FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_map.c
1 /*-
2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
33 *
34 *
35 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
36 * All rights reserved.
37 *
38 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
39 *
40 * Permission to use, copy, modify and distribute this software and
41 * its documentation is hereby granted, provided that both the copyright
42 * notice and this permission notice appear in all copies of the
43 * software, derivative works or modified versions, and any portions
44 * thereof, and that both notices appear in supporting documentation.
45 *
46 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
47 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
48 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
49 *
50 * Carnegie Mellon requests users of this software to return to
51 *
52 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
53 * School of Computer Science
54 * Carnegie Mellon University
55 * Pittsburgh PA 15213-3890
56 *
57 * any improvements or extensions that they make and grant Carnegie the
58 * rights to redistribute these changes.
59 */
60
61 /*
62 * Virtual memory mapping module.
63 */
64
65 #include <sys/cdefs.h>
66 __FBSDID("$FreeBSD: releng/11.1/sys/vm/vm_map.c 320904 2017-07-12 07:00:56Z kib $");
67
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/kernel.h>
71 #include <sys/ktr.h>
72 #include <sys/lock.h>
73 #include <sys/mutex.h>
74 #include <sys/proc.h>
75 #include <sys/vmmeter.h>
76 #include <sys/mman.h>
77 #include <sys/vnode.h>
78 #include <sys/racct.h>
79 #include <sys/resourcevar.h>
80 #include <sys/rwlock.h>
81 #include <sys/file.h>
82 #include <sys/sysctl.h>
83 #include <sys/sysent.h>
84 #include <sys/shm.h>
85
86 #include <vm/vm.h>
87 #include <vm/vm_param.h>
88 #include <vm/pmap.h>
89 #include <vm/vm_map.h>
90 #include <vm/vm_page.h>
91 #include <vm/vm_object.h>
92 #include <vm/vm_pager.h>
93 #include <vm/vm_kern.h>
94 #include <vm/vm_extern.h>
95 #include <vm/vnode_pager.h>
96 #include <vm/swap_pager.h>
97 #include <vm/uma.h>
98
99 /*
100 * Virtual memory maps provide for the mapping, protection,
101 * and sharing of virtual memory objects. In addition,
102 * this module provides for an efficient virtual copy of
103 * memory from one map to another.
104 *
105 * Synchronization is required prior to most operations.
106 *
107 * Maps consist of an ordered doubly-linked list of simple
108 * entries; a self-adjusting binary search tree of these
109 * entries is used to speed up lookups.
110 *
111 * Since portions of maps are specified by start/end addresses,
112 * which may not align with existing map entries, all
113 * routines merely "clip" entries to these start/end values.
114 * [That is, an entry is split into two, bordering at a
115 * start or end value.] Note that these clippings may not
116 * always be necessary (as the two resulting entries are then
117 * not changed); however, the clipping is done for convenience.
118 *
119 * As mentioned above, virtual copy operations are performed
120 * by copying VM object references from one map to
121 * another, and then marking both regions as copy-on-write.
122 */
123
124 static struct mtx map_sleep_mtx;
125 static uma_zone_t mapentzone;
126 static uma_zone_t kmapentzone;
127 static uma_zone_t mapzone;
128 static uma_zone_t vmspace_zone;
129 static int vmspace_zinit(void *mem, int size, int flags);
130 static int vm_map_zinit(void *mem, int ize, int flags);
131 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
132 vm_offset_t max);
133 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
134 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
135 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
136 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
137 vm_map_entry_t gap_entry);
138 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
139 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
140 #ifdef INVARIANTS
141 static void vm_map_zdtor(void *mem, int size, void *arg);
142 static void vmspace_zdtor(void *mem, int size, void *arg);
143 #endif
144 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
145 vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
146 int cow);
147 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
148 vm_offset_t failed_addr);
149
150 #define ENTRY_CHARGED(e) ((e)->cred != NULL || \
151 ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
152 !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
153
154 /*
155 * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
156 * stable.
157 */
158 #define PROC_VMSPACE_LOCK(p) do { } while (0)
159 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
160
161 /*
162 * VM_MAP_RANGE_CHECK: [ internal use only ]
163 *
164 * Asserts that the starting and ending region
165 * addresses fall within the valid range of the map.
166 */
167 #define VM_MAP_RANGE_CHECK(map, start, end) \
168 { \
169 if (start < vm_map_min(map)) \
170 start = vm_map_min(map); \
171 if (end > vm_map_max(map)) \
172 end = vm_map_max(map); \
173 if (start > end) \
174 start = end; \
175 }
176
177 /*
178 * vm_map_startup:
179 *
180 * Initialize the vm_map module. Must be called before
181 * any other vm_map routines.
182 *
183 * Map and entry structures are allocated from the general
184 * purpose memory pool with some exceptions:
185 *
186 * - The kernel map and kmem submap are allocated statically.
187 * - Kernel map entries are allocated out of a static pool.
188 *
189 * These restrictions are necessary since malloc() uses the
190 * maps and requires map entries.
191 */
192
193 void
194 vm_map_startup(void)
195 {
196 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
197 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
198 #ifdef INVARIANTS
199 vm_map_zdtor,
200 #else
201 NULL,
202 #endif
203 vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
204 uma_prealloc(mapzone, MAX_KMAP);
205 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
206 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
207 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
208 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
209 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
210 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
211 #ifdef INVARIANTS
212 vmspace_zdtor,
213 #else
214 NULL,
215 #endif
216 vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
217 }
218
219 static int
220 vmspace_zinit(void *mem, int size, int flags)
221 {
222 struct vmspace *vm;
223
224 vm = (struct vmspace *)mem;
225
226 vm->vm_map.pmap = NULL;
227 (void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
228 PMAP_LOCK_INIT(vmspace_pmap(vm));
229 return (0);
230 }
231
232 static int
233 vm_map_zinit(void *mem, int size, int flags)
234 {
235 vm_map_t map;
236
237 map = (vm_map_t)mem;
238 memset(map, 0, sizeof(*map));
239 mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
240 sx_init(&map->lock, "vm map (user)");
241 return (0);
242 }
243
244 #ifdef INVARIANTS
245 static void
246 vmspace_zdtor(void *mem, int size, void *arg)
247 {
248 struct vmspace *vm;
249
250 vm = (struct vmspace *)mem;
251
252 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
253 }
254 static void
255 vm_map_zdtor(void *mem, int size, void *arg)
256 {
257 vm_map_t map;
258
259 map = (vm_map_t)mem;
260 KASSERT(map->nentries == 0,
261 ("map %p nentries == %d on free.",
262 map, map->nentries));
263 KASSERT(map->size == 0,
264 ("map %p size == %lu on free.",
265 map, (unsigned long)map->size));
266 }
267 #endif /* INVARIANTS */
268
269 /*
270 * Allocate a vmspace structure, including a vm_map and pmap,
271 * and initialize those structures. The refcnt is set to 1.
272 *
273 * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
274 */
275 struct vmspace *
276 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
277 {
278 struct vmspace *vm;
279
280 vm = uma_zalloc(vmspace_zone, M_WAITOK);
281
282 KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
283
284 if (pinit == NULL)
285 pinit = &pmap_pinit;
286
287 if (!pinit(vmspace_pmap(vm))) {
288 uma_zfree(vmspace_zone, vm);
289 return (NULL);
290 }
291 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
292 _vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
293 vm->vm_refcnt = 1;
294 vm->vm_shm = NULL;
295 vm->vm_swrss = 0;
296 vm->vm_tsize = 0;
297 vm->vm_dsize = 0;
298 vm->vm_ssize = 0;
299 vm->vm_taddr = 0;
300 vm->vm_daddr = 0;
301 vm->vm_maxsaddr = 0;
302 return (vm);
303 }
304
305 #ifdef RACCT
306 static void
307 vmspace_container_reset(struct proc *p)
308 {
309
310 PROC_LOCK(p);
311 racct_set(p, RACCT_DATA, 0);
312 racct_set(p, RACCT_STACK, 0);
313 racct_set(p, RACCT_RSS, 0);
314 racct_set(p, RACCT_MEMLOCK, 0);
315 racct_set(p, RACCT_VMEM, 0);
316 PROC_UNLOCK(p);
317 }
318 #endif
319
320 static inline void
321 vmspace_dofree(struct vmspace *vm)
322 {
323
324 CTR1(KTR_VM, "vmspace_free: %p", vm);
325
326 /*
327 * Make sure any SysV shm is freed, it might not have been in
328 * exit1().
329 */
330 shmexit(vm);
331
332 /*
333 * Lock the map, to wait out all other references to it.
334 * Delete all of the mappings and pages they hold, then call
335 * the pmap module to reclaim anything left.
336 */
337 (void)vm_map_remove(&vm->vm_map, vm->vm_map.min_offset,
338 vm->vm_map.max_offset);
339
340 pmap_release(vmspace_pmap(vm));
341 vm->vm_map.pmap = NULL;
342 uma_zfree(vmspace_zone, vm);
343 }
344
345 void
346 vmspace_free(struct vmspace *vm)
347 {
348
349 WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
350 "vmspace_free() called");
351
352 if (vm->vm_refcnt == 0)
353 panic("vmspace_free: attempt to free already freed vmspace");
354
355 if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
356 vmspace_dofree(vm);
357 }
358
359 void
360 vmspace_exitfree(struct proc *p)
361 {
362 struct vmspace *vm;
363
364 PROC_VMSPACE_LOCK(p);
365 vm = p->p_vmspace;
366 p->p_vmspace = NULL;
367 PROC_VMSPACE_UNLOCK(p);
368 KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
369 vmspace_free(vm);
370 }
371
372 void
373 vmspace_exit(struct thread *td)
374 {
375 int refcnt;
376 struct vmspace *vm;
377 struct proc *p;
378
379 /*
380 * Release user portion of address space.
381 * This releases references to vnodes,
382 * which could cause I/O if the file has been unlinked.
383 * Need to do this early enough that we can still sleep.
384 *
385 * The last exiting process to reach this point releases as
386 * much of the environment as it can. vmspace_dofree() is the
387 * slower fallback in case another process had a temporary
388 * reference to the vmspace.
389 */
390
391 p = td->td_proc;
392 vm = p->p_vmspace;
393 atomic_add_int(&vmspace0.vm_refcnt, 1);
394 do {
395 refcnt = vm->vm_refcnt;
396 if (refcnt > 1 && p->p_vmspace != &vmspace0) {
397 /* Switch now since other proc might free vmspace */
398 PROC_VMSPACE_LOCK(p);
399 p->p_vmspace = &vmspace0;
400 PROC_VMSPACE_UNLOCK(p);
401 pmap_activate(td);
402 }
403 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt - 1));
404 if (refcnt == 1) {
405 if (p->p_vmspace != vm) {
406 /* vmspace not yet freed, switch back */
407 PROC_VMSPACE_LOCK(p);
408 p->p_vmspace = vm;
409 PROC_VMSPACE_UNLOCK(p);
410 pmap_activate(td);
411 }
412 pmap_remove_pages(vmspace_pmap(vm));
413 /* Switch now since this proc will free vmspace */
414 PROC_VMSPACE_LOCK(p);
415 p->p_vmspace = &vmspace0;
416 PROC_VMSPACE_UNLOCK(p);
417 pmap_activate(td);
418 vmspace_dofree(vm);
419 }
420 #ifdef RACCT
421 if (racct_enable)
422 vmspace_container_reset(p);
423 #endif
424 }
425
426 /* Acquire reference to vmspace owned by another process. */
427
428 struct vmspace *
429 vmspace_acquire_ref(struct proc *p)
430 {
431 struct vmspace *vm;
432 int refcnt;
433
434 PROC_VMSPACE_LOCK(p);
435 vm = p->p_vmspace;
436 if (vm == NULL) {
437 PROC_VMSPACE_UNLOCK(p);
438 return (NULL);
439 }
440 do {
441 refcnt = vm->vm_refcnt;
442 if (refcnt <= 0) { /* Avoid 0->1 transition */
443 PROC_VMSPACE_UNLOCK(p);
444 return (NULL);
445 }
446 } while (!atomic_cmpset_int(&vm->vm_refcnt, refcnt, refcnt + 1));
447 if (vm != p->p_vmspace) {
448 PROC_VMSPACE_UNLOCK(p);
449 vmspace_free(vm);
450 return (NULL);
451 }
452 PROC_VMSPACE_UNLOCK(p);
453 return (vm);
454 }
455
456 /*
457 * Switch between vmspaces in an AIO kernel process.
458 *
459 * The AIO kernel processes switch to and from a user process's
460 * vmspace while performing an I/O operation on behalf of a user
461 * process. The new vmspace is either the vmspace of a user process
462 * obtained from an active AIO request or the initial vmspace of the
463 * AIO kernel process (when it is idling). Because user processes
464 * will block to drain any active AIO requests before proceeding in
465 * exit() or execve(), the vmspace reference count for these vmspaces
466 * can never be 0. This allows for a much simpler implementation than
467 * the loop in vmspace_acquire_ref() above. Similarly, AIO kernel
468 * processes hold an extra reference on their initial vmspace for the
469 * life of the process so that this guarantee is true for any vmspace
470 * passed as 'newvm'.
471 */
472 void
473 vmspace_switch_aio(struct vmspace *newvm)
474 {
475 struct vmspace *oldvm;
476
477 /* XXX: Need some way to assert that this is an aio daemon. */
478
479 KASSERT(newvm->vm_refcnt > 0,
480 ("vmspace_switch_aio: newvm unreferenced"));
481
482 oldvm = curproc->p_vmspace;
483 if (oldvm == newvm)
484 return;
485
486 /*
487 * Point to the new address space and refer to it.
488 */
489 curproc->p_vmspace = newvm;
490 atomic_add_int(&newvm->vm_refcnt, 1);
491
492 /* Activate the new mapping. */
493 pmap_activate(curthread);
494
495 /* Remove the daemon's reference to the old address space. */
496 KASSERT(oldvm->vm_refcnt > 1,
497 ("vmspace_switch_aio: oldvm dropping last reference"));
498 vmspace_free(oldvm);
499 }
500
501 void
502 _vm_map_lock(vm_map_t map, const char *file, int line)
503 {
504
505 if (map->system_map)
506 mtx_lock_flags_(&map->system_mtx, 0, file, line);
507 else
508 sx_xlock_(&map->lock, file, line);
509 map->timestamp++;
510 }
511
512 static void
513 vm_map_process_deferred(void)
514 {
515 struct thread *td;
516 vm_map_entry_t entry, next;
517 vm_object_t object;
518
519 td = curthread;
520 entry = td->td_map_def_user;
521 td->td_map_def_user = NULL;
522 while (entry != NULL) {
523 next = entry->next;
524 if ((entry->eflags & MAP_ENTRY_VN_WRITECNT) != 0) {
525 /*
526 * Decrement the object's writemappings and
527 * possibly the vnode's v_writecount.
528 */
529 KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
530 ("Submap with writecount"));
531 object = entry->object.vm_object;
532 KASSERT(object != NULL, ("No object for writecount"));
533 vnode_pager_release_writecount(object, entry->start,
534 entry->end);
535 }
536 vm_map_entry_deallocate(entry, FALSE);
537 entry = next;
538 }
539 }
540
541 void
542 _vm_map_unlock(vm_map_t map, const char *file, int line)
543 {
544
545 if (map->system_map)
546 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
547 else {
548 sx_xunlock_(&map->lock, file, line);
549 vm_map_process_deferred();
550 }
551 }
552
553 void
554 _vm_map_lock_read(vm_map_t map, const char *file, int line)
555 {
556
557 if (map->system_map)
558 mtx_lock_flags_(&map->system_mtx, 0, file, line);
559 else
560 sx_slock_(&map->lock, file, line);
561 }
562
563 void
564 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
565 {
566
567 if (map->system_map)
568 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
569 else {
570 sx_sunlock_(&map->lock, file, line);
571 vm_map_process_deferred();
572 }
573 }
574
575 int
576 _vm_map_trylock(vm_map_t map, const char *file, int line)
577 {
578 int error;
579
580 error = map->system_map ?
581 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
582 !sx_try_xlock_(&map->lock, file, line);
583 if (error == 0)
584 map->timestamp++;
585 return (error == 0);
586 }
587
588 int
589 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
590 {
591 int error;
592
593 error = map->system_map ?
594 !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
595 !sx_try_slock_(&map->lock, file, line);
596 return (error == 0);
597 }
598
599 /*
600 * _vm_map_lock_upgrade: [ internal use only ]
601 *
602 * Tries to upgrade a read (shared) lock on the specified map to a write
603 * (exclusive) lock. Returns the value "" if the upgrade succeeds and a
604 * non-zero value if the upgrade fails. If the upgrade fails, the map is
605 * returned without a read or write lock held.
606 *
607 * Requires that the map be read locked.
608 */
609 int
610 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
611 {
612 unsigned int last_timestamp;
613
614 if (map->system_map) {
615 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
616 } else {
617 if (!sx_try_upgrade_(&map->lock, file, line)) {
618 last_timestamp = map->timestamp;
619 sx_sunlock_(&map->lock, file, line);
620 vm_map_process_deferred();
621 /*
622 * If the map's timestamp does not change while the
623 * map is unlocked, then the upgrade succeeds.
624 */
625 sx_xlock_(&map->lock, file, line);
626 if (last_timestamp != map->timestamp) {
627 sx_xunlock_(&map->lock, file, line);
628 return (1);
629 }
630 }
631 }
632 map->timestamp++;
633 return (0);
634 }
635
636 void
637 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
638 {
639
640 if (map->system_map) {
641 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
642 } else
643 sx_downgrade_(&map->lock, file, line);
644 }
645
646 /*
647 * vm_map_locked:
648 *
649 * Returns a non-zero value if the caller holds a write (exclusive) lock
650 * on the specified map and the value "" otherwise.
651 */
652 int
653 vm_map_locked(vm_map_t map)
654 {
655
656 if (map->system_map)
657 return (mtx_owned(&map->system_mtx));
658 else
659 return (sx_xlocked(&map->lock));
660 }
661
662 #ifdef INVARIANTS
663 static void
664 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
665 {
666
667 if (map->system_map)
668 mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
669 else
670 sx_assert_(&map->lock, SA_XLOCKED, file, line);
671 }
672
673 #define VM_MAP_ASSERT_LOCKED(map) \
674 _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
675 #else
676 #define VM_MAP_ASSERT_LOCKED(map)
677 #endif
678
679 /*
680 * _vm_map_unlock_and_wait:
681 *
682 * Atomically releases the lock on the specified map and puts the calling
683 * thread to sleep. The calling thread will remain asleep until either
684 * vm_map_wakeup() is performed on the map or the specified timeout is
685 * exceeded.
686 *
687 * WARNING! This function does not perform deferred deallocations of
688 * objects and map entries. Therefore, the calling thread is expected to
689 * reacquire the map lock after reawakening and later perform an ordinary
690 * unlock operation, such as vm_map_unlock(), before completing its
691 * operation on the map.
692 */
693 int
694 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
695 {
696
697 mtx_lock(&map_sleep_mtx);
698 if (map->system_map)
699 mtx_unlock_flags_(&map->system_mtx, 0, file, line);
700 else
701 sx_xunlock_(&map->lock, file, line);
702 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
703 timo));
704 }
705
706 /*
707 * vm_map_wakeup:
708 *
709 * Awaken any threads that have slept on the map using
710 * vm_map_unlock_and_wait().
711 */
712 void
713 vm_map_wakeup(vm_map_t map)
714 {
715
716 /*
717 * Acquire and release map_sleep_mtx to prevent a wakeup()
718 * from being performed (and lost) between the map unlock
719 * and the msleep() in _vm_map_unlock_and_wait().
720 */
721 mtx_lock(&map_sleep_mtx);
722 mtx_unlock(&map_sleep_mtx);
723 wakeup(&map->root);
724 }
725
726 void
727 vm_map_busy(vm_map_t map)
728 {
729
730 VM_MAP_ASSERT_LOCKED(map);
731 map->busy++;
732 }
733
734 void
735 vm_map_unbusy(vm_map_t map)
736 {
737
738 VM_MAP_ASSERT_LOCKED(map);
739 KASSERT(map->busy, ("vm_map_unbusy: not busy"));
740 if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
741 vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
742 wakeup(&map->busy);
743 }
744 }
745
746 void
747 vm_map_wait_busy(vm_map_t map)
748 {
749
750 VM_MAP_ASSERT_LOCKED(map);
751 while (map->busy) {
752 vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
753 if (map->system_map)
754 msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
755 else
756 sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
757 }
758 map->timestamp++;
759 }
760
761 long
762 vmspace_resident_count(struct vmspace *vmspace)
763 {
764 return pmap_resident_count(vmspace_pmap(vmspace));
765 }
766
767 /*
768 * vm_map_create:
769 *
770 * Creates and returns a new empty VM map with
771 * the given physical map structure, and having
772 * the given lower and upper address bounds.
773 */
774 vm_map_t
775 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
776 {
777 vm_map_t result;
778
779 result = uma_zalloc(mapzone, M_WAITOK);
780 CTR1(KTR_VM, "vm_map_create: %p", result);
781 _vm_map_init(result, pmap, min, max);
782 return (result);
783 }
784
785 /*
786 * Initialize an existing vm_map structure
787 * such as that in the vmspace structure.
788 */
789 static void
790 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
791 {
792
793 map->header.next = map->header.prev = &map->header;
794 map->needs_wakeup = FALSE;
795 map->system_map = 0;
796 map->pmap = pmap;
797 map->min_offset = min;
798 map->max_offset = max;
799 map->flags = 0;
800 map->root = NULL;
801 map->timestamp = 0;
802 map->busy = 0;
803 }
804
805 void
806 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
807 {
808
809 _vm_map_init(map, pmap, min, max);
810 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
811 sx_init(&map->lock, "user map");
812 }
813
814 /*
815 * vm_map_entry_dispose: [ internal use only ]
816 *
817 * Inverse of vm_map_entry_create.
818 */
819 static void
820 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
821 {
822 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
823 }
824
825 /*
826 * vm_map_entry_create: [ internal use only ]
827 *
828 * Allocates a VM map entry for insertion.
829 * No entry fields are filled in.
830 */
831 static vm_map_entry_t
832 vm_map_entry_create(vm_map_t map)
833 {
834 vm_map_entry_t new_entry;
835
836 if (map->system_map)
837 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
838 else
839 new_entry = uma_zalloc(mapentzone, M_WAITOK);
840 if (new_entry == NULL)
841 panic("vm_map_entry_create: kernel resources exhausted");
842 return (new_entry);
843 }
844
845 /*
846 * vm_map_entry_set_behavior:
847 *
848 * Set the expected access behavior, either normal, random, or
849 * sequential.
850 */
851 static inline void
852 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
853 {
854 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
855 (behavior & MAP_ENTRY_BEHAV_MASK);
856 }
857
858 /*
859 * vm_map_entry_set_max_free:
860 *
861 * Set the max_free field in a vm_map_entry.
862 */
863 static inline void
864 vm_map_entry_set_max_free(vm_map_entry_t entry)
865 {
866
867 entry->max_free = entry->adj_free;
868 if (entry->left != NULL && entry->left->max_free > entry->max_free)
869 entry->max_free = entry->left->max_free;
870 if (entry->right != NULL && entry->right->max_free > entry->max_free)
871 entry->max_free = entry->right->max_free;
872 }
873
874 /*
875 * vm_map_entry_splay:
876 *
877 * The Sleator and Tarjan top-down splay algorithm with the
878 * following variation. Max_free must be computed bottom-up, so
879 * on the downward pass, maintain the left and right spines in
880 * reverse order. Then, make a second pass up each side to fix
881 * the pointers and compute max_free. The time bound is O(log n)
882 * amortized.
883 *
884 * The new root is the vm_map_entry containing "addr", or else an
885 * adjacent entry (lower or higher) if addr is not in the tree.
886 *
887 * The map must be locked, and leaves it so.
888 *
889 * Returns: the new root.
890 */
891 static vm_map_entry_t
892 vm_map_entry_splay(vm_offset_t addr, vm_map_entry_t root)
893 {
894 vm_map_entry_t llist, rlist;
895 vm_map_entry_t ltree, rtree;
896 vm_map_entry_t y;
897
898 /* Special case of empty tree. */
899 if (root == NULL)
900 return (root);
901
902 /*
903 * Pass One: Splay down the tree until we find addr or a NULL
904 * pointer where addr would go. llist and rlist are the two
905 * sides in reverse order (bottom-up), with llist linked by
906 * the right pointer and rlist linked by the left pointer in
907 * the vm_map_entry. Wait until Pass Two to set max_free on
908 * the two spines.
909 */
910 llist = NULL;
911 rlist = NULL;
912 for (;;) {
913 /* root is never NULL in here. */
914 if (addr < root->start) {
915 y = root->left;
916 if (y == NULL)
917 break;
918 if (addr < y->start && y->left != NULL) {
919 /* Rotate right and put y on rlist. */
920 root->left = y->right;
921 y->right = root;
922 vm_map_entry_set_max_free(root);
923 root = y->left;
924 y->left = rlist;
925 rlist = y;
926 } else {
927 /* Put root on rlist. */
928 root->left = rlist;
929 rlist = root;
930 root = y;
931 }
932 } else if (addr >= root->end) {
933 y = root->right;
934 if (y == NULL)
935 break;
936 if (addr >= y->end && y->right != NULL) {
937 /* Rotate left and put y on llist. */
938 root->right = y->left;
939 y->left = root;
940 vm_map_entry_set_max_free(root);
941 root = y->right;
942 y->right = llist;
943 llist = y;
944 } else {
945 /* Put root on llist. */
946 root->right = llist;
947 llist = root;
948 root = y;
949 }
950 } else
951 break;
952 }
953
954 /*
955 * Pass Two: Walk back up the two spines, flip the pointers
956 * and set max_free. The subtrees of the root go at the
957 * bottom of llist and rlist.
958 */
959 ltree = root->left;
960 while (llist != NULL) {
961 y = llist->right;
962 llist->right = ltree;
963 vm_map_entry_set_max_free(llist);
964 ltree = llist;
965 llist = y;
966 }
967 rtree = root->right;
968 while (rlist != NULL) {
969 y = rlist->left;
970 rlist->left = rtree;
971 vm_map_entry_set_max_free(rlist);
972 rtree = rlist;
973 rlist = y;
974 }
975
976 /*
977 * Final assembly: add ltree and rtree as subtrees of root.
978 */
979 root->left = ltree;
980 root->right = rtree;
981 vm_map_entry_set_max_free(root);
982
983 return (root);
984 }
985
986 /*
987 * vm_map_entry_{un,}link:
988 *
989 * Insert/remove entries from maps.
990 */
991 static void
992 vm_map_entry_link(vm_map_t map,
993 vm_map_entry_t after_where,
994 vm_map_entry_t entry)
995 {
996
997 CTR4(KTR_VM,
998 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
999 map->nentries, entry, after_where);
1000 VM_MAP_ASSERT_LOCKED(map);
1001 KASSERT(after_where == &map->header ||
1002 after_where->end <= entry->start,
1003 ("vm_map_entry_link: prev end %jx new start %jx overlap",
1004 (uintmax_t)after_where->end, (uintmax_t)entry->start));
1005 KASSERT(after_where->next == &map->header ||
1006 entry->end <= after_where->next->start,
1007 ("vm_map_entry_link: new end %jx next start %jx overlap",
1008 (uintmax_t)entry->end, (uintmax_t)after_where->next->start));
1009
1010 map->nentries++;
1011 entry->prev = after_where;
1012 entry->next = after_where->next;
1013 entry->next->prev = entry;
1014 after_where->next = entry;
1015
1016 if (after_where != &map->header) {
1017 if (after_where != map->root)
1018 vm_map_entry_splay(after_where->start, map->root);
1019 entry->right = after_where->right;
1020 entry->left = after_where;
1021 after_where->right = NULL;
1022 after_where->adj_free = entry->start - after_where->end;
1023 vm_map_entry_set_max_free(after_where);
1024 } else {
1025 entry->right = map->root;
1026 entry->left = NULL;
1027 }
1028 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1029 entry->next->start) - entry->end;
1030 vm_map_entry_set_max_free(entry);
1031 map->root = entry;
1032 }
1033
1034 static void
1035 vm_map_entry_unlink(vm_map_t map,
1036 vm_map_entry_t entry)
1037 {
1038 vm_map_entry_t next, prev, root;
1039
1040 VM_MAP_ASSERT_LOCKED(map);
1041 if (entry != map->root)
1042 vm_map_entry_splay(entry->start, map->root);
1043 if (entry->left == NULL)
1044 root = entry->right;
1045 else {
1046 root = vm_map_entry_splay(entry->start, entry->left);
1047 root->right = entry->right;
1048 root->adj_free = (entry->next == &map->header ? map->max_offset :
1049 entry->next->start) - root->end;
1050 vm_map_entry_set_max_free(root);
1051 }
1052 map->root = root;
1053
1054 prev = entry->prev;
1055 next = entry->next;
1056 next->prev = prev;
1057 prev->next = next;
1058 map->nentries--;
1059 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1060 map->nentries, entry);
1061 }
1062
1063 /*
1064 * vm_map_entry_resize_free:
1065 *
1066 * Recompute the amount of free space following a vm_map_entry
1067 * and propagate that value up the tree. Call this function after
1068 * resizing a map entry in-place, that is, without a call to
1069 * vm_map_entry_link() or _unlink().
1070 *
1071 * The map must be locked, and leaves it so.
1072 */
1073 static void
1074 vm_map_entry_resize_free(vm_map_t map, vm_map_entry_t entry)
1075 {
1076
1077 /*
1078 * Using splay trees without parent pointers, propagating
1079 * max_free up the tree is done by moving the entry to the
1080 * root and making the change there.
1081 */
1082 if (entry != map->root)
1083 map->root = vm_map_entry_splay(entry->start, map->root);
1084
1085 entry->adj_free = (entry->next == &map->header ? map->max_offset :
1086 entry->next->start) - entry->end;
1087 vm_map_entry_set_max_free(entry);
1088 }
1089
1090 /*
1091 * vm_map_lookup_entry: [ internal use only ]
1092 *
1093 * Finds the map entry containing (or
1094 * immediately preceding) the specified address
1095 * in the given map; the entry is returned
1096 * in the "entry" parameter. The boolean
1097 * result indicates whether the address is
1098 * actually contained in the map.
1099 */
1100 boolean_t
1101 vm_map_lookup_entry(
1102 vm_map_t map,
1103 vm_offset_t address,
1104 vm_map_entry_t *entry) /* OUT */
1105 {
1106 vm_map_entry_t cur;
1107 boolean_t locked;
1108
1109 /*
1110 * If the map is empty, then the map entry immediately preceding
1111 * "address" is the map's header.
1112 */
1113 cur = map->root;
1114 if (cur == NULL)
1115 *entry = &map->header;
1116 else if (address >= cur->start && cur->end > address) {
1117 *entry = cur;
1118 return (TRUE);
1119 } else if ((locked = vm_map_locked(map)) ||
1120 sx_try_upgrade(&map->lock)) {
1121 /*
1122 * Splay requires a write lock on the map. However, it only
1123 * restructures the binary search tree; it does not otherwise
1124 * change the map. Thus, the map's timestamp need not change
1125 * on a temporary upgrade.
1126 */
1127 map->root = cur = vm_map_entry_splay(address, cur);
1128 if (!locked)
1129 sx_downgrade(&map->lock);
1130
1131 /*
1132 * If "address" is contained within a map entry, the new root
1133 * is that map entry. Otherwise, the new root is a map entry
1134 * immediately before or after "address".
1135 */
1136 if (address >= cur->start) {
1137 *entry = cur;
1138 if (cur->end > address)
1139 return (TRUE);
1140 } else
1141 *entry = cur->prev;
1142 } else
1143 /*
1144 * Since the map is only locked for read access, perform a
1145 * standard binary search tree lookup for "address".
1146 */
1147 for (;;) {
1148 if (address < cur->start) {
1149 if (cur->left == NULL) {
1150 *entry = cur->prev;
1151 break;
1152 }
1153 cur = cur->left;
1154 } else if (cur->end > address) {
1155 *entry = cur;
1156 return (TRUE);
1157 } else {
1158 if (cur->right == NULL) {
1159 *entry = cur;
1160 break;
1161 }
1162 cur = cur->right;
1163 }
1164 }
1165 return (FALSE);
1166 }
1167
1168 /*
1169 * vm_map_insert:
1170 *
1171 * Inserts the given whole VM object into the target
1172 * map at the specified address range. The object's
1173 * size should match that of the address range.
1174 *
1175 * Requires that the map be locked, and leaves it so.
1176 *
1177 * If object is non-NULL, ref count must be bumped by caller
1178 * prior to making call to account for the new entry.
1179 */
1180 int
1181 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1182 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1183 {
1184 vm_map_entry_t new_entry, prev_entry, temp_entry;
1185 struct ucred *cred;
1186 vm_eflags_t protoeflags;
1187 vm_inherit_t inheritance;
1188
1189 VM_MAP_ASSERT_LOCKED(map);
1190 KASSERT((object != kmem_object && object != kernel_object) ||
1191 (cow & MAP_COPY_ON_WRITE) == 0,
1192 ("vm_map_insert: kmem or kernel object and COW"));
1193 KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1194 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1195 KASSERT((prot & ~max) == 0,
1196 ("prot %#x is not subset of max_prot %#x", prot, max));
1197
1198 /*
1199 * Check that the start and end points are not bogus.
1200 */
1201 if (start < map->min_offset || end > map->max_offset || start >= end)
1202 return (KERN_INVALID_ADDRESS);
1203
1204 /*
1205 * Find the entry prior to the proposed starting address; if it's part
1206 * of an existing entry, this range is bogus.
1207 */
1208 if (vm_map_lookup_entry(map, start, &temp_entry))
1209 return (KERN_NO_SPACE);
1210
1211 prev_entry = temp_entry;
1212
1213 /*
1214 * Assert that the next entry doesn't overlap the end point.
1215 */
1216 if (prev_entry->next != &map->header && prev_entry->next->start < end)
1217 return (KERN_NO_SPACE);
1218
1219 if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1220 max != VM_PROT_NONE))
1221 return (KERN_INVALID_ARGUMENT);
1222
1223 protoeflags = 0;
1224 if (cow & MAP_COPY_ON_WRITE)
1225 protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1226 if (cow & MAP_NOFAULT)
1227 protoeflags |= MAP_ENTRY_NOFAULT;
1228 if (cow & MAP_DISABLE_SYNCER)
1229 protoeflags |= MAP_ENTRY_NOSYNC;
1230 if (cow & MAP_DISABLE_COREDUMP)
1231 protoeflags |= MAP_ENTRY_NOCOREDUMP;
1232 if (cow & MAP_STACK_GROWS_DOWN)
1233 protoeflags |= MAP_ENTRY_GROWS_DOWN;
1234 if (cow & MAP_STACK_GROWS_UP)
1235 protoeflags |= MAP_ENTRY_GROWS_UP;
1236 if (cow & MAP_VN_WRITECOUNT)
1237 protoeflags |= MAP_ENTRY_VN_WRITECNT;
1238 if ((cow & MAP_CREATE_GUARD) != 0)
1239 protoeflags |= MAP_ENTRY_GUARD;
1240 if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1241 protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1242 if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1243 protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1244 if (cow & MAP_INHERIT_SHARE)
1245 inheritance = VM_INHERIT_SHARE;
1246 else
1247 inheritance = VM_INHERIT_DEFAULT;
1248
1249 cred = NULL;
1250 if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1251 goto charged;
1252 if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1253 ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1254 if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1255 return (KERN_RESOURCE_SHORTAGE);
1256 KASSERT(object == NULL ||
1257 (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1258 object->cred == NULL,
1259 ("overcommit: vm_map_insert o %p", object));
1260 cred = curthread->td_ucred;
1261 }
1262
1263 charged:
1264 /* Expand the kernel pmap, if necessary. */
1265 if (map == kernel_map && end > kernel_vm_end)
1266 pmap_growkernel(end);
1267 if (object != NULL) {
1268 /*
1269 * OBJ_ONEMAPPING must be cleared unless this mapping
1270 * is trivially proven to be the only mapping for any
1271 * of the object's pages. (Object granularity
1272 * reference counting is insufficient to recognize
1273 * aliases with precision.)
1274 */
1275 VM_OBJECT_WLOCK(object);
1276 if (object->ref_count > 1 || object->shadow_count != 0)
1277 vm_object_clear_flag(object, OBJ_ONEMAPPING);
1278 VM_OBJECT_WUNLOCK(object);
1279 } else if (prev_entry != &map->header &&
1280 prev_entry->eflags == protoeflags &&
1281 (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 &&
1282 prev_entry->end == start && prev_entry->wired_count == 0 &&
1283 (prev_entry->cred == cred ||
1284 (prev_entry->object.vm_object != NULL &&
1285 prev_entry->object.vm_object->cred == cred)) &&
1286 vm_object_coalesce(prev_entry->object.vm_object,
1287 prev_entry->offset,
1288 (vm_size_t)(prev_entry->end - prev_entry->start),
1289 (vm_size_t)(end - prev_entry->end), cred != NULL &&
1290 (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1291 /*
1292 * We were able to extend the object. Determine if we
1293 * can extend the previous map entry to include the
1294 * new range as well.
1295 */
1296 if (prev_entry->inheritance == inheritance &&
1297 prev_entry->protection == prot &&
1298 prev_entry->max_protection == max) {
1299 if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1300 map->size += end - prev_entry->end;
1301 prev_entry->end = end;
1302 vm_map_entry_resize_free(map, prev_entry);
1303 vm_map_simplify_entry(map, prev_entry);
1304 return (KERN_SUCCESS);
1305 }
1306
1307 /*
1308 * If we can extend the object but cannot extend the
1309 * map entry, we have to create a new map entry. We
1310 * must bump the ref count on the extended object to
1311 * account for it. object may be NULL.
1312 */
1313 object = prev_entry->object.vm_object;
1314 offset = prev_entry->offset +
1315 (prev_entry->end - prev_entry->start);
1316 vm_object_reference(object);
1317 if (cred != NULL && object != NULL && object->cred != NULL &&
1318 !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1319 /* Object already accounts for this uid. */
1320 cred = NULL;
1321 }
1322 }
1323 if (cred != NULL)
1324 crhold(cred);
1325
1326 /*
1327 * Create a new entry
1328 */
1329 new_entry = vm_map_entry_create(map);
1330 new_entry->start = start;
1331 new_entry->end = end;
1332 new_entry->cred = NULL;
1333
1334 new_entry->eflags = protoeflags;
1335 new_entry->object.vm_object = object;
1336 new_entry->offset = offset;
1337
1338 new_entry->inheritance = inheritance;
1339 new_entry->protection = prot;
1340 new_entry->max_protection = max;
1341 new_entry->wired_count = 0;
1342 new_entry->wiring_thread = NULL;
1343 new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1344 new_entry->next_read = start;
1345
1346 KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1347 ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1348 new_entry->cred = cred;
1349
1350 /*
1351 * Insert the new entry into the list
1352 */
1353 vm_map_entry_link(map, prev_entry, new_entry);
1354 if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1355 map->size += new_entry->end - new_entry->start;
1356
1357 /*
1358 * Try to coalesce the new entry with both the previous and next
1359 * entries in the list. Previously, we only attempted to coalesce
1360 * with the previous entry when object is NULL. Here, we handle the
1361 * other cases, which are less common.
1362 */
1363 vm_map_simplify_entry(map, new_entry);
1364
1365 if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1366 vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1367 end - start, cow & MAP_PREFAULT_PARTIAL);
1368 }
1369
1370 return (KERN_SUCCESS);
1371 }
1372
1373 /*
1374 * vm_map_findspace:
1375 *
1376 * Find the first fit (lowest VM address) for "length" free bytes
1377 * beginning at address >= start in the given map.
1378 *
1379 * In a vm_map_entry, "adj_free" is the amount of free space
1380 * adjacent (higher address) to this entry, and "max_free" is the
1381 * maximum amount of contiguous free space in its subtree. This
1382 * allows finding a free region in one path down the tree, so
1383 * O(log n) amortized with splay trees.
1384 *
1385 * The map must be locked, and leaves it so.
1386 *
1387 * Returns: 0 on success, and starting address in *addr,
1388 * 1 if insufficient space.
1389 */
1390 int
1391 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1392 vm_offset_t *addr) /* OUT */
1393 {
1394 vm_map_entry_t entry;
1395 vm_offset_t st;
1396
1397 /*
1398 * Request must fit within min/max VM address and must avoid
1399 * address wrap.
1400 */
1401 if (start < map->min_offset)
1402 start = map->min_offset;
1403 if (start + length > map->max_offset || start + length < start)
1404 return (1);
1405
1406 /* Empty tree means wide open address space. */
1407 if (map->root == NULL) {
1408 *addr = start;
1409 return (0);
1410 }
1411
1412 /*
1413 * After splay, if start comes before root node, then there
1414 * must be a gap from start to the root.
1415 */
1416 map->root = vm_map_entry_splay(start, map->root);
1417 if (start + length <= map->root->start) {
1418 *addr = start;
1419 return (0);
1420 }
1421
1422 /*
1423 * Root is the last node that might begin its gap before
1424 * start, and this is the last comparison where address
1425 * wrap might be a problem.
1426 */
1427 st = (start > map->root->end) ? start : map->root->end;
1428 if (length <= map->root->end + map->root->adj_free - st) {
1429 *addr = st;
1430 return (0);
1431 }
1432
1433 /* With max_free, can immediately tell if no solution. */
1434 entry = map->root->right;
1435 if (entry == NULL || length > entry->max_free)
1436 return (1);
1437
1438 /*
1439 * Search the right subtree in the order: left subtree, root,
1440 * right subtree (first fit). The previous splay implies that
1441 * all regions in the right subtree have addresses > start.
1442 */
1443 while (entry != NULL) {
1444 if (entry->left != NULL && entry->left->max_free >= length)
1445 entry = entry->left;
1446 else if (entry->adj_free >= length) {
1447 *addr = entry->end;
1448 return (0);
1449 } else
1450 entry = entry->right;
1451 }
1452
1453 /* Can't get here, so panic if we do. */
1454 panic("vm_map_findspace: max_free corrupt");
1455 }
1456
1457 int
1458 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1459 vm_offset_t start, vm_size_t length, vm_prot_t prot,
1460 vm_prot_t max, int cow)
1461 {
1462 vm_offset_t end;
1463 int result;
1464
1465 end = start + length;
1466 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1467 object == NULL,
1468 ("vm_map_fixed: non-NULL backing object for stack"));
1469 vm_map_lock(map);
1470 VM_MAP_RANGE_CHECK(map, start, end);
1471 if ((cow & MAP_CHECK_EXCL) == 0)
1472 vm_map_delete(map, start, end);
1473 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1474 result = vm_map_stack_locked(map, start, length, sgrowsiz,
1475 prot, max, cow);
1476 } else {
1477 result = vm_map_insert(map, object, offset, start, end,
1478 prot, max, cow);
1479 }
1480 vm_map_unlock(map);
1481 return (result);
1482 }
1483
1484 /*
1485 * vm_map_find finds an unallocated region in the target address
1486 * map with the given length. The search is defined to be
1487 * first-fit from the specified address; the region found is
1488 * returned in the same parameter.
1489 *
1490 * If object is non-NULL, ref count must be bumped by caller
1491 * prior to making call to account for the new entry.
1492 */
1493 int
1494 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1495 vm_offset_t *addr, /* IN/OUT */
1496 vm_size_t length, vm_offset_t max_addr, int find_space,
1497 vm_prot_t prot, vm_prot_t max, int cow)
1498 {
1499 vm_offset_t alignment, initial_addr, start;
1500 int result;
1501
1502 KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1503 object == NULL,
1504 ("vm_map_find: non-NULL backing object for stack"));
1505 if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
1506 (object->flags & OBJ_COLORED) == 0))
1507 find_space = VMFS_ANY_SPACE;
1508 if (find_space >> 8 != 0) {
1509 KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
1510 alignment = (vm_offset_t)1 << (find_space >> 8);
1511 } else
1512 alignment = 0;
1513 initial_addr = *addr;
1514 again:
1515 start = initial_addr;
1516 vm_map_lock(map);
1517 do {
1518 if (find_space != VMFS_NO_SPACE) {
1519 if (vm_map_findspace(map, start, length, addr) ||
1520 (max_addr != 0 && *addr + length > max_addr)) {
1521 vm_map_unlock(map);
1522 if (find_space == VMFS_OPTIMAL_SPACE) {
1523 find_space = VMFS_ANY_SPACE;
1524 goto again;
1525 }
1526 return (KERN_NO_SPACE);
1527 }
1528 switch (find_space) {
1529 case VMFS_SUPER_SPACE:
1530 case VMFS_OPTIMAL_SPACE:
1531 pmap_align_superpage(object, offset, addr,
1532 length);
1533 break;
1534 case VMFS_ANY_SPACE:
1535 break;
1536 default:
1537 if ((*addr & (alignment - 1)) != 0) {
1538 *addr &= ~(alignment - 1);
1539 *addr += alignment;
1540 }
1541 break;
1542 }
1543
1544 start = *addr;
1545 }
1546 if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1547 result = vm_map_stack_locked(map, start, length,
1548 sgrowsiz, prot, max, cow);
1549 } else {
1550 result = vm_map_insert(map, object, offset, start,
1551 start + length, prot, max, cow);
1552 }
1553 } while (result == KERN_NO_SPACE && find_space != VMFS_NO_SPACE &&
1554 find_space != VMFS_ANY_SPACE);
1555 vm_map_unlock(map);
1556 return (result);
1557 }
1558
1559 int
1560 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1561 vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
1562 vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
1563 int cow)
1564 {
1565 vm_offset_t hint;
1566 int rv;
1567
1568 hint = *addr;
1569 for (;;) {
1570 rv = vm_map_find(map, object, offset, addr, length, max_addr,
1571 find_space, prot, max, cow);
1572 if (rv == KERN_SUCCESS || min_addr >= hint)
1573 return (rv);
1574 *addr = hint = min_addr;
1575 }
1576 }
1577
1578 /*
1579 * vm_map_simplify_entry:
1580 *
1581 * Simplify the given map entry by merging with either neighbor. This
1582 * routine also has the ability to merge with both neighbors.
1583 *
1584 * The map must be locked.
1585 *
1586 * This routine guarantees that the passed entry remains valid (though
1587 * possibly extended). When merging, this routine may delete one or
1588 * both neighbors.
1589 */
1590 void
1591 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1592 {
1593 vm_map_entry_t next, prev;
1594 vm_size_t prevsize, esize;
1595
1596 if ((entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP |
1597 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) != 0)
1598 return;
1599
1600 prev = entry->prev;
1601 if (prev != &map->header) {
1602 prevsize = prev->end - prev->start;
1603 if ( (prev->end == entry->start) &&
1604 (prev->object.vm_object == entry->object.vm_object) &&
1605 (!prev->object.vm_object ||
1606 (prev->offset + prevsize == entry->offset)) &&
1607 (prev->eflags == entry->eflags) &&
1608 (prev->protection == entry->protection) &&
1609 (prev->max_protection == entry->max_protection) &&
1610 (prev->inheritance == entry->inheritance) &&
1611 (prev->wired_count == entry->wired_count) &&
1612 (prev->cred == entry->cred)) {
1613 vm_map_entry_unlink(map, prev);
1614 entry->start = prev->start;
1615 entry->offset = prev->offset;
1616 if (entry->prev != &map->header)
1617 vm_map_entry_resize_free(map, entry->prev);
1618
1619 /*
1620 * If the backing object is a vnode object,
1621 * vm_object_deallocate() calls vrele().
1622 * However, vrele() does not lock the vnode
1623 * because the vnode has additional
1624 * references. Thus, the map lock can be kept
1625 * without causing a lock-order reversal with
1626 * the vnode lock.
1627 *
1628 * Since we count the number of virtual page
1629 * mappings in object->un_pager.vnp.writemappings,
1630 * the writemappings value should not be adjusted
1631 * when the entry is disposed of.
1632 */
1633 if (prev->object.vm_object)
1634 vm_object_deallocate(prev->object.vm_object);
1635 if (prev->cred != NULL)
1636 crfree(prev->cred);
1637 vm_map_entry_dispose(map, prev);
1638 }
1639 }
1640
1641 next = entry->next;
1642 if (next != &map->header) {
1643 esize = entry->end - entry->start;
1644 if ((entry->end == next->start) &&
1645 (next->object.vm_object == entry->object.vm_object) &&
1646 (!entry->object.vm_object ||
1647 (entry->offset + esize == next->offset)) &&
1648 (next->eflags == entry->eflags) &&
1649 (next->protection == entry->protection) &&
1650 (next->max_protection == entry->max_protection) &&
1651 (next->inheritance == entry->inheritance) &&
1652 (next->wired_count == entry->wired_count) &&
1653 (next->cred == entry->cred)) {
1654 vm_map_entry_unlink(map, next);
1655 entry->end = next->end;
1656 vm_map_entry_resize_free(map, entry);
1657
1658 /*
1659 * See comment above.
1660 */
1661 if (next->object.vm_object)
1662 vm_object_deallocate(next->object.vm_object);
1663 if (next->cred != NULL)
1664 crfree(next->cred);
1665 vm_map_entry_dispose(map, next);
1666 }
1667 }
1668 }
1669 /*
1670 * vm_map_clip_start: [ internal use only ]
1671 *
1672 * Asserts that the given entry begins at or after
1673 * the specified address; if necessary,
1674 * it splits the entry into two.
1675 */
1676 #define vm_map_clip_start(map, entry, startaddr) \
1677 { \
1678 if (startaddr > entry->start) \
1679 _vm_map_clip_start(map, entry, startaddr); \
1680 }
1681
1682 /*
1683 * This routine is called only when it is known that
1684 * the entry must be split.
1685 */
1686 static void
1687 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1688 {
1689 vm_map_entry_t new_entry;
1690
1691 VM_MAP_ASSERT_LOCKED(map);
1692 KASSERT(entry->end > start && entry->start < start,
1693 ("_vm_map_clip_start: invalid clip of entry %p", entry));
1694
1695 /*
1696 * Split off the front portion -- note that we must insert the new
1697 * entry BEFORE this one, so that this entry has the specified
1698 * starting address.
1699 */
1700 vm_map_simplify_entry(map, entry);
1701
1702 /*
1703 * If there is no object backing this entry, we might as well create
1704 * one now. If we defer it, an object can get created after the map
1705 * is clipped, and individual objects will be created for the split-up
1706 * map. This is a bit of a hack, but is also about the best place to
1707 * put this improvement.
1708 */
1709 if (entry->object.vm_object == NULL && !map->system_map &&
1710 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1711 vm_object_t object;
1712 object = vm_object_allocate(OBJT_DEFAULT,
1713 atop(entry->end - entry->start));
1714 entry->object.vm_object = object;
1715 entry->offset = 0;
1716 if (entry->cred != NULL) {
1717 object->cred = entry->cred;
1718 object->charge = entry->end - entry->start;
1719 entry->cred = NULL;
1720 }
1721 } else if (entry->object.vm_object != NULL &&
1722 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1723 entry->cred != NULL) {
1724 VM_OBJECT_WLOCK(entry->object.vm_object);
1725 KASSERT(entry->object.vm_object->cred == NULL,
1726 ("OVERCOMMIT: vm_entry_clip_start: both cred e %p", entry));
1727 entry->object.vm_object->cred = entry->cred;
1728 entry->object.vm_object->charge = entry->end - entry->start;
1729 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1730 entry->cred = NULL;
1731 }
1732
1733 new_entry = vm_map_entry_create(map);
1734 *new_entry = *entry;
1735
1736 new_entry->end = start;
1737 entry->offset += (start - entry->start);
1738 entry->start = start;
1739 if (new_entry->cred != NULL)
1740 crhold(entry->cred);
1741
1742 vm_map_entry_link(map, entry->prev, new_entry);
1743
1744 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1745 vm_object_reference(new_entry->object.vm_object);
1746 /*
1747 * The object->un_pager.vnp.writemappings for the
1748 * object of MAP_ENTRY_VN_WRITECNT type entry shall be
1749 * kept as is here. The virtual pages are
1750 * re-distributed among the clipped entries, so the sum is
1751 * left the same.
1752 */
1753 }
1754 }
1755
1756 /*
1757 * vm_map_clip_end: [ internal use only ]
1758 *
1759 * Asserts that the given entry ends at or before
1760 * the specified address; if necessary,
1761 * it splits the entry into two.
1762 */
1763 #define vm_map_clip_end(map, entry, endaddr) \
1764 { \
1765 if ((endaddr) < (entry->end)) \
1766 _vm_map_clip_end((map), (entry), (endaddr)); \
1767 }
1768
1769 /*
1770 * This routine is called only when it is known that
1771 * the entry must be split.
1772 */
1773 static void
1774 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1775 {
1776 vm_map_entry_t new_entry;
1777
1778 VM_MAP_ASSERT_LOCKED(map);
1779 KASSERT(entry->start < end && entry->end > end,
1780 ("_vm_map_clip_end: invalid clip of entry %p", entry));
1781
1782 /*
1783 * If there is no object backing this entry, we might as well create
1784 * one now. If we defer it, an object can get created after the map
1785 * is clipped, and individual objects will be created for the split-up
1786 * map. This is a bit of a hack, but is also about the best place to
1787 * put this improvement.
1788 */
1789 if (entry->object.vm_object == NULL && !map->system_map &&
1790 (entry->eflags & MAP_ENTRY_GUARD) == 0) {
1791 vm_object_t object;
1792 object = vm_object_allocate(OBJT_DEFAULT,
1793 atop(entry->end - entry->start));
1794 entry->object.vm_object = object;
1795 entry->offset = 0;
1796 if (entry->cred != NULL) {
1797 object->cred = entry->cred;
1798 object->charge = entry->end - entry->start;
1799 entry->cred = NULL;
1800 }
1801 } else if (entry->object.vm_object != NULL &&
1802 ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
1803 entry->cred != NULL) {
1804 VM_OBJECT_WLOCK(entry->object.vm_object);
1805 KASSERT(entry->object.vm_object->cred == NULL,
1806 ("OVERCOMMIT: vm_entry_clip_end: both cred e %p", entry));
1807 entry->object.vm_object->cred = entry->cred;
1808 entry->object.vm_object->charge = entry->end - entry->start;
1809 VM_OBJECT_WUNLOCK(entry->object.vm_object);
1810 entry->cred = NULL;
1811 }
1812
1813 /*
1814 * Create a new entry and insert it AFTER the specified entry
1815 */
1816 new_entry = vm_map_entry_create(map);
1817 *new_entry = *entry;
1818
1819 new_entry->start = entry->end = end;
1820 new_entry->offset += (end - entry->start);
1821 if (new_entry->cred != NULL)
1822 crhold(entry->cred);
1823
1824 vm_map_entry_link(map, entry, new_entry);
1825
1826 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1827 vm_object_reference(new_entry->object.vm_object);
1828 }
1829 }
1830
1831 /*
1832 * vm_map_submap: [ kernel use only ]
1833 *
1834 * Mark the given range as handled by a subordinate map.
1835 *
1836 * This range must have been created with vm_map_find,
1837 * and no other operations may have been performed on this
1838 * range prior to calling vm_map_submap.
1839 *
1840 * Only a limited number of operations can be performed
1841 * within this rage after calling vm_map_submap:
1842 * vm_fault
1843 * [Don't try vm_map_copy!]
1844 *
1845 * To remove a submapping, one must first remove the
1846 * range from the superior map, and then destroy the
1847 * submap (if desired). [Better yet, don't try it.]
1848 */
1849 int
1850 vm_map_submap(
1851 vm_map_t map,
1852 vm_offset_t start,
1853 vm_offset_t end,
1854 vm_map_t submap)
1855 {
1856 vm_map_entry_t entry;
1857 int result = KERN_INVALID_ARGUMENT;
1858
1859 vm_map_lock(map);
1860
1861 VM_MAP_RANGE_CHECK(map, start, end);
1862
1863 if (vm_map_lookup_entry(map, start, &entry)) {
1864 vm_map_clip_start(map, entry, start);
1865 } else
1866 entry = entry->next;
1867
1868 vm_map_clip_end(map, entry, end);
1869
1870 if ((entry->start == start) && (entry->end == end) &&
1871 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1872 (entry->object.vm_object == NULL)) {
1873 entry->object.sub_map = submap;
1874 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1875 result = KERN_SUCCESS;
1876 }
1877 vm_map_unlock(map);
1878
1879 return (result);
1880 }
1881
1882 /*
1883 * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
1884 */
1885 #define MAX_INIT_PT 96
1886
1887 /*
1888 * vm_map_pmap_enter:
1889 *
1890 * Preload the specified map's pmap with mappings to the specified
1891 * object's memory-resident pages. No further physical pages are
1892 * allocated, and no further virtual pages are retrieved from secondary
1893 * storage. If the specified flags include MAP_PREFAULT_PARTIAL, then a
1894 * limited number of page mappings are created at the low-end of the
1895 * specified address range. (For this purpose, a superpage mapping
1896 * counts as one page mapping.) Otherwise, all resident pages within
1897 * the specified address range are mapped.
1898 */
1899 static void
1900 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
1901 vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
1902 {
1903 vm_offset_t start;
1904 vm_page_t p, p_start;
1905 vm_pindex_t mask, psize, threshold, tmpidx;
1906
1907 if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
1908 return;
1909 VM_OBJECT_RLOCK(object);
1910 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1911 VM_OBJECT_RUNLOCK(object);
1912 VM_OBJECT_WLOCK(object);
1913 if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
1914 pmap_object_init_pt(map->pmap, addr, object, pindex,
1915 size);
1916 VM_OBJECT_WUNLOCK(object);
1917 return;
1918 }
1919 VM_OBJECT_LOCK_DOWNGRADE(object);
1920 }
1921
1922 psize = atop(size);
1923 if (psize + pindex > object->size) {
1924 if (object->size < pindex) {
1925 VM_OBJECT_RUNLOCK(object);
1926 return;
1927 }
1928 psize = object->size - pindex;
1929 }
1930
1931 start = 0;
1932 p_start = NULL;
1933 threshold = MAX_INIT_PT;
1934
1935 p = vm_page_find_least(object, pindex);
1936 /*
1937 * Assert: the variable p is either (1) the page with the
1938 * least pindex greater than or equal to the parameter pindex
1939 * or (2) NULL.
1940 */
1941 for (;
1942 p != NULL && (tmpidx = p->pindex - pindex) < psize;
1943 p = TAILQ_NEXT(p, listq)) {
1944 /*
1945 * don't allow an madvise to blow away our really
1946 * free pages allocating pv entries.
1947 */
1948 if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
1949 vm_cnt.v_free_count < vm_cnt.v_free_reserved) ||
1950 ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
1951 tmpidx >= threshold)) {
1952 psize = tmpidx;
1953 break;
1954 }
1955 if (p->valid == VM_PAGE_BITS_ALL) {
1956 if (p_start == NULL) {
1957 start = addr + ptoa(tmpidx);
1958 p_start = p;
1959 }
1960 /* Jump ahead if a superpage mapping is possible. */
1961 if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
1962 (pagesizes[p->psind] - 1)) == 0) {
1963 mask = atop(pagesizes[p->psind]) - 1;
1964 if (tmpidx + mask < psize &&
1965 vm_page_ps_is_valid(p)) {
1966 p += mask;
1967 threshold += mask;
1968 }
1969 }
1970 } else if (p_start != NULL) {
1971 pmap_enter_object(map->pmap, start, addr +
1972 ptoa(tmpidx), p_start, prot);
1973 p_start = NULL;
1974 }
1975 }
1976 if (p_start != NULL)
1977 pmap_enter_object(map->pmap, start, addr + ptoa(psize),
1978 p_start, prot);
1979 VM_OBJECT_RUNLOCK(object);
1980 }
1981
1982 /*
1983 * vm_map_protect:
1984 *
1985 * Sets the protection of the specified address
1986 * region in the target map. If "set_max" is
1987 * specified, the maximum protection is to be set;
1988 * otherwise, only the current protection is affected.
1989 */
1990 int
1991 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1992 vm_prot_t new_prot, boolean_t set_max)
1993 {
1994 vm_map_entry_t current, entry;
1995 vm_object_t obj;
1996 struct ucred *cred;
1997 vm_prot_t old_prot;
1998
1999 if (start == end)
2000 return (KERN_SUCCESS);
2001
2002 vm_map_lock(map);
2003
2004 /*
2005 * Ensure that we are not concurrently wiring pages. vm_map_wire() may
2006 * need to fault pages into the map and will drop the map lock while
2007 * doing so, and the VM object may end up in an inconsistent state if we
2008 * update the protection on the map entry in between faults.
2009 */
2010 vm_map_wait_busy(map);
2011
2012 VM_MAP_RANGE_CHECK(map, start, end);
2013
2014 if (vm_map_lookup_entry(map, start, &entry)) {
2015 vm_map_clip_start(map, entry, start);
2016 } else {
2017 entry = entry->next;
2018 }
2019
2020 /*
2021 * Make a first pass to check for protection violations.
2022 */
2023 for (current = entry; current != &map->header && current->start < end;
2024 current = current->next) {
2025 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2026 continue;
2027 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2028 vm_map_unlock(map);
2029 return (KERN_INVALID_ARGUMENT);
2030 }
2031 if ((new_prot & current->max_protection) != new_prot) {
2032 vm_map_unlock(map);
2033 return (KERN_PROTECTION_FAILURE);
2034 }
2035 }
2036
2037 /*
2038 * Do an accounting pass for private read-only mappings that
2039 * now will do cow due to allowed write (e.g. debugger sets
2040 * breakpoint on text segment)
2041 */
2042 for (current = entry; current != &map->header && current->start < end;
2043 current = current->next) {
2044
2045 vm_map_clip_end(map, current, end);
2046
2047 if (set_max ||
2048 ((new_prot & ~(current->protection)) & VM_PROT_WRITE) == 0 ||
2049 ENTRY_CHARGED(current) ||
2050 (current->eflags & MAP_ENTRY_GUARD) != 0) {
2051 continue;
2052 }
2053
2054 cred = curthread->td_ucred;
2055 obj = current->object.vm_object;
2056
2057 if (obj == NULL || (current->eflags & MAP_ENTRY_NEEDS_COPY)) {
2058 if (!swap_reserve(current->end - current->start)) {
2059 vm_map_unlock(map);
2060 return (KERN_RESOURCE_SHORTAGE);
2061 }
2062 crhold(cred);
2063 current->cred = cred;
2064 continue;
2065 }
2066
2067 VM_OBJECT_WLOCK(obj);
2068 if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2069 VM_OBJECT_WUNLOCK(obj);
2070 continue;
2071 }
2072
2073 /*
2074 * Charge for the whole object allocation now, since
2075 * we cannot distinguish between non-charged and
2076 * charged clipped mapping of the same object later.
2077 */
2078 KASSERT(obj->charge == 0,
2079 ("vm_map_protect: object %p overcharged (entry %p)",
2080 obj, current));
2081 if (!swap_reserve(ptoa(obj->size))) {
2082 VM_OBJECT_WUNLOCK(obj);
2083 vm_map_unlock(map);
2084 return (KERN_RESOURCE_SHORTAGE);
2085 }
2086
2087 crhold(cred);
2088 obj->cred = cred;
2089 obj->charge = ptoa(obj->size);
2090 VM_OBJECT_WUNLOCK(obj);
2091 }
2092
2093 /*
2094 * Go back and fix up protections. [Note that clipping is not
2095 * necessary the second time.]
2096 */
2097 for (current = entry; current != &map->header && current->start < end;
2098 current = current->next) {
2099 if ((current->eflags & MAP_ENTRY_GUARD) != 0)
2100 continue;
2101
2102 old_prot = current->protection;
2103
2104 if (set_max)
2105 current->protection =
2106 (current->max_protection = new_prot) &
2107 old_prot;
2108 else
2109 current->protection = new_prot;
2110
2111 /*
2112 * For user wired map entries, the normal lazy evaluation of
2113 * write access upgrades through soft page faults is
2114 * undesirable. Instead, immediately copy any pages that are
2115 * copy-on-write and enable write access in the physical map.
2116 */
2117 if ((current->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2118 (current->protection & VM_PROT_WRITE) != 0 &&
2119 (old_prot & VM_PROT_WRITE) == 0)
2120 vm_fault_copy_entry(map, map, current, current, NULL);
2121
2122 /*
2123 * When restricting access, update the physical map. Worry
2124 * about copy-on-write here.
2125 */
2126 if ((old_prot & ~current->protection) != 0) {
2127 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2128 VM_PROT_ALL)
2129 pmap_protect(map->pmap, current->start,
2130 current->end,
2131 current->protection & MASK(current));
2132 #undef MASK
2133 }
2134 vm_map_simplify_entry(map, current);
2135 }
2136 vm_map_unlock(map);
2137 return (KERN_SUCCESS);
2138 }
2139
2140 /*
2141 * vm_map_madvise:
2142 *
2143 * This routine traverses a processes map handling the madvise
2144 * system call. Advisories are classified as either those effecting
2145 * the vm_map_entry structure, or those effecting the underlying
2146 * objects.
2147 */
2148 int
2149 vm_map_madvise(
2150 vm_map_t map,
2151 vm_offset_t start,
2152 vm_offset_t end,
2153 int behav)
2154 {
2155 vm_map_entry_t current, entry;
2156 int modify_map = 0;
2157
2158 /*
2159 * Some madvise calls directly modify the vm_map_entry, in which case
2160 * we need to use an exclusive lock on the map and we need to perform
2161 * various clipping operations. Otherwise we only need a read-lock
2162 * on the map.
2163 */
2164 switch(behav) {
2165 case MADV_NORMAL:
2166 case MADV_SEQUENTIAL:
2167 case MADV_RANDOM:
2168 case MADV_NOSYNC:
2169 case MADV_AUTOSYNC:
2170 case MADV_NOCORE:
2171 case MADV_CORE:
2172 if (start == end)
2173 return (KERN_SUCCESS);
2174 modify_map = 1;
2175 vm_map_lock(map);
2176 break;
2177 case MADV_WILLNEED:
2178 case MADV_DONTNEED:
2179 case MADV_FREE:
2180 if (start == end)
2181 return (KERN_SUCCESS);
2182 vm_map_lock_read(map);
2183 break;
2184 default:
2185 return (KERN_INVALID_ARGUMENT);
2186 }
2187
2188 /*
2189 * Locate starting entry and clip if necessary.
2190 */
2191 VM_MAP_RANGE_CHECK(map, start, end);
2192
2193 if (vm_map_lookup_entry(map, start, &entry)) {
2194 if (modify_map)
2195 vm_map_clip_start(map, entry, start);
2196 } else {
2197 entry = entry->next;
2198 }
2199
2200 if (modify_map) {
2201 /*
2202 * madvise behaviors that are implemented in the vm_map_entry.
2203 *
2204 * We clip the vm_map_entry so that behavioral changes are
2205 * limited to the specified address range.
2206 */
2207 for (current = entry;
2208 (current != &map->header) && (current->start < end);
2209 current = current->next
2210 ) {
2211 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2212 continue;
2213
2214 vm_map_clip_end(map, current, end);
2215
2216 switch (behav) {
2217 case MADV_NORMAL:
2218 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
2219 break;
2220 case MADV_SEQUENTIAL:
2221 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
2222 break;
2223 case MADV_RANDOM:
2224 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
2225 break;
2226 case MADV_NOSYNC:
2227 current->eflags |= MAP_ENTRY_NOSYNC;
2228 break;
2229 case MADV_AUTOSYNC:
2230 current->eflags &= ~MAP_ENTRY_NOSYNC;
2231 break;
2232 case MADV_NOCORE:
2233 current->eflags |= MAP_ENTRY_NOCOREDUMP;
2234 break;
2235 case MADV_CORE:
2236 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2237 break;
2238 default:
2239 break;
2240 }
2241 vm_map_simplify_entry(map, current);
2242 }
2243 vm_map_unlock(map);
2244 } else {
2245 vm_pindex_t pstart, pend;
2246
2247 /*
2248 * madvise behaviors that are implemented in the underlying
2249 * vm_object.
2250 *
2251 * Since we don't clip the vm_map_entry, we have to clip
2252 * the vm_object pindex and count.
2253 */
2254 for (current = entry;
2255 (current != &map->header) && (current->start < end);
2256 current = current->next
2257 ) {
2258 vm_offset_t useEnd, useStart;
2259
2260 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
2261 continue;
2262
2263 pstart = OFF_TO_IDX(current->offset);
2264 pend = pstart + atop(current->end - current->start);
2265 useStart = current->start;
2266 useEnd = current->end;
2267
2268 if (current->start < start) {
2269 pstart += atop(start - current->start);
2270 useStart = start;
2271 }
2272 if (current->end > end) {
2273 pend -= atop(current->end - end);
2274 useEnd = end;
2275 }
2276
2277 if (pstart >= pend)
2278 continue;
2279
2280 /*
2281 * Perform the pmap_advise() before clearing
2282 * PGA_REFERENCED in vm_page_advise(). Otherwise, a
2283 * concurrent pmap operation, such as pmap_remove(),
2284 * could clear a reference in the pmap and set
2285 * PGA_REFERENCED on the page before the pmap_advise()
2286 * had completed. Consequently, the page would appear
2287 * referenced based upon an old reference that
2288 * occurred before this pmap_advise() ran.
2289 */
2290 if (behav == MADV_DONTNEED || behav == MADV_FREE)
2291 pmap_advise(map->pmap, useStart, useEnd,
2292 behav);
2293
2294 vm_object_madvise(current->object.vm_object, pstart,
2295 pend, behav);
2296
2297 /*
2298 * Pre-populate paging structures in the
2299 * WILLNEED case. For wired entries, the
2300 * paging structures are already populated.
2301 */
2302 if (behav == MADV_WILLNEED &&
2303 current->wired_count == 0) {
2304 vm_map_pmap_enter(map,
2305 useStart,
2306 current->protection,
2307 current->object.vm_object,
2308 pstart,
2309 ptoa(pend - pstart),
2310 MAP_PREFAULT_MADVISE
2311 );
2312 }
2313 }
2314 vm_map_unlock_read(map);
2315 }
2316 return (0);
2317 }
2318
2319
2320 /*
2321 * vm_map_inherit:
2322 *
2323 * Sets the inheritance of the specified address
2324 * range in the target map. Inheritance
2325 * affects how the map will be shared with
2326 * child maps at the time of vmspace_fork.
2327 */
2328 int
2329 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2330 vm_inherit_t new_inheritance)
2331 {
2332 vm_map_entry_t entry;
2333 vm_map_entry_t temp_entry;
2334
2335 switch (new_inheritance) {
2336 case VM_INHERIT_NONE:
2337 case VM_INHERIT_COPY:
2338 case VM_INHERIT_SHARE:
2339 case VM_INHERIT_ZERO:
2340 break;
2341 default:
2342 return (KERN_INVALID_ARGUMENT);
2343 }
2344 if (start == end)
2345 return (KERN_SUCCESS);
2346 vm_map_lock(map);
2347 VM_MAP_RANGE_CHECK(map, start, end);
2348 if (vm_map_lookup_entry(map, start, &temp_entry)) {
2349 entry = temp_entry;
2350 vm_map_clip_start(map, entry, start);
2351 } else
2352 entry = temp_entry->next;
2353 while ((entry != &map->header) && (entry->start < end)) {
2354 vm_map_clip_end(map, entry, end);
2355 if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
2356 new_inheritance != VM_INHERIT_ZERO)
2357 entry->inheritance = new_inheritance;
2358 vm_map_simplify_entry(map, entry);
2359 entry = entry->next;
2360 }
2361 vm_map_unlock(map);
2362 return (KERN_SUCCESS);
2363 }
2364
2365 /*
2366 * vm_map_unwire:
2367 *
2368 * Implements both kernel and user unwiring.
2369 */
2370 int
2371 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2372 int flags)
2373 {
2374 vm_map_entry_t entry, first_entry, tmp_entry;
2375 vm_offset_t saved_start;
2376 unsigned int last_timestamp;
2377 int rv;
2378 boolean_t need_wakeup, result, user_unwire;
2379
2380 if (start == end)
2381 return (KERN_SUCCESS);
2382 user_unwire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2383 vm_map_lock(map);
2384 VM_MAP_RANGE_CHECK(map, start, end);
2385 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2386 if (flags & VM_MAP_WIRE_HOLESOK)
2387 first_entry = first_entry->next;
2388 else {
2389 vm_map_unlock(map);
2390 return (KERN_INVALID_ADDRESS);
2391 }
2392 }
2393 last_timestamp = map->timestamp;
2394 entry = first_entry;
2395 while (entry != &map->header && entry->start < end) {
2396 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2397 /*
2398 * We have not yet clipped the entry.
2399 */
2400 saved_start = (start >= entry->start) ? start :
2401 entry->start;
2402 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2403 if (vm_map_unlock_and_wait(map, 0)) {
2404 /*
2405 * Allow interruption of user unwiring?
2406 */
2407 }
2408 vm_map_lock(map);
2409 if (last_timestamp+1 != map->timestamp) {
2410 /*
2411 * Look again for the entry because the map was
2412 * modified while it was unlocked.
2413 * Specifically, the entry may have been
2414 * clipped, merged, or deleted.
2415 */
2416 if (!vm_map_lookup_entry(map, saved_start,
2417 &tmp_entry)) {
2418 if (flags & VM_MAP_WIRE_HOLESOK)
2419 tmp_entry = tmp_entry->next;
2420 else {
2421 if (saved_start == start) {
2422 /*
2423 * First_entry has been deleted.
2424 */
2425 vm_map_unlock(map);
2426 return (KERN_INVALID_ADDRESS);
2427 }
2428 end = saved_start;
2429 rv = KERN_INVALID_ADDRESS;
2430 goto done;
2431 }
2432 }
2433 if (entry == first_entry)
2434 first_entry = tmp_entry;
2435 else
2436 first_entry = NULL;
2437 entry = tmp_entry;
2438 }
2439 last_timestamp = map->timestamp;
2440 continue;
2441 }
2442 vm_map_clip_start(map, entry, start);
2443 vm_map_clip_end(map, entry, end);
2444 /*
2445 * Mark the entry in case the map lock is released. (See
2446 * above.)
2447 */
2448 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2449 entry->wiring_thread == NULL,
2450 ("owned map entry %p", entry));
2451 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2452 entry->wiring_thread = curthread;
2453 /*
2454 * Check the map for holes in the specified region.
2455 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2456 */
2457 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2458 (entry->end < end && (entry->next == &map->header ||
2459 entry->next->start > entry->end))) {
2460 end = entry->end;
2461 rv = KERN_INVALID_ADDRESS;
2462 goto done;
2463 }
2464 /*
2465 * If system unwiring, require that the entry is system wired.
2466 */
2467 if (!user_unwire &&
2468 vm_map_entry_system_wired_count(entry) == 0) {
2469 end = entry->end;
2470 rv = KERN_INVALID_ARGUMENT;
2471 goto done;
2472 }
2473 entry = entry->next;
2474 }
2475 rv = KERN_SUCCESS;
2476 done:
2477 need_wakeup = FALSE;
2478 if (first_entry == NULL) {
2479 result = vm_map_lookup_entry(map, start, &first_entry);
2480 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2481 first_entry = first_entry->next;
2482 else
2483 KASSERT(result, ("vm_map_unwire: lookup failed"));
2484 }
2485 for (entry = first_entry; entry != &map->header && entry->start < end;
2486 entry = entry->next) {
2487 /*
2488 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2489 * space in the unwired region could have been mapped
2490 * while the map lock was dropped for draining
2491 * MAP_ENTRY_IN_TRANSITION. Moreover, another thread
2492 * could be simultaneously wiring this new mapping
2493 * entry. Detect these cases and skip any entries
2494 * marked as in transition by us.
2495 */
2496 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2497 entry->wiring_thread != curthread) {
2498 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2499 ("vm_map_unwire: !HOLESOK and new/changed entry"));
2500 continue;
2501 }
2502
2503 if (rv == KERN_SUCCESS && (!user_unwire ||
2504 (entry->eflags & MAP_ENTRY_USER_WIRED))) {
2505 if (user_unwire)
2506 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2507 if (entry->wired_count == 1)
2508 vm_map_entry_unwire(map, entry);
2509 else
2510 entry->wired_count--;
2511 }
2512 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2513 ("vm_map_unwire: in-transition flag missing %p", entry));
2514 KASSERT(entry->wiring_thread == curthread,
2515 ("vm_map_unwire: alien wire %p", entry));
2516 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
2517 entry->wiring_thread = NULL;
2518 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2519 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2520 need_wakeup = TRUE;
2521 }
2522 vm_map_simplify_entry(map, entry);
2523 }
2524 vm_map_unlock(map);
2525 if (need_wakeup)
2526 vm_map_wakeup(map);
2527 return (rv);
2528 }
2529
2530 /*
2531 * vm_map_wire_entry_failure:
2532 *
2533 * Handle a wiring failure on the given entry.
2534 *
2535 * The map should be locked.
2536 */
2537 static void
2538 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
2539 vm_offset_t failed_addr)
2540 {
2541
2542 VM_MAP_ASSERT_LOCKED(map);
2543 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
2544 entry->wired_count == 1,
2545 ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
2546 KASSERT(failed_addr < entry->end,
2547 ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
2548
2549 /*
2550 * If any pages at the start of this entry were successfully wired,
2551 * then unwire them.
2552 */
2553 if (failed_addr > entry->start) {
2554 pmap_unwire(map->pmap, entry->start, failed_addr);
2555 vm_object_unwire(entry->object.vm_object, entry->offset,
2556 failed_addr - entry->start, PQ_ACTIVE);
2557 }
2558
2559 /*
2560 * Assign an out-of-range value to represent the failure to wire this
2561 * entry.
2562 */
2563 entry->wired_count = -1;
2564 }
2565
2566 /*
2567 * vm_map_wire:
2568 *
2569 * Implements both kernel and user wiring.
2570 */
2571 int
2572 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
2573 int flags)
2574 {
2575 vm_map_entry_t entry, first_entry, tmp_entry;
2576 vm_offset_t faddr, saved_end, saved_start;
2577 unsigned int last_timestamp;
2578 int rv;
2579 boolean_t need_wakeup, result, user_wire;
2580 vm_prot_t prot;
2581
2582 if (start == end)
2583 return (KERN_SUCCESS);
2584 prot = 0;
2585 if (flags & VM_MAP_WIRE_WRITE)
2586 prot |= VM_PROT_WRITE;
2587 user_wire = (flags & VM_MAP_WIRE_USER) ? TRUE : FALSE;
2588 vm_map_lock(map);
2589 VM_MAP_RANGE_CHECK(map, start, end);
2590 if (!vm_map_lookup_entry(map, start, &first_entry)) {
2591 if (flags & VM_MAP_WIRE_HOLESOK)
2592 first_entry = first_entry->next;
2593 else {
2594 vm_map_unlock(map);
2595 return (KERN_INVALID_ADDRESS);
2596 }
2597 }
2598 last_timestamp = map->timestamp;
2599 entry = first_entry;
2600 while (entry != &map->header && entry->start < end) {
2601 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2602 /*
2603 * We have not yet clipped the entry.
2604 */
2605 saved_start = (start >= entry->start) ? start :
2606 entry->start;
2607 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2608 if (vm_map_unlock_and_wait(map, 0)) {
2609 /*
2610 * Allow interruption of user wiring?
2611 */
2612 }
2613 vm_map_lock(map);
2614 if (last_timestamp + 1 != map->timestamp) {
2615 /*
2616 * Look again for the entry because the map was
2617 * modified while it was unlocked.
2618 * Specifically, the entry may have been
2619 * clipped, merged, or deleted.
2620 */
2621 if (!vm_map_lookup_entry(map, saved_start,
2622 &tmp_entry)) {
2623 if (flags & VM_MAP_WIRE_HOLESOK)
2624 tmp_entry = tmp_entry->next;
2625 else {
2626 if (saved_start == start) {
2627 /*
2628 * first_entry has been deleted.
2629 */
2630 vm_map_unlock(map);
2631 return (KERN_INVALID_ADDRESS);
2632 }
2633 end = saved_start;
2634 rv = KERN_INVALID_ADDRESS;
2635 goto done;
2636 }
2637 }
2638 if (entry == first_entry)
2639 first_entry = tmp_entry;
2640 else
2641 first_entry = NULL;
2642 entry = tmp_entry;
2643 }
2644 last_timestamp = map->timestamp;
2645 continue;
2646 }
2647 vm_map_clip_start(map, entry, start);
2648 vm_map_clip_end(map, entry, end);
2649 /*
2650 * Mark the entry in case the map lock is released. (See
2651 * above.)
2652 */
2653 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
2654 entry->wiring_thread == NULL,
2655 ("owned map entry %p", entry));
2656 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
2657 entry->wiring_thread = curthread;
2658 if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
2659 || (entry->protection & prot) != prot) {
2660 entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
2661 if ((flags & VM_MAP_WIRE_HOLESOK) == 0) {
2662 end = entry->end;
2663 rv = KERN_INVALID_ADDRESS;
2664 goto done;
2665 }
2666 goto next_entry;
2667 }
2668 if (entry->wired_count == 0) {
2669 entry->wired_count++;
2670 saved_start = entry->start;
2671 saved_end = entry->end;
2672
2673 /*
2674 * Release the map lock, relying on the in-transition
2675 * mark. Mark the map busy for fork.
2676 */
2677 vm_map_busy(map);
2678 vm_map_unlock(map);
2679
2680 faddr = saved_start;
2681 do {
2682 /*
2683 * Simulate a fault to get the page and enter
2684 * it into the physical map.
2685 */
2686 if ((rv = vm_fault(map, faddr, VM_PROT_NONE,
2687 VM_FAULT_WIRE)) != KERN_SUCCESS)
2688 break;
2689 } while ((faddr += PAGE_SIZE) < saved_end);
2690 vm_map_lock(map);
2691 vm_map_unbusy(map);
2692 if (last_timestamp + 1 != map->timestamp) {
2693 /*
2694 * Look again for the entry because the map was
2695 * modified while it was unlocked. The entry
2696 * may have been clipped, but NOT merged or
2697 * deleted.
2698 */
2699 result = vm_map_lookup_entry(map, saved_start,
2700 &tmp_entry);
2701 KASSERT(result, ("vm_map_wire: lookup failed"));
2702 if (entry == first_entry)
2703 first_entry = tmp_entry;
2704 else
2705 first_entry = NULL;
2706 entry = tmp_entry;
2707 while (entry->end < saved_end) {
2708 /*
2709 * In case of failure, handle entries
2710 * that were not fully wired here;
2711 * fully wired entries are handled
2712 * later.
2713 */
2714 if (rv != KERN_SUCCESS &&
2715 faddr < entry->end)
2716 vm_map_wire_entry_failure(map,
2717 entry, faddr);
2718 entry = entry->next;
2719 }
2720 }
2721 last_timestamp = map->timestamp;
2722 if (rv != KERN_SUCCESS) {
2723 vm_map_wire_entry_failure(map, entry, faddr);
2724 end = entry->end;
2725 goto done;
2726 }
2727 } else if (!user_wire ||
2728 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2729 entry->wired_count++;
2730 }
2731 /*
2732 * Check the map for holes in the specified region.
2733 * If VM_MAP_WIRE_HOLESOK was specified, skip this check.
2734 */
2735 next_entry:
2736 if (((flags & VM_MAP_WIRE_HOLESOK) == 0) &&
2737 (entry->end < end && (entry->next == &map->header ||
2738 entry->next->start > entry->end))) {
2739 end = entry->end;
2740 rv = KERN_INVALID_ADDRESS;
2741 goto done;
2742 }
2743 entry = entry->next;
2744 }
2745 rv = KERN_SUCCESS;
2746 done:
2747 need_wakeup = FALSE;
2748 if (first_entry == NULL) {
2749 result = vm_map_lookup_entry(map, start, &first_entry);
2750 if (!result && (flags & VM_MAP_WIRE_HOLESOK))
2751 first_entry = first_entry->next;
2752 else
2753 KASSERT(result, ("vm_map_wire: lookup failed"));
2754 }
2755 for (entry = first_entry; entry != &map->header && entry->start < end;
2756 entry = entry->next) {
2757 /*
2758 * If VM_MAP_WIRE_HOLESOK was specified, an empty
2759 * space in the unwired region could have been mapped
2760 * while the map lock was dropped for faulting in the
2761 * pages or draining MAP_ENTRY_IN_TRANSITION.
2762 * Moreover, another thread could be simultaneously
2763 * wiring this new mapping entry. Detect these cases
2764 * and skip any entries marked as in transition not by us.
2765 */
2766 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
2767 entry->wiring_thread != curthread) {
2768 KASSERT((flags & VM_MAP_WIRE_HOLESOK) != 0,
2769 ("vm_map_wire: !HOLESOK and new/changed entry"));
2770 continue;
2771 }
2772
2773 if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0)
2774 goto next_entry_done;
2775
2776 if (rv == KERN_SUCCESS) {
2777 if (user_wire)
2778 entry->eflags |= MAP_ENTRY_USER_WIRED;
2779 } else if (entry->wired_count == -1) {
2780 /*
2781 * Wiring failed on this entry. Thus, unwiring is
2782 * unnecessary.
2783 */
2784 entry->wired_count = 0;
2785 } else if (!user_wire ||
2786 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2787 /*
2788 * Undo the wiring. Wiring succeeded on this entry
2789 * but failed on a later entry.
2790 */
2791 if (entry->wired_count == 1)
2792 vm_map_entry_unwire(map, entry);
2793 else
2794 entry->wired_count--;
2795 }
2796 next_entry_done:
2797 KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
2798 ("vm_map_wire: in-transition flag missing %p", entry));
2799 KASSERT(entry->wiring_thread == curthread,
2800 ("vm_map_wire: alien wire %p", entry));
2801 entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
2802 MAP_ENTRY_WIRE_SKIPPED);
2803 entry->wiring_thread = NULL;
2804 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
2805 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
2806 need_wakeup = TRUE;
2807 }
2808 vm_map_simplify_entry(map, entry);
2809 }
2810 vm_map_unlock(map);
2811 if (need_wakeup)
2812 vm_map_wakeup(map);
2813 return (rv);
2814 }
2815
2816 /*
2817 * vm_map_sync
2818 *
2819 * Push any dirty cached pages in the address range to their pager.
2820 * If syncio is TRUE, dirty pages are written synchronously.
2821 * If invalidate is TRUE, any cached pages are freed as well.
2822 *
2823 * If the size of the region from start to end is zero, we are
2824 * supposed to flush all modified pages within the region containing
2825 * start. Unfortunately, a region can be split or coalesced with
2826 * neighboring regions, making it difficult to determine what the
2827 * original region was. Therefore, we approximate this requirement by
2828 * flushing the current region containing start.
2829 *
2830 * Returns an error if any part of the specified range is not mapped.
2831 */
2832 int
2833 vm_map_sync(
2834 vm_map_t map,
2835 vm_offset_t start,
2836 vm_offset_t end,
2837 boolean_t syncio,
2838 boolean_t invalidate)
2839 {
2840 vm_map_entry_t current;
2841 vm_map_entry_t entry;
2842 vm_size_t size;
2843 vm_object_t object;
2844 vm_ooffset_t offset;
2845 unsigned int last_timestamp;
2846 boolean_t failed;
2847
2848 vm_map_lock_read(map);
2849 VM_MAP_RANGE_CHECK(map, start, end);
2850 if (!vm_map_lookup_entry(map, start, &entry)) {
2851 vm_map_unlock_read(map);
2852 return (KERN_INVALID_ADDRESS);
2853 } else if (start == end) {
2854 start = entry->start;
2855 end = entry->end;
2856 }
2857 /*
2858 * Make a first pass to check for user-wired memory and holes.
2859 */
2860 for (current = entry; current != &map->header && current->start < end;
2861 current = current->next) {
2862 if (invalidate && (current->eflags & MAP_ENTRY_USER_WIRED)) {
2863 vm_map_unlock_read(map);
2864 return (KERN_INVALID_ARGUMENT);
2865 }
2866 if (end > current->end &&
2867 (current->next == &map->header ||
2868 current->end != current->next->start)) {
2869 vm_map_unlock_read(map);
2870 return (KERN_INVALID_ADDRESS);
2871 }
2872 }
2873
2874 if (invalidate)
2875 pmap_remove(map->pmap, start, end);
2876 failed = FALSE;
2877
2878 /*
2879 * Make a second pass, cleaning/uncaching pages from the indicated
2880 * objects as we go.
2881 */
2882 for (current = entry; current != &map->header && current->start < end;) {
2883 offset = current->offset + (start - current->start);
2884 size = (end <= current->end ? end : current->end) - start;
2885 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
2886 vm_map_t smap;
2887 vm_map_entry_t tentry;
2888 vm_size_t tsize;
2889
2890 smap = current->object.sub_map;
2891 vm_map_lock_read(smap);
2892 (void) vm_map_lookup_entry(smap, offset, &tentry);
2893 tsize = tentry->end - offset;
2894 if (tsize < size)
2895 size = tsize;
2896 object = tentry->object.vm_object;
2897 offset = tentry->offset + (offset - tentry->start);
2898 vm_map_unlock_read(smap);
2899 } else {
2900 object = current->object.vm_object;
2901 }
2902 vm_object_reference(object);
2903 last_timestamp = map->timestamp;
2904 vm_map_unlock_read(map);
2905 if (!vm_object_sync(object, offset, size, syncio, invalidate))
2906 failed = TRUE;
2907 start += size;
2908 vm_object_deallocate(object);
2909 vm_map_lock_read(map);
2910 if (last_timestamp == map->timestamp ||
2911 !vm_map_lookup_entry(map, start, ¤t))
2912 current = current->next;
2913 }
2914
2915 vm_map_unlock_read(map);
2916 return (failed ? KERN_FAILURE : KERN_SUCCESS);
2917 }
2918
2919 /*
2920 * vm_map_entry_unwire: [ internal use only ]
2921 *
2922 * Make the region specified by this entry pageable.
2923 *
2924 * The map in question should be locked.
2925 * [This is the reason for this routine's existence.]
2926 */
2927 static void
2928 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2929 {
2930
2931 VM_MAP_ASSERT_LOCKED(map);
2932 KASSERT(entry->wired_count > 0,
2933 ("vm_map_entry_unwire: entry %p isn't wired", entry));
2934 pmap_unwire(map->pmap, entry->start, entry->end);
2935 vm_object_unwire(entry->object.vm_object, entry->offset, entry->end -
2936 entry->start, PQ_ACTIVE);
2937 entry->wired_count = 0;
2938 }
2939
2940 static void
2941 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
2942 {
2943
2944 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
2945 vm_object_deallocate(entry->object.vm_object);
2946 uma_zfree(system_map ? kmapentzone : mapentzone, entry);
2947 }
2948
2949 /*
2950 * vm_map_entry_delete: [ internal use only ]
2951 *
2952 * Deallocate the given entry from the target map.
2953 */
2954 static void
2955 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2956 {
2957 vm_object_t object;
2958 vm_pindex_t offidxstart, offidxend, count, size1;
2959 vm_size_t size;
2960
2961 vm_map_entry_unlink(map, entry);
2962 object = entry->object.vm_object;
2963
2964 if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
2965 MPASS(entry->cred == NULL);
2966 MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
2967 MPASS(object == NULL);
2968 vm_map_entry_deallocate(entry, map->system_map);
2969 return;
2970 }
2971
2972 size = entry->end - entry->start;
2973 map->size -= size;
2974
2975 if (entry->cred != NULL) {
2976 swap_release_by_cred(size, entry->cred);
2977 crfree(entry->cred);
2978 }
2979
2980 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
2981 (object != NULL)) {
2982 KASSERT(entry->cred == NULL || object->cred == NULL ||
2983 (entry->eflags & MAP_ENTRY_NEEDS_COPY),
2984 ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
2985 count = atop(size);
2986 offidxstart = OFF_TO_IDX(entry->offset);
2987 offidxend = offidxstart + count;
2988 VM_OBJECT_WLOCK(object);
2989 if (object->ref_count != 1 && ((object->flags & (OBJ_NOSPLIT |
2990 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING ||
2991 object == kernel_object || object == kmem_object)) {
2992 vm_object_collapse(object);
2993
2994 /*
2995 * The option OBJPR_NOTMAPPED can be passed here
2996 * because vm_map_delete() already performed
2997 * pmap_remove() on the only mapping to this range
2998 * of pages.
2999 */
3000 vm_object_page_remove(object, offidxstart, offidxend,
3001 OBJPR_NOTMAPPED);
3002 if (object->type == OBJT_SWAP)
3003 swap_pager_freespace(object, offidxstart,
3004 count);
3005 if (offidxend >= object->size &&
3006 offidxstart < object->size) {
3007 size1 = object->size;
3008 object->size = offidxstart;
3009 if (object->cred != NULL) {
3010 size1 -= object->size;
3011 KASSERT(object->charge >= ptoa(size1),
3012 ("object %p charge < 0", object));
3013 swap_release_by_cred(ptoa(size1),
3014 object->cred);
3015 object->charge -= ptoa(size1);
3016 }
3017 }
3018 }
3019 VM_OBJECT_WUNLOCK(object);
3020 } else
3021 entry->object.vm_object = NULL;
3022 if (map->system_map)
3023 vm_map_entry_deallocate(entry, TRUE);
3024 else {
3025 entry->next = curthread->td_map_def_user;
3026 curthread->td_map_def_user = entry;
3027 }
3028 }
3029
3030 /*
3031 * vm_map_delete: [ internal use only ]
3032 *
3033 * Deallocates the given address range from the target
3034 * map.
3035 */
3036 int
3037 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3038 {
3039 vm_map_entry_t entry;
3040 vm_map_entry_t first_entry;
3041
3042 VM_MAP_ASSERT_LOCKED(map);
3043 if (start == end)
3044 return (KERN_SUCCESS);
3045
3046 /*
3047 * Find the start of the region, and clip it
3048 */
3049 if (!vm_map_lookup_entry(map, start, &first_entry))
3050 entry = first_entry->next;
3051 else {
3052 entry = first_entry;
3053 vm_map_clip_start(map, entry, start);
3054 }
3055
3056 /*
3057 * Step through all entries in this region
3058 */
3059 while ((entry != &map->header) && (entry->start < end)) {
3060 vm_map_entry_t next;
3061
3062 /*
3063 * Wait for wiring or unwiring of an entry to complete.
3064 * Also wait for any system wirings to disappear on
3065 * user maps.
3066 */
3067 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3068 (vm_map_pmap(map) != kernel_pmap &&
3069 vm_map_entry_system_wired_count(entry) != 0)) {
3070 unsigned int last_timestamp;
3071 vm_offset_t saved_start;
3072 vm_map_entry_t tmp_entry;
3073
3074 saved_start = entry->start;
3075 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3076 last_timestamp = map->timestamp;
3077 (void) vm_map_unlock_and_wait(map, 0);
3078 vm_map_lock(map);
3079 if (last_timestamp + 1 != map->timestamp) {
3080 /*
3081 * Look again for the entry because the map was
3082 * modified while it was unlocked.
3083 * Specifically, the entry may have been
3084 * clipped, merged, or deleted.
3085 */
3086 if (!vm_map_lookup_entry(map, saved_start,
3087 &tmp_entry))
3088 entry = tmp_entry->next;
3089 else {
3090 entry = tmp_entry;
3091 vm_map_clip_start(map, entry,
3092 saved_start);
3093 }
3094 }
3095 continue;
3096 }
3097 vm_map_clip_end(map, entry, end);
3098
3099 next = entry->next;
3100
3101 /*
3102 * Unwire before removing addresses from the pmap; otherwise,
3103 * unwiring will put the entries back in the pmap.
3104 */
3105 if (entry->wired_count != 0) {
3106 vm_map_entry_unwire(map, entry);
3107 }
3108
3109 pmap_remove(map->pmap, entry->start, entry->end);
3110
3111 /*
3112 * Delete the entry only after removing all pmap
3113 * entries pointing to its pages. (Otherwise, its
3114 * page frames may be reallocated, and any modify bits
3115 * will be set in the wrong object!)
3116 */
3117 vm_map_entry_delete(map, entry);
3118 entry = next;
3119 }
3120 return (KERN_SUCCESS);
3121 }
3122
3123 /*
3124 * vm_map_remove:
3125 *
3126 * Remove the given address range from the target map.
3127 * This is the exported form of vm_map_delete.
3128 */
3129 int
3130 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3131 {
3132 int result;
3133
3134 vm_map_lock(map);
3135 VM_MAP_RANGE_CHECK(map, start, end);
3136 result = vm_map_delete(map, start, end);
3137 vm_map_unlock(map);
3138 return (result);
3139 }
3140
3141 /*
3142 * vm_map_check_protection:
3143 *
3144 * Assert that the target map allows the specified privilege on the
3145 * entire address region given. The entire region must be allocated.
3146 *
3147 * WARNING! This code does not and should not check whether the
3148 * contents of the region is accessible. For example a smaller file
3149 * might be mapped into a larger address space.
3150 *
3151 * NOTE! This code is also called by munmap().
3152 *
3153 * The map must be locked. A read lock is sufficient.
3154 */
3155 boolean_t
3156 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3157 vm_prot_t protection)
3158 {
3159 vm_map_entry_t entry;
3160 vm_map_entry_t tmp_entry;
3161
3162 if (!vm_map_lookup_entry(map, start, &tmp_entry))
3163 return (FALSE);
3164 entry = tmp_entry;
3165
3166 while (start < end) {
3167 if (entry == &map->header)
3168 return (FALSE);
3169 /*
3170 * No holes allowed!
3171 */
3172 if (start < entry->start)
3173 return (FALSE);
3174 /*
3175 * Check protection associated with entry.
3176 */
3177 if ((entry->protection & protection) != protection)
3178 return (FALSE);
3179 /* go to next entry */
3180 start = entry->end;
3181 entry = entry->next;
3182 }
3183 return (TRUE);
3184 }
3185
3186 /*
3187 * vm_map_copy_entry:
3188 *
3189 * Copies the contents of the source entry to the destination
3190 * entry. The entries *must* be aligned properly.
3191 */
3192 static void
3193 vm_map_copy_entry(
3194 vm_map_t src_map,
3195 vm_map_t dst_map,
3196 vm_map_entry_t src_entry,
3197 vm_map_entry_t dst_entry,
3198 vm_ooffset_t *fork_charge)
3199 {
3200 vm_object_t src_object;
3201 vm_map_entry_t fake_entry;
3202 vm_offset_t size;
3203 struct ucred *cred;
3204 int charged;
3205
3206 VM_MAP_ASSERT_LOCKED(dst_map);
3207
3208 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3209 return;
3210
3211 if (src_entry->wired_count == 0 ||
3212 (src_entry->protection & VM_PROT_WRITE) == 0) {
3213 /*
3214 * If the source entry is marked needs_copy, it is already
3215 * write-protected.
3216 */
3217 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3218 (src_entry->protection & VM_PROT_WRITE) != 0) {
3219 pmap_protect(src_map->pmap,
3220 src_entry->start,
3221 src_entry->end,
3222 src_entry->protection & ~VM_PROT_WRITE);
3223 }
3224
3225 /*
3226 * Make a copy of the object.
3227 */
3228 size = src_entry->end - src_entry->start;
3229 if ((src_object = src_entry->object.vm_object) != NULL) {
3230 VM_OBJECT_WLOCK(src_object);
3231 charged = ENTRY_CHARGED(src_entry);
3232 if (src_object->handle == NULL &&
3233 (src_object->type == OBJT_DEFAULT ||
3234 src_object->type == OBJT_SWAP)) {
3235 vm_object_collapse(src_object);
3236 if ((src_object->flags & (OBJ_NOSPLIT |
3237 OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
3238 vm_object_split(src_entry);
3239 src_object =
3240 src_entry->object.vm_object;
3241 }
3242 }
3243 vm_object_reference_locked(src_object);
3244 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3245 if (src_entry->cred != NULL &&
3246 !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3247 KASSERT(src_object->cred == NULL,
3248 ("OVERCOMMIT: vm_map_copy_entry: cred %p",
3249 src_object));
3250 src_object->cred = src_entry->cred;
3251 src_object->charge = size;
3252 }
3253 VM_OBJECT_WUNLOCK(src_object);
3254 dst_entry->object.vm_object = src_object;
3255 if (charged) {
3256 cred = curthread->td_ucred;
3257 crhold(cred);
3258 dst_entry->cred = cred;
3259 *fork_charge += size;
3260 if (!(src_entry->eflags &
3261 MAP_ENTRY_NEEDS_COPY)) {
3262 crhold(cred);
3263 src_entry->cred = cred;
3264 *fork_charge += size;
3265 }
3266 }
3267 src_entry->eflags |= MAP_ENTRY_COW |
3268 MAP_ENTRY_NEEDS_COPY;
3269 dst_entry->eflags |= MAP_ENTRY_COW |
3270 MAP_ENTRY_NEEDS_COPY;
3271 dst_entry->offset = src_entry->offset;
3272 if (src_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3273 /*
3274 * MAP_ENTRY_VN_WRITECNT cannot
3275 * indicate write reference from
3276 * src_entry, since the entry is
3277 * marked as needs copy. Allocate a
3278 * fake entry that is used to
3279 * decrement object->un_pager.vnp.writecount
3280 * at the appropriate time. Attach
3281 * fake_entry to the deferred list.
3282 */
3283 fake_entry = vm_map_entry_create(dst_map);
3284 fake_entry->eflags = MAP_ENTRY_VN_WRITECNT;
3285 src_entry->eflags &= ~MAP_ENTRY_VN_WRITECNT;
3286 vm_object_reference(src_object);
3287 fake_entry->object.vm_object = src_object;
3288 fake_entry->start = src_entry->start;
3289 fake_entry->end = src_entry->end;
3290 fake_entry->next = curthread->td_map_def_user;
3291 curthread->td_map_def_user = fake_entry;
3292 }
3293
3294 pmap_copy(dst_map->pmap, src_map->pmap,
3295 dst_entry->start, dst_entry->end - dst_entry->start,
3296 src_entry->start);
3297 } else {
3298 dst_entry->object.vm_object = NULL;
3299 dst_entry->offset = 0;
3300 if (src_entry->cred != NULL) {
3301 dst_entry->cred = curthread->td_ucred;
3302 crhold(dst_entry->cred);
3303 *fork_charge += size;
3304 }
3305 }
3306 } else {
3307 /*
3308 * We don't want to make writeable wired pages copy-on-write.
3309 * Immediately copy these pages into the new map by simulating
3310 * page faults. The new pages are pageable.
3311 */
3312 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
3313 fork_charge);
3314 }
3315 }
3316
3317 /*
3318 * vmspace_map_entry_forked:
3319 * Update the newly-forked vmspace each time a map entry is inherited
3320 * or copied. The values for vm_dsize and vm_tsize are approximate
3321 * (and mostly-obsolete ideas in the face of mmap(2) et al.)
3322 */
3323 static void
3324 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
3325 vm_map_entry_t entry)
3326 {
3327 vm_size_t entrysize;
3328 vm_offset_t newend;
3329
3330 if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
3331 return;
3332 entrysize = entry->end - entry->start;
3333 vm2->vm_map.size += entrysize;
3334 if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
3335 vm2->vm_ssize += btoc(entrysize);
3336 } else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
3337 entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
3338 newend = MIN(entry->end,
3339 (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
3340 vm2->vm_dsize += btoc(newend - entry->start);
3341 } else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
3342 entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
3343 newend = MIN(entry->end,
3344 (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
3345 vm2->vm_tsize += btoc(newend - entry->start);
3346 }
3347 }
3348
3349 /*
3350 * vmspace_fork:
3351 * Create a new process vmspace structure and vm_map
3352 * based on those of an existing process. The new map
3353 * is based on the old map, according to the inheritance
3354 * values on the regions in that map.
3355 *
3356 * XXX It might be worth coalescing the entries added to the new vmspace.
3357 *
3358 * The source map must not be locked.
3359 */
3360 struct vmspace *
3361 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
3362 {
3363 struct vmspace *vm2;
3364 vm_map_t new_map, old_map;
3365 vm_map_entry_t new_entry, old_entry;
3366 vm_object_t object;
3367 int locked;
3368 vm_inherit_t inh;
3369
3370 old_map = &vm1->vm_map;
3371 /* Copy immutable fields of vm1 to vm2. */
3372 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset, NULL);
3373 if (vm2 == NULL)
3374 return (NULL);
3375 vm2->vm_taddr = vm1->vm_taddr;
3376 vm2->vm_daddr = vm1->vm_daddr;
3377 vm2->vm_maxsaddr = vm1->vm_maxsaddr;
3378 vm_map_lock(old_map);
3379 if (old_map->busy)
3380 vm_map_wait_busy(old_map);
3381 new_map = &vm2->vm_map;
3382 locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
3383 KASSERT(locked, ("vmspace_fork: lock failed"));
3384
3385 old_entry = old_map->header.next;
3386
3387 while (old_entry != &old_map->header) {
3388 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
3389 panic("vm_map_fork: encountered a submap");
3390
3391 inh = old_entry->inheritance;
3392 if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
3393 inh != VM_INHERIT_NONE)
3394 inh = VM_INHERIT_COPY;
3395
3396 switch (inh) {
3397 case VM_INHERIT_NONE:
3398 break;
3399
3400 case VM_INHERIT_SHARE:
3401 /*
3402 * Clone the entry, creating the shared object if necessary.
3403 */
3404 object = old_entry->object.vm_object;
3405 if (object == NULL) {
3406 object = vm_object_allocate(OBJT_DEFAULT,
3407 atop(old_entry->end - old_entry->start));
3408 old_entry->object.vm_object = object;
3409 old_entry->offset = 0;
3410 if (old_entry->cred != NULL) {
3411 object->cred = old_entry->cred;
3412 object->charge = old_entry->end -
3413 old_entry->start;
3414 old_entry->cred = NULL;
3415 }
3416 }
3417
3418 /*
3419 * Add the reference before calling vm_object_shadow
3420 * to insure that a shadow object is created.
3421 */
3422 vm_object_reference(object);
3423 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3424 vm_object_shadow(&old_entry->object.vm_object,
3425 &old_entry->offset,
3426 old_entry->end - old_entry->start);
3427 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3428 /* Transfer the second reference too. */
3429 vm_object_reference(
3430 old_entry->object.vm_object);
3431
3432 /*
3433 * As in vm_map_simplify_entry(), the
3434 * vnode lock will not be acquired in
3435 * this call to vm_object_deallocate().
3436 */
3437 vm_object_deallocate(object);
3438 object = old_entry->object.vm_object;
3439 }
3440 VM_OBJECT_WLOCK(object);
3441 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3442 if (old_entry->cred != NULL) {
3443 KASSERT(object->cred == NULL, ("vmspace_fork both cred"));
3444 object->cred = old_entry->cred;
3445 object->charge = old_entry->end - old_entry->start;
3446 old_entry->cred = NULL;
3447 }
3448
3449 /*
3450 * Assert the correct state of the vnode
3451 * v_writecount while the object is locked, to
3452 * not relock it later for the assertion
3453 * correctness.
3454 */
3455 if (old_entry->eflags & MAP_ENTRY_VN_WRITECNT &&
3456 object->type == OBJT_VNODE) {
3457 KASSERT(((struct vnode *)object->handle)->
3458 v_writecount > 0,
3459 ("vmspace_fork: v_writecount %p", object));
3460 KASSERT(object->un_pager.vnp.writemappings > 0,
3461 ("vmspace_fork: vnp.writecount %p",
3462 object));
3463 }
3464 VM_OBJECT_WUNLOCK(object);
3465
3466 /*
3467 * Clone the entry, referencing the shared object.
3468 */
3469 new_entry = vm_map_entry_create(new_map);
3470 *new_entry = *old_entry;
3471 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3472 MAP_ENTRY_IN_TRANSITION);
3473 new_entry->wiring_thread = NULL;
3474 new_entry->wired_count = 0;
3475 if (new_entry->eflags & MAP_ENTRY_VN_WRITECNT) {
3476 vnode_pager_update_writecount(object,
3477 new_entry->start, new_entry->end);
3478 }
3479
3480 /*
3481 * Insert the entry into the new map -- we know we're
3482 * inserting at the end of the new map.
3483 */
3484 vm_map_entry_link(new_map, new_map->header.prev,
3485 new_entry);
3486 vmspace_map_entry_forked(vm1, vm2, new_entry);
3487
3488 /*
3489 * Update the physical map
3490 */
3491 pmap_copy(new_map->pmap, old_map->pmap,
3492 new_entry->start,
3493 (old_entry->end - old_entry->start),
3494 old_entry->start);
3495 break;
3496
3497 case VM_INHERIT_COPY:
3498 /*
3499 * Clone the entry and link into the map.
3500 */
3501 new_entry = vm_map_entry_create(new_map);
3502 *new_entry = *old_entry;
3503 /*
3504 * Copied entry is COW over the old object.
3505 */
3506 new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
3507 MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_VN_WRITECNT);
3508 new_entry->wiring_thread = NULL;
3509 new_entry->wired_count = 0;
3510 new_entry->object.vm_object = NULL;
3511 new_entry->cred = NULL;
3512 vm_map_entry_link(new_map, new_map->header.prev,
3513 new_entry);
3514 vmspace_map_entry_forked(vm1, vm2, new_entry);
3515 vm_map_copy_entry(old_map, new_map, old_entry,
3516 new_entry, fork_charge);
3517 break;
3518
3519 case VM_INHERIT_ZERO:
3520 /*
3521 * Create a new anonymous mapping entry modelled from
3522 * the old one.
3523 */
3524 new_entry = vm_map_entry_create(new_map);
3525 memset(new_entry, 0, sizeof(*new_entry));
3526
3527 new_entry->start = old_entry->start;
3528 new_entry->end = old_entry->end;
3529 new_entry->eflags = old_entry->eflags &
3530 ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
3531 MAP_ENTRY_VN_WRITECNT);
3532 new_entry->protection = old_entry->protection;
3533 new_entry->max_protection = old_entry->max_protection;
3534 new_entry->inheritance = VM_INHERIT_ZERO;
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 new_entry->cred = curthread->td_ucred;
3541 crhold(new_entry->cred);
3542 *fork_charge += (new_entry->end - new_entry->start);
3543
3544 break;
3545 }
3546 old_entry = old_entry->next;
3547 }
3548 /*
3549 * Use inlined vm_map_unlock() to postpone handling the deferred
3550 * map entries, which cannot be done until both old_map and
3551 * new_map locks are released.
3552 */
3553 sx_xunlock(&old_map->lock);
3554 sx_xunlock(&new_map->lock);
3555 vm_map_process_deferred();
3556
3557 return (vm2);
3558 }
3559
3560 int
3561 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3562 vm_prot_t prot, vm_prot_t max, int cow)
3563 {
3564 vm_size_t growsize, init_ssize;
3565 rlim_t lmemlim, vmemlim;
3566 int rv;
3567
3568 growsize = sgrowsiz;
3569 init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
3570 vm_map_lock(map);
3571 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
3572 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3573 if (!old_mlock && map->flags & MAP_WIREFUTURE) {
3574 if (ptoa(pmap_wired_count(map->pmap)) + init_ssize > lmemlim) {
3575 rv = KERN_NO_SPACE;
3576 goto out;
3577 }
3578 }
3579 /* If we would blow our VMEM resource limit, no go */
3580 if (map->size + init_ssize > vmemlim) {
3581 rv = KERN_NO_SPACE;
3582 goto out;
3583 }
3584 rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
3585 max, cow);
3586 out:
3587 vm_map_unlock(map);
3588 return (rv);
3589 }
3590
3591 static int stack_guard_page = 1;
3592 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
3593 &stack_guard_page, 0,
3594 "Specifies the number of guard pages for a stack that grows");
3595
3596 static int
3597 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3598 vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
3599 {
3600 vm_map_entry_t new_entry, prev_entry;
3601 vm_offset_t bot, gap_bot, gap_top, top;
3602 vm_size_t init_ssize, sgp;
3603 int orient, rv;
3604
3605 /*
3606 * The stack orientation is piggybacked with the cow argument.
3607 * Extract it into orient and mask the cow argument so that we
3608 * don't pass it around further.
3609 */
3610 orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
3611 KASSERT(orient != 0, ("No stack grow direction"));
3612 KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
3613 ("bi-dir stack"));
3614
3615 sgp = (vm_size_t)stack_guard_page * PAGE_SIZE;
3616 if (addrbos < vm_map_min(map) ||
3617 addrbos > vm_map_max(map) ||
3618 addrbos + max_ssize < addrbos ||
3619 sgp >= max_ssize)
3620 return (KERN_NO_SPACE);
3621
3622 init_ssize = growsize;
3623 if (max_ssize < init_ssize + sgp)
3624 init_ssize = max_ssize - sgp;
3625
3626 /* If addr is already mapped, no go */
3627 if (vm_map_lookup_entry(map, addrbos, &prev_entry))
3628 return (KERN_NO_SPACE);
3629
3630 /*
3631 * If we can't accommodate max_ssize in the current mapping, no go.
3632 */
3633 if ((prev_entry->next != &map->header) &&
3634 (prev_entry->next->start < addrbos + max_ssize))
3635 return (KERN_NO_SPACE);
3636
3637 /*
3638 * We initially map a stack of only init_ssize. We will grow as
3639 * needed later. Depending on the orientation of the stack (i.e.
3640 * the grow direction) we either map at the top of the range, the
3641 * bottom of the range or in the middle.
3642 *
3643 * Note: we would normally expect prot and max to be VM_PROT_ALL,
3644 * and cow to be 0. Possibly we should eliminate these as input
3645 * parameters, and just pass these values here in the insert call.
3646 */
3647 if (orient == MAP_STACK_GROWS_DOWN) {
3648 bot = addrbos + max_ssize - init_ssize;
3649 top = bot + init_ssize;
3650 gap_bot = addrbos;
3651 gap_top = bot;
3652 } else /* if (orient == MAP_STACK_GROWS_UP) */ {
3653 bot = addrbos;
3654 top = bot + init_ssize;
3655 gap_bot = top;
3656 gap_top = addrbos + max_ssize;
3657 }
3658 rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
3659 if (rv != KERN_SUCCESS)
3660 return (rv);
3661 new_entry = prev_entry->next;
3662 KASSERT(new_entry->end == top || new_entry->start == bot,
3663 ("Bad entry start/end for new stack entry"));
3664 KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
3665 (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
3666 ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
3667 KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
3668 (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
3669 ("new entry lacks MAP_ENTRY_GROWS_UP"));
3670 rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
3671 VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
3672 MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
3673 if (rv != KERN_SUCCESS)
3674 (void)vm_map_delete(map, bot, top);
3675 return (rv);
3676 }
3677
3678 /*
3679 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if we
3680 * successfully grow the stack.
3681 */
3682 static int
3683 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
3684 {
3685 vm_map_entry_t stack_entry;
3686 struct proc *p;
3687 struct vmspace *vm;
3688 struct ucred *cred;
3689 vm_offset_t gap_end, gap_start, grow_start;
3690 size_t grow_amount, guard, max_grow;
3691 rlim_t lmemlim, stacklim, vmemlim;
3692 int rv, rv1;
3693 bool gap_deleted, grow_down, is_procstack;
3694 #ifdef notyet
3695 uint64_t limit;
3696 #endif
3697 #ifdef RACCT
3698 int error;
3699 #endif
3700
3701 p = curproc;
3702 vm = p->p_vmspace;
3703 MPASS(map == &p->p_vmspace->vm_map);
3704 MPASS(!map->system_map);
3705
3706 guard = stack_guard_page * PAGE_SIZE;
3707 lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
3708 stacklim = lim_cur(curthread, RLIMIT_STACK);
3709 vmemlim = lim_cur(curthread, RLIMIT_VMEM);
3710 retry:
3711 /* If addr is not in a hole for a stack grow area, no need to grow. */
3712 if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
3713 return (KERN_FAILURE);
3714 if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
3715 return (KERN_SUCCESS);
3716 if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
3717 stack_entry = gap_entry->next;
3718 if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
3719 stack_entry->start != gap_entry->end)
3720 return (KERN_FAILURE);
3721 grow_amount = round_page(stack_entry->start - addr);
3722 grow_down = true;
3723 } else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
3724 stack_entry = gap_entry->prev;
3725 if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
3726 stack_entry->end != gap_entry->start)
3727 return (KERN_FAILURE);
3728 grow_amount = round_page(addr + 1 - stack_entry->end);
3729 grow_down = false;
3730 } else {
3731 return (KERN_FAILURE);
3732 }
3733 max_grow = gap_entry->end - gap_entry->start;
3734 if (guard > max_grow)
3735 return (KERN_NO_SPACE);
3736 max_grow -= guard;
3737 if (grow_amount > max_grow)
3738 return (KERN_NO_SPACE);
3739
3740 /*
3741 * If this is the main process stack, see if we're over the stack
3742 * limit.
3743 */
3744 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
3745 addr < (vm_offset_t)p->p_sysent->sv_usrstack;
3746 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
3747 return (KERN_NO_SPACE);
3748
3749 #ifdef RACCT
3750 if (racct_enable) {
3751 PROC_LOCK(p);
3752 if (is_procstack && racct_set(p, RACCT_STACK,
3753 ctob(vm->vm_ssize) + grow_amount)) {
3754 PROC_UNLOCK(p);
3755 return (KERN_NO_SPACE);
3756 }
3757 PROC_UNLOCK(p);
3758 }
3759 #endif
3760
3761 grow_amount = roundup(grow_amount, sgrowsiz);
3762 if (grow_amount > max_grow)
3763 grow_amount = max_grow;
3764 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
3765 grow_amount = trunc_page((vm_size_t)stacklim) -
3766 ctob(vm->vm_ssize);
3767 }
3768
3769 #ifdef notyet
3770 PROC_LOCK(p);
3771 limit = racct_get_available(p, RACCT_STACK);
3772 PROC_UNLOCK(p);
3773 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
3774 grow_amount = limit - ctob(vm->vm_ssize);
3775 #endif
3776
3777 if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
3778 if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
3779 rv = KERN_NO_SPACE;
3780 goto out;
3781 }
3782 #ifdef RACCT
3783 if (racct_enable) {
3784 PROC_LOCK(p);
3785 if (racct_set(p, RACCT_MEMLOCK,
3786 ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
3787 PROC_UNLOCK(p);
3788 rv = KERN_NO_SPACE;
3789 goto out;
3790 }
3791 PROC_UNLOCK(p);
3792 }
3793 #endif
3794 }
3795
3796 /* If we would blow our VMEM resource limit, no go */
3797 if (map->size + grow_amount > vmemlim) {
3798 rv = KERN_NO_SPACE;
3799 goto out;
3800 }
3801 #ifdef RACCT
3802 if (racct_enable) {
3803 PROC_LOCK(p);
3804 if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
3805 PROC_UNLOCK(p);
3806 rv = KERN_NO_SPACE;
3807 goto out;
3808 }
3809 PROC_UNLOCK(p);
3810 }
3811 #endif
3812
3813 if (vm_map_lock_upgrade(map)) {
3814 gap_entry = NULL;
3815 vm_map_lock_read(map);
3816 goto retry;
3817 }
3818
3819 if (grow_down) {
3820 grow_start = gap_entry->end - grow_amount;
3821 if (gap_entry->start + grow_amount == gap_entry->end) {
3822 gap_start = gap_entry->start;
3823 gap_end = gap_entry->end;
3824 vm_map_entry_delete(map, gap_entry);
3825 gap_deleted = true;
3826 } else {
3827 MPASS(gap_entry->start < gap_entry->end - grow_amount);
3828 gap_entry->end -= grow_amount;
3829 vm_map_entry_resize_free(map, gap_entry);
3830 gap_deleted = false;
3831 }
3832 rv = vm_map_insert(map, NULL, 0, grow_start,
3833 grow_start + grow_amount,
3834 stack_entry->protection, stack_entry->max_protection,
3835 MAP_STACK_GROWS_DOWN);
3836 if (rv != KERN_SUCCESS) {
3837 if (gap_deleted) {
3838 rv1 = vm_map_insert(map, NULL, 0, gap_start,
3839 gap_end, VM_PROT_NONE, VM_PROT_NONE,
3840 MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
3841 MPASS(rv1 == KERN_SUCCESS);
3842 } else {
3843 gap_entry->end += grow_amount;
3844 vm_map_entry_resize_free(map, gap_entry);
3845 }
3846 }
3847 } else {
3848 grow_start = stack_entry->end;
3849 cred = stack_entry->cred;
3850 if (cred == NULL && stack_entry->object.vm_object != NULL)
3851 cred = stack_entry->object.vm_object->cred;
3852 if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
3853 rv = KERN_NO_SPACE;
3854 /* Grow the underlying object if applicable. */
3855 else if (stack_entry->object.vm_object == NULL ||
3856 vm_object_coalesce(stack_entry->object.vm_object,
3857 stack_entry->offset,
3858 (vm_size_t)(stack_entry->end - stack_entry->start),
3859 (vm_size_t)grow_amount, cred != NULL)) {
3860 if (gap_entry->start + grow_amount == gap_entry->end)
3861 vm_map_entry_delete(map, gap_entry);
3862 else
3863 gap_entry->start += grow_amount;
3864 stack_entry->end += grow_amount;
3865 map->size += grow_amount;
3866 vm_map_entry_resize_free(map, stack_entry);
3867 rv = KERN_SUCCESS;
3868 } else
3869 rv = KERN_FAILURE;
3870 }
3871 if (rv == KERN_SUCCESS && is_procstack)
3872 vm->vm_ssize += btoc(grow_amount);
3873
3874 /*
3875 * Heed the MAP_WIREFUTURE flag if it was set for this process.
3876 */
3877 if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
3878 vm_map_unlock(map);
3879 vm_map_wire(map, grow_start, grow_start + grow_amount,
3880 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
3881 vm_map_lock_read(map);
3882 } else
3883 vm_map_lock_downgrade(map);
3884
3885 out:
3886 #ifdef RACCT
3887 if (racct_enable && rv != KERN_SUCCESS) {
3888 PROC_LOCK(p);
3889 error = racct_set(p, RACCT_VMEM, map->size);
3890 KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
3891 if (!old_mlock) {
3892 error = racct_set(p, RACCT_MEMLOCK,
3893 ptoa(pmap_wired_count(map->pmap)));
3894 KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
3895 }
3896 error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
3897 KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
3898 PROC_UNLOCK(p);
3899 }
3900 #endif
3901
3902 return (rv);
3903 }
3904
3905 /*
3906 * Unshare the specified VM space for exec. If other processes are
3907 * mapped to it, then create a new one. The new vmspace is null.
3908 */
3909 int
3910 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
3911 {
3912 struct vmspace *oldvmspace = p->p_vmspace;
3913 struct vmspace *newvmspace;
3914
3915 KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
3916 ("vmspace_exec recursed"));
3917 newvmspace = vmspace_alloc(minuser, maxuser, NULL);
3918 if (newvmspace == NULL)
3919 return (ENOMEM);
3920 newvmspace->vm_swrss = oldvmspace->vm_swrss;
3921 /*
3922 * This code is written like this for prototype purposes. The
3923 * goal is to avoid running down the vmspace here, but let the
3924 * other process's that are still using the vmspace to finally
3925 * run it down. Even though there is little or no chance of blocking
3926 * here, it is a good idea to keep this form for future mods.
3927 */
3928 PROC_VMSPACE_LOCK(p);
3929 p->p_vmspace = newvmspace;
3930 PROC_VMSPACE_UNLOCK(p);
3931 if (p == curthread->td_proc)
3932 pmap_activate(curthread);
3933 curthread->td_pflags |= TDP_EXECVMSPC;
3934 return (0);
3935 }
3936
3937 /*
3938 * Unshare the specified VM space for forcing COW. This
3939 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3940 */
3941 int
3942 vmspace_unshare(struct proc *p)
3943 {
3944 struct vmspace *oldvmspace = p->p_vmspace;
3945 struct vmspace *newvmspace;
3946 vm_ooffset_t fork_charge;
3947
3948 if (oldvmspace->vm_refcnt == 1)
3949 return (0);
3950 fork_charge = 0;
3951 newvmspace = vmspace_fork(oldvmspace, &fork_charge);
3952 if (newvmspace == NULL)
3953 return (ENOMEM);
3954 if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
3955 vmspace_free(newvmspace);
3956 return (ENOMEM);
3957 }
3958 PROC_VMSPACE_LOCK(p);
3959 p->p_vmspace = newvmspace;
3960 PROC_VMSPACE_UNLOCK(p);
3961 if (p == curthread->td_proc)
3962 pmap_activate(curthread);
3963 vmspace_free(oldvmspace);
3964 return (0);
3965 }
3966
3967 /*
3968 * vm_map_lookup:
3969 *
3970 * Finds the VM object, offset, and
3971 * protection for a given virtual address in the
3972 * specified map, assuming a page fault of the
3973 * type specified.
3974 *
3975 * Leaves the map in question locked for read; return
3976 * values are guaranteed until a vm_map_lookup_done
3977 * call is performed. Note that the map argument
3978 * is in/out; the returned map must be used in
3979 * the call to vm_map_lookup_done.
3980 *
3981 * A handle (out_entry) is returned for use in
3982 * vm_map_lookup_done, to make that fast.
3983 *
3984 * If a lookup is requested with "write protection"
3985 * specified, the map may be changed to perform virtual
3986 * copying operations, although the data referenced will
3987 * remain the same.
3988 */
3989 int
3990 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
3991 vm_offset_t vaddr,
3992 vm_prot_t fault_typea,
3993 vm_map_entry_t *out_entry, /* OUT */
3994 vm_object_t *object, /* OUT */
3995 vm_pindex_t *pindex, /* OUT */
3996 vm_prot_t *out_prot, /* OUT */
3997 boolean_t *wired) /* OUT */
3998 {
3999 vm_map_entry_t entry;
4000 vm_map_t map = *var_map;
4001 vm_prot_t prot;
4002 vm_prot_t fault_type = fault_typea;
4003 vm_object_t eobject;
4004 vm_size_t size;
4005 struct ucred *cred;
4006
4007 RetryLookup:
4008
4009 vm_map_lock_read(map);
4010
4011 RetryLookupLocked:
4012 /*
4013 * Lookup the faulting address.
4014 */
4015 if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4016 vm_map_unlock_read(map);
4017 return (KERN_INVALID_ADDRESS);
4018 }
4019
4020 entry = *out_entry;
4021
4022 /*
4023 * Handle submaps.
4024 */
4025 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4026 vm_map_t old_map = map;
4027
4028 *var_map = map = entry->object.sub_map;
4029 vm_map_unlock_read(old_map);
4030 goto RetryLookup;
4031 }
4032
4033 /*
4034 * Check whether this task is allowed to have this page.
4035 */
4036 prot = entry->protection;
4037 if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4038 fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4039 if (prot == VM_PROT_NONE && map != kernel_map &&
4040 (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4041 (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4042 MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4043 vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4044 goto RetryLookupLocked;
4045 }
4046 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4047 if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4048 vm_map_unlock_read(map);
4049 return (KERN_PROTECTION_FAILURE);
4050 }
4051 KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4052 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4053 (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4054 ("entry %p flags %x", entry, entry->eflags));
4055 if ((fault_typea & VM_PROT_COPY) != 0 &&
4056 (entry->max_protection & VM_PROT_WRITE) == 0 &&
4057 (entry->eflags & MAP_ENTRY_COW) == 0) {
4058 vm_map_unlock_read(map);
4059 return (KERN_PROTECTION_FAILURE);
4060 }
4061
4062 /*
4063 * If this page is not pageable, we have to get it for all possible
4064 * accesses.
4065 */
4066 *wired = (entry->wired_count != 0);
4067 if (*wired)
4068 fault_type = entry->protection;
4069 size = entry->end - entry->start;
4070 /*
4071 * If the entry was copy-on-write, we either ...
4072 */
4073 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4074 /*
4075 * If we want to write the page, we may as well handle that
4076 * now since we've got the map locked.
4077 *
4078 * If we don't need to write the page, we just demote the
4079 * permissions allowed.
4080 */
4081 if ((fault_type & VM_PROT_WRITE) != 0 ||
4082 (fault_typea & VM_PROT_COPY) != 0) {
4083 /*
4084 * Make a new object, and place it in the object
4085 * chain. Note that no new references have appeared
4086 * -- one just moved from the map to the new
4087 * object.
4088 */
4089 if (vm_map_lock_upgrade(map))
4090 goto RetryLookup;
4091
4092 if (entry->cred == NULL) {
4093 /*
4094 * The debugger owner is charged for
4095 * the memory.
4096 */
4097 cred = curthread->td_ucred;
4098 crhold(cred);
4099 if (!swap_reserve_by_cred(size, cred)) {
4100 crfree(cred);
4101 vm_map_unlock(map);
4102 return (KERN_RESOURCE_SHORTAGE);
4103 }
4104 entry->cred = cred;
4105 }
4106 vm_object_shadow(&entry->object.vm_object,
4107 &entry->offset, size);
4108 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4109 eobject = entry->object.vm_object;
4110 if (eobject->cred != NULL) {
4111 /*
4112 * The object was not shadowed.
4113 */
4114 swap_release_by_cred(size, entry->cred);
4115 crfree(entry->cred);
4116 entry->cred = NULL;
4117 } else if (entry->cred != NULL) {
4118 VM_OBJECT_WLOCK(eobject);
4119 eobject->cred = entry->cred;
4120 eobject->charge = size;
4121 VM_OBJECT_WUNLOCK(eobject);
4122 entry->cred = NULL;
4123 }
4124
4125 vm_map_lock_downgrade(map);
4126 } else {
4127 /*
4128 * We're attempting to read a copy-on-write page --
4129 * don't allow writes.
4130 */
4131 prot &= ~VM_PROT_WRITE;
4132 }
4133 }
4134
4135 /*
4136 * Create an object if necessary.
4137 */
4138 if (entry->object.vm_object == NULL &&
4139 !map->system_map) {
4140 if (vm_map_lock_upgrade(map))
4141 goto RetryLookup;
4142 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
4143 atop(size));
4144 entry->offset = 0;
4145 if (entry->cred != NULL) {
4146 VM_OBJECT_WLOCK(entry->object.vm_object);
4147 entry->object.vm_object->cred = entry->cred;
4148 entry->object.vm_object->charge = size;
4149 VM_OBJECT_WUNLOCK(entry->object.vm_object);
4150 entry->cred = NULL;
4151 }
4152 vm_map_lock_downgrade(map);
4153 }
4154
4155 /*
4156 * Return the object/offset from this entry. If the entry was
4157 * copy-on-write or empty, it has been fixed up.
4158 */
4159 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset);
4160 *object = entry->object.vm_object;
4161
4162 *out_prot = prot;
4163 return (KERN_SUCCESS);
4164 }
4165
4166 /*
4167 * vm_map_lookup_locked:
4168 *
4169 * Lookup the faulting address. A version of vm_map_lookup that returns
4170 * KERN_FAILURE instead of blocking on map lock or memory allocation.
4171 */
4172 int
4173 vm_map_lookup_locked(vm_map_t *var_map, /* IN/OUT */
4174 vm_offset_t vaddr,
4175 vm_prot_t fault_typea,
4176 vm_map_entry_t *out_entry, /* OUT */
4177 vm_object_t *object, /* OUT */
4178 vm_pindex_t *pindex, /* OUT */
4179 vm_prot_t *out_prot, /* OUT */
4180 boolean_t *wired) /* OUT */
4181 {
4182 vm_map_entry_t entry;
4183 vm_map_t map = *var_map;
4184 vm_prot_t prot;
4185 vm_prot_t fault_type = fault_typea;
4186
4187 /*
4188 * Lookup the faulting address.
4189 */
4190 if (!vm_map_lookup_entry(map, vaddr, out_entry))
4191 return (KERN_INVALID_ADDRESS);
4192
4193 entry = *out_entry;
4194
4195 /*
4196 * Fail if the entry refers to a submap.
4197 */
4198 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4199 return (KERN_FAILURE);
4200
4201 /*
4202 * Check whether this task is allowed to have this page.
4203 */
4204 prot = entry->protection;
4205 fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4206 if ((fault_type & prot) != fault_type)
4207 return (KERN_PROTECTION_FAILURE);
4208
4209 /*
4210 * If this page is not pageable, we have to get it for all possible
4211 * accesses.
4212 */
4213 *wired = (entry->wired_count != 0);
4214 if (*wired)
4215 fault_type = entry->protection;
4216
4217 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4218 /*
4219 * Fail if the entry was copy-on-write for a write fault.
4220 */
4221 if (fault_type & VM_PROT_WRITE)
4222 return (KERN_FAILURE);
4223 /*
4224 * We're attempting to read a copy-on-write page --
4225 * don't allow writes.
4226 */
4227 prot &= ~VM_PROT_WRITE;
4228 }
4229
4230 /*
4231 * Fail if an object should be created.
4232 */
4233 if (entry->object.vm_object == NULL && !map->system_map)
4234 return (KERN_FAILURE);
4235
4236 /*
4237 * Return the object/offset from this entry. If the entry was
4238 * copy-on-write or empty, it has been fixed up.
4239 */
4240 *pindex = UOFF_TO_IDX((vaddr - entry->start) + entry->offset);
4241 *object = entry->object.vm_object;
4242
4243 *out_prot = prot;
4244 return (KERN_SUCCESS);
4245 }
4246
4247 /*
4248 * vm_map_lookup_done:
4249 *
4250 * Releases locks acquired by a vm_map_lookup
4251 * (according to the handle returned by that lookup).
4252 */
4253 void
4254 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4255 {
4256 /*
4257 * Unlock the main-level map
4258 */
4259 vm_map_unlock_read(map);
4260 }
4261
4262 #include "opt_ddb.h"
4263 #ifdef DDB
4264 #include <sys/kernel.h>
4265
4266 #include <ddb/ddb.h>
4267
4268 static void
4269 vm_map_print(vm_map_t map)
4270 {
4271 vm_map_entry_t entry;
4272
4273 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
4274 (void *)map,
4275 (void *)map->pmap, map->nentries, map->timestamp);
4276
4277 db_indent += 2;
4278 for (entry = map->header.next; entry != &map->header;
4279 entry = entry->next) {
4280 db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
4281 (void *)entry, (void *)entry->start, (void *)entry->end,
4282 entry->eflags);
4283 {
4284 static char *inheritance_name[4] =
4285 {"share", "copy", "none", "donate_copy"};
4286
4287 db_iprintf(" prot=%x/%x/%s",
4288 entry->protection,
4289 entry->max_protection,
4290 inheritance_name[(int)(unsigned char)entry->inheritance]);
4291 if (entry->wired_count != 0)
4292 db_printf(", wired");
4293 }
4294 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4295 db_printf(", share=%p, offset=0x%jx\n",
4296 (void *)entry->object.sub_map,
4297 (uintmax_t)entry->offset);
4298 if ((entry->prev == &map->header) ||
4299 (entry->prev->object.sub_map !=
4300 entry->object.sub_map)) {
4301 db_indent += 2;
4302 vm_map_print((vm_map_t)entry->object.sub_map);
4303 db_indent -= 2;
4304 }
4305 } else {
4306 if (entry->cred != NULL)
4307 db_printf(", ruid %d", entry->cred->cr_ruid);
4308 db_printf(", object=%p, offset=0x%jx",
4309 (void *)entry->object.vm_object,
4310 (uintmax_t)entry->offset);
4311 if (entry->object.vm_object && entry->object.vm_object->cred)
4312 db_printf(", obj ruid %d charge %jx",
4313 entry->object.vm_object->cred->cr_ruid,
4314 (uintmax_t)entry->object.vm_object->charge);
4315 if (entry->eflags & MAP_ENTRY_COW)
4316 db_printf(", copy (%s)",
4317 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
4318 db_printf("\n");
4319
4320 if ((entry->prev == &map->header) ||
4321 (entry->prev->object.vm_object !=
4322 entry->object.vm_object)) {
4323 db_indent += 2;
4324 vm_object_print((db_expr_t)(intptr_t)
4325 entry->object.vm_object,
4326 0, 0, (char *)0);
4327 db_indent -= 2;
4328 }
4329 }
4330 }
4331 db_indent -= 2;
4332 }
4333
4334 DB_SHOW_COMMAND(map, map)
4335 {
4336
4337 if (!have_addr) {
4338 db_printf("usage: show map <addr>\n");
4339 return;
4340 }
4341 vm_map_print((vm_map_t)addr);
4342 }
4343
4344 DB_SHOW_COMMAND(procvm, procvm)
4345 {
4346 struct proc *p;
4347
4348 if (have_addr) {
4349 p = db_lookup_proc(addr);
4350 } else {
4351 p = curproc;
4352 }
4353
4354 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
4355 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
4356 (void *)vmspace_pmap(p->p_vmspace));
4357
4358 vm_map_print((vm_map_t)&p->p_vmspace->vm_map);
4359 }
4360
4361 #endif /* DDB */
Cache object: 1c83292e457530bae75c60cf52c62d42
|