FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_map.c
1 /*-
2 * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
3 *
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
35 *
36 *
37 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38 * All rights reserved.
39 *
40 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
41 *
42 * Permission to use, copy, modify and distribute this software and
43 * its documentation is hereby granted, provided that both the copyright
44 * notice and this permission notice appear in all copies of the
45 * software, derivative works or modified versions, and any portions
46 * thereof, and that both notices appear in supporting documentation.
47 *
48 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
51 *
52 * Carnegie Mellon requests users of this software to return to
53 *
54 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
55 * School of Computer Science
56 * Carnegie Mellon University
57 * Pittsburgh PA 15213-3890
58 *
59 * any improvements or extensions that they make and grant Carnegie the
60 * rights to redistribute these changes.
61 */
62
63 /*
64 * Virtual memory mapping module.
65 */
66
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD: head/sys/vm/vm_map.c 340064 2018-11-02 16:26:44Z markj $");
69
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/kernel.h>
73 #include <sys/ktr.h>
74 #include <sys/lock.h>
75 #include <sys/mutex.h>
76 #include <sys/proc.h>
77 #include <sys/vmmeter.h>
78 #include <sys/mman.h>
79 #include <sys/vnode.h>
80 #include <sys/racct.h>
81 #include <sys/resourcevar.h>
82 #include <sys/rwlock.h>
83 #include <sys/file.h>
84 #include <sys/sysctl.h>
85 #include <sys/sysent.h>
86 #include <sys/shm.h>
87
88 #include <vm/vm.h>
89 #include <vm/vm_param.h>
90 #include <vm/pmap.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_page.h>
93 #include <vm/vm_object.h>
94 #include <vm/vm_pager.h>
95 #include <vm/vm_kern.h>
96 #include <vm/vm_extern.h>
97 #include <vm/vnode_pager.h>
98 #include <vm/swap_pager.h>
99 #include <vm/uma.h>
100
101 /*
102 * Virtual memory maps provide for the mapping, protection,
103 * and sharing of virtual memory objects. In addition,
104 * this module provides for an efficient virtual copy of
105 * memory from one map to another.
106 *
107 * Synchronization is required prior to most operations.
108 *
109 * Maps consist of an ordered doubly-linked list of simple
110 * entries; a self-adjusting binary search tree of these
111 * entries is used to speed up lookups.
112 *
113 * Since portions of maps are specified by start/end addresses,
114 * which may not align with existing map entries, all
115 * routines merely "clip" entries to these start/end values.
116 * [That is, an entry is split into two, bordering at a
117 * start or end value.] Note that these clippings may not
118 * always be necessary (as the two resulting entries are then
119 * not changed); however, the clipping is done for convenience.
120 *
121 * As mentioned above, virtual copy operations are performed
122 * by copying VM object references from one map to
123 * another, and then marking both regions as copy-on-write.
124 */
125
126 static struct mtx map_sleep_mtx;
127 static uma_zone_t mapentzone;
128 static uma_zone_t kmapentzone;
129 static uma_zone_t mapzone;
130 static uma_zone_t vmspace_zone;
131 static int vmspace_zinit(void *mem, int size, int flags);
132 static int vm_map_zinit(void *mem, int ize, int flags);
133 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
134 vm_offset_t max);
135 static int vm_map_alignspace(vm_map_t map, vm_object_t object,
136 vm_ooffset_t offset, vm_offset_t *addr, vm_size_t length,
137 vm_offset_t max_addr, vm_offset_t alignment);
138 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
139 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
140 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
141 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
142 vm_map_entry_t gap_entry);
143 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
144 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
145 #ifdef INVARIANTS
146 static void vm_map_zdtor(void *mem, int size, void *arg);
147 static void vmspace_zdtor(void *mem, int size, void *arg);
148 #endif
149 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
150 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
151 int cow);
152 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
153 vm_offset_t failed_addr);
154
155 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
156 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
157 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
158
159 /*
160 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
161 * stable.
162 */
163 #define PROC_VMSPACE_LOCK(p) do { } while (0)
164 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
165
166 /*
167 * VM_MAP_RANGE_CHECK: [ internal use only ]
168 *
169 * Asserts that the starting and ending region
170 * addresses fall within the valid range of the map.
171 */
172 #define VM_MAP_RANGE_CHECK(map, start, end) \
173 { \
174 if (start < vm_map_min(map)) \
175 start = vm_map_min(map); \
176 if (end > vm_map_max(map)) \
177 end = vm_map_max(map); \
178 if (start > end) \
179 start = end; \
180 }
181
182 /*
183 * vm_map_startup:
184 *
185 * Initialize the vm_map module. Must be called before
186 * any other vm_map routines.
187 *
188 * Map and entry structures are allocated from the general
189 * purpose memory pool with some exceptions:
190 *
191 * - The kernel map and kmem submap are allocated statically.
192 * - Kernel map entries are allocated out of a static pool.
193 *
194 * These restrictions are necessary since malloc() uses the
195 * maps and requires map entries.
196 */
197
198 void
199 vm_map_startup(void)
200 {
201 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
202 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
203 #ifdef INVARIANTS
204 vm_map_zdtor,
205 #else
206 NULL,
207 #endif
208 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
209 uma_prealloc(mapzone, MAX_KMAP);
210 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
211 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
212 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
213 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
214 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
215 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
216 #ifdef INVARIANTS
217 vmspace_zdtor,
218 #else
219 NULL,
220 #endif
221 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
222 }
223
224 static int
225 vmspace_zinit(void *mem, int size, int flags)
226 {
227 struct vmspace *vm;
228
229 vm = (struct vmspace *)mem;
230
231 vm->vm_map.pmap = NULL;
232 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
233 PMAP_LOCK_INIT(vmspace_pmap(vm));
234 return (0);
235 }
236
237 static int
238 vm_map_zinit(void *mem, int size, int flags)
239 {
240 vm_map_t map;
241
242 map = (vm_map_t)mem;
243 memset(map, 0, sizeof(*map));
244 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
245 sx_init(&map->lock, "vm map (user)");
246 return (0);
247 }
248
249 #ifdef INVARIANTS
250 static void
251 vmspace_zdtor(void *mem, int size, void *arg)
252 {
253 struct vmspace *vm;
254
255 vm = (struct vmspace *)mem;
256
257 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
258 }
259 static void
260 vm_map_zdtor(void *mem, int size, void *arg)
261 {
262 vm_map_t map;
263
264 map = (vm_map_t)mem;
265 KASSERT(map->nentries == 0,
266 ("map %p nentries == %d on free.",
267 map, map->nentries));
268 KASSERT(map->size == 0,
269 ("map %p size == %lu on free.",
270 map, (unsigned long)map->size));
271 }
272 #endif /* INVARIANTS */
273
274 /*
275 * Allocate a vmspace structure, including a vm_map and pmap,
276 * and initialize those structures. The refcnt is set to 1.
277 *
278 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
279 */
280 struct vmspace *
281 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
282 {
283 struct vmspace *vm;
284
285 vm = uma_zalloc(vmspace_zone, M_WAITOK);
286
287 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
288
289 if (pinit == NULL)
290 pinit = &pmap_pinit;
291
292 if (!pinit(vmspace_pmap(vm))) {
293 uma_zfree(vmspace_zone, vm);
294 return (NULL);
295 }
296 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
297 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
298 vm->vm_refcnt = 1;
299 vm->vm_shm = NULL;
300 vm->vm_swrss = 0;
301 vm->vm_tsize = 0;
302 vm->vm_dsize = 0;
303 vm->vm_ssize = 0;
304 vm->vm_taddr = 0;
305 vm->vm_daddr = 0;
306 vm->vm_maxsaddr = 0;
307 return (vm);
308 }
309
310 #ifdef RACCT
311 static void
312 vmspace_container_reset(struct proc *p)
313 {
314
315 PROC_LOCK(p);
316 racct_set(p, RACCT_DATA, 0);
317 racct_set(p, RACCT_STACK, 0);
318 racct_set(p, RACCT_RSS, 0);
319 racct_set(p, RACCT_MEMLOCK, 0);
320 racct_set(p, RACCT_VMEM, 0);
321 PROC_UNLOCK(p);
322 }
323 #endif
324
325 static inline void
326 vmspace_dofree(struct vmspace *vm)
327 {
328
329 CTR1(KTR_VM, "vmspace_free: %p", vm);
330
331 /*
332 * Make sure any SysV shm is freed, it might not have been in
333 * exit1().
334 */
335 shmexit(vm);
336
337 /*
338 * Lock the map, to wait out all other references to it.
339 * Delete all of the mappings and pages they hold, then call
340 * the pmap module to reclaim anything left.
341 */
342 (void)vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
343 vm_map_max(&vm->vm_map));
344
345 pmap_release(vmspace_pmap(vm));
346 vm->vm_map.pmap = NULL;
347 uma_zfree(vmspace_zone, vm);
348 }
349
350 void
351 vmspace_free(struct vmspace *vm)
352 {
353
354 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
355 "vmspace_free() called");
356
357 if (vm->vm_refcnt == 0)
358 panic("vmspace_free: attempt to free already freed vmspace");
359
360 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
361 vmspace_dofree(vm);
362 }
363
364 void
365 vmspace_exitfree(struct proc *p)
366 {
367 struct vmspace *vm;
368
369 PROC_VMSPACE_LOCK(p);
370 vm = p->p_vmspace;
371 p->p_vmspace = NULL;
372 PROC_VMSPACE_UNLOCK(p);
373 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
374 vmspace_free(vm);
375 }
376
377 void
378 vmspace_exit(struct thread *td)
379 {
380 int refcnt;
381 struct vmspace *vm;
382 struct proc *p;
383
384 /*
385 * Release user portion of address space.
386 * This releases references to vnodes,
387 * which could cause I/O if the file has been unlinked.
388 * Need to do this early enough that we can still sleep.
389 *
390 * The last exiting process to reach this point releases as
391 * much of the environment as it can. vmspace_dofree() is the
392 * slower fallback in case another process had a temporary
393 * reference to the vmspace.
394 */
395
396 p = td->td_proc;
397 vm = p->p_vmspace;
398 atomic_add_int(&vmspace0.vm_refcnt, 1);
399 do {
400 refcnt = vm->vm_refcnt;
401 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
402 /* Switch now since other proc might free vmspace */
403 PROC_VMSPACE_LOCK(p);
404 p->p_vmspace = &vmspace0;
405 PROC_VMSPACE_UNLOCK(p);
406 pmap_activate(td);
407 }
408 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
409 if (refcnt == 1) {
410 if (p->p_vmspace != vm) {
411 /* vmspace not yet freed, switch back */
412 PROC_VMSPACE_LOCK(p);
413 p->p_vmspace = vm;
414 PROC_VMSPACE_UNLOCK(p);
415 pmap_activate(td);
416 }
417 pmap_remove_pages(vmspace_pmap(vm));
418 /* Switch now since this proc will free vmspace */
419 PROC_VMSPACE_LOCK(p);
420 p->p_vmspace = &vmspace0;
421 PROC_VMSPACE_UNLOCK(p);
422 pmap_activate(td);
423 vmspace_dofree(vm);
424 }
425 #ifdef RACCT
426 if (racct_enable)
427 vmspace_container_reset(p);
428 #endif
429 }
430
431 /* Acquire reference to vmspace owned by another process. */
432
433 struct vmspace *
434 vmspace_acquire_ref(struct proc *p)
435 {
436 struct vmspace *vm;
437 int refcnt;
438
439 PROC_VMSPACE_LOCK(p);
440 vm = p->p_vmspace;
441 if (vm == NULL) {
442 PROC_VMSPACE_UNLOCK(p);
443 return (NULL);
444 }
445 do {
446 refcnt = vm->vm_refcnt;
447 if (refcnt <= 0) { /* Avoid 0->1 transition */
448 PROC_VMSPACE_UNLOCK(p);
449 return (NULL);
450 }
451 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
452 if (vm != p->p_vmspace) {
453 PROC_VMSPACE_UNLOCK(p);
454 vmspace_free(vm);
455 return (NULL);
456 }
457 PROC_VMSPACE_UNLOCK(p);
458 return (vm);
459 }
460
461 /*
462 * Switch between vmspaces in an AIO kernel process.
463 *
464 * The AIO kernel processes switch to and from a user process's
465 * vmspace while performing an I/O operation on behalf of a user
466 * process. The new vmspace is either the vmspace of a user process
467 * obtained from an active AIO request or the initial vmspace of the
468 * AIO kernel process (when it is idling). Because user processes
469 * will block to drain any active AIO requests before proceeding in
470 * exit() or execve(), the vmspace reference count for these vmspaces
471 * can never be 0. This allows for a much simpler implementation than
472 * the loop in vmspace_acquire_ref() above. Similarly, AIO kernel
473 * processes hold an extra reference on their initial vmspace for the
474 * life of the process so that this guarantee is true for any vmspace
475 * passed as 'newvm'.
476 */
477 void
478 vmspace_switch_aio(struct vmspace *newvm)
479 {
480 struct vmspace *oldvm;
481
482 /* XXX: Need some way to assert that this is an aio daemon. */
483
484 KASSERT(newvm->vm_refcnt > 0,
485 ("vmspace_switch_aio: newvm unreferenced"));
486
487 oldvm = curproc->p_vmspace;
488 if (oldvm == newvm)
489 return;
490
491 /*
492 * Point to the new address space and refer to it.
493 */
494 curproc->p_vmspace = newvm;
495 atomic_add_int(&newvm->vm_refcnt, 1);
496
497 /* Activate the new mapping. */
498 pmap_activate(curthread);
499
500 /* Remove the daemon's reference to the old address space. */
501 KASSERT(oldvm->vm_refcnt > 1,
502 ("vmspace_switch_aio: oldvm dropping last reference"));
503 vmspace_free(oldvm);
504 }
505
506 void
507 _vm_map_lock(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_xlock_(&map->lock, file, line);
514 map->timestamp++;
515 }
516
517 static void
518 vm_map_process_deferred(void)
519 {
520 struct thread *td;
521 vm_map_entry_t entry, next;
522 vm_object_t object;
523
524 td = curthread;
525 entry = td->td_map_def_user;
526 td->td_map_def_user = NULL;
527 while (entry != NULL) {
528 next = entry->next;
529 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
530 /*
531 * Decrement the object's writemappings and
532 * possibly the vnode's v_writecount.
533 */
534 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
535 ("Submap with writecount"));
536 object = entry->object.vm_object;
537 KASSERT(object != NULL, ("No object for writecount"));
538 vnode_pager_release_writecount(object, entry->start,
539 entry->end);
540 }
541 vm_map_entry_deallocate(entry, FALSE);
542 entry = next;
543 }
544 }
545
546 void
547 _vm_map_unlock(vm_map_t map, const char *file, int line)
548 {
549
550 if (map->system_map)
551 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
552 else {
553 sx_xunlock_(&map->lock, file, line);
554 vm_map_process_deferred();
555 }
556 }
557
558 void
559 _vm_map_lock_read(vm_map_t map, const char *file, int line)
560 {
561
562 if (map->system_map)
563 mtx_lock_flags_(&map->system_mtx, 0, file, line);
564 else
565 sx_slock_(&map->lock, file, line);
566 }
567
568 void
569 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
570 {
571
572 if (map->system_map)
573 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
574 else {
575 sx_sunlock_(&map->lock, file, line);
576 vm_map_process_deferred();
577 }
578 }
579
580 int
581 _vm_map_trylock(vm_map_t map, const char *file, int line)
582 {
583 int error;
584
585 error = map->system_map ?
586 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
587 !sx_try_xlock_(&map->lock, file, line);
588 if (error == 0)
589 map->timestamp++;
590 return (error == 0);
591 }
592
593 int
594 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
595 {
596 int error;
597
598 error = map->system_map ?
599 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
600 !sx_try_slock_(&map->lock, file, line);
601 return (error == 0);
602 }
603
604 /*
605 * _vm_map_lock_upgrade: [ internal use only ]
606 *
607 * Tries to upgrade a read (shared) lock on the specified map to a write
608 * (exclusive) lock. Returns the value "" if the upgrade succeeds and a
609 * non-zero value if the upgrade fails. If the upgrade fails, the map is
610 * returned without a read or write lock held.
611 *
612 * Requires that the map be read locked.
613 */
614 int
615 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
616 {
617 unsigned int last_timestamp;
618
619 if (map->system_map) {
620 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
621 } else {
622 if (!sx_try_upgrade_(&map->lock, file, line)) {
623 last_timestamp = map->timestamp;
624 sx_sunlock_(&map->lock, file, line);
625 vm_map_process_deferred();
626 /*
627 * If the map's timestamp does not change while the
628 * map is unlocked, then the upgrade succeeds.
629 */
630 sx_xlock_(&map->lock, file, line);
631 if (last_timestamp != map->timestamp) {
632 sx_xunlock_(&map->lock, file, line);
633 return (1);
634 }
635 }
636 }
637 map->timestamp++;
638 return (0);
639 }
640
641 void
642 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
643 {
644
645 if (map->system_map) {
646 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
647 } else
648 sx_downgrade_(&map->lock, file, line);
649 }
650
651 /*
652 * vm_map_locked:
653 *
654 * Returns a non-zero value if the caller holds a write (exclusive) lock
655 * on the specified map and the value "" otherwise.
656 */
657 int
658 vm_map_locked(vm_map_t map)
659 {
660
661 if (map->system_map)
662 return (mtx_owned(&map->system_mtx));
663 else
664 return (sx_xlocked(&map->lock));
665 }
666
667 #ifdef INVARIANTS
668 static void
669 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
670 {
671
672 if (map->system_map)
673 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
674 else
675 sx_assert_(&map->lock, SA_XLOCKED, file, line);
676 }
677
678 #define VM_MAP_ASSERT_LOCKED(map) \
679 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
680 #else
681 #define VM_MAP_ASSERT_LOCKED(map)
682 #endif
683
684 /*
685 * _vm_map_unlock_and_wait:
686 *
687 * Atomically releases the lock on the specified map and puts the calling
688 * thread to sleep. The calling thread will remain asleep until either
689 * vm_map_wakeup() is performed on the map or the specified timeout is
690 * exceeded.
691 *
692 * WARNING! This function does not perform deferred deallocations of
693 * objects and map entries. Therefore, the calling thread is expected to
694 * reacquire the map lock after reawakening and later perform an ordinary
695 * unlock operation, such as vm_map_unlock(), before completing its
696 * operation on the map.
697 */
698 int
699 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
700 {
701
702 mtx_lock(&map_sleep_mtx);
703 if (map->system_map)
704 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
705 else
706 sx_xunlock_(&map->lock, file, line);
707 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
708 timo));
709 }
710
711 /*
712 * vm_map_wakeup:
713 *
714 * Awaken any threads that have slept on the map using
715 * vm_map_unlock_and_wait().
716 */
717 void
718 vm_map_wakeup(vm_map_t map)
719 {
720
721 /*
722 * Acquire and release map_sleep_mtx to prevent a wakeup()
723 * from being performed (and lost) between the map unlock
724 * and the msleep() in _vm_map_unlock_and_wait().
725 */
726 mtx_lock(&map_sleep_mtx);
727 mtx_unlock(&map_sleep_mtx);
728 wakeup(&map->root);
729 }
730
731 void
732 vm_map_busy(vm_map_t map)
733 {
734
735 VM_MAP_ASSERT_LOCKED(map);
736 map->busy++;
737 }
738
739 void
740 vm_map_unbusy(vm_map_t map)
741 {
742
743 VM_MAP_ASSERT_LOCKED(map);
744 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
745 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
746 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
747 wakeup(&map->busy);
748 }
749 }
750
751 void
752 vm_map_wait_busy(vm_map_t map)
753 {
754
755 VM_MAP_ASSERT_LOCKED(map);
756 while (map->busy) {
757 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
758 if (map->system_map)
759 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
760 else
761 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
762 }
763 map->timestamp++;
764 }
765
766 long
767 vmspace_resident_count(struct vmspace *vmspace)
768 {
769 return pmap_resident_count(vmspace_pmap(vmspace));
770 }
771
772 /*
773 * vm_map_create:
774 *
775 * Creates and returns a new empty VM map with
776 * the given physical map structure, and having
777 * the given lower and upper address bounds.
778 */
779 vm_map_t
780 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
781 {
782 vm_map_t result;
783
784 result = uma_zalloc(mapzone, M_WAITOK);
785 CTR1(KTR_VM, "vm_map_create: %p", result);
786 _vm_map_init(result, pmap, min, max);
787 return (result);
788 }
789
790 /*
791 * Initialize an existing vm_map structure
792 * such as that in the vmspace structure.
793 */
794 static void
795 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
796 {
797
798 map->header.next = map->header.prev = &map->header;
799 map->header.eflags = MAP_ENTRY_HEADER;
800 map->needs_wakeup = FALSE;
801 map->system_map = 0;
802 map->pmap = pmap;
803 map->header.end = min;
804 map->header.start = max;
805 map->flags = 0;
806 map->root = NULL;
807 map->timestamp = 0;
808 map->busy = 0;
809 }
810
811 void
812 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
813 {
814
815 _vm_map_init(map, pmap, min, max);
816 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
817 sx_init(&map->lock, "user map");
818 }
819
820 /*
821 * vm_map_entry_dispose: [ internal use only ]
822 *
823 * Inverse of vm_map_entry_create.
824 */
825 static void
826 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
827 {
828 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
829 }
830
831 /*
832 * vm_map_entry_create: [ internal use only ]
833 *
834 * Allocates a VM map entry for insertion.
835 * No entry fields are filled in.
836 */
837 static vm_map_entry_t
838 vm_map_entry_create(vm_map_t map)
839 {
840 vm_map_entry_t new_entry;
841
842 if (map->system_map)
843 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
844 else
845 new_entry = uma_zalloc(mapentzone, M_WAITOK);
846 if (new_entry == NULL)
847 panic("vm_map_entry_create: kernel resources exhausted");
848 return (new_entry);
849 }
850
851 /*
852 * vm_map_entry_set_behavior:
853 *
854 * Set the expected access behavior, either normal, random, or
855 * sequential.
856 */
857 static inline void
858 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
859 {
860 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
861 (behavior & MAP_ENTRY_BEHAV_MASK);
862 }
863
864 /*
865 * vm_map_entry_set_max_free:
866 *
867 * Set the max_free field in a vm_map_entry.
868 */
869 static inline void
870 vm_map_entry_set_max_free(vm_map_entry_t entry)
871 {
872
873 entry->max_free = entry->adj_free;
874 if (entry->left != NULL && entry->left->max_free > entry->max_free)
875 entry->max_free = entry->left->max_free;
876 if (entry->right != NULL && entry->right->max_free > entry->max_free)
877 entry->max_free = entry->right->max_free;
878 }
879
880 /*
881 * vm_map_entry_splay:
882 *
883 * The Sleator and Tarjan top-down splay algorithm with the
884 * following variation. Max_free must be computed bottom-up, so
885 * on the downward pass, maintain the left and right spines in
886 * reverse order. Then, make a second pass up each side to fix
887 * the pointers and compute max_free. The time bound is O(log n)
888 * amortized.
889 *
890 * The new root is the vm_map_entry containing "addr", or else an
891 * adjacent entry (lower or higher) if addr is not in the tree.
892 *
893 * The map must be locked, and leaves it so.
894 *
895 * Returns: the new root.
896 */
897 static vm_map_entry_t
898 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
899 {
900 vm_map_entry_t llist, rlist;
901 vm_map_entry_t ltree, rtree;
902 vm_map_entry_t y;
903
904 /* Special case of empty tree. */
905 if (root == NULL)
906 return (root);
907
908 /*
909 * Pass One: Splay down the tree until we find addr or a NULL
910 * pointer where addr would go. llist and rlist are the two
911 * sides in reverse order (bottom-up), with llist linked by
912 * the right pointer and rlist linked by the left pointer in
913 * the vm_map_entry. Wait until Pass Two to set max_free on
914 * the two spines.
915 */
916 llist = NULL;
917 rlist = NULL;
918 for (;;) {
919 /* root is never NULL in here. */
920 if (addr < root->start) {
921 y = root->left;
922 if (y == NULL)
923 break;
924 if (addr < y->start && y->left != NULL) {
925 /* Rotate right and put y on rlist. */
926 root->left = y->right;
927 y->right = root;
928 vm_map_entry_set_max_free(root);
929 root = y->left;
930 y->left = rlist;
931 rlist = y;
932 } else {
933 /* Put root on rlist. */
934 root->left = rlist;
935 rlist = root;
936 root = y;
937 }
938 } else if (addr >= root->end) {
939 y = root->right;
940 if (y == NULL)
941 break;
942 if (addr >= y->end && y->right != NULL) {
943 /* Rotate left and put y on llist. */
944 root->right = y->left;
945 y->left = root;
946 vm_map_entry_set_max_free(root);
947 root = y->right;
948 y->right = llist;
949 llist = y;
950 } else {
951 /* Put root on llist. */
952 root->right = llist;
953 llist = root;
954 root = y;
955 }
956 } else
957 break;
958 }
959
960 /*
961 * Pass Two: Walk back up the two spines, flip the pointers
962 * and set max_free. The subtrees of the root go at the
963 * bottom of llist and rlist.
964 */
965 ltree = root->left;
966 while (llist != NULL) {
967 y = llist->right;
968 llist->right = ltree;
969 vm_map_entry_set_max_free(llist);
970 ltree = llist;
971 llist = y;
972 }
973 rtree = root->right;
974 while (rlist != NULL) {
975 y = rlist->left;
976 rlist->left = rtree;
977 vm_map_entry_set_max_free(rlist);
978 rtree = rlist;
979 rlist = y;
980 }
981
982 /*
983 * Final assembly: add ltree and rtree as subtrees of root.
984 */
985 root->left = ltree;
986 root->right = rtree;
987 vm_map_entry_set_max_free(root);
988
989 return (root);
990 }
991
992 /*
993 * vm_map_entry_{un,}link:
994 *
995 * Insert/remove entries from maps.
996 */
997 static void
998 vm_map_entry_link(vm_map_t map,
999 vm_map_entry_t after_where,
1000 vm_map_entry_t entry)
1001 {
1002
1003 CTR4(KTR_VM,
1004 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
1005 map->nentries, entry, after_where);
1006 VM_MAP_ASSERT_LOCKED(map);
1007 KASSERT(after_where->end <= entry->start,
1008 ("vm_map_entry_link: prev end %jx new start %jx overlap",
1009 (uintmax_t)after_where->end, (uintmax_t)entry->start));
1010 KASSERT(entry->end <= after_where->next->start,
1011 ("vm_map_entry_link: new end %jx next start %jx overlap",
1012 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
1013
1014 map->nentries++;
1015 entry->prev = after_where;
1016 entry->next = after_where->next;
1017 entry->next->prev = entry;
1018 after_where->next = entry;
1019
1020 if (after_where != &map->header) {
1021 if (after_where != map->root)
1022 vm_map_entry_splay(after_where->start, map->root);
1023 entry->right = after_where->right;
1024 entry->left = after_where;
1025 after_where->right = NULL;
1026 after_where->adj_free = entry->start - after_where->end;
1027 vm_map_entry_set_max_free(after_where);
1028 } else {
1029 entry->right = map->root;
1030 entry->left = NULL;
1031 }
1032 entry->adj_free = entry->next->start - entry->end;
1033 vm_map_entry_set_max_free(entry);
1034 map->root = entry;
1035 }
1036
1037 static void
1038 vm_map_entry_unlink(vm_map_t map,
1039 vm_map_entry_t entry)
1040 {
1041 vm_map_entry_t next, prev, root;
1042
1043 VM_MAP_ASSERT_LOCKED(map);
1044 if (entry != map->root)
1045 vm_map_entry_splay(entry->start, map->root);
1046 if (entry->left == NULL)
1047 root = entry->right;
1048 else {
1049 root = vm_map_entry_splay(entry->start, entry->left);
1050 root->right = entry->right;
1051 root->adj_free = entry->next->start - root->end;
1052 vm_map_entry_set_max_free(root);
1053 }
1054 map->root = root;
1055
1056 prev = entry->prev;
1057 next = entry->next;
1058 next->prev = prev;
1059 prev->next = next;
1060 map->nentries--;
1061 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1062 map->nentries, entry);
1063 }
1064
1065 /*
1066 * vm_map_entry_resize_free:
1067 *
1068 * Recompute the amount of free space following a vm_map_entry
1069 * and propagate that value up the tree. Call this function after
1070 * resizing a map entry in-place, that is, without a call to
1071 * vm_map_entry_link() or _unlink().
1072 *
1073 * The map must be locked, and leaves it so.
1074 */
1075 static void
1076 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1077 {
1078
1079 /*
1080 * Using splay trees without parent pointers, propagating
1081 * max_free up the tree is done by moving the entry to the
1082 * root and making the change there.
1083 */
1084 if (entry != map->root)
1085 map->root = vm_map_entry_splay(entry->start, map->root);
1086
1087 entry->adj_free = entry->next->start - entry->end;
1088 vm_map_entry_set_max_free(entry);
1089 }
1090
1091 /*
1092 * vm_map_lookup_entry: [ internal use only ]
1093 *
1094 * Finds the map entry containing (or
1095 * immediately preceding) the specified address
1096 * in the given map; the entry is returned
1097 * in the "entry" parameter. The boolean
1098 * result indicates whether the address is
1099 * actually contained in the map.
1100 */
1101 boolean_t
1102 vm_map_lookup_entry(
1103 vm_map_t map,
1104 vm_offset_t address,
1105 vm_map_entry_t *entry) /* OUT */
1106 {
1107 vm_map_entry_t cur;
1108 boolean_t locked;
1109
1110 /*
1111 * If the map is empty, then the map entry immediately preceding
1112 * "address" is the map's header.
1113 */
1114 cur = map->root;
1115 if (cur == NULL)
1116 *entry = &map->header;
1117 else if (address >= cur->start && cur->end > address) {
1118 *entry = cur;
1119 return (TRUE);
1120 } else if ((locked = vm_map_locked(map)) ||
1121 sx_try_upgrade(&map->lock)) {
1122 /*
1123 * Splay requires a write lock on the map. However, it only
1124 * restructures the binary search tree; it does not otherwise
1125 * change the map. Thus, the map's timestamp need not change
1126 * on a temporary upgrade.
1127 */
1128 map->root = cur = vm_map_entry_splay(address, cur);
1129 if (!locked)
1130 sx_downgrade(&map->lock);
1131
1132 /*
1133 * If "address" is contained within a map entry, the new root
1134 * is that map entry. Otherwise, the new root is a map entry
1135 * immediately before or after "address".
1136 */
1137 if (address >= cur->start) {
1138 *entry = cur;
1139 if (cur->end > address)
1140 return (TRUE);
1141 } else
1142 *entry = cur->prev;
1143 } else
1144 /*
1145 * Since the map is only locked for read access, perform a
1146 * standard binary search tree lookup for "address".
1147 */
1148 for (;;) {
1149 if (address < cur->start) {
1150 if (cur->left == NULL) {
1151 *entry = cur->prev;
1152 break;
1153 }
1154 cur = cur->left;
1155 } else if (cur->end > address) {
1156 *entry = cur;
1157 return (TRUE);
1158 } else {
1159 if (cur->right == NULL) {
1160 *entry = cur;
1161 break;
1162 }
1163 cur = cur->right;
1164 }
1165 }
1166 return (FALSE);
1167 }
1168
1169 /*
1170 * vm_map_insert:
1171 *
1172 * Inserts the given whole VM object into the target
1173 * map at the specified address range. The object's
1174 * size should match that of the address range.
1175 *
1176 * Requires that the map be locked, and leaves it so.
1177 *
1178 * If object is non-NULL, ref count must be bumped by caller
1179 * prior to making call to account for the new entry.
1180 */
1181 int
1182 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1183 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1184 {
1185 vm_map_entry_t new_entry, prev_entry, temp_entry;
1186 struct ucred *cred;
1187 vm_eflags_t protoeflags;
1188 vm_inherit_t inheritance;
1189
1190 VM_MAP_ASSERT_LOCKED(map);
1191 KASSERT(object != kernel_object ||
1192 (cow & MAP_COPY_ON_WRITE) == 0,
1193 ("vm_map_insert: kernel object and COW"));
1194 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1195 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1196 KASSERT((prot & ~max) == 0,
1197 ("prot %#x is not subset of max_prot %#x", prot, max));
1198
1199 /*
1200 * Check that the start and end points are not bogus.
1201 */
1202 if (start < vm_map_min(map) || end > vm_map_max(map) ||
1203 start >= end)
1204 return (KERN_INVALID_ADDRESS);
1205
1206 /*
1207 * Find the entry prior to the proposed starting address; if it's part
1208 * of an existing entry, this range is bogus.
1209 */
1210 if (vm_map_lookup_entry(map, start, &temp_entry))
1211 return (KERN_NO_SPACE);
1212
1213 prev_entry = temp_entry;
1214
1215 /*
1216 * Assert that the next entry doesn't overlap the end point.
1217 */
1218 if (prev_entry->next->start < end)
1219 return (KERN_NO_SPACE);
1220
1221 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1222 max != VM_PROT_NONE))
1223 return (KERN_INVALID_ARGUMENT);
1224
1225 protoeflags = 0;
1226 if (cow & MAP_COPY_ON_WRITE)
1227 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1228 if (cow & MAP_NOFAULT)
1229 protoeflags |= MAP_ENTRY_NOFAULT;
1230 if (cow & MAP_DISABLE_SYNCER)
1231 protoeflags |= MAP_ENTRY_NOSYNC;
1232 if (cow & MAP_DISABLE_COREDUMP)
1233 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1234 if (cow & MAP_STACK_GROWS_DOWN)
1235 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1236 if (cow & MAP_STACK_GROWS_UP)
1237 protoeflags |= MAP_ENTRY_GROWS_UP;
1238 if (cow & MAP_VN_WRITECOUNT)
1239 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1240 if ((cow & MAP_CREATE_GUARD) != 0)
1241 protoeflags |= MAP_ENTRY_GUARD;
1242 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1243 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1244 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1245 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1246 if (cow & MAP_INHERIT_SHARE)
1247 inheritance = VM_INHERIT_SHARE;
1248 else
1249 inheritance = VM_INHERIT_DEFAULT;
1250
1251 cred = NULL;
1252 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1253 goto charged;
1254 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1255 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1256 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1257 return (KERN_RESOURCE_SHORTAGE);
1258 KASSERT(object == NULL ||
1259 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1260 object->cred == NULL,
1261 ("overcommit: vm_map_insert o %p", object));
1262 cred = curthread->td_ucred;
1263 }
1264
1265 charged:
1266 /* Expand the kernel pmap, if necessary. */
1267 if (map == kernel_map && end > kernel_vm_end)
1268 pmap_growkernel(end);
1269 if (object != NULL) {
1270 /*
1271 * OBJ_ONEMAPPING must be cleared unless this mapping
1272 * is trivially proven to be the only mapping for any
1273 * of the object's pages. (Object granularity
1274 * reference counting is insufficient to recognize
1275 * aliases with precision.)
1276 */
1277 VM_OBJECT_WLOCK(object);
1278 if (object->ref_count > 1 || object->shadow_count != 0)
1279 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1280 VM_OBJECT_WUNLOCK(object);
1281 } else if ((prev_entry->eflags & ~MAP_ENTRY_USER_WIRED) ==
1282 protoeflags &&
1283 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1284 prev_entry->end == start && (prev_entry->cred == cred ||
1285 (prev_entry->object.vm_object != NULL &&
1286 prev_entry->object.vm_object->cred == cred)) &&
1287 vm_object_coalesce(prev_entry->object.vm_object,
1288 prev_entry->offset,
1289 (vm_size_t)(prev_entry->end - prev_entry->start),
1290 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1291 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1292 /*
1293 * We were able to extend the object. Determine if we
1294 * can extend the previous map entry to include the
1295 * new range as well.
1296 */
1297 if (prev_entry->inheritance == inheritance &&
1298 prev_entry->protection == prot &&
1299 prev_entry->max_protection == max &&
1300 prev_entry->wired_count == 0) {
1301 KASSERT((prev_entry->eflags & MAP_ENTRY_USER_WIRED) ==
1302 0, ("prev_entry %p has incoherent wiring",
1303 prev_entry));
1304 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1305 map->size += end - prev_entry->end;
1306 prev_entry->end = end;
1307 vm_map_entry_resize_free(map, prev_entry);
1308 vm_map_simplify_entry(map, prev_entry);
1309 return (KERN_SUCCESS);
1310 }
1311
1312 /*
1313 * If we can extend the object but cannot extend the
1314 * map entry, we have to create a new map entry. We
1315 * must bump the ref count on the extended object to
1316 * account for it. object may be NULL.
1317 */
1318 object = prev_entry->object.vm_object;
1319 offset = prev_entry->offset +
1320 (prev_entry->end - prev_entry->start);
1321 vm_object_reference(object);
1322 if (cred != NULL && object != NULL && object->cred != NULL &&
1323 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1324 /* Object already accounts for this uid. */
1325 cred = NULL;
1326 }
1327 }
1328 if (cred != NULL)
1329 crhold(cred);
1330
1331 /*
1332 * Create a new entry
1333 */
1334 new_entry = vm_map_entry_create(map);
1335 new_entry->start = start;
1336 new_entry->end = end;
1337 new_entry->cred = NULL;
1338
1339 new_entry->eflags = protoeflags;
1340 new_entry->object.vm_object = object;
1341 new_entry->offset = offset;
1342
1343 new_entry->inheritance = inheritance;
1344 new_entry->protection = prot;
1345 new_entry->max_protection = max;
1346 new_entry->wired_count = 0;
1347 new_entry->wiring_thread = NULL;
1348 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1349 new_entry->next_read = start;
1350
1351 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1352 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1353 new_entry->cred = cred;
1354
1355 /*
1356 * Insert the new entry into the list
1357 */
1358 vm_map_entry_link(map, prev_entry, new_entry);
1359 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1360 map->size += new_entry->end - new_entry->start;
1361
1362 /*
1363 * Try to coalesce the new entry with both the previous and next
1364 * entries in the list. Previously, we only attempted to coalesce
1365 * with the previous entry when object is NULL. Here, we handle the
1366 * other cases, which are less common.
1367 */
1368 vm_map_simplify_entry(map, new_entry);
1369
1370 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1371 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1372 end - start, cow & MAP_PREFAULT_PARTIAL);
1373 }
1374
1375 return (KERN_SUCCESS);
1376 }
1377
1378 /*
1379 * vm_map_findspace:
1380 *
1381 * Find the first fit (lowest VM address) for "length" free bytes
1382 * beginning at address >= start in the given map.
1383 *
1384 * In a vm_map_entry, "adj_free" is the amount of free space
1385 * adjacent (higher address) to this entry, and "max_free" is the
1386 * maximum amount of contiguous free space in its subtree. This
1387 * allows finding a free region in one path down the tree, so
1388 * O(log n) amortized with splay trees.
1389 *
1390 * The map must be locked, and leaves it so.
1391 *
1392 * Returns: 0 on success, and starting address in *addr,
1393 * 1 if insufficient space.
1394 */
1395 int
1396 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1397 vm_offset_t *addr) /* OUT */
1398 {
1399 vm_map_entry_t entry;
1400 vm_offset_t st;
1401
1402 /*
1403 * Request must fit within min/max VM address and must avoid
1404 * address wrap.
1405 */
1406 start = MAX(start, vm_map_min(map));
1407 if (start + length > vm_map_max(map) || start + length < start)
1408 return (1);
1409
1410 /* Empty tree means wide open address space. */
1411 if (map->root == NULL) {
1412 *addr = start;
1413 return (0);
1414 }
1415
1416 /*
1417 * After splay, if start comes before root node, then there
1418 * must be a gap from start to the root.
1419 */
1420 map->root = vm_map_entry_splay(start, map->root);
1421 if (start + length <= map->root->start) {
1422 *addr = start;
1423 return (0);
1424 }
1425
1426 /*
1427 * Root is the last node that might begin its gap before
1428 * start, and this is the last comparison where address
1429 * wrap might be a problem.
1430 */
1431 st = (start > map->root->end) ? start : map->root->end;
1432 if (length <= map->root->end + map->root->adj_free - st) {
1433 *addr = st;
1434 return (0);
1435 }
1436
1437 /* With max_free, can immediately tell if no solution. */
1438 entry = map->root->right;
1439 if (entry == NULL || length > entry->max_free)
1440 return (1);
1441
1442 /*
1443 * Search the right subtree in the order: left subtree, root,
1444 * right subtree (first fit). The previous splay implies that
1445 * all regions in the right subtree have addresses > start.
1446 */
1447 while (entry != NULL) {
1448 if (entry->left != NULL && entry->left->max_free >= length)
1449 entry = entry->left;
1450 else if (entry->adj_free >= length) {
1451 *addr = entry->end;
1452 return (0);
1453 } else
1454 entry = entry->right;
1455 }
1456
1457 /* Can't get here, so panic if we do. */
1458 panic("vm_map_findspace: max_free corrupt");
1459 }
1460
1461 int
1462 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1463 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1464 vm_prot_t max, int cow)
1465 {
1466 vm_offset_t end;
1467 int result;
1468
1469 end = start + length;
1470 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1471 object == NULL,
1472 ("vm_map_fixed: non-NULL backing object for stack"));
1473 vm_map_lock(map);
1474 VM_MAP_RANGE_CHECK(map, start, end);
1475 if ((cow & MAP_CHECK_EXCL) == 0)
1476 vm_map_delete(map, start, end);
1477 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1478 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1479 prot, max, cow);
1480 } else {
1481 result = vm_map_insert(map, object, offset, start, end,
1482 prot, max, cow);
1483 }
1484 vm_map_unlock(map);
1485 return (result);
1486 }
1487
1488 /*
1489 * Searches for the specified amount of free space in the given map with the
1490 * specified alignment. Performs an address-ordered, first-fit search from
1491 * the given address "*addr", with an optional upper bound "max_addr". If the
1492 * parameter "alignment" is zero, then the alignment is computed from the
1493 * given (object, offset) pair so as to enable the greatest possible use of
1494 * superpage mappings. Returns KERN_SUCCESS and the address of the free space
1495 * in "*addr" if successful. Otherwise, returns KERN_NO_SPACE.
1496 *
1497 * The map must be locked. Initially, there must be at least "length" bytes
1498 * of free space at the given address.
1499 */
1500 static int
1501 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1502 vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr,
1503 vm_offset_t alignment)
1504 {
1505 vm_offset_t aligned_addr, free_addr;
1506
1507 VM_MAP_ASSERT_LOCKED(map);
1508 free_addr = *addr;
1509 KASSERT(!vm_map_findspace(map, free_addr, length, addr) &&
1510 free_addr == *addr, ("caller provided insufficient free space"));
1511 for (;;) {
1512 /*
1513 * At the start of every iteration, the free space at address
1514 * "*addr" is at least "length" bytes.
1515 */
1516 if (alignment == 0)
1517 pmap_align_superpage(object, offset, addr, length);
1518 else if ((*addr & (alignment - 1)) != 0) {
1519 *addr &= ~(alignment - 1);
1520 *addr += alignment;
1521 }
1522 aligned_addr = *addr;
1523 if (aligned_addr == free_addr) {
1524 /*
1525 * Alignment did not change "*addr", so "*addr" must
1526 * still provide sufficient free space.
1527 */
1528 return (KERN_SUCCESS);
1529 }
1530
1531 /*
1532 * Test for address wrap on "*addr". A wrapped "*addr" could
1533 * be a valid address, in which case vm_map_findspace() cannot
1534 * be relied upon to fail.
1535 */
1536 if (aligned_addr < free_addr ||
1537 vm_map_findspace(map, aligned_addr, length, addr) ||
1538 (max_addr != 0 && *addr + length > max_addr))
1539 return (KERN_NO_SPACE);
1540 free_addr = *addr;
1541 if (free_addr == aligned_addr) {
1542 /*
1543 * If a successful call to vm_map_findspace() did not
1544 * change "*addr", then "*addr" must still be aligned
1545 * and provide sufficient free space.
1546 */
1547 return (KERN_SUCCESS);
1548 }
1549 }
1550 }
1551
1552 /*
1553 * vm_map_find finds an unallocated region in the target address
1554 * map with the given length. The search is defined to be
1555 * first-fit from the specified address; the region found is
1556 * returned in the same parameter.
1557 *
1558 * If object is non-NULL, ref count must be bumped by caller
1559 * prior to making call to account for the new entry.
1560 */
1561 int
1562 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1563 vm_offset_t *addr, /* IN/OUT */
1564 vm_size_t length, vm_offset_t max_addr, int find_space,
1565 vm_prot_t prot, vm_prot_t max, int cow)
1566 {
1567 vm_offset_t alignment, min_addr;
1568 int rv;
1569
1570 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1571 object == NULL,
1572 ("vm_map_find: non-NULL backing object for stack"));
1573 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1574 (object->flags & OBJ_COLORED) == 0))
1575 find_space = VMFS_ANY_SPACE;
1576 if (find_space >> 8 != 0) {
1577 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1578 alignment = (vm_offset_t)1 << (find_space >> 8);
1579 } else
1580 alignment = 0;
1581 vm_map_lock(map);
1582 if (find_space != VMFS_NO_SPACE) {
1583 KASSERT(find_space == VMFS_ANY_SPACE ||
1584 find_space == VMFS_OPTIMAL_SPACE ||
1585 find_space == VMFS_SUPER_SPACE ||
1586 alignment != 0, ("unexpected VMFS flag"));
1587 min_addr = *addr;
1588 again:
1589 if (vm_map_findspace(map, min_addr, length, addr) ||
1590 (max_addr != 0 && *addr + length > max_addr)) {
1591 rv = KERN_NO_SPACE;
1592 goto done;
1593 }
1594 if (find_space != VMFS_ANY_SPACE &&
1595 (rv = vm_map_alignspace(map, object, offset, addr, length,
1596 max_addr, alignment)) != KERN_SUCCESS) {
1597 if (find_space == VMFS_OPTIMAL_SPACE) {
1598 find_space = VMFS_ANY_SPACE;
1599 goto again;
1600 }
1601 goto done;
1602 }
1603 }
1604 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1605 rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot,
1606 max, cow);
1607 } else {
1608 rv = vm_map_insert(map, object, offset, *addr, *addr + length,
1609 prot, max, cow);
1610 }
1611 done:
1612 vm_map_unlock(map);
1613 return (rv);
1614 }
1615
1616 /*
1617 * vm_map_find_min() is a variant of vm_map_find() that takes an
1618 * additional parameter (min_addr) and treats the given address
1619 * (*addr) differently. Specifically, it treats *addr as a hint
1620 * and not as the minimum address where the mapping is created.
1621 *
1622 * This function works in two phases. First, it tries to
1623 * allocate above the hint. If that fails and the hint is
1624 * greater than min_addr, it performs a second pass, replacing
1625 * the hint with min_addr as the minimum address for the
1626 * allocation.
1627 */
1628 int
1629 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1630 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
1631 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
1632 int cow)
1633 {
1634 vm_offset_t hint;
1635 int rv;
1636
1637 hint = *addr;
1638 for (;;) {
1639 rv = vm_map_find(map, object, offset, addr, length, max_addr,
1640 find_space, prot, max, cow);
1641 if (rv == KERN_SUCCESS || min_addr >= hint)
1642 return (rv);
1643 *addr = hint = min_addr;
1644 }
1645 }
1646
1647 static bool
1648 vm_map_mergeable_neighbors(vm_map_entry_t prev, vm_map_entry_t entry)
1649 {
1650 vm_size_t prevsize;
1651
1652 prevsize = prev->end - prev->start;
1653 return (prev->end == entry->start &&
1654 prev->object.vm_object == entry->object.vm_object &&
1655 (prev->object.vm_object == NULL ||
1656 prev->offset + prevsize == entry->offset) &&
1657 prev->eflags == entry->eflags &&
1658 prev->protection == entry->protection &&
1659 prev->max_protection == entry->max_protection &&
1660 prev->inheritance == entry->inheritance &&
1661 prev->wired_count == entry->wired_count &&
1662 prev->cred == entry->cred);
1663 }
1664
1665 static void
1666 vm_map_merged_neighbor_dispose(vm_map_t map, vm_map_entry_t entry)
1667 {
1668
1669 /*
1670 * If the backing object is a vnode object,
1671 * vm_object_deallocate() calls vrele().
1672 * However, vrele() does not lock the vnode
1673 * because the vnode has additional
1674 * references. Thus, the map lock can be kept
1675 * without causing a lock-order reversal with
1676 * the vnode lock.
1677 *
1678 * Since we count the number of virtual page
1679 * mappings in object->un_pager.vnp.writemappings,
1680 * the writemappings value should not be adjusted
1681 * when the entry is disposed of.
1682 */
1683 if (entry->object.vm_object != NULL)
1684 vm_object_deallocate(entry->object.vm_object);
1685 if (entry->cred != NULL)
1686 crfree(entry->cred);
1687 vm_map_entry_dispose(map, entry);
1688 }
1689
1690 /*
1691 * vm_map_simplify_entry:
1692 *
1693 * Simplify the given map entry by merging with either neighbor. This
1694 * routine also has the ability to merge with both neighbors.
1695 *
1696 * The map must be locked.
1697 *
1698 * This routine guarantees that the passed entry remains valid (though
1699 * possibly extended). When merging, this routine may delete one or
1700 * both neighbors.
1701 */
1702 void
1703 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1704 {
1705 vm_map_entry_t next, prev;
1706
1707 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP |
1708 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0)
1709 return;
1710
1711 prev = entry->prev;
1712 if (vm_map_mergeable_neighbors(prev, entry)) {
1713 vm_map_entry_unlink(map, prev);
1714 entry->start = prev->start;
1715 entry->offset = prev->offset;
1716 if (entry->prev != &map->header)
1717 vm_map_entry_resize_free(map, entry->prev);
1718 vm_map_merged_neighbor_dispose(map, prev);
1719 }
1720
1721 next = entry->next;
1722 if (vm_map_mergeable_neighbors(entry, next)) {
1723 vm_map_entry_unlink(map, next);
1724 entry->end = next->end;
1725 vm_map_entry_resize_free(map, entry);
1726 vm_map_merged_neighbor_dispose(map, next);
1727 }
1728 }
1729 /*
1730 * vm_map_clip_start: [ internal use only ]
1731 *
1732 * Asserts that the given entry begins at or after
1733 * the specified address; if necessary,
1734 * it splits the entry into two.
1735 */
1736 #define vm_map_clip_start(map, entry, startaddr) \
1737 { \
1738 if (startaddr > entry->start) \
1739 _vm_map_clip_start(map, entry, startaddr); \
1740 }
1741
1742 /*
1743 * This routine is called only when it is known that
1744 * the entry must be split.
1745 */
1746 static void
1747 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1748 {
1749 vm_map_entry_t new_entry;
1750
1751 VM_MAP_ASSERT_LOCKED(map);
1752 KASSERT(entry->end > start && entry->start < start,
1753 ("_vm_map_clip_start: invalid clip of entry %p", entry));
1754
1755 /*
1756 * Split off the front portion -- note that we must insert the new
1757 * entry BEFORE this one, so that this entry has the specified
1758 * starting address.
1759 */
1760 vm_map_simplify_entry(map, entry);
1761
1762 /*
1763 * If there is no object backing this entry, we might as well create
1764 * one now. If we defer it, an object can get created after the map
1765 * is clipped, and individual objects will be created for the split-up
1766 * map. This is a bit of a hack, but is also about the best place to
1767 * put this improvement.
1768 */
1769 if (entry->object.vm_object == NULL && !map->system_map &&
1770 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1771 vm_object_t object;
1772 object = vm_object_allocate(OBJT_DEFAULT,
1773 atop(entry->end - entry->start));
1774 entry->object.vm_object = object;
1775 entry->offset = 0;
1776 if (entry->cred != NULL) {
1777 object->cred = entry->cred;
1778 object->charge = entry->end - entry->start;
1779 entry->cred = NULL;
1780 }
1781 } else if (entry->object.vm_object != NULL &&
1782 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1783 entry->cred != NULL) {
1784 VM_OBJECT_WLOCK(entry->object.vm_object);
1785 KASSERT(entry->object.vm_object->cred == NULL,
1786 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1787 entry->object.vm_object->cred = entry->cred;
1788 entry->object.vm_object->charge = entry->end - entry->start;
1789 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1790 entry->cred = NULL;
1791 }
1792
1793 new_entry = vm_map_entry_create(map);
1794 *new_entry = *entry;
1795
1796 new_entry->end = start;
1797 entry->offset += (start - entry->start);
1798 entry->start = start;
1799 if (new_entry->cred != NULL)
1800 crhold(entry->cred);
1801
1802 vm_map_entry_link(map, entry->prev, new_entry);
1803
1804 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1805 vm_object_reference(new_entry->object.vm_object);
1806 /*
1807 * The object->un_pager.vnp.writemappings for the
1808 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1809 * kept as is here. The virtual pages are
1810 * re-distributed among the clipped entries, so the sum is
1811 * left the same.
1812 */
1813 }
1814 }
1815
1816 /*
1817 * vm_map_clip_end: [ internal use only ]
1818 *
1819 * Asserts that the given entry ends at or before
1820 * the specified address; if necessary,
1821 * it splits the entry into two.
1822 */
1823 #define vm_map_clip_end(map, entry, endaddr) \
1824 { \
1825 if ((endaddr) < (entry->end)) \
1826 _vm_map_clip_end((map), (entry), (endaddr)); \
1827 }
1828
1829 /*
1830 * This routine is called only when it is known that
1831 * the entry must be split.
1832 */
1833 static void
1834 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1835 {
1836 vm_map_entry_t new_entry;
1837
1838 VM_MAP_ASSERT_LOCKED(map);
1839 KASSERT(entry->start < end && entry->end > end,
1840 ("_vm_map_clip_end: invalid clip of entry %p", entry));
1841
1842 /*
1843 * If there is no object backing this entry, we might as well create
1844 * one now. If we defer it, an object can get created after the map
1845 * is clipped, and individual objects will be created for the split-up
1846 * map. This is a bit of a hack, but is also about the best place to
1847 * put this improvement.
1848 */
1849 if (entry->object.vm_object == NULL && !map->system_map &&
1850 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1851 vm_object_t object;
1852 object = vm_object_allocate(OBJT_DEFAULT,
1853 atop(entry->end - entry->start));
1854 entry->object.vm_object = object;
1855 entry->offset = 0;
1856 if (entry->cred != NULL) {
1857 object->cred = entry->cred;
1858 object->charge = entry->end - entry->start;
1859 entry->cred = NULL;
1860 }
1861 } else if (entry->object.vm_object != NULL &&
1862 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1863 entry->cred != NULL) {
1864 VM_OBJECT_WLOCK(entry->object.vm_object);
1865 KASSERT(entry->object.vm_object->cred == NULL,
1866 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1867 entry->object.vm_object->cred = entry->cred;
1868 entry->object.vm_object->charge = entry->end - entry->start;
1869 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1870 entry->cred = NULL;
1871 }
1872
1873 /*
1874 * Create a new entry and insert it AFTER the specified entry
1875 */
1876 new_entry = vm_map_entry_create(map);
1877 *new_entry = *entry;
1878
1879 new_entry->start = entry->end = end;
1880 new_entry->offset += (end - entry->start);
1881 if (new_entry->cred != NULL)
1882 crhold(entry->cred);
1883
1884 vm_map_entry_link(map, entry, new_entry);
1885
1886 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1887 vm_object_reference(new_entry->object.vm_object);
1888 }
1889 }
1890
1891 /*
1892 * vm_map_submap: [ kernel use only ]
1893 *
1894 * Mark the given range as handled by a subordinate map.
1895 *
1896 * This range must have been created with vm_map_find,
1897 * and no other operations may have been performed on this
1898 * range prior to calling vm_map_submap.
1899 *
1900 * Only a limited number of operations can be performed
1901 * within this rage after calling vm_map_submap:
1902 * vm_fault
1903 * [Don't try vm_map_copy!]
1904 *
1905 * To remove a submapping, one must first remove the
1906 * range from the superior map, and then destroy the
1907 * submap (if desired). [Better yet, don't try it.]
1908 */
1909 int
1910 vm_map_submap(
1911 vm_map_t map,
1912 vm_offset_t start,
1913 vm_offset_t end,
1914 vm_map_t submap)
1915 {
1916 vm_map_entry_t entry;
1917 int result = KERN_INVALID_ARGUMENT;
1918
1919 vm_map_lock(map);
1920
1921 VM_MAP_RANGE_CHECK(map, start, end);
1922
1923 if (vm_map_lookup_entry(map, start, &entry)) {
1924 vm_map_clip_start(map, entry, start);
1925 } else
1926 entry = entry->next;
1927
1928 vm_map_clip_end(map, entry, end);
1929
1930 if ((entry->start == start) && (entry->end == end) &&
1931 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1932 (entry->object.vm_object == NULL)) {
1933 entry->object.sub_map = submap;
1934 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1935 result = KERN_SUCCESS;
1936 }
1937 vm_map_unlock(map);
1938
1939 return (result);
1940 }
1941
1942 /*
1943 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1944 */
1945 #define MAX_INIT_PT 96
1946
1947 /*
1948 * vm_map_pmap_enter:
1949 *
1950 * Preload the specified map's pmap with mappings to the specified
1951 * object's memory-resident pages. No further physical pages are
1952 * allocated, and no further virtual pages are retrieved from secondary
1953 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1954 * limited number of page mappings are created at the low-end of the
1955 * specified address range. (For this purpose, a superpage mapping
1956 * counts as one page mapping.) Otherwise, all resident pages within
1957 * the specified address range are mapped.
1958 */
1959 static void
1960 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1961 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1962 {
1963 vm_offset_t start;
1964 vm_page_t p, p_start;
1965 vm_pindex_t mask, psize, threshold, tmpidx;
1966
1967 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1968 return;
1969 VM_OBJECT_RLOCK(object);
1970 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1971 VM_OBJECT_RUNLOCK(object);
1972 VM_OBJECT_WLOCK(object);
1973 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1974 pmap_object_init_pt(map->pmap, addr, object, pindex,
1975 size);
1976 VM_OBJECT_WUNLOCK(object);
1977 return;
1978 }
1979 VM_OBJECT_LOCK_DOWNGRADE(object);
1980 }
1981
1982 psize = atop(size);
1983 if (psize + pindex > object->size) {
1984 if (object->size < pindex) {
1985 VM_OBJECT_RUNLOCK(object);
1986 return;
1987 }
1988 psize = object->size - pindex;
1989 }
1990
1991 start = 0;
1992 p_start = NULL;
1993 threshold = MAX_INIT_PT;
1994
1995 p = vm_page_find_least(object, pindex);
1996 /*
1997 * Assert: the variable p is either (1) the page with the
1998 * least pindex greater than or equal to the parameter pindex
1999 * or (2) NULL.
2000 */
2001 for (;
2002 p != NULL && (tmpidx = p->pindex - pindex) < psize;
2003 p = TAILQ_NEXT(p, listq)) {
2004 /*
2005 * don't allow an madvise to blow away our really
2006 * free pages allocating pv entries.
2007 */
2008 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
2009 vm_page_count_severe()) ||
2010 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
2011 tmpidx >= threshold)) {
2012 psize = tmpidx;
2013 break;
2014 }
2015 if (p->valid == VM_PAGE_BITS_ALL) {
2016 if (p_start == NULL) {
2017 start = addr + ptoa(tmpidx);
2018 p_start = p;
2019 }
2020 /* Jump ahead if a superpage mapping is possible. */
2021 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
2022 (pagesizes[p->psind] - 1)) == 0) {
2023 mask = atop(pagesizes[p->psind]) - 1;
2024 if (tmpidx + mask < psize &&
2025 vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
2026 p += mask;
2027 threshold += mask;
2028 }
2029 }
2030 } else if (p_start != NULL) {
2031 pmap_enter_object(map->pmap, start, addr +
2032 ptoa(tmpidx), p_start, prot);
2033 p_start = NULL;
2034 }
2035 }
2036 if (p_start != NULL)
2037 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
2038 p_start, prot);
2039 VM_OBJECT_RUNLOCK(object);
2040 }
2041
2042 /*
2043 * vm_map_protect:
2044 *
2045 * Sets the protection of the specified address
2046 * region in the target map. If "set_max" is
2047 * specified, the maximum protection is to be set;
2048 * otherwise, only the current protection is affected.
2049 */
2050 int
2051 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
2052 vm_prot_t new_prot, boolean_t set_max)
2053 {
2054 vm_map_entry_t current, entry;
2055 vm_object_t obj;
2056 struct ucred *cred;
2057 vm_prot_t old_prot;
2058
2059 if (start == end)
2060 return (KERN_SUCCESS);
2061
2062 vm_map_lock(map);
2063
2064 /*
2065 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2066 * need to fault pages into the map and will drop the map lock while
2067 * doing so, and the VM object may end up in an inconsistent state if we
2068 * update the protection on the map entry in between faults.
2069 */
2070 vm_map_wait_busy(map);
2071
2072 VM_MAP_RANGE_CHECK(map, start, end);
2073
2074 if (vm_map_lookup_entry(map, start, &entry)) {
2075 vm_map_clip_start(map, entry, start);
2076 } else {
2077 entry = entry->next;
2078 }
2079
2080 /*
2081 * Make a first pass to check for protection violations.
2082 */
2083 for (current = entry; current->start < end; current = current->next) {
2084 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2085 continue;
2086 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2087 vm_map_unlock(map);
2088 return (KERN_INVALID_ARGUMENT);
2089 }
2090 if ((new_prot & current->max_protection) != new_prot) {
2091 vm_map_unlock(map);
2092 return (KERN_PROTECTION_FAILURE);
2093 }
2094 }
2095
2096 /*
2097 * Do an accounting pass for private read-only mappings that
2098 * now will do cow due to allowed write (e.g. debugger sets
2099 * breakpoint on text segment)
2100 */
2101 for (current = entry; current->start < end; current = current->next) {
2102
2103 vm_map_clip_end(map, current, end);
2104
2105 if (set_max ||
2106 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2107 ENTRY_CHARGED(current) ||
2108 (current->eflags & MAP_ENTRY_GUARD) != 0) {
2109 continue;
2110 }
2111
2112 cred = curthread->td_ucred;
2113 obj = current->object.vm_object;
2114
2115 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2116 if (!swap_reserve(current->end - current->start)) {
2117 vm_map_unlock(map);
2118 return (KERN_RESOURCE_SHORTAGE);
2119 }
2120 crhold(cred);
2121 current->cred = cred;
2122 continue;
2123 }
2124
2125 VM_OBJECT_WLOCK(obj);
2126 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2127 VM_OBJECT_WUNLOCK(obj);
2128 continue;
2129 }
2130
2131 /*
2132 * Charge for the whole object allocation now, since
2133 * we cannot distinguish between non-charged and
2134 * charged clipped mapping of the same object later.
2135 */
2136 KASSERT(obj->charge == 0,
2137 ("vm_map_protect: object %p overcharged (entry %p)",
2138 obj, current));
2139 if (!swap_reserve(ptoa(obj->size))) {
2140 VM_OBJECT_WUNLOCK(obj);
2141 vm_map_unlock(map);
2142 return (KERN_RESOURCE_SHORTAGE);
2143 }
2144
2145 crhold(cred);
2146 obj->cred = cred;
2147 obj->charge = ptoa(obj->size);
2148 VM_OBJECT_WUNLOCK(obj);
2149 }
2150
2151 /*
2152 * Go back and fix up protections. [Note that clipping is not
2153 * necessary the second time.]
2154 */
2155 for (current = entry; current->start < end; current = current->next) {
2156 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2157 continue;
2158
2159 old_prot = current->protection;
2160
2161 if (set_max)
2162 current->protection =
2163 (current->max_protection = new_prot) &
2164 old_prot;
2165 else
2166 current->protection = new_prot;
2167
2168 /*
2169 * For user wired map entries, the normal lazy evaluation of
2170 * write access upgrades through soft page faults is
2171 * undesirable. Instead, immediately copy any pages that are
2172 * copy-on-write and enable write access in the physical map.
2173 */
2174 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2175 (current->protection & VM_PROT_WRITE) != 0 &&
2176 (old_prot & VM_PROT_WRITE) == 0)
2177 vm_fault_copy_entry(map, map, current, current, NULL);
2178
2179 /*
2180 * When restricting access, update the physical map. Worry
2181 * about copy-on-write here.
2182 */
2183 if ((old_prot & ~current->protection) != 0) {
2184 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2185 VM_PROT_ALL)
2186 pmap_protect(map->pmap, current->start,
2187 current->end,
2188 current->protection & MASK(current));
2189 #undef MASK
2190 }
2191 vm_map_simplify_entry(map, current);
2192 }
2193 vm_map_unlock(map);
2194 return (KERN_SUCCESS);
2195 }
2196
2197 /*
2198 * vm_map_madvise:
2199 *
2200 * This routine traverses a processes map handling the madvise
2201 * system call. Advisories are classified as either those effecting
2202 * the vm_map_entry structure, or those effecting the underlying
2203 * objects.
2204 */
2205 int
2206 vm_map_madvise(
2207 vm_map_t map,
2208 vm_offset_t start,
2209 vm_offset_t end,
2210 int behav)
2211 {
2212 vm_map_entry_t current, entry;
2213 bool modify_map;
2214
2215 /*
2216 * Some madvise calls directly modify the vm_map_entry, in which case
2217 * we need to use an exclusive lock on the map and we need to perform
2218 * various clipping operations. Otherwise we only need a read-lock
2219 * on the map.
2220 */
2221 switch(behav) {
2222 case MADV_NORMAL:
2223 case MADV_SEQUENTIAL:
2224 case MADV_RANDOM:
2225 case MADV_NOSYNC:
2226 case MADV_AUTOSYNC:
2227 case MADV_NOCORE:
2228 case MADV_CORE:
2229 if (start == end)
2230 return (0);
2231 modify_map = true;
2232 vm_map_lock(map);
2233 break;
2234 case MADV_WILLNEED:
2235 case MADV_DONTNEED:
2236 case MADV_FREE:
2237 if (start == end)
2238 return (0);
2239 modify_map = false;
2240 vm_map_lock_read(map);
2241 break;
2242 default:
2243 return (EINVAL);
2244 }
2245
2246 /*
2247 * Locate starting entry and clip if necessary.
2248 */
2249 VM_MAP_RANGE_CHECK(map, start, end);
2250
2251 if (vm_map_lookup_entry(map, start, &entry)) {
2252 if (modify_map)
2253 vm_map_clip_start(map, entry, start);
2254 } else {
2255 entry = entry->next;
2256 }
2257
2258 if (modify_map) {
2259 /*
2260 * madvise behaviors that are implemented in the vm_map_entry.
2261 *
2262 * We clip the vm_map_entry so that behavioral changes are
2263 * limited to the specified address range.
2264 */
2265 for (current = entry; current->start < end;
2266 current = current->next) {
2267 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2268 continue;
2269
2270 vm_map_clip_end(map, current, end);
2271
2272 switch (behav) {
2273 case MADV_NORMAL:
2274 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2275 break;
2276 case MADV_SEQUENTIAL:
2277 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2278 break;
2279 case MADV_RANDOM:
2280 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2281 break;
2282 case MADV_NOSYNC:
2283 current->eflags |= MAP_ENTRY_NOSYNC;
2284 break;
2285 case MADV_AUTOSYNC:
2286 current->eflags &= ~MAP_ENTRY_NOSYNC;
2287 break;
2288 case MADV_NOCORE:
2289 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2290 break;
2291 case MADV_CORE:
2292 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2293 break;
2294 default:
2295 break;
2296 }
2297 vm_map_simplify_entry(map, current);
2298 }
2299 vm_map_unlock(map);
2300 } else {
2301 vm_pindex_t pstart, pend;
2302
2303 /*
2304 * madvise behaviors that are implemented in the underlying
2305 * vm_object.
2306 *
2307 * Since we don't clip the vm_map_entry, we have to clip
2308 * the vm_object pindex and count.
2309 */
2310 for (current = entry; current->start < end;
2311 current = current->next) {
2312 vm_offset_t useEnd, useStart;
2313
2314 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2315 continue;
2316
2317 pstart = OFF_TO_IDX(current->offset);
2318 pend = pstart + atop(current->end - current->start);
2319 useStart = current->start;
2320 useEnd = current->end;
2321
2322 if (current->start < start) {
2323 pstart += atop(start - current->start);
2324 useStart = start;
2325 }
2326 if (current->end > end) {
2327 pend -= atop(current->end - end);
2328 useEnd = end;
2329 }
2330
2331 if (pstart >= pend)
2332 continue;
2333
2334 /*
2335 * Perform the pmap_advise() before clearing
2336 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2337 * concurrent pmap operation, such as pmap_remove(),
2338 * could clear a reference in the pmap and set
2339 * PGA_REFERENCED on the page before the pmap_advise()
2340 * had completed. Consequently, the page would appear
2341 * referenced based upon an old reference that
2342 * occurred before this pmap_advise() ran.
2343 */
2344 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2345 pmap_advise(map->pmap, useStart, useEnd,
2346 behav);
2347
2348 vm_object_madvise(current->object.vm_object, pstart,
2349 pend, behav);
2350
2351 /*
2352 * Pre-populate paging structures in the
2353 * WILLNEED case. For wired entries, the
2354 * paging structures are already populated.
2355 */
2356 if (behav == MADV_WILLNEED &&
2357 current->wired_count == 0) {
2358 vm_map_pmap_enter(map,
2359 useStart,
2360 current->protection,
2361 current->object.vm_object,
2362 pstart,
2363 ptoa(pend - pstart),
2364 MAP_PREFAULT_MADVISE
2365 );
2366 }
2367 }
2368 vm_map_unlock_read(map);
2369 }
2370 return (0);
2371 }
2372
2373
2374 /*
2375 * vm_map_inherit:
2376 *
2377 * Sets the inheritance of the specified address
2378 * range in the target map. Inheritance
2379 * affects how the map will be shared with
2380 * child maps at the time of vmspace_fork.
2381 */
2382 int
2383 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2384 vm_inherit_t new_inheritance)
2385 {
2386 vm_map_entry_t entry;
2387 vm_map_entry_t temp_entry;
2388
2389 switch (new_inheritance) {
2390 case VM_INHERIT_NONE:
2391 case VM_INHERIT_COPY:
2392 case VM_INHERIT_SHARE:
2393 case VM_INHERIT_ZERO:
2394 break;
2395 default:
2396 return (KERN_INVALID_ARGUMENT);
2397 }
2398 if (start == end)
2399 return (KERN_SUCCESS);
2400 vm_map_lock(map);
2401 VM_MAP_RANGE_CHECK(map, start, end);
2402 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2403 entry = temp_entry;
2404 vm_map_clip_start(map, entry, start);
2405 } else
2406 entry = temp_entry->next;
2407 while (entry->start < end) {
2408 vm_map_clip_end(map, entry, end);
2409 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2410 new_inheritance != VM_INHERIT_ZERO)
2411 entry->inheritance = new_inheritance;
2412 vm_map_simplify_entry(map, entry);
2413 entry = entry->next;
2414 }
2415 vm_map_unlock(map);
2416 return (KERN_SUCCESS);
2417 }
2418
2419 /*
2420 * vm_map_unwire:
2421 *
2422 * Implements both kernel and user unwiring.
2423 */
2424 int
2425 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2426 int flags)
2427 {
2428 vm_map_entry_t entry, first_entry, tmp_entry;
2429 vm_offset_t saved_start;
2430 unsigned int last_timestamp;
2431 int rv;
2432 boolean_t need_wakeup, result, user_unwire;
2433
2434 if (start == end)
2435 return (KERN_SUCCESS);
2436 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2437 vm_map_lock(map);
2438 VM_MAP_RANGE_CHECK(map, start, end);
2439 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2440 if (flags & VM_MAP_WIRE_HOLESOK)
2441 first_entry = first_entry->next;
2442 else {
2443 vm_map_unlock(map);
2444 return (KERN_INVALID_ADDRESS);
2445 }
2446 }
2447 last_timestamp = map->timestamp;
2448 entry = first_entry;
2449 while (entry->start < end) {
2450 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2451 /*
2452 * We have not yet clipped the entry.
2453 */
2454 saved_start = (start >= entry->start) ? start :
2455 entry->start;
2456 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2457 if (vm_map_unlock_and_wait(map, 0)) {
2458 /*
2459 * Allow interruption of user unwiring?
2460 */
2461 }
2462 vm_map_lock(map);
2463 if (last_timestamp+1 != map->timestamp) {
2464 /*
2465 * Look again for the entry because the map was
2466 * modified while it was unlocked.
2467 * Specifically, the entry may have been
2468 * clipped, merged, or deleted.
2469 */
2470 if (!vm_map_lookup_entry(map, saved_start,
2471 &tmp_entry)) {
2472 if (flags & VM_MAP_WIRE_HOLESOK)
2473 tmp_entry = tmp_entry->next;
2474 else {
2475 if (saved_start == start) {
2476 /*
2477 * First_entry has been deleted.
2478 */
2479 vm_map_unlock(map);
2480 return (KERN_INVALID_ADDRESS);
2481 }
2482 end = saved_start;
2483 rv = KERN_INVALID_ADDRESS;
2484 goto done;
2485 }
2486 }
2487 if (entry == first_entry)
2488 first_entry = tmp_entry;
2489 else
2490 first_entry = NULL;
2491 entry = tmp_entry;
2492 }
2493 last_timestamp = map->timestamp;
2494 continue;
2495 }
2496 vm_map_clip_start(map, entry, start);
2497 vm_map_clip_end(map, entry, end);
2498 /*
2499 * Mark the entry in case the map lock is released. (See
2500 * above.)
2501 */
2502 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2503 entry->wiring_thread == NULL,
2504 ("owned map entry %p", entry));
2505 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2506 entry->wiring_thread = curthread;
2507 /*
2508 * Check the map for holes in the specified region.
2509 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2510 */
2511 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2512 (entry->end < end && entry->next->start > entry->end)) {
2513 end = entry->end;
2514 rv = KERN_INVALID_ADDRESS;
2515 goto done;
2516 }
2517 /*
2518 * If system unwiring, require that the entry is system wired.
2519 */
2520 if (!user_unwire &&
2521 vm_map_entry_system_wired_count(entry) == 0) {
2522 end = entry->end;
2523 rv = KERN_INVALID_ARGUMENT;
2524 goto done;
2525 }
2526 entry = entry->next;
2527 }
2528 rv = KERN_SUCCESS;
2529 done:
2530 need_wakeup = FALSE;
2531 if (first_entry == NULL) {
2532 result = vm_map_lookup_entry(map, start, &first_entry);
2533 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2534 first_entry = first_entry->next;
2535 else
2536 KASSERT(result, ("vm_map_unwire: lookup failed"));
2537 }
2538 for (entry = first_entry; entry->start < end; entry = entry->next) {
2539 /*
2540 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2541 * space in the unwired region could have been mapped
2542 * while the map lock was dropped for draining
2543 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2544 * could be simultaneously wiring this new mapping
2545 * entry. Detect these cases and skip any entries
2546 * marked as in transition by us.
2547 */
2548 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2549 entry->wiring_thread != curthread) {
2550 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2551 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2552 continue;
2553 }
2554
2555 if (rv == KERN_SUCCESS && (!user_unwire ||
2556 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2557 if (user_unwire)
2558 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2559 if (entry->wired_count == 1)
2560 vm_map_entry_unwire(map, entry);
2561 else
2562 entry->wired_count--;
2563 }
2564 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2565 ("vm_map_unwire: in-transition flag missing %p", entry));
2566 KASSERT(entry->wiring_thread == curthread,
2567 ("vm_map_unwire: alien wire %p", entry));
2568 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2569 entry->wiring_thread = NULL;
2570 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2571 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2572 need_wakeup = TRUE;
2573 }
2574 vm_map_simplify_entry(map, entry);
2575 }
2576 vm_map_unlock(map);
2577 if (need_wakeup)
2578 vm_map_wakeup(map);
2579 return (rv);
2580 }
2581
2582 /*
2583 * vm_map_wire_entry_failure:
2584 *
2585 * Handle a wiring failure on the given entry.
2586 *
2587 * The map should be locked.
2588 */
2589 static void
2590 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2591 vm_offset_t failed_addr)
2592 {
2593
2594 VM_MAP_ASSERT_LOCKED(map);
2595 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2596 entry->wired_count == 1,
2597 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2598 KASSERT(failed_addr < entry->end,
2599 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2600
2601 /*
2602 * If any pages at the start of this entry were successfully wired,
2603 * then unwire them.
2604 */
2605 if (failed_addr > entry->start) {
2606 pmap_unwire(map->pmap, entry->start, failed_addr);
2607 vm_object_unwire(entry->object.vm_object, entry->offset,
2608 failed_addr - entry->start, PQ_ACTIVE);
2609 }
2610
2611 /*
2612 * Assign an out-of-range value to represent the failure to wire this
2613 * entry.
2614 */
2615 entry->wired_count = -1;
2616 }
2617
2618 /*
2619 * vm_map_wire:
2620 *
2621 * Implements both kernel and user wiring.
2622 */
2623 int
2624 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2625 int flags)
2626 {
2627 vm_map_entry_t entry, first_entry, tmp_entry;
2628 vm_offset_t faddr, saved_end, saved_start;
2629 unsigned int last_timestamp;
2630 int rv;
2631 boolean_t need_wakeup, result, user_wire;
2632 vm_prot_t prot;
2633
2634 if (start == end)
2635 return (KERN_SUCCESS);
2636 prot = 0;
2637 if (flags & VM_MAP_WIRE_WRITE)
2638 prot |= VM_PROT_WRITE;
2639 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2640 vm_map_lock(map);
2641 VM_MAP_RANGE_CHECK(map, start, end);
2642 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2643 if (flags & VM_MAP_WIRE_HOLESOK)
2644 first_entry = first_entry->next;
2645 else {
2646 vm_map_unlock(map);
2647 return (KERN_INVALID_ADDRESS);
2648 }
2649 }
2650 last_timestamp = map->timestamp;
2651 entry = first_entry;
2652 while (entry->start < end) {
2653 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2654 /*
2655 * We have not yet clipped the entry.
2656 */
2657 saved_start = (start >= entry->start) ? start :
2658 entry->start;
2659 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2660 if (vm_map_unlock_and_wait(map, 0)) {
2661 /*
2662 * Allow interruption of user wiring?
2663 */
2664 }
2665 vm_map_lock(map);
2666 if (last_timestamp + 1 != map->timestamp) {
2667 /*
2668 * Look again for the entry because the map was
2669 * modified while it was unlocked.
2670 * Specifically, the entry may have been
2671 * clipped, merged, or deleted.
2672 */
2673 if (!vm_map_lookup_entry(map, saved_start,
2674 &tmp_entry)) {
2675 if (flags & VM_MAP_WIRE_HOLESOK)
2676 tmp_entry = tmp_entry->next;
2677 else {
2678 if (saved_start == start) {
2679 /*
2680 * first_entry has been deleted.
2681 */
2682 vm_map_unlock(map);
2683 return (KERN_INVALID_ADDRESS);
2684 }
2685 end = saved_start;
2686 rv = KERN_INVALID_ADDRESS;
2687 goto done;
2688 }
2689 }
2690 if (entry == first_entry)
2691 first_entry = tmp_entry;
2692 else
2693 first_entry = NULL;
2694 entry = tmp_entry;
2695 }
2696 last_timestamp = map->timestamp;
2697 continue;
2698 }
2699 vm_map_clip_start(map, entry, start);
2700 vm_map_clip_end(map, entry, end);
2701 /*
2702 * Mark the entry in case the map lock is released. (See
2703 * above.)
2704 */
2705 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2706 entry->wiring_thread == NULL,
2707 ("owned map entry %p", entry));
2708 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2709 entry->wiring_thread = curthread;
2710 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2711 || (entry->protection & prot) != prot) {
2712 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2713 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2714 end = entry->end;
2715 rv = KERN_INVALID_ADDRESS;
2716 goto done;
2717 }
2718 goto next_entry;
2719 }
2720 if (entry->wired_count == 0) {
2721 entry->wired_count++;
2722 saved_start = entry->start;
2723 saved_end = entry->end;
2724
2725 /*
2726 * Release the map lock, relying on the in-transition
2727 * mark. Mark the map busy for fork.
2728 */
2729 vm_map_busy(map);
2730 vm_map_unlock(map);
2731
2732 faddr = saved_start;
2733 do {
2734 /*
2735 * Simulate a fault to get the page and enter
2736 * it into the physical map.
2737 */
2738 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2739 VM_FAULT_WIRE)) != KERN_SUCCESS)
2740 break;
2741 } while ((faddr += PAGE_SIZE) < saved_end);
2742 vm_map_lock(map);
2743 vm_map_unbusy(map);
2744 if (last_timestamp + 1 != map->timestamp) {
2745 /*
2746 * Look again for the entry because the map was
2747 * modified while it was unlocked. The entry
2748 * may have been clipped, but NOT merged or
2749 * deleted.
2750 */
2751 result = vm_map_lookup_entry(map, saved_start,
2752 &tmp_entry);
2753 KASSERT(result, ("vm_map_wire: lookup failed"));
2754 if (entry == first_entry)
2755 first_entry = tmp_entry;
2756 else
2757 first_entry = NULL;
2758 entry = tmp_entry;
2759 while (entry->end < saved_end) {
2760 /*
2761 * In case of failure, handle entries
2762 * that were not fully wired here;
2763 * fully wired entries are handled
2764 * later.
2765 */
2766 if (rv != KERN_SUCCESS &&
2767 faddr < entry->end)
2768 vm_map_wire_entry_failure(map,
2769 entry, faddr);
2770 entry = entry->next;
2771 }
2772 }
2773 last_timestamp = map->timestamp;
2774 if (rv != KERN_SUCCESS) {
2775 vm_map_wire_entry_failure(map, entry, faddr);
2776 end = entry->end;
2777 goto done;
2778 }
2779 } else if (!user_wire ||
2780 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2781 entry->wired_count++;
2782 }
2783 /*
2784 * Check the map for holes in the specified region.
2785 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2786 */
2787 next_entry:
2788 if ((flags & VM_MAP_WIRE_HOLESOK) == 0 &&
2789 entry->end < end && entry->next->start > entry->end) {
2790 end = entry->end;
2791 rv = KERN_INVALID_ADDRESS;
2792 goto done;
2793 }
2794 entry = entry->next;
2795 }
2796 rv = KERN_SUCCESS;
2797 done:
2798 need_wakeup = FALSE;
2799 if (first_entry == NULL) {
2800 result = vm_map_lookup_entry(map, start, &first_entry);
2801 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2802 first_entry = first_entry->next;
2803 else
2804 KASSERT(result, ("vm_map_wire: lookup failed"));
2805 }
2806 for (entry = first_entry; entry->start < end; entry = entry->next) {
2807 /*
2808 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2809 * space in the unwired region could have been mapped
2810 * while the map lock was dropped for faulting in the
2811 * pages or draining MAP_ENTRY_IN_TRANSITION.
2812 * Moreover, another thread could be simultaneously
2813 * wiring this new mapping entry. Detect these cases
2814 * and skip any entries marked as in transition not by us.
2815 */
2816 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2817 entry->wiring_thread != curthread) {
2818 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2819 ("vm_map_wire: !HOLESOK and new/changed entry"));
2820 continue;
2821 }
2822
2823 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2824 goto next_entry_done;
2825
2826 if (rv == KERN_SUCCESS) {
2827 if (user_wire)
2828 entry->eflags |= MAP_ENTRY_USER_WIRED;
2829 } else if (entry->wired_count == -1) {
2830 /*
2831 * Wiring failed on this entry. Thus, unwiring is
2832 * unnecessary.
2833 */
2834 entry->wired_count = 0;
2835 } else if (!user_wire ||
2836 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2837 /*
2838 * Undo the wiring. Wiring succeeded on this entry
2839 * but failed on a later entry.
2840 */
2841 if (entry->wired_count == 1)
2842 vm_map_entry_unwire(map, entry);
2843 else
2844 entry->wired_count--;
2845 }
2846 next_entry_done:
2847 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2848 ("vm_map_wire: in-transition flag missing %p", entry));
2849 KASSERT(entry->wiring_thread == curthread,
2850 ("vm_map_wire: alien wire %p", entry));
2851 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2852 MAP_ENTRY_WIRE_SKIPPED);
2853 entry->wiring_thread = NULL;
2854 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2855 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2856 need_wakeup = TRUE;
2857 }
2858 vm_map_simplify_entry(map, entry);
2859 }
2860 vm_map_unlock(map);
2861 if (need_wakeup)
2862 vm_map_wakeup(map);
2863 return (rv);
2864 }
2865
2866 /*
2867 * vm_map_sync
2868 *
2869 * Push any dirty cached pages in the address range to their pager.
2870 * If syncio is TRUE, dirty pages are written synchronously.
2871 * If invalidate is TRUE, any cached pages are freed as well.
2872 *
2873 * If the size of the region from start to end is zero, we are
2874 * supposed to flush all modified pages within the region containing
2875 * start. Unfortunately, a region can be split or coalesced with
2876 * neighboring regions, making it difficult to determine what the
2877 * original region was. Therefore, we approximate this requirement by
2878 * flushing the current region containing start.
2879 *
2880 * Returns an error if any part of the specified range is not mapped.
2881 */
2882 int
2883 vm_map_sync(
2884 vm_map_t map,
2885 vm_offset_t start,
2886 vm_offset_t end,
2887 boolean_t syncio,
2888 boolean_t invalidate)
2889 {
2890 vm_map_entry_t current;
2891 vm_map_entry_t entry;
2892 vm_size_t size;
2893 vm_object_t object;
2894 vm_ooffset_t offset;
2895 unsigned int last_timestamp;
2896 boolean_t failed;
2897
2898 vm_map_lock_read(map);
2899 VM_MAP_RANGE_CHECK(map, start, end);
2900 if (!vm_map_lookup_entry(map, start, &entry)) {
2901 vm_map_unlock_read(map);
2902 return (KERN_INVALID_ADDRESS);
2903 } else if (start == end) {
2904 start = entry->start;
2905 end = entry->end;
2906 }
2907 /*
2908 * Make a first pass to check for user-wired memory and holes.
2909 */
2910 for (current = entry; current->start < end; current = current->next) {
2911 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2912 vm_map_unlock_read(map);
2913 return (KERN_INVALID_ARGUMENT);
2914 }
2915 if (end > current->end &&
2916 current->end != current->next->start) {
2917 vm_map_unlock_read(map);
2918 return (KERN_INVALID_ADDRESS);
2919 }
2920 }
2921
2922 if (invalidate)
2923 pmap_remove(map->pmap, start, end);
2924 failed = FALSE;
2925
2926 /*
2927 * Make a second pass, cleaning/uncaching pages from the indicated
2928 * objects as we go.
2929 */
2930 for (current = entry; current->start < end;) {
2931 offset = current->offset + (start - current->start);
2932 size = (end <= current->end ? end : current->end) - start;
2933 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2934 vm_map_t smap;
2935 vm_map_entry_t tentry;
2936 vm_size_t tsize;
2937
2938 smap = current->object.sub_map;
2939 vm_map_lock_read(smap);
2940 (void) vm_map_lookup_entry(smap, offset, &tentry);
2941 tsize = tentry->end - offset;
2942 if (tsize < size)
2943 size = tsize;
2944 object = tentry->object.vm_object;
2945 offset = tentry->offset + (offset - tentry->start);
2946 vm_map_unlock_read(smap);
2947 } else {
2948 object = current->object.vm_object;
2949 }
2950 vm_object_reference(object);
2951 last_timestamp = map->timestamp;
2952 vm_map_unlock_read(map);
2953 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2954 failed = TRUE;
2955 start += size;
2956 vm_object_deallocate(object);
2957 vm_map_lock_read(map);
2958 if (last_timestamp == map->timestamp ||
2959 !vm_map_lookup_entry(map, start, ¤t))
2960 current = current->next;
2961 }
2962
2963 vm_map_unlock_read(map);
2964 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2965 }
2966
2967 /*
2968 * vm_map_entry_unwire: [ internal use only ]
2969 *
2970 * Make the region specified by this entry pageable.
2971 *
2972 * The map in question should be locked.
2973 * [This is the reason for this routine's existence.]
2974 */
2975 static void
2976 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2977 {
2978
2979 VM_MAP_ASSERT_LOCKED(map);
2980 KASSERT(entry->wired_count > 0,
2981 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2982 pmap_unwire(map->pmap, entry->start, entry->end);
2983 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2984 entry->start, PQ_ACTIVE);
2985 entry->wired_count = 0;
2986 }
2987
2988 static void
2989 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2990 {
2991
2992 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2993 vm_object_deallocate(entry->object.vm_object);
2994 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2995 }
2996
2997 /*
2998 * vm_map_entry_delete: [ internal use only ]
2999 *
3000 * Deallocate the given entry from the target map.
3001 */
3002 static void
3003 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
3004 {
3005 vm_object_t object;
3006 vm_pindex_t offidxstart, offidxend, count, size1;
3007 vm_size_t size;
3008
3009 vm_map_entry_unlink(map, entry);
3010 object = entry->object.vm_object;
3011
3012 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
3013 MPASS(entry->cred == NULL);
3014 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
3015 MPASS(object == NULL);
3016 vm_map_entry_deallocate(entry, map->system_map);
3017 return;
3018 }
3019
3020 size = entry->end - entry->start;
3021 map->size -= size;
3022
3023 if (entry->cred != NULL) {
3024 swap_release_by_cred(size, entry->cred);
3025 crfree(entry->cred);
3026 }
3027
3028 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
3029 (object != NULL)) {
3030 KASSERT(entry->cred == NULL || object->cred == NULL ||
3031 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
3032 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
3033 count = atop(size);
3034 offidxstart = OFF_TO_IDX(entry->offset);
3035 offidxend = offidxstart + count;
3036 VM_OBJECT_WLOCK(object);
3037 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
3038 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
3039 object == kernel_object)) {
3040 vm_object_collapse(object);
3041
3042 /*
3043 * The option OBJPR_NOTMAPPED can be passed here
3044 * because vm_map_delete() already performed
3045 * pmap_remove() on the only mapping to this range
3046 * of pages.
3047 */
3048 vm_object_page_remove(object, offidxstart, offidxend,
3049 OBJPR_NOTMAPPED);
3050 if (object->type == OBJT_SWAP)
3051 swap_pager_freespace(object, offidxstart,
3052 count);
3053 if (offidxend >= object->size &&
3054 offidxstart < object->size) {
3055 size1 = object->size;
3056 object->size = offidxstart;
3057 if (object->cred != NULL) {
3058 size1 -= object->size;
3059 KASSERT(object->charge >= ptoa(size1),
3060 ("object %p charge < 0", object));
3061 swap_release_by_cred(ptoa(size1),
3062 object->cred);
3063 object->charge -= ptoa(size1);
3064 }
3065 }
3066 }
3067 VM_OBJECT_WUNLOCK(object);
3068 } else
3069 entry->object.vm_object = NULL;
3070 if (map->system_map)
3071 vm_map_entry_deallocate(entry, TRUE);
3072 else {
3073 entry->next = curthread->td_map_def_user;
3074 curthread->td_map_def_user = entry;
3075 }
3076 }
3077
3078 /*
3079 * vm_map_delete: [ internal use only ]
3080 *
3081 * Deallocates the given address range from the target
3082 * map.
3083 */
3084 int
3085 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3086 {
3087 vm_map_entry_t entry;
3088 vm_map_entry_t first_entry;
3089
3090 VM_MAP_ASSERT_LOCKED(map);
3091 if (start == end)
3092 return (KERN_SUCCESS);
3093
3094 /*
3095 * Find the start of the region, and clip it
3096 */
3097 if (!vm_map_lookup_entry(map, start, &first_entry))
3098 entry = first_entry->next;
3099 else {
3100 entry = first_entry;
3101 vm_map_clip_start(map, entry, start);
3102 }
3103
3104 /*
3105 * Step through all entries in this region
3106 */
3107 while (entry->start < end) {
3108 vm_map_entry_t next;
3109
3110 /*
3111 * Wait for wiring or unwiring of an entry to complete.
3112 * Also wait for any system wirings to disappear on
3113 * user maps.
3114 */
3115 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3116 (vm_map_pmap(map) != kernel_pmap &&
3117 vm_map_entry_system_wired_count(entry) != 0)) {
3118 unsigned int last_timestamp;
3119 vm_offset_t saved_start;
3120 vm_map_entry_t tmp_entry;
3121
3122 saved_start = entry->start;
3123 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3124 last_timestamp = map->timestamp;
3125 (void) vm_map_unlock_and_wait(map, 0);
3126 vm_map_lock(map);
3127 if (last_timestamp + 1 != map->timestamp) {
3128 /*
3129 * Look again for the entry because the map was
3130 * modified while it was unlocked.
3131 * Specifically, the entry may have been
3132 * clipped, merged, or deleted.
3133 */
3134 if (!vm_map_lookup_entry(map, saved_start,
3135 &tmp_entry))
3136 entry = tmp_entry->next;
3137 else {
3138 entry = tmp_entry;
3139 vm_map_clip_start(map, entry,
3140 saved_start);
3141 }
3142 }
3143 continue;
3144 }
3145 vm_map_clip_end(map, entry, end);
3146
3147 next = entry->next;
3148
3149 /*
3150 * Unwire before removing addresses from the pmap; otherwise,
3151 * unwiring will put the entries back in the pmap.
3152 */
3153 if (entry->wired_count != 0)
3154 vm_map_entry_unwire(map, entry);
3155
3156 /*
3157 * Remove mappings for the pages, but only if the
3158 * mappings could exist. For instance, it does not
3159 * make sense to call pmap_remove() for guard entries.
3160 */
3161 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3162 entry->object.vm_object != NULL)
3163 pmap_remove(map->pmap, entry->start, entry->end);
3164
3165 /*
3166 * Delete the entry only after removing all pmap
3167 * entries pointing to its pages. (Otherwise, its
3168 * page frames may be reallocated, and any modify bits
3169 * will be set in the wrong object!)
3170 */
3171 vm_map_entry_delete(map, entry);
3172 entry = next;
3173 }
3174 return (KERN_SUCCESS);
3175 }
3176
3177 /*
3178 * vm_map_remove:
3179 *
3180 * Remove the given address range from the target map.
3181 * This is the exported form of vm_map_delete.
3182 */
3183 int
3184 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3185 {
3186 int result;
3187
3188 vm_map_lock(map);
3189 VM_MAP_RANGE_CHECK(map, start, end);
3190 result = vm_map_delete(map, start, end);
3191 vm_map_unlock(map);
3192 return (result);
3193 }
3194
3195 /*
3196 * vm_map_check_protection:
3197 *
3198 * Assert that the target map allows the specified privilege on the
3199 * entire address region given. The entire region must be allocated.
3200 *
3201 * WARNING! This code does not and should not check whether the
3202 * contents of the region is accessible. For example a smaller file
3203 * might be mapped into a larger address space.
3204 *
3205 * NOTE! This code is also called by munmap().
3206 *
3207 * The map must be locked. A read lock is sufficient.
3208 */
3209 boolean_t
3210 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3211 vm_prot_t protection)
3212 {
3213 vm_map_entry_t entry;
3214 vm_map_entry_t tmp_entry;
3215
3216 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3217 return (FALSE);
3218 entry = tmp_entry;
3219
3220 while (start < end) {
3221 /*
3222 * No holes allowed!
3223 */
3224 if (start < entry->start)
3225 return (FALSE);
3226 /*
3227 * Check protection associated with entry.
3228 */
3229 if ((entry->protection & protection) != protection)
3230 return (FALSE);
3231 /* go to next entry */
3232 start = entry->end;
3233 entry = entry->next;
3234 }
3235 return (TRUE);
3236 }
3237
3238 /*
3239 * vm_map_copy_entry:
3240 *
3241 * Copies the contents of the source entry to the destination
3242 * entry. The entries *must* be aligned properly.
3243 */
3244 static void
3245 vm_map_copy_entry(
3246 vm_map_t src_map,
3247 vm_map_t dst_map,
3248 vm_map_entry_t src_entry,
3249 vm_map_entry_t dst_entry,
3250 vm_ooffset_t *fork_charge)
3251 {
3252 vm_object_t src_object;
3253 vm_map_entry_t fake_entry;
3254 vm_offset_t size;
3255 struct ucred *cred;
3256 int charged;
3257
3258 VM_MAP_ASSERT_LOCKED(dst_map);
3259
3260 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3261 return;
3262
3263 if (src_entry->wired_count == 0 ||
3264 (src_entry->protection & VM_PROT_WRITE) == 0) {
3265 /*
3266 * If the source entry is marked needs_copy, it is already
3267 * write-protected.
3268 */
3269 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3270 (src_entry->protection & VM_PROT_WRITE) != 0) {
3271 pmap_protect(src_map->pmap,
3272 src_entry->start,
3273 src_entry->end,
3274 src_entry->protection & ~VM_PROT_WRITE);
3275 }
3276
3277 /*
3278 * Make a copy of the object.
3279 */
3280 size = src_entry->end - src_entry->start;
3281 if ((src_object = src_entry->object.vm_object) != NULL) {
3282 VM_OBJECT_WLOCK(src_object);
3283 charged = ENTRY_CHARGED(src_entry);
3284 if (src_object->handle == NULL &&
3285 (src_object->type == OBJT_DEFAULT ||
3286 src_object->type == OBJT_SWAP)) {
3287 vm_object_collapse(src_object);
3288 if ((src_object->flags & (OBJ_NOSPLIT |
3289 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3290 vm_object_split(src_entry);
3291 src_object =
3292 src_entry->object.vm_object;
3293 }
3294 }
3295 vm_object_reference_locked(src_object);
3296 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3297 if (src_entry->cred != NULL &&
3298 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3299 KASSERT(src_object->cred == NULL,
3300 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3301 src_object));
3302 src_object->cred = src_entry->cred;
3303 src_object->charge = size;
3304 }
3305 VM_OBJECT_WUNLOCK(src_object);
3306 dst_entry->object.vm_object = src_object;
3307 if (charged) {
3308 cred = curthread->td_ucred;
3309 crhold(cred);
3310 dst_entry->cred = cred;
3311 *fork_charge += size;
3312 if (!(src_entry->eflags &
3313 MAP_ENTRY_NEEDS_COPY)) {
3314 crhold(cred);
3315 src_entry->cred = cred;
3316 *fork_charge += size;
3317 }
3318 }
3319 src_entry->eflags |= MAP_ENTRY_COW |
3320 MAP_ENTRY_NEEDS_COPY;
3321 dst_entry->eflags |= MAP_ENTRY_COW |
3322 MAP_ENTRY_NEEDS_COPY;
3323 dst_entry->offset = src_entry->offset;
3324 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3325 /*
3326 * MAP_ENTRY_VN_WRITECNT cannot
3327 * indicate write reference from
3328 * src_entry, since the entry is
3329 * marked as needs copy. Allocate a
3330 * fake entry that is used to
3331 * decrement object->un_pager.vnp.writecount
3332 * at the appropriate time. Attach
3333 * fake_entry to the deferred list.
3334 */
3335 fake_entry = vm_map_entry_create(dst_map);
3336 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3337 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3338 vm_object_reference(src_object);
3339 fake_entry->object.vm_object = src_object;
3340 fake_entry->start = src_entry->start;
3341 fake_entry->end = src_entry->end;
3342 fake_entry->next = curthread->td_map_def_user;
3343 curthread->td_map_def_user = fake_entry;
3344 }
3345
3346 pmap_copy(dst_map->pmap, src_map->pmap,
3347 dst_entry->start, dst_entry->end - dst_entry->start,
3348 src_entry->start);
3349 } else {
3350 dst_entry->object.vm_object = NULL;
3351 dst_entry->offset = 0;
3352 if (src_entry->cred != NULL) {
3353 dst_entry->cred = curthread->td_ucred;
3354 crhold(dst_entry->cred);
3355 *fork_charge += size;
3356 }
3357 }
3358 } else {
3359 /*
3360 * We don't want to make writeable wired pages copy-on-write.
3361 * Immediately copy these pages into the new map by simulating
3362 * page faults. The new pages are pageable.
3363 */
3364 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3365 fork_charge);
3366 }
3367 }
3368
3369 /*
3370 * vmspace_map_entry_forked:
3371 * Update the newly-forked vmspace each time a map entry is inherited
3372 * or copied. The values for vm_dsize and vm_tsize are approximate
3373 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3374 */
3375 static void
3376 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3377 vm_map_entry_t entry)
3378 {
3379 vm_size_t entrysize;
3380 vm_offset_t newend;
3381
3382 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3383 return;
3384 entrysize = entry->end - entry->start;
3385 vm2->vm_map.size += entrysize;
3386 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3387 vm2->vm_ssize += btoc(entrysize);
3388 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3389 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3390 newend = MIN(entry->end,
3391 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3392 vm2->vm_dsize += btoc(newend - entry->start);
3393 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3394 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3395 newend = MIN(entry->end,
3396 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3397 vm2->vm_tsize += btoc(newend - entry->start);
3398 }
3399 }
3400
3401 /*
3402 * vmspace_fork:
3403 * Create a new process vmspace structure and vm_map
3404 * based on those of an existing process. The new map
3405 * is based on the old map, according to the inheritance
3406 * values on the regions in that map.
3407 *
3408 * XXX It might be worth coalescing the entries added to the new vmspace.
3409 *
3410 * The source map must not be locked.
3411 */
3412 struct vmspace *
3413 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3414 {
3415 struct vmspace *vm2;
3416 vm_map_t new_map, old_map;
3417 vm_map_entry_t new_entry, old_entry;
3418 vm_object_t object;
3419 int locked;
3420 vm_inherit_t inh;
3421
3422 old_map = &vm1->vm_map;
3423 /* Copy immutable fields of vm1 to vm2. */
3424 vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map), NULL);
3425 if (vm2 == NULL)
3426 return (NULL);
3427 vm2->vm_taddr = vm1->vm_taddr;
3428 vm2->vm_daddr = vm1->vm_daddr;
3429 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3430 vm_map_lock(old_map);
3431 if (old_map->busy)
3432 vm_map_wait_busy(old_map);
3433 new_map = &vm2->vm_map;
3434 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3435 KASSERT(locked, ("vmspace_fork: lock failed"));
3436
3437 old_entry = old_map->header.next;
3438
3439 while (old_entry != &old_map->header) {
3440 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3441 panic("vm_map_fork: encountered a submap");
3442
3443 inh = old_entry->inheritance;
3444 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3445 inh != VM_INHERIT_NONE)
3446 inh = VM_INHERIT_COPY;
3447
3448 switch (inh) {
3449 case VM_INHERIT_NONE:
3450 break;
3451
3452 case VM_INHERIT_SHARE:
3453 /*
3454 * Clone the entry, creating the shared object if necessary.
3455 */
3456 object = old_entry->object.vm_object;
3457 if (object == NULL) {
3458 object = vm_object_allocate(OBJT_DEFAULT,
3459 atop(old_entry->end - old_entry->start));
3460 old_entry->object.vm_object = object;
3461 old_entry->offset = 0;
3462 if (old_entry->cred != NULL) {
3463 object->cred = old_entry->cred;
3464 object->charge = old_entry->end -
3465 old_entry->start;
3466 old_entry->cred = NULL;
3467 }
3468 }
3469
3470 /*
3471 * Add the reference before calling vm_object_shadow
3472 * to insure that a shadow object is created.
3473 */
3474 vm_object_reference(object);
3475 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3476 vm_object_shadow(&old_entry->object.vm_object,
3477 &old_entry->offset,
3478 old_entry->end - old_entry->start);
3479 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3480 /* Transfer the second reference too. */
3481 vm_object_reference(
3482 old_entry->object.vm_object);
3483
3484 /*
3485 * As in vm_map_simplify_entry(), the
3486 * vnode lock will not be acquired in
3487 * this call to vm_object_deallocate().
3488 */
3489 vm_object_deallocate(object);
3490 object = old_entry->object.vm_object;
3491 }
3492 VM_OBJECT_WLOCK(object);
3493 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3494 if (old_entry->cred != NULL) {
3495 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3496 object->cred = old_entry->cred;
3497 object->charge = old_entry->end - old_entry->start;
3498 old_entry->cred = NULL;
3499 }
3500
3501 /*
3502 * Assert the correct state of the vnode
3503 * v_writecount while the object is locked, to
3504 * not relock it later for the assertion
3505 * correctness.
3506 */
3507 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3508 object->type == OBJT_VNODE) {
3509 KASSERT(((struct vnode *)object->handle)->
3510 v_writecount > 0,
3511 ("vmspace_fork: v_writecount %p", object));
3512 KASSERT(object->un_pager.vnp.writemappings > 0,
3513 ("vmspace_fork: vnp.writecount %p",
3514 object));
3515 }
3516 VM_OBJECT_WUNLOCK(object);
3517
3518 /*
3519 * Clone the entry, referencing the shared object.
3520 */
3521 new_entry = vm_map_entry_create(new_map);
3522 *new_entry = *old_entry;
3523 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3524 MAP_ENTRY_IN_TRANSITION);
3525 new_entry->wiring_thread = NULL;
3526 new_entry->wired_count = 0;
3527 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3528 vnode_pager_update_writecount(object,
3529 new_entry->start, new_entry->end);
3530 }
3531
3532 /*
3533 * Insert the entry into the new map -- we know we're
3534 * inserting at the end of the new map.
3535 */
3536 vm_map_entry_link(new_map, new_map->header.prev,
3537 new_entry);
3538 vmspace_map_entry_forked(vm1, vm2, new_entry);
3539
3540 /*
3541 * Update the physical map
3542 */
3543 pmap_copy(new_map->pmap, old_map->pmap,
3544 new_entry->start,
3545 (old_entry->end - old_entry->start),
3546 old_entry->start);
3547 break;
3548
3549 case VM_INHERIT_COPY:
3550 /*
3551 * Clone the entry and link into the map.
3552 */
3553 new_entry = vm_map_entry_create(new_map);
3554 *new_entry = *old_entry;
3555 /*
3556 * Copied entry is COW over the old object.
3557 */
3558 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3559 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3560 new_entry->wiring_thread = NULL;
3561 new_entry->wired_count = 0;
3562 new_entry->object.vm_object = NULL;
3563 new_entry->cred = NULL;
3564 vm_map_entry_link(new_map, new_map->header.prev,
3565 new_entry);
3566 vmspace_map_entry_forked(vm1, vm2, new_entry);
3567 vm_map_copy_entry(old_map, new_map, old_entry,
3568 new_entry, fork_charge);
3569 break;
3570
3571 case VM_INHERIT_ZERO:
3572 /*
3573 * Create a new anonymous mapping entry modelled from
3574 * the old one.
3575 */
3576 new_entry = vm_map_entry_create(new_map);
3577 memset(new_entry, 0, sizeof(*new_entry));
3578
3579 new_entry->start = old_entry->start;
3580 new_entry->end = old_entry->end;
3581 new_entry->eflags = old_entry->eflags &
3582 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
3583 MAP_ENTRY_VN_WRITECNT);
3584 new_entry->protection = old_entry->protection;
3585 new_entry->max_protection = old_entry->max_protection;
3586 new_entry->inheritance = VM_INHERIT_ZERO;
3587
3588 vm_map_entry_link(new_map, new_map->header.prev,
3589 new_entry);
3590 vmspace_map_entry_forked(vm1, vm2, new_entry);
3591
3592 new_entry->cred = curthread->td_ucred;
3593 crhold(new_entry->cred);
3594 *fork_charge += (new_entry->end - new_entry->start);
3595
3596 break;
3597 }
3598 old_entry = old_entry->next;
3599 }
3600 /*
3601 * Use inlined vm_map_unlock() to postpone handling the deferred
3602 * map entries, which cannot be done until both old_map and
3603 * new_map locks are released.
3604 */
3605 sx_xunlock(&old_map->lock);
3606 sx_xunlock(&new_map->lock);
3607 vm_map_process_deferred();
3608
3609 return (vm2);
3610 }
3611
3612 /*
3613 * Create a process's stack for exec_new_vmspace(). This function is never
3614 * asked to wire the newly created stack.
3615 */
3616 int
3617 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3618 vm_prot_t prot, vm_prot_t max, int cow)
3619 {
3620 vm_size_t growsize, init_ssize;
3621 rlim_t vmemlim;
3622 int rv;
3623
3624 MPASS((map->flags & MAP_WIREFUTURE) == 0);
3625 growsize = sgrowsiz;
3626 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3627 vm_map_lock(map);
3628 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3629 /* If we would blow our VMEM resource limit, no go */
3630 if (map->size + init_ssize > vmemlim) {
3631 rv = KERN_NO_SPACE;
3632 goto out;
3633 }
3634 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3635 max, cow);
3636 out:
3637 vm_map_unlock(map);
3638 return (rv);
3639 }
3640
3641 static int stack_guard_page = 1;
3642 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3643 &stack_guard_page, 0,
3644 "Specifies the number of guard pages for a stack that grows");
3645
3646 static int
3647 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3648 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3649 {
3650 vm_map_entry_t new_entry, prev_entry;
3651 vm_offset_t bot, gap_bot, gap_top, top;
3652 vm_size_t init_ssize, sgp;
3653 int orient, rv;
3654
3655 /*
3656 * The stack orientation is piggybacked with the cow argument.
3657 * Extract it into orient and mask the cow argument so that we
3658 * don't pass it around further.
3659 */
3660 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
3661 KASSERT(orient != 0, ("No stack grow direction"));
3662 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
3663 ("bi-dir stack"));
3664
3665 if (addrbos < vm_map_min(map) ||
3666 addrbos + max_ssize > vm_map_max(map) ||
3667 addrbos + max_ssize <= addrbos)
3668 return (KERN_INVALID_ADDRESS);
3669 sgp = (vm_size_t)stack_guard_page * PAGE_SIZE;
3670 if (sgp >= max_ssize)
3671 return (KERN_INVALID_ARGUMENT);
3672
3673 init_ssize = growsize;
3674 if (max_ssize < init_ssize + sgp)
3675 init_ssize = max_ssize - sgp;
3676
3677 /* If addr is already mapped, no go */
3678 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3679 return (KERN_NO_SPACE);
3680
3681 /*
3682 * If we can't accommodate max_ssize in the current mapping, no go.
3683 */
3684 if (prev_entry->next->start < addrbos + max_ssize)
3685 return (KERN_NO_SPACE);
3686
3687 /*
3688 * We initially map a stack of only init_ssize. We will grow as
3689 * needed later. Depending on the orientation of the stack (i.e.
3690 * the grow direction) we either map at the top of the range, the
3691 * bottom of the range or in the middle.
3692 *
3693 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3694 * and cow to be 0. Possibly we should eliminate these as input
3695 * parameters, and just pass these values here in the insert call.
3696 */
3697 if (orient == MAP_STACK_GROWS_DOWN) {
3698 bot = addrbos + max_ssize - init_ssize;
3699 top = bot + init_ssize;
3700 gap_bot = addrbos;
3701 gap_top = bot;
3702 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
3703 bot = addrbos;
3704 top = bot + init_ssize;
3705 gap_bot = top;
3706 gap_top = addrbos + max_ssize;
3707 }
3708 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3709 if (rv != KERN_SUCCESS)
3710 return (rv);
3711 new_entry = prev_entry->next;
3712 KASSERT(new_entry->end == top || new_entry->start == bot,
3713 ("Bad entry start/end for new stack entry"));
3714 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3715 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3716 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3717 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3718 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3719 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3720 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
3721 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
3722 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
3723 if (rv != KERN_SUCCESS)
3724 (void)vm_map_delete(map, bot, top);
3725 return (rv);
3726 }
3727
3728 /*
3729 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
3730 * successfully grow the stack.
3731 */
3732 static int
3733 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
3734 {
3735 vm_map_entry_t stack_entry;
3736 struct proc *p;
3737 struct vmspace *vm;
3738 struct ucred *cred;
3739 vm_offset_t gap_end, gap_start, grow_start;
3740 size_t grow_amount, guard, max_grow;
3741 rlim_t lmemlim, stacklim, vmemlim;
3742 int rv, rv1;
3743 bool gap_deleted, grow_down, is_procstack;
3744 #ifdef notyet
3745 uint64_t limit;
3746 #endif
3747 #ifdef RACCT
3748 int error;
3749 #endif
3750
3751 p = curproc;
3752 vm = p->p_vmspace;
3753
3754 /*
3755 * Disallow stack growth when the access is performed by a
3756 * debugger or AIO daemon. The reason is that the wrong
3757 * resource limits are applied.
3758 */
3759 if (map != &p->p_vmspace->vm_map || p->p_textvp == NULL)
3760 return (KERN_FAILURE);
3761
3762 MPASS(!map->system_map);
3763
3764 guard = stack_guard_page * PAGE_SIZE;
3765 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
3766 stacklim = lim_cur(curthread, RLIMIT_STACK);
3767 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3768 retry:
3769 /* If addr is not in a hole for a stack grow area, no need to grow. */
3770 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
3771 return (KERN_FAILURE);
3772 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
3773 return (KERN_SUCCESS);
3774 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
3775 stack_entry = gap_entry->next;
3776 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
3777 stack_entry->start != gap_entry->end)
3778 return (KERN_FAILURE);
3779 grow_amount = round_page(stack_entry->start - addr);
3780 grow_down = true;
3781 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
3782 stack_entry = gap_entry->prev;
3783 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
3784 stack_entry->end != gap_entry->start)
3785 return (KERN_FAILURE);
3786 grow_amount = round_page(addr + 1 - stack_entry->end);
3787 grow_down = false;
3788 } else {
3789 return (KERN_FAILURE);
3790 }
3791 max_grow = gap_entry->end - gap_entry->start;
3792 if (guard > max_grow)
3793 return (KERN_NO_SPACE);
3794 max_grow -= guard;
3795 if (grow_amount > max_grow)
3796 return (KERN_NO_SPACE);
3797
3798 /*
3799 * If this is the main process stack, see if we're over the stack
3800 * limit.
3801 */
3802 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
3803 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
3804 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
3805 return (KERN_NO_SPACE);
3806
3807 #ifdef RACCT
3808 if (racct_enable) {
3809 PROC_LOCK(p);
3810 if (is_procstack && racct_set(p, RACCT_STACK,
3811 ctob(vm->vm_ssize) + grow_amount)) {
3812 PROC_UNLOCK(p);
3813 return (KERN_NO_SPACE);
3814 }
3815 PROC_UNLOCK(p);
3816 }
3817 #endif
3818
3819 grow_amount = roundup(grow_amount, sgrowsiz);
3820 if (grow_amount > max_grow)
3821 grow_amount = max_grow;
3822 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3823 grow_amount = trunc_page((vm_size_t)stacklim) -
3824 ctob(vm->vm_ssize);
3825 }
3826
3827 #ifdef notyet
3828 PROC_LOCK(p);
3829 limit = racct_get_available(p, RACCT_STACK);
3830 PROC_UNLOCK(p);
3831 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3832 grow_amount = limit - ctob(vm->vm_ssize);
3833 #endif
3834
3835 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
3836 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3837 rv = KERN_NO_SPACE;
3838 goto out;
3839 }
3840 #ifdef RACCT
3841 if (racct_enable) {
3842 PROC_LOCK(p);
3843 if (racct_set(p, RACCT_MEMLOCK,
3844 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3845 PROC_UNLOCK(p);
3846 rv = KERN_NO_SPACE;
3847 goto out;
3848 }
3849 PROC_UNLOCK(p);
3850 }
3851 #endif
3852 }
3853
3854 /* If we would blow our VMEM resource limit, no go */
3855 if (map->size + grow_amount > vmemlim) {
3856 rv = KERN_NO_SPACE;
3857 goto out;
3858 }
3859 #ifdef RACCT
3860 if (racct_enable) {
3861 PROC_LOCK(p);
3862 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3863 PROC_UNLOCK(p);
3864 rv = KERN_NO_SPACE;
3865 goto out;
3866 }
3867 PROC_UNLOCK(p);
3868 }
3869 #endif
3870
3871 if (vm_map_lock_upgrade(map)) {
3872 gap_entry = NULL;
3873 vm_map_lock_read(map);
3874 goto retry;
3875 }
3876
3877 if (grow_down) {
3878 grow_start = gap_entry->end - grow_amount;
3879 if (gap_entry->start + grow_amount == gap_entry->end) {
3880 gap_start = gap_entry->start;
3881 gap_end = gap_entry->end;
3882 vm_map_entry_delete(map, gap_entry);
3883 gap_deleted = true;
3884 } else {
3885 MPASS(gap_entry->start < gap_entry->end - grow_amount);
3886 gap_entry->end -= grow_amount;
3887 vm_map_entry_resize_free(map, gap_entry);
3888 gap_deleted = false;
3889 }
3890 rv = vm_map_insert(map, NULL, 0, grow_start,
3891 grow_start + grow_amount,
3892 stack_entry->protection, stack_entry->max_protection,
3893 MAP_STACK_GROWS_DOWN);
3894 if (rv != KERN_SUCCESS) {
3895 if (gap_deleted) {
3896 rv1 = vm_map_insert(map, NULL, 0, gap_start,
3897 gap_end, VM_PROT_NONE, VM_PROT_NONE,
3898 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
3899 MPASS(rv1 == KERN_SUCCESS);
3900 } else {
3901 gap_entry->end += grow_amount;
3902 vm_map_entry_resize_free(map, gap_entry);
3903 }
3904 }
3905 } else {
3906 grow_start = stack_entry->end;
3907 cred = stack_entry->cred;
3908 if (cred == NULL && stack_entry->object.vm_object != NULL)
3909 cred = stack_entry->object.vm_object->cred;
3910 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3911 rv = KERN_NO_SPACE;
3912 /* Grow the underlying object if applicable. */
3913 else if (stack_entry->object.vm_object == NULL ||
3914 vm_object_coalesce(stack_entry->object.vm_object,
3915 stack_entry->offset,
3916 (vm_size_t)(stack_entry->end - stack_entry->start),
3917 (vm_size_t)grow_amount, cred != NULL)) {
3918 if (gap_entry->start + grow_amount == gap_entry->end)
3919 vm_map_entry_delete(map, gap_entry);
3920 else
3921 gap_entry->start += grow_amount;
3922 stack_entry->end += grow_amount;
3923 map->size += grow_amount;
3924 vm_map_entry_resize_free(map, stack_entry);
3925 rv = KERN_SUCCESS;
3926 } else
3927 rv = KERN_FAILURE;
3928 }
3929 if (rv == KERN_SUCCESS && is_procstack)
3930 vm->vm_ssize += btoc(grow_amount);
3931
3932 /*
3933 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3934 */
3935 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
3936 vm_map_unlock(map);
3937 vm_map_wire(map, grow_start, grow_start + grow_amount,
3938 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
3939 vm_map_lock_read(map);
3940 } else
3941 vm_map_lock_downgrade(map);
3942
3943 out:
3944 #ifdef RACCT
3945 if (racct_enable && rv != KERN_SUCCESS) {
3946 PROC_LOCK(p);
3947 error = racct_set(p, RACCT_VMEM, map->size);
3948 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3949 if (!old_mlock) {
3950 error = racct_set(p, RACCT_MEMLOCK,
3951 ptoa(pmap_wired_count(map->pmap)));
3952 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3953 }
3954 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3955 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3956 PROC_UNLOCK(p);
3957 }
3958 #endif
3959
3960 return (rv);
3961 }
3962
3963 /*
3964 * Unshare the specified VM space for exec. If other processes are
3965 * mapped to it, then create a new one. The new vmspace is null.
3966 */
3967 int
3968 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3969 {
3970 struct vmspace *oldvmspace = p->p_vmspace;
3971 struct vmspace *newvmspace;
3972
3973 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3974 ("vmspace_exec recursed"));
3975 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3976 if (newvmspace == NULL)
3977 return (ENOMEM);
3978 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3979 /*
3980 * This code is written like this for prototype purposes. The
3981 * goal is to avoid running down the vmspace here, but let the
3982 * other process's that are still using the vmspace to finally
3983 * run it down. Even though there is little or no chance of blocking
3984 * here, it is a good idea to keep this form for future mods.
3985 */
3986 PROC_VMSPACE_LOCK(p);
3987 p->p_vmspace = newvmspace;
3988 PROC_VMSPACE_UNLOCK(p);
3989 if (p == curthread->td_proc)
3990 pmap_activate(curthread);
3991 curthread->td_pflags |= TDP_EXECVMSPC;
3992 return (0);
3993 }
3994
3995 /*
3996 * Unshare the specified VM space for forcing COW. This
3997 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3998 */
3999 int
4000 vmspace_unshare(struct proc *p)
4001 {
4002 struct vmspace *oldvmspace = p->p_vmspace;
4003 struct vmspace *newvmspace;
4004 vm_ooffset_t fork_charge;
4005
4006 if (oldvmspace->vm_refcnt == 1)
4007 return (0);
4008 fork_charge = 0;
4009 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
4010 if (newvmspace == NULL)
4011 return (ENOMEM);
4012 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
4013 vmspace_free(newvmspace);
4014 return (ENOMEM);
4015 }
4016 PROC_VMSPACE_LOCK(p);
4017 p->p_vmspace = newvmspace;
4018 PROC_VMSPACE_UNLOCK(p);
4019 if (p == curthread->td_proc)
4020 pmap_activate(curthread);
4021 vmspace_free(oldvmspace);
4022 return (0);
4023 }
4024
4025 /*
4026 * vm_map_lookup:
4027 *
4028 * Finds the VM object, offset, and
4029 * protection for a given virtual address in the
4030 * specified map, assuming a page fault of the
4031 * type specified.
4032 *
4033 * Leaves the map in question locked for read; return
4034 * values are guaranteed until a vm_map_lookup_done
4035 * call is performed. Note that the map argument
4036 * is in/out; the returned map must be used in
4037 * the call to vm_map_lookup_done.
4038 *
4039 * A handle (out_entry) is returned for use in
4040 * vm_map_lookup_done, to make that fast.
4041 *
4042 * If a lookup is requested with "write protection"
4043 * specified, the map may be changed to perform virtual
4044 * copying operations, although the data referenced will
4045 * remain the same.
4046 */
4047 int
4048 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
4049 vm_offset_t vaddr,
4050 vm_prot_t fault_typea,
4051 vm_map_entry_t *out_entry, /* OUT */
4052 vm_object_t *object, /* OUT */
4053 vm_pindex_t *pindex, /* OUT */
4054 vm_prot_t *out_prot, /* OUT */
4055 boolean_t *wired) /* OUT */
4056 {
4057 vm_map_entry_t entry;
4058 vm_map_t map = *var_map;
4059 vm_prot_t prot;
4060 vm_prot_t fault_type = fault_typea;
4061 vm_object_t eobject;
4062 vm_size_t size;
4063 struct ucred *cred;
4064
4065 RetryLookup:
4066
4067 vm_map_lock_read(map);
4068
4069 RetryLookupLocked:
4070 /*
4071 * Lookup the faulting address.
4072 */
4073 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4074 vm_map_unlock_read(map);
4075 return (KERN_INVALID_ADDRESS);
4076 }
4077
4078 entry = *out_entry;
4079
4080 /*
4081 * Handle submaps.
4082 */
4083 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4084 vm_map_t old_map = map;
4085
4086 *var_map = map = entry->object.sub_map;
4087 vm_map_unlock_read(old_map);
4088 goto RetryLookup;
4089 }
4090
4091 /*
4092 * Check whether this task is allowed to have this page.
4093 */
4094 prot = entry->protection;
4095 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4096 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4097 if (prot == VM_PROT_NONE && map != kernel_map &&
4098 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4099 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4100 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4101 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4102 goto RetryLookupLocked;
4103 }
4104 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4105 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4106 vm_map_unlock_read(map);
4107 return (KERN_PROTECTION_FAILURE);
4108 }
4109 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4110 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4111 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4112 ("entry %p flags %x", entry, entry->eflags));
4113 if ((fault_typea & VM_PROT_COPY) != 0 &&
4114 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4115 (entry->eflags & MAP_ENTRY_COW) == 0) {
4116 vm_map_unlock_read(map);
4117 return (KERN_PROTECTION_FAILURE);
4118 }
4119
4120 /*
4121 * If this page is not pageable, we have to get it for all possible
4122 * accesses.
4123 */
4124 *wired = (entry->wired_count != 0);
4125 if (*wired)
4126 fault_type = entry->protection;
4127 size = entry->end - entry->start;
4128 /*
4129 * If the entry was copy-on-write, we either ...
4130 */
4131 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4132 /*
4133 * If we want to write the page, we may as well handle that
4134 * now since we've got the map locked.
4135 *
4136 * If we don't need to write the page, we just demote the
4137 * permissions allowed.
4138 */
4139 if ((fault_type & VM_PROT_WRITE) != 0 ||
4140 (fault_typea & VM_PROT_COPY) != 0) {
4141 /*
4142 * Make a new object, and place it in the object
4143 * chain. Note that no new references have appeared
4144 * -- one just moved from the map to the new
4145 * object.
4146 */
4147 if (vm_map_lock_upgrade(map))
4148 goto RetryLookup;
4149
4150 if (entry->cred == NULL) {
4151 /*
4152 * The debugger owner is charged for
4153 * the memory.
4154 */
4155 cred = curthread->td_ucred;
4156 crhold(cred);
4157 if (!swap_reserve_by_cred(size, cred)) {
4158 crfree(cred);
4159 vm_map_unlock(map);
4160 return (KERN_RESOURCE_SHORTAGE);
4161 }
4162 entry->cred = cred;
4163 }
4164 vm_object_shadow(&entry->object.vm_object,
4165 &entry->offset, size);
4166 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4167 eobject = entry->object.vm_object;
4168 if (eobject->cred != NULL) {
4169 /*
4170 * The object was not shadowed.
4171 */
4172 swap_release_by_cred(size, entry->cred);
4173 crfree(entry->cred);
4174 entry->cred = NULL;
4175 } else if (entry->cred != NULL) {
4176 VM_OBJECT_WLOCK(eobject);
4177 eobject->cred = entry->cred;
4178 eobject->charge = size;
4179 VM_OBJECT_WUNLOCK(eobject);
4180 entry->cred = NULL;
4181 }
4182
4183 vm_map_lock_downgrade(map);
4184 } else {
4185 /*
4186 * We're attempting to read a copy-on-write page --
4187 * don't allow writes.
4188 */
4189 prot &= ~VM_PROT_WRITE;
4190 }
4191 }
4192
4193 /*
4194 * Create an object if necessary.
4195 */
4196 if (entry->object.vm_object == NULL &&
4197 !map->system_map) {
4198 if (vm_map_lock_upgrade(map))
4199 goto RetryLookup;
4200 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4201 atop(size));
4202 entry->offset = 0;
4203 if (entry->cred != NULL) {
4204 VM_OBJECT_WLOCK(entry->object.vm_object);
4205 entry->object.vm_object->cred = entry->cred;
4206 entry->object.vm_object->charge = size;
4207 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4208 entry->cred = NULL;
4209 }
4210 vm_map_lock_downgrade(map);
4211 }
4212
4213 /*
4214 * Return the object/offset from this entry. If the entry was
4215 * copy-on-write or empty, it has been fixed up.
4216 */
4217 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset);
4218 *object = entry->object.vm_object;
4219
4220 *out_prot = prot;
4221 return (KERN_SUCCESS);
4222 }
4223
4224 /*
4225 * vm_map_lookup_locked:
4226 *
4227 * Lookup the faulting address. A version of vm_map_lookup that returns
4228 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4229 */
4230 int
4231 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4232 vm_offset_t vaddr,
4233 vm_prot_t fault_typea,
4234 vm_map_entry_t *out_entry, /* OUT */
4235 vm_object_t *object, /* OUT */
4236 vm_pindex_t *pindex, /* OUT */
4237 vm_prot_t *out_prot, /* OUT */
4238 boolean_t *wired) /* OUT */
4239 {
4240 vm_map_entry_t entry;
4241 vm_map_t map = *var_map;
4242 vm_prot_t prot;
4243 vm_prot_t fault_type = fault_typea;
4244
4245 /*
4246 * Lookup the faulting address.
4247 */
4248 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4249 return (KERN_INVALID_ADDRESS);
4250
4251 entry = *out_entry;
4252
4253 /*
4254 * Fail if the entry refers to a submap.
4255 */
4256 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4257 return (KERN_FAILURE);
4258
4259 /*
4260 * Check whether this task is allowed to have this page.
4261 */
4262 prot = entry->protection;
4263 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4264 if ((fault_type & prot) != fault_type)
4265 return (KERN_PROTECTION_FAILURE);
4266
4267 /*
4268 * If this page is not pageable, we have to get it for all possible
4269 * accesses.
4270 */
4271 *wired = (entry->wired_count != 0);
4272 if (*wired)
4273 fault_type = entry->protection;
4274
4275 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4276 /*
4277 * Fail if the entry was copy-on-write for a write fault.
4278 */
4279 if (fault_type & VM_PROT_WRITE)
4280 return (KERN_FAILURE);
4281 /*
4282 * We're attempting to read a copy-on-write page --
4283 * don't allow writes.
4284 */
4285 prot &= ~VM_PROT_WRITE;
4286 }
4287
4288 /*
4289 * Fail if an object should be created.
4290 */
4291 if (entry->object.vm_object == NULL && !map->system_map)
4292 return (KERN_FAILURE);
4293
4294 /*
4295 * Return the object/offset from this entry. If the entry was
4296 * copy-on-write or empty, it has been fixed up.
4297 */
4298 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset);
4299 *object = entry->object.vm_object;
4300
4301 *out_prot = prot;
4302 return (KERN_SUCCESS);
4303 }
4304
4305 /*
4306 * vm_map_lookup_done:
4307 *
4308 * Releases locks acquired by a vm_map_lookup
4309 * (according to the handle returned by that lookup).
4310 */
4311 void
4312 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4313 {
4314 /*
4315 * Unlock the main-level map
4316 */
4317 vm_map_unlock_read(map);
4318 }
4319
4320 vm_offset_t
4321 vm_map_max_KBI(const struct vm_map *map)
4322 {
4323
4324 return (vm_map_max(map));
4325 }
4326
4327 vm_offset_t
4328 vm_map_min_KBI(const struct vm_map *map)
4329 {
4330
4331 return (vm_map_min(map));
4332 }
4333
4334 pmap_t
4335 vm_map_pmap_KBI(vm_map_t map)
4336 {
4337
4338 return (map->pmap);
4339 }
4340
4341 #include "opt_ddb.h"
4342 #ifdef DDB
4343 #include <sys/kernel.h>
4344
4345 #include <ddb/ddb.h>
4346
4347 static void
4348 vm_map_print(vm_map_t map)
4349 {
4350 vm_map_entry_t entry;
4351
4352 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4353 (void *)map,
4354 (void *)map->pmap, map->nentries, map->timestamp);
4355
4356 db_indent += 2;
4357 for (entry = map->header.next; entry != &map->header;
4358 entry = entry->next) {
4359 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4360 (void *)entry, (void *)entry->start, (void *)entry->end,
4361 entry->eflags);
4362 {
4363 static char *inheritance_name[4] =
4364 {"share", "copy", "none", "donate_copy"};
4365
4366 db_iprintf(" prot=%x/%x/%s",
4367 entry->protection,
4368 entry->max_protection,
4369 inheritance_name[(int)(unsigned char)entry->inheritance]);
4370 if (entry->wired_count != 0)
4371 db_printf(", wired");
4372 }
4373 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4374 db_printf(", share=%p, offset=0x%jx\n",
4375 (void *)entry->object.sub_map,
4376 (uintmax_t)entry->offset);
4377 if ((entry->prev == &map->header) ||
4378 (entry->prev->object.sub_map !=
4379 entry->object.sub_map)) {
4380 db_indent += 2;
4381 vm_map_print((vm_map_t)entry->object.sub_map);
4382 db_indent -= 2;
4383 }
4384 } else {
4385 if (entry->cred != NULL)
4386 db_printf(", ruid %d", entry->cred->cr_ruid);
4387 db_printf(", object=%p, offset=0x%jx",
4388 (void *)entry->object.vm_object,
4389 (uintmax_t)entry->offset);
4390 if (entry->object.vm_object && entry->object.vm_object->cred)
4391 db_printf(", obj ruid %d charge %jx",
4392 entry->object.vm_object->cred->cr_ruid,
4393 (uintmax_t)entry->object.vm_object->charge);
4394 if (entry->eflags & MAP_ENTRY_COW)
4395 db_printf(", copy (%s)",
4396 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4397 db_printf("\n");
4398
4399 if ((entry->prev == &map->header) ||
4400 (entry->prev->object.vm_object !=
4401 entry->object.vm_object)) {
4402 db_indent += 2;
4403 vm_object_print((db_expr_t)(intptr_t)
4404 entry->object.vm_object,
4405 0, 0, (char *)0);
4406 db_indent -= 2;
4407 }
4408 }
4409 }
4410 db_indent -= 2;
4411 }
4412
4413 DB_SHOW_COMMAND(map, map)
4414 {
4415
4416 if (!have_addr) {
4417 db_printf("usage: show map <addr>\n");
4418 return;
4419 }
4420 vm_map_print((vm_map_t)addr);
4421 }
4422
4423 DB_SHOW_COMMAND(procvm, procvm)
4424 {
4425 struct proc *p;
4426
4427 if (have_addr) {
4428 p = db_lookup_proc(addr);
4429 } else {
4430 p = curproc;
4431 }
4432
4433 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4434 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4435 (void *)vmspace_pmap(p->p_vmspace));
4436
4437 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);
4438 }
4439
4440 #endif /* DDB */
Cache object: 69c846a9afc55e3c1f5f774714ee7bbd
|