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("$FreeBSD$");
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 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
137 vm_map_entry_t gap_entry);
138 #ifdef INVARIANTS
139 static void vm_map_zdtor(void *mem, int size, void *arg);
140 static void vmspace_zdtor(void *mem, int size, void *arg);
141 #endif
142 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
143 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
144 int cow);
145 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
146 vm_offset_t failed_addr);
147
148 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
149 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
150 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
151
152 /*
153 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
154 * stable.
155 */
156 #define PROC_VMSPACE_LOCK(p) do { } while (0)
157 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
158
159 /*
160 * VM_MAP_RANGE_CHECK: [ internal use only ]
161 *
162 * Asserts that the starting and ending region
163 * addresses fall within the valid range of the map.
164 */
165 #define VM_MAP_RANGE_CHECK(map, start, end) \
166 { \
167 if (start < vm_map_min(map)) \
168 start = vm_map_min(map); \
169 if (end > vm_map_max(map)) \
170 end = vm_map_max(map); \
171 if (start > end) \
172 start = end; \
173 }
174
175 /*
176 * vm_map_startup:
177 *
178 * Initialize the vm_map module. Must be called before
179 * any other vm_map routines.
180 *
181 * Map and entry structures are allocated from the general
182 * purpose memory pool with some exceptions:
183 *
184 * - The kernel map and kmem submap are allocated statically.
185 * - Kernel map entries are allocated out of a static pool.
186 *
187 * These restrictions are necessary since malloc() uses the
188 * maps and requires map entries.
189 */
190
191 void
192 vm_map_startup(void)
193 {
194 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
195 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
196 #ifdef INVARIANTS
197 vm_map_zdtor,
198 #else
199 NULL,
200 #endif
201 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
202 uma_prealloc(mapzone, MAX_KMAP);
203 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
204 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
205 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
206 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
207 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
208 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
209 #ifdef INVARIANTS
210 vmspace_zdtor,
211 #else
212 NULL,
213 #endif
214 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
215 }
216
217 static int
218 vmspace_zinit(void *mem, int size, int flags)
219 {
220 struct vmspace *vm;
221
222 vm = (struct vmspace *)mem;
223
224 vm->vm_map.pmap = NULL;
225 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
226 PMAP_LOCK_INIT(vmspace_pmap(vm));
227 return (0);
228 }
229
230 static int
231 vm_map_zinit(void *mem, int size, int flags)
232 {
233 vm_map_t map;
234
235 map = (vm_map_t)mem;
236 memset(map, 0, sizeof(*map));
237 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
238 sx_init(&map->lock, "vm map (user)");
239 return (0);
240 }
241
242 #ifdef INVARIANTS
243 static void
244 vmspace_zdtor(void *mem, int size, void *arg)
245 {
246 struct vmspace *vm;
247
248 vm = (struct vmspace *)mem;
249
250 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
251 }
252 static void
253 vm_map_zdtor(void *mem, int size, void *arg)
254 {
255 vm_map_t map;
256
257 map = (vm_map_t)mem;
258 KASSERT(map->nentries == 0,
259 ("map %p nentries == %d on free.",
260 map, map->nentries));
261 KASSERT(map->size == 0,
262 ("map %p size == %lu on free.",
263 map, (unsigned long)map->size));
264 }
265 #endif /* INVARIANTS */
266
267 /*
268 * Allocate a vmspace structure, including a vm_map and pmap,
269 * and initialize those structures. The refcnt is set to 1.
270 *
271 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
272 */
273 struct vmspace *
274 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
275 {
276 struct vmspace *vm;
277
278 vm = uma_zalloc(vmspace_zone, M_WAITOK);
279
280 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
281
282 if (pinit == NULL)
283 pinit = &pmap_pinit;
284
285 if (!pinit(vmspace_pmap(vm))) {
286 uma_zfree(vmspace_zone, vm);
287 return (NULL);
288 }
289 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
290 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
291 vm->vm_refcnt = 1;
292 vm->vm_shm = NULL;
293 vm->vm_swrss = 0;
294 vm->vm_tsize = 0;
295 vm->vm_dsize = 0;
296 vm->vm_ssize = 0;
297 vm->vm_taddr = 0;
298 vm->vm_daddr = 0;
299 vm->vm_maxsaddr = 0;
300 return (vm);
301 }
302
303 #ifdef RACCT
304 static void
305 vmspace_container_reset(struct proc *p)
306 {
307
308 PROC_LOCK(p);
309 racct_set(p, RACCT_DATA, 0);
310 racct_set(p, RACCT_STACK, 0);
311 racct_set(p, RACCT_RSS, 0);
312 racct_set(p, RACCT_MEMLOCK, 0);
313 racct_set(p, RACCT_VMEM, 0);
314 PROC_UNLOCK(p);
315 }
316 #endif
317
318 static inline void
319 vmspace_dofree(struct vmspace *vm)
320 {
321
322 CTR1(KTR_VM, "vmspace_free: %p", vm);
323
324 /*
325 * Make sure any SysV shm is freed, it might not have been in
326 * exit1().
327 */
328 shmexit(vm);
329
330 /*
331 * Lock the map, to wait out all other references to it.
332 * Delete all of the mappings and pages they hold, then call
333 * the pmap module to reclaim anything left.
334 */
335 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
336 vm->vm_map.max_offset);
337
338 pmap_release(vmspace_pmap(vm));
339 vm->vm_map.pmap = NULL;
340 uma_zfree(vmspace_zone, vm);
341 }
342
343 void
344 vmspace_free(struct vmspace *vm)
345 {
346
347 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
348 "vmspace_free() called with non-sleepable lock held");
349
350 if (vm->vm_refcnt == 0)
351 panic("vmspace_free: attempt to free already freed vmspace");
352
353 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
354 vmspace_dofree(vm);
355 }
356
357 void
358 vmspace_exitfree(struct proc *p)
359 {
360 struct vmspace *vm;
361
362 PROC_VMSPACE_LOCK(p);
363 vm = p->p_vmspace;
364 p->p_vmspace = NULL;
365 PROC_VMSPACE_UNLOCK(p);
366 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
367 vmspace_free(vm);
368 }
369
370 void
371 vmspace_exit(struct thread *td)
372 {
373 int refcnt;
374 struct vmspace *vm;
375 struct proc *p;
376
377 /*
378 * Release user portion of address space.
379 * This releases references to vnodes,
380 * which could cause I/O if the file has been unlinked.
381 * Need to do this early enough that we can still sleep.
382 *
383 * The last exiting process to reach this point releases as
384 * much of the environment as it can. vmspace_dofree() is the
385 * slower fallback in case another process had a temporary
386 * reference to the vmspace.
387 */
388
389 p = td->td_proc;
390 vm = p->p_vmspace;
391 atomic_add_int(&vmspace0.vm_refcnt, 1);
392 do {
393 refcnt = vm->vm_refcnt;
394 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
395 /* Switch now since other proc might free vmspace */
396 PROC_VMSPACE_LOCK(p);
397 p->p_vmspace = &vmspace0;
398 PROC_VMSPACE_UNLOCK(p);
399 pmap_activate(td);
400 }
401 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
402 if (refcnt == 1) {
403 if (p->p_vmspace != vm) {
404 /* vmspace not yet freed, switch back */
405 PROC_VMSPACE_LOCK(p);
406 p->p_vmspace = vm;
407 PROC_VMSPACE_UNLOCK(p);
408 pmap_activate(td);
409 }
410 pmap_remove_pages(vmspace_pmap(vm));
411 /* Switch now since this proc will free vmspace */
412 PROC_VMSPACE_LOCK(p);
413 p->p_vmspace = &vmspace0;
414 PROC_VMSPACE_UNLOCK(p);
415 pmap_activate(td);
416 vmspace_dofree(vm);
417 }
418 #ifdef RACCT
419 if (racct_enable)
420 vmspace_container_reset(p);
421 #endif
422 }
423
424 /* Acquire reference to vmspace owned by another process. */
425
426 struct vmspace *
427 vmspace_acquire_ref(struct proc *p)
428 {
429 struct vmspace *vm;
430 int refcnt;
431
432 PROC_VMSPACE_LOCK(p);
433 vm = p->p_vmspace;
434 if (vm == NULL) {
435 PROC_VMSPACE_UNLOCK(p);
436 return (NULL);
437 }
438 do {
439 refcnt = vm->vm_refcnt;
440 if (refcnt <= 0) { /* Avoid 0->1 transition */
441 PROC_VMSPACE_UNLOCK(p);
442 return (NULL);
443 }
444 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
445 if (vm != p->p_vmspace) {
446 PROC_VMSPACE_UNLOCK(p);
447 vmspace_free(vm);
448 return (NULL);
449 }
450 PROC_VMSPACE_UNLOCK(p);
451 return (vm);
452 }
453
454 void
455 _vm_map_lock(vm_map_t map, const char *file, int line)
456 {
457
458 if (map->system_map)
459 mtx_lock_flags_(&map->system_mtx, 0, file, line);
460 else
461 sx_xlock_(&map->lock, file, line);
462 map->timestamp++;
463 }
464
465 static void
466 vm_map_process_deferred(void)
467 {
468 struct thread *td;
469 vm_map_entry_t entry, next;
470 vm_object_t object;
471
472 td = curthread;
473 entry = td->td_map_def_user;
474 td->td_map_def_user = NULL;
475 while (entry != NULL) {
476 next = entry->next;
477 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
478 /*
479 * Decrement the object's writemappings and
480 * possibly the vnode's v_writecount.
481 */
482 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
483 ("Submap with writecount"));
484 object = entry->object.vm_object;
485 KASSERT(object != NULL, ("No object for writecount"));
486 vnode_pager_release_writecount(object, entry->start,
487 entry->end);
488 }
489 vm_map_entry_deallocate(entry, FALSE);
490 entry = next;
491 }
492 }
493
494 void
495 _vm_map_unlock(vm_map_t map, const char *file, int line)
496 {
497
498 if (map->system_map)
499 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
500 else {
501 sx_xunlock_(&map->lock, file, line);
502 vm_map_process_deferred();
503 }
504 }
505
506 void
507 _vm_map_lock_read(vm_map_t map, const char *file, int line)
508 {
509
510 if (map->system_map)
511 mtx_lock_flags_(&map->system_mtx, 0, file, line);
512 else
513 sx_slock_(&map->lock, file, line);
514 }
515
516 void
517 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
518 {
519
520 if (map->system_map)
521 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
522 else {
523 sx_sunlock_(&map->lock, file, line);
524 vm_map_process_deferred();
525 }
526 }
527
528 int
529 _vm_map_trylock(vm_map_t map, const char *file, int line)
530 {
531 int error;
532
533 error = map->system_map ?
534 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
535 !sx_try_xlock_(&map->lock, file, line);
536 if (error == 0)
537 map->timestamp++;
538 return (error == 0);
539 }
540
541 int
542 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
543 {
544 int error;
545
546 error = map->system_map ?
547 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
548 !sx_try_slock_(&map->lock, file, line);
549 return (error == 0);
550 }
551
552 /*
553 * _vm_map_lock_upgrade: [ internal use only ]
554 *
555 * Tries to upgrade a read (shared) lock on the specified map to a write
556 * (exclusive) lock. Returns the value "" if the upgrade succeeds and a
557 * non-zero value if the upgrade fails. If the upgrade fails, the map is
558 * returned without a read or write lock held.
559 *
560 * Requires that the map be read locked.
561 */
562 int
563 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
564 {
565 unsigned int last_timestamp;
566
567 if (map->system_map) {
568 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
569 } else {
570 if (!sx_try_upgrade_(&map->lock, file, line)) {
571 last_timestamp = map->timestamp;
572 sx_sunlock_(&map->lock, file, line);
573 vm_map_process_deferred();
574 /*
575 * If the map's timestamp does not change while the
576 * map is unlocked, then the upgrade succeeds.
577 */
578 sx_xlock_(&map->lock, file, line);
579 if (last_timestamp != map->timestamp) {
580 sx_xunlock_(&map->lock, file, line);
581 return (1);
582 }
583 }
584 }
585 map->timestamp++;
586 return (0);
587 }
588
589 void
590 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
591 {
592
593 if (map->system_map) {
594 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
595 } else
596 sx_downgrade_(&map->lock, file, line);
597 }
598
599 /*
600 * vm_map_locked:
601 *
602 * Returns a non-zero value if the caller holds a write (exclusive) lock
603 * on the specified map and the value "" otherwise.
604 */
605 int
606 vm_map_locked(vm_map_t map)
607 {
608
609 if (map->system_map)
610 return (mtx_owned(&map->system_mtx));
611 else
612 return (sx_xlocked(&map->lock));
613 }
614
615 #ifdef INVARIANTS
616 static void
617 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
618 {
619
620 if (map->system_map)
621 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
622 else
623 sx_assert_(&map->lock, SA_XLOCKED, file, line);
624 }
625
626 #define VM_MAP_ASSERT_LOCKED(map) \
627 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
628 #else
629 #define VM_MAP_ASSERT_LOCKED(map)
630 #endif
631
632 /*
633 * _vm_map_unlock_and_wait:
634 *
635 * Atomically releases the lock on the specified map and puts the calling
636 * thread to sleep. The calling thread will remain asleep until either
637 * vm_map_wakeup() is performed on the map or the specified timeout is
638 * exceeded.
639 *
640 * WARNING! This function does not perform deferred deallocations of
641 * objects and map entries. Therefore, the calling thread is expected to
642 * reacquire the map lock after reawakening and later perform an ordinary
643 * unlock operation, such as vm_map_unlock(), before completing its
644 * operation on the map.
645 */
646 int
647 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
648 {
649
650 mtx_lock(&map_sleep_mtx);
651 if (map->system_map)
652 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
653 else
654 sx_xunlock_(&map->lock, file, line);
655 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
656 timo));
657 }
658
659 /*
660 * vm_map_wakeup:
661 *
662 * Awaken any threads that have slept on the map using
663 * vm_map_unlock_and_wait().
664 */
665 void
666 vm_map_wakeup(vm_map_t map)
667 {
668
669 /*
670 * Acquire and release map_sleep_mtx to prevent a wakeup()
671 * from being performed (and lost) between the map unlock
672 * and the msleep() in _vm_map_unlock_and_wait().
673 */
674 mtx_lock(&map_sleep_mtx);
675 mtx_unlock(&map_sleep_mtx);
676 wakeup(&map->root);
677 }
678
679 void
680 vm_map_busy(vm_map_t map)
681 {
682
683 VM_MAP_ASSERT_LOCKED(map);
684 map->busy++;
685 }
686
687 void
688 vm_map_unbusy(vm_map_t map)
689 {
690
691 VM_MAP_ASSERT_LOCKED(map);
692 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
693 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
694 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
695 wakeup(&map->busy);
696 }
697 }
698
699 void
700 vm_map_wait_busy(vm_map_t map)
701 {
702
703 VM_MAP_ASSERT_LOCKED(map);
704 while (map->busy) {
705 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
706 if (map->system_map)
707 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
708 else
709 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
710 }
711 map->timestamp++;
712 }
713
714 long
715 vmspace_resident_count(struct vmspace *vmspace)
716 {
717 return pmap_resident_count(vmspace_pmap(vmspace));
718 }
719
720 /*
721 * vm_map_create:
722 *
723 * Creates and returns a new empty VM map with
724 * the given physical map structure, and having
725 * the given lower and upper address bounds.
726 */
727 vm_map_t
728 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
729 {
730 vm_map_t result;
731
732 result = uma_zalloc(mapzone, M_WAITOK);
733 CTR1(KTR_VM, "vm_map_create: %p", result);
734 _vm_map_init(result, pmap, min, max);
735 return (result);
736 }
737
738 /*
739 * Initialize an existing vm_map structure
740 * such as that in the vmspace structure.
741 */
742 static void
743 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
744 {
745
746 map->header.next = map->header.prev = &map->header;
747 map->needs_wakeup = FALSE;
748 map->system_map = 0;
749 map->pmap = pmap;
750 map->min_offset = min;
751 map->max_offset = max;
752 map->flags = 0;
753 map->root = NULL;
754 map->timestamp = 0;
755 map->busy = 0;
756 }
757
758 void
759 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
760 {
761
762 _vm_map_init(map, pmap, min, max);
763 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
764 sx_init(&map->lock, "user map");
765 }
766
767 /*
768 * vm_map_entry_dispose: [ internal use only ]
769 *
770 * Inverse of vm_map_entry_create.
771 */
772 static void
773 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
774 {
775 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
776 }
777
778 /*
779 * vm_map_entry_create: [ internal use only ]
780 *
781 * Allocates a VM map entry for insertion.
782 * No entry fields are filled in.
783 */
784 static vm_map_entry_t
785 vm_map_entry_create(vm_map_t map)
786 {
787 vm_map_entry_t new_entry;
788
789 if (map->system_map)
790 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
791 else
792 new_entry = uma_zalloc(mapentzone, M_WAITOK);
793 if (new_entry == NULL)
794 panic("vm_map_entry_create: kernel resources exhausted");
795 return (new_entry);
796 }
797
798 /*
799 * vm_map_entry_set_behavior:
800 *
801 * Set the expected access behavior, either normal, random, or
802 * sequential.
803 */
804 static inline void
805 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
806 {
807 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
808 (behavior & MAP_ENTRY_BEHAV_MASK);
809 }
810
811 /*
812 * vm_map_entry_set_max_free:
813 *
814 * Set the max_free field in a vm_map_entry.
815 */
816 static inline void
817 vm_map_entry_set_max_free(vm_map_entry_t entry)
818 {
819
820 entry->max_free = entry->adj_free;
821 if (entry->left != NULL && entry->left->max_free > entry->max_free)
822 entry->max_free = entry->left->max_free;
823 if (entry->right != NULL && entry->right->max_free > entry->max_free)
824 entry->max_free = entry->right->max_free;
825 }
826
827 /*
828 * vm_map_entry_splay:
829 *
830 * The Sleator and Tarjan top-down splay algorithm with the
831 * following variation. Max_free must be computed bottom-up, so
832 * on the downward pass, maintain the left and right spines in
833 * reverse order. Then, make a second pass up each side to fix
834 * the pointers and compute max_free. The time bound is O(log n)
835 * amortized.
836 *
837 * The new root is the vm_map_entry containing "addr", or else an
838 * adjacent entry (lower or higher) if addr is not in the tree.
839 *
840 * The map must be locked, and leaves it so.
841 *
842 * Returns: the new root.
843 */
844 static vm_map_entry_t
845 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
846 {
847 vm_map_entry_t llist, rlist;
848 vm_map_entry_t ltree, rtree;
849 vm_map_entry_t y;
850
851 /* Special case of empty tree. */
852 if (root == NULL)
853 return (root);
854
855 /*
856 * Pass One: Splay down the tree until we find addr or a NULL
857 * pointer where addr would go. llist and rlist are the two
858 * sides in reverse order (bottom-up), with llist linked by
859 * the right pointer and rlist linked by the left pointer in
860 * the vm_map_entry. Wait until Pass Two to set max_free on
861 * the two spines.
862 */
863 llist = NULL;
864 rlist = NULL;
865 for (;;) {
866 /* root is never NULL in here. */
867 if (addr < root->start) {
868 y = root->left;
869 if (y == NULL)
870 break;
871 if (addr < y->start && y->left != NULL) {
872 /* Rotate right and put y on rlist. */
873 root->left = y->right;
874 y->right = root;
875 vm_map_entry_set_max_free(root);
876 root = y->left;
877 y->left = rlist;
878 rlist = y;
879 } else {
880 /* Put root on rlist. */
881 root->left = rlist;
882 rlist = root;
883 root = y;
884 }
885 } else if (addr >= root->end) {
886 y = root->right;
887 if (y == NULL)
888 break;
889 if (addr >= y->end && y->right != NULL) {
890 /* Rotate left and put y on llist. */
891 root->right = y->left;
892 y->left = root;
893 vm_map_entry_set_max_free(root);
894 root = y->right;
895 y->right = llist;
896 llist = y;
897 } else {
898 /* Put root on llist. */
899 root->right = llist;
900 llist = root;
901 root = y;
902 }
903 } else
904 break;
905 }
906
907 /*
908 * Pass Two: Walk back up the two spines, flip the pointers
909 * and set max_free. The subtrees of the root go at the
910 * bottom of llist and rlist.
911 */
912 ltree = root->left;
913 while (llist != NULL) {
914 y = llist->right;
915 llist->right = ltree;
916 vm_map_entry_set_max_free(llist);
917 ltree = llist;
918 llist = y;
919 }
920 rtree = root->right;
921 while (rlist != NULL) {
922 y = rlist->left;
923 rlist->left = rtree;
924 vm_map_entry_set_max_free(rlist);
925 rtree = rlist;
926 rlist = y;
927 }
928
929 /*
930 * Final assembly: add ltree and rtree as subtrees of root.
931 */
932 root->left = ltree;
933 root->right = rtree;
934 vm_map_entry_set_max_free(root);
935
936 return (root);
937 }
938
939 /*
940 * vm_map_entry_{un,}link:
941 *
942 * Insert/remove entries from maps.
943 */
944 static void
945 vm_map_entry_link(vm_map_t map,
946 vm_map_entry_t after_where,
947 vm_map_entry_t entry)
948 {
949
950 CTR4(KTR_VM,
951 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
952 map->nentries, entry, after_where);
953 VM_MAP_ASSERT_LOCKED(map);
954 KASSERT(after_where == &map->header ||
955 after_where->end <= entry->start,
956 ("vm_map_entry_link: prev end %jx new start %jx overlap",
957 (uintmax_t)after_where->end, (uintmax_t)entry->start));
958 KASSERT(after_where->next == &map->header ||
959 entry->end <= after_where->next->start,
960 ("vm_map_entry_link: new end %jx next start %jx overlap",
961 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
962
963 map->nentries++;
964 entry->prev = after_where;
965 entry->next = after_where->next;
966 entry->next->prev = entry;
967 after_where->next = entry;
968
969 if (after_where != &map->header) {
970 if (after_where != map->root)
971 vm_map_entry_splay(after_where->start, map->root);
972 entry->right = after_where->right;
973 entry->left = after_where;
974 after_where->right = NULL;
975 after_where->adj_free = entry->start - after_where->end;
976 vm_map_entry_set_max_free(after_where);
977 } else {
978 entry->right = map->root;
979 entry->left = NULL;
980 }
981 entry->adj_free = (entry->next == &map->header ? map->max_offset :
982 entry->next->start) - entry->end;
983 vm_map_entry_set_max_free(entry);
984 map->root = entry;
985 }
986
987 static void
988 vm_map_entry_unlink(vm_map_t map,
989 vm_map_entry_t entry)
990 {
991 vm_map_entry_t next, prev, root;
992
993 VM_MAP_ASSERT_LOCKED(map);
994 if (entry != map->root)
995 vm_map_entry_splay(entry->start, map->root);
996 if (entry->left == NULL)
997 root = entry->right;
998 else {
999 root = vm_map_entry_splay(entry->start, entry->left);
1000 root->right = entry->right;
1001 root->adj_free = (entry->next == &map->header ? map->max_offset :
1002 entry->next->start) - root->end;
1003 vm_map_entry_set_max_free(root);
1004 }
1005 map->root = root;
1006
1007 prev = entry->prev;
1008 next = entry->next;
1009 next->prev = prev;
1010 prev->next = next;
1011 map->nentries--;
1012 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1013 map->nentries, entry);
1014 }
1015
1016 /*
1017 * vm_map_entry_resize_free:
1018 *
1019 * Recompute the amount of free space following a vm_map_entry
1020 * and propagate that value up the tree. Call this function after
1021 * resizing a map entry in-place, that is, without a call to
1022 * vm_map_entry_link() or _unlink().
1023 *
1024 * The map must be locked, and leaves it so.
1025 */
1026 static void
1027 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1028 {
1029
1030 /*
1031 * Using splay trees without parent pointers, propagating
1032 * max_free up the tree is done by moving the entry to the
1033 * root and making the change there.
1034 */
1035 if (entry != map->root)
1036 map->root = vm_map_entry_splay(entry->start, map->root);
1037
1038 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1039 entry->next->start) - entry->end;
1040 vm_map_entry_set_max_free(entry);
1041 }
1042
1043 /*
1044 * vm_map_lookup_entry: [ internal use only ]
1045 *
1046 * Finds the map entry containing (or
1047 * immediately preceding) the specified address
1048 * in the given map; the entry is returned
1049 * in the "entry" parameter. The boolean
1050 * result indicates whether the address is
1051 * actually contained in the map.
1052 */
1053 boolean_t
1054 vm_map_lookup_entry(
1055 vm_map_t map,
1056 vm_offset_t address,
1057 vm_map_entry_t *entry) /* OUT */
1058 {
1059 vm_map_entry_t cur;
1060 boolean_t locked;
1061
1062 /*
1063 * If the map is empty, then the map entry immediately preceding
1064 * "address" is the map's header.
1065 */
1066 cur = map->root;
1067 if (cur == NULL)
1068 *entry = &map->header;
1069 else if (address >= cur->start && cur->end > address) {
1070 *entry = cur;
1071 return (TRUE);
1072 } else if ((locked = vm_map_locked(map)) ||
1073 sx_try_upgrade(&map->lock)) {
1074 /*
1075 * Splay requires a write lock on the map. However, it only
1076 * restructures the binary search tree; it does not otherwise
1077 * change the map. Thus, the map's timestamp need not change
1078 * on a temporary upgrade.
1079 */
1080 map->root = cur = vm_map_entry_splay(address, cur);
1081 if (!locked)
1082 sx_downgrade(&map->lock);
1083
1084 /*
1085 * If "address" is contained within a map entry, the new root
1086 * is that map entry. Otherwise, the new root is a map entry
1087 * immediately before or after "address".
1088 */
1089 if (address >= cur->start) {
1090 *entry = cur;
1091 if (cur->end > address)
1092 return (TRUE);
1093 } else
1094 *entry = cur->prev;
1095 } else
1096 /*
1097 * Since the map is only locked for read access, perform a
1098 * standard binary search tree lookup for "address".
1099 */
1100 for (;;) {
1101 if (address < cur->start) {
1102 if (cur->left == NULL) {
1103 *entry = cur->prev;
1104 break;
1105 }
1106 cur = cur->left;
1107 } else if (cur->end > address) {
1108 *entry = cur;
1109 return (TRUE);
1110 } else {
1111 if (cur->right == NULL) {
1112 *entry = cur;
1113 break;
1114 }
1115 cur = cur->right;
1116 }
1117 }
1118 return (FALSE);
1119 }
1120
1121 /*
1122 * vm_map_insert:
1123 *
1124 * Inserts the given whole VM object into the target
1125 * map at the specified address range. The object's
1126 * size should match that of the address range.
1127 *
1128 * Requires that the map be locked, and leaves it so.
1129 *
1130 * If object is non-NULL, ref count must be bumped by caller
1131 * prior to making call to account for the new entry.
1132 */
1133 int
1134 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1135 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1136 {
1137 vm_map_entry_t new_entry, prev_entry, temp_entry;
1138 struct ucred *cred;
1139 vm_eflags_t protoeflags;
1140 vm_inherit_t inheritance;
1141
1142 VM_MAP_ASSERT_LOCKED(map);
1143 KASSERT((object != kmem_object && object != kernel_object) ||
1144 (cow & MAP_COPY_ON_WRITE) == 0,
1145 ("vm_map_insert: kmem or kernel object and COW"));
1146 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1147 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1148 KASSERT((prot & ~max) == 0,
1149 ("prot %#x is not subset of max_prot %#x", prot, max));
1150
1151 /*
1152 * Check that the start and end points are not bogus.
1153 */
1154 if (start < map->min_offset || end > map->max_offset || start >= end)
1155 return (KERN_INVALID_ADDRESS);
1156
1157 /*
1158 * Find the entry prior to the proposed starting address; if it's part
1159 * of an existing entry, this range is bogus.
1160 */
1161 if (vm_map_lookup_entry(map, start, &temp_entry))
1162 return (KERN_NO_SPACE);
1163
1164 prev_entry = temp_entry;
1165
1166 /*
1167 * Assert that the next entry doesn't overlap the end point.
1168 */
1169 if (prev_entry->next != &map->header && prev_entry->next->start < end)
1170 return (KERN_NO_SPACE);
1171
1172 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1173 max != VM_PROT_NONE))
1174 return (KERN_INVALID_ARGUMENT);
1175
1176 protoeflags = 0;
1177 if (cow & MAP_COPY_ON_WRITE)
1178 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1179 if (cow & MAP_NOFAULT)
1180 protoeflags |= MAP_ENTRY_NOFAULT;
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_STACK_GROWS_DOWN)
1186 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1187 if (cow & MAP_STACK_GROWS_UP)
1188 protoeflags |= MAP_ENTRY_GROWS_UP;
1189 if (cow & MAP_VN_WRITECOUNT)
1190 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1191 if ((cow & MAP_CREATE_GUARD) != 0)
1192 protoeflags |= MAP_ENTRY_GUARD;
1193 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1194 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1195 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1196 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1197 if (cow & MAP_INHERIT_SHARE)
1198 inheritance = VM_INHERIT_SHARE;
1199 else
1200 inheritance = VM_INHERIT_DEFAULT;
1201
1202 cred = NULL;
1203 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1204 goto charged;
1205 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1206 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1207 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1208 return (KERN_RESOURCE_SHORTAGE);
1209 KASSERT(object == NULL ||
1210 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1211 object->cred == NULL,
1212 ("overcommit: vm_map_insert o %p", object));
1213 cred = curthread->td_ucred;
1214 }
1215
1216 charged:
1217 /* Expand the kernel pmap, if necessary. */
1218 if (map == kernel_map && end > kernel_vm_end)
1219 pmap_growkernel(end);
1220 if (object != NULL) {
1221 /*
1222 * OBJ_ONEMAPPING must be cleared unless this mapping
1223 * is trivially proven to be the only mapping for any
1224 * of the object's pages. (Object granularity
1225 * reference counting is insufficient to recognize
1226 * aliases with precision.)
1227 */
1228 VM_OBJECT_WLOCK(object);
1229 if (object->ref_count > 1 || object->shadow_count != 0)
1230 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1231 VM_OBJECT_WUNLOCK(object);
1232 } else if (prev_entry != &map->header &&
1233 prev_entry->eflags == protoeflags &&
1234 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1235 prev_entry->end == start && prev_entry->wired_count == 0 &&
1236 (prev_entry->cred == cred ||
1237 (prev_entry->object.vm_object != NULL &&
1238 prev_entry->object.vm_object->cred == cred)) &&
1239 vm_object_coalesce(prev_entry->object.vm_object,
1240 prev_entry->offset,
1241 (vm_size_t)(prev_entry->end - prev_entry->start),
1242 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1243 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1244 /*
1245 * We were able to extend the object. Determine if we
1246 * can extend the previous map entry to include the
1247 * new range as well.
1248 */
1249 if (prev_entry->inheritance == inheritance &&
1250 prev_entry->protection == prot &&
1251 prev_entry->max_protection == max) {
1252 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1253 map->size += end - prev_entry->end;
1254 prev_entry->end = end;
1255 vm_map_entry_resize_free(map, prev_entry);
1256 vm_map_simplify_entry(map, prev_entry);
1257 return (KERN_SUCCESS);
1258 }
1259
1260 /*
1261 * If we can extend the object but cannot extend the
1262 * map entry, we have to create a new map entry. We
1263 * must bump the ref count on the extended object to
1264 * account for it. object may be NULL.
1265 */
1266 object = prev_entry->object.vm_object;
1267 offset = prev_entry->offset +
1268 (prev_entry->end - prev_entry->start);
1269 vm_object_reference(object);
1270 if (cred != NULL && object != NULL && object->cred != NULL &&
1271 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1272 /* Object already accounts for this uid. */
1273 cred = NULL;
1274 }
1275 }
1276 if (cred != NULL)
1277 crhold(cred);
1278
1279 /*
1280 * Create a new entry
1281 */
1282 new_entry = vm_map_entry_create(map);
1283 new_entry->start = start;
1284 new_entry->end = end;
1285 new_entry->cred = NULL;
1286
1287 new_entry->eflags = protoeflags;
1288 new_entry->object.vm_object = object;
1289 new_entry->offset = offset;
1290
1291 new_entry->inheritance = inheritance;
1292 new_entry->protection = prot;
1293 new_entry->max_protection = max;
1294 new_entry->wired_count = 0;
1295 new_entry->wiring_thread = NULL;
1296 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1297 new_entry->next_read = OFF_TO_IDX(offset);
1298
1299 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1300 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1301 new_entry->cred = cred;
1302
1303 /*
1304 * Insert the new entry into the list
1305 */
1306 vm_map_entry_link(map, prev_entry, new_entry);
1307 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1308 map->size += new_entry->end - new_entry->start;
1309
1310 /*
1311 * Try to coalesce the new entry with both the previous and next
1312 * entries in the list. Previously, we only attempted to coalesce
1313 * with the previous entry when object is NULL. Here, we handle the
1314 * other cases, which are less common.
1315 */
1316 vm_map_simplify_entry(map, new_entry);
1317
1318 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1319 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1320 end - start, cow & MAP_PREFAULT_PARTIAL);
1321 }
1322
1323 return (KERN_SUCCESS);
1324 }
1325
1326 /*
1327 * vm_map_findspace:
1328 *
1329 * Find the first fit (lowest VM address) for "length" free bytes
1330 * beginning at address >= start in the given map.
1331 *
1332 * In a vm_map_entry, "adj_free" is the amount of free space
1333 * adjacent (higher address) to this entry, and "max_free" is the
1334 * maximum amount of contiguous free space in its subtree. This
1335 * allows finding a free region in one path down the tree, so
1336 * O(log n) amortized with splay trees.
1337 *
1338 * The map must be locked, and leaves it so.
1339 *
1340 * Returns: 0 on success, and starting address in *addr,
1341 * 1 if insufficient space.
1342 */
1343 int
1344 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1345 vm_offset_t *addr) /* OUT */
1346 {
1347 vm_map_entry_t entry;
1348 vm_offset_t st;
1349
1350 /*
1351 * Request must fit within min/max VM address and must avoid
1352 * address wrap.
1353 */
1354 if (start < map->min_offset)
1355 start = map->min_offset;
1356 if (start + length > map->max_offset || start + length < start)
1357 return (1);
1358
1359 /* Empty tree means wide open address space. */
1360 if (map->root == NULL) {
1361 *addr = start;
1362 return (0);
1363 }
1364
1365 /*
1366 * After splay, if start comes before root node, then there
1367 * must be a gap from start to the root.
1368 */
1369 map->root = vm_map_entry_splay(start, map->root);
1370 if (start + length <= map->root->start) {
1371 *addr = start;
1372 return (0);
1373 }
1374
1375 /*
1376 * Root is the last node that might begin its gap before
1377 * start, and this is the last comparison where address
1378 * wrap might be a problem.
1379 */
1380 st = (start > map->root->end) ? start : map->root->end;
1381 if (length <= map->root->end + map->root->adj_free - st) {
1382 *addr = st;
1383 return (0);
1384 }
1385
1386 /* With max_free, can immediately tell if no solution. */
1387 entry = map->root->right;
1388 if (entry == NULL || length > entry->max_free)
1389 return (1);
1390
1391 /*
1392 * Search the right subtree in the order: left subtree, root,
1393 * right subtree (first fit). The previous splay implies that
1394 * all regions in the right subtree have addresses > start.
1395 */
1396 while (entry != NULL) {
1397 if (entry->left != NULL && entry->left->max_free >= length)
1398 entry = entry->left;
1399 else if (entry->adj_free >= length) {
1400 *addr = entry->end;
1401 return (0);
1402 } else
1403 entry = entry->right;
1404 }
1405
1406 /* Can't get here, so panic if we do. */
1407 panic("vm_map_findspace: max_free corrupt");
1408 }
1409
1410 int
1411 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1412 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1413 vm_prot_t max, int cow)
1414 {
1415 vm_offset_t end;
1416 int result;
1417
1418 end = start + length;
1419 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1420 object == NULL,
1421 ("vm_map_fixed: non-NULL backing object for stack"));
1422 vm_map_lock(map);
1423 VM_MAP_RANGE_CHECK(map, start, end);
1424 if ((cow & MAP_CHECK_EXCL) == 0)
1425 vm_map_delete(map, start, end);
1426 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1427 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1428 prot, max, cow);
1429 } else {
1430 result = vm_map_insert(map, object, offset, start, end,
1431 prot, max, cow);
1432 }
1433 vm_map_unlock(map);
1434 return (result);
1435 }
1436
1437 /*
1438 * vm_map_find finds an unallocated region in the target address
1439 * map with the given length. The search is defined to be
1440 * first-fit from the specified address; the region found is
1441 * returned in the same parameter.
1442 *
1443 * If object is non-NULL, ref count must be bumped by caller
1444 * prior to making call to account for the new entry.
1445 */
1446 int
1447 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1448 vm_offset_t *addr, /* IN/OUT */
1449 vm_size_t length, vm_offset_t max_addr, int find_space,
1450 vm_prot_t prot, vm_prot_t max, int cow)
1451 {
1452 vm_offset_t alignment, initial_addr, start;
1453 int result;
1454
1455 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1456 object == NULL,
1457 ("vm_map_find: non-NULL backing object for stack"));
1458 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1459 (object->flags & OBJ_COLORED) == 0))
1460 find_space = VMFS_ANY_SPACE;
1461 if (find_space >> 8 != 0) {
1462 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1463 alignment = (vm_offset_t)1 << (find_space >> 8);
1464 } else
1465 alignment = 0;
1466 initial_addr = *addr;
1467 again:
1468 start = initial_addr;
1469 vm_map_lock(map);
1470 do {
1471 if (find_space != VMFS_NO_SPACE) {
1472 if (vm_map_findspace(map, start, length, addr) ||
1473 (max_addr != 0 && *addr + length > max_addr)) {
1474 vm_map_unlock(map);
1475 if (find_space == VMFS_OPTIMAL_SPACE) {
1476 find_space = VMFS_ANY_SPACE;
1477 goto again;
1478 }
1479 return (KERN_NO_SPACE);
1480 }
1481 switch (find_space) {
1482 case VMFS_SUPER_SPACE:
1483 case VMFS_OPTIMAL_SPACE:
1484 pmap_align_superpage(object, offset, addr,
1485 length);
1486 break;
1487 case VMFS_ANY_SPACE:
1488 break;
1489 default:
1490 if ((*addr & (alignment - 1)) != 0) {
1491 *addr &= ~(alignment - 1);
1492 *addr += alignment;
1493 }
1494 break;
1495 }
1496
1497 start = *addr;
1498 }
1499 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1500 result = vm_map_stack_locked(map, start, length,
1501 sgrowsiz, prot, max, cow);
1502 } else {
1503 result = vm_map_insert(map, object, offset, start,
1504 start + length, prot, max, cow);
1505 }
1506 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1507 find_space != VMFS_ANY_SPACE);
1508 vm_map_unlock(map);
1509 return (result);
1510 }
1511
1512 /*
1513 * vm_map_find_min() is a variant of vm_map_find() that takes an
1514 * additional parameter (min_addr) and treats the given address
1515 * (*addr) differently. Specifically, it treats *addr as a hint
1516 * and not as the minimum address where the mapping is created.
1517 *
1518 * This function works in two phases. First, it tries to
1519 * allocate above the hint. If that fails and the hint is
1520 * greater than min_addr, it performs a second pass, replacing
1521 * the hint with min_addr as the minimum address for the
1522 * allocation.
1523 */
1524 int
1525 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1526 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
1527 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
1528 int cow)
1529 {
1530 vm_offset_t hint;
1531 int rv;
1532
1533 hint = *addr;
1534 for (;;) {
1535 rv = vm_map_find(map, object, offset, addr, length, max_addr,
1536 find_space, prot, max, cow);
1537 if (rv == KERN_SUCCESS || min_addr >= hint)
1538 return (rv);
1539 *addr = hint = min_addr;
1540 }
1541 }
1542
1543 /*
1544 * vm_map_simplify_entry:
1545 *
1546 * Simplify the given map entry by merging with either neighbor. This
1547 * routine also has the ability to merge with both neighbors.
1548 *
1549 * The map must be locked.
1550 *
1551 * This routine guarentees that the passed entry remains valid (though
1552 * possibly extended). When merging, this routine may delete one or
1553 * both neighbors.
1554 */
1555 void
1556 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1557 {
1558 vm_map_entry_t next, prev;
1559 vm_size_t prevsize, esize;
1560
1561 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP |
1562 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0)
1563 return;
1564
1565 prev = entry->prev;
1566 if (prev != &map->header) {
1567 prevsize = prev->end - prev->start;
1568 if ( (prev->end == entry->start) &&
1569 (prev->object.vm_object == entry->object.vm_object) &&
1570 (!prev->object.vm_object ||
1571 (prev->offset + prevsize == entry->offset)) &&
1572 (prev->eflags == entry->eflags) &&
1573 (prev->protection == entry->protection) &&
1574 (prev->max_protection == entry->max_protection) &&
1575 (prev->inheritance == entry->inheritance) &&
1576 (prev->wired_count == entry->wired_count) &&
1577 (prev->cred == entry->cred)) {
1578 vm_map_entry_unlink(map, prev);
1579 entry->start = prev->start;
1580 entry->offset = prev->offset;
1581 if (entry->prev != &map->header)
1582 vm_map_entry_resize_free(map, entry->prev);
1583
1584 /*
1585 * If the backing object is a vnode object,
1586 * vm_object_deallocate() calls vrele().
1587 * However, vrele() does not lock the vnode
1588 * because the vnode has additional
1589 * references. Thus, the map lock can be kept
1590 * without causing a lock-order reversal with
1591 * the vnode lock.
1592 *
1593 * Since we count the number of virtual page
1594 * mappings in object->un_pager.vnp.writemappings,
1595 * the writemappings value should not be adjusted
1596 * when the entry is disposed of.
1597 */
1598 if (prev->object.vm_object)
1599 vm_object_deallocate(prev->object.vm_object);
1600 if (prev->cred != NULL)
1601 crfree(prev->cred);
1602 vm_map_entry_dispose(map, prev);
1603 }
1604 }
1605
1606 next = entry->next;
1607 if (next != &map->header) {
1608 esize = entry->end - entry->start;
1609 if ((entry->end == next->start) &&
1610 (next->object.vm_object == entry->object.vm_object) &&
1611 (!entry->object.vm_object ||
1612 (entry->offset + esize == next->offset)) &&
1613 (next->eflags == entry->eflags) &&
1614 (next->protection == entry->protection) &&
1615 (next->max_protection == entry->max_protection) &&
1616 (next->inheritance == entry->inheritance) &&
1617 (next->wired_count == entry->wired_count) &&
1618 (next->cred == entry->cred)) {
1619 vm_map_entry_unlink(map, next);
1620 entry->end = next->end;
1621 vm_map_entry_resize_free(map, entry);
1622
1623 /*
1624 * See comment above.
1625 */
1626 if (next->object.vm_object)
1627 vm_object_deallocate(next->object.vm_object);
1628 if (next->cred != NULL)
1629 crfree(next->cred);
1630 vm_map_entry_dispose(map, next);
1631 }
1632 }
1633 }
1634 /*
1635 * vm_map_clip_start: [ internal use only ]
1636 *
1637 * Asserts that the given entry begins at or after
1638 * the specified address; if necessary,
1639 * it splits the entry into two.
1640 */
1641 #define vm_map_clip_start(map, entry, startaddr) \
1642 { \
1643 if (startaddr > entry->start) \
1644 _vm_map_clip_start(map, entry, startaddr); \
1645 }
1646
1647 /*
1648 * This routine is called only when it is known that
1649 * the entry must be split.
1650 */
1651 static void
1652 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1653 {
1654 vm_map_entry_t new_entry;
1655
1656 VM_MAP_ASSERT_LOCKED(map);
1657
1658 /*
1659 * Split off the front portion -- note that we must insert the new
1660 * entry BEFORE this one, so that this entry has the specified
1661 * starting address.
1662 */
1663 vm_map_simplify_entry(map, entry);
1664
1665 /*
1666 * If there is no object backing this entry, we might as well create
1667 * one now. If we defer it, an object can get created after the map
1668 * is clipped, and individual objects will be created for the split-up
1669 * map. This is a bit of a hack, but is also about the best place to
1670 * put this improvement.
1671 */
1672 if (entry->object.vm_object == NULL && !map->system_map &&
1673 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1674 vm_object_t object;
1675 object = vm_object_allocate(OBJT_DEFAULT,
1676 atop(entry->end - entry->start));
1677 entry->object.vm_object = object;
1678 entry->offset = 0;
1679 if (entry->cred != NULL) {
1680 object->cred = entry->cred;
1681 object->charge = entry->end - entry->start;
1682 entry->cred = NULL;
1683 }
1684 } else if (entry->object.vm_object != NULL &&
1685 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1686 entry->cred != NULL) {
1687 VM_OBJECT_WLOCK(entry->object.vm_object);
1688 KASSERT(entry->object.vm_object->cred == NULL,
1689 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1690 entry->object.vm_object->cred = entry->cred;
1691 entry->object.vm_object->charge = entry->end - entry->start;
1692 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1693 entry->cred = NULL;
1694 }
1695
1696 new_entry = vm_map_entry_create(map);
1697 *new_entry = *entry;
1698
1699 new_entry->end = start;
1700 entry->offset += (start - entry->start);
1701 entry->start = start;
1702 if (new_entry->cred != NULL)
1703 crhold(entry->cred);
1704
1705 vm_map_entry_link(map, entry->prev, new_entry);
1706
1707 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1708 vm_object_reference(new_entry->object.vm_object);
1709 /*
1710 * The object->un_pager.vnp.writemappings for the
1711 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1712 * kept as is here. The virtual pages are
1713 * re-distributed among the clipped entries, so the sum is
1714 * left the same.
1715 */
1716 }
1717 }
1718
1719 /*
1720 * vm_map_clip_end: [ internal use only ]
1721 *
1722 * Asserts that the given entry ends at or before
1723 * the specified address; if necessary,
1724 * it splits the entry into two.
1725 */
1726 #define vm_map_clip_end(map, entry, endaddr) \
1727 { \
1728 if ((endaddr) < (entry->end)) \
1729 _vm_map_clip_end((map), (entry), (endaddr)); \
1730 }
1731
1732 /*
1733 * This routine is called only when it is known that
1734 * the entry must be split.
1735 */
1736 static void
1737 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1738 {
1739 vm_map_entry_t new_entry;
1740
1741 VM_MAP_ASSERT_LOCKED(map);
1742
1743 /*
1744 * If there is no object backing this entry, we might as well create
1745 * one now. If we defer it, an object can get created after the map
1746 * is clipped, and individual objects will be created for the split-up
1747 * map. This is a bit of a hack, but is also about the best place to
1748 * put this improvement.
1749 */
1750 if (entry->object.vm_object == NULL && !map->system_map &&
1751 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1752 vm_object_t object;
1753 object = vm_object_allocate(OBJT_DEFAULT,
1754 atop(entry->end - entry->start));
1755 entry->object.vm_object = object;
1756 entry->offset = 0;
1757 if (entry->cred != NULL) {
1758 object->cred = entry->cred;
1759 object->charge = entry->end - entry->start;
1760 entry->cred = NULL;
1761 }
1762 } else if (entry->object.vm_object != NULL &&
1763 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1764 entry->cred != NULL) {
1765 VM_OBJECT_WLOCK(entry->object.vm_object);
1766 KASSERT(entry->object.vm_object->cred == NULL,
1767 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1768 entry->object.vm_object->cred = entry->cred;
1769 entry->object.vm_object->charge = entry->end - entry->start;
1770 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1771 entry->cred = NULL;
1772 }
1773
1774 /*
1775 * Create a new entry and insert it AFTER the specified entry
1776 */
1777 new_entry = vm_map_entry_create(map);
1778 *new_entry = *entry;
1779
1780 new_entry->start = entry->end = end;
1781 new_entry->offset += (end - entry->start);
1782 if (new_entry->cred != NULL)
1783 crhold(entry->cred);
1784
1785 vm_map_entry_link(map, entry, new_entry);
1786
1787 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1788 vm_object_reference(new_entry->object.vm_object);
1789 }
1790 }
1791
1792 /*
1793 * vm_map_submap: [ kernel use only ]
1794 *
1795 * Mark the given range as handled by a subordinate map.
1796 *
1797 * This range must have been created with vm_map_find,
1798 * and no other operations may have been performed on this
1799 * range prior to calling vm_map_submap.
1800 *
1801 * Only a limited number of operations can be performed
1802 * within this rage after calling vm_map_submap:
1803 * vm_fault
1804 * [Don't try vm_map_copy!]
1805 *
1806 * To remove a submapping, one must first remove the
1807 * range from the superior map, and then destroy the
1808 * submap (if desired). [Better yet, don't try it.]
1809 */
1810 int
1811 vm_map_submap(
1812 vm_map_t map,
1813 vm_offset_t start,
1814 vm_offset_t end,
1815 vm_map_t submap)
1816 {
1817 vm_map_entry_t entry;
1818 int result = KERN_INVALID_ARGUMENT;
1819
1820 vm_map_lock(map);
1821
1822 VM_MAP_RANGE_CHECK(map, start, end);
1823
1824 if (vm_map_lookup_entry(map, start, &entry)) {
1825 vm_map_clip_start(map, entry, start);
1826 } else
1827 entry = entry->next;
1828
1829 vm_map_clip_end(map, entry, end);
1830
1831 if ((entry->start == start) && (entry->end == end) &&
1832 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1833 (entry->object.vm_object == NULL)) {
1834 entry->object.sub_map = submap;
1835 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1836 result = KERN_SUCCESS;
1837 }
1838 vm_map_unlock(map);
1839
1840 return (result);
1841 }
1842
1843 /*
1844 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1845 */
1846 #define MAX_INIT_PT 96
1847
1848 /*
1849 * vm_map_pmap_enter:
1850 *
1851 * Preload the specified map's pmap with mappings to the specified
1852 * object's memory-resident pages. No further physical pages are
1853 * allocated, and no further virtual pages are retrieved from secondary
1854 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1855 * limited number of page mappings are created at the low-end of the
1856 * specified address range. (For this purpose, a superpage mapping
1857 * counts as one page mapping.) Otherwise, all resident pages within
1858 * the specified address range are mapped. Because these mappings are
1859 * being created speculatively, cached pages are not reactivated and
1860 * mapped.
1861 */
1862 void
1863 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1864 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1865 {
1866 vm_offset_t start;
1867 vm_page_t p, p_start;
1868 vm_pindex_t mask, psize, threshold, tmpidx;
1869
1870 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1871 return;
1872 VM_OBJECT_RLOCK(object);
1873 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1874 VM_OBJECT_RUNLOCK(object);
1875 VM_OBJECT_WLOCK(object);
1876 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1877 pmap_object_init_pt(map->pmap, addr, object, pindex,
1878 size);
1879 VM_OBJECT_WUNLOCK(object);
1880 return;
1881 }
1882 VM_OBJECT_LOCK_DOWNGRADE(object);
1883 }
1884
1885 psize = atop(size);
1886 if (psize + pindex > object->size) {
1887 if (object->size < pindex) {
1888 VM_OBJECT_RUNLOCK(object);
1889 return;
1890 }
1891 psize = object->size - pindex;
1892 }
1893
1894 start = 0;
1895 p_start = NULL;
1896 threshold = MAX_INIT_PT;
1897
1898 p = vm_page_find_least(object, pindex);
1899 /*
1900 * Assert: the variable p is either (1) the page with the
1901 * least pindex greater than or equal to the parameter pindex
1902 * or (2) NULL.
1903 */
1904 for (;
1905 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1906 p = TAILQ_NEXT(p, listq)) {
1907 /*
1908 * don't allow an madvise to blow away our really
1909 * free pages allocating pv entries.
1910 */
1911 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1912 cnt.v_free_count < cnt.v_free_reserved) ||
1913 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1914 tmpidx >= threshold)) {
1915 psize = tmpidx;
1916 break;
1917 }
1918 if (p->valid == VM_PAGE_BITS_ALL) {
1919 if (p_start == NULL) {
1920 start = addr + ptoa(tmpidx);
1921 p_start = p;
1922 }
1923 /* Jump ahead if a superpage mapping is possible. */
1924 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1925 (pagesizes[p->psind] - 1)) == 0) {
1926 mask = atop(pagesizes[p->psind]) - 1;
1927 if (tmpidx + mask < psize &&
1928 vm_page_ps_is_valid(p)) {
1929 p += mask;
1930 threshold += mask;
1931 }
1932 }
1933 } else if (p_start != NULL) {
1934 pmap_enter_object(map->pmap, start, addr +
1935 ptoa(tmpidx), p_start, prot);
1936 p_start = NULL;
1937 }
1938 }
1939 if (p_start != NULL)
1940 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1941 p_start, prot);
1942 VM_OBJECT_RUNLOCK(object);
1943 }
1944
1945 /*
1946 * vm_map_protect:
1947 *
1948 * Sets the protection of the specified address
1949 * region in the target map. If "set_max" is
1950 * specified, the maximum protection is to be set;
1951 * otherwise, only the current protection is affected.
1952 */
1953 int
1954 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1955 vm_prot_t new_prot, boolean_t set_max)
1956 {
1957 vm_map_entry_t current, entry;
1958 vm_object_t obj;
1959 struct ucred *cred;
1960 vm_prot_t old_prot;
1961
1962 if (start == end)
1963 return (KERN_SUCCESS);
1964
1965 vm_map_lock(map);
1966
1967 VM_MAP_RANGE_CHECK(map, start, end);
1968
1969 if (vm_map_lookup_entry(map, start, &entry)) {
1970 vm_map_clip_start(map, entry, start);
1971 } else {
1972 entry = entry->next;
1973 }
1974
1975 /*
1976 * Make a first pass to check for protection violations.
1977 */
1978 for (current = entry; current != &map->header && current->start < end;
1979 current = current->next) {
1980 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
1981 continue;
1982 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1983 vm_map_unlock(map);
1984 return (KERN_INVALID_ARGUMENT);
1985 }
1986 if ((new_prot & current->max_protection) != new_prot) {
1987 vm_map_unlock(map);
1988 return (KERN_PROTECTION_FAILURE);
1989 }
1990 }
1991
1992 /*
1993 * Do an accounting pass for private read-only mappings that
1994 * now will do cow due to allowed write (e.g. debugger sets
1995 * breakpoint on text segment)
1996 */
1997 for (current = entry; current != &map->header && current->start < end;
1998 current = current->next) {
1999
2000 vm_map_clip_end(map, current, end);
2001
2002 if (set_max ||
2003 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2004 ENTRY_CHARGED(current) ||
2005 (current->eflags & MAP_ENTRY_GUARD) != 0) {
2006 continue;
2007 }
2008
2009 cred = curthread->td_ucred;
2010 obj = current->object.vm_object;
2011
2012 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2013 if (!swap_reserve(current->end - current->start)) {
2014 vm_map_unlock(map);
2015 return (KERN_RESOURCE_SHORTAGE);
2016 }
2017 crhold(cred);
2018 current->cred = cred;
2019 continue;
2020 }
2021
2022 VM_OBJECT_WLOCK(obj);
2023 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2024 VM_OBJECT_WUNLOCK(obj);
2025 continue;
2026 }
2027
2028 /*
2029 * Charge for the whole object allocation now, since
2030 * we cannot distinguish between non-charged and
2031 * charged clipped mapping of the same object later.
2032 */
2033 KASSERT(obj->charge == 0,
2034 ("vm_map_protect: object %p overcharged (entry %p)",
2035 obj, current));
2036 if (!swap_reserve(ptoa(obj->size))) {
2037 VM_OBJECT_WUNLOCK(obj);
2038 vm_map_unlock(map);
2039 return (KERN_RESOURCE_SHORTAGE);
2040 }
2041
2042 crhold(cred);
2043 obj->cred = cred;
2044 obj->charge = ptoa(obj->size);
2045 VM_OBJECT_WUNLOCK(obj);
2046 }
2047
2048 /*
2049 * Go back and fix up protections. [Note that clipping is not
2050 * necessary the second time.]
2051 */
2052 for (current = entry; current != &map->header && current->start < end;
2053 current = current->next) {
2054 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2055 continue;
2056
2057 old_prot = current->protection;
2058
2059 if (set_max)
2060 current->protection =
2061 (current->max_protection = new_prot) &
2062 old_prot;
2063 else
2064 current->protection = new_prot;
2065
2066 /*
2067 * For user wired map entries, the normal lazy evaluation of
2068 * write access upgrades through soft page faults is
2069 * undesirable. Instead, immediately copy any pages that are
2070 * copy-on-write and enable write access in the physical map.
2071 */
2072 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2073 (current->protection & VM_PROT_WRITE) != 0 &&
2074 (old_prot & VM_PROT_WRITE) == 0)
2075 vm_fault_copy_entry(map, map, current, current, NULL);
2076
2077 /*
2078 * When restricting access, update the physical map. Worry
2079 * about copy-on-write here.
2080 */
2081 if ((old_prot & ~current->protection) != 0) {
2082 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2083 VM_PROT_ALL)
2084 pmap_protect(map->pmap, current->start,
2085 current->end,
2086 current->protection & MASK(current));
2087 #undef MASK
2088 }
2089 vm_map_simplify_entry(map, current);
2090 }
2091 vm_map_unlock(map);
2092 return (KERN_SUCCESS);
2093 }
2094
2095 /*
2096 * vm_map_madvise:
2097 *
2098 * This routine traverses a processes map handling the madvise
2099 * system call. Advisories are classified as either those effecting
2100 * the vm_map_entry structure, or those effecting the underlying
2101 * objects.
2102 */
2103 int
2104 vm_map_madvise(
2105 vm_map_t map,
2106 vm_offset_t start,
2107 vm_offset_t end,
2108 int behav)
2109 {
2110 vm_map_entry_t current, entry;
2111 int modify_map = 0;
2112
2113 /*
2114 * Some madvise calls directly modify the vm_map_entry, in which case
2115 * we need to use an exclusive lock on the map and we need to perform
2116 * various clipping operations. Otherwise we only need a read-lock
2117 * on the map.
2118 */
2119 switch(behav) {
2120 case MADV_NORMAL:
2121 case MADV_SEQUENTIAL:
2122 case MADV_RANDOM:
2123 case MADV_NOSYNC:
2124 case MADV_AUTOSYNC:
2125 case MADV_NOCORE:
2126 case MADV_CORE:
2127 if (start == end)
2128 return (KERN_SUCCESS);
2129 modify_map = 1;
2130 vm_map_lock(map);
2131 break;
2132 case MADV_WILLNEED:
2133 case MADV_DONTNEED:
2134 case MADV_FREE:
2135 if (start == end)
2136 return (KERN_SUCCESS);
2137 vm_map_lock_read(map);
2138 break;
2139 default:
2140 return (KERN_INVALID_ARGUMENT);
2141 }
2142
2143 /*
2144 * Locate starting entry and clip if necessary.
2145 */
2146 VM_MAP_RANGE_CHECK(map, start, end);
2147
2148 if (vm_map_lookup_entry(map, start, &entry)) {
2149 if (modify_map)
2150 vm_map_clip_start(map, entry, start);
2151 } else {
2152 entry = entry->next;
2153 }
2154
2155 if (modify_map) {
2156 /*
2157 * madvise behaviors that are implemented in the vm_map_entry.
2158 *
2159 * We clip the vm_map_entry so that behavioral changes are
2160 * limited to the specified address range.
2161 */
2162 for (current = entry;
2163 (current != &map->header) && (current->start < end);
2164 current = current->next
2165 ) {
2166 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2167 continue;
2168
2169 vm_map_clip_end(map, current, end);
2170
2171 switch (behav) {
2172 case MADV_NORMAL:
2173 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2174 break;
2175 case MADV_SEQUENTIAL:
2176 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2177 break;
2178 case MADV_RANDOM:
2179 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2180 break;
2181 case MADV_NOSYNC:
2182 current->eflags |= MAP_ENTRY_NOSYNC;
2183 break;
2184 case MADV_AUTOSYNC:
2185 current->eflags &= ~MAP_ENTRY_NOSYNC;
2186 break;
2187 case MADV_NOCORE:
2188 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2189 break;
2190 case MADV_CORE:
2191 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2192 break;
2193 default:
2194 break;
2195 }
2196 vm_map_simplify_entry(map, current);
2197 }
2198 vm_map_unlock(map);
2199 } else {
2200 vm_pindex_t pstart, pend;
2201
2202 /*
2203 * madvise behaviors that are implemented in the underlying
2204 * vm_object.
2205 *
2206 * Since we don't clip the vm_map_entry, we have to clip
2207 * the vm_object pindex and count.
2208 */
2209 for (current = entry;
2210 (current != &map->header) && (current->start < end);
2211 current = current->next
2212 ) {
2213 vm_offset_t useEnd, useStart;
2214
2215 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2216 continue;
2217
2218 pstart = OFF_TO_IDX(current->offset);
2219 pend = pstart + atop(current->end - current->start);
2220 useStart = current->start;
2221 useEnd = current->end;
2222
2223 if (current->start < start) {
2224 pstart += atop(start - current->start);
2225 useStart = start;
2226 }
2227 if (current->end > end) {
2228 pend -= atop(current->end - end);
2229 useEnd = end;
2230 }
2231
2232 if (pstart >= pend)
2233 continue;
2234
2235 /*
2236 * Perform the pmap_advise() before clearing
2237 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2238 * concurrent pmap operation, such as pmap_remove(),
2239 * could clear a reference in the pmap and set
2240 * PGA_REFERENCED on the page before the pmap_advise()
2241 * had completed. Consequently, the page would appear
2242 * referenced based upon an old reference that
2243 * occurred before this pmap_advise() ran.
2244 */
2245 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2246 pmap_advise(map->pmap, useStart, useEnd,
2247 behav);
2248
2249 vm_object_madvise(current->object.vm_object, pstart,
2250 pend, behav);
2251
2252 /*
2253 * Pre-populate paging structures in the
2254 * WILLNEED case. For wired entries, the
2255 * paging structures are already populated.
2256 */
2257 if (behav == MADV_WILLNEED &&
2258 current->wired_count == 0) {
2259 vm_map_pmap_enter(map,
2260 useStart,
2261 current->protection,
2262 current->object.vm_object,
2263 pstart,
2264 ptoa(pend - pstart),
2265 MAP_PREFAULT_MADVISE
2266 );
2267 }
2268 }
2269 vm_map_unlock_read(map);
2270 }
2271 return (0);
2272 }
2273
2274
2275 /*
2276 * vm_map_inherit:
2277 *
2278 * Sets the inheritance of the specified address
2279 * range in the target map. Inheritance
2280 * affects how the map will be shared with
2281 * child maps at the time of vmspace_fork.
2282 */
2283 int
2284 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2285 vm_inherit_t new_inheritance)
2286 {
2287 vm_map_entry_t entry;
2288 vm_map_entry_t temp_entry;
2289
2290 switch (new_inheritance) {
2291 case VM_INHERIT_NONE:
2292 case VM_INHERIT_COPY:
2293 case VM_INHERIT_SHARE:
2294 case VM_INHERIT_ZERO:
2295 break;
2296 default:
2297 return (KERN_INVALID_ARGUMENT);
2298 }
2299 if (start == end)
2300 return (KERN_SUCCESS);
2301 vm_map_lock(map);
2302 VM_MAP_RANGE_CHECK(map, start, end);
2303 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2304 entry = temp_entry;
2305 vm_map_clip_start(map, entry, start);
2306 } else
2307 entry = temp_entry->next;
2308 while ((entry != &map->header) && (entry->start < end)) {
2309 vm_map_clip_end(map, entry, end);
2310 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2311 new_inheritance != VM_INHERIT_ZERO)
2312 entry->inheritance = new_inheritance;
2313 vm_map_simplify_entry(map, entry);
2314 entry = entry->next;
2315 }
2316 vm_map_unlock(map);
2317 return (KERN_SUCCESS);
2318 }
2319
2320 /*
2321 * vm_map_unwire:
2322 *
2323 * Implements both kernel and user unwiring.
2324 */
2325 int
2326 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2327 int flags)
2328 {
2329 vm_map_entry_t entry, first_entry, tmp_entry;
2330 vm_offset_t saved_start;
2331 unsigned int last_timestamp;
2332 int rv;
2333 boolean_t need_wakeup, result, user_unwire;
2334
2335 if (start == end)
2336 return (KERN_SUCCESS);
2337 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2338 vm_map_lock(map);
2339 VM_MAP_RANGE_CHECK(map, start, end);
2340 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2341 if (flags & VM_MAP_WIRE_HOLESOK)
2342 first_entry = first_entry->next;
2343 else {
2344 vm_map_unlock(map);
2345 return (KERN_INVALID_ADDRESS);
2346 }
2347 }
2348 last_timestamp = map->timestamp;
2349 entry = first_entry;
2350 while (entry != &map->header && entry->start < end) {
2351 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2352 /*
2353 * We have not yet clipped the entry.
2354 */
2355 saved_start = (start >= entry->start) ? start :
2356 entry->start;
2357 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2358 if (vm_map_unlock_and_wait(map, 0)) {
2359 /*
2360 * Allow interruption of user unwiring?
2361 */
2362 }
2363 vm_map_lock(map);
2364 if (last_timestamp+1 != map->timestamp) {
2365 /*
2366 * Look again for the entry because the map was
2367 * modified while it was unlocked.
2368 * Specifically, the entry may have been
2369 * clipped, merged, or deleted.
2370 */
2371 if (!vm_map_lookup_entry(map, saved_start,
2372 &tmp_entry)) {
2373 if (flags & VM_MAP_WIRE_HOLESOK)
2374 tmp_entry = tmp_entry->next;
2375 else {
2376 if (saved_start == start) {
2377 /*
2378 * First_entry has been deleted.
2379 */
2380 vm_map_unlock(map);
2381 return (KERN_INVALID_ADDRESS);
2382 }
2383 end = saved_start;
2384 rv = KERN_INVALID_ADDRESS;
2385 goto done;
2386 }
2387 }
2388 if (entry == first_entry)
2389 first_entry = tmp_entry;
2390 else
2391 first_entry = NULL;
2392 entry = tmp_entry;
2393 }
2394 last_timestamp = map->timestamp;
2395 continue;
2396 }
2397 vm_map_clip_start(map, entry, start);
2398 vm_map_clip_end(map, entry, end);
2399 /*
2400 * Mark the entry in case the map lock is released. (See
2401 * above.)
2402 */
2403 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2404 entry->wiring_thread == NULL,
2405 ("owned map entry %p", entry));
2406 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2407 entry->wiring_thread = curthread;
2408 /*
2409 * Check the map for holes in the specified region.
2410 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2411 */
2412 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2413 (entry->end < end && (entry->next == &map->header ||
2414 entry->next->start > entry->end))) {
2415 end = entry->end;
2416 rv = KERN_INVALID_ADDRESS;
2417 goto done;
2418 }
2419 /*
2420 * If system unwiring, require that the entry is system wired.
2421 */
2422 if (!user_unwire &&
2423 vm_map_entry_system_wired_count(entry) == 0) {
2424 end = entry->end;
2425 rv = KERN_INVALID_ARGUMENT;
2426 goto done;
2427 }
2428 entry = entry->next;
2429 }
2430 rv = KERN_SUCCESS;
2431 done:
2432 need_wakeup = FALSE;
2433 if (first_entry == NULL) {
2434 result = vm_map_lookup_entry(map, start, &first_entry);
2435 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2436 first_entry = first_entry->next;
2437 else
2438 KASSERT(result, ("vm_map_unwire: lookup failed"));
2439 }
2440 for (entry = first_entry; entry != &map->header && entry->start < end;
2441 entry = entry->next) {
2442 /*
2443 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2444 * space in the unwired region could have been mapped
2445 * while the map lock was dropped for draining
2446 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2447 * could be simultaneously wiring this new mapping
2448 * entry. Detect these cases and skip any entries
2449 * marked as in transition by us.
2450 */
2451 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2452 entry->wiring_thread != curthread) {
2453 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2454 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2455 continue;
2456 }
2457
2458 if (rv == KERN_SUCCESS && (!user_unwire ||
2459 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2460 if (user_unwire)
2461 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2462 if (entry->wired_count == 1)
2463 vm_map_entry_unwire(map, entry);
2464 else
2465 entry->wired_count--;
2466 }
2467 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2468 ("vm_map_unwire: in-transition flag missing %p", entry));
2469 KASSERT(entry->wiring_thread == curthread,
2470 ("vm_map_unwire: alien wire %p", entry));
2471 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2472 entry->wiring_thread = NULL;
2473 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2474 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2475 need_wakeup = TRUE;
2476 }
2477 vm_map_simplify_entry(map, entry);
2478 }
2479 vm_map_unlock(map);
2480 if (need_wakeup)
2481 vm_map_wakeup(map);
2482 return (rv);
2483 }
2484
2485 /*
2486 * vm_map_wire_entry_failure:
2487 *
2488 * Handle a wiring failure on the given entry.
2489 *
2490 * The map should be locked.
2491 */
2492 static void
2493 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2494 vm_offset_t failed_addr)
2495 {
2496
2497 VM_MAP_ASSERT_LOCKED(map);
2498 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2499 entry->wired_count == 1,
2500 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2501 KASSERT(failed_addr < entry->end,
2502 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2503
2504 /*
2505 * If any pages at the start of this entry were successfully wired,
2506 * then unwire them.
2507 */
2508 if (failed_addr > entry->start) {
2509 pmap_unwire(map->pmap, entry->start, failed_addr);
2510 vm_object_unwire(entry->object.vm_object, entry->offset,
2511 failed_addr - entry->start, PQ_ACTIVE);
2512 }
2513
2514 /*
2515 * Assign an out-of-range value to represent the failure to wire this
2516 * entry.
2517 */
2518 entry->wired_count = -1;
2519 }
2520
2521 /*
2522 * vm_map_wire:
2523 *
2524 * Implements both kernel and user wiring.
2525 */
2526 int
2527 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2528 int flags)
2529 {
2530 vm_map_entry_t entry, first_entry, tmp_entry;
2531 vm_offset_t faddr, saved_end, saved_start;
2532 unsigned int last_timestamp;
2533 int rv;
2534 boolean_t need_wakeup, result, user_wire;
2535 vm_prot_t prot;
2536
2537 if (start == end)
2538 return (KERN_SUCCESS);
2539 prot = 0;
2540 if (flags & VM_MAP_WIRE_WRITE)
2541 prot |= VM_PROT_WRITE;
2542 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2543 vm_map_lock(map);
2544 VM_MAP_RANGE_CHECK(map, start, end);
2545 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2546 if (flags & VM_MAP_WIRE_HOLESOK)
2547 first_entry = first_entry->next;
2548 else {
2549 vm_map_unlock(map);
2550 return (KERN_INVALID_ADDRESS);
2551 }
2552 }
2553 last_timestamp = map->timestamp;
2554 entry = first_entry;
2555 while (entry != &map->header && entry->start < end) {
2556 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2557 /*
2558 * We have not yet clipped the entry.
2559 */
2560 saved_start = (start >= entry->start) ? start :
2561 entry->start;
2562 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2563 if (vm_map_unlock_and_wait(map, 0)) {
2564 /*
2565 * Allow interruption of user wiring?
2566 */
2567 }
2568 vm_map_lock(map);
2569 if (last_timestamp + 1 != map->timestamp) {
2570 /*
2571 * Look again for the entry because the map was
2572 * modified while it was unlocked.
2573 * Specifically, the entry may have been
2574 * clipped, merged, or deleted.
2575 */
2576 if (!vm_map_lookup_entry(map, saved_start,
2577 &tmp_entry)) {
2578 if (flags & VM_MAP_WIRE_HOLESOK)
2579 tmp_entry = tmp_entry->next;
2580 else {
2581 if (saved_start == start) {
2582 /*
2583 * first_entry has been deleted.
2584 */
2585 vm_map_unlock(map);
2586 return (KERN_INVALID_ADDRESS);
2587 }
2588 end = saved_start;
2589 rv = KERN_INVALID_ADDRESS;
2590 goto done;
2591 }
2592 }
2593 if (entry == first_entry)
2594 first_entry = tmp_entry;
2595 else
2596 first_entry = NULL;
2597 entry = tmp_entry;
2598 }
2599 last_timestamp = map->timestamp;
2600 continue;
2601 }
2602 vm_map_clip_start(map, entry, start);
2603 vm_map_clip_end(map, entry, end);
2604 /*
2605 * Mark the entry in case the map lock is released. (See
2606 * above.)
2607 */
2608 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2609 entry->wiring_thread == NULL,
2610 ("owned map entry %p", entry));
2611 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2612 entry->wiring_thread = curthread;
2613 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2614 || (entry->protection & prot) != prot) {
2615 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2616 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2617 end = entry->end;
2618 rv = KERN_INVALID_ADDRESS;
2619 goto done;
2620 }
2621 goto next_entry;
2622 }
2623 if (entry->wired_count == 0) {
2624 entry->wired_count++;
2625 saved_start = entry->start;
2626 saved_end = entry->end;
2627
2628 /*
2629 * Release the map lock, relying on the in-transition
2630 * mark. Mark the map busy for fork.
2631 */
2632 vm_map_busy(map);
2633 vm_map_unlock(map);
2634
2635 faddr = saved_start;
2636 do {
2637 /*
2638 * Simulate a fault to get the page and enter
2639 * it into the physical map.
2640 */
2641 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2642 VM_FAULT_WIRE)) != KERN_SUCCESS)
2643 break;
2644 } while ((faddr += PAGE_SIZE) < saved_end);
2645 vm_map_lock(map);
2646 vm_map_unbusy(map);
2647 if (last_timestamp + 1 != map->timestamp) {
2648 /*
2649 * Look again for the entry because the map was
2650 * modified while it was unlocked. The entry
2651 * may have been clipped, but NOT merged or
2652 * deleted.
2653 */
2654 result = vm_map_lookup_entry(map, saved_start,
2655 &tmp_entry);
2656 KASSERT(result, ("vm_map_wire: lookup failed"));
2657 if (entry == first_entry)
2658 first_entry = tmp_entry;
2659 else
2660 first_entry = NULL;
2661 entry = tmp_entry;
2662 while (entry->end < saved_end) {
2663 /*
2664 * In case of failure, handle entries
2665 * that were not fully wired here;
2666 * fully wired entries are handled
2667 * later.
2668 */
2669 if (rv != KERN_SUCCESS &&
2670 faddr < entry->end)
2671 vm_map_wire_entry_failure(map,
2672 entry, faddr);
2673 entry = entry->next;
2674 }
2675 }
2676 last_timestamp = map->timestamp;
2677 if (rv != KERN_SUCCESS) {
2678 vm_map_wire_entry_failure(map, entry, faddr);
2679 end = entry->end;
2680 goto done;
2681 }
2682 } else if (!user_wire ||
2683 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2684 entry->wired_count++;
2685 }
2686 /*
2687 * Check the map for holes in the specified region.
2688 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2689 */
2690 next_entry:
2691 if ((flags & VM_MAP_WIRE_HOLESOK) == 0 &&
2692 entry->end < end && (entry->next == &map->header ||
2693 entry->next->start > entry->end)) {
2694 end = entry->end;
2695 rv = KERN_INVALID_ADDRESS;
2696 goto done;
2697 }
2698 entry = entry->next;
2699 }
2700 rv = KERN_SUCCESS;
2701 done:
2702 need_wakeup = FALSE;
2703 if (first_entry == NULL) {
2704 result = vm_map_lookup_entry(map, start, &first_entry);
2705 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2706 first_entry = first_entry->next;
2707 else
2708 KASSERT(result, ("vm_map_wire: lookup failed"));
2709 }
2710 for (entry = first_entry; entry != &map->header && entry->start < end;
2711 entry = entry->next) {
2712 /*
2713 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2714 * space in the unwired region could have been mapped
2715 * while the map lock was dropped for faulting in the
2716 * pages or draining MAP_ENTRY_IN_TRANSITION.
2717 * Moreover, another thread could be simultaneously
2718 * wiring this new mapping entry. Detect these cases
2719 * and skip any entries marked as in transition not by us.
2720 */
2721 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2722 entry->wiring_thread != curthread) {
2723 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2724 ("vm_map_wire: !HOLESOK and new/changed entry"));
2725 continue;
2726 }
2727
2728 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2729 goto next_entry_done;
2730
2731 if (rv == KERN_SUCCESS) {
2732 if (user_wire)
2733 entry->eflags |= MAP_ENTRY_USER_WIRED;
2734 } else if (entry->wired_count == -1) {
2735 /*
2736 * Wiring failed on this entry. Thus, unwiring is
2737 * unnecessary.
2738 */
2739 entry->wired_count = 0;
2740 } else if (!user_wire ||
2741 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2742 /*
2743 * Undo the wiring. Wiring succeeded on this entry
2744 * but failed on a later entry.
2745 */
2746 if (entry->wired_count == 1)
2747 vm_map_entry_unwire(map, entry);
2748 else
2749 entry->wired_count--;
2750 }
2751 next_entry_done:
2752 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2753 ("vm_map_wire: in-transition flag missing %p", entry));
2754 KASSERT(entry->wiring_thread == curthread,
2755 ("vm_map_wire: alien wire %p", entry));
2756 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2757 MAP_ENTRY_WIRE_SKIPPED);
2758 entry->wiring_thread = NULL;
2759 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2760 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2761 need_wakeup = TRUE;
2762 }
2763 vm_map_simplify_entry(map, entry);
2764 }
2765 vm_map_unlock(map);
2766 if (need_wakeup)
2767 vm_map_wakeup(map);
2768 return (rv);
2769 }
2770
2771 /*
2772 * vm_map_sync
2773 *
2774 * Push any dirty cached pages in the address range to their pager.
2775 * If syncio is TRUE, dirty pages are written synchronously.
2776 * If invalidate is TRUE, any cached pages are freed as well.
2777 *
2778 * If the size of the region from start to end is zero, we are
2779 * supposed to flush all modified pages within the region containing
2780 * start. Unfortunately, a region can be split or coalesced with
2781 * neighboring regions, making it difficult to determine what the
2782 * original region was. Therefore, we approximate this requirement by
2783 * flushing the current region containing start.
2784 *
2785 * Returns an error if any part of the specified range is not mapped.
2786 */
2787 int
2788 vm_map_sync(
2789 vm_map_t map,
2790 vm_offset_t start,
2791 vm_offset_t end,
2792 boolean_t syncio,
2793 boolean_t invalidate)
2794 {
2795 vm_map_entry_t current;
2796 vm_map_entry_t entry;
2797 vm_size_t size;
2798 vm_object_t object;
2799 vm_ooffset_t offset;
2800 unsigned int last_timestamp;
2801 boolean_t failed;
2802
2803 vm_map_lock_read(map);
2804 VM_MAP_RANGE_CHECK(map, start, end);
2805 if (!vm_map_lookup_entry(map, start, &entry)) {
2806 vm_map_unlock_read(map);
2807 return (KERN_INVALID_ADDRESS);
2808 } else if (start == end) {
2809 start = entry->start;
2810 end = entry->end;
2811 }
2812 /*
2813 * Make a first pass to check for user-wired memory and holes.
2814 */
2815 for (current = entry; current != &map->header && current->start < end;
2816 current = current->next) {
2817 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2818 vm_map_unlock_read(map);
2819 return (KERN_INVALID_ARGUMENT);
2820 }
2821 if (end > current->end &&
2822 (current->next == &map->header ||
2823 current->end != current->next->start)) {
2824 vm_map_unlock_read(map);
2825 return (KERN_INVALID_ADDRESS);
2826 }
2827 }
2828
2829 if (invalidate)
2830 pmap_remove(map->pmap, start, end);
2831 failed = FALSE;
2832
2833 /*
2834 * Make a second pass, cleaning/uncaching pages from the indicated
2835 * objects as we go.
2836 */
2837 for (current = entry; current != &map->header && current->start < end;) {
2838 offset = current->offset + (start - current->start);
2839 size = (end <= current->end ? end : current->end) - start;
2840 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2841 vm_map_t smap;
2842 vm_map_entry_t tentry;
2843 vm_size_t tsize;
2844
2845 smap = current->object.sub_map;
2846 vm_map_lock_read(smap);
2847 (void) vm_map_lookup_entry(smap, offset, &tentry);
2848 tsize = tentry->end - offset;
2849 if (tsize < size)
2850 size = tsize;
2851 object = tentry->object.vm_object;
2852 offset = tentry->offset + (offset - tentry->start);
2853 vm_map_unlock_read(smap);
2854 } else {
2855 object = current->object.vm_object;
2856 }
2857 vm_object_reference(object);
2858 last_timestamp = map->timestamp;
2859 vm_map_unlock_read(map);
2860 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2861 failed = TRUE;
2862 start += size;
2863 vm_object_deallocate(object);
2864 vm_map_lock_read(map);
2865 if (last_timestamp == map->timestamp ||
2866 !vm_map_lookup_entry(map, start, ¤t))
2867 current = current->next;
2868 }
2869
2870 vm_map_unlock_read(map);
2871 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2872 }
2873
2874 /*
2875 * vm_map_entry_unwire: [ internal use only ]
2876 *
2877 * Make the region specified by this entry pageable.
2878 *
2879 * The map in question should be locked.
2880 * [This is the reason for this routine's existence.]
2881 */
2882 static void
2883 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2884 {
2885
2886 VM_MAP_ASSERT_LOCKED(map);
2887 KASSERT(entry->wired_count > 0,
2888 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2889 pmap_unwire(map->pmap, entry->start, entry->end);
2890 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2891 entry->start, PQ_ACTIVE);
2892 entry->wired_count = 0;
2893 }
2894
2895 static void
2896 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2897 {
2898
2899 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2900 vm_object_deallocate(entry->object.vm_object);
2901 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2902 }
2903
2904 /*
2905 * vm_map_entry_delete: [ internal use only ]
2906 *
2907 * Deallocate the given entry from the target map.
2908 */
2909 static void
2910 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2911 {
2912 vm_object_t object;
2913 vm_pindex_t offidxstart, offidxend, count, size1;
2914 vm_ooffset_t size;
2915
2916 vm_map_entry_unlink(map, entry);
2917 object = entry->object.vm_object;
2918
2919 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
2920 MPASS(entry->cred == NULL);
2921 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
2922 MPASS(object == NULL);
2923 vm_map_entry_deallocate(entry, map->system_map);
2924 return;
2925 }
2926
2927 size = entry->end - entry->start;
2928 map->size -= size;
2929
2930 if (entry->cred != NULL) {
2931 swap_release_by_cred(size, entry->cred);
2932 crfree(entry->cred);
2933 }
2934
2935 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2936 (object != NULL)) {
2937 KASSERT(entry->cred == NULL || object->cred == NULL ||
2938 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2939 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2940 count = OFF_TO_IDX(size);
2941 offidxstart = OFF_TO_IDX(entry->offset);
2942 offidxend = offidxstart + count;
2943 VM_OBJECT_WLOCK(object);
2944 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
2945 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2946 object == kernel_object || object == kmem_object)) {
2947 vm_object_collapse(object);
2948
2949 /*
2950 * The option OBJPR_NOTMAPPED can be passed here
2951 * because vm_map_delete() already performed
2952 * pmap_remove() on the only mapping to this range
2953 * of pages.
2954 */
2955 vm_object_page_remove(object, offidxstart, offidxend,
2956 OBJPR_NOTMAPPED);
2957 if (object->type == OBJT_SWAP)
2958 swap_pager_freespace(object, offidxstart,
2959 count);
2960 if (offidxend >= object->size &&
2961 offidxstart < object->size) {
2962 size1 = object->size;
2963 object->size = offidxstart;
2964 if (object->cred != NULL) {
2965 size1 -= object->size;
2966 KASSERT(object->charge >= ptoa(size1),
2967 ("object %p charge < 0", object));
2968 swap_release_by_cred(ptoa(size1),
2969 object->cred);
2970 object->charge -= ptoa(size1);
2971 }
2972 }
2973 }
2974 VM_OBJECT_WUNLOCK(object);
2975 } else
2976 entry->object.vm_object = NULL;
2977 if (map->system_map)
2978 vm_map_entry_deallocate(entry, TRUE);
2979 else {
2980 entry->next = curthread->td_map_def_user;
2981 curthread->td_map_def_user = entry;
2982 }
2983 }
2984
2985 /*
2986 * vm_map_delete: [ internal use only ]
2987 *
2988 * Deallocates the given address range from the target
2989 * map.
2990 */
2991 int
2992 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2993 {
2994 vm_map_entry_t entry;
2995 vm_map_entry_t first_entry;
2996
2997 VM_MAP_ASSERT_LOCKED(map);
2998 if (start == end)
2999 return (KERN_SUCCESS);
3000
3001 /*
3002 * Find the start of the region, and clip it
3003 */
3004 if (!vm_map_lookup_entry(map, start, &first_entry))
3005 entry = first_entry->next;
3006 else {
3007 entry = first_entry;
3008 vm_map_clip_start(map, entry, start);
3009 }
3010
3011 /*
3012 * Step through all entries in this region
3013 */
3014 while ((entry != &map->header) && (entry->start < end)) {
3015 vm_map_entry_t next;
3016
3017 /*
3018 * Wait for wiring or unwiring of an entry to complete.
3019 * Also wait for any system wirings to disappear on
3020 * user maps.
3021 */
3022 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3023 (vm_map_pmap(map) != kernel_pmap &&
3024 vm_map_entry_system_wired_count(entry) != 0)) {
3025 unsigned int last_timestamp;
3026 vm_offset_t saved_start;
3027 vm_map_entry_t tmp_entry;
3028
3029 saved_start = entry->start;
3030 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3031 last_timestamp = map->timestamp;
3032 (void) vm_map_unlock_and_wait(map, 0);
3033 vm_map_lock(map);
3034 if (last_timestamp + 1 != map->timestamp) {
3035 /*
3036 * Look again for the entry because the map was
3037 * modified while it was unlocked.
3038 * Specifically, the entry may have been
3039 * clipped, merged, or deleted.
3040 */
3041 if (!vm_map_lookup_entry(map, saved_start,
3042 &tmp_entry))
3043 entry = tmp_entry->next;
3044 else {
3045 entry = tmp_entry;
3046 vm_map_clip_start(map, entry,
3047 saved_start);
3048 }
3049 }
3050 continue;
3051 }
3052 vm_map_clip_end(map, entry, end);
3053
3054 next = entry->next;
3055
3056 /*
3057 * Unwire before removing addresses from the pmap; otherwise,
3058 * unwiring will put the entries back in the pmap.
3059 */
3060 if (entry->wired_count != 0) {
3061 vm_map_entry_unwire(map, entry);
3062 }
3063
3064 pmap_remove(map->pmap, entry->start, entry->end);
3065
3066 /*
3067 * Delete the entry only after removing all pmap
3068 * entries pointing to its pages. (Otherwise, its
3069 * page frames may be reallocated, and any modify bits
3070 * will be set in the wrong object!)
3071 */
3072 vm_map_entry_delete(map, entry);
3073 entry = next;
3074 }
3075 return (KERN_SUCCESS);
3076 }
3077
3078 /*
3079 * vm_map_remove:
3080 *
3081 * Remove the given address range from the target map.
3082 * This is the exported form of vm_map_delete.
3083 */
3084 int
3085 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3086 {
3087 int result;
3088
3089 vm_map_lock(map);
3090 VM_MAP_RANGE_CHECK(map, start, end);
3091 result = vm_map_delete(map, start, end);
3092 vm_map_unlock(map);
3093 return (result);
3094 }
3095
3096 /*
3097 * vm_map_check_protection:
3098 *
3099 * Assert that the target map allows the specified privilege on the
3100 * entire address region given. The entire region must be allocated.
3101 *
3102 * WARNING! This code does not and should not check whether the
3103 * contents of the region is accessible. For example a smaller file
3104 * might be mapped into a larger address space.
3105 *
3106 * NOTE! This code is also called by munmap().
3107 *
3108 * The map must be locked. A read lock is sufficient.
3109 */
3110 boolean_t
3111 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3112 vm_prot_t protection)
3113 {
3114 vm_map_entry_t entry;
3115 vm_map_entry_t tmp_entry;
3116
3117 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3118 return (FALSE);
3119 entry = tmp_entry;
3120
3121 while (start < end) {
3122 if (entry == &map->header)
3123 return (FALSE);
3124 /*
3125 * No holes allowed!
3126 */
3127 if (start < entry->start)
3128 return (FALSE);
3129 /*
3130 * Check protection associated with entry.
3131 */
3132 if ((entry->protection & protection) != protection)
3133 return (FALSE);
3134 /* go to next entry */
3135 start = entry->end;
3136 entry = entry->next;
3137 }
3138 return (TRUE);
3139 }
3140
3141 /*
3142 * vm_map_copy_entry:
3143 *
3144 * Copies the contents of the source entry to the destination
3145 * entry. The entries *must* be aligned properly.
3146 */
3147 static void
3148 vm_map_copy_entry(
3149 vm_map_t src_map,
3150 vm_map_t dst_map,
3151 vm_map_entry_t src_entry,
3152 vm_map_entry_t dst_entry,
3153 vm_ooffset_t *fork_charge)
3154 {
3155 vm_object_t src_object;
3156 vm_map_entry_t fake_entry;
3157 vm_offset_t size;
3158 struct ucred *cred;
3159 int charged;
3160
3161 VM_MAP_ASSERT_LOCKED(dst_map);
3162
3163 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3164 return;
3165
3166 if (src_entry->wired_count == 0 ||
3167 (src_entry->protection & VM_PROT_WRITE) == 0) {
3168 /*
3169 * If the source entry is marked needs_copy, it is already
3170 * write-protected.
3171 */
3172 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3173 (src_entry->protection & VM_PROT_WRITE) != 0) {
3174 pmap_protect(src_map->pmap,
3175 src_entry->start,
3176 src_entry->end,
3177 src_entry->protection & ~VM_PROT_WRITE);
3178 }
3179
3180 /*
3181 * Make a copy of the object.
3182 */
3183 size = src_entry->end - src_entry->start;
3184 if ((src_object = src_entry->object.vm_object) != NULL) {
3185 VM_OBJECT_WLOCK(src_object);
3186 charged = ENTRY_CHARGED(src_entry);
3187 if (src_object->handle == NULL &&
3188 (src_object->type == OBJT_DEFAULT ||
3189 src_object->type == OBJT_SWAP)) {
3190 vm_object_collapse(src_object);
3191 if ((src_object->flags & (OBJ_NOSPLIT |
3192 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3193 vm_object_split(src_entry);
3194 src_object =
3195 src_entry->object.vm_object;
3196 }
3197 }
3198 vm_object_reference_locked(src_object);
3199 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3200 if (src_entry->cred != NULL &&
3201 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3202 KASSERT(src_object->cred == NULL,
3203 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3204 src_object));
3205 src_object->cred = src_entry->cred;
3206 src_object->charge = size;
3207 }
3208 VM_OBJECT_WUNLOCK(src_object);
3209 dst_entry->object.vm_object = src_object;
3210 if (charged) {
3211 cred = curthread->td_ucred;
3212 crhold(cred);
3213 dst_entry->cred = cred;
3214 *fork_charge += size;
3215 if (!(src_entry->eflags &
3216 MAP_ENTRY_NEEDS_COPY)) {
3217 crhold(cred);
3218 src_entry->cred = cred;
3219 *fork_charge += size;
3220 }
3221 }
3222 src_entry->eflags |= MAP_ENTRY_COW |
3223 MAP_ENTRY_NEEDS_COPY;
3224 dst_entry->eflags |= MAP_ENTRY_COW |
3225 MAP_ENTRY_NEEDS_COPY;
3226 dst_entry->offset = src_entry->offset;
3227 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3228 /*
3229 * MAP_ENTRY_VN_WRITECNT cannot
3230 * indicate write reference from
3231 * src_entry, since the entry is
3232 * marked as needs copy. Allocate a
3233 * fake entry that is used to
3234 * decrement object->un_pager.vnp.writecount
3235 * at the appropriate time. Attach
3236 * fake_entry to the deferred list.
3237 */
3238 fake_entry = vm_map_entry_create(dst_map);
3239 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3240 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3241 vm_object_reference(src_object);
3242 fake_entry->object.vm_object = src_object;
3243 fake_entry->start = src_entry->start;
3244 fake_entry->end = src_entry->end;
3245 fake_entry->next = curthread->td_map_def_user;
3246 curthread->td_map_def_user = fake_entry;
3247 }
3248
3249 pmap_copy(dst_map->pmap, src_map->pmap,
3250 dst_entry->start, dst_entry->end - dst_entry->start,
3251 src_entry->start);
3252 } else {
3253 dst_entry->object.vm_object = NULL;
3254 dst_entry->offset = 0;
3255 if (src_entry->cred != NULL) {
3256 dst_entry->cred = curthread->td_ucred;
3257 crhold(dst_entry->cred);
3258 *fork_charge += size;
3259 }
3260 }
3261 } else {
3262 /*
3263 * We don't want to make writeable wired pages copy-on-write.
3264 * Immediately copy these pages into the new map by simulating
3265 * page faults. The new pages are pageable.
3266 */
3267 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3268 fork_charge);
3269 }
3270 }
3271
3272 /*
3273 * vmspace_map_entry_forked:
3274 * Update the newly-forked vmspace each time a map entry is inherited
3275 * or copied. The values for vm_dsize and vm_tsize are approximate
3276 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3277 */
3278 static void
3279 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3280 vm_map_entry_t entry)
3281 {
3282 vm_size_t entrysize;
3283 vm_offset_t newend;
3284
3285 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3286 return;
3287 entrysize = entry->end - entry->start;
3288 vm2->vm_map.size += entrysize;
3289 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3290 vm2->vm_ssize += btoc(entrysize);
3291 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3292 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3293 newend = MIN(entry->end,
3294 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3295 vm2->vm_dsize += btoc(newend - entry->start);
3296 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3297 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3298 newend = MIN(entry->end,
3299 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3300 vm2->vm_tsize += btoc(newend - entry->start);
3301 }
3302 }
3303
3304 /*
3305 * vmspace_fork:
3306 * Create a new process vmspace structure and vm_map
3307 * based on those of an existing process. The new map
3308 * is based on the old map, according to the inheritance
3309 * values on the regions in that map.
3310 *
3311 * XXX It might be worth coalescing the entries added to the new vmspace.
3312 *
3313 * The source map must not be locked.
3314 */
3315 struct vmspace *
3316 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3317 {
3318 struct vmspace *vm2;
3319 vm_map_t new_map, old_map;
3320 vm_map_entry_t new_entry, old_entry;
3321 vm_object_t object;
3322 int locked;
3323 vm_inherit_t inh;
3324
3325 old_map = &vm1->vm_map;
3326 /* Copy immutable fields of vm1 to vm2. */
3327 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3328 if (vm2 == NULL)
3329 return (NULL);
3330 vm2->vm_taddr = vm1->vm_taddr;
3331 vm2->vm_daddr = vm1->vm_daddr;
3332 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3333 vm_map_lock(old_map);
3334 if (old_map->busy)
3335 vm_map_wait_busy(old_map);
3336 new_map = &vm2->vm_map;
3337 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3338 KASSERT(locked, ("vmspace_fork: lock failed"));
3339
3340 old_entry = old_map->header.next;
3341
3342 while (old_entry != &old_map->header) {
3343 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3344 panic("vm_map_fork: encountered a submap");
3345
3346 inh = old_entry->inheritance;
3347 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3348 inh != VM_INHERIT_NONE)
3349 inh = VM_INHERIT_COPY;
3350
3351 switch (inh) {
3352 case VM_INHERIT_NONE:
3353 break;
3354
3355 case VM_INHERIT_SHARE:
3356 /*
3357 * Clone the entry, creating the shared object if necessary.
3358 */
3359 object = old_entry->object.vm_object;
3360 if (object == NULL) {
3361 object = vm_object_allocate(OBJT_DEFAULT,
3362 atop(old_entry->end - old_entry->start));
3363 old_entry->object.vm_object = object;
3364 old_entry->offset = 0;
3365 if (old_entry->cred != NULL) {
3366 object->cred = old_entry->cred;
3367 object->charge = old_entry->end -
3368 old_entry->start;
3369 old_entry->cred = NULL;
3370 }
3371 }
3372
3373 /*
3374 * Add the reference before calling vm_object_shadow
3375 * to insure that a shadow object is created.
3376 */
3377 vm_object_reference(object);
3378 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3379 vm_object_shadow(&old_entry->object.vm_object,
3380 &old_entry->offset,
3381 old_entry->end - old_entry->start);
3382 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3383 /* Transfer the second reference too. */
3384 vm_object_reference(
3385 old_entry->object.vm_object);
3386
3387 /*
3388 * As in vm_map_simplify_entry(), the
3389 * vnode lock will not be acquired in
3390 * this call to vm_object_deallocate().
3391 */
3392 vm_object_deallocate(object);
3393 object = old_entry->object.vm_object;
3394 }
3395 VM_OBJECT_WLOCK(object);
3396 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3397 if (old_entry->cred != NULL) {
3398 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3399 object->cred = old_entry->cred;
3400 object->charge = old_entry->end - old_entry->start;
3401 old_entry->cred = NULL;
3402 }
3403
3404 /*
3405 * Assert the correct state of the vnode
3406 * v_writecount while the object is locked, to
3407 * not relock it later for the assertion
3408 * correctness.
3409 */
3410 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3411 object->type == OBJT_VNODE) {
3412 KASSERT(((struct vnode *)object->handle)->
3413 v_writecount > 0,
3414 ("vmspace_fork: v_writecount %p", object));
3415 KASSERT(object->un_pager.vnp.writemappings > 0,
3416 ("vmspace_fork: vnp.writecount %p",
3417 object));
3418 }
3419 VM_OBJECT_WUNLOCK(object);
3420
3421 /*
3422 * Clone the entry, referencing the shared object.
3423 */
3424 new_entry = vm_map_entry_create(new_map);
3425 *new_entry = *old_entry;
3426 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3427 MAP_ENTRY_IN_TRANSITION);
3428 new_entry->wiring_thread = NULL;
3429 new_entry->wired_count = 0;
3430 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3431 vnode_pager_update_writecount(object,
3432 new_entry->start, new_entry->end);
3433 }
3434
3435 /*
3436 * Insert the entry into the new map -- we know we're
3437 * inserting at the end of the new map.
3438 */
3439 vm_map_entry_link(new_map, new_map->header.prev,
3440 new_entry);
3441 vmspace_map_entry_forked(vm1, vm2, new_entry);
3442
3443 /*
3444 * Update the physical map
3445 */
3446 pmap_copy(new_map->pmap, old_map->pmap,
3447 new_entry->start,
3448 (old_entry->end - old_entry->start),
3449 old_entry->start);
3450 break;
3451
3452 case VM_INHERIT_COPY:
3453 /*
3454 * Clone the entry and link into the map.
3455 */
3456 new_entry = vm_map_entry_create(new_map);
3457 *new_entry = *old_entry;
3458 /*
3459 * Copied entry is COW over the old object.
3460 */
3461 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3462 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3463 new_entry->wiring_thread = NULL;
3464 new_entry->wired_count = 0;
3465 new_entry->object.vm_object = NULL;
3466 new_entry->cred = NULL;
3467 vm_map_entry_link(new_map, new_map->header.prev,
3468 new_entry);
3469 vmspace_map_entry_forked(vm1, vm2, new_entry);
3470 vm_map_copy_entry(old_map, new_map, old_entry,
3471 new_entry, fork_charge);
3472 break;
3473
3474 case VM_INHERIT_ZERO:
3475 /*
3476 * Create a new anonymous mapping entry modelled from
3477 * the old one.
3478 */
3479 new_entry = vm_map_entry_create(new_map);
3480 memset(new_entry, 0, sizeof(*new_entry));
3481
3482 new_entry->start = old_entry->start;
3483 new_entry->end = old_entry->end;
3484 new_entry->eflags = old_entry->eflags &
3485 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
3486 MAP_ENTRY_VN_WRITECNT);
3487 new_entry->protection = old_entry->protection;
3488 new_entry->max_protection = old_entry->max_protection;
3489 new_entry->inheritance = VM_INHERIT_ZERO;
3490
3491 vm_map_entry_link(new_map, new_map->header.prev,
3492 new_entry);
3493 vmspace_map_entry_forked(vm1, vm2, new_entry);
3494
3495 new_entry->cred = curthread->td_ucred;
3496 crhold(new_entry->cred);
3497 *fork_charge += (new_entry->end - new_entry->start);
3498
3499 break;
3500 }
3501 old_entry = old_entry->next;
3502 }
3503 /*
3504 * Use inlined vm_map_unlock() to postpone handling the deferred
3505 * map entries, which cannot be done until both old_map and
3506 * new_map locks are released.
3507 */
3508 sx_xunlock(&old_map->lock);
3509 sx_xunlock(&new_map->lock);
3510 vm_map_process_deferred();
3511
3512 return (vm2);
3513 }
3514
3515 /*
3516 * Create a process's stack for exec_new_vmspace(). This function is never
3517 * asked to wire the newly created stack.
3518 */
3519 int
3520 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3521 vm_prot_t prot, vm_prot_t max, int cow)
3522 {
3523 vm_size_t growsize, init_ssize;
3524 rlim_t vmemlim;
3525 int rv;
3526
3527 MPASS((map->flags & MAP_WIREFUTURE) == 0);
3528 growsize = sgrowsiz;
3529 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3530 vm_map_lock(map);
3531 PROC_LOCK(curproc);
3532 vmemlim = lim_cur(curproc, RLIMIT_VMEM);
3533 PROC_UNLOCK(curproc);
3534 /* If we would blow our VMEM resource limit, no go */
3535 if (map->size + init_ssize > vmemlim) {
3536 rv = KERN_NO_SPACE;
3537 goto out;
3538 }
3539 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3540 max, cow);
3541 out:
3542 vm_map_unlock(map);
3543 return (rv);
3544 }
3545
3546 static int stack_guard_page = 1;
3547 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3548 &stack_guard_page, 0,
3549 "Specifies the number of guard pages for a stack that grows");
3550
3551 static int
3552 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3553 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3554 {
3555 vm_map_entry_t new_entry, prev_entry;
3556 vm_offset_t bot, gap_bot, gap_top, top;
3557 vm_size_t init_ssize, sgp;
3558 int orient, rv;
3559
3560 /*
3561 * The stack orientation is piggybacked with the cow argument.
3562 * Extract it into orient and mask the cow argument so that we
3563 * don't pass it around further.
3564 */
3565 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
3566 KASSERT(orient != 0, ("No stack grow direction"));
3567 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
3568 ("bi-dir stack"));
3569
3570 if (addrbos < vm_map_min(map) ||
3571 addrbos + max_ssize > vm_map_max(map) ||
3572 addrbos + max_ssize <= addrbos)
3573 return (KERN_INVALID_ADDRESS);
3574 sgp = (vm_size_t)stack_guard_page * PAGE_SIZE;
3575 if (sgp >= max_ssize)
3576 return (KERN_INVALID_ARGUMENT);
3577
3578 init_ssize = growsize;
3579 if (max_ssize < init_ssize + sgp)
3580 init_ssize = max_ssize - sgp;
3581
3582 /* If addr is already mapped, no go */
3583 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3584 return (KERN_NO_SPACE);
3585
3586 /*
3587 * If we can't accomodate max_ssize in the current mapping, no go.
3588 */
3589 if ((prev_entry->next != &map->header) &&
3590 (prev_entry->next->start < addrbos + max_ssize))
3591 return (KERN_NO_SPACE);
3592
3593 /*
3594 * We initially map a stack of only init_ssize. We will grow as
3595 * needed later. Depending on the orientation of the stack (i.e.
3596 * the grow direction) we either map at the top of the range, the
3597 * bottom of the range or in the middle.
3598 *
3599 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3600 * and cow to be 0. Possibly we should eliminate these as input
3601 * parameters, and just pass these values here in the insert call.
3602 */
3603 if (orient == MAP_STACK_GROWS_DOWN) {
3604 bot = addrbos + max_ssize - init_ssize;
3605 top = bot + init_ssize;
3606 gap_bot = addrbos;
3607 gap_top = bot;
3608 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
3609 bot = addrbos;
3610 top = bot + init_ssize;
3611 gap_bot = top;
3612 gap_top = addrbos + max_ssize;
3613 }
3614 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3615 if (rv != KERN_SUCCESS)
3616 return (rv);
3617 new_entry = prev_entry->next;
3618 KASSERT(new_entry->end == top || new_entry->start == bot,
3619 ("Bad entry start/end for new stack entry"));
3620 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3621 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3622 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3623 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3624 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3625 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3626 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
3627 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
3628 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
3629 if (rv != KERN_SUCCESS)
3630 (void)vm_map_delete(map, bot, top);
3631 return (rv);
3632 }
3633
3634 /*
3635 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
3636 * successfully grow the stack.
3637 */
3638 static int
3639 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
3640 {
3641 vm_map_entry_t stack_entry;
3642 struct proc *p;
3643 struct vmspace *vm;
3644 struct ucred *cred;
3645 vm_offset_t gap_end, gap_start, grow_start;
3646 size_t grow_amount, guard, max_grow;
3647 rlim_t lmemlim, stacklim, vmemlim;
3648 int rv, rv1;
3649 bool gap_deleted, grow_down, is_procstack;
3650 #ifdef notyet
3651 uint64_t limit;
3652 #endif
3653 #ifdef RACCT
3654 int error;
3655 #endif
3656
3657 p = curproc;
3658 vm = p->p_vmspace;
3659
3660 /*
3661 * Disallow stack growth when the access is performed by a
3662 * debugger or AIO daemon. The reason is that the wrong
3663 * resource limits are applied.
3664 */
3665 if (map != &p->p_vmspace->vm_map || p->p_textvp == NULL)
3666 return (KERN_FAILURE);
3667
3668 MPASS(!map->system_map);
3669
3670 guard = stack_guard_page * PAGE_SIZE;
3671 PROC_LOCK(p);
3672 lmemlim = lim_cur(p, RLIMIT_MEMLOCK);
3673 stacklim = lim_cur(p, RLIMIT_STACK);
3674 vmemlim = lim_cur(p, RLIMIT_VMEM);
3675 PROC_UNLOCK(p);
3676 retry:
3677 /* If addr is not in a hole for a stack grow area, no need to grow. */
3678 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
3679 return (KERN_FAILURE);
3680 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
3681 return (KERN_SUCCESS);
3682 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
3683 stack_entry = gap_entry->next;
3684 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
3685 stack_entry->start != gap_entry->end)
3686 return (KERN_FAILURE);
3687 grow_amount = round_page(stack_entry->start - addr);
3688 grow_down = true;
3689 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
3690 stack_entry = gap_entry->prev;
3691 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
3692 stack_entry->end != gap_entry->start)
3693 return (KERN_FAILURE);
3694 grow_amount = round_page(addr + 1 - stack_entry->end);
3695 grow_down = false;
3696 } else {
3697 return (KERN_FAILURE);
3698 }
3699 max_grow = gap_entry->end - gap_entry->start;
3700 if (guard > max_grow)
3701 return (KERN_NO_SPACE);
3702 max_grow -= guard;
3703 if (grow_amount > max_grow)
3704 return (KERN_NO_SPACE);
3705
3706 /*
3707 * If this is the main process stack, see if we're over the stack
3708 * limit.
3709 */
3710 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
3711 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
3712 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
3713 return (KERN_NO_SPACE);
3714
3715 #ifdef RACCT
3716 if (racct_enable) {
3717 PROC_LOCK(p);
3718 if (is_procstack && racct_set(p, RACCT_STACK,
3719 ctob(vm->vm_ssize) + grow_amount)) {
3720 PROC_UNLOCK(p);
3721 return (KERN_NO_SPACE);
3722 }
3723 PROC_UNLOCK(p);
3724 }
3725 #endif
3726
3727 grow_amount = roundup(grow_amount, sgrowsiz);
3728 if (grow_amount > max_grow)
3729 grow_amount = max_grow;
3730 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3731 grow_amount = trunc_page((vm_size_t)stacklim) -
3732 ctob(vm->vm_ssize);
3733 }
3734
3735 #ifdef notyet
3736 PROC_LOCK(p);
3737 limit = racct_get_available(p, RACCT_STACK);
3738 PROC_UNLOCK(p);
3739 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3740 grow_amount = limit - ctob(vm->vm_ssize);
3741 #endif
3742
3743 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
3744 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3745 rv = KERN_NO_SPACE;
3746 goto out;
3747 }
3748 #ifdef RACCT
3749 if (racct_enable) {
3750 PROC_LOCK(p);
3751 if (racct_set(p, RACCT_MEMLOCK,
3752 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3753 PROC_UNLOCK(p);
3754 rv = KERN_NO_SPACE;
3755 goto out;
3756 }
3757 PROC_UNLOCK(p);
3758 }
3759 #endif
3760 }
3761
3762 /* If we would blow our VMEM resource limit, no go */
3763 if (map->size + grow_amount > vmemlim) {
3764 rv = KERN_NO_SPACE;
3765 goto out;
3766 }
3767 #ifdef RACCT
3768 if (racct_enable) {
3769 PROC_LOCK(p);
3770 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3771 PROC_UNLOCK(p);
3772 rv = KERN_NO_SPACE;
3773 goto out;
3774 }
3775 PROC_UNLOCK(p);
3776 }
3777 #endif
3778
3779 if (vm_map_lock_upgrade(map)) {
3780 gap_entry = NULL;
3781 vm_map_lock_read(map);
3782 goto retry;
3783 }
3784
3785 if (grow_down) {
3786 grow_start = gap_entry->end - grow_amount;
3787 if (gap_entry->start + grow_amount == gap_entry->end) {
3788 gap_start = gap_entry->start;
3789 gap_end = gap_entry->end;
3790 vm_map_entry_delete(map, gap_entry);
3791 gap_deleted = true;
3792 } else {
3793 MPASS(gap_entry->start < gap_entry->end - grow_amount);
3794 gap_entry->end -= grow_amount;
3795 vm_map_entry_resize_free(map, gap_entry);
3796 gap_deleted = false;
3797 }
3798 rv = vm_map_insert(map, NULL, 0, grow_start,
3799 grow_start + grow_amount,
3800 stack_entry->protection, stack_entry->max_protection,
3801 MAP_STACK_GROWS_DOWN);
3802 if (rv != KERN_SUCCESS) {
3803 if (gap_deleted) {
3804 rv1 = vm_map_insert(map, NULL, 0, gap_start,
3805 gap_end, VM_PROT_NONE, VM_PROT_NONE,
3806 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
3807 MPASS(rv1 == KERN_SUCCESS);
3808 } else {
3809 gap_entry->end += grow_amount;
3810 vm_map_entry_resize_free(map, gap_entry);
3811 }
3812 }
3813 } else {
3814 grow_start = stack_entry->end;
3815 cred = stack_entry->cred;
3816 if (cred == NULL && stack_entry->object.vm_object != NULL)
3817 cred = stack_entry->object.vm_object->cred;
3818 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3819 rv = KERN_NO_SPACE;
3820 /* Grow the underlying object if applicable. */
3821 else if (stack_entry->object.vm_object == NULL ||
3822 vm_object_coalesce(stack_entry->object.vm_object,
3823 stack_entry->offset,
3824 (vm_size_t)(stack_entry->end - stack_entry->start),
3825 (vm_size_t)grow_amount, cred != NULL)) {
3826 if (gap_entry->start + grow_amount == gap_entry->end)
3827 vm_map_entry_delete(map, gap_entry);
3828 else
3829 gap_entry->start += grow_amount;
3830 stack_entry->end += grow_amount;
3831 map->size += grow_amount;
3832 vm_map_entry_resize_free(map, stack_entry);
3833 rv = KERN_SUCCESS;
3834 } else
3835 rv = KERN_FAILURE;
3836 }
3837 if (rv == KERN_SUCCESS && is_procstack)
3838 vm->vm_ssize += btoc(grow_amount);
3839
3840 /*
3841 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3842 */
3843 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
3844 vm_map_unlock(map);
3845 vm_map_wire(map, grow_start, grow_start + grow_amount,
3846 (p->p_flag & P_SYSTEM)
3847 ? VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES
3848 : VM_MAP_WIRE_USER|VM_MAP_WIRE_NOHOLES);
3849 vm_map_lock_read(map);
3850 } else
3851 vm_map_lock_downgrade(map);
3852
3853 out:
3854 #ifdef RACCT
3855 if (racct_enable && rv != KERN_SUCCESS) {
3856 PROC_LOCK(p);
3857 error = racct_set(p, RACCT_VMEM, map->size);
3858 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3859 if (!old_mlock) {
3860 error = racct_set(p, RACCT_MEMLOCK,
3861 ptoa(pmap_wired_count(map->pmap)));
3862 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3863 }
3864 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3865 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3866 PROC_UNLOCK(p);
3867 }
3868 #endif
3869
3870 return (rv);
3871 }
3872
3873 /*
3874 * Unshare the specified VM space for exec. If other processes are
3875 * mapped to it, then create a new one. The new vmspace is null.
3876 */
3877 int
3878 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3879 {
3880 struct vmspace *oldvmspace = p->p_vmspace;
3881 struct vmspace *newvmspace;
3882
3883 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3884 ("vmspace_exec recursed"));
3885 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3886 if (newvmspace == NULL)
3887 return (ENOMEM);
3888 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3889 /*
3890 * This code is written like this for prototype purposes. The
3891 * goal is to avoid running down the vmspace here, but let the
3892 * other process's that are still using the vmspace to finally
3893 * run it down. Even though there is little or no chance of blocking
3894 * here, it is a good idea to keep this form for future mods.
3895 */
3896 PROC_VMSPACE_LOCK(p);
3897 p->p_vmspace = newvmspace;
3898 PROC_VMSPACE_UNLOCK(p);
3899 if (p == curthread->td_proc)
3900 pmap_activate(curthread);
3901 curthread->td_pflags |= TDP_EXECVMSPC;
3902 return (0);
3903 }
3904
3905 /*
3906 * Unshare the specified VM space for forcing COW. This
3907 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3908 */
3909 int
3910 vmspace_unshare(struct proc *p)
3911 {
3912 struct vmspace *oldvmspace = p->p_vmspace;
3913 struct vmspace *newvmspace;
3914 vm_ooffset_t fork_charge;
3915
3916 if (oldvmspace->vm_refcnt == 1)
3917 return (0);
3918 fork_charge = 0;
3919 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3920 if (newvmspace == NULL)
3921 return (ENOMEM);
3922 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3923 vmspace_free(newvmspace);
3924 return (ENOMEM);
3925 }
3926 PROC_VMSPACE_LOCK(p);
3927 p->p_vmspace = newvmspace;
3928 PROC_VMSPACE_UNLOCK(p);
3929 if (p == curthread->td_proc)
3930 pmap_activate(curthread);
3931 vmspace_free(oldvmspace);
3932 return (0);
3933 }
3934
3935 /*
3936 * vm_map_lookup:
3937 *
3938 * Finds the VM object, offset, and
3939 * protection for a given virtual address in the
3940 * specified map, assuming a page fault of the
3941 * type specified.
3942 *
3943 * Leaves the map in question locked for read; return
3944 * values are guaranteed until a vm_map_lookup_done
3945 * call is performed. Note that the map argument
3946 * is in/out; the returned map must be used in
3947 * the call to vm_map_lookup_done.
3948 *
3949 * A handle (out_entry) is returned for use in
3950 * vm_map_lookup_done, to make that fast.
3951 *
3952 * If a lookup is requested with "write protection"
3953 * specified, the map may be changed to perform virtual
3954 * copying operations, although the data referenced will
3955 * remain the same.
3956 */
3957 int
3958 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3959 vm_offset_t vaddr,
3960 vm_prot_t fault_typea,
3961 vm_map_entry_t *out_entry, /* OUT */
3962 vm_object_t *object, /* OUT */
3963 vm_pindex_t *pindex, /* OUT */
3964 vm_prot_t *out_prot, /* OUT */
3965 boolean_t *wired) /* OUT */
3966 {
3967 vm_map_entry_t entry;
3968 vm_map_t map = *var_map;
3969 vm_prot_t prot;
3970 vm_prot_t fault_type = fault_typea;
3971 vm_object_t eobject;
3972 vm_size_t size;
3973 struct ucred *cred;
3974
3975 RetryLookup:
3976
3977 vm_map_lock_read(map);
3978
3979 RetryLookupLocked:
3980 /*
3981 * Lookup the faulting address.
3982 */
3983 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
3984 vm_map_unlock_read(map);
3985 return (KERN_INVALID_ADDRESS);
3986 }
3987
3988 entry = *out_entry;
3989
3990 /*
3991 * Handle submaps.
3992 */
3993 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3994 vm_map_t old_map = map;
3995
3996 *var_map = map = entry->object.sub_map;
3997 vm_map_unlock_read(old_map);
3998 goto RetryLookup;
3999 }
4000
4001 /*
4002 * Check whether this task is allowed to have this page.
4003 */
4004 prot = entry->protection;
4005 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4006 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4007 if (prot == VM_PROT_NONE && map != kernel_map &&
4008 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4009 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4010 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4011 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4012 goto RetryLookupLocked;
4013 }
4014 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4015 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4016 vm_map_unlock_read(map);
4017 return (KERN_PROTECTION_FAILURE);
4018 }
4019 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4020 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4021 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4022 ("entry %p flags %x", entry, entry->eflags));
4023 if ((fault_typea & VM_PROT_COPY) != 0 &&
4024 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4025 (entry->eflags & MAP_ENTRY_COW) == 0) {
4026 vm_map_unlock_read(map);
4027 return (KERN_PROTECTION_FAILURE);
4028 }
4029
4030 /*
4031 * If this page is not pageable, we have to get it for all possible
4032 * accesses.
4033 */
4034 *wired = (entry->wired_count != 0);
4035 if (*wired)
4036 fault_type = entry->protection;
4037 size = entry->end - entry->start;
4038 /*
4039 * If the entry was copy-on-write, we either ...
4040 */
4041 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4042 /*
4043 * If we want to write the page, we may as well handle that
4044 * now since we've got the map locked.
4045 *
4046 * If we don't need to write the page, we just demote the
4047 * permissions allowed.
4048 */
4049 if ((fault_type & VM_PROT_WRITE) != 0 ||
4050 (fault_typea & VM_PROT_COPY) != 0) {
4051 /*
4052 * Make a new object, and place it in the object
4053 * chain. Note that no new references have appeared
4054 * -- one just moved from the map to the new
4055 * object.
4056 */
4057 if (vm_map_lock_upgrade(map))
4058 goto RetryLookup;
4059
4060 if (entry->cred == NULL) {
4061 /*
4062 * The debugger owner is charged for
4063 * the memory.
4064 */
4065 cred = curthread->td_ucred;
4066 crhold(cred);
4067 if (!swap_reserve_by_cred(size, cred)) {
4068 crfree(cred);
4069 vm_map_unlock(map);
4070 return (KERN_RESOURCE_SHORTAGE);
4071 }
4072 entry->cred = cred;
4073 }
4074 vm_object_shadow(&entry->object.vm_object,
4075 &entry->offset, size);
4076 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4077 eobject = entry->object.vm_object;
4078 if (eobject->cred != NULL) {
4079 /*
4080 * The object was not shadowed.
4081 */
4082 swap_release_by_cred(size, entry->cred);
4083 crfree(entry->cred);
4084 entry->cred = NULL;
4085 } else if (entry->cred != NULL) {
4086 VM_OBJECT_WLOCK(eobject);
4087 eobject->cred = entry->cred;
4088 eobject->charge = size;
4089 VM_OBJECT_WUNLOCK(eobject);
4090 entry->cred = NULL;
4091 }
4092
4093 vm_map_lock_downgrade(map);
4094 } else {
4095 /*
4096 * We're attempting to read a copy-on-write page --
4097 * don't allow writes.
4098 */
4099 prot &= ~VM_PROT_WRITE;
4100 }
4101 }
4102
4103 /*
4104 * Create an object if necessary.
4105 */
4106 if (entry->object.vm_object == NULL &&
4107 !map->system_map) {
4108 if (vm_map_lock_upgrade(map))
4109 goto RetryLookup;
4110 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4111 atop(size));
4112 entry->offset = 0;
4113 if (entry->cred != NULL) {
4114 VM_OBJECT_WLOCK(entry->object.vm_object);
4115 entry->object.vm_object->cred = entry->cred;
4116 entry->object.vm_object->charge = size;
4117 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4118 entry->cred = NULL;
4119 }
4120 vm_map_lock_downgrade(map);
4121 }
4122
4123 /*
4124 * Return the object/offset from this entry. If the entry was
4125 * copy-on-write or empty, it has been fixed up.
4126 */
4127 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4128 *object = entry->object.vm_object;
4129
4130 *out_prot = prot;
4131 return (KERN_SUCCESS);
4132 }
4133
4134 /*
4135 * vm_map_lookup_locked:
4136 *
4137 * Lookup the faulting address. A version of vm_map_lookup that returns
4138 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4139 */
4140 int
4141 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4142 vm_offset_t vaddr,
4143 vm_prot_t fault_typea,
4144 vm_map_entry_t *out_entry, /* OUT */
4145 vm_object_t *object, /* OUT */
4146 vm_pindex_t *pindex, /* OUT */
4147 vm_prot_t *out_prot, /* OUT */
4148 boolean_t *wired) /* OUT */
4149 {
4150 vm_map_entry_t entry;
4151 vm_map_t map = *var_map;
4152 vm_prot_t prot;
4153 vm_prot_t fault_type = fault_typea;
4154
4155 /*
4156 * Lookup the faulting address.
4157 */
4158 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4159 return (KERN_INVALID_ADDRESS);
4160
4161 entry = *out_entry;
4162
4163 /*
4164 * Fail if the entry refers to a submap.
4165 */
4166 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4167 return (KERN_FAILURE);
4168
4169 /*
4170 * Check whether this task is allowed to have this page.
4171 */
4172 prot = entry->protection;
4173 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4174 if ((fault_type & prot) != fault_type)
4175 return (KERN_PROTECTION_FAILURE);
4176
4177 /*
4178 * If this page is not pageable, we have to get it for all possible
4179 * accesses.
4180 */
4181 *wired = (entry->wired_count != 0);
4182 if (*wired)
4183 fault_type = entry->protection;
4184
4185 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4186 /*
4187 * Fail if the entry was copy-on-write for a write fault.
4188 */
4189 if (fault_type & VM_PROT_WRITE)
4190 return (KERN_FAILURE);
4191 /*
4192 * We're attempting to read a copy-on-write page --
4193 * don't allow writes.
4194 */
4195 prot &= ~VM_PROT_WRITE;
4196 }
4197
4198 /*
4199 * Fail if an object should be created.
4200 */
4201 if (entry->object.vm_object == NULL && !map->system_map)
4202 return (KERN_FAILURE);
4203
4204 /*
4205 * Return the object/offset from this entry. If the entry was
4206 * copy-on-write or empty, it has been fixed up.
4207 */
4208 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4209 *object = entry->object.vm_object;
4210
4211 *out_prot = prot;
4212 return (KERN_SUCCESS);
4213 }
4214
4215 /*
4216 * vm_map_lookup_done:
4217 *
4218 * Releases locks acquired by a vm_map_lookup
4219 * (according to the handle returned by that lookup).
4220 */
4221 void
4222 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4223 {
4224 /*
4225 * Unlock the main-level map
4226 */
4227 vm_map_unlock_read(map);
4228 }
4229
4230 #include "opt_ddb.h"
4231 #ifdef DDB
4232 #include <sys/kernel.h>
4233
4234 #include <ddb/ddb.h>
4235
4236 static void
4237 vm_map_print(vm_map_t map)
4238 {
4239 vm_map_entry_t entry;
4240
4241 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4242 (void *)map,
4243 (void *)map->pmap, map->nentries, map->timestamp);
4244
4245 db_indent += 2;
4246 for (entry = map->header.next; entry != &map->header;
4247 entry = entry->next) {
4248 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4249 (void *)entry, (void *)entry->start, (void *)entry->end,
4250 entry->eflags);
4251 {
4252 static char *inheritance_name[4] =
4253 {"share", "copy", "none", "donate_copy"};
4254
4255 db_iprintf(" prot=%x/%x/%s",
4256 entry->protection,
4257 entry->max_protection,
4258 inheritance_name[(int)(unsigned char)entry->inheritance]);
4259 if (entry->wired_count != 0)
4260 db_printf(", wired");
4261 }
4262 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4263 db_printf(", share=%p, offset=0x%jx\n",
4264 (void *)entry->object.sub_map,
4265 (uintmax_t)entry->offset);
4266 if ((entry->prev == &map->header) ||
4267 (entry->prev->object.sub_map !=
4268 entry->object.sub_map)) {
4269 db_indent += 2;
4270 vm_map_print((vm_map_t)entry->object.sub_map);
4271 db_indent -= 2;
4272 }
4273 } else {
4274 if (entry->cred != NULL)
4275 db_printf(", ruid %d", entry->cred->cr_ruid);
4276 db_printf(", object=%p, offset=0x%jx",
4277 (void *)entry->object.vm_object,
4278 (uintmax_t)entry->offset);
4279 if (entry->object.vm_object && entry->object.vm_object->cred)
4280 db_printf(", obj ruid %d charge %jx",
4281 entry->object.vm_object->cred->cr_ruid,
4282 (uintmax_t)entry->object.vm_object->charge);
4283 if (entry->eflags & MAP_ENTRY_COW)
4284 db_printf(", copy (%s)",
4285 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4286 db_printf("\n");
4287
4288 if ((entry->prev == &map->header) ||
4289 (entry->prev->object.vm_object !=
4290 entry->object.vm_object)) {
4291 db_indent += 2;
4292 vm_object_print((db_expr_t)(intptr_t)
4293 entry->object.vm_object,
4294 0, 0, (char *)0);
4295 db_indent -= 2;
4296 }
4297 }
4298 }
4299 db_indent -= 2;
4300 }
4301
4302 DB_SHOW_COMMAND(map, map)
4303 {
4304
4305 if (!have_addr) {
4306 db_printf("usage: show map <addr>\n");
4307 return;
4308 }
4309 vm_map_print((vm_map_t)addr);
4310 }
4311
4312 DB_SHOW_COMMAND(procvm, procvm)
4313 {
4314 struct proc *p;
4315
4316 if (have_addr) {
4317 p = db_lookup_proc(addr);
4318 } else {
4319 p = curproc;
4320 }
4321
4322 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4323 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4324 (void *)vmspace_pmap(p->p_vmspace));
4325
4326 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);
4327 }
4328
4329 #endif /* DDB */
Cache object: c00dc60ab4185bcfb60dd218218a2fc4
|