blob: 39ae5a9bf382615120e14e3fbe4efd3bc7cda153 [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
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700141static void anon_vma_ctor(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 */
Hugh Dickinsbd845e32008-03-04 14:29:01 -0800324 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
Balbir Singhbed71612008-02-07 00:14:01 -0800325 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.
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700338 * @mem_cont: target memory controller
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 *
340 * For an object-based mapped page, find all the places it is mapped and
341 * check/clear the referenced flag. This is done by following the page->mapping
342 * pointer, then walking the chain of vmas it holds. It returns the number
343 * of references it found.
344 *
345 * This function is only called from page_referenced for object-based pages.
346 */
Balbir Singhbed71612008-02-07 00:14:01 -0800347static int page_referenced_file(struct page *page,
348 struct mem_cgroup *mem_cont)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349{
350 unsigned int mapcount;
351 struct address_space *mapping = page->mapping;
352 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
353 struct vm_area_struct *vma;
354 struct prio_tree_iter iter;
355 int referenced = 0;
356
357 /*
358 * The caller's checks on page->mapping and !PageAnon have made
359 * sure that this is a file page: the check for page->mapping
360 * excludes the case just before it gets set on an anon page.
361 */
362 BUG_ON(PageAnon(page));
363
364 /*
365 * The page lock not only makes sure that page->mapping cannot
366 * suddenly be NULLified by truncation, it makes sure that the
367 * structure at mapping cannot be freed and reused yet,
368 * so we can safely take mapping->i_mmap_lock.
369 */
370 BUG_ON(!PageLocked(page));
371
372 spin_lock(&mapping->i_mmap_lock);
373
374 /*
375 * i_mmap_lock does not stabilize mapcount at all, but mapcount
376 * is more likely to be accurate if we note it after spinning.
377 */
378 mapcount = page_mapcount(page);
379
380 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Balbir Singhbed71612008-02-07 00:14:01 -0800381 /*
382 * If we are reclaiming on behalf of a cgroup, skip
383 * counting on behalf of references from different
384 * cgroups
385 */
Hugh Dickinsbd845e32008-03-04 14:29:01 -0800386 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
Balbir Singhbed71612008-02-07 00:14:01 -0800387 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 if ((vma->vm_flags & (VM_LOCKED|VM_MAYSHARE))
389 == (VM_LOCKED|VM_MAYSHARE)) {
390 referenced++;
391 break;
392 }
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800393 referenced += page_referenced_one(page, vma, &mapcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 if (!mapcount)
395 break;
396 }
397
398 spin_unlock(&mapping->i_mmap_lock);
399 return referenced;
400}
401
402/**
403 * page_referenced - test if the page was referenced
404 * @page: the page to test
405 * @is_locked: caller holds lock on the page
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700406 * @mem_cont: target memory controller
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 *
408 * Quick test_and_clear_referenced for all mappings to a page,
409 * returns the number of ptes which referenced the page.
410 */
Balbir Singhbed71612008-02-07 00:14:01 -0800411int page_referenced(struct page *page, int is_locked,
412 struct mem_cgroup *mem_cont)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413{
414 int referenced = 0;
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 if (TestClearPageReferenced(page))
417 referenced++;
418
419 if (page_mapped(page) && page->mapping) {
420 if (PageAnon(page))
Balbir Singhbed71612008-02-07 00:14:01 -0800421 referenced += page_referenced_anon(page, mem_cont);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 else if (is_locked)
Balbir Singhbed71612008-02-07 00:14:01 -0800423 referenced += page_referenced_file(page, mem_cont);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 else if (TestSetPageLocked(page))
425 referenced++;
426 else {
427 if (page->mapping)
Balbir Singhbed71612008-02-07 00:14:01 -0800428 referenced +=
429 page_referenced_file(page, mem_cont);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 unlock_page(page);
431 }
432 }
Christian Borntraeger5b7baf02008-03-25 18:47:12 +0100433
434 if (page_test_and_clear_young(page))
435 referenced++;
436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 return referenced;
438}
439
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700440static int page_mkclean_one(struct page *page, struct vm_area_struct *vma)
441{
442 struct mm_struct *mm = vma->vm_mm;
443 unsigned long address;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100444 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700445 spinlock_t *ptl;
446 int ret = 0;
447
448 address = vma_address(page, vma);
449 if (address == -EFAULT)
450 goto out;
451
452 pte = page_check_address(page, mm, address, &ptl);
453 if (!pte)
454 goto out;
455
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100456 if (pte_dirty(*pte) || pte_write(*pte)) {
457 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700458
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100459 flush_cache_page(vma, address, pte_pfn(*pte));
460 entry = ptep_clear_flush(vma, address, pte);
461 entry = pte_wrprotect(entry);
462 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800463 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100464 ret = 1;
465 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700466
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700467 pte_unmap_unlock(pte, ptl);
468out:
469 return ret;
470}
471
472static int page_mkclean_file(struct address_space *mapping, struct page *page)
473{
474 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
475 struct vm_area_struct *vma;
476 struct prio_tree_iter iter;
477 int ret = 0;
478
479 BUG_ON(PageAnon(page));
480
481 spin_lock(&mapping->i_mmap_lock);
482 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
483 if (vma->vm_flags & VM_SHARED)
484 ret += page_mkclean_one(page, vma);
485 }
486 spin_unlock(&mapping->i_mmap_lock);
487 return ret;
488}
489
490int page_mkclean(struct page *page)
491{
492 int ret = 0;
493
494 BUG_ON(!PageLocked(page));
495
496 if (page_mapped(page)) {
497 struct address_space *mapping = page_mapping(page);
Christian Borntraegerce7e9fa2007-11-20 11:13:36 +0100498 if (mapping) {
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700499 ret = page_mkclean_file(mapping, page);
Christian Borntraegerce7e9fa2007-11-20 11:13:36 +0100500 if (page_test_dirty(page)) {
501 page_clear_dirty(page);
502 ret = 1;
503 }
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200504 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700505 }
506
507 return ret;
508}
Jaya Kumar60b59be2007-05-08 00:37:37 -0700509EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700512 * __page_set_anon_rmap - setup new anonymous rmap
Nick Piggin9617d952006-01-06 00:11:12 -0800513 * @page: the page to add the mapping to
514 * @vma: the vm area in which the mapping is added
515 * @address: the user virtual address mapped
516 */
517static void __page_set_anon_rmap(struct page *page,
518 struct vm_area_struct *vma, unsigned long address)
519{
520 struct anon_vma *anon_vma = vma->anon_vma;
521
522 BUG_ON(!anon_vma);
523 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
524 page->mapping = (struct address_space *) anon_vma;
525
526 page->index = linear_page_index(vma, address);
527
Nick Piggina74609f2006-01-06 00:11:20 -0800528 /*
529 * nr_mapped state can be updated without turning off
530 * interrupts because it is not modified via interrupt.
531 */
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700532 __inc_zone_page_state(page, NR_ANON_PAGES);
Nick Piggin9617d952006-01-06 00:11:12 -0800533}
534
535/**
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700536 * __page_check_anon_rmap - sanity check anonymous rmap addition
Nick Pigginc97a9e12007-05-16 22:11:21 -0700537 * @page: the page to add the mapping to
538 * @vma: the vm area in which the mapping is added
539 * @address: the user virtual address mapped
540 */
541static void __page_check_anon_rmap(struct page *page,
542 struct vm_area_struct *vma, unsigned long address)
543{
544#ifdef CONFIG_DEBUG_VM
545 /*
546 * The page's anon-rmap details (mapping and index) are guaranteed to
547 * be set up correctly at this point.
548 *
549 * We have exclusion against page_add_anon_rmap because the caller
550 * always holds the page locked, except if called from page_dup_rmap,
551 * in which case the page is already known to be setup.
552 *
553 * We have exclusion against page_add_new_anon_rmap because those pages
554 * are initially only visible via the pagetables, and the pte is locked
555 * over the call to page_add_new_anon_rmap.
556 */
557 struct anon_vma *anon_vma = vma->anon_vma;
558 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
559 BUG_ON(page->mapping != (struct address_space *)anon_vma);
560 BUG_ON(page->index != linear_page_index(vma, address));
561#endif
562}
563
564/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 * page_add_anon_rmap - add pte mapping to an anonymous page
566 * @page: the page to add the mapping to
567 * @vma: the vm area in which the mapping is added
568 * @address: the user virtual address mapped
569 *
Nick Pigginc97a9e12007-05-16 22:11:21 -0700570 * The caller needs to hold the pte lock and the page must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 */
572void page_add_anon_rmap(struct page *page,
573 struct vm_area_struct *vma, unsigned long address)
574{
Nick Pigginc97a9e12007-05-16 22:11:21 -0700575 VM_BUG_ON(!PageLocked(page));
576 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Nick Piggin9617d952006-01-06 00:11:12 -0800577 if (atomic_inc_and_test(&page->_mapcount))
578 __page_set_anon_rmap(page, vma, address);
KAMEZAWA Hiroyuki69029cd2008-07-25 01:47:14 -0700579 else
Nick Pigginc97a9e12007-05-16 22:11:21 -0700580 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581}
582
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700583/**
Nick Piggin9617d952006-01-06 00:11:12 -0800584 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
585 * @page: the page to add the mapping to
586 * @vma: the vm area in which the mapping is added
587 * @address: the user virtual address mapped
588 *
589 * Same as page_add_anon_rmap but must only be called on *new* pages.
590 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -0700591 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -0800592 */
593void page_add_new_anon_rmap(struct page *page,
594 struct vm_area_struct *vma, unsigned long address)
595{
Nick Pigginc97a9e12007-05-16 22:11:21 -0700596 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Nick Piggin9617d952006-01-06 00:11:12 -0800597 atomic_set(&page->_mapcount, 0); /* elevate count by 1 (starts at -1) */
598 __page_set_anon_rmap(page, vma, address);
599}
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601/**
602 * page_add_file_rmap - add pte mapping to a file page
603 * @page: the page to add the mapping to
604 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700605 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 */
607void page_add_file_rmap(struct page *page)
608{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 if (atomic_inc_and_test(&page->_mapcount))
Christoph Lameter65ba55f2006-06-30 01:55:34 -0700610 __inc_zone_page_state(page, NR_FILE_MAPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611}
612
Nick Pigginc97a9e12007-05-16 22:11:21 -0700613#ifdef CONFIG_DEBUG_VM
614/**
615 * page_dup_rmap - duplicate pte mapping to a page
616 * @page: the page to add the mapping to
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700617 * @vma: the vm area being duplicated
618 * @address: the user virtual address mapped
Nick Pigginc97a9e12007-05-16 22:11:21 -0700619 *
620 * For copy_page_range only: minimal extract from page_add_file_rmap /
621 * page_add_anon_rmap, avoiding unnecessary tests (already checked) so it's
622 * quicker.
623 *
624 * The caller needs to hold the pte lock.
625 */
626void page_dup_rmap(struct page *page, struct vm_area_struct *vma, unsigned long address)
627{
628 BUG_ON(page_mapcount(page) == 0);
629 if (PageAnon(page))
630 __page_check_anon_rmap(page, vma, address);
631 atomic_inc(&page->_mapcount);
632}
633#endif
634
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635/**
636 * page_remove_rmap - take down pte mapping from a page
637 * @page: page to remove mapping from
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700638 * @vma: the vm area in which the mapping is removed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700640 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 */
Nick Piggin7de6b802006-12-22 01:09:33 -0800642void page_remove_rmap(struct page *page, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 if (atomic_add_negative(-1, &page->_mapcount)) {
Nick Pigginb7ab7952006-03-22 00:08:42 -0800645 if (unlikely(page_mapcount(page) < 0)) {
Dave Jonesef2bf0d2006-01-08 01:01:00 -0800646 printk (KERN_EMERG "Eeek! page_mapcount(page) went negative! (%d)\n", page_mapcount(page));
Nick Piggin7de6b802006-12-22 01:09:33 -0800647 printk (KERN_EMERG " page pfn = %lx\n", page_to_pfn(page));
Dave Jonesef2bf0d2006-01-08 01:01:00 -0800648 printk (KERN_EMERG " page->flags = %lx\n", page->flags);
649 printk (KERN_EMERG " page->count = %x\n", page_count(page));
650 printk (KERN_EMERG " page->mapping = %p\n", page->mapping);
Nick Piggin7de6b802006-12-22 01:09:33 -0800651 print_symbol (KERN_EMERG " vma->vm_ops = %s\n", (unsigned long)vma->vm_ops);
Nick Piggin54cb8822007-07-19 01:46:59 -0700652 if (vma->vm_ops) {
Nick Piggin54cb8822007-07-19 01:46:59 -0700653 print_symbol (KERN_EMERG " vma->vm_ops->fault = %s\n", (unsigned long)vma->vm_ops->fault);
654 }
Nick Piggin7de6b802006-12-22 01:09:33 -0800655 if (vma->vm_file && vma->vm_file->f_op)
656 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 -0700657 BUG();
Dave Jonesef2bf0d2006-01-08 01:01:00 -0800658 }
Dave Jonesb16bc642006-10-11 01:21:27 -0700659
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 /*
661 * It would be tidy to reset the PageAnon mapping here,
662 * but that might overwrite a racing page_add_anon_rmap
663 * which increments mapcount after us but sets mapping
664 * before us: so leave the reset to free_hot_cold_page,
665 * and remember that it's only reliable while mapped.
666 * Leaving it set also helps swapoff to reinstate ptes
667 * faster for those pages still in swapcache.
668 */
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200669 if (page_test_dirty(page)) {
670 page_clear_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 set_page_dirty(page);
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200672 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800673 mem_cgroup_uncharge_page(page);
674
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700675 __dec_zone_page_state(page,
676 PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 }
678}
679
680/*
681 * Subfunctions of try_to_unmap: try_to_unmap_one called
682 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
683 */
Christoph Lametera48d07a2006-02-01 03:05:38 -0800684static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
Christoph Lameter73523492006-06-23 02:03:27 -0700685 int migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686{
687 struct mm_struct *mm = vma->vm_mm;
688 unsigned long address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 pte_t *pte;
690 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700691 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 int ret = SWAP_AGAIN;
693
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 address = vma_address(page, vma);
695 if (address == -EFAULT)
696 goto out;
697
Hugh Dickinsc0718802005-10-29 18:16:31 -0700698 pte = page_check_address(page, mm, address, &ptl);
699 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -0700700 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
702 /*
703 * If the page is mlock()d, we cannot swap it out.
704 * If it's recently referenced (perhaps page_referenced
705 * skipped over this mm) then we should reactivate it.
706 */
Christoph Lametere6a15302006-06-25 05:46:49 -0700707 if (!migration && ((vma->vm_flags & VM_LOCKED) ||
708 (ptep_clear_flush_young(vma, address, pte)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 ret = SWAP_FAIL;
710 goto out_unmap;
711 }
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 /* Nuke the page table entry. */
714 flush_cache_page(vma, address, page_to_pfn(page));
715 pteval = ptep_clear_flush(vma, address, pte);
716
717 /* Move the dirty bit to the physical page now the pte is gone. */
718 if (pte_dirty(pteval))
719 set_page_dirty(page);
720
Hugh Dickins365e9c872005-10-29 18:16:18 -0700721 /* Update high watermark before we lower rss */
722 update_hiwater_rss(mm);
723
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700725 swp_entry_t entry = { .val = page_private(page) };
Christoph Lameter06972122006-06-23 02:03:35 -0700726
727 if (PageSwapCache(page)) {
728 /*
729 * Store the swap location in the pte.
730 * See handle_pte_fault() ...
731 */
732 swap_duplicate(entry);
733 if (list_empty(&mm->mmlist)) {
734 spin_lock(&mmlist_lock);
735 if (list_empty(&mm->mmlist))
736 list_add(&mm->mmlist, &init_mm.mmlist);
737 spin_unlock(&mmlist_lock);
738 }
Christoph Lameter442c9132006-06-23 02:03:38 -0700739 dec_mm_counter(mm, anon_rss);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700740#ifdef CONFIG_MIGRATION
Christoph Lameter06972122006-06-23 02:03:35 -0700741 } else {
742 /*
743 * Store the pfn of the page in a special migration
744 * pte. do_swap_page() will wait until the migration
745 * pte is removed and then restart fault handling.
746 */
747 BUG_ON(!migration);
748 entry = make_migration_entry(page, pte_write(pteval));
Christoph Lameter04e62a22006-06-23 02:03:38 -0700749#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 }
751 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
752 BUG_ON(pte_file(*pte));
Hugh Dickins42946212005-10-29 18:16:05 -0700753 } else
Christoph Lameter04e62a22006-06-23 02:03:38 -0700754#ifdef CONFIG_MIGRATION
755 if (migration) {
756 /* Establish migration entry for a file page */
757 swp_entry_t entry;
758 entry = make_migration_entry(page, pte_write(pteval));
759 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
760 } else
761#endif
Hugh Dickins42946212005-10-29 18:16:05 -0700762 dec_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
Christoph Lameter04e62a22006-06-23 02:03:38 -0700764
Nick Piggin7de6b802006-12-22 01:09:33 -0800765 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 page_cache_release(page);
767
768out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700769 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770out:
771 return ret;
772}
773
774/*
775 * objrmap doesn't work for nonlinear VMAs because the assumption that
776 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
777 * Consequently, given a particular page and its ->index, we cannot locate the
778 * ptes which are mapping that page without an exhaustive linear search.
779 *
780 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
781 * maps the file to which the target page belongs. The ->vm_private_data field
782 * holds the current cursor into that scan. Successive searches will circulate
783 * around the vma's virtual address space.
784 *
785 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
786 * more scanning pressure is placed against them as well. Eventually pages
787 * will become fully unmapped and are eligible for eviction.
788 *
789 * For very sparsely populated VMAs this is a little inefficient - chances are
790 * there there won't be many ptes located within the scan cluster. In this case
791 * maybe we could scan further - to the end of the pte page, perhaps.
792 */
793#define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
794#define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
795
796static void try_to_unmap_cluster(unsigned long cursor,
797 unsigned int *mapcount, struct vm_area_struct *vma)
798{
799 struct mm_struct *mm = vma->vm_mm;
800 pgd_t *pgd;
801 pud_t *pud;
802 pmd_t *pmd;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700803 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700805 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 struct page *page;
807 unsigned long address;
808 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 address = (vma->vm_start + cursor) & CLUSTER_MASK;
811 end = address + CLUSTER_SIZE;
812 if (address < vma->vm_start)
813 address = vma->vm_start;
814 if (end > vma->vm_end)
815 end = vma->vm_end;
816
817 pgd = pgd_offset(mm, address);
818 if (!pgd_present(*pgd))
Hugh Dickinsc0718802005-10-29 18:16:31 -0700819 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
821 pud = pud_offset(pgd, address);
822 if (!pud_present(*pud))
Hugh Dickinsc0718802005-10-29 18:16:31 -0700823 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
825 pmd = pmd_offset(pud, address);
826 if (!pmd_present(*pmd))
Hugh Dickinsc0718802005-10-29 18:16:31 -0700827 return;
828
829 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
Hugh Dickins365e9c872005-10-29 18:16:18 -0700831 /* Update high watermark before we lower rss */
832 update_hiwater_rss(mm);
833
Hugh Dickinsc0718802005-10-29 18:16:31 -0700834 for (; address < end; pte++, address += PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 if (!pte_present(*pte))
836 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800837 page = vm_normal_page(vma, address, *pte);
838 BUG_ON(!page || PageAnon(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
840 if (ptep_clear_flush_young(vma, address, pte))
841 continue;
842
843 /* Nuke the page table entry. */
Ben Collinseca35132005-11-29 11:45:26 -0800844 flush_cache_page(vma, address, pte_pfn(*pte));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 pteval = ptep_clear_flush(vma, address, pte);
846
847 /* If nonlinear, store the file page offset in the pte. */
848 if (page->index != linear_page_index(vma, address))
849 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
850
851 /* Move the dirty bit to the physical page now the pte is gone. */
852 if (pte_dirty(pteval))
853 set_page_dirty(page);
854
Nick Piggin7de6b802006-12-22 01:09:33 -0800855 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 page_cache_release(page);
Hugh Dickins42946212005-10-29 18:16:05 -0700857 dec_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 (*mapcount)--;
859 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700860 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861}
862
Christoph Lameter73523492006-06-23 02:03:27 -0700863static int try_to_unmap_anon(struct page *page, int migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864{
865 struct anon_vma *anon_vma;
866 struct vm_area_struct *vma;
867 int ret = SWAP_AGAIN;
868
869 anon_vma = page_lock_anon_vma(page);
870 if (!anon_vma)
871 return ret;
872
873 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
Christoph Lameter73523492006-06-23 02:03:27 -0700874 ret = try_to_unmap_one(page, vma, migration);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 if (ret == SWAP_FAIL || !page_mapped(page))
876 break;
877 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800878
879 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 return ret;
881}
882
883/**
884 * try_to_unmap_file - unmap file page using the object-based rmap method
885 * @page: the page to unmap
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700886 * @migration: migration flag
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 *
888 * Find all the mappings of a page using the mapping pointer and the vma chains
889 * contained in the address_space struct it points to.
890 *
891 * This function is only called from try_to_unmap for object-based pages.
892 */
Christoph Lameter73523492006-06-23 02:03:27 -0700893static int try_to_unmap_file(struct page *page, int migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894{
895 struct address_space *mapping = page->mapping;
896 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
897 struct vm_area_struct *vma;
898 struct prio_tree_iter iter;
899 int ret = SWAP_AGAIN;
900 unsigned long cursor;
901 unsigned long max_nl_cursor = 0;
902 unsigned long max_nl_size = 0;
903 unsigned int mapcount;
904
905 spin_lock(&mapping->i_mmap_lock);
906 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Christoph Lameter73523492006-06-23 02:03:27 -0700907 ret = try_to_unmap_one(page, vma, migration);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 if (ret == SWAP_FAIL || !page_mapped(page))
909 goto out;
910 }
911
912 if (list_empty(&mapping->i_mmap_nonlinear))
913 goto out;
914
915 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
916 shared.vm_set.list) {
Christoph Lametere6a15302006-06-25 05:46:49 -0700917 if ((vma->vm_flags & VM_LOCKED) && !migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 continue;
919 cursor = (unsigned long) vma->vm_private_data;
920 if (cursor > max_nl_cursor)
921 max_nl_cursor = cursor;
922 cursor = vma->vm_end - vma->vm_start;
923 if (cursor > max_nl_size)
924 max_nl_size = cursor;
925 }
926
927 if (max_nl_size == 0) { /* any nonlinears locked or reserved */
928 ret = SWAP_FAIL;
929 goto out;
930 }
931
932 /*
933 * We don't try to search for this page in the nonlinear vmas,
934 * and page_referenced wouldn't have found it anyway. Instead
935 * just walk the nonlinear vmas trying to age and unmap some.
936 * The mapcount of the page we came in with is irrelevant,
937 * but even so use it as a guide to how hard we should try?
938 */
939 mapcount = page_mapcount(page);
940 if (!mapcount)
941 goto out;
942 cond_resched_lock(&mapping->i_mmap_lock);
943
944 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
945 if (max_nl_cursor == 0)
946 max_nl_cursor = CLUSTER_SIZE;
947
948 do {
949 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
950 shared.vm_set.list) {
Christoph Lametere6a15302006-06-25 05:46:49 -0700951 if ((vma->vm_flags & VM_LOCKED) && !migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 continue;
953 cursor = (unsigned long) vma->vm_private_data;
Hugh Dickins839b9682005-09-03 15:54:43 -0700954 while ( cursor < max_nl_cursor &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 cursor < vma->vm_end - vma->vm_start) {
956 try_to_unmap_cluster(cursor, &mapcount, vma);
957 cursor += CLUSTER_SIZE;
958 vma->vm_private_data = (void *) cursor;
959 if ((int)mapcount <= 0)
960 goto out;
961 }
962 vma->vm_private_data = (void *) max_nl_cursor;
963 }
964 cond_resched_lock(&mapping->i_mmap_lock);
965 max_nl_cursor += CLUSTER_SIZE;
966 } while (max_nl_cursor <= max_nl_size);
967
968 /*
969 * Don't loop forever (perhaps all the remaining pages are
970 * in locked vmas). Reset cursor on all unreserved nonlinear
971 * vmas, now forgetting on which ones it had fallen behind.
972 */
Hugh Dickins101d2be2005-11-21 21:32:16 -0800973 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
974 vma->vm_private_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975out:
976 spin_unlock(&mapping->i_mmap_lock);
977 return ret;
978}
979
980/**
981 * try_to_unmap - try to remove all page table mappings to a page
982 * @page: the page to get unmapped
Randy Dunlap43d8eac2008-03-19 17:00:43 -0700983 * @migration: migration flag
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 *
985 * Tries to remove all the page table entries which are mapping this
986 * page, used in the pageout path. Caller must hold the page lock.
987 * Return values are:
988 *
989 * SWAP_SUCCESS - we succeeded in removing all mappings
990 * SWAP_AGAIN - we missed a mapping, try again later
991 * SWAP_FAIL - the page is unswappable
992 */
Christoph Lameter73523492006-06-23 02:03:27 -0700993int try_to_unmap(struct page *page, int migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
995 int ret;
996
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 BUG_ON(!PageLocked(page));
998
999 if (PageAnon(page))
Christoph Lameter73523492006-06-23 02:03:27 -07001000 ret = try_to_unmap_anon(page, migration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 else
Christoph Lameter73523492006-06-23 02:03:27 -07001002 ret = try_to_unmap_file(page, migration);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
1004 if (!page_mapped(page))
1005 ret = SWAP_SUCCESS;
1006 return ret;
1007}
Nikita Danilov81b40822005-05-01 08:58:36 -07001008