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/10.3/sys/vm/vm_map.c 287846 2015-09-16 04:20:39Z kib $");
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 #ifdef RACCT
302 static void
303 vmspace_container_reset(struct proc *p)
304 {
305
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 }
314 #endif
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 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
346 "vmspace_free() called with non-sleepable lock held");
347
348 if (vm->vm_refcnt == 0)
349 panic("vmspace_free: attempt to free already freed vmspace");
350
351 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
352 vmspace_dofree(vm);
353 }
354
355 void
356 vmspace_exitfree(struct proc *p)
357 {
358 struct vmspace *vm;
359
360 PROC_VMSPACE_LOCK(p);
361 vm = p->p_vmspace;
362 p->p_vmspace = NULL;
363 PROC_VMSPACE_UNLOCK(p);
364 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
365 vmspace_free(vm);
366 }
367
368 void
369 vmspace_exit(struct thread *td)
370 {
371 int refcnt;
372 struct vmspace *vm;
373 struct proc *p;
374
375 /*
376 * Release user portion of address space.
377 * This releases references to vnodes,
378 * which could cause I/O if the file has been unlinked.
379 * Need to do this early enough that we can still sleep.
380 *
381 * The last exiting process to reach this point releases as
382 * much of the environment as it can. vmspace_dofree() is the
383 * slower fallback in case another process had a temporary
384 * reference to the vmspace.
385 */
386
387 p = td->td_proc;
388 vm = p->p_vmspace;
389 atomic_add_int(&vmspace0.vm_refcnt, 1);
390 do {
391 refcnt = vm->vm_refcnt;
392 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
393 /* Switch now since other proc might free vmspace */
394 PROC_VMSPACE_LOCK(p);
395 p->p_vmspace = &vmspace0;
396 PROC_VMSPACE_UNLOCK(p);
397 pmap_activate(td);
398 }
399 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
400 if (refcnt == 1) {
401 if (p->p_vmspace != vm) {
402 /* vmspace not yet freed, switch back */
403 PROC_VMSPACE_LOCK(p);
404 p->p_vmspace = vm;
405 PROC_VMSPACE_UNLOCK(p);
406 pmap_activate(td);
407 }
408 pmap_remove_pages(vmspace_pmap(vm));
409 /* Switch now since this proc will free vmspace */
410 PROC_VMSPACE_LOCK(p);
411 p->p_vmspace = &vmspace0;
412 PROC_VMSPACE_UNLOCK(p);
413 pmap_activate(td);
414 vmspace_dofree(vm);
415 }
416 #ifdef RACCT
417 if (racct_enable)
418 vmspace_container_reset(p);
419 #endif
420 }
421
422 /* Acquire reference to vmspace owned by another process. */
423
424 struct vmspace *
425 vmspace_acquire_ref(struct proc *p)
426 {
427 struct vmspace *vm;
428 int refcnt;
429
430 PROC_VMSPACE_LOCK(p);
431 vm = p->p_vmspace;
432 if (vm == NULL) {
433 PROC_VMSPACE_UNLOCK(p);
434 return (NULL);
435 }
436 do {
437 refcnt = vm->vm_refcnt;
438 if (refcnt <= 0) { /* Avoid 0->1 transition */
439 PROC_VMSPACE_UNLOCK(p);
440 return (NULL);
441 }
442 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
443 if (vm != p->p_vmspace) {
444 PROC_VMSPACE_UNLOCK(p);
445 vmspace_free(vm);
446 return (NULL);
447 }
448 PROC_VMSPACE_UNLOCK(p);
449 return (vm);
450 }
451
452 void
453 _vm_map_lock(vm_map_t map, const char *file, int line)
454 {
455
456 if (map->system_map)
457 mtx_lock_flags_(&map->system_mtx, 0, file, line);
458 else
459 sx_xlock_(&map->lock, file, line);
460 map->timestamp++;
461 }
462
463 static void
464 vm_map_process_deferred(void)
465 {
466 struct thread *td;
467 vm_map_entry_t entry, next;
468 vm_object_t object;
469
470 td = curthread;
471 entry = td->td_map_def_user;
472 td->td_map_def_user = NULL;
473 while (entry != NULL) {
474 next = entry->next;
475 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
476 /*
477 * Decrement the object's writemappings and
478 * possibly the vnode's v_writecount.
479 */
480 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
481 ("Submap with writecount"));
482 object = entry->object.vm_object;
483 KASSERT(object != NULL, ("No object for writecount"));
484 vnode_pager_release_writecount(object, entry->start,
485 entry->end);
486 }
487 vm_map_entry_deallocate(entry, FALSE);
488 entry = next;
489 }
490 }
491
492 void
493 _vm_map_unlock(vm_map_t map, const char *file, int line)
494 {
495
496 if (map->system_map)
497 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
498 else {
499 sx_xunlock_(&map->lock, file, line);
500 vm_map_process_deferred();
501 }
502 }
503
504 void
505 _vm_map_lock_read(vm_map_t map, const char *file, int line)
506 {
507
508 if (map->system_map)
509 mtx_lock_flags_(&map->system_mtx, 0, file, line);
510 else
511 sx_slock_(&map->lock, file, line);
512 }
513
514 void
515 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
516 {
517
518 if (map->system_map)
519 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
520 else {
521 sx_sunlock_(&map->lock, file, line);
522 vm_map_process_deferred();
523 }
524 }
525
526 int
527 _vm_map_trylock(vm_map_t map, const char *file, int line)
528 {
529 int error;
530
531 error = map->system_map ?
532 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
533 !sx_try_xlock_(&map->lock, file, line);
534 if (error == 0)
535 map->timestamp++;
536 return (error == 0);
537 }
538
539 int
540 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
541 {
542 int error;
543
544 error = map->system_map ?
545 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
546 !sx_try_slock_(&map->lock, file, line);
547 return (error == 0);
548 }
549
550 /*
551 * _vm_map_lock_upgrade: [ internal use only ]
552 *
553 * Tries to upgrade a read (shared) lock on the specified map to a write
554 * (exclusive) lock. Returns the value "" if the upgrade succeeds and a
555 * non-zero value if the upgrade fails. If the upgrade fails, the map is
556 * returned without a read or write lock held.
557 *
558 * Requires that the map be read locked.
559 */
560 int
561 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
562 {
563 unsigned int last_timestamp;
564
565 if (map->system_map) {
566 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
567 } else {
568 if (!sx_try_upgrade_(&map->lock, file, line)) {
569 last_timestamp = map->timestamp;
570 sx_sunlock_(&map->lock, file, line);
571 vm_map_process_deferred();
572 /*
573 * If the map's timestamp does not change while the
574 * map is unlocked, then the upgrade succeeds.
575 */
576 sx_xlock_(&map->lock, file, line);
577 if (last_timestamp != map->timestamp) {
578 sx_xunlock_(&map->lock, file, line);
579 return (1);
580 }
581 }
582 }
583 map->timestamp++;
584 return (0);
585 }
586
587 void
588 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
589 {
590
591 if (map->system_map) {
592 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
593 } else
594 sx_downgrade_(&map->lock, file, line);
595 }
596
597 /*
598 * vm_map_locked:
599 *
600 * Returns a non-zero value if the caller holds a write (exclusive) lock
601 * on the specified map and the value "" otherwise.
602 */
603 int
604 vm_map_locked(vm_map_t map)
605 {
606
607 if (map->system_map)
608 return (mtx_owned(&map->system_mtx));
609 else
610 return (sx_xlocked(&map->lock));
611 }
612
613 #ifdef INVARIANTS
614 static void
615 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
616 {
617
618 if (map->system_map)
619 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
620 else
621 sx_assert_(&map->lock, SA_XLOCKED, file, line);
622 }
623
624 #define VM_MAP_ASSERT_LOCKED(map) \
625 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
626 #else
627 #define VM_MAP_ASSERT_LOCKED(map)
628 #endif
629
630 /*
631 * _vm_map_unlock_and_wait:
632 *
633 * Atomically releases the lock on the specified map and puts the calling
634 * thread to sleep. The calling thread will remain asleep until either
635 * vm_map_wakeup() is performed on the map or the specified timeout is
636 * exceeded.
637 *
638 * WARNING! This function does not perform deferred deallocations of
639 * objects and map entries. Therefore, the calling thread is expected to
640 * reacquire the map lock after reawakening and later perform an ordinary
641 * unlock operation, such as vm_map_unlock(), before completing its
642 * operation on the map.
643 */
644 int
645 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
646 {
647
648 mtx_lock(&map_sleep_mtx);
649 if (map->system_map)
650 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
651 else
652 sx_xunlock_(&map->lock, file, line);
653 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
654 timo));
655 }
656
657 /*
658 * vm_map_wakeup:
659 *
660 * Awaken any threads that have slept on the map using
661 * vm_map_unlock_and_wait().
662 */
663 void
664 vm_map_wakeup(vm_map_t map)
665 {
666
667 /*
668 * Acquire and release map_sleep_mtx to prevent a wakeup()
669 * from being performed (and lost) between the map unlock
670 * and the msleep() in _vm_map_unlock_and_wait().
671 */
672 mtx_lock(&map_sleep_mtx);
673 mtx_unlock(&map_sleep_mtx);
674 wakeup(&map->root);
675 }
676
677 void
678 vm_map_busy(vm_map_t map)
679 {
680
681 VM_MAP_ASSERT_LOCKED(map);
682 map->busy++;
683 }
684
685 void
686 vm_map_unbusy(vm_map_t map)
687 {
688
689 VM_MAP_ASSERT_LOCKED(map);
690 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
691 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
692 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
693 wakeup(&map->busy);
694 }
695 }
696
697 void
698 vm_map_wait_busy(vm_map_t map)
699 {
700
701 VM_MAP_ASSERT_LOCKED(map);
702 while (map->busy) {
703 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
704 if (map->system_map)
705 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
706 else
707 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
708 }
709 map->timestamp++;
710 }
711
712 long
713 vmspace_resident_count(struct vmspace *vmspace)
714 {
715 return pmap_resident_count(vmspace_pmap(vmspace));
716 }
717
718 /*
719 * vm_map_create:
720 *
721 * Creates and returns a new empty VM map with
722 * the given physical map structure, and having
723 * the given lower and upper address bounds.
724 */
725 vm_map_t
726 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
727 {
728 vm_map_t result;
729
730 result = uma_zalloc(mapzone, M_WAITOK);
731 CTR1(KTR_VM, "vm_map_create: %p", result);
732 _vm_map_init(result, pmap, min, max);
733 return (result);
734 }
735
736 /*
737 * Initialize an existing vm_map structure
738 * such as that in the vmspace structure.
739 */
740 static void
741 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
742 {
743
744 map->header.next = map->header.prev = &map->header;
745 map->needs_wakeup = FALSE;
746 map->system_map = 0;
747 map->pmap = pmap;
748 map->min_offset = min;
749 map->max_offset = max;
750 map->flags = 0;
751 map->root = NULL;
752 map->timestamp = 0;
753 map->busy = 0;
754 }
755
756 void
757 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
758 {
759
760 _vm_map_init(map, pmap, min, max);
761 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
762 sx_init(&map->lock, "user map");
763 }
764
765 /*
766 * vm_map_entry_dispose: [ internal use only ]
767 *
768 * Inverse of vm_map_entry_create.
769 */
770 static void
771 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
772 {
773 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
774 }
775
776 /*
777 * vm_map_entry_create: [ internal use only ]
778 *
779 * Allocates a VM map entry for insertion.
780 * No entry fields are filled in.
781 */
782 static vm_map_entry_t
783 vm_map_entry_create(vm_map_t map)
784 {
785 vm_map_entry_t new_entry;
786
787 if (map->system_map)
788 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
789 else
790 new_entry = uma_zalloc(mapentzone, M_WAITOK);
791 if (new_entry == NULL)
792 panic("vm_map_entry_create: kernel resources exhausted");
793 return (new_entry);
794 }
795
796 /*
797 * vm_map_entry_set_behavior:
798 *
799 * Set the expected access behavior, either normal, random, or
800 * sequential.
801 */
802 static inline void
803 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
804 {
805 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
806 (behavior & MAP_ENTRY_BEHAV_MASK);
807 }
808
809 /*
810 * vm_map_entry_set_max_free:
811 *
812 * Set the max_free field in a vm_map_entry.
813 */
814 static inline void
815 vm_map_entry_set_max_free(vm_map_entry_t entry)
816 {
817
818 entry->max_free = entry->adj_free;
819 if (entry->left != NULL && entry->left->max_free > entry->max_free)
820 entry->max_free = entry->left->max_free;
821 if (entry->right != NULL && entry->right->max_free > entry->max_free)
822 entry->max_free = entry->right->max_free;
823 }
824
825 /*
826 * vm_map_entry_splay:
827 *
828 * The Sleator and Tarjan top-down splay algorithm with the
829 * following variation. Max_free must be computed bottom-up, so
830 * on the downward pass, maintain the left and right spines in
831 * reverse order. Then, make a second pass up each side to fix
832 * the pointers and compute max_free. The time bound is O(log n)
833 * amortized.
834 *
835 * The new root is the vm_map_entry containing "addr", or else an
836 * adjacent entry (lower or higher) if addr is not in the tree.
837 *
838 * The map must be locked, and leaves it so.
839 *
840 * Returns: the new root.
841 */
842 static vm_map_entry_t
843 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
844 {
845 vm_map_entry_t llist, rlist;
846 vm_map_entry_t ltree, rtree;
847 vm_map_entry_t y;
848
849 /* Special case of empty tree. */
850 if (root == NULL)
851 return (root);
852
853 /*
854 * Pass One: Splay down the tree until we find addr or a NULL
855 * pointer where addr would go. llist and rlist are the two
856 * sides in reverse order (bottom-up), with llist linked by
857 * the right pointer and rlist linked by the left pointer in
858 * the vm_map_entry. Wait until Pass Two to set max_free on
859 * the two spines.
860 */
861 llist = NULL;
862 rlist = NULL;
863 for (;;) {
864 /* root is never NULL in here. */
865 if (addr < root->start) {
866 y = root->left;
867 if (y == NULL)
868 break;
869 if (addr < y->start && y->left != NULL) {
870 /* Rotate right and put y on rlist. */
871 root->left = y->right;
872 y->right = root;
873 vm_map_entry_set_max_free(root);
874 root = y->left;
875 y->left = rlist;
876 rlist = y;
877 } else {
878 /* Put root on rlist. */
879 root->left = rlist;
880 rlist = root;
881 root = y;
882 }
883 } else if (addr >= root->end) {
884 y = root->right;
885 if (y == NULL)
886 break;
887 if (addr >= y->end && y->right != NULL) {
888 /* Rotate left and put y on llist. */
889 root->right = y->left;
890 y->left = root;
891 vm_map_entry_set_max_free(root);
892 root = y->right;
893 y->right = llist;
894 llist = y;
895 } else {
896 /* Put root on llist. */
897 root->right = llist;
898 llist = root;
899 root = y;
900 }
901 } else
902 break;
903 }
904
905 /*
906 * Pass Two: Walk back up the two spines, flip the pointers
907 * and set max_free. The subtrees of the root go at the
908 * bottom of llist and rlist.
909 */
910 ltree = root->left;
911 while (llist != NULL) {
912 y = llist->right;
913 llist->right = ltree;
914 vm_map_entry_set_max_free(llist);
915 ltree = llist;
916 llist = y;
917 }
918 rtree = root->right;
919 while (rlist != NULL) {
920 y = rlist->left;
921 rlist->left = rtree;
922 vm_map_entry_set_max_free(rlist);
923 rtree = rlist;
924 rlist = y;
925 }
926
927 /*
928 * Final assembly: add ltree and rtree as subtrees of root.
929 */
930 root->left = ltree;
931 root->right = rtree;
932 vm_map_entry_set_max_free(root);
933
934 return (root);
935 }
936
937 /*
938 * vm_map_entry_{un,}link:
939 *
940 * Insert/remove entries from maps.
941 */
942 static void
943 vm_map_entry_link(vm_map_t map,
944 vm_map_entry_t after_where,
945 vm_map_entry_t entry)
946 {
947
948 CTR4(KTR_VM,
949 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
950 map->nentries, entry, after_where);
951 VM_MAP_ASSERT_LOCKED(map);
952 KASSERT(after_where == &map->header ||
953 after_where->end <= entry->start,
954 ("vm_map_entry_link: prev end %jx new start %jx overlap",
955 (uintmax_t)after_where->end, (uintmax_t)entry->start));
956 KASSERT(after_where->next == &map->header ||
957 entry->end <= after_where->next->start,
958 ("vm_map_entry_link: new end %jx next start %jx overlap",
959 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
960
961 map->nentries++;
962 entry->prev = after_where;
963 entry->next = after_where->next;
964 entry->next->prev = entry;
965 after_where->next = entry;
966
967 if (after_where != &map->header) {
968 if (after_where != map->root)
969 vm_map_entry_splay(after_where->start, map->root);
970 entry->right = after_where->right;
971 entry->left = after_where;
972 after_where->right = NULL;
973 after_where->adj_free = entry->start - after_where->end;
974 vm_map_entry_set_max_free(after_where);
975 } else {
976 entry->right = map->root;
977 entry->left = NULL;
978 }
979 entry->adj_free = (entry->next == &map->header ? map->max_offset :
980 entry->next->start) - entry->end;
981 vm_map_entry_set_max_free(entry);
982 map->root = entry;
983 }
984
985 static void
986 vm_map_entry_unlink(vm_map_t map,
987 vm_map_entry_t entry)
988 {
989 vm_map_entry_t next, prev, root;
990
991 VM_MAP_ASSERT_LOCKED(map);
992 if (entry != map->root)
993 vm_map_entry_splay(entry->start, map->root);
994 if (entry->left == NULL)
995 root = entry->right;
996 else {
997 root = vm_map_entry_splay(entry->start, entry->left);
998 root->right = entry->right;
999 root->adj_free = (entry->next == &map->header ? map->max_offset :
1000 entry->next->start) - root->end;
1001 vm_map_entry_set_max_free(root);
1002 }
1003 map->root = root;
1004
1005 prev = entry->prev;
1006 next = entry->next;
1007 next->prev = prev;
1008 prev->next = next;
1009 map->nentries--;
1010 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1011 map->nentries, entry);
1012 }
1013
1014 /*
1015 * vm_map_entry_resize_free:
1016 *
1017 * Recompute the amount of free space following a vm_map_entry
1018 * and propagate that value up the tree. Call this function after
1019 * resizing a map entry in-place, that is, without a call to
1020 * vm_map_entry_link() or _unlink().
1021 *
1022 * The map must be locked, and leaves it so.
1023 */
1024 static void
1025 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1026 {
1027
1028 /*
1029 * Using splay trees without parent pointers, propagating
1030 * max_free up the tree is done by moving the entry to the
1031 * root and making the change there.
1032 */
1033 if (entry != map->root)
1034 map->root = vm_map_entry_splay(entry->start, map->root);
1035
1036 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1037 entry->next->start) - entry->end;
1038 vm_map_entry_set_max_free(entry);
1039 }
1040
1041 /*
1042 * vm_map_lookup_entry: [ internal use only ]
1043 *
1044 * Finds the map entry containing (or
1045 * immediately preceding) the specified address
1046 * in the given map; the entry is returned
1047 * in the "entry" parameter. The boolean
1048 * result indicates whether the address is
1049 * actually contained in the map.
1050 */
1051 boolean_t
1052 vm_map_lookup_entry(
1053 vm_map_t map,
1054 vm_offset_t address,
1055 vm_map_entry_t *entry) /* OUT */
1056 {
1057 vm_map_entry_t cur;
1058 boolean_t locked;
1059
1060 /*
1061 * If the map is empty, then the map entry immediately preceding
1062 * "address" is the map's header.
1063 */
1064 cur = map->root;
1065 if (cur == NULL)
1066 *entry = &map->header;
1067 else if (address >= cur->start && cur->end > address) {
1068 *entry = cur;
1069 return (TRUE);
1070 } else if ((locked = vm_map_locked(map)) ||
1071 sx_try_upgrade(&map->lock)) {
1072 /*
1073 * Splay requires a write lock on the map. However, it only
1074 * restructures the binary search tree; it does not otherwise
1075 * change the map. Thus, the map's timestamp need not change
1076 * on a temporary upgrade.
1077 */
1078 map->root = cur = vm_map_entry_splay(address, cur);
1079 if (!locked)
1080 sx_downgrade(&map->lock);
1081
1082 /*
1083 * If "address" is contained within a map entry, the new root
1084 * is that map entry. Otherwise, the new root is a map entry
1085 * immediately before or after "address".
1086 */
1087 if (address >= cur->start) {
1088 *entry = cur;
1089 if (cur->end > address)
1090 return (TRUE);
1091 } else
1092 *entry = cur->prev;
1093 } else
1094 /*
1095 * Since the map is only locked for read access, perform a
1096 * standard binary search tree lookup for "address".
1097 */
1098 for (;;) {
1099 if (address < cur->start) {
1100 if (cur->left == NULL) {
1101 *entry = cur->prev;
1102 break;
1103 }
1104 cur = cur->left;
1105 } else if (cur->end > address) {
1106 *entry = cur;
1107 return (TRUE);
1108 } else {
1109 if (cur->right == NULL) {
1110 *entry = cur;
1111 break;
1112 }
1113 cur = cur->right;
1114 }
1115 }
1116 return (FALSE);
1117 }
1118
1119 /*
1120 * vm_map_insert:
1121 *
1122 * Inserts the given whole VM object into the target
1123 * map at the specified address range. The object's
1124 * size should match that of the address range.
1125 *
1126 * Requires that the map be locked, and leaves it so.
1127 *
1128 * If object is non-NULL, ref count must be bumped by caller
1129 * prior to making call to account for the new entry.
1130 */
1131 int
1132 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1133 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
1134 int cow)
1135 {
1136 vm_map_entry_t new_entry;
1137 vm_map_entry_t prev_entry;
1138 vm_map_entry_t temp_entry;
1139 vm_eflags_t protoeflags;
1140 struct ucred *cred;
1141 vm_inherit_t inheritance;
1142 boolean_t charge_prev_obj;
1143
1144 VM_MAP_ASSERT_LOCKED(map);
1145
1146 /*
1147 * Check that the start and end points are not bogus.
1148 */
1149 if ((start < map->min_offset) || (end > map->max_offset) ||
1150 (start >= end))
1151 return (KERN_INVALID_ADDRESS);
1152
1153 /*
1154 * Find the entry prior to the proposed starting address; if it's part
1155 * of an existing entry, this range is bogus.
1156 */
1157 if (vm_map_lookup_entry(map, start, &temp_entry))
1158 return (KERN_NO_SPACE);
1159
1160 prev_entry = temp_entry;
1161
1162 /*
1163 * Assert that the next entry doesn't overlap the end point.
1164 */
1165 if ((prev_entry->next != &map->header) &&
1166 (prev_entry->next->start < end))
1167 return (KERN_NO_SPACE);
1168
1169 protoeflags = 0;
1170 charge_prev_obj = FALSE;
1171
1172 if (cow & MAP_COPY_ON_WRITE)
1173 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
1174
1175 if (cow & MAP_NOFAULT) {
1176 protoeflags |= MAP_ENTRY_NOFAULT;
1177
1178 KASSERT(object == NULL,
1179 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1180 }
1181 if (cow & MAP_DISABLE_SYNCER)
1182 protoeflags |= MAP_ENTRY_NOSYNC;
1183 if (cow & MAP_DISABLE_COREDUMP)
1184 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1185 if (cow & MAP_VN_WRITECOUNT)
1186 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1187 if (cow & MAP_INHERIT_SHARE)
1188 inheritance = VM_INHERIT_SHARE;
1189 else
1190 inheritance = VM_INHERIT_DEFAULT;
1191
1192 cred = NULL;
1193 KASSERT((object != kmem_object && object != kernel_object) ||
1194 ((object == kmem_object || object == kernel_object) &&
1195 !(protoeflags & MAP_ENTRY_NEEDS_COPY)),
1196 ("kmem or kernel object and cow"));
1197 if (cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT))
1198 goto charged;
1199 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1200 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1201 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1202 return (KERN_RESOURCE_SHORTAGE);
1203 KASSERT(object == NULL || (protoeflags & MAP_ENTRY_NEEDS_COPY) ||
1204 object->cred == NULL,
1205 ("OVERCOMMIT: vm_map_insert o %p", object));
1206 cred = curthread->td_ucred;
1207 crhold(cred);
1208 if (object == NULL && !(protoeflags & MAP_ENTRY_NEEDS_COPY))
1209 charge_prev_obj = TRUE;
1210 }
1211
1212 charged:
1213 /* Expand the kernel pmap, if necessary. */
1214 if (map == kernel_map && end > kernel_vm_end)
1215 pmap_growkernel(end);
1216 if (object != NULL) {
1217 /*
1218 * OBJ_ONEMAPPING must be cleared unless this mapping
1219 * is trivially proven to be the only mapping for any
1220 * of the object's pages. (Object granularity
1221 * reference counting is insufficient to recognize
1222 * aliases with precision.)
1223 */
1224 VM_OBJECT_WLOCK(object);
1225 if (object->ref_count > 1 || object->shadow_count != 0)
1226 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1227 VM_OBJECT_WUNLOCK(object);
1228 }
1229 else if ((prev_entry != &map->header) &&
1230 (prev_entry->eflags == protoeflags) &&
1231 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1232 (prev_entry->end == start) &&
1233 (prev_entry->wired_count == 0) &&
1234 (prev_entry->cred == cred ||
1235 (prev_entry->object.vm_object != NULL &&
1236 (prev_entry->object.vm_object->cred == cred))) &&
1237 vm_object_coalesce(prev_entry->object.vm_object,
1238 prev_entry->offset,
1239 (vm_size_t)(prev_entry->end - prev_entry->start),
1240 (vm_size_t)(end - prev_entry->end), charge_prev_obj)) {
1241 /*
1242 * We were able to extend the object. Determine if we
1243 * can extend the previous map entry to include the
1244 * new range as well.
1245 */
1246 if ((prev_entry->inheritance == inheritance) &&
1247 (prev_entry->protection == prot) &&
1248 (prev_entry->max_protection == max)) {
1249 map->size += (end - prev_entry->end);
1250 prev_entry->end = end;
1251 vm_map_entry_resize_free(map, prev_entry);
1252 vm_map_simplify_entry(map, prev_entry);
1253 if (cred != NULL)
1254 crfree(cred);
1255 return (KERN_SUCCESS);
1256 }
1257
1258 /*
1259 * If we can extend the object but cannot extend the
1260 * map entry, we have to create a new map entry. We
1261 * must bump the ref count on the extended object to
1262 * account for it. object may be NULL.
1263 */
1264 object = prev_entry->object.vm_object;
1265 offset = prev_entry->offset +
1266 (prev_entry->end - prev_entry->start);
1267 vm_object_reference(object);
1268 if (cred != NULL && object != NULL && object->cred != NULL &&
1269 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1270 /* Object already accounts for this uid. */
1271 crfree(cred);
1272 cred = NULL;
1273 }
1274 }
1275
1276 /*
1277 * NOTE: if conditionals fail, object can be NULL here. This occurs
1278 * in things like the buffer map where we manage kva but do not manage
1279 * backing objects.
1280 */
1281
1282 /*
1283 * Create a new entry
1284 */
1285 new_entry = vm_map_entry_create(map);
1286 new_entry->start = start;
1287 new_entry->end = end;
1288 new_entry->cred = NULL;
1289
1290 new_entry->eflags = protoeflags;
1291 new_entry->object.vm_object = object;
1292 new_entry->offset = offset;
1293 new_entry->avail_ssize = 0;
1294
1295 new_entry->inheritance = inheritance;
1296 new_entry->protection = prot;
1297 new_entry->max_protection = max;
1298 new_entry->wired_count = 0;
1299 new_entry->wiring_thread = NULL;
1300 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1301 new_entry->next_read = OFF_TO_IDX(offset);
1302
1303 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1304 ("OVERCOMMIT: vm_map_insert leaks vm_map %p", new_entry));
1305 new_entry->cred = cred;
1306
1307 /*
1308 * Insert the new entry into the list
1309 */
1310 vm_map_entry_link(map, prev_entry, new_entry);
1311 map->size += new_entry->end - new_entry->start;
1312
1313 /*
1314 * It may be possible to merge the new entry with the next and/or
1315 * previous entries. However, due to MAP_STACK_* being a hack, a
1316 * panic can result from merging such entries.
1317 */
1318 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0)
1319 vm_map_simplify_entry(map, new_entry);
1320
1321 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
1322 vm_map_pmap_enter(map, start, prot,
1323 object, OFF_TO_IDX(offset), end - start,
1324 cow & MAP_PREFAULT_PARTIAL);
1325 }
1326
1327 return (KERN_SUCCESS);
1328 }
1329
1330 /*
1331 * vm_map_findspace:
1332 *
1333 * Find the first fit (lowest VM address) for "length" free bytes
1334 * beginning at address >= start in the given map.
1335 *
1336 * In a vm_map_entry, "adj_free" is the amount of free space
1337 * adjacent (higher address) to this entry, and "max_free" is the
1338 * maximum amount of contiguous free space in its subtree. This
1339 * allows finding a free region in one path down the tree, so
1340 * O(log n) amortized with splay trees.
1341 *
1342 * The map must be locked, and leaves it so.
1343 *
1344 * Returns: 0 on success, and starting address in *addr,
1345 * 1 if insufficient space.
1346 */
1347 int
1348 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1349 vm_offset_t *addr) /* OUT */
1350 {
1351 vm_map_entry_t entry;
1352 vm_offset_t st;
1353
1354 /*
1355 * Request must fit within min/max VM address and must avoid
1356 * address wrap.
1357 */
1358 if (start < map->min_offset)
1359 start = map->min_offset;
1360 if (start + length > map->max_offset || start + length < start)
1361 return (1);
1362
1363 /* Empty tree means wide open address space. */
1364 if (map->root == NULL) {
1365 *addr = start;
1366 return (0);
1367 }
1368
1369 /*
1370 * After splay, if start comes before root node, then there
1371 * must be a gap from start to the root.
1372 */
1373 map->root = vm_map_entry_splay(start, map->root);
1374 if (start + length <= map->root->start) {
1375 *addr = start;
1376 return (0);
1377 }
1378
1379 /*
1380 * Root is the last node that might begin its gap before
1381 * start, and this is the last comparison where address
1382 * wrap might be a problem.
1383 */
1384 st = (start > map->root->end) ? start : map->root->end;
1385 if (length <= map->root->end + map->root->adj_free - st) {
1386 *addr = st;
1387 return (0);
1388 }
1389
1390 /* With max_free, can immediately tell if no solution. */
1391 entry = map->root->right;
1392 if (entry == NULL || length > entry->max_free)
1393 return (1);
1394
1395 /*
1396 * Search the right subtree in the order: left subtree, root,
1397 * right subtree (first fit). The previous splay implies that
1398 * all regions in the right subtree have addresses > start.
1399 */
1400 while (entry != NULL) {
1401 if (entry->left != NULL && entry->left->max_free >= length)
1402 entry = entry->left;
1403 else if (entry->adj_free >= length) {
1404 *addr = entry->end;
1405 return (0);
1406 } else
1407 entry = entry->right;
1408 }
1409
1410 /* Can't get here, so panic if we do. */
1411 panic("vm_map_findspace: max_free corrupt");
1412 }
1413
1414 int
1415 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1416 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1417 vm_prot_t max, int cow)
1418 {
1419 vm_offset_t end;
1420 int result;
1421
1422 end = start + length;
1423 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1424 object == NULL,
1425 ("vm_map_fixed: non-NULL backing object for stack"));
1426 vm_map_lock(map);
1427 VM_MAP_RANGE_CHECK(map, start, end);
1428 if ((cow & MAP_CHECK_EXCL) == 0)
1429 vm_map_delete(map, start, end);
1430 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1431 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1432 prot, max, cow);
1433 } else {
1434 result = vm_map_insert(map, object, offset, start, end,
1435 prot, max, cow);
1436 }
1437 vm_map_unlock(map);
1438 return (result);
1439 }
1440
1441 /*
1442 * vm_map_find finds an unallocated region in the target address
1443 * map with the given length. The search is defined to be
1444 * first-fit from the specified address; the region found is
1445 * returned in the same parameter.
1446 *
1447 * If object is non-NULL, ref count must be bumped by caller
1448 * prior to making call to account for the new entry.
1449 */
1450 int
1451 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1452 vm_offset_t *addr, /* IN/OUT */
1453 vm_size_t length, vm_offset_t max_addr, int find_space,
1454 vm_prot_t prot, vm_prot_t max, int cow)
1455 {
1456 vm_offset_t alignment, initial_addr, start;
1457 int result;
1458
1459 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1460 object == NULL,
1461 ("vm_map_find: non-NULL backing object for stack"));
1462 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1463 (object->flags & OBJ_COLORED) == 0))
1464 find_space = VMFS_ANY_SPACE;
1465 if (find_space >> 8 != 0) {
1466 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1467 alignment = (vm_offset_t)1 << (find_space >> 8);
1468 } else
1469 alignment = 0;
1470 initial_addr = *addr;
1471 again:
1472 start = initial_addr;
1473 vm_map_lock(map);
1474 do {
1475 if (find_space != VMFS_NO_SPACE) {
1476 if (vm_map_findspace(map, start, length, addr) ||
1477 (max_addr != 0 && *addr + length > max_addr)) {
1478 vm_map_unlock(map);
1479 if (find_space == VMFS_OPTIMAL_SPACE) {
1480 find_space = VMFS_ANY_SPACE;
1481 goto again;
1482 }
1483 return (KERN_NO_SPACE);
1484 }
1485 switch (find_space) {
1486 case VMFS_SUPER_SPACE:
1487 case VMFS_OPTIMAL_SPACE:
1488 pmap_align_superpage(object, offset, addr,
1489 length);
1490 break;
1491 case VMFS_ANY_SPACE:
1492 break;
1493 default:
1494 if ((*addr & (alignment - 1)) != 0) {
1495 *addr &= ~(alignment - 1);
1496 *addr += alignment;
1497 }
1498 break;
1499 }
1500
1501 start = *addr;
1502 }
1503 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1504 result = vm_map_stack_locked(map, start, length,
1505 sgrowsiz, prot, max, cow);
1506 } else {
1507 result = vm_map_insert(map, object, offset, start,
1508 start + length, prot, max, cow);
1509 }
1510 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1511 find_space != VMFS_ANY_SPACE);
1512 vm_map_unlock(map);
1513 return (result);
1514 }
1515
1516 /*
1517 * vm_map_simplify_entry:
1518 *
1519 * Simplify the given map entry by merging with either neighbor. This
1520 * routine also has the ability to merge with both neighbors.
1521 *
1522 * The map must be locked.
1523 *
1524 * This routine guarentees that the passed entry remains valid (though
1525 * possibly extended). When merging, this routine may delete one or
1526 * both neighbors.
1527 */
1528 void
1529 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1530 {
1531 vm_map_entry_t next, prev;
1532 vm_size_t prevsize, esize;
1533
1534 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1535 return;
1536
1537 prev = entry->prev;
1538 if (prev != &map->header) {
1539 prevsize = prev->end - prev->start;
1540 if ( (prev->end == entry->start) &&
1541 (prev->object.vm_object == entry->object.vm_object) &&
1542 (!prev->object.vm_object ||
1543 (prev->offset + prevsize == entry->offset)) &&
1544 (prev->eflags == entry->eflags) &&
1545 (prev->protection == entry->protection) &&
1546 (prev->max_protection == entry->max_protection) &&
1547 (prev->inheritance == entry->inheritance) &&
1548 (prev->wired_count == entry->wired_count) &&
1549 (prev->cred == entry->cred)) {
1550 vm_map_entry_unlink(map, prev);
1551 entry->start = prev->start;
1552 entry->offset = prev->offset;
1553 if (entry->prev != &map->header)
1554 vm_map_entry_resize_free(map, entry->prev);
1555
1556 /*
1557 * If the backing object is a vnode object,
1558 * vm_object_deallocate() calls vrele().
1559 * However, vrele() does not lock the vnode
1560 * because the vnode has additional
1561 * references. Thus, the map lock can be kept
1562 * without causing a lock-order reversal with
1563 * the vnode lock.
1564 *
1565 * Since we count the number of virtual page
1566 * mappings in object->un_pager.vnp.writemappings,
1567 * the writemappings value should not be adjusted
1568 * when the entry is disposed of.
1569 */
1570 if (prev->object.vm_object)
1571 vm_object_deallocate(prev->object.vm_object);
1572 if (prev->cred != NULL)
1573 crfree(prev->cred);
1574 vm_map_entry_dispose(map, prev);
1575 }
1576 }
1577
1578 next = entry->next;
1579 if (next != &map->header) {
1580 esize = entry->end - entry->start;
1581 if ((entry->end == next->start) &&
1582 (next->object.vm_object == entry->object.vm_object) &&
1583 (!entry->object.vm_object ||
1584 (entry->offset + esize == next->offset)) &&
1585 (next->eflags == entry->eflags) &&
1586 (next->protection == entry->protection) &&
1587 (next->max_protection == entry->max_protection) &&
1588 (next->inheritance == entry->inheritance) &&
1589 (next->wired_count == entry->wired_count) &&
1590 (next->cred == entry->cred)) {
1591 vm_map_entry_unlink(map, next);
1592 entry->end = next->end;
1593 vm_map_entry_resize_free(map, entry);
1594
1595 /*
1596 * See comment above.
1597 */
1598 if (next->object.vm_object)
1599 vm_object_deallocate(next->object.vm_object);
1600 if (next->cred != NULL)
1601 crfree(next->cred);
1602 vm_map_entry_dispose(map, next);
1603 }
1604 }
1605 }
1606 /*
1607 * vm_map_clip_start: [ internal use only ]
1608 *
1609 * Asserts that the given entry begins at or after
1610 * the specified address; if necessary,
1611 * it splits the entry into two.
1612 */
1613 #define vm_map_clip_start(map, entry, startaddr) \
1614 { \
1615 if (startaddr > entry->start) \
1616 _vm_map_clip_start(map, entry, startaddr); \
1617 }
1618
1619 /*
1620 * This routine is called only when it is known that
1621 * the entry must be split.
1622 */
1623 static void
1624 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1625 {
1626 vm_map_entry_t new_entry;
1627
1628 VM_MAP_ASSERT_LOCKED(map);
1629
1630 /*
1631 * Split off the front portion -- note that we must insert the new
1632 * entry BEFORE this one, so that this entry has the specified
1633 * starting address.
1634 */
1635 vm_map_simplify_entry(map, entry);
1636
1637 /*
1638 * If there is no object backing this entry, we might as well create
1639 * one now. If we defer it, an object can get created after the map
1640 * is clipped, and individual objects will be created for the split-up
1641 * map. This is a bit of a hack, but is also about the best place to
1642 * put this improvement.
1643 */
1644 if (entry->object.vm_object == NULL && !map->system_map) {
1645 vm_object_t object;
1646 object = vm_object_allocate(OBJT_DEFAULT,
1647 atop(entry->end - entry->start));
1648 entry->object.vm_object = object;
1649 entry->offset = 0;
1650 if (entry->cred != NULL) {
1651 object->cred = entry->cred;
1652 object->charge = entry->end - entry->start;
1653 entry->cred = NULL;
1654 }
1655 } else if (entry->object.vm_object != NULL &&
1656 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1657 entry->cred != NULL) {
1658 VM_OBJECT_WLOCK(entry->object.vm_object);
1659 KASSERT(entry->object.vm_object->cred == NULL,
1660 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1661 entry->object.vm_object->cred = entry->cred;
1662 entry->object.vm_object->charge = entry->end - entry->start;
1663 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1664 entry->cred = NULL;
1665 }
1666
1667 new_entry = vm_map_entry_create(map);
1668 *new_entry = *entry;
1669
1670 new_entry->end = start;
1671 entry->offset += (start - entry->start);
1672 entry->start = start;
1673 if (new_entry->cred != NULL)
1674 crhold(entry->cred);
1675
1676 vm_map_entry_link(map, entry->prev, new_entry);
1677
1678 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1679 vm_object_reference(new_entry->object.vm_object);
1680 /*
1681 * The object->un_pager.vnp.writemappings for the
1682 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1683 * kept as is here. The virtual pages are
1684 * re-distributed among the clipped entries, so the sum is
1685 * left the same.
1686 */
1687 }
1688 }
1689
1690 /*
1691 * vm_map_clip_end: [ internal use only ]
1692 *
1693 * Asserts that the given entry ends at or before
1694 * the specified address; if necessary,
1695 * it splits the entry into two.
1696 */
1697 #define vm_map_clip_end(map, entry, endaddr) \
1698 { \
1699 if ((endaddr) < (entry->end)) \
1700 _vm_map_clip_end((map), (entry), (endaddr)); \
1701 }
1702
1703 /*
1704 * This routine is called only when it is known that
1705 * the entry must be split.
1706 */
1707 static void
1708 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1709 {
1710 vm_map_entry_t new_entry;
1711
1712 VM_MAP_ASSERT_LOCKED(map);
1713
1714 /*
1715 * If there is no object backing this entry, we might as well create
1716 * one now. If we defer it, an object can get created after the map
1717 * is clipped, and individual objects will be created for the split-up
1718 * map. This is a bit of a hack, but is also about the best place to
1719 * put this improvement.
1720 */
1721 if (entry->object.vm_object == NULL && !map->system_map) {
1722 vm_object_t object;
1723 object = vm_object_allocate(OBJT_DEFAULT,
1724 atop(entry->end - entry->start));
1725 entry->object.vm_object = object;
1726 entry->offset = 0;
1727 if (entry->cred != NULL) {
1728 object->cred = entry->cred;
1729 object->charge = entry->end - entry->start;
1730 entry->cred = NULL;
1731 }
1732 } else if (entry->object.vm_object != NULL &&
1733 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1734 entry->cred != NULL) {
1735 VM_OBJECT_WLOCK(entry->object.vm_object);
1736 KASSERT(entry->object.vm_object->cred == NULL,
1737 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1738 entry->object.vm_object->cred = entry->cred;
1739 entry->object.vm_object->charge = entry->end - entry->start;
1740 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1741 entry->cred = NULL;
1742 }
1743
1744 /*
1745 * Create a new entry and insert it AFTER the specified entry
1746 */
1747 new_entry = vm_map_entry_create(map);
1748 *new_entry = *entry;
1749
1750 new_entry->start = entry->end = end;
1751 new_entry->offset += (end - entry->start);
1752 if (new_entry->cred != NULL)
1753 crhold(entry->cred);
1754
1755 vm_map_entry_link(map, entry, new_entry);
1756
1757 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1758 vm_object_reference(new_entry->object.vm_object);
1759 }
1760 }
1761
1762 /*
1763 * vm_map_submap: [ kernel use only ]
1764 *
1765 * Mark the given range as handled by a subordinate map.
1766 *
1767 * This range must have been created with vm_map_find,
1768 * and no other operations may have been performed on this
1769 * range prior to calling vm_map_submap.
1770 *
1771 * Only a limited number of operations can be performed
1772 * within this rage after calling vm_map_submap:
1773 * vm_fault
1774 * [Don't try vm_map_copy!]
1775 *
1776 * To remove a submapping, one must first remove the
1777 * range from the superior map, and then destroy the
1778 * submap (if desired). [Better yet, don't try it.]
1779 */
1780 int
1781 vm_map_submap(
1782 vm_map_t map,
1783 vm_offset_t start,
1784 vm_offset_t end,
1785 vm_map_t submap)
1786 {
1787 vm_map_entry_t entry;
1788 int result = KERN_INVALID_ARGUMENT;
1789
1790 vm_map_lock(map);
1791
1792 VM_MAP_RANGE_CHECK(map, start, end);
1793
1794 if (vm_map_lookup_entry(map, start, &entry)) {
1795 vm_map_clip_start(map, entry, start);
1796 } else
1797 entry = entry->next;
1798
1799 vm_map_clip_end(map, entry, end);
1800
1801 if ((entry->start == start) && (entry->end == end) &&
1802 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1803 (entry->object.vm_object == NULL)) {
1804 entry->object.sub_map = submap;
1805 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1806 result = KERN_SUCCESS;
1807 }
1808 vm_map_unlock(map);
1809
1810 return (result);
1811 }
1812
1813 /*
1814 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1815 */
1816 #define MAX_INIT_PT 96
1817
1818 /*
1819 * vm_map_pmap_enter:
1820 *
1821 * Preload the specified map's pmap with mappings to the specified
1822 * object's memory-resident pages. No further physical pages are
1823 * allocated, and no further virtual pages are retrieved from secondary
1824 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1825 * limited number of page mappings are created at the low-end of the
1826 * specified address range. (For this purpose, a superpage mapping
1827 * counts as one page mapping.) Otherwise, all resident pages within
1828 * the specified address range are mapped. Because these mappings are
1829 * being created speculatively, cached pages are not reactivated and
1830 * mapped.
1831 */
1832 void
1833 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1834 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1835 {
1836 vm_offset_t start;
1837 vm_page_t p, p_start;
1838 vm_pindex_t mask, psize, threshold, tmpidx;
1839
1840 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1841 return;
1842 VM_OBJECT_RLOCK(object);
1843 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1844 VM_OBJECT_RUNLOCK(object);
1845 VM_OBJECT_WLOCK(object);
1846 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1847 pmap_object_init_pt(map->pmap, addr, object, pindex,
1848 size);
1849 VM_OBJECT_WUNLOCK(object);
1850 return;
1851 }
1852 VM_OBJECT_LOCK_DOWNGRADE(object);
1853 }
1854
1855 psize = atop(size);
1856 if (psize + pindex > object->size) {
1857 if (object->size < pindex) {
1858 VM_OBJECT_RUNLOCK(object);
1859 return;
1860 }
1861 psize = object->size - pindex;
1862 }
1863
1864 start = 0;
1865 p_start = NULL;
1866 threshold = MAX_INIT_PT;
1867
1868 p = vm_page_find_least(object, pindex);
1869 /*
1870 * Assert: the variable p is either (1) the page with the
1871 * least pindex greater than or equal to the parameter pindex
1872 * or (2) NULL.
1873 */
1874 for (;
1875 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1876 p = TAILQ_NEXT(p, listq)) {
1877 /*
1878 * don't allow an madvise to blow away our really
1879 * free pages allocating pv entries.
1880 */
1881 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1882 cnt.v_free_count < cnt.v_free_reserved) ||
1883 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1884 tmpidx >= threshold)) {
1885 psize = tmpidx;
1886 break;
1887 }
1888 if (p->valid == VM_PAGE_BITS_ALL) {
1889 if (p_start == NULL) {
1890 start = addr + ptoa(tmpidx);
1891 p_start = p;
1892 }
1893 /* Jump ahead if a superpage mapping is possible. */
1894 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1895 (pagesizes[p->psind] - 1)) == 0) {
1896 mask = atop(pagesizes[p->psind]) - 1;
1897 if (tmpidx + mask < psize &&
1898 vm_page_ps_is_valid(p)) {
1899 p += mask;
1900 threshold += mask;
1901 }
1902 }
1903 } else if (p_start != NULL) {
1904 pmap_enter_object(map->pmap, start, addr +
1905 ptoa(tmpidx), p_start, prot);
1906 p_start = NULL;
1907 }
1908 }
1909 if (p_start != NULL)
1910 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1911 p_start, prot);
1912 VM_OBJECT_RUNLOCK(object);
1913 }
1914
1915 /*
1916 * vm_map_protect:
1917 *
1918 * Sets the protection of the specified address
1919 * region in the target map. If "set_max" is
1920 * specified, the maximum protection is to be set;
1921 * otherwise, only the current protection is affected.
1922 */
1923 int
1924 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1925 vm_prot_t new_prot, boolean_t set_max)
1926 {
1927 vm_map_entry_t current, entry;
1928 vm_object_t obj;
1929 struct ucred *cred;
1930 vm_prot_t old_prot;
1931
1932 if (start == end)
1933 return (KERN_SUCCESS);
1934
1935 vm_map_lock(map);
1936
1937 VM_MAP_RANGE_CHECK(map, start, end);
1938
1939 if (vm_map_lookup_entry(map, start, &entry)) {
1940 vm_map_clip_start(map, entry, start);
1941 } else {
1942 entry = entry->next;
1943 }
1944
1945 /*
1946 * Make a first pass to check for protection violations.
1947 */
1948 current = entry;
1949 while ((current != &map->header) && (current->start < end)) {
1950 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1951 vm_map_unlock(map);
1952 return (KERN_INVALID_ARGUMENT);
1953 }
1954 if ((new_prot & current->max_protection) != new_prot) {
1955 vm_map_unlock(map);
1956 return (KERN_PROTECTION_FAILURE);
1957 }
1958 current = current->next;
1959 }
1960
1961
1962 /*
1963 * Do an accounting pass for private read-only mappings that
1964 * now will do cow due to allowed write (e.g. debugger sets
1965 * breakpoint on text segment)
1966 */
1967 for (current = entry; (current != &map->header) &&
1968 (current->start < end); current = current->next) {
1969
1970 vm_map_clip_end(map, current, end);
1971
1972 if (set_max ||
1973 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
1974 ENTRY_CHARGED(current)) {
1975 continue;
1976 }
1977
1978 cred = curthread->td_ucred;
1979 obj = current->object.vm_object;
1980
1981 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
1982 if (!swap_reserve(current->end - current->start)) {
1983 vm_map_unlock(map);
1984 return (KERN_RESOURCE_SHORTAGE);
1985 }
1986 crhold(cred);
1987 current->cred = cred;
1988 continue;
1989 }
1990
1991 VM_OBJECT_WLOCK(obj);
1992 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
1993 VM_OBJECT_WUNLOCK(obj);
1994 continue;
1995 }
1996
1997 /*
1998 * Charge for the whole object allocation now, since
1999 * we cannot distinguish between non-charged and
2000 * charged clipped mapping of the same object later.
2001 */
2002 KASSERT(obj->charge == 0,
2003 ("vm_map_protect: object %p overcharged (entry %p)",
2004 obj, current));
2005 if (!swap_reserve(ptoa(obj->size))) {
2006 VM_OBJECT_WUNLOCK(obj);
2007 vm_map_unlock(map);
2008 return (KERN_RESOURCE_SHORTAGE);
2009 }
2010
2011 crhold(cred);
2012 obj->cred = cred;
2013 obj->charge = ptoa(obj->size);
2014 VM_OBJECT_WUNLOCK(obj);
2015 }
2016
2017 /*
2018 * Go back and fix up protections. [Note that clipping is not
2019 * necessary the second time.]
2020 */
2021 current = entry;
2022 while ((current != &map->header) && (current->start < end)) {
2023 old_prot = current->protection;
2024
2025 if (set_max)
2026 current->protection =
2027 (current->max_protection = new_prot) &
2028 old_prot;
2029 else
2030 current->protection = new_prot;
2031
2032 /*
2033 * For user wired map entries, the normal lazy evaluation of
2034 * write access upgrades through soft page faults is
2035 * undesirable. Instead, immediately copy any pages that are
2036 * copy-on-write and enable write access in the physical map.
2037 */
2038 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2039 (current->protection & VM_PROT_WRITE) != 0 &&
2040 (old_prot & VM_PROT_WRITE) == 0)
2041 vm_fault_copy_entry(map, map, current, current, NULL);
2042
2043 /*
2044 * When restricting access, update the physical map. Worry
2045 * about copy-on-write here.
2046 */
2047 if ((old_prot & ~current->protection) != 0) {
2048 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2049 VM_PROT_ALL)
2050 pmap_protect(map->pmap, current->start,
2051 current->end,
2052 current->protection & MASK(current));
2053 #undef MASK
2054 }
2055 vm_map_simplify_entry(map, current);
2056 current = current->next;
2057 }
2058 vm_map_unlock(map);
2059 return (KERN_SUCCESS);
2060 }
2061
2062 /*
2063 * vm_map_madvise:
2064 *
2065 * This routine traverses a processes map handling the madvise
2066 * system call. Advisories are classified as either those effecting
2067 * the vm_map_entry structure, or those effecting the underlying
2068 * objects.
2069 */
2070 int
2071 vm_map_madvise(
2072 vm_map_t map,
2073 vm_offset_t start,
2074 vm_offset_t end,
2075 int behav)
2076 {
2077 vm_map_entry_t current, entry;
2078 int modify_map = 0;
2079
2080 /*
2081 * Some madvise calls directly modify the vm_map_entry, in which case
2082 * we need to use an exclusive lock on the map and we need to perform
2083 * various clipping operations. Otherwise we only need a read-lock
2084 * on the map.
2085 */
2086 switch(behav) {
2087 case MADV_NORMAL:
2088 case MADV_SEQUENTIAL:
2089 case MADV_RANDOM:
2090 case MADV_NOSYNC:
2091 case MADV_AUTOSYNC:
2092 case MADV_NOCORE:
2093 case MADV_CORE:
2094 if (start == end)
2095 return (KERN_SUCCESS);
2096 modify_map = 1;
2097 vm_map_lock(map);
2098 break;
2099 case MADV_WILLNEED:
2100 case MADV_DONTNEED:
2101 case MADV_FREE:
2102 if (start == end)
2103 return (KERN_SUCCESS);
2104 vm_map_lock_read(map);
2105 break;
2106 default:
2107 return (KERN_INVALID_ARGUMENT);
2108 }
2109
2110 /*
2111 * Locate starting entry and clip if necessary.
2112 */
2113 VM_MAP_RANGE_CHECK(map, start, end);
2114
2115 if (vm_map_lookup_entry(map, start, &entry)) {
2116 if (modify_map)
2117 vm_map_clip_start(map, entry, start);
2118 } else {
2119 entry = entry->next;
2120 }
2121
2122 if (modify_map) {
2123 /*
2124 * madvise behaviors that are implemented in the vm_map_entry.
2125 *
2126 * We clip the vm_map_entry so that behavioral changes are
2127 * limited to the specified address range.
2128 */
2129 for (current = entry;
2130 (current != &map->header) && (current->start < end);
2131 current = current->next
2132 ) {
2133 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2134 continue;
2135
2136 vm_map_clip_end(map, current, end);
2137
2138 switch (behav) {
2139 case MADV_NORMAL:
2140 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2141 break;
2142 case MADV_SEQUENTIAL:
2143 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2144 break;
2145 case MADV_RANDOM:
2146 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2147 break;
2148 case MADV_NOSYNC:
2149 current->eflags |= MAP_ENTRY_NOSYNC;
2150 break;
2151 case MADV_AUTOSYNC:
2152 current->eflags &= ~MAP_ENTRY_NOSYNC;
2153 break;
2154 case MADV_NOCORE:
2155 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2156 break;
2157 case MADV_CORE:
2158 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2159 break;
2160 default:
2161 break;
2162 }
2163 vm_map_simplify_entry(map, current);
2164 }
2165 vm_map_unlock(map);
2166 } else {
2167 vm_pindex_t pstart, pend;
2168
2169 /*
2170 * madvise behaviors that are implemented in the underlying
2171 * vm_object.
2172 *
2173 * Since we don't clip the vm_map_entry, we have to clip
2174 * the vm_object pindex and count.
2175 */
2176 for (current = entry;
2177 (current != &map->header) && (current->start < end);
2178 current = current->next
2179 ) {
2180 vm_offset_t useEnd, useStart;
2181
2182 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2183 continue;
2184
2185 pstart = OFF_TO_IDX(current->offset);
2186 pend = pstart + atop(current->end - current->start);
2187 useStart = current->start;
2188 useEnd = current->end;
2189
2190 if (current->start < start) {
2191 pstart += atop(start - current->start);
2192 useStart = start;
2193 }
2194 if (current->end > end) {
2195 pend -= atop(current->end - end);
2196 useEnd = end;
2197 }
2198
2199 if (pstart >= pend)
2200 continue;
2201
2202 /*
2203 * Perform the pmap_advise() before clearing
2204 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2205 * concurrent pmap operation, such as pmap_remove(),
2206 * could clear a reference in the pmap and set
2207 * PGA_REFERENCED on the page before the pmap_advise()
2208 * had completed. Consequently, the page would appear
2209 * referenced based upon an old reference that
2210 * occurred before this pmap_advise() ran.
2211 */
2212 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2213 pmap_advise(map->pmap, useStart, useEnd,
2214 behav);
2215
2216 vm_object_madvise(current->object.vm_object, pstart,
2217 pend, behav);
2218
2219 /*
2220 * Pre-populate paging structures in the
2221 * WILLNEED case. For wired entries, the
2222 * paging structures are already populated.
2223 */
2224 if (behav == MADV_WILLNEED &&
2225 current->wired_count == 0) {
2226 vm_map_pmap_enter(map,
2227 useStart,
2228 current->protection,
2229 current->object.vm_object,
2230 pstart,
2231 ptoa(pend - pstart),
2232 MAP_PREFAULT_MADVISE
2233 );
2234 }
2235 }
2236 vm_map_unlock_read(map);
2237 }
2238 return (0);
2239 }
2240
2241
2242 /*
2243 * vm_map_inherit:
2244 *
2245 * Sets the inheritance of the specified address
2246 * range in the target map. Inheritance
2247 * affects how the map will be shared with
2248 * child maps at the time of vmspace_fork.
2249 */
2250 int
2251 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2252 vm_inherit_t new_inheritance)
2253 {
2254 vm_map_entry_t entry;
2255 vm_map_entry_t temp_entry;
2256
2257 switch (new_inheritance) {
2258 case VM_INHERIT_NONE:
2259 case VM_INHERIT_COPY:
2260 case VM_INHERIT_SHARE:
2261 break;
2262 default:
2263 return (KERN_INVALID_ARGUMENT);
2264 }
2265 if (start == end)
2266 return (KERN_SUCCESS);
2267 vm_map_lock(map);
2268 VM_MAP_RANGE_CHECK(map, start, end);
2269 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2270 entry = temp_entry;
2271 vm_map_clip_start(map, entry, start);
2272 } else
2273 entry = temp_entry->next;
2274 while ((entry != &map->header) && (entry->start < end)) {
2275 vm_map_clip_end(map, entry, end);
2276 entry->inheritance = new_inheritance;
2277 vm_map_simplify_entry(map, entry);
2278 entry = entry->next;
2279 }
2280 vm_map_unlock(map);
2281 return (KERN_SUCCESS);
2282 }
2283
2284 /*
2285 * vm_map_unwire:
2286 *
2287 * Implements both kernel and user unwiring.
2288 */
2289 int
2290 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2291 int flags)
2292 {
2293 vm_map_entry_t entry, first_entry, tmp_entry;
2294 vm_offset_t saved_start;
2295 unsigned int last_timestamp;
2296 int rv;
2297 boolean_t need_wakeup, result, user_unwire;
2298
2299 if (start == end)
2300 return (KERN_SUCCESS);
2301 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2302 vm_map_lock(map);
2303 VM_MAP_RANGE_CHECK(map, start, end);
2304 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2305 if (flags & VM_MAP_WIRE_HOLESOK)
2306 first_entry = first_entry->next;
2307 else {
2308 vm_map_unlock(map);
2309 return (KERN_INVALID_ADDRESS);
2310 }
2311 }
2312 last_timestamp = map->timestamp;
2313 entry = first_entry;
2314 while (entry != &map->header && entry->start < end) {
2315 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2316 /*
2317 * We have not yet clipped the entry.
2318 */
2319 saved_start = (start >= entry->start) ? start :
2320 entry->start;
2321 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2322 if (vm_map_unlock_and_wait(map, 0)) {
2323 /*
2324 * Allow interruption of user unwiring?
2325 */
2326 }
2327 vm_map_lock(map);
2328 if (last_timestamp+1 != map->timestamp) {
2329 /*
2330 * Look again for the entry because the map was
2331 * modified while it was unlocked.
2332 * Specifically, the entry may have been
2333 * clipped, merged, or deleted.
2334 */
2335 if (!vm_map_lookup_entry(map, saved_start,
2336 &tmp_entry)) {
2337 if (flags & VM_MAP_WIRE_HOLESOK)
2338 tmp_entry = tmp_entry->next;
2339 else {
2340 if (saved_start == start) {
2341 /*
2342 * First_entry has been deleted.
2343 */
2344 vm_map_unlock(map);
2345 return (KERN_INVALID_ADDRESS);
2346 }
2347 end = saved_start;
2348 rv = KERN_INVALID_ADDRESS;
2349 goto done;
2350 }
2351 }
2352 if (entry == first_entry)
2353 first_entry = tmp_entry;
2354 else
2355 first_entry = NULL;
2356 entry = tmp_entry;
2357 }
2358 last_timestamp = map->timestamp;
2359 continue;
2360 }
2361 vm_map_clip_start(map, entry, start);
2362 vm_map_clip_end(map, entry, end);
2363 /*
2364 * Mark the entry in case the map lock is released. (See
2365 * above.)
2366 */
2367 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2368 entry->wiring_thread == NULL,
2369 ("owned map entry %p", entry));
2370 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2371 entry->wiring_thread = curthread;
2372 /*
2373 * Check the map for holes in the specified region.
2374 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2375 */
2376 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2377 (entry->end < end && (entry->next == &map->header ||
2378 entry->next->start > entry->end))) {
2379 end = entry->end;
2380 rv = KERN_INVALID_ADDRESS;
2381 goto done;
2382 }
2383 /*
2384 * If system unwiring, require that the entry is system wired.
2385 */
2386 if (!user_unwire &&
2387 vm_map_entry_system_wired_count(entry) == 0) {
2388 end = entry->end;
2389 rv = KERN_INVALID_ARGUMENT;
2390 goto done;
2391 }
2392 entry = entry->next;
2393 }
2394 rv = KERN_SUCCESS;
2395 done:
2396 need_wakeup = FALSE;
2397 if (first_entry == NULL) {
2398 result = vm_map_lookup_entry(map, start, &first_entry);
2399 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2400 first_entry = first_entry->next;
2401 else
2402 KASSERT(result, ("vm_map_unwire: lookup failed"));
2403 }
2404 for (entry = first_entry; entry != &map->header && entry->start < end;
2405 entry = entry->next) {
2406 /*
2407 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2408 * space in the unwired region could have been mapped
2409 * while the map lock was dropped for draining
2410 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2411 * could be simultaneously wiring this new mapping
2412 * entry. Detect these cases and skip any entries
2413 * marked as in transition by us.
2414 */
2415 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2416 entry->wiring_thread != curthread) {
2417 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2418 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2419 continue;
2420 }
2421
2422 if (rv == KERN_SUCCESS && (!user_unwire ||
2423 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2424 if (user_unwire)
2425 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2426 if (entry->wired_count == 1)
2427 vm_map_entry_unwire(map, entry);
2428 else
2429 entry->wired_count--;
2430 }
2431 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2432 ("vm_map_unwire: in-transition flag missing %p", entry));
2433 KASSERT(entry->wiring_thread == curthread,
2434 ("vm_map_unwire: alien wire %p", entry));
2435 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2436 entry->wiring_thread = NULL;
2437 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2438 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2439 need_wakeup = TRUE;
2440 }
2441 vm_map_simplify_entry(map, entry);
2442 }
2443 vm_map_unlock(map);
2444 if (need_wakeup)
2445 vm_map_wakeup(map);
2446 return (rv);
2447 }
2448
2449 /*
2450 * vm_map_wire_entry_failure:
2451 *
2452 * Handle a wiring failure on the given entry.
2453 *
2454 * The map should be locked.
2455 */
2456 static void
2457 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2458 vm_offset_t failed_addr)
2459 {
2460
2461 VM_MAP_ASSERT_LOCKED(map);
2462 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2463 entry->wired_count == 1,
2464 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2465 KASSERT(failed_addr < entry->end,
2466 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2467
2468 /*
2469 * If any pages at the start of this entry were successfully wired,
2470 * then unwire them.
2471 */
2472 if (failed_addr > entry->start) {
2473 pmap_unwire(map->pmap, entry->start, failed_addr);
2474 vm_object_unwire(entry->object.vm_object, entry->offset,
2475 failed_addr - entry->start, PQ_ACTIVE);
2476 }
2477
2478 /*
2479 * Assign an out-of-range value to represent the failure to wire this
2480 * entry.
2481 */
2482 entry->wired_count = -1;
2483 }
2484
2485 /*
2486 * vm_map_wire:
2487 *
2488 * Implements both kernel and user wiring.
2489 */
2490 int
2491 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2492 int flags)
2493 {
2494 vm_map_entry_t entry, first_entry, tmp_entry;
2495 vm_offset_t faddr, saved_end, saved_start;
2496 unsigned int last_timestamp;
2497 int rv;
2498 boolean_t need_wakeup, result, user_wire;
2499 vm_prot_t prot;
2500
2501 if (start == end)
2502 return (KERN_SUCCESS);
2503 prot = 0;
2504 if (flags & VM_MAP_WIRE_WRITE)
2505 prot |= VM_PROT_WRITE;
2506 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2507 vm_map_lock(map);
2508 VM_MAP_RANGE_CHECK(map, start, end);
2509 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2510 if (flags & VM_MAP_WIRE_HOLESOK)
2511 first_entry = first_entry->next;
2512 else {
2513 vm_map_unlock(map);
2514 return (KERN_INVALID_ADDRESS);
2515 }
2516 }
2517 last_timestamp = map->timestamp;
2518 entry = first_entry;
2519 while (entry != &map->header && entry->start < end) {
2520 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2521 /*
2522 * We have not yet clipped the entry.
2523 */
2524 saved_start = (start >= entry->start) ? start :
2525 entry->start;
2526 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2527 if (vm_map_unlock_and_wait(map, 0)) {
2528 /*
2529 * Allow interruption of user wiring?
2530 */
2531 }
2532 vm_map_lock(map);
2533 if (last_timestamp + 1 != map->timestamp) {
2534 /*
2535 * Look again for the entry because the map was
2536 * modified while it was unlocked.
2537 * Specifically, the entry may have been
2538 * clipped, merged, or deleted.
2539 */
2540 if (!vm_map_lookup_entry(map, saved_start,
2541 &tmp_entry)) {
2542 if (flags & VM_MAP_WIRE_HOLESOK)
2543 tmp_entry = tmp_entry->next;
2544 else {
2545 if (saved_start == start) {
2546 /*
2547 * first_entry has been deleted.
2548 */
2549 vm_map_unlock(map);
2550 return (KERN_INVALID_ADDRESS);
2551 }
2552 end = saved_start;
2553 rv = KERN_INVALID_ADDRESS;
2554 goto done;
2555 }
2556 }
2557 if (entry == first_entry)
2558 first_entry = tmp_entry;
2559 else
2560 first_entry = NULL;
2561 entry = tmp_entry;
2562 }
2563 last_timestamp = map->timestamp;
2564 continue;
2565 }
2566 vm_map_clip_start(map, entry, start);
2567 vm_map_clip_end(map, entry, end);
2568 /*
2569 * Mark the entry in case the map lock is released. (See
2570 * above.)
2571 */
2572 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2573 entry->wiring_thread == NULL,
2574 ("owned map entry %p", entry));
2575 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2576 entry->wiring_thread = curthread;
2577 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2578 || (entry->protection & prot) != prot) {
2579 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2580 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2581 end = entry->end;
2582 rv = KERN_INVALID_ADDRESS;
2583 goto done;
2584 }
2585 goto next_entry;
2586 }
2587 if (entry->wired_count == 0) {
2588 entry->wired_count++;
2589 saved_start = entry->start;
2590 saved_end = entry->end;
2591
2592 /*
2593 * Release the map lock, relying on the in-transition
2594 * mark. Mark the map busy for fork.
2595 */
2596 vm_map_busy(map);
2597 vm_map_unlock(map);
2598
2599 faddr = saved_start;
2600 do {
2601 /*
2602 * Simulate a fault to get the page and enter
2603 * it into the physical map.
2604 */
2605 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2606 VM_FAULT_WIRE)) != KERN_SUCCESS)
2607 break;
2608 } while ((faddr += PAGE_SIZE) < saved_end);
2609 vm_map_lock(map);
2610 vm_map_unbusy(map);
2611 if (last_timestamp + 1 != map->timestamp) {
2612 /*
2613 * Look again for the entry because the map was
2614 * modified while it was unlocked. The entry
2615 * may have been clipped, but NOT merged or
2616 * deleted.
2617 */
2618 result = vm_map_lookup_entry(map, saved_start,
2619 &tmp_entry);
2620 KASSERT(result, ("vm_map_wire: lookup failed"));
2621 if (entry == first_entry)
2622 first_entry = tmp_entry;
2623 else
2624 first_entry = NULL;
2625 entry = tmp_entry;
2626 while (entry->end < saved_end) {
2627 /*
2628 * In case of failure, handle entries
2629 * that were not fully wired here;
2630 * fully wired entries are handled
2631 * later.
2632 */
2633 if (rv != KERN_SUCCESS &&
2634 faddr < entry->end)
2635 vm_map_wire_entry_failure(map,
2636 entry, faddr);
2637 entry = entry->next;
2638 }
2639 }
2640 last_timestamp = map->timestamp;
2641 if (rv != KERN_SUCCESS) {
2642 vm_map_wire_entry_failure(map, entry, faddr);
2643 end = entry->end;
2644 goto done;
2645 }
2646 } else if (!user_wire ||
2647 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2648 entry->wired_count++;
2649 }
2650 /*
2651 * Check the map for holes in the specified region.
2652 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2653 */
2654 next_entry:
2655 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2656 (entry->end < end && (entry->next == &map->header ||
2657 entry->next->start > entry->end))) {
2658 end = entry->end;
2659 rv = KERN_INVALID_ADDRESS;
2660 goto done;
2661 }
2662 entry = entry->next;
2663 }
2664 rv = KERN_SUCCESS;
2665 done:
2666 need_wakeup = FALSE;
2667 if (first_entry == NULL) {
2668 result = vm_map_lookup_entry(map, start, &first_entry);
2669 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2670 first_entry = first_entry->next;
2671 else
2672 KASSERT(result, ("vm_map_wire: lookup failed"));
2673 }
2674 for (entry = first_entry; entry != &map->header && entry->start < end;
2675 entry = entry->next) {
2676 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2677 goto next_entry_done;
2678
2679 /*
2680 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2681 * space in the unwired region could have been mapped
2682 * while the map lock was dropped for faulting in the
2683 * pages or draining MAP_ENTRY_IN_TRANSITION.
2684 * Moreover, another thread could be simultaneously
2685 * wiring this new mapping entry. Detect these cases
2686 * and skip any entries marked as in transition by us.
2687 */
2688 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2689 entry->wiring_thread != curthread) {
2690 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2691 ("vm_map_wire: !HOLESOK and new/changed entry"));
2692 continue;
2693 }
2694
2695 if (rv == KERN_SUCCESS) {
2696 if (user_wire)
2697 entry->eflags |= MAP_ENTRY_USER_WIRED;
2698 } else if (entry->wired_count == -1) {
2699 /*
2700 * Wiring failed on this entry. Thus, unwiring is
2701 * unnecessary.
2702 */
2703 entry->wired_count = 0;
2704 } else if (!user_wire ||
2705 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2706 /*
2707 * Undo the wiring. Wiring succeeded on this entry
2708 * but failed on a later entry.
2709 */
2710 if (entry->wired_count == 1)
2711 vm_map_entry_unwire(map, entry);
2712 else
2713 entry->wired_count--;
2714 }
2715 next_entry_done:
2716 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2717 ("vm_map_wire: in-transition flag missing %p", entry));
2718 KASSERT(entry->wiring_thread == curthread,
2719 ("vm_map_wire: alien wire %p", entry));
2720 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2721 MAP_ENTRY_WIRE_SKIPPED);
2722 entry->wiring_thread = NULL;
2723 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2724 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2725 need_wakeup = TRUE;
2726 }
2727 vm_map_simplify_entry(map, entry);
2728 }
2729 vm_map_unlock(map);
2730 if (need_wakeup)
2731 vm_map_wakeup(map);
2732 return (rv);
2733 }
2734
2735 /*
2736 * vm_map_sync
2737 *
2738 * Push any dirty cached pages in the address range to their pager.
2739 * If syncio is TRUE, dirty pages are written synchronously.
2740 * If invalidate is TRUE, any cached pages are freed as well.
2741 *
2742 * If the size of the region from start to end is zero, we are
2743 * supposed to flush all modified pages within the region containing
2744 * start. Unfortunately, a region can be split or coalesced with
2745 * neighboring regions, making it difficult to determine what the
2746 * original region was. Therefore, we approximate this requirement by
2747 * flushing the current region containing start.
2748 *
2749 * Returns an error if any part of the specified range is not mapped.
2750 */
2751 int
2752 vm_map_sync(
2753 vm_map_t map,
2754 vm_offset_t start,
2755 vm_offset_t end,
2756 boolean_t syncio,
2757 boolean_t invalidate)
2758 {
2759 vm_map_entry_t current;
2760 vm_map_entry_t entry;
2761 vm_size_t size;
2762 vm_object_t object;
2763 vm_ooffset_t offset;
2764 unsigned int last_timestamp;
2765 boolean_t failed;
2766
2767 vm_map_lock_read(map);
2768 VM_MAP_RANGE_CHECK(map, start, end);
2769 if (!vm_map_lookup_entry(map, start, &entry)) {
2770 vm_map_unlock_read(map);
2771 return (KERN_INVALID_ADDRESS);
2772 } else if (start == end) {
2773 start = entry->start;
2774 end = entry->end;
2775 }
2776 /*
2777 * Make a first pass to check for user-wired memory and holes.
2778 */
2779 for (current = entry; current != &map->header && current->start < end;
2780 current = current->next) {
2781 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2782 vm_map_unlock_read(map);
2783 return (KERN_INVALID_ARGUMENT);
2784 }
2785 if (end > current->end &&
2786 (current->next == &map->header ||
2787 current->end != current->next->start)) {
2788 vm_map_unlock_read(map);
2789 return (KERN_INVALID_ADDRESS);
2790 }
2791 }
2792
2793 if (invalidate)
2794 pmap_remove(map->pmap, start, end);
2795 failed = FALSE;
2796
2797 /*
2798 * Make a second pass, cleaning/uncaching pages from the indicated
2799 * objects as we go.
2800 */
2801 for (current = entry; current != &map->header && current->start < end;) {
2802 offset = current->offset + (start - current->start);
2803 size = (end <= current->end ? end : current->end) - start;
2804 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2805 vm_map_t smap;
2806 vm_map_entry_t tentry;
2807 vm_size_t tsize;
2808
2809 smap = current->object.sub_map;
2810 vm_map_lock_read(smap);
2811 (void) vm_map_lookup_entry(smap, offset, &tentry);
2812 tsize = tentry->end - offset;
2813 if (tsize < size)
2814 size = tsize;
2815 object = tentry->object.vm_object;
2816 offset = tentry->offset + (offset - tentry->start);
2817 vm_map_unlock_read(smap);
2818 } else {
2819 object = current->object.vm_object;
2820 }
2821 vm_object_reference(object);
2822 last_timestamp = map->timestamp;
2823 vm_map_unlock_read(map);
2824 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2825 failed = TRUE;
2826 start += size;
2827 vm_object_deallocate(object);
2828 vm_map_lock_read(map);
2829 if (last_timestamp == map->timestamp ||
2830 !vm_map_lookup_entry(map, start, ¤t))
2831 current = current->next;
2832 }
2833
2834 vm_map_unlock_read(map);
2835 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2836 }
2837
2838 /*
2839 * vm_map_entry_unwire: [ internal use only ]
2840 *
2841 * Make the region specified by this entry pageable.
2842 *
2843 * The map in question should be locked.
2844 * [This is the reason for this routine's existence.]
2845 */
2846 static void
2847 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2848 {
2849
2850 VM_MAP_ASSERT_LOCKED(map);
2851 KASSERT(entry->wired_count > 0,
2852 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2853 pmap_unwire(map->pmap, entry->start, entry->end);
2854 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2855 entry->start, PQ_ACTIVE);
2856 entry->wired_count = 0;
2857 }
2858
2859 static void
2860 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2861 {
2862
2863 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2864 vm_object_deallocate(entry->object.vm_object);
2865 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2866 }
2867
2868 /*
2869 * vm_map_entry_delete: [ internal use only ]
2870 *
2871 * Deallocate the given entry from the target map.
2872 */
2873 static void
2874 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2875 {
2876 vm_object_t object;
2877 vm_pindex_t offidxstart, offidxend, count, size1;
2878 vm_ooffset_t size;
2879
2880 vm_map_entry_unlink(map, entry);
2881 object = entry->object.vm_object;
2882 size = entry->end - entry->start;
2883 map->size -= size;
2884
2885 if (entry->cred != NULL) {
2886 swap_release_by_cred(size, entry->cred);
2887 crfree(entry->cred);
2888 }
2889
2890 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2891 (object != NULL)) {
2892 KASSERT(entry->cred == NULL || object->cred == NULL ||
2893 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2894 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2895 count = OFF_TO_IDX(size);
2896 offidxstart = OFF_TO_IDX(entry->offset);
2897 offidxend = offidxstart + count;
2898 VM_OBJECT_WLOCK(object);
2899 if (object->ref_count != 1 &&
2900 ((object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2901 object == kernel_object || object == kmem_object)) {
2902 vm_object_collapse(object);
2903
2904 /*
2905 * The option OBJPR_NOTMAPPED can be passed here
2906 * because vm_map_delete() already performed
2907 * pmap_remove() on the only mapping to this range
2908 * of pages.
2909 */
2910 vm_object_page_remove(object, offidxstart, offidxend,
2911 OBJPR_NOTMAPPED);
2912 if (object->type == OBJT_SWAP)
2913 swap_pager_freespace(object, offidxstart, count);
2914 if (offidxend >= object->size &&
2915 offidxstart < object->size) {
2916 size1 = object->size;
2917 object->size = offidxstart;
2918 if (object->cred != NULL) {
2919 size1 -= object->size;
2920 KASSERT(object->charge >= ptoa(size1),
2921 ("vm_map_entry_delete: object->charge < 0"));
2922 swap_release_by_cred(ptoa(size1), object->cred);
2923 object->charge -= ptoa(size1);
2924 }
2925 }
2926 }
2927 VM_OBJECT_WUNLOCK(object);
2928 } else
2929 entry->object.vm_object = NULL;
2930 if (map->system_map)
2931 vm_map_entry_deallocate(entry, TRUE);
2932 else {
2933 entry->next = curthread->td_map_def_user;
2934 curthread->td_map_def_user = entry;
2935 }
2936 }
2937
2938 /*
2939 * vm_map_delete: [ internal use only ]
2940 *
2941 * Deallocates the given address range from the target
2942 * map.
2943 */
2944 int
2945 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2946 {
2947 vm_map_entry_t entry;
2948 vm_map_entry_t first_entry;
2949
2950 VM_MAP_ASSERT_LOCKED(map);
2951 if (start == end)
2952 return (KERN_SUCCESS);
2953
2954 /*
2955 * Find the start of the region, and clip it
2956 */
2957 if (!vm_map_lookup_entry(map, start, &first_entry))
2958 entry = first_entry->next;
2959 else {
2960 entry = first_entry;
2961 vm_map_clip_start(map, entry, start);
2962 }
2963
2964 /*
2965 * Step through all entries in this region
2966 */
2967 while ((entry != &map->header) && (entry->start < end)) {
2968 vm_map_entry_t next;
2969
2970 /*
2971 * Wait for wiring or unwiring of an entry to complete.
2972 * Also wait for any system wirings to disappear on
2973 * user maps.
2974 */
2975 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
2976 (vm_map_pmap(map) != kernel_pmap &&
2977 vm_map_entry_system_wired_count(entry) != 0)) {
2978 unsigned int last_timestamp;
2979 vm_offset_t saved_start;
2980 vm_map_entry_t tmp_entry;
2981
2982 saved_start = entry->start;
2983 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2984 last_timestamp = map->timestamp;
2985 (void) vm_map_unlock_and_wait(map, 0);
2986 vm_map_lock(map);
2987 if (last_timestamp + 1 != map->timestamp) {
2988 /*
2989 * Look again for the entry because the map was
2990 * modified while it was unlocked.
2991 * Specifically, the entry may have been
2992 * clipped, merged, or deleted.
2993 */
2994 if (!vm_map_lookup_entry(map, saved_start,
2995 &tmp_entry))
2996 entry = tmp_entry->next;
2997 else {
2998 entry = tmp_entry;
2999 vm_map_clip_start(map, entry,
3000 saved_start);
3001 }
3002 }
3003 continue;
3004 }
3005 vm_map_clip_end(map, entry, end);
3006
3007 next = entry->next;
3008
3009 /*
3010 * Unwire before removing addresses from the pmap; otherwise,
3011 * unwiring will put the entries back in the pmap.
3012 */
3013 if (entry->wired_count != 0) {
3014 vm_map_entry_unwire(map, entry);
3015 }
3016
3017 pmap_remove(map->pmap, entry->start, entry->end);
3018
3019 /*
3020 * Delete the entry only after removing all pmap
3021 * entries pointing to its pages. (Otherwise, its
3022 * page frames may be reallocated, and any modify bits
3023 * will be set in the wrong object!)
3024 */
3025 vm_map_entry_delete(map, entry);
3026 entry = next;
3027 }
3028 return (KERN_SUCCESS);
3029 }
3030
3031 /*
3032 * vm_map_remove:
3033 *
3034 * Remove the given address range from the target map.
3035 * This is the exported form of vm_map_delete.
3036 */
3037 int
3038 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3039 {
3040 int result;
3041
3042 vm_map_lock(map);
3043 VM_MAP_RANGE_CHECK(map, start, end);
3044 result = vm_map_delete(map, start, end);
3045 vm_map_unlock(map);
3046 return (result);
3047 }
3048
3049 /*
3050 * vm_map_check_protection:
3051 *
3052 * Assert that the target map allows the specified privilege on the
3053 * entire address region given. The entire region must be allocated.
3054 *
3055 * WARNING! This code does not and should not check whether the
3056 * contents of the region is accessible. For example a smaller file
3057 * might be mapped into a larger address space.
3058 *
3059 * NOTE! This code is also called by munmap().
3060 *
3061 * The map must be locked. A read lock is sufficient.
3062 */
3063 boolean_t
3064 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3065 vm_prot_t protection)
3066 {
3067 vm_map_entry_t entry;
3068 vm_map_entry_t tmp_entry;
3069
3070 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3071 return (FALSE);
3072 entry = tmp_entry;
3073
3074 while (start < end) {
3075 if (entry == &map->header)
3076 return (FALSE);
3077 /*
3078 * No holes allowed!
3079 */
3080 if (start < entry->start)
3081 return (FALSE);
3082 /*
3083 * Check protection associated with entry.
3084 */
3085 if ((entry->protection & protection) != protection)
3086 return (FALSE);
3087 /* go to next entry */
3088 start = entry->end;
3089 entry = entry->next;
3090 }
3091 return (TRUE);
3092 }
3093
3094 /*
3095 * vm_map_copy_entry:
3096 *
3097 * Copies the contents of the source entry to the destination
3098 * entry. The entries *must* be aligned properly.
3099 */
3100 static void
3101 vm_map_copy_entry(
3102 vm_map_t src_map,
3103 vm_map_t dst_map,
3104 vm_map_entry_t src_entry,
3105 vm_map_entry_t dst_entry,
3106 vm_ooffset_t *fork_charge)
3107 {
3108 vm_object_t src_object;
3109 vm_map_entry_t fake_entry;
3110 vm_offset_t size;
3111 struct ucred *cred;
3112 int charged;
3113
3114 VM_MAP_ASSERT_LOCKED(dst_map);
3115
3116 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3117 return;
3118
3119 if (src_entry->wired_count == 0 ||
3120 (src_entry->protection & VM_PROT_WRITE) == 0) {
3121 /*
3122 * If the source entry is marked needs_copy, it is already
3123 * write-protected.
3124 */
3125 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3126 (src_entry->protection & VM_PROT_WRITE) != 0) {
3127 pmap_protect(src_map->pmap,
3128 src_entry->start,
3129 src_entry->end,
3130 src_entry->protection & ~VM_PROT_WRITE);
3131 }
3132
3133 /*
3134 * Make a copy of the object.
3135 */
3136 size = src_entry->end - src_entry->start;
3137 if ((src_object = src_entry->object.vm_object) != NULL) {
3138 VM_OBJECT_WLOCK(src_object);
3139 charged = ENTRY_CHARGED(src_entry);
3140 if ((src_object->handle == NULL) &&
3141 (src_object->type == OBJT_DEFAULT ||
3142 src_object->type == OBJT_SWAP)) {
3143 vm_object_collapse(src_object);
3144 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3145 vm_object_split(src_entry);
3146 src_object = src_entry->object.vm_object;
3147 }
3148 }
3149 vm_object_reference_locked(src_object);
3150 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3151 if (src_entry->cred != NULL &&
3152 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3153 KASSERT(src_object->cred == NULL,
3154 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3155 src_object));
3156 src_object->cred = src_entry->cred;
3157 src_object->charge = size;
3158 }
3159 VM_OBJECT_WUNLOCK(src_object);
3160 dst_entry->object.vm_object = src_object;
3161 if (charged) {
3162 cred = curthread->td_ucred;
3163 crhold(cred);
3164 dst_entry->cred = cred;
3165 *fork_charge += size;
3166 if (!(src_entry->eflags &
3167 MAP_ENTRY_NEEDS_COPY)) {
3168 crhold(cred);
3169 src_entry->cred = cred;
3170 *fork_charge += size;
3171 }
3172 }
3173 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3174 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3175 dst_entry->offset = src_entry->offset;
3176 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3177 /*
3178 * MAP_ENTRY_VN_WRITECNT cannot
3179 * indicate write reference from
3180 * src_entry, since the entry is
3181 * marked as needs copy. Allocate a
3182 * fake entry that is used to
3183 * decrement object->un_pager.vnp.writecount
3184 * at the appropriate time. Attach
3185 * fake_entry to the deferred list.
3186 */
3187 fake_entry = vm_map_entry_create(dst_map);
3188 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3189 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3190 vm_object_reference(src_object);
3191 fake_entry->object.vm_object = src_object;
3192 fake_entry->start = src_entry->start;
3193 fake_entry->end = src_entry->end;
3194 fake_entry->next = curthread->td_map_def_user;
3195 curthread->td_map_def_user = fake_entry;
3196 }
3197 } else {
3198 dst_entry->object.vm_object = NULL;
3199 dst_entry->offset = 0;
3200 if (src_entry->cred != NULL) {
3201 dst_entry->cred = curthread->td_ucred;
3202 crhold(dst_entry->cred);
3203 *fork_charge += size;
3204 }
3205 }
3206
3207 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3208 dst_entry->end - dst_entry->start, src_entry->start);
3209 } else {
3210 /*
3211 * We don't want to make writeable wired pages copy-on-write.
3212 * Immediately copy these pages into the new map by simulating
3213 * page faults. The new pages are pageable.
3214 */
3215 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3216 fork_charge);
3217 }
3218 }
3219
3220 /*
3221 * vmspace_map_entry_forked:
3222 * Update the newly-forked vmspace each time a map entry is inherited
3223 * or copied. The values for vm_dsize and vm_tsize are approximate
3224 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3225 */
3226 static void
3227 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3228 vm_map_entry_t entry)
3229 {
3230 vm_size_t entrysize;
3231 vm_offset_t newend;
3232
3233 entrysize = entry->end - entry->start;
3234 vm2->vm_map.size += entrysize;
3235 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3236 vm2->vm_ssize += btoc(entrysize);
3237 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3238 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3239 newend = MIN(entry->end,
3240 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3241 vm2->vm_dsize += btoc(newend - entry->start);
3242 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3243 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3244 newend = MIN(entry->end,
3245 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3246 vm2->vm_tsize += btoc(newend - entry->start);
3247 }
3248 }
3249
3250 /*
3251 * vmspace_fork:
3252 * Create a new process vmspace structure and vm_map
3253 * based on those of an existing process. The new map
3254 * is based on the old map, according to the inheritance
3255 * values on the regions in that map.
3256 *
3257 * XXX It might be worth coalescing the entries added to the new vmspace.
3258 *
3259 * The source map must not be locked.
3260 */
3261 struct vmspace *
3262 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3263 {
3264 struct vmspace *vm2;
3265 vm_map_t new_map, old_map;
3266 vm_map_entry_t new_entry, old_entry;
3267 vm_object_t object;
3268 int locked;
3269
3270 old_map = &vm1->vm_map;
3271 /* Copy immutable fields of vm1 to vm2. */
3272 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3273 if (vm2 == NULL)
3274 return (NULL);
3275 vm2->vm_taddr = vm1->vm_taddr;
3276 vm2->vm_daddr = vm1->vm_daddr;
3277 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3278 vm_map_lock(old_map);
3279 if (old_map->busy)
3280 vm_map_wait_busy(old_map);
3281 new_map = &vm2->vm_map;
3282 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3283 KASSERT(locked, ("vmspace_fork: lock failed"));
3284
3285 old_entry = old_map->header.next;
3286
3287 while (old_entry != &old_map->header) {
3288 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3289 panic("vm_map_fork: encountered a submap");
3290
3291 switch (old_entry->inheritance) {
3292 case VM_INHERIT_NONE:
3293 break;
3294
3295 case VM_INHERIT_SHARE:
3296 /*
3297 * Clone the entry, creating the shared object if necessary.
3298 */
3299 object = old_entry->object.vm_object;
3300 if (object == NULL) {
3301 object = vm_object_allocate(OBJT_DEFAULT,
3302 atop(old_entry->end - old_entry->start));
3303 old_entry->object.vm_object = object;
3304 old_entry->offset = 0;
3305 if (old_entry->cred != NULL) {
3306 object->cred = old_entry->cred;
3307 object->charge = old_entry->end -
3308 old_entry->start;
3309 old_entry->cred = NULL;
3310 }
3311 }
3312
3313 /*
3314 * Add the reference before calling vm_object_shadow
3315 * to insure that a shadow object is created.
3316 */
3317 vm_object_reference(object);
3318 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3319 vm_object_shadow(&old_entry->object.vm_object,
3320 &old_entry->offset,
3321 old_entry->end - old_entry->start);
3322 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3323 /* Transfer the second reference too. */
3324 vm_object_reference(
3325 old_entry->object.vm_object);
3326
3327 /*
3328 * As in vm_map_simplify_entry(), the
3329 * vnode lock will not be acquired in
3330 * this call to vm_object_deallocate().
3331 */
3332 vm_object_deallocate(object);
3333 object = old_entry->object.vm_object;
3334 }
3335 VM_OBJECT_WLOCK(object);
3336 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3337 if (old_entry->cred != NULL) {
3338 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3339 object->cred = old_entry->cred;
3340 object->charge = old_entry->end - old_entry->start;
3341 old_entry->cred = NULL;
3342 }
3343
3344 /*
3345 * Assert the correct state of the vnode
3346 * v_writecount while the object is locked, to
3347 * not relock it later for the assertion
3348 * correctness.
3349 */
3350 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3351 object->type == OBJT_VNODE) {
3352 KASSERT(((struct vnode *)object->handle)->
3353 v_writecount > 0,
3354 ("vmspace_fork: v_writecount %p", object));
3355 KASSERT(object->un_pager.vnp.writemappings > 0,
3356 ("vmspace_fork: vnp.writecount %p",
3357 object));
3358 }
3359 VM_OBJECT_WUNLOCK(object);
3360
3361 /*
3362 * Clone the entry, referencing the shared object.
3363 */
3364 new_entry = vm_map_entry_create(new_map);
3365 *new_entry = *old_entry;
3366 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3367 MAP_ENTRY_IN_TRANSITION);
3368 new_entry->wiring_thread = NULL;
3369 new_entry->wired_count = 0;
3370 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3371 vnode_pager_update_writecount(object,
3372 new_entry->start, new_entry->end);
3373 }
3374
3375 /*
3376 * Insert the entry into the new map -- we know we're
3377 * inserting at the end of the new map.
3378 */
3379 vm_map_entry_link(new_map, new_map->header.prev,
3380 new_entry);
3381 vmspace_map_entry_forked(vm1, vm2, new_entry);
3382
3383 /*
3384 * Update the physical map
3385 */
3386 pmap_copy(new_map->pmap, old_map->pmap,
3387 new_entry->start,
3388 (old_entry->end - old_entry->start),
3389 old_entry->start);
3390 break;
3391
3392 case VM_INHERIT_COPY:
3393 /*
3394 * Clone the entry and link into the map.
3395 */
3396 new_entry = vm_map_entry_create(new_map);
3397 *new_entry = *old_entry;
3398 /*
3399 * Copied entry is COW over the old object.
3400 */
3401 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3402 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3403 new_entry->wiring_thread = NULL;
3404 new_entry->wired_count = 0;
3405 new_entry->object.vm_object = NULL;
3406 new_entry->cred = NULL;
3407 vm_map_entry_link(new_map, new_map->header.prev,
3408 new_entry);
3409 vmspace_map_entry_forked(vm1, vm2, new_entry);
3410 vm_map_copy_entry(old_map, new_map, old_entry,
3411 new_entry, fork_charge);
3412 break;
3413 }
3414 old_entry = old_entry->next;
3415 }
3416 /*
3417 * Use inlined vm_map_unlock() to postpone handling the deferred
3418 * map entries, which cannot be done until both old_map and
3419 * new_map locks are released.
3420 */
3421 sx_xunlock(&old_map->lock);
3422 sx_xunlock(&new_map->lock);
3423 vm_map_process_deferred();
3424
3425 return (vm2);
3426 }
3427
3428 int
3429 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3430 vm_prot_t prot, vm_prot_t max, int cow)
3431 {
3432 vm_size_t growsize, init_ssize;
3433 rlim_t lmemlim, vmemlim;
3434 int rv;
3435
3436 growsize = sgrowsiz;
3437 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3438 vm_map_lock(map);
3439 PROC_LOCK(curproc);
3440 lmemlim = lim_cur(curproc, RLIMIT_MEMLOCK);
3441 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3442 PROC_UNLOCK(curproc);
3443 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3444 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3445 rv = KERN_NO_SPACE;
3446 goto out;
3447 }
3448 }
3449 /* If we would blow our VMEM resource limit, no go */
3450 if (map->size + init_ssize > vmemlim) {
3451 rv = KERN_NO_SPACE;
3452 goto out;
3453 }
3454 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3455 max, cow);
3456 out:
3457 vm_map_unlock(map);
3458 return (rv);
3459 }
3460
3461 static int
3462 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3463 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3464 {
3465 vm_map_entry_t new_entry, prev_entry;
3466 vm_offset_t bot, top;
3467 vm_size_t init_ssize;
3468 int orient, rv;
3469
3470 /*
3471 * The stack orientation is piggybacked with the cow argument.
3472 * Extract it into orient and mask the cow argument so that we
3473 * don't pass it around further.
3474 * NOTE: We explicitly allow bi-directional stacks.
3475 */
3476 orient = cow & (MAP_STACK_GROWS_DOWN|MAP_STACK_GROWS_UP);
3477 KASSERT(orient != 0, ("No stack grow direction"));
3478
3479 if (addrbos < vm_map_min(map) ||
3480 addrbos > vm_map_max(map) ||
3481 addrbos + max_ssize < addrbos)
3482 return (KERN_NO_SPACE);
3483
3484 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3485
3486 /* If addr is already mapped, no go */
3487 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3488 return (KERN_NO_SPACE);
3489
3490 /*
3491 * If we can't accomodate max_ssize in the current mapping, no go.
3492 * However, we need to be aware that subsequent user mappings might
3493 * map into the space we have reserved for stack, and currently this
3494 * space is not protected.
3495 *
3496 * Hopefully we will at least detect this condition when we try to
3497 * grow the stack.
3498 */
3499 if ((prev_entry->next != &map->header) &&
3500 (prev_entry->next->start < addrbos + max_ssize))
3501 return (KERN_NO_SPACE);
3502
3503 /*
3504 * We initially map a stack of only init_ssize. We will grow as
3505 * needed later. Depending on the orientation of the stack (i.e.
3506 * the grow direction) we either map at the top of the range, the
3507 * bottom of the range or in the middle.
3508 *
3509 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3510 * and cow to be 0. Possibly we should eliminate these as input
3511 * parameters, and just pass these values here in the insert call.
3512 */
3513 if (orient == MAP_STACK_GROWS_DOWN)
3514 bot = addrbos + max_ssize - init_ssize;
3515 else if (orient == MAP_STACK_GROWS_UP)
3516 bot = addrbos;
3517 else
3518 bot = round_page(addrbos + max_ssize/2 - init_ssize/2);
3519 top = bot + init_ssize;
3520 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3521
3522 /* Now set the avail_ssize amount. */
3523 if (rv == KERN_SUCCESS) {
3524 if (prev_entry != &map->header)
3525 vm_map_clip_end(map, prev_entry, bot);
3526 new_entry = prev_entry->next;
3527 if (new_entry->end != top || new_entry->start != bot)
3528 panic("Bad entry start/end for new stack entry");
3529
3530 new_entry->avail_ssize = max_ssize - init_ssize;
3531 if (orient & MAP_STACK_GROWS_DOWN)
3532 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3533 if (orient & MAP_STACK_GROWS_UP)
3534 new_entry->eflags |= MAP_ENTRY_GROWS_UP;
3535 }
3536
3537 return (rv);
3538 }
3539
3540 static int stack_guard_page = 0;
3541 TUNABLE_INT("security.bsd.stack_guard_page", &stack_guard_page);
3542 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RW,
3543 &stack_guard_page, 0,
3544 "Insert stack guard page ahead of the growable segments.");
3545
3546 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
3547 * desired address is already mapped, or if we successfully grow
3548 * the stack. Also returns KERN_SUCCESS if addr is outside the
3549 * stack range (this is strange, but preserves compatibility with
3550 * the grow function in vm_machdep.c).
3551 */
3552 int
3553 vm_map_growstack(struct proc *p, vm_offset_t addr)
3554 {
3555 vm_map_entry_t next_entry, prev_entry;
3556 vm_map_entry_t new_entry, stack_entry;
3557 struct vmspace *vm = p->p_vmspace;
3558 vm_map_t map = &vm->vm_map;
3559 vm_offset_t end;
3560 vm_size_t growsize;
3561 size_t grow_amount, max_grow;
3562 rlim_t lmemlim, stacklim, vmemlim;
3563 int is_procstack, rv;
3564 struct ucred *cred;
3565 #ifdef notyet
3566 uint64_t limit;
3567 #endif
3568 #ifdef RACCT
3569 int error;
3570 #endif
3571
3572 Retry:
3573 PROC_LOCK(p);
3574 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3575 stacklim = lim_cur(p, RLIMIT_STACK);
3576 vmemlim = lim_cur(p, RLIMIT_VMEM);
3577 PROC_UNLOCK(p);
3578
3579 vm_map_lock_read(map);
3580
3581 /* If addr is already in the entry range, no need to grow.*/
3582 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
3583 vm_map_unlock_read(map);
3584 return (KERN_SUCCESS);
3585 }
3586
3587 next_entry = prev_entry->next;
3588 if (!(prev_entry->eflags & MAP_ENTRY_GROWS_UP)) {
3589 /*
3590 * This entry does not grow upwards. Since the address lies
3591 * beyond this entry, the next entry (if one exists) has to
3592 * be a downward growable entry. The entry list header is
3593 * never a growable entry, so it suffices to check the flags.
3594 */
3595 if (!(next_entry->eflags & MAP_ENTRY_GROWS_DOWN)) {
3596 vm_map_unlock_read(map);
3597 return (KERN_SUCCESS);
3598 }
3599 stack_entry = next_entry;
3600 } else {
3601 /*
3602 * This entry grows upward. If the next entry does not at
3603 * least grow downwards, this is the entry we need to grow.
3604 * otherwise we have two possible choices and we have to
3605 * select one.
3606 */
3607 if (next_entry->eflags & MAP_ENTRY_GROWS_DOWN) {
3608 /*
3609 * We have two choices; grow the entry closest to
3610 * the address to minimize the amount of growth.
3611 */
3612 if (addr - prev_entry->end <= next_entry->start - addr)
3613 stack_entry = prev_entry;
3614 else
3615 stack_entry = next_entry;
3616 } else
3617 stack_entry = prev_entry;
3618 }
3619
3620 if (stack_entry == next_entry) {
3621 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_DOWN, ("foo"));
3622 KASSERT(addr < stack_entry->start, ("foo"));
3623 end = (prev_entry != &map->header) ? prev_entry->end :
3624 stack_entry->start - stack_entry->avail_ssize;
3625 grow_amount = roundup(stack_entry->start - addr, PAGE_SIZE);
3626 max_grow = stack_entry->start - end;
3627 } else {
3628 KASSERT(stack_entry->eflags & MAP_ENTRY_GROWS_UP, ("foo"));
3629 KASSERT(addr >= stack_entry->end, ("foo"));
3630 end = (next_entry != &map->header) ? next_entry->start :
3631 stack_entry->end + stack_entry->avail_ssize;
3632 grow_amount = roundup(addr + 1 - stack_entry->end, PAGE_SIZE);
3633 max_grow = end - stack_entry->end;
3634 }
3635
3636 if (grow_amount > stack_entry->avail_ssize) {
3637 vm_map_unlock_read(map);
3638 return (KERN_NO_SPACE);
3639 }
3640
3641 /*
3642 * If there is no longer enough space between the entries nogo, and
3643 * adjust the available space. Note: this should only happen if the
3644 * user has mapped into the stack area after the stack was created,
3645 * and is probably an error.
3646 *
3647 * This also effectively destroys any guard page the user might have
3648 * intended by limiting the stack size.
3649 */
3650 if (grow_amount + (stack_guard_page ? PAGE_SIZE : 0) > max_grow) {
3651 if (vm_map_lock_upgrade(map))
3652 goto Retry;
3653
3654 stack_entry->avail_ssize = max_grow;
3655
3656 vm_map_unlock(map);
3657 return (KERN_NO_SPACE);
3658 }
3659
3660 is_procstack = (addr >= (vm_offset_t)vm->vm_maxsaddr &&
3661 addr < (vm_offset_t)p->p_sysent->sv_usrstack) ? 1 : 0;
3662
3663 /*
3664 * If this is the main process stack, see if we're over the stack
3665 * limit.
3666 */
3667 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3668 vm_map_unlock_read(map);
3669 return (KERN_NO_SPACE);
3670 }
3671 #ifdef RACCT
3672 if (racct_enable) {
3673 PROC_LOCK(p);
3674 if (is_procstack && racct_set(p, RACCT_STACK,
3675 ctob(vm->vm_ssize) + grow_amount)) {
3676 PROC_UNLOCK(p);
3677 vm_map_unlock_read(map);
3678 return (KERN_NO_SPACE);
3679 }
3680 PROC_UNLOCK(p);
3681 }
3682 #endif
3683
3684 /* Round up the grow amount modulo sgrowsiz */
3685 growsize = sgrowsiz;
3686 grow_amount = roundup(grow_amount, growsize);
3687 if (grow_amount > stack_entry->avail_ssize)
3688 grow_amount = stack_entry->avail_ssize;
3689 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3690 grow_amount = trunc_page((vm_size_t)stacklim) -
3691 ctob(vm->vm_ssize);
3692 }
3693 #ifdef notyet
3694 PROC_LOCK(p);
3695 limit = racct_get_available(p, RACCT_STACK);
3696 PROC_UNLOCK(p);
3697 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3698 grow_amount = limit - ctob(vm->vm_ssize);
3699 #endif
3700 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3701 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3702 vm_map_unlock_read(map);
3703 rv = KERN_NO_SPACE;
3704 goto out;
3705 }
3706 #ifdef RACCT
3707 if (racct_enable) {
3708 PROC_LOCK(p);
3709 if (racct_set(p, RACCT_MEMLOCK,
3710 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3711 PROC_UNLOCK(p);
3712 vm_map_unlock_read(map);
3713 rv = KERN_NO_SPACE;
3714 goto out;
3715 }
3716 PROC_UNLOCK(p);
3717 }
3718 #endif
3719 }
3720 /* If we would blow our VMEM resource limit, no go */
3721 if (map->size + grow_amount > vmemlim) {
3722 vm_map_unlock_read(map);
3723 rv = KERN_NO_SPACE;
3724 goto out;
3725 }
3726 #ifdef RACCT
3727 if (racct_enable) {
3728 PROC_LOCK(p);
3729 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3730 PROC_UNLOCK(p);
3731 vm_map_unlock_read(map);
3732 rv = KERN_NO_SPACE;
3733 goto out;
3734 }
3735 PROC_UNLOCK(p);
3736 }
3737 #endif
3738
3739 if (vm_map_lock_upgrade(map))
3740 goto Retry;
3741
3742 if (stack_entry == next_entry) {
3743 /*
3744 * Growing downward.
3745 */
3746 /* Get the preliminary new entry start value */
3747 addr = stack_entry->start - grow_amount;
3748
3749 /*
3750 * If this puts us into the previous entry, cut back our
3751 * growth to the available space. Also, see the note above.
3752 */
3753 if (addr < end) {
3754 stack_entry->avail_ssize = max_grow;
3755 addr = end;
3756 if (stack_guard_page)
3757 addr += PAGE_SIZE;
3758 }
3759
3760 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
3761 next_entry->protection, next_entry->max_protection, 0);
3762
3763 /* Adjust the available stack space by the amount we grew. */
3764 if (rv == KERN_SUCCESS) {
3765 if (prev_entry != &map->header)
3766 vm_map_clip_end(map, prev_entry, addr);
3767 new_entry = prev_entry->next;
3768 KASSERT(new_entry == stack_entry->prev, ("foo"));
3769 KASSERT(new_entry->end == stack_entry->start, ("foo"));
3770 KASSERT(new_entry->start == addr, ("foo"));
3771 grow_amount = new_entry->end - new_entry->start;
3772 new_entry->avail_ssize = stack_entry->avail_ssize -
3773 grow_amount;
3774 stack_entry->eflags &= ~MAP_ENTRY_GROWS_DOWN;
3775 new_entry->eflags |= MAP_ENTRY_GROWS_DOWN;
3776 }
3777 } else {
3778 /*
3779 * Growing upward.
3780 */
3781 addr = stack_entry->end + grow_amount;
3782
3783 /*
3784 * If this puts us into the next entry, cut back our growth
3785 * to the available space. Also, see the note above.
3786 */
3787 if (addr > end) {
3788 stack_entry->avail_ssize = end - stack_entry->end;
3789 addr = end;
3790 if (stack_guard_page)
3791 addr -= PAGE_SIZE;
3792 }
3793
3794 grow_amount = addr - stack_entry->end;
3795 cred = stack_entry->cred;
3796 if (cred == NULL && stack_entry->object.vm_object != NULL)
3797 cred = stack_entry->object.vm_object->cred;
3798 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3799 rv = KERN_NO_SPACE;
3800 /* Grow the underlying object if applicable. */
3801 else if (stack_entry->object.vm_object == NULL ||
3802 vm_object_coalesce(stack_entry->object.vm_object,
3803 stack_entry->offset,
3804 (vm_size_t)(stack_entry->end - stack_entry->start),
3805 (vm_size_t)grow_amount, cred != NULL)) {
3806 map->size += (addr - stack_entry->end);
3807 /* Update the current entry. */
3808 stack_entry->end = addr;
3809 stack_entry->avail_ssize -= grow_amount;
3810 vm_map_entry_resize_free(map, stack_entry);
3811 rv = KERN_SUCCESS;
3812
3813 if (next_entry != &map->header)
3814 vm_map_clip_start(map, next_entry, addr);
3815 } else
3816 rv = KERN_FAILURE;
3817 }
3818
3819 if (rv == KERN_SUCCESS && is_procstack)
3820 vm->vm_ssize += btoc(grow_amount);
3821
3822 vm_map_unlock(map);
3823
3824 /*
3825 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3826 */
3827 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE)) {
3828 vm_map_wire(map,
3829 (stack_entry == next_entry) ? addr : addr - grow_amount,
3830 (stack_entry == next_entry) ? stack_entry->start : addr,
3831 (p->p_flag & P_SYSTEM)
3832 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3833 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3834 }
3835
3836 out:
3837 #ifdef RACCT
3838 if (racct_enable && rv != KERN_SUCCESS) {
3839 PROC_LOCK(p);
3840 error = racct_set(p, RACCT_VMEM, map->size);
3841 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3842 if (!old_mlock) {
3843 error = racct_set(p, RACCT_MEMLOCK,
3844 ptoa(pmap_wired_count(map->pmap)));
3845 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3846 }
3847 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3848 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3849 PROC_UNLOCK(p);
3850 }
3851 #endif
3852
3853 return (rv);
3854 }
3855
3856 /*
3857 * Unshare the specified VM space for exec. If other processes are
3858 * mapped to it, then create a new one. The new vmspace is null.
3859 */
3860 int
3861 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3862 {
3863 struct vmspace *oldvmspace = p->p_vmspace;
3864 struct vmspace *newvmspace;
3865
3866 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3867 ("vmspace_exec recursed"));
3868 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3869 if (newvmspace == NULL)
3870 return (ENOMEM);
3871 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3872 /*
3873 * This code is written like this for prototype purposes. The
3874 * goal is to avoid running down the vmspace here, but let the
3875 * other process's that are still using the vmspace to finally
3876 * run it down. Even though there is little or no chance of blocking
3877 * here, it is a good idea to keep this form for future mods.
3878 */
3879 PROC_VMSPACE_LOCK(p);
3880 p->p_vmspace = newvmspace;
3881 PROC_VMSPACE_UNLOCK(p);
3882 if (p == curthread->td_proc)
3883 pmap_activate(curthread);
3884 curthread->td_pflags |= TDP_EXECVMSPC;
3885 return (0);
3886 }
3887
3888 /*
3889 * Unshare the specified VM space for forcing COW. This
3890 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3891 */
3892 int
3893 vmspace_unshare(struct proc *p)
3894 {
3895 struct vmspace *oldvmspace = p->p_vmspace;
3896 struct vmspace *newvmspace;
3897 vm_ooffset_t fork_charge;
3898
3899 if (oldvmspace->vm_refcnt == 1)
3900 return (0);
3901 fork_charge = 0;
3902 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3903 if (newvmspace == NULL)
3904 return (ENOMEM);
3905 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3906 vmspace_free(newvmspace);
3907 return (ENOMEM);
3908 }
3909 PROC_VMSPACE_LOCK(p);
3910 p->p_vmspace = newvmspace;
3911 PROC_VMSPACE_UNLOCK(p);
3912 if (p == curthread->td_proc)
3913 pmap_activate(curthread);
3914 vmspace_free(oldvmspace);
3915 return (0);
3916 }
3917
3918 /*
3919 * vm_map_lookup:
3920 *
3921 * Finds the VM object, offset, and
3922 * protection for a given virtual address in the
3923 * specified map, assuming a page fault of the
3924 * type specified.
3925 *
3926 * Leaves the map in question locked for read; return
3927 * values are guaranteed until a vm_map_lookup_done
3928 * call is performed. Note that the map argument
3929 * is in/out; the returned map must be used in
3930 * the call to vm_map_lookup_done.
3931 *
3932 * A handle (out_entry) is returned for use in
3933 * vm_map_lookup_done, to make that fast.
3934 *
3935 * If a lookup is requested with "write protection"
3936 * specified, the map may be changed to perform virtual
3937 * copying operations, although the data referenced will
3938 * remain the same.
3939 */
3940 int
3941 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3942 vm_offset_t vaddr,
3943 vm_prot_t fault_typea,
3944 vm_map_entry_t *out_entry, /* OUT */
3945 vm_object_t *object, /* OUT */
3946 vm_pindex_t *pindex, /* OUT */
3947 vm_prot_t *out_prot, /* OUT */
3948 boolean_t *wired) /* OUT */
3949 {
3950 vm_map_entry_t entry;
3951 vm_map_t map = *var_map;
3952 vm_prot_t prot;
3953 vm_prot_t fault_type = fault_typea;
3954 vm_object_t eobject;
3955 vm_size_t size;
3956 struct ucred *cred;
3957
3958 RetryLookup:;
3959
3960 vm_map_lock_read(map);
3961
3962 /*
3963 * Lookup the faulting address.
3964 */
3965 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3966 vm_map_unlock_read(map);
3967 return (KERN_INVALID_ADDRESS);
3968 }
3969
3970 entry = *out_entry;
3971
3972 /*
3973 * Handle submaps.
3974 */
3975 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3976 vm_map_t old_map = map;
3977
3978 *var_map = map = entry->object.sub_map;
3979 vm_map_unlock_read(old_map);
3980 goto RetryLookup;
3981 }
3982
3983 /*
3984 * Check whether this task is allowed to have this page.
3985 */
3986 prot = entry->protection;
3987 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3988 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
3989 vm_map_unlock_read(map);
3990 return (KERN_PROTECTION_FAILURE);
3991 }
3992 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
3993 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
3994 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
3995 ("entry %p flags %x", entry, entry->eflags));
3996 if ((fault_typea & VM_PROT_COPY) != 0 &&
3997 (entry->max_protection & VM_PROT_WRITE) == 0 &&
3998 (entry->eflags & MAP_ENTRY_COW) == 0) {
3999 vm_map_unlock_read(map);
4000 return (KERN_PROTECTION_FAILURE);
4001 }
4002
4003 /*
4004 * If this page is not pageable, we have to get it for all possible
4005 * accesses.
4006 */
4007 *wired = (entry->wired_count != 0);
4008 if (*wired)
4009 fault_type = entry->protection;
4010 size = entry->end - entry->start;
4011 /*
4012 * If the entry was copy-on-write, we either ...
4013 */
4014 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4015 /*
4016 * If we want to write the page, we may as well handle that
4017 * now since we've got the map locked.
4018 *
4019 * If we don't need to write the page, we just demote the
4020 * permissions allowed.
4021 */
4022 if ((fault_type & VM_PROT_WRITE) != 0 ||
4023 (fault_typea & VM_PROT_COPY) != 0) {
4024 /*
4025 * Make a new object, and place it in the object
4026 * chain. Note that no new references have appeared
4027 * -- one just moved from the map to the new
4028 * object.
4029 */
4030 if (vm_map_lock_upgrade(map))
4031 goto RetryLookup;
4032
4033 if (entry->cred == NULL) {
4034 /*
4035 * The debugger owner is charged for
4036 * the memory.
4037 */
4038 cred = curthread->td_ucred;
4039 crhold(cred);
4040 if (!swap_reserve_by_cred(size, cred)) {
4041 crfree(cred);
4042 vm_map_unlock(map);
4043 return (KERN_RESOURCE_SHORTAGE);
4044 }
4045 entry->cred = cred;
4046 }
4047 vm_object_shadow(&entry->object.vm_object,
4048 &entry->offset, size);
4049 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4050 eobject = entry->object.vm_object;
4051 if (eobject->cred != NULL) {
4052 /*
4053 * The object was not shadowed.
4054 */
4055 swap_release_by_cred(size, entry->cred);
4056 crfree(entry->cred);
4057 entry->cred = NULL;
4058 } else if (entry->cred != NULL) {
4059 VM_OBJECT_WLOCK(eobject);
4060 eobject->cred = entry->cred;
4061 eobject->charge = size;
4062 VM_OBJECT_WUNLOCK(eobject);
4063 entry->cred = NULL;
4064 }
4065
4066 vm_map_lock_downgrade(map);
4067 } else {
4068 /*
4069 * We're attempting to read a copy-on-write page --
4070 * don't allow writes.
4071 */
4072 prot &= ~VM_PROT_WRITE;
4073 }
4074 }
4075
4076 /*
4077 * Create an object if necessary.
4078 */
4079 if (entry->object.vm_object == NULL &&
4080 !map->system_map) {
4081 if (vm_map_lock_upgrade(map))
4082 goto RetryLookup;
4083 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4084 atop(size));
4085 entry->offset = 0;
4086 if (entry->cred != NULL) {
4087 VM_OBJECT_WLOCK(entry->object.vm_object);
4088 entry->object.vm_object->cred = entry->cred;
4089 entry->object.vm_object->charge = size;
4090 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4091 entry->cred = NULL;
4092 }
4093 vm_map_lock_downgrade(map);
4094 }
4095
4096 /*
4097 * Return the object/offset from this entry. If the entry was
4098 * copy-on-write or empty, it has been fixed up.
4099 */
4100 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4101 *object = entry->object.vm_object;
4102
4103 *out_prot = prot;
4104 return (KERN_SUCCESS);
4105 }
4106
4107 /*
4108 * vm_map_lookup_locked:
4109 *
4110 * Lookup the faulting address. A version of vm_map_lookup that returns
4111 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4112 */
4113 int
4114 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4115 vm_offset_t vaddr,
4116 vm_prot_t fault_typea,
4117 vm_map_entry_t *out_entry, /* OUT */
4118 vm_object_t *object, /* OUT */
4119 vm_pindex_t *pindex, /* OUT */
4120 vm_prot_t *out_prot, /* OUT */
4121 boolean_t *wired) /* OUT */
4122 {
4123 vm_map_entry_t entry;
4124 vm_map_t map = *var_map;
4125 vm_prot_t prot;
4126 vm_prot_t fault_type = fault_typea;
4127
4128 /*
4129 * Lookup the faulting address.
4130 */
4131 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4132 return (KERN_INVALID_ADDRESS);
4133
4134 entry = *out_entry;
4135
4136 /*
4137 * Fail if the entry refers to a submap.
4138 */
4139 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4140 return (KERN_FAILURE);
4141
4142 /*
4143 * Check whether this task is allowed to have this page.
4144 */
4145 prot = entry->protection;
4146 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4147 if ((fault_type & prot) != fault_type)
4148 return (KERN_PROTECTION_FAILURE);
4149
4150 /*
4151 * If this page is not pageable, we have to get it for all possible
4152 * accesses.
4153 */
4154 *wired = (entry->wired_count != 0);
4155 if (*wired)
4156 fault_type = entry->protection;
4157
4158 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4159 /*
4160 * Fail if the entry was copy-on-write for a write fault.
4161 */
4162 if (fault_type & VM_PROT_WRITE)
4163 return (KERN_FAILURE);
4164 /*
4165 * We're attempting to read a copy-on-write page --
4166 * don't allow writes.
4167 */
4168 prot &= ~VM_PROT_WRITE;
4169 }
4170
4171 /*
4172 * Fail if an object should be created.
4173 */
4174 if (entry->object.vm_object == NULL && !map->system_map)
4175 return (KERN_FAILURE);
4176
4177 /*
4178 * Return the object/offset from this entry. If the entry was
4179 * copy-on-write or empty, it has been fixed up.
4180 */
4181 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4182 *object = entry->object.vm_object;
4183
4184 *out_prot = prot;
4185 return (KERN_SUCCESS);
4186 }
4187
4188 /*
4189 * vm_map_lookup_done:
4190 *
4191 * Releases locks acquired by a vm_map_lookup
4192 * (according to the handle returned by that lookup).
4193 */
4194 void
4195 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4196 {
4197 /*
4198 * Unlock the main-level map
4199 */
4200 vm_map_unlock_read(map);
4201 }
4202
4203 #include "opt_ddb.h"
4204 #ifdef DDB
4205 #include <sys/kernel.h>
4206
4207 #include <ddb/ddb.h>
4208
4209 static void
4210 vm_map_print(vm_map_t map)
4211 {
4212 vm_map_entry_t entry;
4213
4214 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4215 (void *)map,
4216 (void *)map->pmap, map->nentries, map->timestamp);
4217
4218 db_indent += 2;
4219 for (entry = map->header.next; entry != &map->header;
4220 entry = entry->next) {
4221 db_iprintf("map entry %p: start=%p, end=%p\n",
4222 (void *)entry, (void *)entry->start, (void *)entry->end);
4223 {
4224 static char *inheritance_name[4] =
4225 {"share", "copy", "none", "donate_copy"};
4226
4227 db_iprintf(" prot=%x/%x/%s",
4228 entry->protection,
4229 entry->max_protection,
4230 inheritance_name[(int)(unsigned char)entry->inheritance]);
4231 if (entry->wired_count != 0)
4232 db_printf(", wired");
4233 }
4234 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4235 db_printf(", share=%p, offset=0x%jx\n",
4236 (void *)entry->object.sub_map,
4237 (uintmax_t)entry->offset);
4238 if ((entry->prev == &map->header) ||
4239 (entry->prev->object.sub_map !=
4240 entry->object.sub_map)) {
4241 db_indent += 2;
4242 vm_map_print((vm_map_t)entry->object.sub_map);
4243 db_indent -= 2;
4244 }
4245 } else {
4246 if (entry->cred != NULL)
4247 db_printf(", ruid %d", entry->cred->cr_ruid);
4248 db_printf(", object=%p, offset=0x%jx",
4249 (void *)entry->object.vm_object,
4250 (uintmax_t)entry->offset);
4251 if (entry->object.vm_object && entry->object.vm_object->cred)
4252 db_printf(", obj ruid %d charge %jx",
4253 entry->object.vm_object->cred->cr_ruid,
4254 (uintmax_t)entry->object.vm_object->charge);
4255 if (entry->eflags & MAP_ENTRY_COW)
4256 db_printf(", copy (%s)",
4257 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4258 db_printf("\n");
4259
4260 if ((entry->prev == &map->header) ||
4261 (entry->prev->object.vm_object !=
4262 entry->object.vm_object)) {
4263 db_indent += 2;
4264 vm_object_print((db_expr_t)(intptr_t)
4265 entry->object.vm_object,
4266 0, 0, (char *)0);
4267 db_indent -= 2;
4268 }
4269 }
4270 }
4271 db_indent -= 2;
4272 }
4273
4274 DB_SHOW_COMMAND(map, map)
4275 {
4276
4277 if (!have_addr) {
4278 db_printf("usage: show map <addr>\n");
4279 return;
4280 }
4281 vm_map_print((vm_map_t)addr);
4282 }
4283
4284 DB_SHOW_COMMAND(procvm, procvm)
4285 {
4286 struct proc *p;
4287
4288 if (have_addr) {
4289 p = (struct proc *) addr;
4290 } else {
4291 p = curproc;
4292 }
4293
4294 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4295 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4296 (void *)vmspace_pmap(p->p_vmspace));
4297
4298 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);
4299 }
4300
4301 #endif /* DDB */
Cache object: 0dc75d891d2054da11c1548ab3a00a39
|