kernel - Fix lapic mask for > 16 apic ids
[dragonfly.git] / sys / vm / vm_map.c
CommitLineData
984263bc 1/*
46754a20
MD
2 * (MPSAFE)
3 *
984263bc
MD
4 * Copyright (c) 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * The Mach Operating System project at Carnegie-Mellon University.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
39 *
40 *
41 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
42 * All rights reserved.
43 *
44 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
45 *
46 * Permission to use, copy, modify and distribute this software and
47 * its documentation is hereby granted, provided that both the copyright
48 * notice and this permission notice appear in all copies of the
49 * software, derivative works or modified versions, and any portions
50 * thereof, and that both notices appear in supporting documentation.
51 *
52 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
53 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
54 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
55 *
56 * Carnegie Mellon requests users of this software to return to
57 *
58 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
59 * School of Computer Science
60 * Carnegie Mellon University
61 * Pittsburgh PA 15213-3890
62 *
63 * any improvements or extensions that they make and grant Carnegie the
64 * rights to redistribute these changes.
65 *
66 * $FreeBSD: src/sys/vm/vm_map.c,v 1.187.2.19 2003/05/27 00:47:02 alc Exp $
e3161323 67 * $DragonFly: src/sys/vm/vm_map.c,v 1.56 2007/04/29 18:25:41 dillon Exp $
984263bc
MD
68 */
69
70/*
71 * Virtual memory mapping module.
72 */
73
74#include <sys/param.h>
75#include <sys/systm.h>
e3161323 76#include <sys/kernel.h>
984263bc 77#include <sys/proc.h>
ff13bc52 78#include <sys/serialize.h>
fef0fdf2 79#include <sys/lock.h>
984263bc
MD
80#include <sys/vmmeter.h>
81#include <sys/mman.h>
82#include <sys/vnode.h>
83#include <sys/resourcevar.h>
fef0fdf2 84#include <sys/shm.h>
686dbf64 85#include <sys/tree.h>
e3161323 86#include <sys/malloc.h>
984263bc
MD
87
88#include <vm/vm.h>
89#include <vm/vm_param.h>
984263bc
MD
90#include <vm/pmap.h>
91#include <vm/vm_map.h>
92#include <vm/vm_page.h>
93#include <vm/vm_object.h>
94#include <vm/vm_pager.h>
95#include <vm/vm_kern.h>
96#include <vm/vm_extern.h>
97#include <vm/swap_pager.h>
98#include <vm/vm_zone.h>
99
a108bf71 100#include <sys/thread2.h>
e3161323 101#include <sys/sysref2.h>
911e30e2
AH
102#include <sys/random.h>
103#include <sys/sysctl.h>
a108bf71 104
984263bc 105/*
46754a20
MD
106 * Virtual memory maps provide for the mapping, protection, and sharing
107 * of virtual memory objects. In addition, this module provides for an
108 * efficient virtual copy of memory from one map to another.
984263bc 109 *
46754a20 110 * Synchronization is required prior to most operations.
984263bc 111 *
46754a20
MD
112 * Maps consist of an ordered doubly-linked list of simple entries.
113 * A hint and a RB tree is used to speed-up lookups.
984263bc 114 *
46754a20
MD
115 * Callers looking to modify maps specify start/end addresses which cause
116 * the related map entry to be clipped if necessary, and then later
117 * recombined if the pieces remained compatible.
984263bc 118 *
46754a20
MD
119 * Virtual copy operations are performed by copying VM object references
120 * from one map to another, and then marking both regions as copy-on-write.
984263bc 121 */
e3161323 122static void vmspace_terminate(struct vmspace *vm);
e654922c
MD
123static void vmspace_lock(struct vmspace *vm);
124static void vmspace_unlock(struct vmspace *vm);
e3161323
MD
125static void vmspace_dtor(void *obj, void *private);
126
127MALLOC_DEFINE(M_VMSPACE, "vmspace", "vmspace objcache backingstore");
128
129struct sysref_class vmspace_sysref_class = {
130 .name = "vmspace",
131 .mtype = M_VMSPACE,
132 .proto = SYSREF_PROTO_VMSPACE,
133 .offset = offsetof(struct vmspace, vm_sysref),
134 .objsize = sizeof(struct vmspace),
135 .mag_capacity = 32,
136 .flags = SRC_MANAGEDINIT,
137 .dtor = vmspace_dtor,
138 .ops = {
e654922c
MD
139 .terminate = (sysref_terminate_func_t)vmspace_terminate,
140 .lock = (sysref_lock_func_t)vmspace_lock,
141 .unlock = (sysref_lock_func_t)vmspace_unlock
e3161323
MD
142 }
143};
984263bc 144
c4ae567f
MD
145#define VMEPERCPU 2
146
a108bf71 147static struct vm_zone mapentzone_store, mapzone_store;
e3161323 148static vm_zone_t mapentzone, mapzone;
a108bf71 149static struct vm_object mapentobj, mapobj;
984263bc
MD
150
151static struct vm_map_entry map_entry_init[MAX_MAPENT];
c4ae567f 152static struct vm_map_entry cpu_map_entry_init[MAXCPU][VMEPERCPU];
984263bc
MD
153static struct vm_map map_init[MAX_KMAP];
154
911e30e2
AH
155static int randomize_mmap;
156SYSCTL_INT(_vm, OID_AUTO, randomize_mmap, CTLFLAG_RW, &randomize_mmap, 0,
157 "Randomize mmap offsets");
158
53025830 159static void vm_map_entry_shadow(vm_map_entry_t entry);
a108bf71
MD
160static vm_map_entry_t vm_map_entry_create(vm_map_t map, int *);
161static void vm_map_entry_dispose (vm_map_t map, vm_map_entry_t entry, int *);
162static void _vm_map_clip_end (vm_map_t, vm_map_entry_t, vm_offset_t, int *);
163static void _vm_map_clip_start (vm_map_t, vm_map_entry_t, vm_offset_t, int *);
164static void vm_map_entry_delete (vm_map_t, vm_map_entry_t, int *);
1388df65
RG
165static void vm_map_entry_unwire (vm_map_t, vm_map_entry_t);
166static void vm_map_copy_entry (vm_map_t, vm_map_t, vm_map_entry_t,
167 vm_map_entry_t);
168static void vm_map_split (vm_map_entry_t);
a108bf71 169static void vm_map_unclip_range (vm_map_t map, vm_map_entry_t start_entry, vm_offset_t start, vm_offset_t end, int *count, int flags);
984263bc 170
e3161323 171/*
46754a20
MD
172 * Initialize the vm_map module. Must be called before any other vm_map
173 * routines.
e3161323 174 *
46754a20
MD
175 * Map and entry structures are allocated from the general purpose
176 * memory pool with some exceptions:
e3161323 177 *
46754a20
MD
178 * - The kernel map is allocated statically.
179 * - Initial kernel map entries are allocated out of a static pool.
e3161323
MD
180 *
181 * These restrictions are necessary since malloc() uses the
182 * maps and requires map entries.
46754a20
MD
183 *
184 * Called from the low level boot code only.
e3161323 185 */
984263bc 186void
57e43348 187vm_map_startup(void)
984263bc
MD
188{
189 mapzone = &mapzone_store;
190 zbootinit(mapzone, "MAP", sizeof (struct vm_map),
191 map_init, MAX_KMAP);
984263bc
MD
192 mapentzone = &mapentzone_store;
193 zbootinit(mapentzone, "MAP ENTRY", sizeof (struct vm_map_entry),
194 map_entry_init, MAX_MAPENT);
195}
196
e3161323 197/*
46754a20
MD
198 * Called prior to any vmspace allocations.
199 *
200 * Called from the low level boot code only.
e3161323
MD
201 */
202void
203vm_init2(void)
204{
205 zinitna(mapentzone, &mapentobj, NULL, 0, 0,
206 ZONE_USE_RESERVE | ZONE_SPECIAL, 1);
207 zinitna(mapzone, &mapobj, NULL, 0, 0, 0, 1);
208 pmap_init2();
209 vm_object_init2();
210}
211
212
686dbf64
MD
213/*
214 * Red black tree functions
46754a20
MD
215 *
216 * The caller must hold the related map lock.
686dbf64
MD
217 */
218static int rb_vm_map_compare(vm_map_entry_t a, vm_map_entry_t b);
219RB_GENERATE(vm_map_rb_tree, vm_map_entry, rb_entry, rb_vm_map_compare);
220
221/* a->start is address, and the only field has to be initialized */
222static int
223rb_vm_map_compare(vm_map_entry_t a, vm_map_entry_t b)
224{
225 if (a->start < b->start)
226 return(-1);
227 else if (a->start > b->start)
228 return(1);
229 return(0);
230}
231
984263bc 232/*
e3161323
MD
233 * Allocate a vmspace structure, including a vm_map and pmap.
234 * Initialize numerous fields. While the initial allocation is zerod,
235 * subsequence reuse from the objcache leaves elements of the structure
236 * intact (particularly the pmap), so portions must be zerod.
237 *
238 * The structure is not considered activated until we call sysref_activate().
46754a20
MD
239 *
240 * No requirements.
984263bc
MD
241 */
242struct vmspace *
57e43348 243vmspace_alloc(vm_offset_t min, vm_offset_t max)
984263bc
MD
244{
245 struct vmspace *vm;
246
46754a20 247 lwkt_gettoken(&vmspace_token);
e3161323 248 vm = sysref_alloc(&vmspace_sysref_class);
54a764e8 249 bzero(&vm->vm_startcopy,
e3161323 250 (char *)&vm->vm_endcopy - (char *)&vm->vm_startcopy);
e4846942 251 vm_map_init(&vm->vm_map, min, max, NULL);
e3161323 252 pmap_pinit(vmspace_pmap(vm)); /* (some fields reused) */
984263bc 253 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
984263bc
MD
254 vm->vm_shm = NULL;
255 vm->vm_exitingcnt = 0;
135d7199 256 cpu_vmspace_alloc(vm);
e3161323 257 sysref_activate(&vm->vm_sysref);
46754a20
MD
258 lwkt_reltoken(&vmspace_token);
259
984263bc
MD
260 return (vm);
261}
262
e3161323
MD
263/*
264 * dtor function - Some elements of the pmap are retained in the
265 * free-cached vmspaces to improve performance. We have to clean them up
266 * here before returning the vmspace to the memory pool.
46754a20
MD
267 *
268 * No requirements.
e3161323
MD
269 */
270static void
271vmspace_dtor(void *obj, void *private)
a108bf71 272{
e3161323
MD
273 struct vmspace *vm = obj;
274
275 pmap_puninit(vmspace_pmap(vm));
984263bc
MD
276}
277
e3161323
MD
278/*
279 * Called in two cases:
280 *
281 * (1) When the last sysref is dropped, but exitingcnt might still be
282 * non-zero.
283 *
284 * (2) When there are no sysrefs (i.e. refcnt is negative) left and the
285 * exitingcnt becomes zero
286 *
287 * sysref will not scrap the object until we call sysref_put() once more
288 * after the last ref has been dropped.
46754a20
MD
289 *
290 * Interlocked by the sysref API.
e3161323
MD
291 */
292static void
293vmspace_terminate(struct vmspace *vm)
984263bc 294{
a108bf71
MD
295 int count;
296
e3161323
MD
297 /*
298 * If exitingcnt is non-zero we can't get rid of the entire vmspace
299 * yet, but we can scrap user memory.
300 */
46754a20 301 lwkt_gettoken(&vmspace_token);
e3161323
MD
302 if (vm->vm_exitingcnt) {
303 shmexit(vm);
304 pmap_remove_pages(vmspace_pmap(vm), VM_MIN_USER_ADDRESS,
305 VM_MAX_USER_ADDRESS);
306 vm_map_remove(&vm->vm_map, VM_MIN_USER_ADDRESS,
307 VM_MAX_USER_ADDRESS);
46754a20 308 lwkt_reltoken(&vmspace_token);
e3161323
MD
309 return;
310 }
135d7199
MD
311 cpu_vmspace_free(vm);
312
fef0fdf2
MD
313 /*
314 * Make sure any SysV shm is freed, it might not have in
315 * exit1()
316 */
317 shmexit(vm);
318
a722be49
MD
319 KKASSERT(vm->vm_upcalls == NULL);
320
984263bc
MD
321 /*
322 * Lock the map, to wait out all other references to it.
323 * Delete all of the mappings and pages they hold, then call
324 * the pmap module to reclaim anything left.
325 */
a108bf71 326 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc 327 vm_map_lock(&vm->vm_map);
a108bf71
MD
328 vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
329 vm->vm_map.max_offset, &count);
984263bc 330 vm_map_unlock(&vm->vm_map);
a108bf71 331 vm_map_entry_release(count);
984263bc
MD
332
333 pmap_release(vmspace_pmap(vm));
e3161323 334 sysref_put(&vm->vm_sysref);
46754a20 335 lwkt_reltoken(&vmspace_token);
984263bc
MD
336}
337
46754a20
MD
338/*
339 * vmspaces are not currently locked.
340 */
e654922c
MD
341static void
342vmspace_lock(struct vmspace *vm __unused)
343{
344}
345
346static void
347vmspace_unlock(struct vmspace *vm __unused)
348{
349}
350
46754a20
MD
351/*
352 * This is called during exit indicating that the vmspace is no
353 * longer in used by an exiting process, but the process has not yet
354 * been cleaned up.
355 *
356 * No requirements.
357 */
358void
359vmspace_exitbump(struct vmspace *vm)
360{
361 lwkt_gettoken(&vmspace_token);
362 ++vm->vm_exitingcnt;
363 lwkt_reltoken(&vmspace_token);
364}
365
e3161323
MD
366/*
367 * This is called in the wait*() handling code. The vmspace can be terminated
368 * after the last wait is finished using it.
46754a20
MD
369 *
370 * No requirements.
e3161323 371 */
984263bc
MD
372void
373vmspace_exitfree(struct proc *p)
374{
375 struct vmspace *vm;
376
46754a20 377 lwkt_gettoken(&vmspace_token);
984263bc
MD
378 vm = p->p_vmspace;
379 p->p_vmspace = NULL;
380
e3161323
MD
381 if (--vm->vm_exitingcnt == 0 && sysref_isinactive(&vm->vm_sysref))
382 vmspace_terminate(vm);
46754a20 383 lwkt_reltoken(&vmspace_token);
984263bc
MD
384}
385
386/*
46754a20
MD
387 * Swap useage is determined by taking the proportional swap used by
388 * VM objects backing the VM map. To make up for fractional losses,
389 * if the VM object has any swap use at all the associated map entries
390 * count for at least 1 swap page.
984263bc 391 *
46754a20 392 * No requirements.
984263bc
MD
393 */
394int
395vmspace_swap_count(struct vmspace *vmspace)
396{
397 vm_map_t map = &vmspace->vm_map;
398 vm_map_entry_t cur;
1b874851 399 vm_object_t object;
984263bc 400 int count = 0;
1b874851 401 int n;
984263bc 402
46754a20 403 lwkt_gettoken(&vmspace_token);
984263bc 404 for (cur = map->header.next; cur != &map->header; cur = cur->next) {
1b874851
MD
405 switch(cur->maptype) {
406 case VM_MAPTYPE_NORMAL:
407 case VM_MAPTYPE_VPAGETABLE:
408 if ((object = cur->object.vm_object) == NULL)
409 break;
96adc753
MD
410 if (object->swblock_count) {
411 n = (cur->end - cur->start) / PAGE_SIZE;
412 count += object->swblock_count *
984263bc
MD
413 SWAP_META_PAGES * n / object->size + 1;
414 }
1b874851
MD
415 break;
416 default:
417 break;
984263bc
MD
418 }
419 }
46754a20 420 lwkt_reltoken(&vmspace_token);
984263bc
MD
421 return(count);
422}
423
20479584 424/*
46754a20
MD
425 * Calculate the approximate number of anonymous pages in use by
426 * this vmspace. To make up for fractional losses, we count each
427 * VM object as having at least 1 anonymous page.
20479584 428 *
46754a20 429 * No requirements.
20479584
MD
430 */
431int
432vmspace_anonymous_count(struct vmspace *vmspace)
433{
434 vm_map_t map = &vmspace->vm_map;
435 vm_map_entry_t cur;
436 vm_object_t object;
437 int count = 0;
438
46754a20 439 lwkt_gettoken(&vmspace_token);
20479584
MD
440 for (cur = map->header.next; cur != &map->header; cur = cur->next) {
441 switch(cur->maptype) {
442 case VM_MAPTYPE_NORMAL:
443 case VM_MAPTYPE_VPAGETABLE:
444 if ((object = cur->object.vm_object) == NULL)
445 break;
446 if (object->type != OBJT_DEFAULT &&
447 object->type != OBJT_SWAP) {
448 break;
449 }
450 count += object->resident_page_count;
451 break;
452 default:
453 break;
454 }
455 }
46754a20 456 lwkt_reltoken(&vmspace_token);
20479584
MD
457 return(count);
458}
459
984263bc 460/*
46754a20
MD
461 * Creates and returns a new empty VM map with the given physical map
462 * structure, and having the given lower and upper address bounds.
984263bc 463 *
46754a20 464 * No requirements.
984263bc
MD
465 */
466vm_map_t
e4846942 467vm_map_create(vm_map_t result, pmap_t pmap, vm_offset_t min, vm_offset_t max)
984263bc 468{
e4846942
MD
469 if (result == NULL)
470 result = zalloc(mapzone);
471 vm_map_init(result, min, max, pmap);
984263bc
MD
472 return (result);
473}
474
475/*
46754a20
MD
476 * Initialize an existing vm_map structure such as that in the vmspace
477 * structure. The pmap is initialized elsewhere.
478 *
479 * No requirements.
984263bc
MD
480 */
481void
e4846942 482vm_map_init(struct vm_map *map, vm_offset_t min, vm_offset_t max, pmap_t pmap)
984263bc
MD
483{
484 map->header.next = map->header.prev = &map->header;
686dbf64 485 RB_INIT(&map->rb_root);
984263bc
MD
486 map->nentries = 0;
487 map->size = 0;
488 map->system_map = 0;
489 map->infork = 0;
490 map->min_offset = min;
491 map->max_offset = max;
e4846942 492 map->pmap = pmap;
984263bc
MD
493 map->first_free = &map->header;
494 map->hint = &map->header;
495 map->timestamp = 0;
69e16e2a 496 map->flags = 0;
ab6f251b 497 lockinit(&map->lock, "thrd_sleep", 0, 0);
984263bc
MD
498}
499
53025830
MD
500/*
501 * Shadow the vm_map_entry's object. This typically needs to be done when
502 * a write fault is taken on an entry which had previously been cloned by
503 * fork(). The shared object (which might be NULL) must become private so
504 * we add a shadow layer above it.
505 *
506 * Object allocation for anonymous mappings is defered as long as possible.
507 * When creating a shadow, however, the underlying object must be instantiated
508 * so it can be shared.
509 *
510 * If the map segment is governed by a virtual page table then it is
511 * possible to address offsets beyond the mapped area. Just allocate
512 * a maximally sized object for this case.
46754a20
MD
513 *
514 * The vm_map must be exclusively locked.
515 * No other requirements.
53025830
MD
516 */
517static
518void
519vm_map_entry_shadow(vm_map_entry_t entry)
520{
521 if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
522 vm_object_shadow(&entry->object.vm_object, &entry->offset,
523 0x7FFFFFFF); /* XXX */
524 } else {
525 vm_object_shadow(&entry->object.vm_object, &entry->offset,
526 atop(entry->end - entry->start));
527 }
528 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
529}
530
531/*
532 * Allocate an object for a vm_map_entry.
533 *
534 * Object allocation for anonymous mappings is defered as long as possible.
535 * This function is called when we can defer no longer, generally when a map
536 * entry might be split or forked or takes a page fault.
537 *
538 * If the map segment is governed by a virtual page table then it is
539 * possible to address offsets beyond the mapped area. Just allocate
540 * a maximally sized object for this case.
46754a20
MD
541 *
542 * The vm_map must be exclusively locked.
543 * No other requirements.
53025830
MD
544 */
545void
546vm_map_entry_allocate_object(vm_map_entry_t entry)
547{
548 vm_object_t obj;
549
550 if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
551 obj = vm_object_allocate(OBJT_DEFAULT, 0x7FFFFFFF); /* XXX */
552 } else {
553 obj = vm_object_allocate(OBJT_DEFAULT,
554 atop(entry->end - entry->start));
555 }
556 entry->object.vm_object = obj;
557 entry->offset = 0;
558}
559
41a01a4d 560/*
46754a20
MD
561 * Set an initial negative count so the first attempt to reserve
562 * space preloads a bunch of vm_map_entry's for this cpu. Also
563 * pre-allocate 2 vm_map_entries which will be needed by zalloc() to
564 * map a new page for vm_map_entry structures. SMP systems are
565 * particularly sensitive.
c4ae567f 566 *
46754a20
MD
567 * This routine is called in early boot so we cannot just call
568 * vm_map_entry_reserve().
41a01a4d 569 *
46754a20 570 * Called from the low level boot code only (for each cpu)
41a01a4d
MD
571 */
572void
573vm_map_entry_reserve_cpu_init(globaldata_t gd)
574{
c4ae567f
MD
575 vm_map_entry_t entry;
576 int i;
577
41a01a4d 578 gd->gd_vme_avail -= MAP_RESERVE_COUNT * 2;
c4ae567f
MD
579 entry = &cpu_map_entry_init[gd->gd_cpuid][0];
580 for (i = 0; i < VMEPERCPU; ++i, ++entry) {
581 entry->next = gd->gd_vme_base;
582 gd->gd_vme_base = entry;
583 }
41a01a4d
MD
584}
585
a108bf71 586/*
46754a20
MD
587 * Reserves vm_map_entry structures so code later on can manipulate
588 * map_entry structures within a locked map without blocking trying
589 * to allocate a new vm_map_entry.
a108bf71 590 *
46754a20 591 * No requirements.
a108bf71
MD
592 */
593int
594vm_map_entry_reserve(int count)
595{
596 struct globaldata *gd = mycpu;
597 vm_map_entry_t entry;
598
a108bf71
MD
599 /*
600 * Make sure we have enough structures in gd_vme_base to handle
601 * the reservation request.
602 */
46754a20 603 crit_enter();
ac13eccd 604 while (gd->gd_vme_avail < count) {
a108bf71
MD
605 entry = zalloc(mapentzone);
606 entry->next = gd->gd_vme_base;
607 gd->gd_vme_base = entry;
608 ++gd->gd_vme_avail;
609 }
ac13eccd 610 gd->gd_vme_avail -= count;
a108bf71 611 crit_exit();
46754a20 612
a108bf71
MD
613 return(count);
614}
615
616/*
46754a20
MD
617 * Releases previously reserved vm_map_entry structures that were not
618 * used. If we have too much junk in our per-cpu cache clean some of
619 * it out.
a108bf71 620 *
46754a20 621 * No requirements.
a108bf71
MD
622 */
623void
624vm_map_entry_release(int count)
625{
626 struct globaldata *gd = mycpu;
627 vm_map_entry_t entry;
628
629 crit_enter();
630 gd->gd_vme_avail += count;
631 while (gd->gd_vme_avail > MAP_RESERVE_SLOP) {
632 entry = gd->gd_vme_base;
633 KKASSERT(entry != NULL);
634 gd->gd_vme_base = entry->next;
635 --gd->gd_vme_avail;
636 crit_exit();
637 zfree(mapentzone, entry);
638 crit_enter();
639 }
640 crit_exit();
641}
642
643/*
46754a20
MD
644 * Reserve map entry structures for use in kernel_map itself. These
645 * entries have *ALREADY* been reserved on a per-cpu basis when the map
646 * was inited. This function is used by zalloc() to avoid a recursion
647 * when zalloc() itself needs to allocate additional kernel memory.
a108bf71 648 *
46754a20
MD
649 * This function works like the normal reserve but does not load the
650 * vm_map_entry cache (because that would result in an infinite
651 * recursion). Note that gd_vme_avail may go negative. This is expected.
c4ae567f 652 *
46754a20
MD
653 * Any caller of this function must be sure to renormalize after
654 * potentially eating entries to ensure that the reserve supply
655 * remains intact.
a108bf71 656 *
46754a20 657 * No requirements.
a108bf71
MD
658 */
659int
660vm_map_entry_kreserve(int count)
661{
662 struct globaldata *gd = mycpu;
663
664 crit_enter();
c4ae567f 665 gd->gd_vme_avail -= count;
a108bf71 666 crit_exit();
46754a20
MD
667 KASSERT(gd->gd_vme_base != NULL,
668 ("no reserved entries left, gd_vme_avail = %d\n",
669 gd->gd_vme_avail));
a108bf71
MD
670 return(count);
671}
672
673/*
46754a20
MD
674 * Release previously reserved map entries for kernel_map. We do not
675 * attempt to clean up like the normal release function as this would
676 * cause an unnecessary (but probably not fatal) deep procedure call.
a108bf71 677 *
46754a20 678 * No requirements.
a108bf71
MD
679 */
680void
681vm_map_entry_krelease(int count)
682{
683 struct globaldata *gd = mycpu;
684
685 crit_enter();
c4ae567f 686 gd->gd_vme_avail += count;
a108bf71
MD
687 crit_exit();
688}
689
984263bc 690/*
46754a20 691 * Allocates a VM map entry for insertion. No entry fields are filled in.
984263bc 692 *
46754a20
MD
693 * The entries should have previously been reserved. The reservation count
694 * is tracked in (*countp).
a108bf71 695 *
46754a20 696 * No requirements.
984263bc 697 */
8a8d5d85 698static vm_map_entry_t
a108bf71 699vm_map_entry_create(vm_map_t map, int *countp)
984263bc 700{
a108bf71
MD
701 struct globaldata *gd = mycpu;
702 vm_map_entry_t entry;
8a8d5d85 703
a108bf71
MD
704 KKASSERT(*countp > 0);
705 --*countp;
706 crit_enter();
707 entry = gd->gd_vme_base;
708 KASSERT(entry != NULL, ("gd_vme_base NULL! count %d", *countp));
709 gd->gd_vme_base = entry->next;
710 crit_exit();
46754a20 711
a108bf71 712 return(entry);
984263bc
MD
713}
714
715/*
46754a20 716 * Dispose of a vm_map_entry that is no longer being referenced.
984263bc 717 *
46754a20 718 * No requirements.
984263bc 719 */
8a8d5d85 720static void
a108bf71 721vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc 722{
a108bf71
MD
723 struct globaldata *gd = mycpu;
724
686dbf64
MD
725 KKASSERT(map->hint != entry);
726 KKASSERT(map->first_free != entry);
727
a108bf71
MD
728 ++*countp;
729 crit_enter();
730 entry->next = gd->gd_vme_base;
731 gd->gd_vme_base = entry;
732 crit_exit();
984263bc
MD
733}
734
8a8d5d85 735
984263bc 736/*
46754a20 737 * Insert/remove entries from maps.
984263bc 738 *
46754a20
MD
739 * The related map must be exclusively locked.
740 * No other requirements.
741 *
742 * NOTE! We currently acquire the vmspace_token only to avoid races
743 * against the pageout daemon's calls to vmspace_*_count(), which
744 * are unable to safely lock the vm_map without potentially
745 * deadlocking.
984263bc
MD
746 */
747static __inline void
748vm_map_entry_link(vm_map_t map,
749 vm_map_entry_t after_where,
750 vm_map_entry_t entry)
751{
46754a20
MD
752 ASSERT_VM_MAP_LOCKED(map);
753
754 lwkt_gettoken(&vmspace_token);
984263bc
MD
755 map->nentries++;
756 entry->prev = after_where;
757 entry->next = after_where->next;
758 entry->next->prev = entry;
759 after_where->next = entry;
0cd275af
MD
760 if (vm_map_rb_tree_RB_INSERT(&map->rb_root, entry))
761 panic("vm_map_entry_link: dup addr map %p ent %p", map, entry);
46754a20 762 lwkt_reltoken(&vmspace_token);
984263bc
MD
763}
764
765static __inline void
766vm_map_entry_unlink(vm_map_t map,
767 vm_map_entry_t entry)
768{
769 vm_map_entry_t prev;
770 vm_map_entry_t next;
771
46754a20
MD
772 ASSERT_VM_MAP_LOCKED(map);
773
774 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
775 panic("vm_map_entry_unlink: attempt to mess with "
776 "locked entry! %p", entry);
777 }
778 lwkt_gettoken(&vmspace_token);
984263bc
MD
779 prev = entry->prev;
780 next = entry->next;
781 next->prev = prev;
782 prev->next = next;
686dbf64 783 vm_map_rb_tree_RB_REMOVE(&map->rb_root, entry);
984263bc 784 map->nentries--;
46754a20 785 lwkt_reltoken(&vmspace_token);
984263bc
MD
786}
787
984263bc 788/*
46754a20
MD
789 * Finds the map entry containing (or immediately preceding) the specified
790 * address in the given map. The entry is returned in (*entry).
791 *
792 * The boolean result indicates whether the address is actually contained
793 * in the map.
984263bc 794 *
46754a20
MD
795 * The related map must be locked.
796 * No other requirements.
984263bc
MD
797 */
798boolean_t
46754a20 799vm_map_lookup_entry(vm_map_t map, vm_offset_t address, vm_map_entry_t *entry)
984263bc 800{
686dbf64 801 vm_map_entry_t tmp;
984263bc
MD
802 vm_map_entry_t last;
803
46754a20 804 ASSERT_VM_MAP_LOCKED(map);
686dbf64 805#if 0
984263bc 806 /*
686dbf64
MD
807 * XXX TEMPORARILY DISABLED. For some reason our attempt to revive
808 * the hint code with the red-black lookup meets with system crashes
809 * and lockups. We do not yet know why.
810 *
811 * It is possible that the problem is related to the setting
812 * of the hint during map_entry deletion, in the code specified
813 * at the GGG comment later on in this file.
984263bc 814 */
686dbf64
MD
815 /*
816 * Quickly check the cached hint, there's a good chance of a match.
817 */
818 if (map->hint != &map->header) {
819 tmp = map->hint;
820 if (address >= tmp->start && address < tmp->end) {
821 *entry = tmp;
822 return(TRUE);
984263bc 823 }
984263bc 824 }
686dbf64 825#endif
984263bc
MD
826
827 /*
686dbf64
MD
828 * Locate the record from the top of the tree. 'last' tracks the
829 * closest prior record and is returned if no match is found, which
830 * in binary tree terms means tracking the most recent right-branch
831 * taken. If there is no prior record, &map->header is returned.
984263bc 832 */
686dbf64
MD
833 last = &map->header;
834 tmp = RB_ROOT(&map->rb_root);
835
836 while (tmp) {
837 if (address >= tmp->start) {
838 if (address < tmp->end) {
839 *entry = tmp;
840 map->hint = tmp;
841 return(TRUE);
984263bc 842 }
686dbf64
MD
843 last = tmp;
844 tmp = RB_RIGHT(tmp, rb_entry);
845 } else {
846 tmp = RB_LEFT(tmp, rb_entry);
984263bc 847 }
984263bc 848 }
686dbf64 849 *entry = last;
984263bc
MD
850 return (FALSE);
851}
852
853/*
46754a20
MD
854 * Inserts the given whole VM object into the target map at the specified
855 * address range. The object's size should match that of the address range.
984263bc 856 *
46754a20
MD
857 * The map must be exclusively locked.
858 * The caller must have reserved sufficient vm_map_entry structures.
984263bc 859 *
46754a20
MD
860 * If object is non-NULL, ref count must be bumped by caller
861 * prior to making call to account for the new entry.
984263bc
MD
862 */
863int
a108bf71
MD
864vm_map_insert(vm_map_t map, int *countp,
865 vm_object_t object, vm_ooffset_t offset,
1b874851
MD
866 vm_offset_t start, vm_offset_t end,
867 vm_maptype_t maptype,
868 vm_prot_t prot, vm_prot_t max,
984263bc
MD
869 int cow)
870{
871 vm_map_entry_t new_entry;
872 vm_map_entry_t prev_entry;
873 vm_map_entry_t temp_entry;
874 vm_eflags_t protoeflags;
875
46754a20
MD
876 ASSERT_VM_MAP_LOCKED(map);
877
984263bc
MD
878 /*
879 * Check that the start and end points are not bogus.
880 */
984263bc
MD
881 if ((start < map->min_offset) || (end > map->max_offset) ||
882 (start >= end))
883 return (KERN_INVALID_ADDRESS);
884
885 /*
886 * Find the entry prior to the proposed starting address; if it's part
887 * of an existing entry, this range is bogus.
888 */
984263bc
MD
889 if (vm_map_lookup_entry(map, start, &temp_entry))
890 return (KERN_NO_SPACE);
891
892 prev_entry = temp_entry;
893
894 /*
895 * Assert that the next entry doesn't overlap the end point.
896 */
897
898 if ((prev_entry->next != &map->header) &&
899 (prev_entry->next->start < end))
900 return (KERN_NO_SPACE);
901
902 protoeflags = 0;
903
904 if (cow & MAP_COPY_ON_WRITE)
905 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
906
907 if (cow & MAP_NOFAULT) {
908 protoeflags |= MAP_ENTRY_NOFAULT;
909
910 KASSERT(object == NULL,
911 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
912 }
913 if (cow & MAP_DISABLE_SYNCER)
914 protoeflags |= MAP_ENTRY_NOSYNC;
915 if (cow & MAP_DISABLE_COREDUMP)
916 protoeflags |= MAP_ENTRY_NOCOREDUMP;
c809941b
MD
917 if (cow & MAP_IS_STACK)
918 protoeflags |= MAP_ENTRY_STACK;
e40cfbd7
MD
919 if (cow & MAP_IS_KSTACK)
920 protoeflags |= MAP_ENTRY_KSTACK;
984263bc 921
2de4f77e
MD
922 lwkt_gettoken(&vm_token);
923 lwkt_gettoken(&vmobj_token);
924
984263bc
MD
925 if (object) {
926 /*
927 * When object is non-NULL, it could be shared with another
928 * process. We have to set or clear OBJ_ONEMAPPING
929 * appropriately.
930 */
931 if ((object->ref_count > 1) || (object->shadow_count != 0)) {
932 vm_object_clear_flag(object, OBJ_ONEMAPPING);
933 }
934 }
935 else if ((prev_entry != &map->header) &&
936 (prev_entry->eflags == protoeflags) &&
937 (prev_entry->end == start) &&
938 (prev_entry->wired_count == 0) &&
1b874851 939 prev_entry->maptype == maptype &&
984263bc
MD
940 ((prev_entry->object.vm_object == NULL) ||
941 vm_object_coalesce(prev_entry->object.vm_object,
942 OFF_TO_IDX(prev_entry->offset),
943 (vm_size_t)(prev_entry->end - prev_entry->start),
944 (vm_size_t)(end - prev_entry->end)))) {
945 /*
946 * We were able to extend the object. Determine if we
947 * can extend the previous map entry to include the
948 * new range as well.
949 */
950 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
951 (prev_entry->protection == prot) &&
952 (prev_entry->max_protection == max)) {
2de4f77e
MD
953 lwkt_reltoken(&vmobj_token);
954 lwkt_reltoken(&vm_token);
984263bc
MD
955 map->size += (end - prev_entry->end);
956 prev_entry->end = end;
a108bf71 957 vm_map_simplify_entry(map, prev_entry, countp);
984263bc
MD
958 return (KERN_SUCCESS);
959 }
960
961 /*
962 * If we can extend the object but cannot extend the
963 * map entry, we have to create a new map entry. We
964 * must bump the ref count on the extended object to
965 * account for it. object may be NULL.
966 */
967 object = prev_entry->object.vm_object;
968 offset = prev_entry->offset +
969 (prev_entry->end - prev_entry->start);
2de4f77e 970 vm_object_reference_locked(object);
984263bc
MD
971 }
972
2de4f77e
MD
973 lwkt_reltoken(&vmobj_token);
974 lwkt_reltoken(&vm_token);
975
984263bc
MD
976 /*
977 * NOTE: if conditionals fail, object can be NULL here. This occurs
978 * in things like the buffer map where we manage kva but do not manage
979 * backing objects.
980 */
981
982 /*
983 * Create a new entry
984 */
985
a108bf71 986 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
987 new_entry->start = start;
988 new_entry->end = end;
989
1b874851 990 new_entry->maptype = maptype;
984263bc
MD
991 new_entry->eflags = protoeflags;
992 new_entry->object.vm_object = object;
993 new_entry->offset = offset;
afeabdca 994 new_entry->aux.master_pde = 0;
984263bc
MD
995
996 new_entry->inheritance = VM_INHERIT_DEFAULT;
997 new_entry->protection = prot;
998 new_entry->max_protection = max;
999 new_entry->wired_count = 0;
1000
1001 /*
1002 * Insert the new entry into the list
1003 */
1004
1005 vm_map_entry_link(map, prev_entry, new_entry);
1006 map->size += new_entry->end - new_entry->start;
1007
1008 /*
791c6551
MD
1009 * Update the free space hint. Entries cannot overlap.
1010 * An exact comparison is needed to avoid matching
1011 * against the map->header.
984263bc
MD
1012 */
1013 if ((map->first_free == prev_entry) &&
791c6551 1014 (prev_entry->end == new_entry->start)) {
984263bc
MD
1015 map->first_free = new_entry;
1016 }
1017
1018#if 0
1019 /*
1020 * Temporarily removed to avoid MAP_STACK panic, due to
1021 * MAP_STACK being a huge hack. Will be added back in
1022 * when MAP_STACK (and the user stack mapping) is fixed.
1023 */
1024 /*
1025 * It may be possible to simplify the entry
1026 */
a108bf71 1027 vm_map_simplify_entry(map, new_entry, countp);
984263bc
MD
1028#endif
1029
afeabdca
MD
1030 /*
1031 * Try to pre-populate the page table. Mappings governed by virtual
1032 * page tables cannot be prepopulated without a lot of work, so
1033 * don't try.
1034 */
1035 if ((cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) &&
1036 maptype != VM_MAPTYPE_VPAGETABLE) {
083a7402 1037 pmap_object_init_pt(map->pmap, start, prot,
984263bc
MD
1038 object, OFF_TO_IDX(offset), end - start,
1039 cow & MAP_PREFAULT_PARTIAL);
1040 }
1041
1042 return (KERN_SUCCESS);
1043}
1044
1045/*
1046 * Find sufficient space for `length' bytes in the given map, starting at
46754a20 1047 * `start'. Returns 0 on success, 1 on no space.
e9bb90e8
MD
1048 *
1049 * This function will returned an arbitrarily aligned pointer. If no
1050 * particular alignment is required you should pass align as 1. Note that
1051 * the map may return PAGE_SIZE aligned pointers if all the lengths used in
1052 * the map are a multiple of PAGE_SIZE, even if you pass a smaller align
1053 * argument.
1054 *
1055 * 'align' should be a power of 2 but is not required to be.
46754a20
MD
1056 *
1057 * The map must be exclusively locked.
1058 * No other requirements.
984263bc
MD
1059 */
1060int
c809941b 1061vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
9388fcaa 1062 vm_size_t align, int flags, vm_offset_t *addr)
984263bc
MD
1063{
1064 vm_map_entry_t entry, next;
1065 vm_offset_t end;
e9bb90e8 1066 vm_offset_t align_mask;
984263bc
MD
1067
1068 if (start < map->min_offset)
1069 start = map->min_offset;
1070 if (start > map->max_offset)
1071 return (1);
1072
e9bb90e8
MD
1073 /*
1074 * If the alignment is not a power of 2 we will have to use
1075 * a mod/division, set align_mask to a special value.
1076 */
1077 if ((align | (align - 1)) + 1 != (align << 1))
1078 align_mask = (vm_offset_t)-1;
1079 else
1080 align_mask = align - 1;
1081
984263bc
MD
1082 /*
1083 * Look for the first possible address; if there's already something
1084 * at this address, we have to start after it.
1085 */
1086 if (start == map->min_offset) {
1087 if ((entry = map->first_free) != &map->header)
1088 start = entry->end;
1089 } else {
1090 vm_map_entry_t tmp;
1091
1092 if (vm_map_lookup_entry(map, start, &tmp))
1093 start = tmp->end;
1094 entry = tmp;
1095 }
1096
1097 /*
1098 * Look through the rest of the map, trying to fit a new region in the
1099 * gap between existing regions, or after the very last region.
1100 */
1101 for (;; start = (entry = next)->end) {
e9bb90e8
MD
1102 /*
1103 * Adjust the proposed start by the requested alignment,
1104 * be sure that we didn't wrap the address.
1105 */
1106 if (align_mask == (vm_offset_t)-1)
1107 end = ((start + align - 1) / align) * align;
1108 else
1109 end = (start + align_mask) & ~align_mask;
1110 if (end < start)
1111 return (1);
1112 start = end;
984263bc
MD
1113 /*
1114 * Find the end of the proposed new region. Be sure we didn't
e9bb90e8
MD
1115 * go beyond the end of the map, or wrap around the address.
1116 * Then check to see if this is the last entry or if the
1117 * proposed end fits in the gap between this and the next
1118 * entry.
984263bc
MD
1119 */
1120 end = start + length;
1121 if (end > map->max_offset || end < start)
1122 return (1);
1123 next = entry->next;
c809941b
MD
1124
1125 /*
1126 * If the next entry's start address is beyond the desired
1127 * end address we may have found a good entry.
1128 *
1129 * If the next entry is a stack mapping we do not map into
1130 * the stack's reserved space.
1131 *
1132 * XXX continue to allow mapping into the stack's reserved
1133 * space if doing a MAP_STACK mapping inside a MAP_STACK
1134 * mapping, for backwards compatibility. But the caller
1135 * really should use MAP_STACK | MAP_TRYFIXED if they
1136 * want to do that.
1137 */
1138 if (next == &map->header)
984263bc 1139 break;
c809941b
MD
1140 if (next->start >= end) {
1141 if ((next->eflags & MAP_ENTRY_STACK) == 0)
1142 break;
1143 if (flags & MAP_STACK)
1144 break;
1145 if (next->start - next->aux.avail_ssize >= end)
1146 break;
1147 }
984263bc 1148 }
686dbf64 1149 map->hint = entry;
a8cf2878
MD
1150
1151 /*
1152 * Grow the kernel_map if necessary. pmap_growkernel() will panic
1153 * if it fails. The kernel_map is locked and nothing can steal
1154 * our address space if pmap_growkernel() blocks.
1155 *
1156 * NOTE: This may be unconditionally called for kldload areas on
1157 * x86_64 because these do not bump kernel_vm_end (which would
1158 * fill 128G worth of page tables!). Therefore we must not
1159 * retry.
1160 */
e4846942 1161 if (map == &kernel_map) {
a8cf2878
MD
1162 vm_offset_t kstop;
1163
1164 kstop = round_page(start + length);
1165 if (kstop > kernel_vm_end)
1166 pmap_growkernel(start, kstop);
984263bc 1167 }
a108bf71 1168 *addr = start;
984263bc
MD
1169 return (0);
1170}
1171
1172/*
46754a20
MD
1173 * vm_map_find finds an unallocated region in the target address map with
1174 * the given length. The search is defined to be first-fit from the
1175 * specified address; the region found is returned in the same parameter.
984263bc 1176 *
46754a20
MD
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 * No requirements. This function will lock the map temporarily.
984263bc
MD
1181 */
1182int
1183vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
9388fcaa 1184 vm_offset_t *addr, vm_size_t length, vm_size_t align,
c809941b 1185 boolean_t fitit,
1b874851
MD
1186 vm_maptype_t maptype,
1187 vm_prot_t prot, vm_prot_t max,
1188 int cow)
984263bc
MD
1189{
1190 vm_offset_t start;
03aa8d99 1191 int result;
a108bf71 1192 int count;
984263bc
MD
1193
1194 start = *addr;
1195
a108bf71 1196 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc 1197 vm_map_lock(map);
c809941b 1198 if (fitit) {
9388fcaa 1199 if (vm_map_findspace(map, start, length, align, 0, addr)) {
984263bc 1200 vm_map_unlock(map);
a108bf71 1201 vm_map_entry_release(count);
984263bc
MD
1202 return (KERN_NO_SPACE);
1203 }
1204 start = *addr;
1205 }
a108bf71 1206 result = vm_map_insert(map, &count, object, offset,
1b874851
MD
1207 start, start + length,
1208 maptype,
1209 prot, max,
1210 cow);
984263bc 1211 vm_map_unlock(map);
a108bf71 1212 vm_map_entry_release(count);
984263bc 1213
984263bc
MD
1214 return (result);
1215}
1216
1217/*
46754a20
MD
1218 * Simplify the given map entry by merging with either neighbor. This
1219 * routine also has the ability to merge with both neighbors.
984263bc 1220 *
46754a20
MD
1221 * This routine guarentees that the passed entry remains valid (though
1222 * possibly extended). When merging, this routine may delete one or
1223 * both neighbors. No action is taken on entries which have their
1224 * in-transition flag set.
984263bc 1225 *
46754a20 1226 * The map must be exclusively locked.
984263bc
MD
1227 */
1228void
a108bf71 1229vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc
MD
1230{
1231 vm_map_entry_t next, prev;
1232 vm_size_t prevsize, esize;
1233
1b874851 1234 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
12e4aaff 1235 ++mycpu->gd_cnt.v_intrans_coll;
984263bc
MD
1236 return;
1237 }
1238
1b874851
MD
1239 if (entry->maptype == VM_MAPTYPE_SUBMAP)
1240 return;
1241
984263bc
MD
1242 prev = entry->prev;
1243 if (prev != &map->header) {
1244 prevsize = prev->end - prev->start;
1245 if ( (prev->end == entry->start) &&
1b874851 1246 (prev->maptype == entry->maptype) &&
984263bc
MD
1247 (prev->object.vm_object == entry->object.vm_object) &&
1248 (!prev->object.vm_object ||
1249 (prev->offset + prevsize == entry->offset)) &&
1250 (prev->eflags == entry->eflags) &&
1251 (prev->protection == entry->protection) &&
1252 (prev->max_protection == entry->max_protection) &&
1253 (prev->inheritance == entry->inheritance) &&
1254 (prev->wired_count == entry->wired_count)) {
1255 if (map->first_free == prev)
1256 map->first_free = entry;
1257 if (map->hint == prev)
1258 map->hint = entry;
1259 vm_map_entry_unlink(map, prev);
1260 entry->start = prev->start;
1261 entry->offset = prev->offset;
1262 if (prev->object.vm_object)
1263 vm_object_deallocate(prev->object.vm_object);
a108bf71 1264 vm_map_entry_dispose(map, prev, countp);
984263bc
MD
1265 }
1266 }
1267
1268 next = entry->next;
1269 if (next != &map->header) {
1270 esize = entry->end - entry->start;
1271 if ((entry->end == next->start) &&
1b874851 1272 (next->maptype == entry->maptype) &&
984263bc
MD
1273 (next->object.vm_object == entry->object.vm_object) &&
1274 (!entry->object.vm_object ||
1275 (entry->offset + esize == next->offset)) &&
1276 (next->eflags == entry->eflags) &&
1277 (next->protection == entry->protection) &&
1278 (next->max_protection == entry->max_protection) &&
1279 (next->inheritance == entry->inheritance) &&
1280 (next->wired_count == entry->wired_count)) {
1281 if (map->first_free == next)
1282 map->first_free = entry;
1283 if (map->hint == next)
1284 map->hint = entry;
1285 vm_map_entry_unlink(map, next);
1286 entry->end = next->end;
1287 if (next->object.vm_object)
1288 vm_object_deallocate(next->object.vm_object);
a108bf71 1289 vm_map_entry_dispose(map, next, countp);
984263bc
MD
1290 }
1291 }
1292}
46754a20 1293
984263bc 1294/*
46754a20
MD
1295 * Asserts that the given entry begins at or after the specified address.
1296 * If necessary, it splits the entry into two.
984263bc 1297 */
46754a20
MD
1298#define vm_map_clip_start(map, entry, startaddr, countp) \
1299{ \
1300 if (startaddr > entry->start) \
1301 _vm_map_clip_start(map, entry, startaddr, countp); \
984263bc
MD
1302}
1303
1304/*
46754a20
MD
1305 * This routine is called only when it is known that the entry must be split.
1306 *
1307 * The map must be exclusively locked.
984263bc
MD
1308 */
1309static void
46754a20
MD
1310_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start,
1311 int *countp)
984263bc
MD
1312{
1313 vm_map_entry_t new_entry;
1314
1315 /*
1316 * Split off the front portion -- note that we must insert the new
1317 * entry BEFORE this one, so that this entry has the specified
1318 * starting address.
1319 */
1320
a108bf71 1321 vm_map_simplify_entry(map, entry, countp);
984263bc
MD
1322
1323 /*
1324 * If there is no object backing this entry, we might as well create
1325 * one now. If we defer it, an object can get created after the map
1326 * is clipped, and individual objects will be created for the split-up
1327 * map. This is a bit of a hack, but is also about the best place to
1328 * put this improvement.
1329 */
984263bc 1330 if (entry->object.vm_object == NULL && !map->system_map) {
53025830 1331 vm_map_entry_allocate_object(entry);
984263bc
MD
1332 }
1333
a108bf71 1334 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
1335 *new_entry = *entry;
1336
1337 new_entry->end = start;
1338 entry->offset += (start - entry->start);
1339 entry->start = start;
1340
1341 vm_map_entry_link(map, entry->prev, new_entry);
1342
1b874851
MD
1343 switch(entry->maptype) {
1344 case VM_MAPTYPE_NORMAL:
1345 case VM_MAPTYPE_VPAGETABLE:
984263bc 1346 vm_object_reference(new_entry->object.vm_object);
1b874851
MD
1347 break;
1348 default:
1349 break;
984263bc
MD
1350 }
1351}
1352
1353/*
46754a20
MD
1354 * Asserts that the given entry ends at or before the specified address.
1355 * If necessary, it splits the entry into two.
984263bc 1356 *
46754a20 1357 * The map must be exclusively locked.
984263bc 1358 */
46754a20
MD
1359#define vm_map_clip_end(map, entry, endaddr, countp) \
1360{ \
1361 if (endaddr < entry->end) \
1362 _vm_map_clip_end(map, entry, endaddr, countp); \
984263bc
MD
1363}
1364
1365/*
46754a20
MD
1366 * This routine is called only when it is known that the entry must be split.
1367 *
1368 * The map must be exclusively locked.
984263bc
MD
1369 */
1370static void
46754a20
MD
1371_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end,
1372 int *countp)
984263bc
MD
1373{
1374 vm_map_entry_t new_entry;
1375
1376 /*
1377 * If there is no object backing this entry, we might as well create
1378 * one now. If we defer it, an object can get created after the map
1379 * is clipped, and individual objects will be created for the split-up
1380 * map. This is a bit of a hack, but is also about the best place to
1381 * put this improvement.
1382 */
1383
1384 if (entry->object.vm_object == NULL && !map->system_map) {
53025830 1385 vm_map_entry_allocate_object(entry);
984263bc
MD
1386 }
1387
1388 /*
1389 * Create a new entry and insert it AFTER the specified entry
1390 */
1391
a108bf71 1392 new_entry = vm_map_entry_create(map, countp);
984263bc
MD
1393 *new_entry = *entry;
1394
1395 new_entry->start = entry->end = end;
1396 new_entry->offset += (end - entry->start);
1397
1398 vm_map_entry_link(map, entry, new_entry);
1399
1b874851
MD
1400 switch(entry->maptype) {
1401 case VM_MAPTYPE_NORMAL:
1402 case VM_MAPTYPE_VPAGETABLE:
984263bc 1403 vm_object_reference(new_entry->object.vm_object);
1b874851
MD
1404 break;
1405 default:
1406 break;
984263bc
MD
1407 }
1408}
1409
1410/*
46754a20
MD
1411 * Asserts that the starting and ending region addresses fall within the
1412 * valid range for the map.
984263bc 1413 */
46754a20
MD
1414#define VM_MAP_RANGE_CHECK(map, start, end) \
1415{ \
1416 if (start < vm_map_min(map)) \
1417 start = vm_map_min(map); \
1418 if (end > vm_map_max(map)) \
1419 end = vm_map_max(map); \
1420 if (start > end) \
1421 start = end; \
1422}
984263bc
MD
1423
1424/*
46754a20
MD
1425 * Used to block when an in-transition collison occurs. The map
1426 * is unlocked for the sleep and relocked before the return.
984263bc 1427 */
984263bc
MD
1428void
1429vm_map_transition_wait(vm_map_t map)
1430{
ff13bc52 1431 tsleep_interlock(map, 0);
984263bc 1432 vm_map_unlock(map);
ff13bc52 1433 tsleep(map, PINTERLOCKED, "vment", 0);
984263bc
MD
1434 vm_map_lock(map);
1435}
1436
1437/*
46754a20
MD
1438 * When we do blocking operations with the map lock held it is
1439 * possible that a clip might have occured on our in-transit entry,
1440 * requiring an adjustment to the entry in our loop. These macros
1441 * help the pageable and clip_range code deal with the case. The
1442 * conditional costs virtually nothing if no clipping has occured.
984263bc
MD
1443 */
1444
1445#define CLIP_CHECK_BACK(entry, save_start) \
1446 do { \
1447 while (entry->start != save_start) { \
1448 entry = entry->prev; \
1449 KASSERT(entry != &map->header, ("bad entry clip")); \
1450 } \
1451 } while(0)
1452
1453#define CLIP_CHECK_FWD(entry, save_end) \
1454 do { \
1455 while (entry->end != save_end) { \
1456 entry = entry->next; \
1457 KASSERT(entry != &map->header, ("bad entry clip")); \
1458 } \
1459 } while(0)
1460
1461
1462/*
46754a20
MD
1463 * Clip the specified range and return the base entry. The
1464 * range may cover several entries starting at the returned base
1465 * and the first and last entry in the covering sequence will be
1466 * properly clipped to the requested start and end address.
1467 *
1468 * If no holes are allowed you should pass the MAP_CLIP_NO_HOLES
1469 * flag.
1470 *
1471 * The MAP_ENTRY_IN_TRANSITION flag will be set for the entries
1472 * covered by the requested range.
1473 *
1474 * The map must be exclusively locked on entry and will remain locked
1475 * on return. If no range exists or the range contains holes and you
1476 * specified that no holes were allowed, NULL will be returned. This
1477 * routine may temporarily unlock the map in order avoid a deadlock when
1478 * sleeping.
984263bc
MD
1479 */
1480static
1481vm_map_entry_t
a108bf71 1482vm_map_clip_range(vm_map_t map, vm_offset_t start, vm_offset_t end,
46754a20 1483 int *countp, int flags)
984263bc
MD
1484{
1485 vm_map_entry_t start_entry;
1486 vm_map_entry_t entry;
1487
1488 /*
1489 * Locate the entry and effect initial clipping. The in-transition
1490 * case does not occur very often so do not try to optimize it.
1491 */
1492again:
1493 if (vm_map_lookup_entry(map, start, &start_entry) == FALSE)
1494 return (NULL);
1495 entry = start_entry;
1496 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1497 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
12e4aaff
MD
1498 ++mycpu->gd_cnt.v_intrans_coll;
1499 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
1500 vm_map_transition_wait(map);
1501 /*
1502 * entry and/or start_entry may have been clipped while
1503 * we slept, or may have gone away entirely. We have
1504 * to restart from the lookup.
1505 */
1506 goto again;
1507 }
46754a20 1508
984263bc
MD
1509 /*
1510 * Since we hold an exclusive map lock we do not have to restart
1511 * after clipping, even though clipping may block in zalloc.
1512 */
a108bf71
MD
1513 vm_map_clip_start(map, entry, start, countp);
1514 vm_map_clip_end(map, entry, end, countp);
984263bc
MD
1515 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1516
1517 /*
1518 * Scan entries covered by the range. When working on the next
1519 * entry a restart need only re-loop on the current entry which
1520 * we have already locked, since 'next' may have changed. Also,
1521 * even though entry is safe, it may have been clipped so we
1522 * have to iterate forwards through the clip after sleeping.
1523 */
1524 while (entry->next != &map->header && entry->next->start < end) {
1525 vm_map_entry_t next = entry->next;
1526
1527 if (flags & MAP_CLIP_NO_HOLES) {
1528 if (next->start > entry->end) {
1529 vm_map_unclip_range(map, start_entry,
a108bf71 1530 start, entry->end, countp, flags);
984263bc
MD
1531 return(NULL);
1532 }
1533 }
1534
1535 if (next->eflags & MAP_ENTRY_IN_TRANSITION) {
1536 vm_offset_t save_end = entry->end;
1537 next->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
12e4aaff
MD
1538 ++mycpu->gd_cnt.v_intrans_coll;
1539 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
1540 vm_map_transition_wait(map);
1541
1542 /*
1543 * clips might have occured while we blocked.
1544 */
1545 CLIP_CHECK_FWD(entry, save_end);
1546 CLIP_CHECK_BACK(start_entry, start);
1547 continue;
1548 }
1549 /*
1550 * No restart necessary even though clip_end may block, we
1551 * are holding the map lock.
1552 */
a108bf71 1553 vm_map_clip_end(map, next, end, countp);
984263bc
MD
1554 next->eflags |= MAP_ENTRY_IN_TRANSITION;
1555 entry = next;
1556 }
1557 if (flags & MAP_CLIP_NO_HOLES) {
1558 if (entry->end != end) {
1559 vm_map_unclip_range(map, start_entry,
a108bf71 1560 start, entry->end, countp, flags);
984263bc
MD
1561 return(NULL);
1562 }
1563 }
1564 return(start_entry);
1565}
1566
1567/*
46754a20
MD
1568 * Undo the effect of vm_map_clip_range(). You should pass the same
1569 * flags and the same range that you passed to vm_map_clip_range().
1570 * This code will clear the in-transition flag on the entries and
1571 * wake up anyone waiting. This code will also simplify the sequence
1572 * and attempt to merge it with entries before and after the sequence.
1573 *
1574 * The map must be locked on entry and will remain locked on return.
1575 *
1576 * Note that you should also pass the start_entry returned by
1577 * vm_map_clip_range(). However, if you block between the two calls
1578 * with the map unlocked please be aware that the start_entry may
1579 * have been clipped and you may need to scan it backwards to find
1580 * the entry corresponding with the original start address. You are
1581 * responsible for this, vm_map_unclip_range() expects the correct
1582 * start_entry to be passed to it and will KASSERT otherwise.
984263bc
MD
1583 */
1584static
1585void
46754a20
MD
1586vm_map_unclip_range(vm_map_t map, vm_map_entry_t start_entry,
1587 vm_offset_t start, vm_offset_t end,
1588 int *countp, int flags)
984263bc
MD
1589{
1590 vm_map_entry_t entry;
1591
1592 entry = start_entry;
1593
1594 KASSERT(entry->start == start, ("unclip_range: illegal base entry"));
1595 while (entry != &map->header && entry->start < end) {
46754a20
MD
1596 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
1597 ("in-transition flag not set during unclip on: %p",
1598 entry));
1599 KASSERT(entry->end <= end,
1600 ("unclip_range: tail wasn't clipped"));
984263bc
MD
1601 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1602 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1603 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1604 wakeup(map);
1605 }
1606 entry = entry->next;
1607 }
1608
1609 /*
1610 * Simplification does not block so there is no restart case.
1611 */
1612 entry = start_entry;
1613 while (entry != &map->header && entry->start < end) {
a108bf71 1614 vm_map_simplify_entry(map, entry, countp);
984263bc
MD
1615 entry = entry->next;
1616 }
1617}
1618
1619/*
46754a20 1620 * Mark the given range as handled by a subordinate map.
984263bc 1621 *
46754a20
MD
1622 * This range must have been created with vm_map_find(), and no other
1623 * operations may have been performed on this range prior to calling
1624 * vm_map_submap().
984263bc 1625 *
46754a20 1626 * Submappings cannot be removed.
984263bc 1627 *
46754a20 1628 * No requirements.
984263bc
MD
1629 */
1630int
a108bf71 1631vm_map_submap(vm_map_t map, vm_offset_t start, vm_offset_t end, vm_map_t submap)
984263bc
MD
1632{
1633 vm_map_entry_t entry;
1634 int result = KERN_INVALID_ARGUMENT;
a108bf71 1635 int count;
984263bc 1636
a108bf71 1637 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1638 vm_map_lock(map);
1639
1640 VM_MAP_RANGE_CHECK(map, start, end);
1641
1642 if (vm_map_lookup_entry(map, start, &entry)) {
a108bf71 1643 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1644 } else {
1645 entry = entry->next;
1646 }
1647
a108bf71 1648 vm_map_clip_end(map, entry, end, &count);
984263bc
MD
1649
1650 if ((entry->start == start) && (entry->end == end) &&
1651 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1652 (entry->object.vm_object == NULL)) {
1653 entry->object.sub_map = submap;
1b874851 1654 entry->maptype = VM_MAPTYPE_SUBMAP;
984263bc
MD
1655 result = KERN_SUCCESS;
1656 }
1657 vm_map_unlock(map);
a108bf71 1658 vm_map_entry_release(count);
984263bc
MD
1659
1660 return (result);
1661}
1662
1663/*
1b874851
MD
1664 * Sets the protection of the specified address region in the target map.
1665 * If "set_max" is specified, the maximum protection is to be set;
1666 * otherwise, only the current protection is affected.
1667 *
1668 * The protection is not applicable to submaps, but is applicable to normal
1669 * maps and maps governed by virtual page tables. For example, when operating
1670 * on a virtual page table our protection basically controls how COW occurs
1671 * on the backing object, whereas the virtual page table abstraction itself
1672 * is an abstraction for userland.
46754a20
MD
1673 *
1674 * No requirements.
984263bc
MD
1675 */
1676int
1677vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1678 vm_prot_t new_prot, boolean_t set_max)
1679{
1680 vm_map_entry_t current;
1681 vm_map_entry_t entry;
a108bf71 1682 int count;
984263bc 1683
a108bf71 1684 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1685 vm_map_lock(map);
1686
1687 VM_MAP_RANGE_CHECK(map, start, end);
1688
1689 if (vm_map_lookup_entry(map, start, &entry)) {
a108bf71 1690 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1691 } else {
1692 entry = entry->next;
1693 }
1694
1695 /*
1696 * Make a first pass to check for protection violations.
1697 */
984263bc
MD
1698 current = entry;
1699 while ((current != &map->header) && (current->start < end)) {
1b874851 1700 if (current->maptype == VM_MAPTYPE_SUBMAP) {
984263bc 1701 vm_map_unlock(map);
a108bf71 1702 vm_map_entry_release(count);
984263bc
MD
1703 return (KERN_INVALID_ARGUMENT);
1704 }
1705 if ((new_prot & current->max_protection) != new_prot) {
1706 vm_map_unlock(map);
a108bf71 1707 vm_map_entry_release(count);
984263bc
MD
1708 return (KERN_PROTECTION_FAILURE);
1709 }
1710 current = current->next;
1711 }
1712
1713 /*
1714 * Go back and fix up protections. [Note that clipping is not
1715 * necessary the second time.]
1716 */
984263bc
MD
1717 current = entry;
1718
1719 while ((current != &map->header) && (current->start < end)) {
1720 vm_prot_t old_prot;
1721
a108bf71 1722 vm_map_clip_end(map, current, end, &count);
984263bc
MD
1723
1724 old_prot = current->protection;
1b874851 1725 if (set_max) {
984263bc
MD
1726 current->protection =
1727 (current->max_protection = new_prot) &
1728 old_prot;
1b874851 1729 } else {
984263bc 1730 current->protection = new_prot;
1b874851 1731 }
984263bc
MD
1732
1733 /*
1734 * Update physical map if necessary. Worry about copy-on-write
1735 * here -- CHECK THIS XXX
1736 */
1737
1738 if (current->protection != old_prot) {
1739#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1740 VM_PROT_ALL)
1741
1742 pmap_protect(map->pmap, current->start,
1743 current->end,
1744 current->protection & MASK(current));
1745#undef MASK
1746 }
1747
a108bf71 1748 vm_map_simplify_entry(map, current, &count);
984263bc
MD
1749
1750 current = current->next;
1751 }
1752
1753 vm_map_unlock(map);
a108bf71 1754 vm_map_entry_release(count);
984263bc
MD
1755 return (KERN_SUCCESS);
1756}
1757
1758/*
46754a20
MD
1759 * This routine traverses a processes map handling the madvise
1760 * system call. Advisories are classified as either those effecting
1761 * the vm_map_entry structure, or those effecting the underlying
1762 * objects.
984263bc 1763 *
46754a20 1764 * The <value> argument is used for extended madvise calls.
afeabdca 1765 *
46754a20 1766 * No requirements.
984263bc 1767 */
984263bc 1768int
afeabdca
MD
1769vm_map_madvise(vm_map_t map, vm_offset_t start, vm_offset_t end,
1770 int behav, off_t value)
984263bc
MD
1771{
1772 vm_map_entry_t current, entry;
1773 int modify_map = 0;
afeabdca 1774 int error = 0;
a108bf71 1775 int count;
984263bc
MD
1776
1777 /*
1778 * Some madvise calls directly modify the vm_map_entry, in which case
1779 * we need to use an exclusive lock on the map and we need to perform
1780 * various clipping operations. Otherwise we only need a read-lock
1781 * on the map.
1782 */
1783
a108bf71
MD
1784 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1785
984263bc
MD
1786 switch(behav) {
1787 case MADV_NORMAL:
1788 case MADV_SEQUENTIAL:
1789 case MADV_RANDOM:
1790 case MADV_NOSYNC:
1791 case MADV_AUTOSYNC:
1792 case MADV_NOCORE:
1793 case MADV_CORE:
afeabdca
MD
1794 case MADV_SETMAP:
1795 case MADV_INVAL:
984263bc
MD
1796 modify_map = 1;
1797 vm_map_lock(map);
1798 break;
1799 case MADV_WILLNEED:
1800 case MADV_DONTNEED:
1801 case MADV_FREE:
1802 vm_map_lock_read(map);
1803 break;
1804 default:
a108bf71 1805 vm_map_entry_release(count);
afeabdca 1806 return (EINVAL);
984263bc
MD
1807 }
1808
1809 /*
1810 * Locate starting entry and clip if necessary.
1811 */
1812
1813 VM_MAP_RANGE_CHECK(map, start, end);
1814
1815 if (vm_map_lookup_entry(map, start, &entry)) {
1816 if (modify_map)
a108bf71 1817 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1818 } else {
1819 entry = entry->next;
1820 }
1821
1822 if (modify_map) {
1823 /*
1824 * madvise behaviors that are implemented in the vm_map_entry.
1825 *
1826 * We clip the vm_map_entry so that behavioral changes are
1827 * limited to the specified address range.
1828 */
1829 for (current = entry;
1830 (current != &map->header) && (current->start < end);
1831 current = current->next
1832 ) {
1b874851 1833 if (current->maptype == VM_MAPTYPE_SUBMAP)
984263bc
MD
1834 continue;
1835
a108bf71 1836 vm_map_clip_end(map, current, end, &count);
984263bc
MD
1837
1838 switch (behav) {
1839 case MADV_NORMAL:
1840 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1841 break;
1842 case MADV_SEQUENTIAL:
1843 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1844 break;
1845 case MADV_RANDOM:
1846 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1847 break;
1848 case MADV_NOSYNC:
1849 current->eflags |= MAP_ENTRY_NOSYNC;
1850 break;
1851 case MADV_AUTOSYNC:
1852 current->eflags &= ~MAP_ENTRY_NOSYNC;
1853 break;
1854 case MADV_NOCORE:
1855 current->eflags |= MAP_ENTRY_NOCOREDUMP;
1856 break;
1857 case MADV_CORE:
1858 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1859 break;
afeabdca
MD
1860 case MADV_INVAL:
1861 /*
1862 * Invalidate the related pmap entries, used
1863 * to flush portions of the real kernel's
1864 * pmap when the caller has removed or
1865 * modified existing mappings in a virtual
1866 * page table.
1867 */
1868 pmap_remove(map->pmap,
1869 current->start, current->end);
1870 break;
1871 case MADV_SETMAP:
1872 /*
1873 * Set the page directory page for a map
1874 * governed by a virtual page table. Mark
1875 * the entry as being governed by a virtual
1876 * page table if it is not.
1877 *
1878 * XXX the page directory page is stored
1879 * in the avail_ssize field if the map_entry.
1880 *
1881 * XXX the map simplification code does not
1882 * compare this field so weird things may
1883 * happen if you do not apply this function
1884 * to the entire mapping governed by the
1885 * virtual page table.
1886 */
1887 if (current->maptype != VM_MAPTYPE_VPAGETABLE) {
1888 error = EINVAL;
1889 break;
1890 }
1891 current->aux.master_pde = value;
1892 pmap_remove(map->pmap,
1893 current->start, current->end);
1894 break;
984263bc 1895 default:
afeabdca 1896 error = EINVAL;
984263bc
MD
1897 break;
1898 }
a108bf71 1899 vm_map_simplify_entry(map, current, &count);
984263bc
MD
1900 }
1901 vm_map_unlock(map);
1902 } else {
1903 vm_pindex_t pindex;
1904 int count;
1905
1906 /*
1907 * madvise behaviors that are implemented in the underlying
1908 * vm_object.
1909 *
1910 * Since we don't clip the vm_map_entry, we have to clip
1911 * the vm_object pindex and count.
1b874851
MD
1912 *
1913 * NOTE! We currently do not support these functions on
1914 * virtual page tables.
984263bc
MD
1915 */
1916 for (current = entry;
1917 (current != &map->header) && (current->start < end);
1918 current = current->next
1919 ) {
1920 vm_offset_t useStart;
1921
1b874851 1922 if (current->maptype != VM_MAPTYPE_NORMAL)
984263bc
MD
1923 continue;
1924
1925 pindex = OFF_TO_IDX(current->offset);
1926 count = atop(current->end - current->start);
1927 useStart = current->start;
1928
1929 if (current->start < start) {
1930 pindex += atop(start - current->start);
1931 count -= atop(start - current->start);
1932 useStart = start;
1933 }
1934 if (current->end > end)
1935 count -= atop(current->end - end);
1936
1937 if (count <= 0)
1938 continue;
1939
1940 vm_object_madvise(current->object.vm_object,
1941 pindex, count, behav);
afeabdca
MD
1942
1943 /*
1944 * Try to populate the page table. Mappings governed
1945 * by virtual page tables cannot be pre-populated
1946 * without a lot of work so don't try.
1947 */
1948 if (behav == MADV_WILLNEED &&
1949 current->maptype != VM_MAPTYPE_VPAGETABLE) {
984263bc
MD
1950 pmap_object_init_pt(
1951 map->pmap,
1952 useStart,
083a7402 1953 current->protection,
984263bc
MD
1954 current->object.vm_object,
1955 pindex,
1956 (count << PAGE_SHIFT),
1957 MAP_PREFAULT_MADVISE
1958 );
1959 }
1960 }
1961 vm_map_unlock_read(map);
1962 }
a108bf71 1963 vm_map_entry_release(count);
afeabdca 1964 return(error);
984263bc
MD
1965}
1966
1967
1968/*
46754a20
MD
1969 * Sets the inheritance of the specified address range in the target map.
1970 * Inheritance affects how the map will be shared with child maps at the
1971 * time of vm_map_fork.
984263bc
MD
1972 */
1973int
1974vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1975 vm_inherit_t new_inheritance)
1976{
1977 vm_map_entry_t entry;
1978 vm_map_entry_t temp_entry;
a108bf71 1979 int count;
984263bc
MD
1980
1981 switch (new_inheritance) {
1982 case VM_INHERIT_NONE:
1983 case VM_INHERIT_COPY:
1984 case VM_INHERIT_SHARE:
1985 break;
1986 default:
1987 return (KERN_INVALID_ARGUMENT);
1988 }
1989
a108bf71 1990 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
1991 vm_map_lock(map);
1992
1993 VM_MAP_RANGE_CHECK(map, start, end);
1994
1995 if (vm_map_lookup_entry(map, start, &temp_entry)) {
1996 entry = temp_entry;
a108bf71 1997 vm_map_clip_start(map, entry, start, &count);
984263bc
MD
1998 } else
1999 entry = temp_entry->next;
2000
2001 while ((entry != &map->header) && (entry->start < end)) {
a108bf71 2002 vm_map_clip_end(map, entry, end, &count);
984263bc
MD
2003
2004 entry->inheritance = new_inheritance;
2005
a108bf71 2006 vm_map_simplify_entry(map, entry, &count);
984263bc
MD
2007
2008 entry = entry->next;
2009 }
984263bc 2010 vm_map_unlock(map);
a108bf71 2011 vm_map_entry_release(count);
984263bc
MD
2012 return (KERN_SUCCESS);
2013}
2014
2015/*
2016 * Implement the semantics of mlock
2017 */
2018int
57e43348 2019vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t real_end,
46754a20 2020 boolean_t new_pageable)
984263bc
MD
2021{
2022 vm_map_entry_t entry;
2023 vm_map_entry_t start_entry;
2024 vm_offset_t end;
2025 int rv = KERN_SUCCESS;
a108bf71 2026 int count;
984263bc 2027
a108bf71 2028 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2029 vm_map_lock(map);
2030 VM_MAP_RANGE_CHECK(map, start, real_end);
2031 end = real_end;
2032
46754a20
MD
2033 start_entry = vm_map_clip_range(map, start, end, &count,
2034 MAP_CLIP_NO_HOLES);
984263bc
MD
2035 if (start_entry == NULL) {
2036 vm_map_unlock(map);
a108bf71 2037 vm_map_entry_release(count);
984263bc
MD
2038 return (KERN_INVALID_ADDRESS);
2039 }
2040
2041 if (new_pageable == 0) {
2042 entry = start_entry;
2043 while ((entry != &map->header) && (entry->start < end)) {
2044 vm_offset_t save_start;
2045 vm_offset_t save_end;
2046
2047 /*
2048 * Already user wired or hard wired (trivial cases)
2049 */
2050 if (entry->eflags & MAP_ENTRY_USER_WIRED) {
2051 entry = entry->next;
2052 continue;
2053 }
2054 if (entry->wired_count != 0) {
2055 entry->wired_count++;
2056 entry->eflags |= MAP_ENTRY_USER_WIRED;
2057 entry = entry->next;
2058 continue;
2059 }
2060
2061 /*
2062 * A new wiring requires instantiation of appropriate
2063 * management structures and the faulting in of the
2064 * page.
2065 */
1b874851 2066 if (entry->maptype != VM_MAPTYPE_SUBMAP) {
46754a20
MD
2067 int copyflag = entry->eflags &
2068 MAP_ENTRY_NEEDS_COPY;
2069 if (copyflag && ((entry->protection &
2070 VM_PROT_WRITE) != 0)) {
53025830 2071 vm_map_entry_shadow(entry);
984263bc
MD
2072 } else if (entry->object.vm_object == NULL &&
2073 !map->system_map) {
53025830 2074 vm_map_entry_allocate_object(entry);
984263bc
MD
2075 }
2076 }
2077 entry->wired_count++;
2078 entry->eflags |= MAP_ENTRY_USER_WIRED;
2079
2080 /*
f2d22ebf
MD
2081 * Now fault in the area. Note that vm_fault_wire()
2082 * may release the map lock temporarily, it will be
2083 * relocked on return. The in-transition
984263bc
MD
2084 * flag protects the entries.
2085 */
2086 save_start = entry->start;
2087 save_end = entry->end;
f2d22ebf 2088 rv = vm_fault_wire(map, entry, TRUE);
984263bc
MD
2089 if (rv) {
2090 CLIP_CHECK_BACK(entry, save_start);
2091 for (;;) {
2092 KASSERT(entry->wired_count == 1, ("bad wired_count on entry"));
2093 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2094 entry->wired_count = 0;
2095 if (entry->end == save_end)
2096 break;
2097 entry = entry->next;
2098 KASSERT(entry != &map->header, ("bad entry clip during backout"));
2099 }
2100 end = save_start; /* unwire the rest */
2101 break;
2102 }
2103 /*
2104 * note that even though the entry might have been
2105 * clipped, the USER_WIRED flag we set prevents
2106 * duplication so we do not have to do a
2107 * clip check.
2108 */
2109 entry = entry->next;
2110 }
2111
2112 /*
2113 * If we failed fall through to the unwiring section to
2114 * unwire what we had wired so far. 'end' has already
2115 * been adjusted.
2116 */
2117 if (rv)
2118 new_pageable = 1;
2119
2120 /*
2121 * start_entry might have been clipped if we unlocked the
2122 * map and blocked. No matter how clipped it has gotten
2123 * there should be a fragment that is on our start boundary.
2124 */
2125 CLIP_CHECK_BACK(start_entry, start);
2126 }
2127
2128 /*
2129 * Deal with the unwiring case.
2130 */
2131 if (new_pageable) {
2132 /*
2133 * This is the unwiring case. We must first ensure that the
2134 * range to be unwired is really wired down. We know there
2135 * are no holes.
2136 */
2137 entry = start_entry;
2138 while ((entry != &map->header) && (entry->start < end)) {
2139 if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2140 rv = KERN_INVALID_ARGUMENT;
2141 goto done;
2142 }
2143 KASSERT(entry->wired_count != 0, ("wired count was 0 with USER_WIRED set! %p", entry));
2144 entry = entry->next;
2145 }
2146
2147 /*
2148 * Now decrement the wiring count for each region. If a region
2149 * becomes completely unwired, unwire its physical pages and
2150 * mappings.
2151 */
b4eddbac
DR
2152 /*
2153 * The map entries are processed in a loop, checking to
2154 * make sure the entry is wired and asserting it has a wired
2155 * count. However, another loop was inserted more-or-less in
2156 * the middle of the unwiring path. This loop picks up the
2157 * "entry" loop variable from the first loop without first
2158 * setting it to start_entry. Naturally, the secound loop
2159 * is never entered and the pages backing the entries are
2160 * never unwired. This can lead to a leak of wired pages.
2161 */
2162 entry = start_entry;
984263bc 2163 while ((entry != &map->header) && (entry->start < end)) {
f2d22ebf
MD
2164 KASSERT(entry->eflags & MAP_ENTRY_USER_WIRED,
2165 ("expected USER_WIRED on entry %p", entry));
984263bc
MD
2166 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2167 entry->wired_count--;
2168 if (entry->wired_count == 0)
f2d22ebf 2169 vm_fault_unwire(map, entry);
984263bc
MD
2170 entry = entry->next;
2171 }
2172 }
2173done:
a108bf71 2174 vm_map_unclip_range(map, start_entry, start, real_end, &count,
984263bc
MD
2175 MAP_CLIP_NO_HOLES);
2176 map->timestamp++;
2177 vm_map_unlock(map);
a108bf71 2178 vm_map_entry_release(count);
984263bc
MD
2179 return (rv);
2180}
2181
2182/*
46754a20
MD
2183 * Sets the pageability of the specified address range in the target map.
2184 * Regions specified as not pageable require locked-down physical
2185 * memory and physical page maps.
984263bc 2186 *
46754a20
MD
2187 * The map must not be locked, but a reference must remain to the map
2188 * throughout the call.
984263bc 2189 *
46754a20
MD
2190 * This function may be called via the zalloc path and must properly
2191 * reserve map entries for kernel_map.
a108bf71 2192 *
46754a20 2193 * No requirements.
984263bc
MD
2194 */
2195int
e1359933 2196vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t real_end, int kmflags)
984263bc
MD
2197{
2198 vm_map_entry_t entry;
2199 vm_map_entry_t start_entry;
2200 vm_offset_t end;
2201 int rv = KERN_SUCCESS;
a108bf71 2202 int count;
984263bc 2203
e1359933 2204 if (kmflags & KM_KRESERVE)
a108bf71 2205 count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
a108bf71
MD
2206 else
2207 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2208 vm_map_lock(map);
2209 VM_MAP_RANGE_CHECK(map, start, real_end);
2210 end = real_end;
2211
46754a20
MD
2212 start_entry = vm_map_clip_range(map, start, end, &count,
2213 MAP_CLIP_NO_HOLES);
984263bc
MD
2214 if (start_entry == NULL) {
2215 vm_map_unlock(map);
a108bf71
MD
2216 rv = KERN_INVALID_ADDRESS;
2217 goto failure;
984263bc 2218 }
e1359933 2219 if ((kmflags & KM_PAGEABLE) == 0) {
984263bc
MD
2220 /*
2221 * Wiring.
2222 *
2223 * 1. Holding the write lock, we create any shadow or zero-fill
2224 * objects that need to be created. Then we clip each map
2225 * entry to the region to be wired and increment its wiring
2226 * count. We create objects before clipping the map entries
2227 * to avoid object proliferation.
2228 *
2229 * 2. We downgrade to a read lock, and call vm_fault_wire to
2230 * fault in the pages for any newly wired area (wired_count is
2231 * 1).
2232 *
2233 * Downgrading to a read lock for vm_fault_wire avoids a
2234 * possible deadlock with another process that may have faulted
2235 * on one of the pages to be wired (it would mark the page busy,
2236 * blocking us, then in turn block on the map lock that we
2237 * hold). Because of problems in the recursive lock package,
2238 * we cannot upgrade to a write lock in vm_map_lookup. Thus,
2239 * any actions that require the write lock must be done
2240 * beforehand. Because we keep the read lock on the map, the
2241 * copy-on-write status of the entries we modify here cannot
2242 * change.
2243 */
984263bc
MD
2244 entry = start_entry;
2245 while ((entry != &map->header) && (entry->start < end)) {
2246 /*
2247 * Trivial case if the entry is already wired
2248 */
2249 if (entry->wired_count) {
2250 entry->wired_count++;
2251 entry = entry->next;
2252 continue;
2253 }
2254
2255 /*
2256 * The entry is being newly wired, we have to setup
2257 * appropriate management structures. A shadow
2258 * object is required for a copy-on-write region,
2259 * or a normal object for a zero-fill region. We
2260 * do not have to do this for entries that point to sub
2261 * maps because we won't hold the lock on the sub map.
2262 */
1b874851 2263 if (entry->maptype != VM_MAPTYPE_SUBMAP) {
46754a20
MD
2264 int copyflag = entry->eflags &
2265 MAP_ENTRY_NEEDS_COPY;
2266 if (copyflag && ((entry->protection &
2267 VM_PROT_WRITE) != 0)) {
53025830 2268 vm_map_entry_shadow(entry);
984263bc
MD
2269 } else if (entry->object.vm_object == NULL &&
2270 !map->system_map) {
53025830 2271 vm_map_entry_allocate_object(entry);
984263bc
MD
2272 }
2273 }
2274
2275 entry->wired_count++;
2276 entry = entry->next;
2277 }
2278
2279 /*
2280 * Pass 2.
2281 */
2282
2283 /*
2284 * HACK HACK HACK HACK
2285 *
46754a20
MD
2286 * vm_fault_wire() temporarily unlocks the map to avoid
2287 * deadlocks. The in-transition flag from vm_map_clip_range
2288 * call should protect us from changes while the map is
2289 * unlocked. T
2290 *
2291 * NOTE: Previously this comment stated that clipping might
2292 * still occur while the entry is unlocked, but from
2293 * what I can tell it actually cannot.
2294 *
2295 * It is unclear whether the CLIP_CHECK_*() calls
2296 * are still needed but we keep them in anyway.
984263bc
MD
2297 *
2298 * HACK HACK HACK HACK
2299 */
2300
984263bc
MD
2301 entry = start_entry;
2302 while (entry != &map->header && entry->start < end) {
2303 /*
2304 * If vm_fault_wire fails for any page we need to undo
2305 * what has been done. We decrement the wiring count
2306 * for those pages which have not yet been wired (now)
2307 * and unwire those that have (later).
2308 */
2309 vm_offset_t save_start = entry->start;
2310 vm_offset_t save_end = entry->end;
2311
2312 if (entry->wired_count == 1)
f2d22ebf 2313 rv = vm_fault_wire(map, entry, FALSE);
984263bc
MD
2314 if (rv) {
2315 CLIP_CHECK_BACK(entry, save_start);
2316 for (;;) {
2317 KASSERT(entry->wired_count == 1, ("wired_count changed unexpectedly"));
2318 entry->wired_count = 0;
2319 if (entry->end == save_end)
2320 break;
2321 entry = entry->next;
2322 KASSERT(entry != &map->header, ("bad entry clip during backout"));
2323 }
2324 end = save_start;
2325 break;
2326 }
2327 CLIP_CHECK_FWD(entry, save_end);
2328 entry = entry->next;
2329 }
984263bc 2330
984263bc
MD
2331 /*
2332 * If a failure occured undo everything by falling through
2333 * to the unwiring code. 'end' has already been adjusted
2334 * appropriately.
2335 */
2336 if (rv)
e1359933 2337 kmflags |= KM_PAGEABLE;
984263bc
MD
2338
2339 /*
f2d22ebf
MD
2340 * start_entry is still IN_TRANSITION but may have been
2341 * clipped since vm_fault_wire() unlocks and relocks the
2342 * map. No matter how clipped it has gotten there should
2343 * be a fragment that is on our start boundary.
984263bc
MD
2344 */
2345 CLIP_CHECK_BACK(start_entry, start);
2346 }
2347
e1359933 2348 if (kmflags & KM_PAGEABLE) {
984263bc
MD
2349 /*
2350 * This is the unwiring case. We must first ensure that the
2351 * range to be unwired is really wired down. We know there
2352 * are no holes.
2353 */
2354 entry = start_entry;
2355 while ((entry != &map->header) && (entry->start < end)) {
2356 if (entry->wired_count == 0) {
2357 rv = KERN_INVALID_ARGUMENT;
2358 goto done;
2359 }
2360 entry = entry->next;
2361 }
2362
2363 /*
2364 * Now decrement the wiring count for each region. If a region
2365 * becomes completely unwired, unwire its physical pages and
2366 * mappings.
2367 */
2368 entry = start_entry;
2369 while ((entry != &map->header) && (entry->start < end)) {
2370 entry->wired_count--;
2371 if (entry->wired_count == 0)
f2d22ebf 2372 vm_fault_unwire(map, entry);
984263bc
MD
2373 entry = entry->next;
2374 }
2375 }
2376done:
46754a20
MD
2377 vm_map_unclip_range(map, start_entry, start, real_end,
2378 &count, MAP_CLIP_NO_HOLES);
984263bc
MD
2379 map->timestamp++;
2380 vm_map_unlock(map);
a108bf71 2381failure:
e1359933 2382 if (kmflags & KM_KRESERVE)
a108bf71 2383 vm_map_entry_krelease(count);
a108bf71
MD
2384 else
2385 vm_map_entry_release(count);
984263bc
MD
2386 return (rv);
2387}
2388
a108bf71 2389/*
46754a20
MD
2390 * Mark a newly allocated address range as wired but do not fault in
2391 * the pages. The caller is expected to load the pages into the object.
a108bf71 2392 *
46754a20
MD
2393 * The map must be locked on entry and will remain locked on return.
2394 * No other requirements.
a108bf71
MD
2395 */
2396void
46754a20
MD
2397vm_map_set_wired_quick(vm_map_t map, vm_offset_t addr, vm_size_t size,
2398 int *countp)
a108bf71
MD
2399{
2400 vm_map_entry_t scan;
2401 vm_map_entry_t entry;
2402
46754a20
MD
2403 entry = vm_map_clip_range(map, addr, addr + size,
2404 countp, MAP_CLIP_NO_HOLES);
2405 for (scan = entry;
2406 scan != &map->header && scan->start < addr + size;
2407 scan = scan->next) {
a108bf71
MD
2408 KKASSERT(entry->wired_count == 0);
2409 entry->wired_count = 1;
2410 }
46754a20
MD
2411 vm_map_unclip_range(map, entry, addr, addr + size,
2412 countp, MAP_CLIP_NO_HOLES);
a108bf71
MD
2413}
2414
984263bc 2415/*
984263bc
MD
2416 * Push any dirty cached pages in the address range to their pager.
2417 * If syncio is TRUE, dirty pages are written synchronously.
2418 * If invalidate is TRUE, any cached pages are freed as well.
2419 *
2bc7505b
MD
2420 * This routine is called by sys_msync()
2421 *
984263bc 2422 * Returns an error if any part of the specified range is not mapped.
46754a20
MD
2423 *
2424 * No requirements.
984263bc
MD
2425 */
2426int
2bc7505b
MD
2427vm_map_clean(vm_map_t map, vm_offset_t start, vm_offset_t end,
2428 boolean_t syncio, boolean_t invalidate)
984263bc
MD
2429{
2430 vm_map_entry_t current;
2431 vm_map_entry_t entry;
2432 vm_size_t size;
2433 vm_object_t object;
2434 vm_ooffset_t offset;
2435
2436 vm_map_lock_read(map);
2437 VM_MAP_RANGE_CHECK(map, start, end);
2438 if (!vm_map_lookup_entry(map, start, &entry)) {
2439 vm_map_unlock_read(map);
2440 return (KERN_INVALID_ADDRESS);
2441 }
2442 /*
2443 * Make a first pass to check for holes.
2444 */
2445 for (current = entry; current->start < end; current = current->next) {
1b874851 2446 if (current->maptype == VM_MAPTYPE_SUBMAP) {
984263bc
MD
2447 vm_map_unlock_read(map);
2448 return (KERN_INVALID_ARGUMENT);
2449 }
2450 if (end > current->end &&
2451 (current->next == &map->header ||
2452 current->end != current->next->start)) {
2453 vm_map_unlock_read(map);
2454 return (KERN_INVALID_ADDRESS);
2455 }
2456 }
2457
2458 if (invalidate)
2459 pmap_remove(vm_map_pmap(map), start, end);
46754a20 2460
984263bc
MD
2461 /*
2462 * Make a second pass, cleaning/uncaching pages from the indicated
2463 * objects as we go.
46754a20
MD
2464 *
2465 * Hold vm_token to avoid blocking in vm_object_reference()
984263bc 2466 */
46754a20 2467 lwkt_gettoken(&vm_token);
2de4f77e
MD
2468 lwkt_gettoken(&vmobj_token);
2469
984263bc
MD
2470 for (current = entry; current->start < end; current = current->next) {
2471 offset = current->offset + (start - current->start);
2472 size = (end <= current->end ? end : current->end) - start;
1b874851 2473 if (current->maptype == VM_MAPTYPE_SUBMAP) {
984263bc
MD
2474 vm_map_t smap;
2475 vm_map_entry_t tentry;
2476 vm_size_t tsize;
2477
2478 smap = current->object.sub_map;
2479 vm_map_lock_read(smap);
418ff780 2480 vm_map_lookup_entry(smap, offset, &tentry);
984263bc
MD
2481 tsize = tentry->end - offset;
2482 if (tsize < size)
2483 size = tsize;
2484 object = tentry->object.vm_object;
2485 offset = tentry->offset + (offset - tentry->start);
2486 vm_map_unlock_read(smap);
2487 } else {
2488 object = current->object.vm_object;
2489 }
2490 /*
2491 * Note that there is absolutely no sense in writing out
2492 * anonymous objects, so we track down the vnode object
2493 * to write out.
2494 * We invalidate (remove) all pages from the address space
2495 * anyway, for semantic correctness.
2496 *
2497 * note: certain anonymous maps, such as MAP_NOSYNC maps,
2498 * may start out with a NULL object.
2499 */
2500 while (object && object->backing_object) {
984263bc 2501 offset += object->backing_object_offset;
7cad6903 2502 object = object->backing_object;
984263bc
MD
2503 if (object->size < OFF_TO_IDX( offset + size))
2504 size = IDX_TO_OFF(object->size) - offset;
2505 }
2506 if (object && (object->type == OBJT_VNODE) &&
2bc7505b
MD
2507 (current->protection & VM_PROT_WRITE) &&
2508 (object->flags & OBJ_NOMSYNC) == 0) {
984263bc
MD
2509 /*
2510 * Flush pages if writing is allowed, invalidate them
2511 * if invalidation requested. Pages undergoing I/O
2512 * will be ignored by vm_object_page_remove().
2513 *
2514 * We cannot lock the vnode and then wait for paging
2515 * to complete without deadlocking against vm_fault.
2516 * Instead we simply call vm_object_page_remove() and
2517 * allow it to block internally on a page-by-page
2518 * basis when it encounters pages undergoing async
2519 * I/O.
2520 */
2521 int flags;
2522
2de4f77e 2523 vm_object_reference_locked(object);
ca466bae 2524 vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY);
984263bc
MD
2525 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
2526 flags |= invalidate ? OBJPC_INVAL : 0;
1b874851
MD
2527
2528 /*
2529 * When operating on a virtual page table just
2530 * flush the whole object. XXX we probably ought
2531 * to
2532 */
2533 switch(current->maptype) {
2534 case VM_MAPTYPE_NORMAL:
2535 vm_object_page_clean(object,
2536 OFF_TO_IDX(offset),
2537 OFF_TO_IDX(offset + size + PAGE_MASK),
2538 flags);
2539 break;
2540 case VM_MAPTYPE_VPAGETABLE:
2541 vm_object_page_clean(object, 0, 0, flags);
2542 break;
2543 }
a11aaa81 2544 vn_unlock(((struct vnode *)object->handle));
2de4f77e 2545 vm_object_deallocate_locked(object);
984263bc
MD
2546 }
2547 if (object && invalidate &&
2548 ((object->type == OBJT_VNODE) ||
2549 (object->type == OBJT_DEVICE))) {
2f1821ca
MD
2550 int clean_only =
2551 (object->type == OBJT_DEVICE) ? FALSE : TRUE;
2de4f77e 2552 vm_object_reference_locked(object);
1b874851
MD
2553 switch(current->maptype) {
2554 case VM_MAPTYPE_NORMAL:
2555 vm_object_page_remove(object,
2556 OFF_TO_IDX(offset),
2557 OFF_TO_IDX(offset + size + PAGE_MASK),
2558 clean_only);
2559 break;
2560 case VM_MAPTYPE_VPAGETABLE:
2561 vm_object_page_remove(object, 0, 0, clean_only);
2562 break;
2563 }
2de4f77e 2564 vm_object_deallocate_locked(object);
984263bc
MD
2565 }
2566 start += size;
2567 }
2de4f77e
MD
2568
2569 lwkt_reltoken(&vmobj_token);
46754a20 2570 lwkt_reltoken(&vm_token);
2de4f77e 2571 vm_map_unlock_read(map);
46754a20 2572
984263bc
MD
2573 return (KERN_SUCCESS);
2574}
2575
2576/*
46754a20 2577 * Make the region specified by this entry pageable.
984263bc 2578 *
46754a20 2579 * The vm_map must be exclusively locked.
984263bc
MD
2580 */
2581static void
a108bf71 2582vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
984263bc 2583{
f2d22ebf 2584 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
984263bc 2585 entry->wired_count = 0;
f2d22ebf 2586 vm_fault_unwire(map, entry);
984263bc
MD
2587}
2588
2589/*
46754a20 2590 * Deallocate the given entry from the target map.
984263bc 2591 *
46754a20 2592 * The vm_map must be exclusively locked.
984263bc
MD
2593 */
2594static void
a108bf71 2595vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry, int *countp)
984263bc
MD
2596{
2597 vm_map_entry_unlink(map, entry);
2598 map->size -= entry->end - entry->start;
2599
1b874851
MD
2600 switch(entry->maptype) {
2601 case VM_MAPTYPE_NORMAL:
2602 case VM_MAPTYPE_VPAGETABLE:
984263bc 2603 vm_object_deallocate(entry->object.vm_object);
1b874851
MD
2604 break;
2605 default:
2606 break;
984263bc
MD
2607 }
2608
a108bf71 2609 vm_map_entry_dispose(map, entry, countp);
984263bc
MD
2610}
2611
2612/*
46754a20 2613 * Deallocates the given address range from the target map.
984263bc 2614 *
46754a20 2615 * The vm_map must be exclusively locked.
984263bc
MD
2616 */
2617int
a108bf71 2618vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end, int *countp)
984263bc
MD
2619{
2620 vm_object_t object;
2621 vm_map_entry_t entry;
2622 vm_map_entry_t first_entry;
2623
46754a20 2624 ASSERT_VM_MAP_LOCKED(map);
686dbf64 2625again:
984263bc 2626 /*
686dbf64
MD
2627 * Find the start of the region, and clip it. Set entry to point
2628 * at the first record containing the requested address or, if no
2629 * such record exists, the next record with a greater address. The
2630 * loop will run from this point until a record beyond the termination
2631 * address is encountered.
2632 *
2633 * map->hint must be adjusted to not point to anything we delete,
2634 * so set it to the entry prior to the one being deleted.
2635 *
2636 * GGG see other GGG comment.
984263bc 2637 */
686dbf64 2638 if (vm_map_lookup_entry(map, start, &first_entry)) {
984263bc 2639 entry = first_entry;
a108bf71 2640 vm_map_clip_start(map, entry, start, countp);
686dbf64
MD
2641 map->hint = entry->prev; /* possible problem XXX */
2642 } else {
2643 map->hint = first_entry; /* possible problem XXX */
2644 entry = first_entry->next;
984263bc
MD
2645 }
2646
2647 /*
686dbf64
MD
2648 * If a hole opens up prior to the current first_free then
2649 * adjust first_free. As with map->hint, map->first_free
2650 * cannot be left set to anything we might delete.
984263bc 2651 */
984263bc
MD
2652 if (entry == &map->header) {
2653 map->first_free = &map->header;
2654 } else if (map->first_free->start >= start) {
2655 map->first_free = entry->prev;
2656 }
2657
2658 /*
2659 * Step through all entries in this region
2660 */
984263bc
MD
2661 while ((entry != &map->header) && (entry->start < end)) {
2662 vm_map_entry_t next;
2663 vm_offset_t s, e;
2664 vm_pindex_t offidxstart, offidxend, count;
2665
2666 /*
2667 * If we hit an in-transition entry we have to sleep and
2668 * retry. It's easier (and not really slower) to just retry
2669 * since this case occurs so rarely and the hint is already
2670 * pointing at the right place. We have to reset the
2671 * start offset so as not to accidently delete an entry
2672 * another process just created in vacated space.
2673 */
2674 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2675 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2676 start = entry->start;
12e4aaff
MD
2677 ++mycpu->gd_cnt.v_intrans_coll;
2678 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
2679 vm_map_transition_wait(map);
2680 goto again;
2681 }
a108bf71 2682 vm_map_clip_end(map, entry, end, countp);
984263bc
MD
2683
2684 s = entry->start;
2685 e = entry->end;
2686 next = entry->next;
2687
2688 offidxstart = OFF_TO_IDX(entry->offset);
2689 count = OFF_TO_IDX(e - s);
2690 object = entry->object.vm_object;
2691
2692 /*
2693 * Unwire before removing addresses from the pmap; otherwise,
2694 * unwiring will put the entries back in the pmap.
2695 */
f2d22ebf 2696 if (entry->wired_count != 0)
984263bc 2697 vm_map_entry_unwire(map, entry);
984263bc
MD
2698
2699 offidxend = offidxstart + count;
2700
46754a20 2701 /*
2de4f77e
MD
2702 * Hold vm_token when manipulating vm_objects,
2703 *
2704 * Hold vmobj_token when potentially adding or removing
2705 * objects (collapse requires both).
46754a20
MD
2706 */
2707 lwkt_gettoken(&vm_token);
2de4f77e
MD
2708 lwkt_gettoken(&vmobj_token);
2709
c439ad8f 2710 if (object == &kernel_object) {
46754a20
MD
2711 vm_object_page_remove(object, offidxstart,
2712 offidxend, FALSE);
984263bc
MD
2713 } else {
2714 pmap_remove(map->pmap, s, e);
2de4f77e 2715
984263bc
MD
2716 if (object != NULL &&
2717 object->ref_count != 1 &&
46754a20
MD
2718 (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) ==
2719 OBJ_ONEMAPPING &&
2720 (object->type == OBJT_DEFAULT ||
2721 object->type == OBJT_SWAP)) {
984263bc 2722 vm_object_collapse(object);
46754a20
MD
2723 vm_object_page_remove(object, offidxstart,
2724 offidxend, FALSE);
984263bc 2725 if (object->type == OBJT_SWAP) {
46754a20
MD
2726 swap_pager_freespace(object,
2727 offidxstart,
2728 count);
984263bc
MD
2729 }
2730 if (offidxend >= object->size &&
2731 offidxstart < object->size) {
2732 object->size = offidxstart;
2733 }
2734 }
2735 }
2de4f77e 2736 lwkt_reltoken(&vmobj_token);
46754a20 2737 lwkt_reltoken(&vm_token);
984263bc
MD
2738
2739 /*
2740 * Delete the entry (which may delete the object) only after
2741 * removing all pmap entries pointing to its pages.
2742 * (Otherwise, its page frames may be reallocated, and any
2743 * modify bits will be set in the wrong object!)
2744 */
a108bf71 2745 vm_map_entry_delete(map, entry, countp);
984263bc
MD
2746 entry = next;
2747 }
2748 return (KERN_SUCCESS);
2749}
2750
2751/*
46754a20
MD
2752 * Remove the given address range from the target map.
2753 * This is the exported form of vm_map_delete.
984263bc 2754 *
46754a20 2755 * No requirements.
984263bc
MD
2756 */
2757int
a108bf71 2758vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
984263bc 2759{
03aa8d99 2760 int result;
a108bf71 2761 int count;
984263bc 2762
a108bf71 2763 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
2764 vm_map_lock(map);
2765 VM_MAP_RANGE_CHECK(map, start, end);
a108bf71 2766 result = vm_map_delete(map, start, end, &count);
984263bc 2767 vm_map_unlock(map);
a108bf71 2768 vm_map_entry_release(count);
984263bc 2769
984263bc
MD
2770 return (result);
2771}
2772
2773/*
46754a20
MD
2774 * Assert that the target map allows the specified privilege on the
2775 * entire address region given. The entire region must be allocated.
984263bc 2776 *
46754a20 2777 * The caller must specify whether the vm_map is already locked or not.
984263bc
MD
2778 */
2779boolean_t
2780vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
46754a20 2781 vm_prot_t protection, boolean_t have_lock)
984263bc
MD
2782{
2783 vm_map_entry_t entry;
2784 vm_map_entry_t tmp_entry;
46754a20
MD
2785 boolean_t result;
2786
2787 if (have_lock == FALSE)
2788 vm_map_lock_read(map);
984263bc
MD
2789
2790 if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
46754a20
MD
2791 if (have_lock == FALSE)
2792 vm_map_unlock_read(map);
984263bc
MD
2793 return (FALSE);
2794 }
2795 entry = tmp_entry;
2796
46754a20 2797 result = TRUE;
984263bc
MD
2798 while (start < end) {
2799 if (entry == &map->header) {
46754a20
MD
2800 result = FALSE;
2801 break;
984263bc
MD
2802 }
2803 /*
2804 * No holes allowed!
2805 */
2806
2807 if (start < entry->start) {
46754a20
MD
2808 result = FALSE;
2809 break;
984263bc
MD
2810 }
2811 /*
2812 * Check protection associated with entry.
2813 */
2814
2815 if ((entry->protection & protection) != protection) {
46754a20
MD
2816 result = FALSE;
2817 break;
984263bc
MD
2818 }
2819 /* go to next entry */
2820
2821 start = entry->end;
2822 entry = entry->next;
2823 }
46754a20
MD
2824 if (have_lock == FALSE)
2825 vm_map_unlock_read(map);
2826 return (result);
984263bc
MD
2827}
2828
2829/*
2830 * Split the pages in a map entry into a new object. This affords
2831 * easier removal of unused pages, and keeps object inheritance from
2832 * being a negative impact on memory usage.
46754a20
MD
2833 *
2834 * The vm_map must be exclusively locked.
984263bc
MD
2835 */
2836static void
a108bf71 2837vm_map_split(vm_map_entry_t entry)
984263bc
MD
2838{
2839 vm_page_t m;
2840 vm_object_t orig_object, new_object, source;
2841 vm_offset_t s, e;
2842 vm_pindex_t offidxstart, offidxend, idx;
2843 vm_size_t size;
2844 vm_ooffset_t offset;
2845
2846 orig_object = entry->object.vm_object;
2847 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
2848 return;
2849 if (orig_object->ref_count <= 1)
2850 return;
2851
2852 offset = entry->offset;
2853 s = entry->start;
2854 e = entry->end;
2855
2856 offidxstart = OFF_TO_IDX(offset);
2857 offidxend = offidxstart + OFF_TO_IDX(e - s);
2858 size = offidxend - offidxstart;
2859
5a648714
MD
2860 switch(orig_object->type) {
2861 case OBJT_DEFAULT:
2862 new_object = default_pager_alloc(NULL, IDX_TO_OFF(size),
2863 VM_PROT_ALL, 0);
2864 break;
2865 case OBJT_SWAP:
2866 new_object = swap_pager_alloc(NULL, IDX_TO_OFF(size),
2867 VM_PROT_ALL, 0);
2868 break;
2869 default:
2870 /* not reached */
1570194c 2871 new_object = NULL;
5a648714
MD
2872 KKASSERT(0);
2873 }
984263bc
MD
2874 if (new_object == NULL)
2875 return;
2876
46754a20
MD
2877 /*
2878 * vm_token required when manipulating vm_objects.
2879 */
2880 lwkt_gettoken(&vm_token);
2de4f77e 2881 lwkt_gettoken(&vmobj_token);
46754a20 2882
984263bc
MD
2883 source = orig_object->backing_object;
2884 if (source != NULL) {
2de4f77e
MD
2885 /* Referenced by new_object */
2886 vm_object_reference_locked(source);
984263bc 2887 LIST_INSERT_HEAD(&source->shadow_head,
2de4f77e 2888 new_object, shadow_list);
984263bc
MD
2889 vm_object_clear_flag(source, OBJ_ONEMAPPING);
2890 new_object->backing_object_offset =
2de4f77e
MD
2891 orig_object->backing_object_offset +
2892 IDX_TO_OFF(offidxstart);
984263bc
MD
2893 new_object->backing_object = source;
2894 source->shadow_count++;
2895 source->generation++;
2896 }
2897
2898 for (idx = 0; idx < size; idx++) {
2899 vm_page_t m;
2900
2901 retry:
2902 m = vm_page_lookup(orig_object, offidxstart + idx);
2de4f77e 2903 if (m == NULL)
984263bc
MD
2904 continue;
2905
2906 /*
2907 * We must wait for pending I/O to complete before we can
2908 * rename the page.
2909 *
2910 * We do not have to VM_PROT_NONE the page as mappings should
2911 * not be changed by this operation.
2912 */
2913 if (vm_page_sleep_busy(m, TRUE, "spltwt"))
2914 goto retry;
984263bc
MD
2915 vm_page_busy(m);
2916 vm_page_rename(m, new_object, idx);
2917 /* page automatically made dirty by rename and cache handled */
2918 vm_page_busy(m);
2919 }
2920
2921 if (orig_object->type == OBJT_SWAP) {
2922 vm_object_pip_add(orig_object, 1);
2923 /*
2924 * copy orig_object pages into new_object
2925 * and destroy unneeded pages in
2926 * shadow object.
2927 */
2928 swap_pager_copy(orig_object, new_object, offidxstart, 0);
2929 vm_object_pip_wakeup(orig_object);
2930 }
2931
06ecca5a
MD
2932 /*
2933 * Wakeup the pages we played with. No spl protection is needed
2934 * for a simple wakeup.
2935 */
984263bc
MD
2936 for (idx = 0; idx < size; idx++) {
2937 m = vm_page_lookup(new_object, idx);
06ecca5a 2938 if (m)
984263bc 2939 vm_page_wakeup(m);
984263bc
MD
2940 }
2941
2942 entry->object.vm_object = new_object;
2943 entry->offset = 0LL;
2de4f77e
MD
2944 vm_object_deallocate_locked(orig_object);
2945 lwkt_reltoken(&vmobj_token);
46754a20 2946 lwkt_reltoken(&vm_token);
984263bc
MD
2947}
2948
2949/*
46754a20
MD
2950 * Copies the contents of the source entry to the destination
2951 * entry. The entries *must* be aligned properly.
984263bc 2952 *
46754a20 2953 * The vm_map must be exclusively locked.
2de4f77e 2954 * vm_token must be held
984263bc
MD
2955 */
2956static void
a108bf71
MD
2957vm_map_copy_entry(vm_map_t src_map, vm_map_t dst_map,
2958 vm_map_entry_t src_entry, vm_map_entry_t dst_entry)
984263bc
MD
2959{
2960 vm_object_t src_object;
2961
1b874851
MD
2962 if (dst_entry->maptype == VM_MAPTYPE_SUBMAP)
2963 return;
2964 if (src_entry->maptype == VM_MAPTYPE_SUBMAP)
984263bc
MD
2965 return;
2966
2de4f77e
MD
2967 ASSERT_LWKT_TOKEN_HELD(&vm_token);
2968 lwkt_gettoken(&vmobj_token); /* required for collapse */
2969
984263bc 2970 if (src_entry->wired_count == 0) {
984263bc
MD
2971 /*
2972 * If the source entry is marked needs_copy, it is already
2973 * write-protected.
2974 */
2975 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2976 pmap_protect(src_map->pmap,
2977 src_entry->start,
2978 src_entry->end,
2979 src_entry->protection & ~VM_PROT_WRITE);
2980 }
2981
2982 /*
2983 * Make a copy of the object.
2984 */
2985 if ((src_object = src_entry->object.vm_object) != NULL) {
984263bc
MD
2986 if ((src_object->handle == NULL) &&
2987 (src_object->type == OBJT_DEFAULT ||
2988 src_object->type == OBJT_SWAP)) {
2989 vm_object_collapse(src_object);
2990 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2991 vm_map_split(src_entry);
2992 src_object = src_entry->object.vm_object;
2993 }
2994 }
2995
2de4f77e 2996 vm_object_reference_locked(src_object);
984263bc
MD
2997 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2998 dst_entry->object.vm_object = src_object;
2999 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3000 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
3001 dst_entry->offset = src_entry->offset;
3002 } else {
3003 dst_entry->object.vm_object = NULL;
3004 dst_entry->offset = 0;
3005 }
3006
3007 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
3008 dst_entry->end - dst_entry->start, src_entry->start);
3009 } else {
3010 /*
3011 * Of course, wired down pages can't be set copy-on-write.
3012 * Cause wired pages to be copied into the new map by
3013 * simulating faults (the new pages are pageable)
3014 */
3015 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
3016 }
2de4f77e 3017 lwkt_reltoken(&vmobj_token);
984263bc
MD
3018}
3019
3020/*
3021 * vmspace_fork:
3022 * Create a new process vmspace structure and vm_map
3023 * based on those of an existing process. The new map
3024 * is based on the old map, according to the inheritance
3025 * values on the regions in that map.
3026 *
3027 * The source map must not be locked.
46754a20 3028 * No requirements.
984263bc
MD
3029 */
3030struct vmspace *
a108bf71 3031vmspace_fork(struct vmspace *vm1)
984263bc
MD
3032{
3033 struct vmspace *vm2;
3034 vm_map_t old_map = &vm1->vm_map;
3035 vm_map_t new_map;
3036 vm_map_entry_t old_entry;
3037 vm_map_entry_t new_entry;
3038 vm_object_t object;
a108bf71 3039 int count;
984263bc 3040
46754a20
MD
3041 lwkt_gettoken(&vm_token);
3042 lwkt_gettoken(&vmspace_token);
2de4f77e 3043 lwkt_gettoken(&vmobj_token);
984263bc
MD
3044 vm_map_lock(old_map);
3045 old_map->infork = 1;
3046
239b4df9
MD
3047 /*
3048 * XXX Note: upcalls are not copied.
3049 */
984263bc
MD
3050 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3051 bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
239b4df9 3052 (caddr_t)&vm1->vm_endcopy - (caddr_t)&vm1->vm_startcopy);
984263bc
MD
3053 new_map = &vm2->vm_map; /* XXX */
3054 new_map->timestamp = 1;
3055
46754a20
MD
3056 vm_map_lock(new_map);
3057
a108bf71 3058 count = 0;
984263bc 3059 old_entry = old_map->header.next;
a108bf71
MD
3060 while (old_entry != &old_map->header) {
3061 ++count;
3062 old_entry = old_entry->next;
3063 }
984263bc 3064
a108bf71
MD
3065 count = vm_map_entry_reserve(count + MAP_RESERVE_COUNT);
3066
3067 old_entry = old_map->header.next;
984263bc 3068 while (old_entry != &old_map->header) {
1b874851 3069 if (old_entry->maptype == VM_MAPTYPE_SUBMAP)
984263bc
MD
3070 panic("vm_map_fork: encountered a submap");
3071
3072 switch (old_entry->inheritance) {
3073 case VM_INHERIT_NONE:
3074 break;
984263bc
MD
3075 case VM_INHERIT_SHARE:
3076 /*
1b874851
MD
3077 * Clone the entry, creating the shared object if
3078 * necessary.
984263bc
MD
3079 */
3080 object = old_entry->object.vm_object;
3081 if (object == NULL) {
53025830
MD
3082 vm_map_entry_allocate_object(old_entry);
3083 object = old_entry->object.vm_object;
984263bc
MD
3084 }
3085
3086 /*
53025830 3087 * Add the reference before calling vm_map_entry_shadow
984263bc
MD
3088 * to insure that a shadow object is created.
3089 */
2de4f77e 3090 vm_object_reference_locked(object);
984263bc 3091 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
53025830 3092 vm_map_entry_shadow(old_entry);
984263bc 3093 /* Transfer the second reference too. */
2de4f77e 3094 vm_object_reference_locked(
984263bc 3095 old_entry->object.vm_object);
2de4f77e 3096 vm_object_deallocate_locked(object);
984263bc
MD
3097 object = old_entry->object.vm_object;
3098 }
3099 vm_object_clear_flag(object, OBJ_ONEMAPPING);
3100
3101 /*
3102 * Clone the entry, referencing the shared object.
3103 */
a108bf71 3104 new_entry = vm_map_entry_create(new_map, &count);
984263bc
MD
3105 *new_entry = *old_entry;
3106 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3107 new_entry->wired_count = 0;
3108
3109 /*
3110 * Insert the entry into the new map -- we know we're
3111 * inserting at the end of the new map.
3112 */
3113
3114 vm_map_entry_link(new_map, new_map->header.prev,
46754a20 3115 new_entry);
984263bc
MD
3116
3117 /*
3118 * Update the physical map
3119 */
984263bc
MD
3120 pmap_copy(new_map->pmap, old_map->pmap,
3121 new_entry->start,
3122 (old_entry->end - old_entry->start),
3123 old_entry->start);
3124 break;
984263bc
MD
3125 case VM_INHERIT_COPY:
3126 /*
3127 * Clone the entry and link into the map.
3128 */
a108bf71 3129 new_entry = vm_map_entry_create(new_map, &count);
984263bc
MD
3130 *new_entry = *old_entry;
3131 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3132 new_entry->wired_count = 0;
3133 new_entry->object.vm_object = NULL;
3134 vm_map_entry_link(new_map, new_map->header.prev,
46754a20 3135 new_entry);
984263bc 3136 vm_map_copy_entry(old_map, new_map, old_entry,
46754a20 3137 new_entry);
984263bc
MD
3138 break;
3139 }
3140 old_entry = old_entry->next;
3141 }
3142
3143 new_map->size = old_map->size;
3144 old_map->infork = 0;
3145 vm_map_unlock(old_map);
46754a20 3146 vm_map_unlock(new_map);
a108bf71 3147 vm_map_entry_release(count);
2de4f77e
MD
3148
3149 lwkt_reltoken(&vmobj_token);
46754a20
MD
3150 lwkt_reltoken(&vmspace_token);
3151 lwkt_reltoken(&vm_token);
984263bc
MD
3152
3153 return (vm2);
3154}
3155
46754a20
MD
3156/*
3157 * Create an auto-grow stack entry
3158 *
3159 * No requirements.
3160 */
984263bc
MD
3161int
3162vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
c809941b 3163 int flags, vm_prot_t prot, vm_prot_t max, int cow)
984263bc 3164{
85d25bcf
MD
3165 vm_map_entry_t prev_entry;
3166 vm_map_entry_t new_stack_entry;
3167 vm_size_t init_ssize;
3168 int rv;
a108bf71 3169 int count;
85d25bcf 3170 vm_offset_t tmpaddr;
984263bc 3171
c809941b 3172 cow |= MAP_IS_STACK;
984263bc
MD
3173
3174 if (max_ssize < sgrowsiz)
3175 init_ssize = max_ssize;
3176 else
3177 init_ssize = sgrowsiz;
3178
a108bf71 3179 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
3180 vm_map_lock(map);
3181
85d25bcf
MD
3182 /*
3183 * Find space for the mapping
3184 */
cadb984b 3185 if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) {
c809941b
MD
3186 if (vm_map_findspace(map, addrbos, max_ssize, 1,
3187 flags, &tmpaddr)) {
85d25bcf
MD
3188 vm_map_unlock(map);
3189 vm_map_entry_release(count);
3190 return (KERN_NO_SPACE);
3191 }
3192 addrbos = tmpaddr;
3193 }
3194
984263bc
MD
3195 /* If addr is already mapped, no go */
3196 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3197 vm_map_unlock(map);
a108bf71 3198 vm_map_entry_release(count);
984263bc
MD
3199 return (KERN_NO_SPACE);
3200 }
3201
85d25bcf
MD
3202#if 0
3203 /* XXX already handled by kern_mmap() */
984263bc
MD
3204 /* If we would blow our VMEM resource limit, no go */
3205 if (map->size + init_ssize >
3206 curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
3207 vm_map_unlock(map);
a108bf71 3208 vm_map_entry_release(count);
984263bc
MD
3209 return (KERN_NO_SPACE);
3210 }
85d25bcf 3211#endif
984263bc 3212
85d25bcf
MD
3213 /*
3214 * If we can't accomodate max_ssize in the current mapping,
984263bc
MD
3215 * no go. However, we need to be aware that subsequent user
3216 * mappings might map into the space we have reserved for
3217 * stack, and currently this space is not protected.
3218 *
3219 * Hopefully we will at least detect this condition
3220 * when we try to grow the stack.
3221 */
3222 if ((prev_entry->next != &map->header) &&
3223 (prev_entry->next->start < addrbos + max_ssize)) {
3224 vm_map_unlock(map);
a108bf71 3225 vm_map_entry_release(count);
984263bc
MD
3226 return (KERN_NO_SPACE);
3227 }
3228
85d25bcf
MD
3229 /*
3230 * We initially map a stack of only init_ssize. We will
984263bc
MD
3231 * grow as needed later. Since this is to be a grow
3232 * down stack, we map at the top of the range.
3233 *
3234 * Note: we would normally expect prot and max to be
3235 * VM_PROT_ALL, and cow to be 0. Possibly we should
3236 * eliminate these as input parameters, and just
3237 * pass these values here in the insert call.
3238 */
a108bf71
MD
3239 rv = vm_map_insert(map, &count,
3240 NULL, 0, addrbos + max_ssize - init_ssize,
1b874851
MD
3241 addrbos + max_ssize,
3242 VM_MAPTYPE_NORMAL,
3243 prot, max,
3244 cow);
984263bc
MD
3245
3246 /* Now set the avail_ssize amount */
517e1666 3247 if (rv == KERN_SUCCESS) {
984263bc 3248 if (prev_entry != &map->header)
a108bf71 3249 vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize, &count);
984263bc
MD
3250 new_stack_entry = prev_entry->next;
3251 if (new_stack_entry->end != addrbos + max_ssize ||
3252 new_stack_entry->start != addrbos + max_ssize - init_ssize)
3253 panic ("Bad entry start/end for new stack entry");
3254 else
afeabdca 3255 new_stack_entry->aux.avail_ssize = max_ssize - init_ssize;
984263bc
MD
3256 }
3257
3258 vm_map_unlock(map);
a108bf71 3259 vm_map_entry_release(count);
984263bc
MD
3260 return (rv);
3261}
3262
46754a20
MD
3263/*
3264 * Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
984263bc
MD
3265 * desired address is already mapped, or if we successfully grow
3266 * the stack. Also returns KERN_SUCCESS if addr is outside the
3267 * stack range (this is strange, but preserves compatibility with
3268 * the grow function in vm_machdep.c).
46754a20
MD
3269 *
3270 * No requirements.
984263bc
MD
3271 */
3272int
3273vm_map_growstack (struct proc *p, vm_offset_t addr)
3274{
3275 vm_map_entry_t prev_entry;
3276 vm_map_entry_t stack_entry;
3277 vm_map_entry_t new_stack_entry;
3278 struct vmspace *vm = p->p_vmspace;
3279 vm_map_t map = &vm->vm_map;
3280 vm_offset_t end;
a108bf71
MD
3281 int grow_amount;
3282 int rv = KERN_SUCCESS;
3283 int is_procstack;
3284 int use_read_lock = 1;
3285 int count;
984263bc 3286
a108bf71 3287 count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
984263bc
MD
3288Retry:
3289 if (use_read_lock)
3290 vm_map_lock_read(map);
3291 else
3292 vm_map_lock(map);
3293
3294 /* If addr is already in the entry range, no need to grow.*/
3295 if (vm_map_lookup_entry(map, addr, &prev_entry))
3296 goto done;
3297
3298 if ((stack_entry = prev_entry->next) == &map->header)
3299 goto done;
3300 if (prev_entry == &map->header)
afeabdca 3301 end = stack_entry->start - stack_entry->aux.avail_ssize;
984263bc
MD
3302 else
3303 end = prev_entry->end;
3304
c809941b
MD
3305 /*
3306 * This next test mimics the old grow function in vm_machdep.c.
984263bc
MD
3307 * It really doesn't quite make sense, but we do it anyway
3308 * for compatibility.
3309 *
3310 * If not growable stack, return success. This signals the
3311 * caller to proceed as he would normally with normal vm.
3312 */
afeabdca 3313 if (stack_entry->aux.avail_ssize < 1 ||
984263bc 3314 addr >= stack_entry->start ||
afeabdca 3315 addr < stack_entry->start - stack_entry->aux.avail_ssize) {
984263bc
MD
3316 goto done;
3317 }
3318
3319 /* Find the minimum grow amount */
3320 grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
afeabdca 3321 if (grow_amount > stack_entry->aux.avail_ssize) {
984263bc
MD
3322 rv = KERN_NO_SPACE;
3323 goto done;
3324 }
3325
c809941b
MD
3326 /*
3327 * If there is no longer enough space between the entries
984263bc
MD
3328 * nogo, and adjust the available space. Note: this
3329 * should only happen if the user has mapped into the
3330 * stack area after the stack was created, and is
3331 * probably an error.
3332 *
3333 * This also effectively destroys any guard page the user
3334 * might have intended by limiting the stack size.
3335 */
3336 if (grow_amount > stack_entry->start - end) {
3337 if (use_read_lock && vm_map_lock_upgrade(map)) {
3338 use_read_lock = 0;
3339 goto Retry;
3340 }
3341 use_read_lock = 0;
afeabdca 3342 stack_entry->aux.avail_ssize = stack_entry->start - end;
984263bc
MD
3343 rv = KERN_NO_SPACE;
3344 goto done;
3345 }
3346
3347 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
3348
3349 /* If this is the main process stack, see if we're over the
3350 * stack limit.
3351 */
3352 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
3353 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
3354 rv = KERN_NO_SPACE;
3355 goto done;
3356 }
3357
3358 /* Round up the grow amount modulo SGROWSIZ */
3359 grow_amount = roundup (grow_amount, sgrowsiz);
afeabdca
MD
3360 if (grow_amount > stack_entry->aux.avail_ssize) {
3361 grow_amount = stack_entry->aux.avail_ssize;
984263bc
MD
3362 }
3363 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
3364 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
3365 grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
3366 ctob(vm->vm_ssize);
3367 }
3368
3369 /* If we would blow our VMEM resource limit, no go */
be77b5f9 3370 if (map->size + grow_amount > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
984263bc
MD
3371 rv = KERN_NO_SPACE;
3372 goto done;
3373 }
3374
3375 if (use_read_lock && vm_map_lock_upgrade(map)) {
3376 use_read_lock = 0;
3377 goto Retry;
3378 }
3379 use_read_lock = 0;
3380
3381 /* Get the preliminary new entry start value */
3382 addr = stack_entry->start - grow_amount;
3383
3384 /* If this puts us into the previous entry, cut back our growth
3385 * to the available space. Also, see the note above.
3386 */
3387 if (addr < end) {
afeabdca 3388 stack_entry->aux.avail_ssize = stack_entry->start - end;
984263bc
MD
3389 addr = end;
3390 }
3391
a108bf71
MD
3392 rv = vm_map_insert(map, &count,
3393 NULL, 0, addr, stack_entry->start,
1b874851
MD
3394 VM_MAPTYPE_NORMAL,
3395 VM_PROT_ALL, VM_PROT_ALL,
984263bc
MD
3396 0);
3397
3398 /* Adjust the available stack space by the amount we grew. */
3399 if (rv == KERN_SUCCESS) {
3400 if (prev_entry != &map->header)
a108bf71 3401 vm_map_clip_end(map, prev_entry, addr, &count);
984263bc
MD
3402 new_stack_entry = prev_entry->next;
3403 if (new_stack_entry->end != stack_entry->start ||
3404 new_stack_entry->start != addr)
3405 panic ("Bad stack grow start/end in new stack entry");
3406 else {
afeabdca
MD
3407 new_stack_entry->aux.avail_ssize =
3408 stack_entry->aux.avail_ssize -
3409 (new_stack_entry->end - new_stack_entry->start);
984263bc
MD
3410 if (is_procstack)
3411 vm->vm_ssize += btoc(new_stack_entry->end -
3412 new_stack_entry->start);
3413 }
7c553423
VS
3414
3415 if (map->flags & MAP_WIREFUTURE)
3416 vm_map_unwire(map, new_stack_entry->start,
3417 new_stack_entry->end, FALSE);
984263bc
MD
3418 }
3419
3420done:
3421 if (use_read_lock)
3422 vm_map_unlock_read(map);
3423 else
3424 vm_map_unlock(map);
a108bf71 3425 vm_map_entry_release(count);
984263bc
MD
3426 return (rv);
3427}
3428
3429/*
3430 * Unshare the specified VM space for exec. If other processes are
3431 * mapped to it, then create a new one. The new vmspace is null.
46754a20
MD
3432 *
3433 * No requirements.
984263bc 3434 */
984263bc 3435void
29802dbb 3436vmspace_exec(struct proc *p, struct vmspace *vmcopy)
a108bf71 3437{
984263bc
MD
3438 struct vmspace *oldvmspace = p->p_vmspace;
3439 struct vmspace *newvmspace;
3440 vm_map_t map = &p->p_vmspace->vm_map;
3441
29802dbb
MD
3442 /*
3443 * If we are execing a resident vmspace we fork it, otherwise
239b4df9
MD
3444 * we create a new vmspace. Note that exitingcnt and upcalls
3445 * are not copied to the new vmspace.
29802dbb 3446 */
46754a20 3447 lwkt_gettoken(&vmspace_token);
29802dbb 3448 if (vmcopy) {
46754a20 3449 newvmspace = vmspace_fork(vmcopy);
29802dbb 3450 } else {
46754a20
MD
3451 newvmspace = vmspace_alloc(map->min_offset, map->max_offset);
3452 bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
3453 (caddr_t)&oldvmspace->vm_endcopy -
3454 (caddr_t)&oldvmspace->vm_startcopy);
29802dbb
MD
3455 }
3456
984263bc 3457 /*
e3161323
MD<