blob: a0e92a263d12eb734e8a8171e9d575e35dc0d768 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
17 * Contributions by Hugh Dickins <hugh@veritas.com> 2003, 2004
18 */
19
20/*
21 * Lock ordering in mm:
22 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -080023 * inode->i_mutex (while writing or truncating, not reading or faulting)
Nick Piggin82591e62006-10-19 23:29:10 -070024 * 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)
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 */
40
41#include <linux/mm.h>
42#include <linux/pagemap.h>
43#include <linux/swap.h>
44#include <linux/swapops.h>
45#include <linux/slab.h>
46#include <linux/init.h>
47#include <linux/rmap.h>
48#include <linux/rcupdate.h>
Christoph Lametera48d07a2006-02-01 03:05:38 -080049#include <linux/module.h>
Nick Piggin7de6b802006-12-22 01:09:33 -080050#include <linux/kallsyms.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080051#include <linux/memcontrol.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53#include <asm/tlbflush.h>
54
Pekka Enbergfcc234f2006-03-22 00:08:13 -080055struct kmem_cache *anon_vma_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Linus Torvalds1da177e2005-04-16 15:20:36 -070057/* This must be called under the mmap_sem. */
58int anon_vma_prepare(struct vm_area_struct *vma)
59{
60 struct anon_vma *anon_vma = vma->anon_vma;
61
62 might_sleep();
63 if (unlikely(!anon_vma)) {
64 struct mm_struct *mm = vma->vm_mm;
65 struct anon_vma *allocated, *locked;
66
67 anon_vma = find_mergeable_anon_vma(vma);
68 if (anon_vma) {
69 allocated = NULL;
70 locked = anon_vma;
71 spin_lock(&locked->lock);
72 } else {
73 anon_vma = anon_vma_alloc();
74 if (unlikely(!anon_vma))
75 return -ENOMEM;
76 allocated = anon_vma;
77 locked = NULL;
78 }
79
80 /* page_table_lock to protect against threads */
81 spin_lock(&mm->page_table_lock);
82 if (likely(!vma->anon_vma)) {
83 vma->anon_vma = anon_vma;
Christoph Lameter06972122006-06-23 02:03:35 -070084 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 allocated = NULL;
86 }
87 spin_unlock(&mm->page_table_lock);
88
89 if (locked)
90 spin_unlock(&locked->lock);
91 if (unlikely(allocated))
92 anon_vma_free(allocated);
93 }
94 return 0;
95}
96
97void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next)
98{
99 BUG_ON(vma->anon_vma != next->anon_vma);
100 list_del(&next->anon_vma_node);
101}
102
103void __anon_vma_link(struct vm_area_struct *vma)
104{
105 struct anon_vma *anon_vma = vma->anon_vma;
106
Hugh Dickins30acbab2007-06-27 14:09:53 -0700107 if (anon_vma)
Christoph Lameter06972122006-06-23 02:03:35 -0700108 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109}
110
111void anon_vma_link(struct vm_area_struct *vma)
112{
113 struct anon_vma *anon_vma = vma->anon_vma;
114
115 if (anon_vma) {
116 spin_lock(&anon_vma->lock);
Christoph Lameter06972122006-06-23 02:03:35 -0700117 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 spin_unlock(&anon_vma->lock);
119 }
120}
121
122void anon_vma_unlink(struct vm_area_struct *vma)
123{
124 struct anon_vma *anon_vma = vma->anon_vma;
125 int empty;
126
127 if (!anon_vma)
128 return;
129
130 spin_lock(&anon_vma->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 list_del(&vma->anon_vma_node);
132
133 /* We must garbage collect the anon_vma if it's empty */
134 empty = list_empty(&anon_vma->head);
135 spin_unlock(&anon_vma->lock);
136
137 if (empty)
138 anon_vma_free(anon_vma);
139}
140
Christoph Lameter4ba9b9d2007-10-16 23:25:51 -0700141static void anon_vma_ctor(struct kmem_cache *cachep, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142{
Christoph Lametera35afb82007-05-16 22:10:57 -0700143 struct anon_vma *anon_vma = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
Christoph Lametera35afb82007-05-16 22:10:57 -0700145 spin_lock_init(&anon_vma->lock);
146 INIT_LIST_HEAD(&anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
148
149void __init anon_vma_init(void)
150{
151 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
Paul Mundt20c2df82007-07-20 10:11:58 +0900152 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153}
154
155/*
156 * Getting a lock on a stable anon_vma from a page off the LRU is
157 * tricky: page_lock_anon_vma rely on RCU to guard against the races.
158 */
159static struct anon_vma *page_lock_anon_vma(struct page *page)
160{
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800161 struct anon_vma *anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 unsigned long anon_mapping;
163
164 rcu_read_lock();
165 anon_mapping = (unsigned long) page->mapping;
166 if (!(anon_mapping & PAGE_MAPPING_ANON))
167 goto out;
168 if (!page_mapped(page))
169 goto out;
170
171 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
172 spin_lock(&anon_vma->lock);
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800173 return anon_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174out:
175 rcu_read_unlock();
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800176 return NULL;
177}
178
179static void page_unlock_anon_vma(struct anon_vma *anon_vma)
180{
181 spin_unlock(&anon_vma->lock);
182 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
184
185/*
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800186 * At what user virtual address is page expected in @vma?
187 * Returns virtual address or -EFAULT if page's index/offset is not
188 * within the range mapped the @vma.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 */
190static inline unsigned long
191vma_address(struct page *page, struct vm_area_struct *vma)
192{
193 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
194 unsigned long address;
195
196 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
197 if (unlikely(address < vma->vm_start || address >= vma->vm_end)) {
Lee Schermerhorn3ad33b22007-11-14 16:59:10 -0800198 /* page should be within @vma mapping range */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 return -EFAULT;
200 }
201 return address;
202}
203
204/*
205 * At what user virtual address is page expected in vma? checking that the
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800206 * page matches the vma: currently only used on anon pages, by unuse_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 */
208unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
209{
210 if (PageAnon(page)) {
211 if ((void *)vma->anon_vma !=
212 (void *)page->mapping - PAGE_MAPPING_ANON)
213 return -EFAULT;
214 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800215 if (!vma->vm_file ||
216 vma->vm_file->f_mapping != page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 return -EFAULT;
218 } else
219 return -EFAULT;
220 return vma_address(page, vma);
221}
222
223/*
Nikita Danilov81b40822005-05-01 08:58:36 -0700224 * Check that @page is mapped at @address into @mm.
225 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700226 * On success returns with pte mapped and locked.
Nikita Danilov81b40822005-05-01 08:58:36 -0700227 */
Carsten Otteceffc072005-06-23 22:05:25 -0700228pte_t *page_check_address(struct page *page, struct mm_struct *mm,
Hugh Dickinsc0718802005-10-29 18:16:31 -0700229 unsigned long address, spinlock_t **ptlp)
Nikita Danilov81b40822005-05-01 08:58:36 -0700230{
231 pgd_t *pgd;
232 pud_t *pud;
233 pmd_t *pmd;
234 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700235 spinlock_t *ptl;
Nikita Danilov81b40822005-05-01 08:58:36 -0700236
Nikita Danilov81b40822005-05-01 08:58:36 -0700237 pgd = pgd_offset(mm, address);
Hugh Dickinsc0718802005-10-29 18:16:31 -0700238 if (!pgd_present(*pgd))
239 return NULL;
240
241 pud = pud_offset(pgd, address);
242 if (!pud_present(*pud))
243 return NULL;
244
245 pmd = pmd_offset(pud, address);
246 if (!pmd_present(*pmd))
247 return NULL;
248
249 pte = pte_offset_map(pmd, address);
250 /* Make a quick check before getting the lock */
251 if (!pte_present(*pte)) {
252 pte_unmap(pte);
253 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700254 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700255
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700256 ptl = pte_lockptr(mm, pmd);
Hugh Dickinsc0718802005-10-29 18:16:31 -0700257 spin_lock(ptl);
258 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
259 *ptlp = ptl;
260 return pte;
261 }
262 pte_unmap_unlock(pte, ptl);
263 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700264}
265
266/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 * Subfunctions of page_referenced: page_referenced_one called
268 * repeatedly from either page_referenced_anon or page_referenced_file.
269 */
270static int page_referenced_one(struct page *page,
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800271 struct vm_area_struct *vma, unsigned int *mapcount)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272{
273 struct mm_struct *mm = vma->vm_mm;
274 unsigned long address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700276 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 int referenced = 0;
278
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 address = vma_address(page, vma);
280 if (address == -EFAULT)
281 goto out;
282
Hugh Dickinsc0718802005-10-29 18:16:31 -0700283 pte = page_check_address(page, mm, address, &ptl);
284 if (!pte)
285 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
Hugh Dickins5a9bbdc2008-02-04 22:29:23 -0800287 if (vma->vm_flags & VM_LOCKED) {
288 referenced++;
289 *mapcount = 1; /* break early from loop */
290 } else if (ptep_clear_flush_young(vma, address, pte))
Hugh Dickinsc0718802005-10-29 18:16:31 -0700291 referenced++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
Hugh Dickinsc0718802005-10-29 18:16:31 -0700293 /* Pretend the page is referenced if the task has the
294 swap token and is in the middle of a page fault. */
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800295 if (mm != current->mm && has_swap_token(mm) &&
Hugh Dickinsc0718802005-10-29 18:16:31 -0700296 rwsem_is_locked(&mm->mmap_sem))
297 referenced++;
298
299 (*mapcount)--;
300 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301out:
302 return referenced;
303}
304
Balbir Singhbed71612008-02-07 00:14:01 -0800305static int page_referenced_anon(struct page *page,
306 struct mem_cgroup *mem_cont)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
308 unsigned int mapcount;
309 struct anon_vma *anon_vma;
310 struct vm_area_struct *vma;
311 int referenced = 0;
312
313 anon_vma = page_lock_anon_vma(page);
314 if (!anon_vma)
315 return referenced;
316
317 mapcount = page_mapcount(page);
318 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
Balbir Singhbed71612008-02-07 00:14:01 -0800319 /*
320 * If we are reclaiming on behalf of a cgroup, skip
321 * counting on behalf of references from different
322 * cgroups
323 */
324 if (mem_cont && (mm_cgroup(vma->vm_mm) != mem_cont))
325 continue;
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800326 referenced += page_referenced_one(page, vma, &mapcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 if (!mapcount)
328 break;
329 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800330
331 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 return referenced;
333}
334
335/**
336 * page_referenced_file - referenced check for object-based rmap
337 * @page: the page we're checking references on.
338 *
339 * For an object-based mapped page, find all the places it is mapped and
340 * check/clear the referenced flag. This is done by following the page->mapping
341 * pointer, then walking the chain of vmas it holds. It returns the number
342 * of references it found.
343 *
344 * This function is only called from page_referenced for object-based pages.
345 */
Balbir Singhbed71612008-02-07 00:14:01 -0800346static int page_referenced_file(struct page *page,
347 struct mem_cgroup *mem_cont)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
349 unsigned int mapcount;
350 struct address_space *mapping = page->mapping;
351 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
352 struct vm_area_struct *vma;
353 struct prio_tree_iter iter;
354 int referenced = 0;
355
356 /*
357 * The caller's checks on page->mapping and !PageAnon have made
358 * sure that this is a file page: the check for page->mapping
359 * excludes the case just before it gets set on an anon page.
360 */
361 BUG_ON(PageAnon(page));
362
363 /*
364 * The page lock not only makes sure that page->mapping cannot
365 * suddenly be NULLified by truncation, it makes sure that the
366 * structure at mapping cannot be freed and reused yet,
367 * so we can safely take mapping->i_mmap_lock.
368 */
369 BUG_ON(!PageLocked(page));
370
371 spin_lock(&mapping->i_mmap_lock);
372
373 /*
374 * i_mmap_lock does not stabilize mapcount at all, but mapcount
375 * is more likely to be accurate if we note it after spinning.
376 */
377 mapcount = page_mapcount(page);
378
379 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Balbir Singhbed71612008-02-07 00:14:01 -0800380 /*
381 * If we are reclaiming on behalf of a cgroup, skip
382 * counting on behalf of references from different
383 * cgroups
384 */
385 if (mem_cont && (mm_cgroup(vma->vm_mm) != mem_cont))
386 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 if ((vma->vm_flags & (VM_LOCKED|VM_MAYSHARE))
388 == (VM_LOCKED|VM_MAYSHARE)) {
389 referenced++;
390 break;
391 }
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800392 referenced += page_referenced_one(page, vma, &mapcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 if (!mapcount)
394 break;
395 }
396
397 spin_unlock(&mapping->i_mmap_lock);
398 return referenced;
399}
400
401/**
402 * page_referenced - test if the page was referenced
403 * @page: the page to test
404 * @is_locked: caller holds lock on the page
405 *
406 * Quick test_and_clear_referenced for all mappings to a page,
407 * returns the number of ptes which referenced the page.
408 */
Balbir Singhbed71612008-02-07 00:14:01 -0800409int page_referenced(struct page *page, int is_locked,
410 struct mem_cgroup *mem_cont)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
412 int referenced = 0;
413
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 if (page_test_and_clear_young(page))
415 referenced++;
416
417 if (TestClearPageReferenced(page))
418 referenced++;
419
420 if (page_mapped(page) && page->mapping) {
421 if (PageAnon(page))
Balbir Singhbed71612008-02-07 00:14:01 -0800422 referenced += page_referenced_anon(page, mem_cont);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 else if (is_locked)
Balbir Singhbed71612008-02-07 00:14:01 -0800424 referenced += page_referenced_file(page, mem_cont);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 else if (TestSetPageLocked(page))
426 referenced++;
427 else {
428 if (page->mapping)
Balbir Singhbed71612008-02-07 00:14:01 -0800429 referenced +=
430 page_referenced_file(page, mem_cont);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 unlock_page(page);
432 }
433 }
434 return referenced;
435}
436
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700437static int page_mkclean_one(struct page *page, struct vm_area_struct *vma)
438{
439 struct mm_struct *mm = vma->vm_mm;
440 unsigned long address;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100441 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700442 spinlock_t *ptl;
443 int ret = 0;
444
445 address = vma_address(page, vma);
446 if (address == -EFAULT)
447 goto out;
448
449 pte = page_check_address(page, mm, address, &ptl);
450 if (!pte)
451 goto out;
452
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100453 if (pte_dirty(*pte) || pte_write(*pte)) {
454 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700455
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100456 flush_cache_page(vma, address, pte_pfn(*pte));
457 entry = ptep_clear_flush(vma, address, pte);
458 entry = pte_wrprotect(entry);
459 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800460 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100461 ret = 1;
462 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700463
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700464 pte_unmap_unlock(pte, ptl);
465out:
466 return ret;
467}
468
469static int page_mkclean_file(struct address_space *mapping, struct page *page)
470{
471 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
472 struct vm_area_struct *vma;
473 struct prio_tree_iter iter;
474 int ret = 0;
475
476 BUG_ON(PageAnon(page));
477
478 spin_lock(&mapping->i_mmap_lock);
479 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
480 if (vma->vm_flags & VM_SHARED)
481 ret += page_mkclean_one(page, vma);
482 }
483 spin_unlock(&mapping->i_mmap_lock);
484 return ret;
485}
486
487int page_mkclean(struct page *page)
488{
489 int ret = 0;
490
491 BUG_ON(!PageLocked(page));
492
493 if (page_mapped(page)) {
494 struct address_space *mapping = page_mapping(page);
Christian Borntraegerce7e9fa2007-11-20 11:13:36 +0100495 if (mapping) {
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700496 ret = page_mkclean_file(mapping, page);
Christian Borntraegerce7e9fa2007-11-20 11:13:36 +0100497 if (page_test_dirty(page)) {
498 page_clear_dirty(page);
499 ret = 1;
500 }
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200501 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700502 }
503
504 return ret;
505}
Jaya Kumar60b59be2007-05-08 00:37:37 -0700506EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700507
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508/**
Nick Piggin9617d952006-01-06 00:11:12 -0800509 * page_set_anon_rmap - setup new anonymous rmap
510 * @page: the page to add the mapping to
511 * @vma: the vm area in which the mapping is added
512 * @address: the user virtual address mapped
513 */
514static void __page_set_anon_rmap(struct page *page,
515 struct vm_area_struct *vma, unsigned long address)
516{
517 struct anon_vma *anon_vma = vma->anon_vma;
518
519 BUG_ON(!anon_vma);
520 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
521 page->mapping = (struct address_space *) anon_vma;
522
523 page->index = linear_page_index(vma, address);
524
Nick Piggina74609f2006-01-06 00:11:20 -0800525 /*
526 * nr_mapped state can be updated without turning off
527 * interrupts because it is not modified via interrupt.
528 */
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700529 __inc_zone_page_state(page, NR_ANON_PAGES);
Nick Piggin9617d952006-01-06 00:11:12 -0800530}
531
532/**
Nick Pigginc97a9e12007-05-16 22:11:21 -0700533 * page_set_anon_rmap - sanity check anonymous rmap addition
534 * @page: the page to add the mapping to
535 * @vma: the vm area in which the mapping is added
536 * @address: the user virtual address mapped
537 */
538static void __page_check_anon_rmap(struct page *page,
539 struct vm_area_struct *vma, unsigned long address)
540{
541#ifdef CONFIG_DEBUG_VM
542 /*
543 * The page's anon-rmap details (mapping and index) are guaranteed to
544 * be set up correctly at this point.
545 *
546 * We have exclusion against page_add_anon_rmap because the caller
547 * always holds the page locked, except if called from page_dup_rmap,
548 * in which case the page is already known to be setup.
549 *
550 * We have exclusion against page_add_new_anon_rmap because those pages
551 * are initially only visible via the pagetables, and the pte is locked
552 * over the call to page_add_new_anon_rmap.
553 */
554 struct anon_vma *anon_vma = vma->anon_vma;
555 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
556 BUG_ON(page->mapping != (struct address_space *)anon_vma);
557 BUG_ON(page->index != linear_page_index(vma, address));
558#endif
559}
560
561/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 * page_add_anon_rmap - add pte mapping to an anonymous page
563 * @page: the page to add the mapping to
564 * @vma: the vm area in which the mapping is added
565 * @address: the user virtual address mapped
566 *
Nick Pigginc97a9e12007-05-16 22:11:21 -0700567 * The caller needs to hold the pte lock and the page must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 */
569void page_add_anon_rmap(struct page *page,
570 struct vm_area_struct *vma, unsigned long address)
571{
Nick Pigginc97a9e12007-05-16 22:11:21 -0700572 VM_BUG_ON(!PageLocked(page));
573 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Nick Piggin9617d952006-01-06 00:11:12 -0800574 if (atomic_inc_and_test(&page->_mapcount))
575 __page_set_anon_rmap(page, vma, address);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800576 else {
Nick Pigginc97a9e12007-05-16 22:11:21 -0700577 __page_check_anon_rmap(page, vma, address);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800578 /*
579 * We unconditionally charged during prepare, we uncharge here
580 * This takes care of balancing the reference counts
581 */
582 mem_cgroup_uncharge_page(page);
583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584}
585
Nick Piggin9617d952006-01-06 00:11:12 -0800586/*
587 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
588 * @page: the page to add the mapping to
589 * @vma: the vm area in which the mapping is added
590 * @address: the user virtual address mapped
591 *
592 * Same as page_add_anon_rmap but must only be called on *new* pages.
593 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -0700594 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -0800595 */
596void page_add_new_anon_rmap(struct page *page,
597 struct vm_area_struct *vma, unsigned long address)
598{
Nick Pigginc97a9e12007-05-16 22:11:21 -0700599 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Nick Piggin9617d952006-01-06 00:11:12 -0800600 atomic_set(&page->_mapcount, 0); /* elevate count by 1 (starts at -1) */
601 __page_set_anon_rmap(page, vma, address);
602}
603
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604/**
605 * page_add_file_rmap - add pte mapping to a file page
606 * @page: the page to add the mapping to
607 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700608 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 */
610void page_add_file_rmap(struct page *page)
611{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 if (atomic_inc_and_test(&page->_mapcount))
Christoph Lameter65ba55f2006-06-30 01:55:34 -0700613 __inc_zone_page_state(page, NR_FILE_MAPPED);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800614 else
615 /*
616 * We unconditionally charged during prepare, we uncharge here
617 * This takes care of balancing the reference counts
618 */
619 mem_cgroup_uncharge_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
621
Nick Pigginc97a9e12007-05-16 22:11:21 -0700622#ifdef CONFIG_DEBUG_VM
623/**
624 * page_dup_rmap - duplicate pte mapping to a page
625 * @page: the page to add the mapping to
626 *
627 * For copy_page_range only: minimal extract from page_add_file_rmap /
628 * page_add_anon_rmap, avoiding unnecessary tests (already checked) so it's
629 * quicker.
630 *
631 * The caller needs to hold the pte lock.
632 */
633void page_dup_rmap(struct page *page, struct vm_area_struct *vma, unsigned long address)
634{
635 BUG_ON(page_mapcount(page) == 0);
636 if (PageAnon(page))
637 __page_check_anon_rmap(page, vma, address);
638 atomic_inc(&page->_mapcount);
639}
640#endif
641
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642/**
643 * page_remove_rmap - take down pte mapping from a page
644 * @page: page to remove mapping from
645 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700646 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 */
Nick Piggin7de6b802006-12-22 01:09:33 -0800648void page_remove_rmap(struct page *page, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 if (atomic_add_negative(-1, &page->_mapcount)) {
Nick Pigginb7ab7952006-03-22 00:08:42 -0800651 if (unlikely(page_mapcount(page) < 0)) {
Dave Jonesef2bf0d2006-01-08 01:01:00 -0800652 printk (KERN_EMERG "Eeek! page_mapcount(page) went negative! (%d)\n", page_mapcount(page));
Nick Piggin7de6b802006-12-22 01:09:33 -0800653 printk (KERN_EMERG " page pfn = %lx\n", page_to_pfn(page));
Dave Jonesef2bf0d2006-01-08 01:01:00 -0800654 printk (KERN_EMERG " page->flags = %lx\n", page->flags);
655 printk (KERN_EMERG " page->count = %x\n", page_count(page));
656 printk (KERN_EMERG " page->mapping = %p\n", page->mapping);
Nick Piggin7de6b802006-12-22 01:09:33 -0800657 print_symbol (KERN_EMERG " vma->vm_ops = %s\n", (unsigned long)vma->vm_ops);
Nick Piggin54cb8822007-07-19 01:46:59 -0700658 if (vma->vm_ops) {
Nick Piggin7de6b802006-12-22 01:09:33 -0800659 print_symbol (KERN_EMERG " vma->vm_ops->nopage = %s\n", (unsigned long)vma->vm_ops->nopage);
Nick Piggin54cb8822007-07-19 01:46:59 -0700660 print_symbol (KERN_EMERG " vma->vm_ops->fault = %s\n", (unsigned long)vma->vm_ops->fault);
661 }
Nick Piggin7de6b802006-12-22 01:09:33 -0800662 if (vma->vm_file && vma->vm_file->f_op)
663 print_symbol (KERN_EMERG " vma->vm_file->f_op->mmap = %s\n", (unsigned long)vma->vm_file->f_op->mmap);
Dave Jonesb16bc642006-10-11 01:21:27 -0700664 BUG();
Dave Jonesef2bf0d2006-01-08 01:01:00 -0800665 }
Dave Jonesb16bc642006-10-11 01:21:27 -0700666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 /*
668 * It would be tidy to reset the PageAnon mapping here,
669 * but that might overwrite a racing page_add_anon_rmap
670 * which increments mapcount after us but sets mapping
671 * before us: so leave the reset to free_hot_cold_page,
672 * and remember that it's only reliable while mapped.
673 * Leaving it set also helps swapoff to reinstate ptes
674 * faster for those pages still in swapcache.
675 */
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200676 if (page_test_dirty(page)) {
677 page_clear_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 set_page_dirty(page);
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200679 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800680 mem_cgroup_uncharge_page(page);
681
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700682 __dec_zone_page_state(page,
683 PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 }
685}
686
687/*
688 * Subfunctions of try_to_unmap: try_to_unmap_one called
689 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
690 */
Christoph Lametera48d07a2006-02-01 03:05:38 -0800691static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
Christoph Lameter73523492006-06-23 02:03:27 -0700692 int migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693{
694 struct mm_struct *mm = vma->vm_mm;
695 unsigned long address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 pte_t *pte;
697 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700698 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 int ret = SWAP_AGAIN;
700
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 address = vma_address(page, vma);
702 if (address == -EFAULT)
703 goto out;
704
Hugh Dickinsc0718802005-10-29 18:16:31 -0700705 pte = page_check_address(page, mm, address, &ptl);
706 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -0700707 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708
709 /*
710 * If the page is mlock()d, we cannot swap it out.
711 * If it's recently referenced (perhaps page_referenced
712 * skipped over this mm) then we should reactivate it.
713 */
Christoph Lametere6a15302006-06-25 05:46:49 -0700714 if (!migration && ((vma->vm_flags & VM_LOCKED) ||
715 (ptep_clear_flush_young(vma, address, pte)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 ret = SWAP_FAIL;
717 goto out_unmap;
718 }
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 /* Nuke the page table entry. */
721 flush_cache_page(vma, address, page_to_pfn(page));
722 pteval = ptep_clear_flush(vma, address, pte);
723
724 /* Move the dirty bit to the physical page now the pte is gone. */
725 if (pte_dirty(pteval))
726 set_page_dirty(page);
727
Hugh Dickins365e9c872005-10-29 18:16:18 -0700728 /* Update high watermark before we lower rss */
729 update_hiwater_rss(mm);
730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700732 swp_entry_t entry = { .val = page_private(page) };
Christoph Lameter06972122006-06-23 02:03:35 -0700733
734 if (PageSwapCache(page)) {
735 /*
736 * Store the swap location in the pte.
737 * See handle_pte_fault() ...
738 */
739 swap_duplicate(entry);
740 if (list_empty(&mm->mmlist)) {
741 spin_lock(&mmlist_lock);
742 if (list_empty(&mm->mmlist))
743 list_add(&mm->mmlist, &init_mm.mmlist);
744 spin_unlock(&mmlist_lock);
745 }
Christoph Lameter442c9132006-06-23 02:03:38 -0700746 dec_mm_counter(mm, anon_rss);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700747#ifdef CONFIG_MIGRATION
Christoph Lameter06972122006-06-23 02:03:35 -0700748 } else {
749 /*
750 * Store the pfn of the page in a special migration
751 * pte. do_swap_page() will wait until the migration
752 * pte is removed and then restart fault handling.
753 */
754 BUG_ON(!migration);
755 entry = make_migration_entry(page, pte_write(pteval));
Christoph Lameter04e62a22006-06-23 02:03:38 -0700756#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 }
758 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
759 BUG_ON(pte_file(*pte));
Hugh Dickins42946212005-10-29 18:16:05 -0700760 } else
Christoph Lameter04e62a22006-06-23 02:03:38 -0700761#ifdef CONFIG_MIGRATION
762 if (migration) {
763 /* Establish migration entry for a file page */
764 swp_entry_t entry;
765 entry = make_migration_entry(page, pte_write(pteval));
766 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
767 } else
768#endif
Hugh Dickins42946212005-10-29 18:16:05 -0700769 dec_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Christoph Lameter04e62a22006-06-23 02:03:38 -0700771
Nick Piggin7de6b802006-12-22 01:09:33 -0800772 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 page_cache_release(page);
774
775out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700776 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777out:
778 return ret;
779}
780
781/*
782 * objrmap doesn't work for nonlinear VMAs because the assumption that
783 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
784 * Consequently, given a particular page and its ->index, we cannot locate the
785 * ptes which are mapping that page without an exhaustive linear search.
786 *
787 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
788 * maps the file to which the target page belongs. The ->vm_private_data field
789 * holds the current cursor into that scan. Successive searches will circulate
790 * around the vma's virtual address space.
791 *
792 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
793 * more scanning pressure is placed against them as well. Eventually pages
794 * will become fully unmapped and are eligible for eviction.
795 *
796 * For very sparsely populated VMAs this is a little inefficient - chances are
797 * there there won't be many ptes located within the scan cluster. In this case
798 * maybe we could scan further - to the end of the pte page, perhaps.
799 */
800#define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
801#define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
802
803static void try_to_unmap_cluster(unsigned long cursor,
804 unsigned int *mapcount, struct vm_area_struct *vma)
805{
806 struct mm_struct *mm = vma->vm_mm;
807 pgd_t *pgd;
808 pud_t *pud;
809 pmd_t *pmd;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700810 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700812 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 struct page *page;
814 unsigned long address;
815 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 address = (vma->vm_start + cursor) & CLUSTER_MASK;
818 end = address + CLUSTER_SIZE;
819 if (address < vma->vm_start)
820 address = vma->vm_start;
821 if (end > vma->vm_end)
822 end = vma->vm_end;
823
824 pgd = pgd_offset(mm, address);
825 if (!pgd_present(*pgd))
Hugh Dickinsc0718802005-10-29 18:16:31 -0700826 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
828 pud = pud_offset(pgd, address);
829 if (!pud_present(*pud))
Hugh Dickinsc0718802005-10-29 18:16:31 -0700830 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831
832 pmd = pmd_offset(pud, address);
833 if (!pmd_present(*pmd))
Hugh Dickinsc0718802005-10-29 18:16:31 -0700834 return;
835
836 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837
Hugh Dickins365e9c872005-10-29 18:16:18 -0700838 /* Update high watermark before we lower rss */
839 update_hiwater_rss(mm);
840
Hugh Dickinsc0718802005-10-29 18:16:31 -0700841 for (; address < end; pte++, address += PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 if (!pte_present(*pte))
843 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800844 page = vm_normal_page(vma, address, *pte);
845 BUG_ON(!page || PageAnon(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
847 if (ptep_clear_flush_young(vma, address, pte))
848 continue;
849
850 /* Nuke the page table entry. */
Ben Collinseca35132005-11-29 11:45:26 -0800851 flush_cache_page(vma, address, pte_pfn(*pte));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 pteval = ptep_clear_flush(vma, address, pte);
853
854 /* If nonlinear, store the file page offset in the pte. */
855 if (page->index != linear_page_index(vma, address))
856 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
857
858 /* Move the dirty bit to the physical page now the pte is gone. */
859 if (pte_dirty(pteval))
860 set_page_dirty(page);
861
Nick Piggin7de6b802006-12-22 01:09:33 -0800862 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 page_cache_release(page);
Hugh Dickins42946212005-10-29 18:16:05 -0700864 dec_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 (*mapcount)--;
866 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700867 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868}
869
Christoph Lameter73523492006-06-23 02:03:27 -0700870static int try_to_unmap_anon(struct page *page, int migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871{
872 struct anon_vma *anon_vma;
873 struct vm_area_struct *vma;
874 int ret = SWAP_AGAIN;
875
876 anon_vma = page_lock_anon_vma(page);
877 if (!anon_vma)
878 return ret;
879
880 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
Christoph Lameter73523492006-06-23 02:03:27 -0700881 ret = try_to_unmap_one(page, vma, migration);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 if (ret == SWAP_FAIL || !page_mapped(page))
883 break;
884 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800885
886 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 return ret;
888}
889
890/**
891 * try_to_unmap_file - unmap file page using the object-based rmap method
892 * @page: the page to unmap
893 *
894 * Find all the mappings of a page using the mapping pointer and the vma chains
895 * contained in the address_space struct it points to.
896 *
897 * This function is only called from try_to_unmap for object-based pages.
898 */
Christoph Lameter73523492006-06-23 02:03:27 -0700899static int try_to_unmap_file(struct page *page, int migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900{
901 struct address_space *mapping = page->mapping;
902 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
903 struct vm_area_struct *vma;
904 struct prio_tree_iter iter;
905 int ret = SWAP_AGAIN;
906 unsigned long cursor;
907 unsigned long max_nl_cursor = 0;
908 unsigned long max_nl_size = 0;
909 unsigned int mapcount;
910
911 spin_lock(&mapping->i_mmap_lock);
912 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Christoph Lameter73523492006-06-23 02:03:27 -0700913 ret = try_to_unmap_one(page, vma, migration);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 if (ret == SWAP_FAIL || !page_mapped(page))
915 goto out;
916 }
917
918 if (list_empty(&mapping->i_mmap_nonlinear))
919 goto out;
920
921 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
922 shared.vm_set.list) {
Christoph Lametere6a15302006-06-25 05:46:49 -0700923 if ((vma->vm_flags & VM_LOCKED) && !migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 continue;
925 cursor = (unsigned long) vma->vm_private_data;
926 if (cursor > max_nl_cursor)
927 max_nl_cursor = cursor;
928 cursor = vma->vm_end - vma->vm_start;
929 if (cursor > max_nl_size)
930 max_nl_size = cursor;
931 }
932
933 if (max_nl_size == 0) { /* any nonlinears locked or reserved */
934 ret = SWAP_FAIL;
935 goto out;
936 }
937
938 /*
939 * We don't try to search for this page in the nonlinear vmas,
940 * and page_referenced wouldn't have found it anyway. Instead
941 * just walk the nonlinear vmas trying to age and unmap some.
942 * The mapcount of the page we came in with is irrelevant,
943 * but even so use it as a guide to how hard we should try?
944 */
945 mapcount = page_mapcount(page);
946 if (!mapcount)
947 goto out;
948 cond_resched_lock(&mapping->i_mmap_lock);
949
950 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
951 if (max_nl_cursor == 0)
952 max_nl_cursor = CLUSTER_SIZE;
953
954 do {
955 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
956 shared.vm_set.list) {
Christoph Lametere6a15302006-06-25 05:46:49 -0700957 if ((vma->vm_flags & VM_LOCKED) && !migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 continue;
959 cursor = (unsigned long) vma->vm_private_data;
Hugh Dickins839b9682005-09-03 15:54:43 -0700960 while ( cursor < max_nl_cursor &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 cursor < vma->vm_end - vma->vm_start) {
962 try_to_unmap_cluster(cursor, &mapcount, vma);
963 cursor += CLUSTER_SIZE;
964 vma->vm_private_data = (void *) cursor;
965 if ((int)mapcount <= 0)
966 goto out;
967 }
968 vma->vm_private_data = (void *) max_nl_cursor;
969 }
970 cond_resched_lock(&mapping->i_mmap_lock);
971 max_nl_cursor += CLUSTER_SIZE;
972 } while (max_nl_cursor <= max_nl_size);
973
974 /*
975 * Don't loop forever (perhaps all the remaining pages are
976 * in locked vmas). Reset cursor on all unreserved nonlinear
977 * vmas, now forgetting on which ones it had fallen behind.
978 */
Hugh Dickins101d2be2005-11-21 21:32:16 -0800979 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
980 vma->vm_private_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981out:
982 spin_unlock(&mapping->i_mmap_lock);
983 return ret;
984}
985
986/**
987 * try_to_unmap - try to remove all page table mappings to a page
988 * @page: the page to get unmapped
989 *
990 * Tries to remove all the page table entries which are mapping this
991 * page, used in the pageout path. Caller must hold the page lock.
992 * Return values are:
993 *
994 * SWAP_SUCCESS - we succeeded in removing all mappings
995 * SWAP_AGAIN - we missed a mapping, try again later
996 * SWAP_FAIL - the page is unswappable
997 */
Christoph Lameter73523492006-06-23 02:03:27 -0700998int try_to_unmap(struct page *page, int migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999{
1000 int ret;
1001
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 BUG_ON(!PageLocked(page));
1003
1004 if (PageAnon(page))
Christoph Lameter73523492006-06-23 02:03:27 -07001005 ret = try_to_unmap_anon(page, migration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 else
Christoph Lameter73523492006-06-23 02:03:27 -07001007 ret = try_to_unmap_file(page, migration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009 if (!page_mapped(page))
1010 ret = SWAP_SUCCESS;
1011 return ret;
1012}
Nikita Danilov81b40822005-05-01 08:58:36 -07001013