3265 /* Copy immutable fields of vm1 to vm2. */
3266 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3267 if (vm2 == NULL)
3268 return (NULL);
3269 vm2->vm_taddr = vm1->vm_taddr;
3270 vm2->vm_daddr = vm1->vm_daddr;
3271 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3272 vm_map_lock(old_map);
3273 if (old_map->busy)
3274 vm_map_wait_busy(old_map);
3275 new_map = &vm2->vm_map;
3276 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3277 KASSERT(locked, ("vmspace_fork: lock failed"));
3278
3279 old_entry = old_map->header.next;
3280
3281 while (old_entry != &old_map->header) {
3282 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3283 panic("vm_map_fork: encountered a submap");
3284
3285 switch (old_entry->inheritance) {
3286 case VM_INHERIT_NONE:
3287 break;
3288
3289 case VM_INHERIT_SHARE:
3290 /*
3291 * Clone the entry, creating the shared object if necessary.
3292 */
3293 object = old_entry->object.vm_object;
3294 if (object == NULL) {
3295 object = vm_object_allocate(OBJT_DEFAULT,
3296 atop(old_entry->end - old_entry->start));
3297 old_entry->object.vm_object = object;
3298 old_entry->offset = 0;
3299 if (old_entry->cred != NULL) {
3300 object->cred = old_entry->cred;
3301 object->charge = old_entry->end -
3302 old_entry->start;
3303 old_entry->cred = NULL;
3304 }
3305 }
3306
3307 /*
3308 * Add the reference before calling vm_object_shadow
3309 * to insure that a shadow object is created.
3310 */
3311 vm_object_reference(object);
3312 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3313 vm_object_shadow(&old_entry->object.vm_object,
3314 &old_entry->offset,
3315 old_entry->end - old_entry->start);
3316 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3317 /* Transfer the second reference too. */
3318 vm_object_reference(
3319 old_entry->object.vm_object);
3320
3321 /*
3322 * As in vm_map_simplify_entry(), the
3323 * vnode lock will not be acquired in
3324 * this call to vm_object_deallocate().
3325 */
3326 vm_object_deallocate(object);
3327 object = old_entry->object.vm_object;
3328 }
3329 VM_OBJECT_WLOCK(object);
3330 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3331 if (old_entry->cred != NULL) {
3332 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3333 object->cred = old_entry->cred;
3334 object->charge = old_entry->end - old_entry->start;
3335 old_entry->cred = NULL;
3336 }
3337
3338 /*
3339 * Assert the correct state of the vnode
3340 * v_writecount while the object is locked, to
3341 * not relock it later for the assertion
3342 * correctness.
3343 */
3344 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3345 object->type == OBJT_VNODE) {
3346 KASSERT(((struct vnode *)object->handle)->
3347 v_writecount > 0,
3348 ("vmspace_fork: v_writecount %p", object));
3349 KASSERT(object->un_pager.vnp.writemappings > 0,
3350 ("vmspace_fork: vnp.writecount %p",
3351 object));
3352 }
3353 VM_OBJECT_WUNLOCK(object);
3354
3355 /*
3356 * Clone the entry, referencing the shared object.
3357 */
3358 new_entry = vm_map_entry_create(new_map);
3359 *new_entry = *old_entry;
3360 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3361 MAP_ENTRY_IN_TRANSITION);
3362 new_entry->wiring_thread = NULL;
3363 new_entry->wired_count = 0;
3364 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3365 vnode_pager_update_writecount(object,
3366 new_entry->start, new_entry->end);
3367 }
3368
3369 /*
3370 * Insert the entry into the new map -- we know we're
3371 * inserting at the end of the new map.
3372 */
3373 vm_map_entry_link(new_map, new_map->header.prev,
3374 new_entry);
3375 vmspace_map_entry_forked(vm1, vm2, new_entry);
3376
3377 /*
3378 * Update the physical map
3379 */
3380 pmap_copy(new_map->pmap, old_map->pmap,
3381 new_entry->start,
3382 (old_entry->end - old_entry->start),
3383 old_entry->start);
3384 break;
3385
3386 case VM_INHERIT_COPY:
3387 /*
3388 * Clone the entry and link into the map.
3389 */
3390 new_entry = vm_map_entry_create(new_map);
3391 *new_entry = *old_entry;
3392 /*
3393 * Copied entry is COW over the old object.
3394 */
3395 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3396 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3397 new_entry->wiring_thread = NULL;
3398 new_entry->wired_count = 0;
3399 new_entry->object.vm_object = NULL;
3400 new_entry->cred = NULL;
3401 vm_map_entry_link(new_map, new_map->header.prev,
3402 new_entry);
3403 vmspace_map_entry_forked(vm1, vm2, new_entry);
3404 vm_map_copy_entry(old_map, new_map, old_entry,
3405 new_entry, fork_charge);
3406 break;
3407 }
3408 old_entry = old_entry->next;
3409 }
3410 /*
3411 * Use inlined vm_map_unlock() to postpone handling the deferred
3412 * map entries, which cannot be done until both old_map and
3413 * new_map locks are released.
3414 */
3415 sx_xunlock(&old_map->lock);
3416 sx_xunlock(&new_map->lock);
3417 vm_map_process_deferred();
3418
3419 return (vm2);
3420 }
3421
3422 int
3423 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3424 vm_prot_t prot, vm_prot_t max, int cow)
3425 {
3426 vm_size_t growsize, init_ssize;
3427 rlim_t lmemlim, vmemlim;
3428 int rv;
3429
3430 growsize = sgrowsiz;
3431 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3432 vm_map_lock(map);
3433 PROC_LOCK(curproc);
3434 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK);
3435 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3436 PROC_UNLOCK(curproc);
3437 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3438 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3439 rv = KERN_NO_SPACE;
3440 goto out;
3441 }
3442 }
3443 /* If we would blow our VMEM resource limit, no go */
3444 if (map->size + init_ssize > vmemlim) {
3445 rv = KERN_NO_SPACE;
3446 goto out;
3447 }
3448 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3449 max, cow);
3450 out:
3451 vm_map_unlock(map);
3452 return (rv);
3453 }
3454
3455 static int
3456 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3457 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3458 {
3459 vm_map_entry_t new_entry, prev_entry;
3460 vm_offset_t bot, top;
3461 vm_size_t init_ssize;
3462 int orient, rv;
3463
3464 /*
3465 * The stack orientation is piggybacked with the cow argument.
3466 * Extract it into orient and mask the cow argument so that we
3467 * don't pass it around further.
3468 * NOTE: We explicitly allow bi-directional stacks.
3469 */
3470 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3471 KASSERT(orient != 0, ("No stack grow direction"));
3472
3473 if (addrbos < vm_map_min(map) ||
3474 addrbos > vm_map_max(map) ||
3475 addrbos + max_ssize < addrbos)
3476 return (KERN_NO_SPACE);
3477
3478 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3479
3480 /* If addr is already mapped, no go */
3481 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3482 return (KERN_NO_SPACE);
3483
3484 /*
3485 * If we can't accomodate max_ssize in the current mapping, no go.
3486 * However, we need to be aware that subsequent user mappings might
3487 * map into the space we have reserved for stack, and currently this
3488 * space is not protected.
3489 *
3490 * Hopefully we will at least detect this condition when we try to
3491 * grow the stack.
3492 */
3493 if ((prev_entry->next != &map->header) &&
3494 (prev_entry->next->start < addrbos + max_ssize))
3495 return (KERN_NO_SPACE);
3496
3497 /*
3498 * We initially map a stack of only init_ssize. We will grow as
3499 * needed later. Depending on the orientation of the stack (i.e.
3500 * the grow direction) we either map at the top of the range, the
3501 * bottom of the range or in the middle.
3502 *
3503 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3504 * and cow to be 0. Possibly we should eliminate these as input
3505 * parameters, and just pass these values here in the insert call.
3506 */
3507 if (orient == MAP_STACK_GROWS_DOWN)
3508 bot = addrbos + max_ssize - init_ssize;
3509 else if (orient == MAP_STACK_GROWS_UP)
3510 bot = addrbos;
3511 else
3512 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3513 top = bot + init_ssize;
3514 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3515
3516 /* Now set the avail_ssize amount. */
3517 if (rv == KERN_SUCCESS) {
3518 if (prev_entry != &map->header)
3519 vm_map_clip_end(map, prev_entry, bot);
3520 new_entry = prev_entry->next;
3521 if (new_entry->end != top || new_entry->start != bot)
3522 panic("Bad entry start/end for new stack entry");
3523
3524 new_entry->avail_ssize = max_ssize - init_ssize;
3525 if (orient & MAP_STACK_GROWS_DOWN)
3526 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3527 if (orient & MAP_STACK_GROWS_UP)
3528 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3529 }
3530
3531 return (rv);
3532 }
3533
3534 static int stack_guard_page = 0;
3535 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3536 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3537 &stack_guard_page, 0,
3538 "Insert stack guard page ahead of the growable segments.");
3539
3540 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3541 * desired address is already mapped, or if we successfully grow
3542 * the stack. Also returns KERN_SUCCESS if addr is outside the
3543 * stack range (this is strange, but preserves compatibility with
3544 * the grow function in vm_machdep.c).
3545 */
3546 int
3547 vm_map_growstack(struct proc *p, vm_offset_t addr)
3548 {
3549 vm_map_entry_t next_entry, prev_entry;
3550 vm_map_entry_t new_entry, stack_entry;
3551 struct vmspace *vm = p->p_vmspace;
3552 vm_map_t map = &vm->vm_map;
3553 vm_offset_t end;
3554 vm_size_t growsize;
3555 size_t grow_amount, max_grow;
3556 rlim_t lmemlim, stacklim, vmemlim;
3557 int is_procstack, rv;
3558 struct ucred *cred;
3559 #ifdef notyet
3560 uint64_t limit;
3561 #endif
3562 #ifdef RACCT
3563 int error;
3564 #endif
3565
3566 Retry:
3567 PROC_LOCK(p);
3568 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3569 stacklim = lim_cur(p, RLIMIT_STACK);
3570 vmemlim = lim_cur(p, RLIMIT_VMEM);
3571 PROC_UNLOCK(p);
3572
3573 vm_map_lock_read(map);
3574
3575 /* If addr is already in the entry range, no need to grow.*/
3576 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3577 vm_map_unlock_read(map);
3578 return (KERN_SUCCESS);
3579 }
3580
3581 next_entry = prev_entry->next;
3582 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3583 /*
3584 * This entry does not grow upwards. Since the address lies
3585 * beyond this entry, the next entry (if one exists) has to
3586 * be a downward growable entry. The entry list header is
3587 * never a growable entry, so it suffices to check the flags.
3588 */
3589 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3590 vm_map_unlock_read(map);
3591 return (KERN_SUCCESS);
3592 }
3593 stack_entry = next_entry;
3594 } else {
3595 /*
3596 * This entry grows upward. If the next entry does not at
3597 * least grow downwards, this is the entry we need to grow.
3598 * otherwise we have two possible choices and we have to
3599 * select one.
3600 */
3601 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3602 /*
3603 * We have two choices; grow the entry closest to
3604 * the address to minimize the amount of growth.
3605 */
3606 if (addr - prev_entry->end <= next_entry->start - addr)
3607 stack_entry = prev_entry;
3608 else
3609 stack_entry = next_entry;
3610 } else
3611 stack_entry = prev_entry;
3612 }
3613
3614 if (stack_entry == next_entry) {
3615 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3616 KASSERT(addr < stack_entry->start, ("foo"));
3617 end = (prev_entry != &map->header) ? prev_entry->end :
3618 stack_entry->start - stack_entry->avail_ssize;
3619 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3620 max_grow = stack_entry->start - end;
3621 } else {
3622 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3623 KASSERT(addr >= stack_entry->end, ("foo"));
3624 end = (next_entry != &map->header) ? next_entry->start :
3625 stack_entry->end + stack_entry->avail_ssize;
3626 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3627 max_grow = end - stack_entry->end;
3628 }
3629
3630 if (grow_amount > stack_entry->avail_ssize) {
3631 vm_map_unlock_read(map);
3632 return (KERN_NO_SPACE);
3633 }
3634
3635 /*
3636 * If there is no longer enough space between the entries nogo, and
3637 * adjust the available space. Note: this should only happen if the
3638 * user has mapped into the stack area after the stack was created,
3639 * and is probably an error.
3640 *
3641 * This also effectively destroys any guard page the user might have
3642 * intended by limiting the stack size.
3643 */
3644 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3645 if (vm_map_lock_upgrade(map))
3646 goto Retry;
3647
3648 stack_entry->avail_ssize = max_grow;
3649
3650 vm_map_unlock(map);
3651 return (KERN_NO_SPACE);
3652 }
3653
3654 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr) ? 1 : 0;
3655
3656 /*
3657 * If this is the main process stack, see if we're over the stack
3658 * limit.
3659 */
3660 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3661 vm_map_unlock_read(map);
3662 return (KERN_NO_SPACE);
3663 }
3664 #ifdef RACCT
3665 PROC_LOCK(p);
3666 if (is_procstack &&
3667 racct_set(p, RACCT_STACK, ctob(vm->vm_ssize) + grow_amount)) {
3668 PROC_UNLOCK(p);
3669 vm_map_unlock_read(map);
3670 return (KERN_NO_SPACE);
3671 }
3672 PROC_UNLOCK(p);
3673 #endif
3674
3675 /* Round up the grow amount modulo sgrowsiz */
3676 growsize = sgrowsiz;
3677 grow_amount = roundup(grow_amount, growsize);
3678 if (grow_amount > stack_entry->avail_ssize)
3679 grow_amount = stack_entry->avail_ssize;
3680 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3681 grow_amount = trunc_page((vm_size_t)stacklim) -
3682 ctob(vm->vm_ssize);
3683 }
3684 #ifdef notyet
3685 PROC_LOCK(p);
3686 limit = racct_get_available(p, RACCT_STACK);
3687 PROC_UNLOCK(p);
3688 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3689 grow_amount = limit - ctob(vm->vm_ssize);
3690 #endif
3691 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3692 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3693 vm_map_unlock_read(map);
3694 rv = KERN_NO_SPACE;
3695 goto out;
3696 }
3697 #ifdef RACCT
3698 PROC_LOCK(p);
3699 if (racct_set(p, RACCT_MEMLOCK,
3700 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3701 PROC_UNLOCK(p);
3702 vm_map_unlock_read(map);
3703 rv = KERN_NO_SPACE;
3704 goto out;
3705 }
3706 PROC_UNLOCK(p);
3707 #endif
3708 }
3709 /* If we would blow our VMEM resource limit, no go */
3710 if (map->size + grow_amount > vmemlim) {
3711 vm_map_unlock_read(map);
3712 rv = KERN_NO_SPACE;
3713 goto out;
3714 }
3715 #ifdef RACCT
3716 PROC_LOCK(p);
3717 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3718 PROC_UNLOCK(p);
3719 vm_map_unlock_read(map);
3720 rv = KERN_NO_SPACE;
3721 goto out;
3722 }
3723 PROC_UNLOCK(p);
3724 #endif
3725
3726 if (vm_map_lock_upgrade(map))
3727 goto Retry;
3728
3729 if (stack_entry == next_entry) {
3730 /*
3731 * Growing downward.
3732 */
3733 /* Get the preliminary new entry start value */
3734 addr = stack_entry->start - grow_amount;
3735
3736 /*
3737 * If this puts us into the previous entry, cut back our
3738 * growth to the available space. Also, see the note above.
3739 */
3740 if (addr < end) {
3741 stack_entry->avail_ssize = max_grow;
3742 addr = end;
3743 if (stack_guard_page)
3744 addr += PAGE_SIZE;
3745 }
3746
3747 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3748 next_entry->protection, next_entry->max_protection, 0);
3749
3750 /* Adjust the available stack space by the amount we grew. */
3751 if (rv == KERN_SUCCESS) {
3752 if (prev_entry != &map->header)
3753 vm_map_clip_end(map, prev_entry, addr);
3754 new_entry = prev_entry->next;
3755 KASSERT(new_entry == stack_entry->prev, ("foo"));
3756 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3757 KASSERT(new_entry->start == addr, ("foo"));
3758 grow_amount = new_entry->end - new_entry->start;
3759 new_entry->avail_ssize = stack_entry->avail_ssize -
3760 grow_amount;
3761 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3762 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3763 }
3764 } else {
3765 /*
3766 * Growing upward.
3767 */
3768 addr = stack_entry->end + grow_amount;
3769
3770 /*
3771 * If this puts us into the next entry, cut back our growth
3772 * to the available space. Also, see the note above.
3773 */
3774 if (addr > end) {
3775 stack_entry->avail_ssize = end - stack_entry->end;
3776 addr = end;
3777 if (stack_guard_page)
3778 addr -= PAGE_SIZE;
3779 }
3780
3781 grow_amount = addr - stack_entry->end;
3782 cred = stack_entry->cred;
3783 if (cred == NULL && stack_entry->object.vm_object != NULL)
3784 cred = stack_entry->object.vm_object->cred;
3785 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3786 rv = KERN_NO_SPACE;
3787 /* Grow the underlying object if applicable. */
3788 else if (stack_entry->object.vm_object == NULL ||
3789 vm_object_coalesce(stack_entry->object.vm_object,
3790 stack_entry->offset,
3791 (vm_size_t)(stack_entry->end - stack_entry->start),
3792 (vm_size_t)grow_amount, cred != NULL)) {
3793 map->size += (addr - stack_entry->end);
3794 /* Update the current entry. */
3795 stack_entry->end = addr;
3796 stack_entry->avail_ssize -= grow_amount;
3797 vm_map_entry_resize_free(map, stack_entry);
3798 rv = KERN_SUCCESS;
3799
3800 if (next_entry != &map->header)
3801 vm_map_clip_start(map, next_entry, addr);
3802 } else
3803 rv = KERN_FAILURE;
3804 }
3805
3806 if (rv == KERN_SUCCESS && is_procstack)
3807 vm->vm_ssize += btoc(grow_amount);
3808
3809 vm_map_unlock(map);
3810
3811 /*
3812 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3813 */
3814 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3815 vm_map_wire(map,
3816 (stack_entry == next_entry) ? addr : addr - grow_amount,
3817 (stack_entry == next_entry) ? stack_entry->start : addr,
3818 (p->p_flag & P_SYSTEM)
3819 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3820 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3821 }
3822
3823 out:
3824 #ifdef RACCT
3825 if (rv != KERN_SUCCESS) {
3826 PROC_LOCK(p);
3827 error = racct_set(p, RACCT_VMEM, map->size);
3828 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3829 if (!old_mlock) {
3830 error = racct_set(p, RACCT_MEMLOCK,
3831 ptoa(pmap_wired_count(map->pmap)));
3832 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3833 }
3834 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3835 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3836 PROC_UNLOCK(p);
3837 }
3838 #endif
3839
3840 return (rv);
3841 }
3842
3843 /*
3844 * Unshare the specified VM space for exec. If other processes are
3845 * mapped to it, then create a new one. The new vmspace is null.
3846 */
3847 int
3848 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3849 {
3850 struct vmspace *oldvmspace = p->p_vmspace;
3851 struct vmspace *newvmspace;
3852
3853 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3854 ("vmspace_exec recursed"));
3855 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3856 if (newvmspace == NULL)
3857 return (ENOMEM);
3858 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3859 /*
3860 * This code is written like this for prototype purposes. The
3861 * goal is to avoid running down the vmspace here, but let the
3862 * other process's that are still using the vmspace to finally
3863 * run it down. Even though there is little or no chance of blocking
3864 * here, it is a good idea to keep this form for future mods.
3865 */
3866 PROC_VMSPACE_LOCK(p);
3867 p->p_vmspace = newvmspace;
3868 PROC_VMSPACE_UNLOCK(p);
3869 if (p == curthread->td_proc)
3870 pmap_activate(curthread);
3871 curthread->td_pflags |= TDP_EXECVMSPC;
3872 return (0);
3873 }
3874
3875 /*
3876 * Unshare the specified VM space for forcing COW. This
3877 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3878 */
3879 int
3880 vmspace_unshare(struct proc *p)
3881 {
3882 struct vmspace *oldvmspace = p->p_vmspace;
3883 struct vmspace *newvmspace;
3884 vm_ooffset_t fork_charge;
3885
3886 if (oldvmspace->vm_refcnt == 1)
3887 return (0);
3888 fork_charge = 0;
3889 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3890 if (newvmspace == NULL)
3891 return (ENOMEM);
3892 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3893 vmspace_free(newvmspace);
3894 return (ENOMEM);
3895 }
3896 PROC_VMSPACE_LOCK(p);
3897 p->p_vmspace = newvmspace;
3898 PROC_VMSPACE_UNLOCK(p);
3899 if (p == curthread->td_proc)
3900 pmap_activate(curthread);
3901 vmspace_free(oldvmspace);
3902 return (0);
3903 }
3904
3905 /*
3906 * vm_map_lookup:
3907 *
3908 * Finds the VM object, offset, and
3909 * protection for a given virtual address in the
3910 * specified map, assuming a page fault of the
3911 * type specified.
3912 *
3913 * Leaves the map in question locked for read; return
3914 * values are guaranteed until a vm_map_lookup_done
3915 * call is performed. Note that the map argument
3916 * is in/out; the returned map must be used in
3917 * the call to vm_map_lookup_done.
3918 *
3919 * A handle (out_entry) is returned for use in
3920 * vm_map_lookup_done, to make that fast.
3921 *
3922 * If a lookup is requested with "write protection"
3923 * specified, the map may be changed to perform virtual
3924 * copying operations, although the data referenced will
3925 * remain the same.
3926 */
3927 int
3928 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3929 vm_offset_t vaddr,
3930 vm_prot_t fault_typea,
3931 vm_map_entry_t *out_entry, /* OUT */
3932 vm_object_t *object, /* OUT */
3933 vm_pindex_t *pindex, /* OUT */
3934 vm_prot_t *out_prot, /* OUT */
3935 boolean_t *wired) /* OUT */
3936 {
3937 vm_map_entry_t entry;
3938 vm_map_t map = *var_map;
3939 vm_prot_t prot;
3940 vm_prot_t fault_type = fault_typea;
3941 vm_object_t eobject;
3942 vm_size_t size;
3943 struct ucred *cred;
3944
3945 RetryLookup:;
3946
3947 vm_map_lock_read(map);
3948
3949 /*
3950 * Lookup the faulting address.
3951 */
3952 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3953 vm_map_unlock_read(map);
3954 return (KERN_INVALID_ADDRESS);
3955 }
3956
3957 entry = *out_entry;
3958
3959 /*
3960 * Handle submaps.
3961 */
3962 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3963 vm_map_t old_map = map;
3964
3965 *var_map = map = entry->object.sub_map;
3966 vm_map_unlock_read(old_map);
3967 goto RetryLookup;
3968 }
3969
3970 /*
3971 * Check whether this task is allowed to have this page.
3972 */
3973 prot = entry->protection;
3974 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3975 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3976 vm_map_unlock_read(map);
3977 return (KERN_PROTECTION_FAILURE);
3978 }
3979 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
3980 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
3981 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
3982 ("entry %p flags %x", entry, entry->eflags));
3983 if ((fault_typea & VM_PROT_COPY) != 0 &&
3984 (entry->max_protection & VM_PROT_WRITE) == 0 &&
3985 (entry->eflags & MAP_ENTRY_COW) == 0) {
3986 vm_map_unlock_read(map);
3987 return (KERN_PROTECTION_FAILURE);
3988 }
3989
3990 /*
3991 * If this page is not pageable, we have to get it for all possible
3992 * accesses.
3993 */
3994 *wired = (entry->wired_count != 0);
3995 if (*wired)
3996 fault_type = entry->protection;
3997 size = entry->end - entry->start;
3998 /*
3999 * If the entry was copy-on-write, we either ...
4000 */
4001 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4002 /*
4003 * If we want to write the page, we may as well handle that
4004 * now since we've got the map locked.
4005 *
4006 * If we don't need to write the page, we just demote the
4007 * permissions allowed.
4008 */
4009 if ((fault_type & VM_PROT_WRITE) != 0 ||
4010 (fault_typea & VM_PROT_COPY) != 0) {
4011 /*
4012 * Make a new object, and place it in the object
4013 * chain. Note that no new references have appeared
4014 * -- one just moved from the map to the new
4015 * object.
4016 */
4017 if (vm_map_lock_upgrade(map))
4018 goto RetryLookup;
4019
4020 if (entry->cred == NULL) {
4021 /*
4022 * The debugger owner is charged for
4023 * the memory.
4024 */
4025 cred = curthread->td_ucred;
4026 crhold(cred);
4027 if (!swap_reserve_by_cred(size, cred)) {
4028 crfree(cred);
4029 vm_map_unlock(map);
4030 return (KERN_RESOURCE_SHORTAGE);
4031 }
4032 entry->cred = cred;
4033 }
4034 vm_object_shadow(&entry->object.vm_object,
4035 &entry->offset, size);
4036 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4037 eobject = entry->object.vm_object;
4038 if (eobject->cred != NULL) {
4039 /*
4040 * The object was not shadowed.
4041 */
4042 swap_release_by_cred(size, entry->cred);
4043 crfree(entry->cred);
4044 entry->cred = NULL;
4045 } else if (entry->cred != NULL) {
4046 VM_OBJECT_WLOCK(eobject);
4047 eobject->cred = entry->cred;
4048 eobject->charge = size;
4049 VM_OBJECT_WUNLOCK(eobject);
4050 entry->cred = NULL;
4051 }
4052
4053 vm_map_lock_downgrade(map);
4054 } else {
4055 /*
4056 * We're attempting to read a copy-on-write page --
4057 * don't allow writes.
4058 */
4059 prot &= ~VM_PROT_WRITE;
4060 }
4061 }
4062
4063 /*
4064 * Create an object if necessary.
4065 */
4066 if (entry->object.vm_object == NULL &&
4067 !map->system_map) {
4068 if (vm_map_lock_upgrade(map))
4069 goto RetryLookup;
4070 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4071 atop(size));
4072 entry->offset = 0;
4073 if (entry->cred != NULL) {
4074 VM_OBJECT_WLOCK(entry->object.vm_object);
4075 entry->object.vm_object->cred = entry->cred;
4076 entry->object.vm_object->charge = size;
4077 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4078 entry->cred = NULL;
4079 }
4080 vm_map_lock_downgrade(map);
4081 }
4082
4083 /*
4084 * Return the object/offset from this entry. If the entry was
4085 * copy-on-write or empty, it has been fixed up.
4086 */
4087 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4088 *object = entry->object.vm_object;
4089
4090 *out_prot = prot;
4091 return (KERN_SUCCESS);
4092 }
4093
4094 /*
4095 * vm_map_lookup_locked:
4096 *
4097 * Lookup the faulting address. A version of vm_map_lookup that returns
4098 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4099 */
4100 int
4101 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4102 vm_offset_t vaddr,
4103 vm_prot_t fault_typea,
4104 vm_map_entry_t *out_entry, /* OUT */
4105 vm_object_t *object, /* OUT */
4106 vm_pindex_t *pindex, /* OUT */
4107 vm_prot_t *out_prot, /* OUT */
4108 boolean_t *wired) /* OUT */
4109 {
4110 vm_map_entry_t entry;
4111 vm_map_t map = *var_map;
4112 vm_prot_t prot;
4113 vm_prot_t fault_type = fault_typea;
4114
4115 /*
4116 * Lookup the faulting address.
4117 */
4118 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4119 return (KERN_INVALID_ADDRESS);
4120
4121 entry = *out_entry;
4122
4123 /*
4124 * Fail if the entry refers to a submap.
4125 */
4126 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4127 return (KERN_FAILURE);
4128
4129 /*
4130 * Check whether this task is allowed to have this page.
4131 */
4132 prot = entry->protection;
4133 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4134 if ((fault_type & prot) != fault_type)
4135 return (KERN_PROTECTION_FAILURE);
4136
4137 /*
4138 * If this page is not pageable, we have to get it for all possible
4139 * accesses.
4140 */
4141 *wired = (entry->wired_count != 0);
4142 if (*wired)
4143 fault_type = entry->protection;
4144
4145 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4146 /*
4147 * Fail if the entry was copy-on-write for a write fault.
4148 */
4149 if (fault_type & VM_PROT_WRITE)
4150 return (KERN_FAILURE);
4151 /*
4152 * We're attempting to read a copy-on-write page --
4153 * don't allow writes.
4154 */
4155 prot &= ~VM_PROT_WRITE;
4156 }
4157
4158 /*
4159 * Fail if an object should be created.
4160 */
4161 if (entry->object.vm_object == NULL && !map->system_map)
4162 return (KERN_FAILURE);
4163
4164 /*
4165 * Return the object/offset from this entry. If the entry was
4166 * copy-on-write or empty, it has been fixed up.
4167 */
4168 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4169 *object = entry->object.vm_object;
4170
4171 *out_prot = prot;
4172 return (KERN_SUCCESS);
4173 }
4174
4175 /*
4176 * vm_map_lookup_done:
4177 *
4178 * Releases locks acquired by a vm_map_lookup
4179 * (according to the handle returned by that lookup).
4180 */
4181 void
4182 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4183 {
4184 /*
4185 * Unlock the main-level map
4186 */
4187 vm_map_unlock_read(map);
4188 }
4189
4190 #include "opt_ddb.h"
4191 #ifdef DDB
4192 #include <sys/kernel.h>
4193
4194 #include <ddb/ddb.h>
4195
4196 static void
4197 vm_map_print(vm_map_t map)
4198 {
4199 vm_map_entry_t entry;
4200
4201 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4202 (void *)map,
4203 (void *)map->pmap, map->nentries, map->timestamp);
4204
4205 db_indent += 2;
4206 for (entry = map->header.next; entry != &map->header;
4207 entry = entry->next) {
4208 db_iprintf("map entry %p: start=%p, end=%p\n",
4209 (void *)entry, (void *)entry->start, (void *)entry->end);
4210 {
4211 static char *inheritance_name[4] =
4212 {"share", "copy", "none", "donate_copy"};
4213
4214 db_iprintf(" prot=%x/%x/%s",
4215 entry->protection,
4216 entry->max_protection,
4217 inheritance_name[(int)(unsigned char)entry->inheritance]);
4218 if (entry->wired_count != 0)
4219 db_printf(", wired");
4220 }
4221 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4222 db_printf(", share=%p, offset=0x%jx\n",
4223 (void *)entry->object.sub_map,
4224 (uintmax_t)entry->offset);
4225 if ((entry->prev == &map->header) ||
4226 (entry->prev->object.sub_map !=
4227 entry->object.sub_map)) {
4228 db_indent += 2;
4229 vm_map_print((vm_map_t)entry->object.sub_map);
4230 db_indent -= 2;
4231 }
4232 } else {
4233 if (entry->cred != NULL)
4234 db_printf(", ruid %d", entry->cred->cr_ruid);
4235 db_printf(", object=%p, offset=0x%jx",
4236 (void *)entry->object.vm_object,
4237 (uintmax_t)entry->offset);
4238 if (entry->object.vm_object && entry->object.vm_object->cred)
4239 db_printf(", obj ruid %d charge %jx",
4240 entry->object.vm_object->cred->cr_ruid,
4241 (uintmax_t)entry->object.vm_object->charge);
4242 if (entry->eflags & MAP_ENTRY_COW)
4243 db_printf(", copy (%s)",
4244 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4245 db_printf("\n");
4246
4247 if ((entry->prev == &map->header) ||
4248 (entry->prev->object.vm_object !=
4249 entry->object.vm_object)) {
4250 db_indent += 2;
4251 vm_object_print((db_expr_t)(intptr_t)
4252 entry->object.vm_object,
4253 0, 0, (char *)0);
4254 db_indent -= 2;
4255 }
4256 }
4257 }
4258 db_indent -= 2;
4259 }
4260
4261 DB_SHOW_COMMAND(map, map)
4262 {
4263
4264 if (!have_addr) {
4265 db_printf("usage: show map <addr>\n");
4266 return;
4267 }
4268 vm_map_print((vm_map_t)addr);
4269 }
4270
4271 DB_SHOW_COMMAND(procvm, procvm)
4272 {
4273 struct proc *p;
4274
4275 if (have_addr) {
4276 p = (struct proc *) addr;
4277 } else {
4278 p = curproc;
4279 }
4280
4281 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4282 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4283 (void *)vmspace_pmap(p->p_vmspace));
4284
4285 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);
4286 }
4287
4288 #endif /* DDB */
Cache object: 03ed3e82f91ca89b330510512aac95de
|