]> bbs.cooldavid.org Git - net-next-2.6.git/blame - mm/rmap.c
mm: fix section mismatch in memory_hotplug.c
[net-next-2.6.git] / mm / rmap.c
CommitLineData
1da177e4
LT
1/*
2 * mm/rmap.c - physical to virtual reverse mappings
3 *
4 * Copyright 2001, Rik van Riel <riel@conectiva.com.br>
5 * Released under the General Public License (GPL).
6 *
7 * Simple, low overhead reverse mapping scheme.
8 * Please try to keep this thing as modular as possible.
9 *
10 * Provides methods for unmapping each kind of mapped page:
11 * the anon methods track anonymous pages, and
12 * the file methods track pages belonging to an inode.
13 *
14 * Original design by Rik van Riel <riel@conectiva.com.br> 2001
15 * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004
16 * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004
98f32602 17 * Contributions by Hugh Dickins 2003, 2004
1da177e4
LT
18 */
19
20/*
21 * Lock ordering in mm:
22 *
1b1dcc1b 23 * inode->i_mutex (while writing or truncating, not reading or faulting)
82591e6e
NP
24 * inode->i_alloc_sem (vmtruncate_range)
25 * mm->mmap_sem
26 * page->flags PG_locked (lock_page)
27 * mapping->i_mmap_lock
28 * anon_vma->lock
29 * mm->page_table_lock or pte_lock
30 * zone->lru_lock (in mark_page_accessed, isolate_lru_page)
31 * swap_lock (in swap_duplicate, swap_info_get)
32 * mmlist_lock (in mmput, drain_mmlist and others)
33 * mapping->private_lock (in __set_page_dirty_buffers)
34 * inode_lock (in set_page_dirty's __mark_inode_dirty)
35 * sb_lock (within inode_lock in fs/fs-writeback.c)
36 * mapping->tree_lock (widely used, in set_page_dirty,
37 * in arch-dependent flush_dcache_mmap_lock,
38 * within inode_lock in __sync_single_inode)
6a46079c
AK
39 *
40 * (code doesn't rely on that order so it could be switched around)
41 * ->tasklist_lock
42 * anon_vma->lock (memory_failure, collect_procs_anon)
43 * pte map lock
1da177e4
LT
44 */
45
46#include <linux/mm.h>
47#include <linux/pagemap.h>
48#include <linux/swap.h>
49#include <linux/swapops.h>
50#include <linux/slab.h>
51#include <linux/init.h>
5ad64688 52#include <linux/ksm.h>
1da177e4
LT
53#include <linux/rmap.h>
54#include <linux/rcupdate.h>
a48d07af 55#include <linux/module.h>
8a9f3ccd 56#include <linux/memcontrol.h>
cddb8a5c 57#include <linux/mmu_notifier.h>
64cdd548 58#include <linux/migrate.h>
1da177e4
LT
59
60#include <asm/tlbflush.h>
61
b291f000
NP
62#include "internal.h"
63
fdd2e5f8
AB
64static struct kmem_cache *anon_vma_cachep;
65
66static inline struct anon_vma *anon_vma_alloc(void)
67{
68 return kmem_cache_alloc(anon_vma_cachep, GFP_KERNEL);
69}
70
db114b83 71void anon_vma_free(struct anon_vma *anon_vma)
fdd2e5f8
AB
72{
73 kmem_cache_free(anon_vma_cachep, anon_vma);
74}
1da177e4 75
d9d332e0
LT
76/**
77 * anon_vma_prepare - attach an anon_vma to a memory region
78 * @vma: the memory region in question
79 *
80 * This makes sure the memory mapping described by 'vma' has
81 * an 'anon_vma' attached to it, so that we can associate the
82 * anonymous pages mapped into it with that anon_vma.
83 *
84 * The common case will be that we already have one, but if
85 * if not we either need to find an adjacent mapping that we
86 * can re-use the anon_vma from (very common when the only
87 * reason for splitting a vma has been mprotect()), or we
88 * allocate a new one.
89 *
90 * Anon-vma allocations are very subtle, because we may have
91 * optimistically looked up an anon_vma in page_lock_anon_vma()
92 * and that may actually touch the spinlock even in the newly
93 * allocated vma (it depends on RCU to make sure that the
94 * anon_vma isn't actually destroyed).
95 *
96 * As a result, we need to do proper anon_vma locking even
97 * for the new allocation. At the same time, we do not want
98 * to do any locking for the common case of already having
99 * an anon_vma.
100 *
101 * This must be called with the mmap_sem held for reading.
102 */
1da177e4
LT
103int anon_vma_prepare(struct vm_area_struct *vma)
104{
105 struct anon_vma *anon_vma = vma->anon_vma;
106
107 might_sleep();
108 if (unlikely(!anon_vma)) {
109 struct mm_struct *mm = vma->vm_mm;
d9d332e0 110 struct anon_vma *allocated;
1da177e4
LT
111
112 anon_vma = find_mergeable_anon_vma(vma);
d9d332e0
LT
113 allocated = NULL;
114 if (!anon_vma) {
1da177e4
LT
115 anon_vma = anon_vma_alloc();
116 if (unlikely(!anon_vma))
117 return -ENOMEM;
118 allocated = anon_vma;
1da177e4 119 }
d9d332e0 120 spin_lock(&anon_vma->lock);
1da177e4
LT
121
122 /* page_table_lock to protect against threads */
123 spin_lock(&mm->page_table_lock);
124 if (likely(!vma->anon_vma)) {
125 vma->anon_vma = anon_vma;
0697212a 126 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
1da177e4
LT
127 allocated = NULL;
128 }
129 spin_unlock(&mm->page_table_lock);
130
d9d332e0 131 spin_unlock(&anon_vma->lock);
1da177e4
LT
132 if (unlikely(allocated))
133 anon_vma_free(allocated);
134 }
135 return 0;
136}
137
138void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next)
139{
140 BUG_ON(vma->anon_vma != next->anon_vma);
141 list_del(&next->anon_vma_node);
142}
143
144void __anon_vma_link(struct vm_area_struct *vma)
145{
146 struct anon_vma *anon_vma = vma->anon_vma;
147
30acbaba 148 if (anon_vma)
0697212a 149 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
1da177e4
LT
150}
151
152void anon_vma_link(struct vm_area_struct *vma)
153{
154 struct anon_vma *anon_vma = vma->anon_vma;
155
156 if (anon_vma) {
157 spin_lock(&anon_vma->lock);
0697212a 158 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
1da177e4
LT
159 spin_unlock(&anon_vma->lock);
160 }
161}
162
163void anon_vma_unlink(struct vm_area_struct *vma)
164{
165 struct anon_vma *anon_vma = vma->anon_vma;
166 int empty;
167
168 if (!anon_vma)
169 return;
170
171 spin_lock(&anon_vma->lock);
1da177e4
LT
172 list_del(&vma->anon_vma_node);
173
174 /* We must garbage collect the anon_vma if it's empty */
db114b83 175 empty = list_empty(&anon_vma->head) && !ksm_refcount(anon_vma);
1da177e4
LT
176 spin_unlock(&anon_vma->lock);
177
178 if (empty)
179 anon_vma_free(anon_vma);
180}
181
51cc5068 182static void anon_vma_ctor(void *data)
1da177e4 183{
a35afb83 184 struct anon_vma *anon_vma = data;
1da177e4 185
a35afb83 186 spin_lock_init(&anon_vma->lock);
db114b83 187 ksm_refcount_init(anon_vma);
a35afb83 188 INIT_LIST_HEAD(&anon_vma->head);
1da177e4
LT
189}
190
191void __init anon_vma_init(void)
192{
193 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
20c2df83 194 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
1da177e4
LT
195}
196
197/*
198 * Getting a lock on a stable anon_vma from a page off the LRU is
199 * tricky: page_lock_anon_vma rely on RCU to guard against the races.
200 */
10be22df 201struct anon_vma *page_lock_anon_vma(struct page *page)
1da177e4 202{
34bbd704 203 struct anon_vma *anon_vma;
1da177e4
LT
204 unsigned long anon_mapping;
205
206 rcu_read_lock();
80e14822 207 anon_mapping = (unsigned long) ACCESS_ONCE(page->mapping);
3ca7b3c5 208 if ((anon_mapping & PAGE_MAPPING_FLAGS) != PAGE_MAPPING_ANON)
1da177e4
LT
209 goto out;
210 if (!page_mapped(page))
211 goto out;
212
213 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
214 spin_lock(&anon_vma->lock);
34bbd704 215 return anon_vma;
1da177e4
LT
216out:
217 rcu_read_unlock();
34bbd704
ON
218 return NULL;
219}
220
10be22df 221void page_unlock_anon_vma(struct anon_vma *anon_vma)
34bbd704
ON
222{
223 spin_unlock(&anon_vma->lock);
224 rcu_read_unlock();
1da177e4
LT
225}
226
227/*
3ad33b24
LS
228 * At what user virtual address is page expected in @vma?
229 * Returns virtual address or -EFAULT if page's index/offset is not
230 * within the range mapped the @vma.
1da177e4
LT
231 */
232static inline unsigned long
233vma_address(struct page *page, struct vm_area_struct *vma)
234{
235 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
236 unsigned long address;
237
238 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
239 if (unlikely(address < vma->vm_start || address >= vma->vm_end)) {
3ad33b24 240 /* page should be within @vma mapping range */
1da177e4
LT
241 return -EFAULT;
242 }
243 return address;
244}
245
246/*
bf89c8c8
HS
247 * At what user virtual address is page expected in vma?
248 * checking that the page matches the vma.
1da177e4
LT
249 */
250unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
251{
252 if (PageAnon(page)) {
3ca7b3c5 253 if (vma->anon_vma != page_anon_vma(page))
1da177e4
LT
254 return -EFAULT;
255 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
ee498ed7
HD
256 if (!vma->vm_file ||
257 vma->vm_file->f_mapping != page->mapping)
1da177e4
LT
258 return -EFAULT;
259 } else
260 return -EFAULT;
261 return vma_address(page, vma);
262}
263
81b4082d
ND
264/*
265 * Check that @page is mapped at @address into @mm.
266 *
479db0bf
NP
267 * If @sync is false, page_check_address may perform a racy check to avoid
268 * the page table lock when the pte is not present (helpful when reclaiming
269 * highly shared pages).
270 *
b8072f09 271 * On success returns with pte mapped and locked.
81b4082d 272 */
ceffc078 273pte_t *page_check_address(struct page *page, struct mm_struct *mm,
479db0bf 274 unsigned long address, spinlock_t **ptlp, int sync)
81b4082d
ND
275{
276 pgd_t *pgd;
277 pud_t *pud;
278 pmd_t *pmd;
279 pte_t *pte;
c0718806 280 spinlock_t *ptl;
81b4082d 281
81b4082d 282 pgd = pgd_offset(mm, address);
c0718806
HD
283 if (!pgd_present(*pgd))
284 return NULL;
285
286 pud = pud_offset(pgd, address);
287 if (!pud_present(*pud))
288 return NULL;
289
290 pmd = pmd_offset(pud, address);
291 if (!pmd_present(*pmd))
292 return NULL;
293
294 pte = pte_offset_map(pmd, address);
295 /* Make a quick check before getting the lock */
479db0bf 296 if (!sync && !pte_present(*pte)) {
c0718806
HD
297 pte_unmap(pte);
298 return NULL;
299 }
300
4c21e2f2 301 ptl = pte_lockptr(mm, pmd);
c0718806
HD
302 spin_lock(ptl);
303 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
304 *ptlp = ptl;
305 return pte;
81b4082d 306 }
c0718806
HD
307 pte_unmap_unlock(pte, ptl);
308 return NULL;
81b4082d
ND
309}
310
b291f000
NP
311/**
312 * page_mapped_in_vma - check whether a page is really mapped in a VMA
313 * @page: the page to test
314 * @vma: the VMA to test
315 *
316 * Returns 1 if the page is mapped into the page tables of the VMA, 0
317 * if the page is not mapped into the page tables of this VMA. Only
318 * valid for normal file or anonymous VMAs.
319 */
6a46079c 320int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
b291f000
NP
321{
322 unsigned long address;
323 pte_t *pte;
324 spinlock_t *ptl;
325
326 address = vma_address(page, vma);
327 if (address == -EFAULT) /* out of vma range */
328 return 0;
329 pte = page_check_address(page, vma->vm_mm, address, &ptl, 1);
330 if (!pte) /* the page is not in this mm */
331 return 0;
332 pte_unmap_unlock(pte, ptl);
333
334 return 1;
335}
336
1da177e4
LT
337/*
338 * Subfunctions of page_referenced: page_referenced_one called
339 * repeatedly from either page_referenced_anon or page_referenced_file.
340 */
5ad64688
HD
341int page_referenced_one(struct page *page, struct vm_area_struct *vma,
342 unsigned long address, unsigned int *mapcount,
343 unsigned long *vm_flags)
1da177e4
LT
344{
345 struct mm_struct *mm = vma->vm_mm;
1da177e4 346 pte_t *pte;
c0718806 347 spinlock_t *ptl;
1da177e4
LT
348 int referenced = 0;
349
479db0bf 350 pte = page_check_address(page, mm, address, &ptl, 0);
c0718806
HD
351 if (!pte)
352 goto out;
1da177e4 353
b291f000
NP
354 /*
355 * Don't want to elevate referenced for mlocked page that gets this far,
356 * in order that it progresses to try_to_unmap and is moved to the
357 * unevictable list.
358 */
5a9bbdcd 359 if (vma->vm_flags & VM_LOCKED) {
5a9bbdcd 360 *mapcount = 1; /* break early from loop */
03ef83af 361 *vm_flags |= VM_LOCKED;
b291f000
NP
362 goto out_unmap;
363 }
364
4917e5d0
JW
365 if (ptep_clear_flush_young_notify(vma, address, pte)) {
366 /*
367 * Don't treat a reference through a sequentially read
368 * mapping as such. If the page has been used in
369 * another mapping, we will catch it; if this other
370 * mapping is already gone, the unmap path will have
371 * set PG_referenced or activated the page.
372 */
373 if (likely(!VM_SequentialReadHint(vma)))
374 referenced++;
375 }
1da177e4 376
c0718806
HD
377 /* Pretend the page is referenced if the task has the
378 swap token and is in the middle of a page fault. */
f7b7fd8f 379 if (mm != current->mm && has_swap_token(mm) &&
c0718806
HD
380 rwsem_is_locked(&mm->mmap_sem))
381 referenced++;
382
b291f000 383out_unmap:
c0718806
HD
384 (*mapcount)--;
385 pte_unmap_unlock(pte, ptl);
273f047e 386
6fe6b7e3
WF
387 if (referenced)
388 *vm_flags |= vma->vm_flags;
273f047e 389out:
1da177e4
LT
390 return referenced;
391}
392
bed7161a 393static int page_referenced_anon(struct page *page,
6fe6b7e3
WF
394 struct mem_cgroup *mem_cont,
395 unsigned long *vm_flags)
1da177e4
LT
396{
397 unsigned int mapcount;
398 struct anon_vma *anon_vma;
399 struct vm_area_struct *vma;
400 int referenced = 0;
401
402 anon_vma = page_lock_anon_vma(page);
403 if (!anon_vma)
404 return referenced;
405
406 mapcount = page_mapcount(page);
407 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
1cb1729b
HD
408 unsigned long address = vma_address(page, vma);
409 if (address == -EFAULT)
410 continue;
bed7161a
BS
411 /*
412 * If we are reclaiming on behalf of a cgroup, skip
413 * counting on behalf of references from different
414 * cgroups
415 */
bd845e38 416 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
bed7161a 417 continue;
1cb1729b 418 referenced += page_referenced_one(page, vma, address,
6fe6b7e3 419 &mapcount, vm_flags);
1da177e4
LT
420 if (!mapcount)
421 break;
422 }
34bbd704
ON
423
424 page_unlock_anon_vma(anon_vma);
1da177e4
LT
425 return referenced;
426}
427
428/**
429 * page_referenced_file - referenced check for object-based rmap
430 * @page: the page we're checking references on.
43d8eac4 431 * @mem_cont: target memory controller
6fe6b7e3 432 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
1da177e4
LT
433 *
434 * For an object-based mapped page, find all the places it is mapped and
435 * check/clear the referenced flag. This is done by following the page->mapping
436 * pointer, then walking the chain of vmas it holds. It returns the number
437 * of references it found.
438 *
439 * This function is only called from page_referenced for object-based pages.
440 */
bed7161a 441static int page_referenced_file(struct page *page,
6fe6b7e3
WF
442 struct mem_cgroup *mem_cont,
443 unsigned long *vm_flags)
1da177e4
LT
444{
445 unsigned int mapcount;
446 struct address_space *mapping = page->mapping;
447 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
448 struct vm_area_struct *vma;
449 struct prio_tree_iter iter;
450 int referenced = 0;
451
452 /*
453 * The caller's checks on page->mapping and !PageAnon have made
454 * sure that this is a file page: the check for page->mapping
455 * excludes the case just before it gets set on an anon page.
456 */
457 BUG_ON(PageAnon(page));
458
459 /*
460 * The page lock not only makes sure that page->mapping cannot
461 * suddenly be NULLified by truncation, it makes sure that the
462 * structure at mapping cannot be freed and reused yet,
463 * so we can safely take mapping->i_mmap_lock.
464 */
465 BUG_ON(!PageLocked(page));
466
467 spin_lock(&mapping->i_mmap_lock);
468
469 /*
470 * i_mmap_lock does not stabilize mapcount at all, but mapcount
471 * is more likely to be accurate if we note it after spinning.
472 */
473 mapcount = page_mapcount(page);
474
475 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
1cb1729b
HD
476 unsigned long address = vma_address(page, vma);
477 if (address == -EFAULT)
478 continue;
bed7161a
BS
479 /*
480 * If we are reclaiming on behalf of a cgroup, skip
481 * counting on behalf of references from different
482 * cgroups
483 */
bd845e38 484 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
bed7161a 485 continue;
1cb1729b 486 referenced += page_referenced_one(page, vma, address,
6fe6b7e3 487 &mapcount, vm_flags);
1da177e4
LT
488 if (!mapcount)
489 break;
490 }
491
492 spin_unlock(&mapping->i_mmap_lock);
493 return referenced;
494}
495
496/**
497 * page_referenced - test if the page was referenced
498 * @page: the page to test
499 * @is_locked: caller holds lock on the page
43d8eac4 500 * @mem_cont: target memory controller
6fe6b7e3 501 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
1da177e4
LT
502 *
503 * Quick test_and_clear_referenced for all mappings to a page,
504 * returns the number of ptes which referenced the page.
505 */
6fe6b7e3
WF
506int page_referenced(struct page *page,
507 int is_locked,
508 struct mem_cgroup *mem_cont,
509 unsigned long *vm_flags)
1da177e4
LT
510{
511 int referenced = 0;
5ad64688 512 int we_locked = 0;
1da177e4 513
1da177e4
LT
514 if (TestClearPageReferenced(page))
515 referenced++;
516
6fe6b7e3 517 *vm_flags = 0;
3ca7b3c5 518 if (page_mapped(page) && page_rmapping(page)) {
5ad64688
HD
519 if (!is_locked && (!PageAnon(page) || PageKsm(page))) {
520 we_locked = trylock_page(page);
521 if (!we_locked) {
522 referenced++;
523 goto out;
524 }
525 }
526 if (unlikely(PageKsm(page)))
527 referenced += page_referenced_ksm(page, mem_cont,
528 vm_flags);
529 else if (PageAnon(page))
6fe6b7e3
WF
530 referenced += page_referenced_anon(page, mem_cont,
531 vm_flags);
5ad64688 532 else if (page->mapping)
6fe6b7e3
WF
533 referenced += page_referenced_file(page, mem_cont,
534 vm_flags);
5ad64688 535 if (we_locked)
1da177e4 536 unlock_page(page);
1da177e4 537 }
5ad64688 538out:
5b7baf05
CB
539 if (page_test_and_clear_young(page))
540 referenced++;
541
1da177e4
LT
542 return referenced;
543}
544
1cb1729b
HD
545static int page_mkclean_one(struct page *page, struct vm_area_struct *vma,
546 unsigned long address)
d08b3851
PZ
547{
548 struct mm_struct *mm = vma->vm_mm;
c2fda5fe 549 pte_t *pte;
d08b3851
PZ
550 spinlock_t *ptl;
551 int ret = 0;
552
479db0bf 553 pte = page_check_address(page, mm, address, &ptl, 1);
d08b3851
PZ
554 if (!pte)
555 goto out;
556
c2fda5fe
PZ
557 if (pte_dirty(*pte) || pte_write(*pte)) {
558 pte_t entry;
d08b3851 559
c2fda5fe 560 flush_cache_page(vma, address, pte_pfn(*pte));
cddb8a5c 561 entry = ptep_clear_flush_notify(vma, address, pte);
c2fda5fe
PZ
562 entry = pte_wrprotect(entry);
563 entry = pte_mkclean(entry);
d6e88e67 564 set_pte_at(mm, address, pte, entry);
c2fda5fe
PZ
565 ret = 1;
566 }
d08b3851 567
d08b3851
PZ
568 pte_unmap_unlock(pte, ptl);
569out:
570 return ret;
571}
572
573static int page_mkclean_file(struct address_space *mapping, struct page *page)
574{
575 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
576 struct vm_area_struct *vma;
577 struct prio_tree_iter iter;
578 int ret = 0;
579
580 BUG_ON(PageAnon(page));
581
582 spin_lock(&mapping->i_mmap_lock);
583 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
1cb1729b
HD
584 if (vma->vm_flags & VM_SHARED) {
585 unsigned long address = vma_address(page, vma);
586 if (address == -EFAULT)
587 continue;
588 ret += page_mkclean_one(page, vma, address);
589 }
d08b3851
PZ
590 }
591 spin_unlock(&mapping->i_mmap_lock);
592 return ret;
593}
594
595int page_mkclean(struct page *page)
596{
597 int ret = 0;
598
599 BUG_ON(!PageLocked(page));
600
601 if (page_mapped(page)) {
602 struct address_space *mapping = page_mapping(page);
ce7e9fae 603 if (mapping) {
d08b3851 604 ret = page_mkclean_file(mapping, page);
ce7e9fae
CB
605 if (page_test_dirty(page)) {
606 page_clear_dirty(page);
607 ret = 1;
608 }
6c210482 609 }
d08b3851
PZ
610 }
611
612 return ret;
613}
60b59bea 614EXPORT_SYMBOL_GPL(page_mkclean);
d08b3851 615
9617d95e 616/**
43d8eac4 617 * __page_set_anon_rmap - setup new anonymous rmap
9617d95e
NP
618 * @page: the page to add the mapping to
619 * @vma: the vm area in which the mapping is added
620 * @address: the user virtual address mapped
621 */
622static void __page_set_anon_rmap(struct page *page,
623 struct vm_area_struct *vma, unsigned long address)
624{
625 struct anon_vma *anon_vma = vma->anon_vma;
626
627 BUG_ON(!anon_vma);
628 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
629 page->mapping = (struct address_space *) anon_vma;
9617d95e 630 page->index = linear_page_index(vma, address);
9617d95e
NP
631}
632
c97a9e10 633/**
43d8eac4 634 * __page_check_anon_rmap - sanity check anonymous rmap addition
c97a9e10
NP
635 * @page: the page to add the mapping to
636 * @vma: the vm area in which the mapping is added
637 * @address: the user virtual address mapped
638 */
639static void __page_check_anon_rmap(struct page *page,
640 struct vm_area_struct *vma, unsigned long address)
641{
642#ifdef CONFIG_DEBUG_VM
643 /*
644 * The page's anon-rmap details (mapping and index) are guaranteed to
645 * be set up correctly at this point.
646 *
647 * We have exclusion against page_add_anon_rmap because the caller
648 * always holds the page locked, except if called from page_dup_rmap,
649 * in which case the page is already known to be setup.
650 *
651 * We have exclusion against page_add_new_anon_rmap because those pages
652 * are initially only visible via the pagetables, and the pte is locked
653 * over the call to page_add_new_anon_rmap.
654 */
655 struct anon_vma *anon_vma = vma->anon_vma;
656 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
657 BUG_ON(page->mapping != (struct address_space *)anon_vma);
658 BUG_ON(page->index != linear_page_index(vma, address));
659#endif
660}
661
1da177e4
LT
662/**
663 * page_add_anon_rmap - add pte mapping to an anonymous page
664 * @page: the page to add the mapping to
665 * @vma: the vm area in which the mapping is added
666 * @address: the user virtual address mapped
667 *
5ad64688 668 * The caller needs to hold the pte lock, and the page must be locked in
80e14822
HD
669 * the anon_vma case: to serialize mapping,index checking after setting,
670 * and to ensure that PageAnon is not being upgraded racily to PageKsm
671 * (but PageKsm is never downgraded to PageAnon).
1da177e4
LT
672 */
673void page_add_anon_rmap(struct page *page,
674 struct vm_area_struct *vma, unsigned long address)
675{
5ad64688
HD
676 int first = atomic_inc_and_test(&page->_mapcount);
677 if (first)
678 __inc_zone_page_state(page, NR_ANON_PAGES);
679 if (unlikely(PageKsm(page)))
680 return;
681
c97a9e10
NP
682 VM_BUG_ON(!PageLocked(page));
683 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
5ad64688 684 if (first)
9617d95e 685 __page_set_anon_rmap(page, vma, address);
69029cd5 686 else
c97a9e10 687 __page_check_anon_rmap(page, vma, address);
1da177e4
LT
688}
689
43d8eac4 690/**
9617d95e
NP
691 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
692 * @page: the page to add the mapping to
693 * @vma: the vm area in which the mapping is added
694 * @address: the user virtual address mapped
695 *
696 * Same as page_add_anon_rmap but must only be called on *new* pages.
697 * This means the inc-and-test can be bypassed.
c97a9e10 698 * Page does not have to be locked.
9617d95e
NP
699 */
700void page_add_new_anon_rmap(struct page *page,
701 struct vm_area_struct *vma, unsigned long address)
702{
b5934c53 703 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
cbf84b7a
HD
704 SetPageSwapBacked(page);
705 atomic_set(&page->_mapcount, 0); /* increment count (starts at -1) */
5ad64688 706 __inc_zone_page_state(page, NR_ANON_PAGES);
9617d95e 707 __page_set_anon_rmap(page, vma, address);
b5934c53 708 if (page_evictable(page, vma))
cbf84b7a 709 lru_cache_add_lru(page, LRU_ACTIVE_ANON);
b5934c53
HD
710 else
711 add_page_to_unevictable_list(page);
9617d95e
NP
712}
713
1da177e4
LT
714/**
715 * page_add_file_rmap - add pte mapping to a file page
716 * @page: the page to add the mapping to
717 *
b8072f09 718 * The caller needs to hold the pte lock.
1da177e4
LT
719 */
720void page_add_file_rmap(struct page *page)
721{
d69b042f 722 if (atomic_inc_and_test(&page->_mapcount)) {
65ba55f5 723 __inc_zone_page_state(page, NR_FILE_MAPPED);
d69b042f
BS
724 mem_cgroup_update_mapped_file_stat(page, 1);
725 }
1da177e4
LT
726}
727
728/**
729 * page_remove_rmap - take down pte mapping from a page
730 * @page: page to remove mapping from
731 *
b8072f09 732 * The caller needs to hold the pte lock.
1da177e4 733 */
edc315fd 734void page_remove_rmap(struct page *page)
1da177e4 735{
b904dcfe
KM
736 /* page still mapped by someone else? */
737 if (!atomic_add_negative(-1, &page->_mapcount))
738 return;
739
740 /*
741 * Now that the last pte has gone, s390 must transfer dirty
742 * flag from storage key to struct page. We can usually skip
743 * this if the page is anon, so about to be freed; but perhaps
744 * not if it's in swapcache - there might be another pte slot
745 * containing the swap entry, but page not yet written to swap.
746 */
747 if ((!PageAnon(page) || PageSwapCache(page)) && page_test_dirty(page)) {
748 page_clear_dirty(page);
749 set_page_dirty(page);
1da177e4 750 }
b904dcfe
KM
751 if (PageAnon(page)) {
752 mem_cgroup_uncharge_page(page);
753 __dec_zone_page_state(page, NR_ANON_PAGES);
754 } else {
755 __dec_zone_page_state(page, NR_FILE_MAPPED);
756 }
757 mem_cgroup_update_mapped_file_stat(page, -1);
758 /*
759 * It would be tidy to reset the PageAnon mapping here,
760 * but that might overwrite a racing page_add_anon_rmap
761 * which increments mapcount after us but sets mapping
762 * before us: so leave the reset to free_hot_cold_page,
763 * and remember that it's only reliable while mapped.
764 * Leaving it set also helps swapoff to reinstate ptes
765 * faster for those pages still in swapcache.
766 */
1da177e4
LT
767}
768
769/*
770 * Subfunctions of try_to_unmap: try_to_unmap_one called
771 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
772 */
5ad64688
HD
773int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
774 unsigned long address, enum ttu_flags flags)
1da177e4
LT
775{
776 struct mm_struct *mm = vma->vm_mm;
1da177e4
LT
777 pte_t *pte;
778 pte_t pteval;
c0718806 779 spinlock_t *ptl;
1da177e4
LT
780 int ret = SWAP_AGAIN;
781
479db0bf 782 pte = page_check_address(page, mm, address, &ptl, 0);
c0718806 783 if (!pte)
81b4082d 784 goto out;
1da177e4
LT
785
786 /*
787 * If the page is mlock()d, we cannot swap it out.
788 * If it's recently referenced (perhaps page_referenced
789 * skipped over this mm) then we should reactivate it.
790 */
14fa31b8 791 if (!(flags & TTU_IGNORE_MLOCK)) {
b291f000
NP
792 if (vma->vm_flags & VM_LOCKED) {
793 ret = SWAP_MLOCK;
794 goto out_unmap;
795 }
af8e3354 796 if (TTU_ACTION(flags) == TTU_MUNLOCK)
53f79acb 797 goto out_unmap;
14fa31b8
AK
798 }
799 if (!(flags & TTU_IGNORE_ACCESS)) {
b291f000
NP
800 if (ptep_clear_flush_young_notify(vma, address, pte)) {
801 ret = SWAP_FAIL;
802 goto out_unmap;
803 }
804 }
1da177e4 805
1da177e4
LT
806 /* Nuke the page table entry. */
807 flush_cache_page(vma, address, page_to_pfn(page));
cddb8a5c 808 pteval = ptep_clear_flush_notify(vma, address, pte);
1da177e4
LT
809
810 /* Move the dirty bit to the physical page now the pte is gone. */
811 if (pte_dirty(pteval))
812 set_page_dirty(page);
813
365e9c87
HD
814 /* Update high watermark before we lower rss */
815 update_hiwater_rss(mm);
816
888b9f7c
AK
817 if (PageHWPoison(page) && !(flags & TTU_IGNORE_HWPOISON)) {
818 if (PageAnon(page))
819 dec_mm_counter(mm, anon_rss);
820 else
821 dec_mm_counter(mm, file_rss);
822 set_pte_at(mm, address, pte,
823 swp_entry_to_pte(make_hwpoison_entry(page)));
824 } else if (PageAnon(page)) {
4c21e2f2 825 swp_entry_t entry = { .val = page_private(page) };
0697212a
CL
826
827 if (PageSwapCache(page)) {
828 /*
829 * Store the swap location in the pte.
830 * See handle_pte_fault() ...
831 */
570a335b
HD
832 if (swap_duplicate(entry) < 0) {
833 set_pte_at(mm, address, pte, pteval);
834 ret = SWAP_FAIL;
835 goto out_unmap;
836 }
0697212a
CL
837 if (list_empty(&mm->mmlist)) {
838 spin_lock(&mmlist_lock);
839 if (list_empty(&mm->mmlist))
840 list_add(&mm->mmlist, &init_mm.mmlist);
841 spin_unlock(&mmlist_lock);
842 }
442c9137 843 dec_mm_counter(mm, anon_rss);
64cdd548 844 } else if (PAGE_MIGRATION) {
0697212a
CL
845 /*
846 * Store the pfn of the page in a special migration
847 * pte. do_swap_page() will wait until the migration
848 * pte is removed and then restart fault handling.
849 */
14fa31b8 850 BUG_ON(TTU_ACTION(flags) != TTU_MIGRATION);
0697212a 851 entry = make_migration_entry(page, pte_write(pteval));
1da177e4
LT
852 }
853 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
854 BUG_ON(pte_file(*pte));
14fa31b8 855 } else if (PAGE_MIGRATION && (TTU_ACTION(flags) == TTU_MIGRATION)) {
04e62a29
CL
856 /* Establish migration entry for a file page */
857 swp_entry_t entry;
858 entry = make_migration_entry(page, pte_write(pteval));
859 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
860 } else
4294621f 861 dec_mm_counter(mm, file_rss);
1da177e4 862
edc315fd 863 page_remove_rmap(page);
1da177e4
LT
864 page_cache_release(page);
865
866out_unmap:
c0718806 867 pte_unmap_unlock(pte, ptl);
53f79acb 868
af8e3354 869 if (ret == SWAP_MLOCK) {
53f79acb
HD
870 ret = SWAP_AGAIN;
871 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
872 if (vma->vm_flags & VM_LOCKED) {
873 mlock_vma_page(page);
874 ret = SWAP_MLOCK;
875 }
876 up_read(&vma->vm_mm->mmap_sem);
877 }
878 }
1da177e4
LT
879out:
880 return ret;
881}
882
883/*
884 * objrmap doesn't work for nonlinear VMAs because the assumption that
885 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
886 * Consequently, given a particular page and its ->index, we cannot locate the
887 * ptes which are mapping that page without an exhaustive linear search.
888 *
889 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
890 * maps the file to which the target page belongs. The ->vm_private_data field
891 * holds the current cursor into that scan. Successive searches will circulate
892 * around the vma's virtual address space.
893 *
894 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
895 * more scanning pressure is placed against them as well. Eventually pages
896 * will become fully unmapped and are eligible for eviction.
897 *
898 * For very sparsely populated VMAs this is a little inefficient - chances are
899 * there there won't be many ptes located within the scan cluster. In this case
900 * maybe we could scan further - to the end of the pte page, perhaps.
b291f000
NP
901 *
902 * Mlocked pages: check VM_LOCKED under mmap_sem held for read, if we can
903 * acquire it without blocking. If vma locked, mlock the pages in the cluster,
904 * rather than unmapping them. If we encounter the "check_page" that vmscan is
905 * trying to unmap, return SWAP_MLOCK, else default SWAP_AGAIN.
1da177e4
LT
906 */
907#define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
908#define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
909
b291f000
NP
910static int try_to_unmap_cluster(unsigned long cursor, unsigned int *mapcount,
911 struct vm_area_struct *vma, struct page *check_page)
1da177e4
LT
912{
913 struct mm_struct *mm = vma->vm_mm;
914 pgd_t *pgd;
915 pud_t *pud;
916 pmd_t *pmd;
c0718806 917 pte_t *pte;
1da177e4 918 pte_t pteval;
c0718806 919 spinlock_t *ptl;
1da177e4
LT
920 struct page *page;
921 unsigned long address;
922 unsigned long end;
b291f000
NP
923 int ret = SWAP_AGAIN;
924 int locked_vma = 0;
1da177e4 925
1da177e4
LT
926 address = (vma->vm_start + cursor) & CLUSTER_MASK;
927 end = address + CLUSTER_SIZE;
928 if (address < vma->vm_start)
929 address = vma->vm_start;
930 if (end > vma->vm_end)
931 end = vma->vm_end;
932
933 pgd = pgd_offset(mm, address);
934 if (!pgd_present(*pgd))
b291f000 935 return ret;
1da177e4
LT
936
937 pud = pud_offset(pgd, address);
938 if (!pud_present(*pud))
b291f000 939 return ret;
1da177e4
LT
940
941 pmd = pmd_offset(pud, address);
942 if (!pmd_present(*pmd))
b291f000
NP
943 return ret;
944
945 /*
af8e3354 946 * If we can acquire the mmap_sem for read, and vma is VM_LOCKED,
b291f000
NP
947 * keep the sem while scanning the cluster for mlocking pages.
948 */
af8e3354 949 if (down_read_trylock(&vma->vm_mm->mmap_sem)) {
b291f000
NP
950 locked_vma = (vma->vm_flags & VM_LOCKED);
951 if (!locked_vma)
952 up_read(&vma->vm_mm->mmap_sem); /* don't need it */
953 }
c0718806
HD
954
955 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
1da177e4 956
365e9c87
HD
957 /* Update high watermark before we lower rss */
958 update_hiwater_rss(mm);
959
c0718806 960 for (; address < end; pte++, address += PAGE_SIZE) {
1da177e4
LT
961 if (!pte_present(*pte))
962 continue;
6aab341e
LT
963 page = vm_normal_page(vma, address, *pte);
964 BUG_ON(!page || PageAnon(page));
1da177e4 965
b291f000
NP
966 if (locked_vma) {
967 mlock_vma_page(page); /* no-op if already mlocked */
968 if (page == check_page)
969 ret = SWAP_MLOCK;
970 continue; /* don't unmap */
971 }
972
cddb8a5c 973 if (ptep_clear_flush_young_notify(vma, address, pte))
1da177e4
LT
974 continue;
975
976 /* Nuke the page table entry. */
eca35133 977 flush_cache_page(vma, address, pte_pfn(*pte));
cddb8a5c 978 pteval = ptep_clear_flush_notify(vma, address, pte);
1da177e4
LT
979
980 /* If nonlinear, store the file page offset in the pte. */
981 if (page->index != linear_page_index(vma, address))
982 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
983
984 /* Move the dirty bit to the physical page now the pte is gone. */
985 if (pte_dirty(pteval))
986 set_page_dirty(page);
987
edc315fd 988 page_remove_rmap(page);
1da177e4 989 page_cache_release(page);
4294621f 990 dec_mm_counter(mm, file_rss);
1da177e4
LT
991 (*mapcount)--;
992 }
c0718806 993 pte_unmap_unlock(pte - 1, ptl);
b291f000
NP
994 if (locked_vma)
995 up_read(&vma->vm_mm->mmap_sem);
996 return ret;
1da177e4
LT
997}
998
b291f000
NP
999/**
1000 * try_to_unmap_anon - unmap or unlock anonymous page using the object-based
1001 * rmap method
1002 * @page: the page to unmap/unlock
8051be5e 1003 * @flags: action and flags
b291f000
NP
1004 *
1005 * Find all the mappings of a page using the mapping pointer and the vma chains
1006 * contained in the anon_vma struct it points to.
1007 *
1008 * This function is only called from try_to_unmap/try_to_munlock for
1009 * anonymous pages.
1010 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1011 * where the page was found will be held for write. So, we won't recheck
1012 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1013 * 'LOCKED.
1014 */
14fa31b8 1015static int try_to_unmap_anon(struct page *page, enum ttu_flags flags)
1da177e4
LT
1016{
1017 struct anon_vma *anon_vma;
1018 struct vm_area_struct *vma;
1019 int ret = SWAP_AGAIN;
b291f000 1020
1da177e4
LT
1021 anon_vma = page_lock_anon_vma(page);
1022 if (!anon_vma)
1023 return ret;
1024
1025 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
1cb1729b
HD
1026 unsigned long address = vma_address(page, vma);
1027 if (address == -EFAULT)
1028 continue;
1029 ret = try_to_unmap_one(page, vma, address, flags);
53f79acb
HD
1030 if (ret != SWAP_AGAIN || !page_mapped(page))
1031 break;
1da177e4 1032 }
34bbd704
ON
1033
1034 page_unlock_anon_vma(anon_vma);
1da177e4
LT
1035 return ret;
1036}
1037
1038/**
b291f000
NP
1039 * try_to_unmap_file - unmap/unlock file page using the object-based rmap method
1040 * @page: the page to unmap/unlock
14fa31b8 1041 * @flags: action and flags
1da177e4
LT
1042 *
1043 * Find all the mappings of a page using the mapping pointer and the vma chains
1044 * contained in the address_space struct it points to.
1045 *
b291f000
NP
1046 * This function is only called from try_to_unmap/try_to_munlock for
1047 * object-based pages.
1048 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1049 * where the page was found will be held for write. So, we won't recheck
1050 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1051 * 'LOCKED.
1da177e4 1052 */
14fa31b8 1053static int try_to_unmap_file(struct page *page, enum ttu_flags flags)
1da177e4
LT
1054{
1055 struct address_space *mapping = page->mapping;
1056 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1057 struct vm_area_struct *vma;
1058 struct prio_tree_iter iter;
1059 int ret = SWAP_AGAIN;
1060 unsigned long cursor;
1061 unsigned long max_nl_cursor = 0;
1062 unsigned long max_nl_size = 0;
1063 unsigned int mapcount;
1064
1065 spin_lock(&mapping->i_mmap_lock);
1066 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
1cb1729b
HD
1067 unsigned long address = vma_address(page, vma);
1068 if (address == -EFAULT)
1069 continue;
1070 ret = try_to_unmap_one(page, vma, address, flags);
53f79acb
HD
1071 if (ret != SWAP_AGAIN || !page_mapped(page))
1072 goto out;
1da177e4
LT
1073 }
1074
1075 if (list_empty(&mapping->i_mmap_nonlinear))
1076 goto out;
1077
53f79acb
HD
1078 /*
1079 * We don't bother to try to find the munlocked page in nonlinears.
1080 * It's costly. Instead, later, page reclaim logic may call
1081 * try_to_unmap(TTU_MUNLOCK) and recover PG_mlocked lazily.
1082 */
1083 if (TTU_ACTION(flags) == TTU_MUNLOCK)
1084 goto out;
1085
1da177e4
LT
1086 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1087 shared.vm_set.list) {
1da177e4
LT
1088 cursor = (unsigned long) vma->vm_private_data;
1089 if (cursor > max_nl_cursor)
1090 max_nl_cursor = cursor;
1091 cursor = vma->vm_end - vma->vm_start;
1092 if (cursor > max_nl_size)
1093 max_nl_size = cursor;
1094 }
1095
b291f000 1096 if (max_nl_size == 0) { /* all nonlinears locked or reserved ? */
1da177e4
LT
1097 ret = SWAP_FAIL;
1098 goto out;
1099 }
1100
1101 /*
1102 * We don't try to search for this page in the nonlinear vmas,
1103 * and page_referenced wouldn't have found it anyway. Instead
1104 * just walk the nonlinear vmas trying to age and unmap some.
1105 * The mapcount of the page we came in with is irrelevant,
1106 * but even so use it as a guide to how hard we should try?
1107 */
1108 mapcount = page_mapcount(page);
1109 if (!mapcount)
1110 goto out;
1111 cond_resched_lock(&mapping->i_mmap_lock);
1112
1113 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
1114 if (max_nl_cursor == 0)
1115 max_nl_cursor = CLUSTER_SIZE;
1116
1117 do {
1118 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
1119 shared.vm_set.list) {
1da177e4 1120 cursor = (unsigned long) vma->vm_private_data;
839b9685 1121 while ( cursor < max_nl_cursor &&
1da177e4 1122 cursor < vma->vm_end - vma->vm_start) {
53f79acb
HD
1123 if (try_to_unmap_cluster(cursor, &mapcount,
1124 vma, page) == SWAP_MLOCK)
1125 ret = SWAP_MLOCK;
1da177e4
LT
1126 cursor += CLUSTER_SIZE;
1127 vma->vm_private_data = (void *) cursor;
1128 if ((int)mapcount <= 0)
1129 goto out;
1130 }
1131 vma->vm_private_data = (void *) max_nl_cursor;
1132 }
1133 cond_resched_lock(&mapping->i_mmap_lock);
1134 max_nl_cursor += CLUSTER_SIZE;
1135 } while (max_nl_cursor <= max_nl_size);
1136
1137 /*
1138 * Don't loop forever (perhaps all the remaining pages are
1139 * in locked vmas). Reset cursor on all unreserved nonlinear
1140 * vmas, now forgetting on which ones it had fallen behind.
1141 */
101d2be7
HD
1142 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1143 vma->vm_private_data = NULL;
1da177e4
LT
1144out:
1145 spin_unlock(&mapping->i_mmap_lock);
1146 return ret;
1147}
1148
1149/**
1150 * try_to_unmap - try to remove all page table mappings to a page
1151 * @page: the page to get unmapped
14fa31b8 1152 * @flags: action and flags
1da177e4
LT
1153 *
1154 * Tries to remove all the page table entries which are mapping this
1155 * page, used in the pageout path. Caller must hold the page lock.
1156 * Return values are:
1157 *
1158 * SWAP_SUCCESS - we succeeded in removing all mappings
1159 * SWAP_AGAIN - we missed a mapping, try again later
1160 * SWAP_FAIL - the page is unswappable
b291f000 1161 * SWAP_MLOCK - page is mlocked.
1da177e4 1162 */
14fa31b8 1163int try_to_unmap(struct page *page, enum ttu_flags flags)
1da177e4
LT
1164{
1165 int ret;
1166
1da177e4
LT
1167 BUG_ON(!PageLocked(page));
1168
5ad64688
HD
1169 if (unlikely(PageKsm(page)))
1170 ret = try_to_unmap_ksm(page, flags);
1171 else if (PageAnon(page))
14fa31b8 1172 ret = try_to_unmap_anon(page, flags);
1da177e4 1173 else
14fa31b8 1174 ret = try_to_unmap_file(page, flags);
b291f000 1175 if (ret != SWAP_MLOCK && !page_mapped(page))
1da177e4
LT
1176 ret = SWAP_SUCCESS;
1177 return ret;
1178}
81b4082d 1179
b291f000
NP
1180/**
1181 * try_to_munlock - try to munlock a page
1182 * @page: the page to be munlocked
1183 *
1184 * Called from munlock code. Checks all of the VMAs mapping the page
1185 * to make sure nobody else has this page mlocked. The page will be
1186 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1187 *
1188 * Return values are:
1189 *
53f79acb 1190 * SWAP_AGAIN - no vma is holding page mlocked, or,
b291f000 1191 * SWAP_AGAIN - page mapped in mlocked vma -- couldn't acquire mmap sem
5ad64688 1192 * SWAP_FAIL - page cannot be located at present
b291f000
NP
1193 * SWAP_MLOCK - page is now mlocked.
1194 */
1195int try_to_munlock(struct page *page)
1196{
1197 VM_BUG_ON(!PageLocked(page) || PageLRU(page));
1198
5ad64688
HD
1199 if (unlikely(PageKsm(page)))
1200 return try_to_unmap_ksm(page, TTU_MUNLOCK);
1201 else if (PageAnon(page))
14fa31b8 1202 return try_to_unmap_anon(page, TTU_MUNLOCK);
b291f000 1203 else
14fa31b8 1204 return try_to_unmap_file(page, TTU_MUNLOCK);
b291f000 1205}
e9995ef9
HD
1206
1207#ifdef CONFIG_MIGRATION
1208/*
1209 * rmap_walk() and its helpers rmap_walk_anon() and rmap_walk_file():
1210 * Called by migrate.c to remove migration ptes, but might be used more later.
1211 */
1212static int rmap_walk_anon(struct page *page, int (*rmap_one)(struct page *,
1213 struct vm_area_struct *, unsigned long, void *), void *arg)
1214{
1215 struct anon_vma *anon_vma;
1216 struct vm_area_struct *vma;
1217 int ret = SWAP_AGAIN;
1218
1219 /*
1220 * Note: remove_migration_ptes() cannot use page_lock_anon_vma()
1221 * because that depends on page_mapped(); but not all its usages
1222 * are holding mmap_sem, which also gave the necessary guarantee
1223 * (that this anon_vma's slab has not already been destroyed).
1224 * This needs to be reviewed later: avoiding page_lock_anon_vma()
1225 * is risky, and currently limits the usefulness of rmap_walk().
1226 */
1227 anon_vma = page_anon_vma(page);
1228 if (!anon_vma)
1229 return ret;
1230 spin_lock(&anon_vma->lock);
1231 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
1232 unsigned long address = vma_address(page, vma);
1233 if (address == -EFAULT)
1234 continue;
1235 ret = rmap_one(page, vma, address, arg);
1236 if (ret != SWAP_AGAIN)
1237 break;
1238 }
1239 spin_unlock(&anon_vma->lock);
1240 return ret;
1241}
1242
1243static int rmap_walk_file(struct page *page, int (*rmap_one)(struct page *,
1244 struct vm_area_struct *, unsigned long, void *), void *arg)
1245{
1246 struct address_space *mapping = page->mapping;
1247 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
1248 struct vm_area_struct *vma;
1249 struct prio_tree_iter iter;
1250 int ret = SWAP_AGAIN;
1251
1252 if (!mapping)
1253 return ret;
1254 spin_lock(&mapping->i_mmap_lock);
1255 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
1256 unsigned long address = vma_address(page, vma);
1257 if (address == -EFAULT)
1258 continue;
1259 ret = rmap_one(page, vma, address, arg);
1260 if (ret != SWAP_AGAIN)
1261 break;
1262 }
1263 /*
1264 * No nonlinear handling: being always shared, nonlinear vmas
1265 * never contain migration ptes. Decide what to do about this
1266 * limitation to linear when we need rmap_walk() on nonlinear.
1267 */
1268 spin_unlock(&mapping->i_mmap_lock);
1269 return ret;
1270}
1271
1272int rmap_walk(struct page *page, int (*rmap_one)(struct page *,
1273 struct vm_area_struct *, unsigned long, void *), void *arg)
1274{
1275 VM_BUG_ON(!PageLocked(page));
1276
1277 if (unlikely(PageKsm(page)))
1278 return rmap_walk_ksm(page, rmap_one, arg);
1279 else if (PageAnon(page))
1280 return rmap_walk_anon(page, rmap_one, arg);
1281 else
1282 return rmap_walk_file(page, rmap_one, arg);
1283}
1284#endif /* CONFIG_MIGRATION */