blob: 61e492597a0ba4be07634c60b427031743f533d4 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
52#include <asm/tlbflush.h>
53
Pekka Enbergfcc234f2006-03-22 00:08:13 -080054struct kmem_cache *anon_vma_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Linus Torvalds1da177e2005-04-16 15:20:36 -070056/* This must be called under the mmap_sem. */
57int anon_vma_prepare(struct vm_area_struct *vma)
58{
59 struct anon_vma *anon_vma = vma->anon_vma;
60
61 might_sleep();
62 if (unlikely(!anon_vma)) {
63 struct mm_struct *mm = vma->vm_mm;
64 struct anon_vma *allocated, *locked;
65
66 anon_vma = find_mergeable_anon_vma(vma);
67 if (anon_vma) {
68 allocated = NULL;
69 locked = anon_vma;
70 spin_lock(&locked->lock);
71 } else {
72 anon_vma = anon_vma_alloc();
73 if (unlikely(!anon_vma))
74 return -ENOMEM;
75 allocated = anon_vma;
76 locked = NULL;
77 }
78
79 /* page_table_lock to protect against threads */
80 spin_lock(&mm->page_table_lock);
81 if (likely(!vma->anon_vma)) {
82 vma->anon_vma = anon_vma;
Christoph Lameter06972122006-06-23 02:03:35 -070083 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 allocated = NULL;
85 }
86 spin_unlock(&mm->page_table_lock);
87
88 if (locked)
89 spin_unlock(&locked->lock);
90 if (unlikely(allocated))
91 anon_vma_free(allocated);
92 }
93 return 0;
94}
95
96void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next)
97{
98 BUG_ON(vma->anon_vma != next->anon_vma);
99 list_del(&next->anon_vma_node);
100}
101
102void __anon_vma_link(struct vm_area_struct *vma)
103{
104 struct anon_vma *anon_vma = vma->anon_vma;
105
Hugh Dickins30acbab2007-06-27 14:09:53 -0700106 if (anon_vma)
Christoph Lameter06972122006-06-23 02:03:35 -0700107 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108}
109
110void anon_vma_link(struct vm_area_struct *vma)
111{
112 struct anon_vma *anon_vma = vma->anon_vma;
113
114 if (anon_vma) {
115 spin_lock(&anon_vma->lock);
Christoph Lameter06972122006-06-23 02:03:35 -0700116 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 spin_unlock(&anon_vma->lock);
118 }
119}
120
121void anon_vma_unlink(struct vm_area_struct *vma)
122{
123 struct anon_vma *anon_vma = vma->anon_vma;
124 int empty;
125
126 if (!anon_vma)
127 return;
128
129 spin_lock(&anon_vma->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 list_del(&vma->anon_vma_node);
131
132 /* We must garbage collect the anon_vma if it's empty */
133 empty = list_empty(&anon_vma->head);
134 spin_unlock(&anon_vma->lock);
135
136 if (empty)
137 anon_vma_free(anon_vma);
138}
139
Pekka Enbergfcc234f2006-03-22 00:08:13 -0800140static void anon_vma_ctor(void *data, struct kmem_cache *cachep,
141 unsigned long flags)
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),
152 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor, NULL);
153}
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/*
186 * At what user virtual address is page expected in vma?
187 */
188static inline unsigned long
189vma_address(struct page *page, struct vm_area_struct *vma)
190{
191 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
192 unsigned long address;
193
194 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
195 if (unlikely(address < vma->vm_start || address >= vma->vm_end)) {
196 /* page should be within any vma from prio_tree_next */
197 BUG_ON(!PageAnon(page));
198 return -EFAULT;
199 }
200 return address;
201}
202
203/*
204 * At what user virtual address is page expected in vma? checking that the
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800205 * page matches the vma: currently only used on anon pages, by unuse_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 */
207unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
208{
209 if (PageAnon(page)) {
210 if ((void *)vma->anon_vma !=
211 (void *)page->mapping - PAGE_MAPPING_ANON)
212 return -EFAULT;
213 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
Hugh Dickinsee498ed2005-11-21 21:32:18 -0800214 if (!vma->vm_file ||
215 vma->vm_file->f_mapping != page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 return -EFAULT;
217 } else
218 return -EFAULT;
219 return vma_address(page, vma);
220}
221
222/*
Nikita Danilov81b40822005-05-01 08:58:36 -0700223 * Check that @page is mapped at @address into @mm.
224 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700225 * On success returns with pte mapped and locked.
Nikita Danilov81b40822005-05-01 08:58:36 -0700226 */
Carsten Otteceffc072005-06-23 22:05:25 -0700227pte_t *page_check_address(struct page *page, struct mm_struct *mm,
Hugh Dickinsc0718802005-10-29 18:16:31 -0700228 unsigned long address, spinlock_t **ptlp)
Nikita Danilov81b40822005-05-01 08:58:36 -0700229{
230 pgd_t *pgd;
231 pud_t *pud;
232 pmd_t *pmd;
233 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700234 spinlock_t *ptl;
Nikita Danilov81b40822005-05-01 08:58:36 -0700235
Nikita Danilov81b40822005-05-01 08:58:36 -0700236 pgd = pgd_offset(mm, address);
Hugh Dickinsc0718802005-10-29 18:16:31 -0700237 if (!pgd_present(*pgd))
238 return NULL;
239
240 pud = pud_offset(pgd, address);
241 if (!pud_present(*pud))
242 return NULL;
243
244 pmd = pmd_offset(pud, address);
245 if (!pmd_present(*pmd))
246 return NULL;
247
248 pte = pte_offset_map(pmd, address);
249 /* Make a quick check before getting the lock */
250 if (!pte_present(*pte)) {
251 pte_unmap(pte);
252 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700253 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700254
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700255 ptl = pte_lockptr(mm, pmd);
Hugh Dickinsc0718802005-10-29 18:16:31 -0700256 spin_lock(ptl);
257 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
258 *ptlp = ptl;
259 return pte;
260 }
261 pte_unmap_unlock(pte, ptl);
262 return NULL;
Nikita Danilov81b40822005-05-01 08:58:36 -0700263}
264
265/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 * Subfunctions of page_referenced: page_referenced_one called
267 * repeatedly from either page_referenced_anon or page_referenced_file.
268 */
269static int page_referenced_one(struct page *page,
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800270 struct vm_area_struct *vma, unsigned int *mapcount)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
272 struct mm_struct *mm = vma->vm_mm;
273 unsigned long address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 pte_t *pte;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700275 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 int referenced = 0;
277
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 address = vma_address(page, vma);
279 if (address == -EFAULT)
280 goto out;
281
Hugh Dickinsc0718802005-10-29 18:16:31 -0700282 pte = page_check_address(page, mm, address, &ptl);
283 if (!pte)
284 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285
Hugh Dickinsc0718802005-10-29 18:16:31 -0700286 if (ptep_clear_flush_young(vma, address, pte))
287 referenced++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
Hugh Dickinsc0718802005-10-29 18:16:31 -0700289 /* Pretend the page is referenced if the task has the
290 swap token and is in the middle of a page fault. */
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800291 if (mm != current->mm && has_swap_token(mm) &&
Hugh Dickinsc0718802005-10-29 18:16:31 -0700292 rwsem_is_locked(&mm->mmap_sem))
293 referenced++;
294
295 (*mapcount)--;
296 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297out:
298 return referenced;
299}
300
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800301static int page_referenced_anon(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302{
303 unsigned int mapcount;
304 struct anon_vma *anon_vma;
305 struct vm_area_struct *vma;
306 int referenced = 0;
307
308 anon_vma = page_lock_anon_vma(page);
309 if (!anon_vma)
310 return referenced;
311
312 mapcount = page_mapcount(page);
313 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800314 referenced += page_referenced_one(page, vma, &mapcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 if (!mapcount)
316 break;
317 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800318
319 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 return referenced;
321}
322
323/**
324 * page_referenced_file - referenced check for object-based rmap
325 * @page: the page we're checking references on.
326 *
327 * For an object-based mapped page, find all the places it is mapped and
328 * check/clear the referenced flag. This is done by following the page->mapping
329 * pointer, then walking the chain of vmas it holds. It returns the number
330 * of references it found.
331 *
332 * This function is only called from page_referenced for object-based pages.
333 */
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800334static int page_referenced_file(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335{
336 unsigned int mapcount;
337 struct address_space *mapping = page->mapping;
338 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
339 struct vm_area_struct *vma;
340 struct prio_tree_iter iter;
341 int referenced = 0;
342
343 /*
344 * The caller's checks on page->mapping and !PageAnon have made
345 * sure that this is a file page: the check for page->mapping
346 * excludes the case just before it gets set on an anon page.
347 */
348 BUG_ON(PageAnon(page));
349
350 /*
351 * The page lock not only makes sure that page->mapping cannot
352 * suddenly be NULLified by truncation, it makes sure that the
353 * structure at mapping cannot be freed and reused yet,
354 * so we can safely take mapping->i_mmap_lock.
355 */
356 BUG_ON(!PageLocked(page));
357
358 spin_lock(&mapping->i_mmap_lock);
359
360 /*
361 * i_mmap_lock does not stabilize mapcount at all, but mapcount
362 * is more likely to be accurate if we note it after spinning.
363 */
364 mapcount = page_mapcount(page);
365
366 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
367 if ((vma->vm_flags & (VM_LOCKED|VM_MAYSHARE))
368 == (VM_LOCKED|VM_MAYSHARE)) {
369 referenced++;
370 break;
371 }
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800372 referenced += page_referenced_one(page, vma, &mapcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 if (!mapcount)
374 break;
375 }
376
377 spin_unlock(&mapping->i_mmap_lock);
378 return referenced;
379}
380
381/**
382 * page_referenced - test if the page was referenced
383 * @page: the page to test
384 * @is_locked: caller holds lock on the page
385 *
386 * Quick test_and_clear_referenced for all mappings to a page,
387 * returns the number of ptes which referenced the page.
388 */
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800389int page_referenced(struct page *page, int is_locked)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390{
391 int referenced = 0;
392
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 if (page_test_and_clear_young(page))
394 referenced++;
395
396 if (TestClearPageReferenced(page))
397 referenced++;
398
399 if (page_mapped(page) && page->mapping) {
400 if (PageAnon(page))
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800401 referenced += page_referenced_anon(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 else if (is_locked)
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800403 referenced += page_referenced_file(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 else if (TestSetPageLocked(page))
405 referenced++;
406 else {
407 if (page->mapping)
Rik van Rielf7b7fd82005-11-28 13:44:07 -0800408 referenced += page_referenced_file(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 unlock_page(page);
410 }
411 }
412 return referenced;
413}
414
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700415static int page_mkclean_one(struct page *page, struct vm_area_struct *vma)
416{
417 struct mm_struct *mm = vma->vm_mm;
418 unsigned long address;
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100419 pte_t *pte;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700420 spinlock_t *ptl;
421 int ret = 0;
422
423 address = vma_address(page, vma);
424 if (address == -EFAULT)
425 goto out;
426
427 pte = page_check_address(page, mm, address, &ptl);
428 if (!pte)
429 goto out;
430
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100431 if (pte_dirty(*pte) || pte_write(*pte)) {
432 pte_t entry;
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700433
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100434 flush_cache_page(vma, address, pte_pfn(*pte));
435 entry = ptep_clear_flush(vma, address, pte);
436 entry = pte_wrprotect(entry);
437 entry = pte_mkclean(entry);
Al Virod6e88e62006-12-29 16:48:35 -0800438 set_pte_at(mm, address, pte, entry);
Peter Zijlstrac2fda5f2006-12-22 14:25:52 +0100439 lazy_mmu_prot_update(entry);
440 ret = 1;
441 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700442
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700443 pte_unmap_unlock(pte, ptl);
444out:
445 return ret;
446}
447
448static int page_mkclean_file(struct address_space *mapping, struct page *page)
449{
450 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
451 struct vm_area_struct *vma;
452 struct prio_tree_iter iter;
453 int ret = 0;
454
455 BUG_ON(PageAnon(page));
456
457 spin_lock(&mapping->i_mmap_lock);
458 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
459 if (vma->vm_flags & VM_SHARED)
460 ret += page_mkclean_one(page, vma);
461 }
462 spin_unlock(&mapping->i_mmap_lock);
463 return ret;
464}
465
466int page_mkclean(struct page *page)
467{
468 int ret = 0;
469
470 BUG_ON(!PageLocked(page));
471
472 if (page_mapped(page)) {
473 struct address_space *mapping = page_mapping(page);
474 if (mapping)
475 ret = page_mkclean_file(mapping, page);
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200476 if (page_test_dirty(page)) {
477 page_clear_dirty(page);
Martin Schwidefsky6e1beb32007-04-04 14:37:10 +0200478 ret = 1;
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200479 }
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700480 }
481
482 return ret;
483}
Jaya Kumar60b59be2007-05-08 00:37:37 -0700484EXPORT_SYMBOL_GPL(page_mkclean);
Peter Zijlstrad08b3852006-09-25 23:30:57 -0700485
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486/**
Nick Piggin9617d952006-01-06 00:11:12 -0800487 * page_set_anon_rmap - setup new anonymous rmap
488 * @page: the page to add the mapping to
489 * @vma: the vm area in which the mapping is added
490 * @address: the user virtual address mapped
491 */
492static void __page_set_anon_rmap(struct page *page,
493 struct vm_area_struct *vma, unsigned long address)
494{
495 struct anon_vma *anon_vma = vma->anon_vma;
496
497 BUG_ON(!anon_vma);
498 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
499 page->mapping = (struct address_space *) anon_vma;
500
501 page->index = linear_page_index(vma, address);
502
Nick Piggina74609f2006-01-06 00:11:20 -0800503 /*
504 * nr_mapped state can be updated without turning off
505 * interrupts because it is not modified via interrupt.
506 */
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700507 __inc_zone_page_state(page, NR_ANON_PAGES);
Nick Piggin9617d952006-01-06 00:11:12 -0800508}
509
510/**
Nick Pigginc97a9e12007-05-16 22:11:21 -0700511 * page_set_anon_rmap - sanity check anonymous rmap addition
512 * @page: the page to add the mapping to
513 * @vma: the vm area in which the mapping is added
514 * @address: the user virtual address mapped
515 */
516static void __page_check_anon_rmap(struct page *page,
517 struct vm_area_struct *vma, unsigned long address)
518{
519#ifdef CONFIG_DEBUG_VM
520 /*
521 * The page's anon-rmap details (mapping and index) are guaranteed to
522 * be set up correctly at this point.
523 *
524 * We have exclusion against page_add_anon_rmap because the caller
525 * always holds the page locked, except if called from page_dup_rmap,
526 * in which case the page is already known to be setup.
527 *
528 * We have exclusion against page_add_new_anon_rmap because those pages
529 * are initially only visible via the pagetables, and the pte is locked
530 * over the call to page_add_new_anon_rmap.
531 */
532 struct anon_vma *anon_vma = vma->anon_vma;
533 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
534 BUG_ON(page->mapping != (struct address_space *)anon_vma);
535 BUG_ON(page->index != linear_page_index(vma, address));
536#endif
537}
538
539/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 * page_add_anon_rmap - add pte mapping to an anonymous page
541 * @page: the page to add the mapping to
542 * @vma: the vm area in which the mapping is added
543 * @address: the user virtual address mapped
544 *
Nick Pigginc97a9e12007-05-16 22:11:21 -0700545 * The caller needs to hold the pte lock and the page must be locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 */
547void page_add_anon_rmap(struct page *page,
548 struct vm_area_struct *vma, unsigned long address)
549{
Nick Pigginc97a9e12007-05-16 22:11:21 -0700550 VM_BUG_ON(!PageLocked(page));
551 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Nick Piggin9617d952006-01-06 00:11:12 -0800552 if (atomic_inc_and_test(&page->_mapcount))
553 __page_set_anon_rmap(page, vma, address);
Nick Pigginc97a9e12007-05-16 22:11:21 -0700554 else
555 __page_check_anon_rmap(page, vma, address);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556}
557
Nick Piggin9617d952006-01-06 00:11:12 -0800558/*
559 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
560 * @page: the page to add the mapping to
561 * @vma: the vm area in which the mapping is added
562 * @address: the user virtual address mapped
563 *
564 * Same as page_add_anon_rmap but must only be called on *new* pages.
565 * This means the inc-and-test can be bypassed.
Nick Pigginc97a9e12007-05-16 22:11:21 -0700566 * Page does not have to be locked.
Nick Piggin9617d952006-01-06 00:11:12 -0800567 */
568void page_add_new_anon_rmap(struct page *page,
569 struct vm_area_struct *vma, unsigned long address)
570{
Nick Pigginc97a9e12007-05-16 22:11:21 -0700571 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
Nick Piggin9617d952006-01-06 00:11:12 -0800572 atomic_set(&page->_mapcount, 0); /* elevate count by 1 (starts at -1) */
573 __page_set_anon_rmap(page, vma, address);
574}
575
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576/**
577 * page_add_file_rmap - add pte mapping to a file page
578 * @page: the page to add the mapping to
579 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700580 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 */
582void page_add_file_rmap(struct page *page)
583{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 if (atomic_inc_and_test(&page->_mapcount))
Christoph Lameter65ba55f2006-06-30 01:55:34 -0700585 __inc_zone_page_state(page, NR_FILE_MAPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586}
587
Nick Pigginc97a9e12007-05-16 22:11:21 -0700588#ifdef CONFIG_DEBUG_VM
589/**
590 * page_dup_rmap - duplicate pte mapping to a page
591 * @page: the page to add the mapping to
592 *
593 * For copy_page_range only: minimal extract from page_add_file_rmap /
594 * page_add_anon_rmap, avoiding unnecessary tests (already checked) so it's
595 * quicker.
596 *
597 * The caller needs to hold the pte lock.
598 */
599void page_dup_rmap(struct page *page, struct vm_area_struct *vma, unsigned long address)
600{
601 BUG_ON(page_mapcount(page) == 0);
602 if (PageAnon(page))
603 __page_check_anon_rmap(page, vma, address);
604 atomic_inc(&page->_mapcount);
605}
606#endif
607
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608/**
609 * page_remove_rmap - take down pte mapping from a page
610 * @page: page to remove mapping from
611 *
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700612 * The caller needs to hold the pte lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 */
Nick Piggin7de6b802006-12-22 01:09:33 -0800614void page_remove_rmap(struct page *page, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 if (atomic_add_negative(-1, &page->_mapcount)) {
Nick Pigginb7ab7952006-03-22 00:08:42 -0800617 if (unlikely(page_mapcount(page) < 0)) {
Dave Jonesef2bf0d2006-01-08 01:01:00 -0800618 printk (KERN_EMERG "Eeek! page_mapcount(page) went negative! (%d)\n", page_mapcount(page));
Nick Piggin7de6b802006-12-22 01:09:33 -0800619 printk (KERN_EMERG " page pfn = %lx\n", page_to_pfn(page));
Dave Jonesef2bf0d2006-01-08 01:01:00 -0800620 printk (KERN_EMERG " page->flags = %lx\n", page->flags);
621 printk (KERN_EMERG " page->count = %x\n", page_count(page));
622 printk (KERN_EMERG " page->mapping = %p\n", page->mapping);
Nick Piggin7de6b802006-12-22 01:09:33 -0800623 print_symbol (KERN_EMERG " vma->vm_ops = %s\n", (unsigned long)vma->vm_ops);
624 if (vma->vm_ops)
625 print_symbol (KERN_EMERG " vma->vm_ops->nopage = %s\n", (unsigned long)vma->vm_ops->nopage);
626 if (vma->vm_file && vma->vm_file->f_op)
627 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 -0700628 BUG();
Dave Jonesef2bf0d2006-01-08 01:01:00 -0800629 }
Dave Jonesb16bc642006-10-11 01:21:27 -0700630
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 /*
632 * It would be tidy to reset the PageAnon mapping here,
633 * but that might overwrite a racing page_add_anon_rmap
634 * which increments mapcount after us but sets mapping
635 * before us: so leave the reset to free_hot_cold_page,
636 * and remember that it's only reliable while mapped.
637 * Leaving it set also helps swapoff to reinstate ptes
638 * faster for those pages still in swapcache.
639 */
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200640 if (page_test_dirty(page)) {
641 page_clear_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 set_page_dirty(page);
Martin Schwidefsky6c210482007-04-27 16:01:57 +0200643 }
Christoph Lameterf3dbd342006-06-30 01:55:36 -0700644 __dec_zone_page_state(page,
645 PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 }
647}
648
649/*
650 * Subfunctions of try_to_unmap: try_to_unmap_one called
651 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
652 */
Christoph Lametera48d07a2006-02-01 03:05:38 -0800653static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
Christoph Lameter73523492006-06-23 02:03:27 -0700654 int migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655{
656 struct mm_struct *mm = vma->vm_mm;
657 unsigned long address;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 pte_t *pte;
659 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700660 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 int ret = SWAP_AGAIN;
662
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 address = vma_address(page, vma);
664 if (address == -EFAULT)
665 goto out;
666
Hugh Dickinsc0718802005-10-29 18:16:31 -0700667 pte = page_check_address(page, mm, address, &ptl);
668 if (!pte)
Nikita Danilov81b40822005-05-01 08:58:36 -0700669 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
671 /*
672 * If the page is mlock()d, we cannot swap it out.
673 * If it's recently referenced (perhaps page_referenced
674 * skipped over this mm) then we should reactivate it.
675 */
Christoph Lametere6a15302006-06-25 05:46:49 -0700676 if (!migration && ((vma->vm_flags & VM_LOCKED) ||
677 (ptep_clear_flush_young(vma, address, pte)))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 ret = SWAP_FAIL;
679 goto out_unmap;
680 }
681
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 /* Nuke the page table entry. */
683 flush_cache_page(vma, address, page_to_pfn(page));
684 pteval = ptep_clear_flush(vma, address, pte);
685
686 /* Move the dirty bit to the physical page now the pte is gone. */
687 if (pte_dirty(pteval))
688 set_page_dirty(page);
689
Hugh Dickins365e9c872005-10-29 18:16:18 -0700690 /* Update high watermark before we lower rss */
691 update_hiwater_rss(mm);
692
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 if (PageAnon(page)) {
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700694 swp_entry_t entry = { .val = page_private(page) };
Christoph Lameter06972122006-06-23 02:03:35 -0700695
696 if (PageSwapCache(page)) {
697 /*
698 * Store the swap location in the pte.
699 * See handle_pte_fault() ...
700 */
701 swap_duplicate(entry);
702 if (list_empty(&mm->mmlist)) {
703 spin_lock(&mmlist_lock);
704 if (list_empty(&mm->mmlist))
705 list_add(&mm->mmlist, &init_mm.mmlist);
706 spin_unlock(&mmlist_lock);
707 }
Christoph Lameter442c9132006-06-23 02:03:38 -0700708 dec_mm_counter(mm, anon_rss);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700709#ifdef CONFIG_MIGRATION
Christoph Lameter06972122006-06-23 02:03:35 -0700710 } else {
711 /*
712 * Store the pfn of the page in a special migration
713 * pte. do_swap_page() will wait until the migration
714 * pte is removed and then restart fault handling.
715 */
716 BUG_ON(!migration);
717 entry = make_migration_entry(page, pte_write(pteval));
Christoph Lameter04e62a22006-06-23 02:03:38 -0700718#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 }
720 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
721 BUG_ON(pte_file(*pte));
Hugh Dickins42946212005-10-29 18:16:05 -0700722 } else
Christoph Lameter04e62a22006-06-23 02:03:38 -0700723#ifdef CONFIG_MIGRATION
724 if (migration) {
725 /* Establish migration entry for a file page */
726 swp_entry_t entry;
727 entry = make_migration_entry(page, pte_write(pteval));
728 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
729 } else
730#endif
Hugh Dickins42946212005-10-29 18:16:05 -0700731 dec_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Christoph Lameter04e62a22006-06-23 02:03:38 -0700733
Nick Piggin7de6b802006-12-22 01:09:33 -0800734 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 page_cache_release(page);
736
737out_unmap:
Hugh Dickinsc0718802005-10-29 18:16:31 -0700738 pte_unmap_unlock(pte, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739out:
740 return ret;
741}
742
743/*
744 * objrmap doesn't work for nonlinear VMAs because the assumption that
745 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
746 * Consequently, given a particular page and its ->index, we cannot locate the
747 * ptes which are mapping that page without an exhaustive linear search.
748 *
749 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
750 * maps the file to which the target page belongs. The ->vm_private_data field
751 * holds the current cursor into that scan. Successive searches will circulate
752 * around the vma's virtual address space.
753 *
754 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
755 * more scanning pressure is placed against them as well. Eventually pages
756 * will become fully unmapped and are eligible for eviction.
757 *
758 * For very sparsely populated VMAs this is a little inefficient - chances are
759 * there there won't be many ptes located within the scan cluster. In this case
760 * maybe we could scan further - to the end of the pte page, perhaps.
761 */
762#define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
763#define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
764
765static void try_to_unmap_cluster(unsigned long cursor,
766 unsigned int *mapcount, struct vm_area_struct *vma)
767{
768 struct mm_struct *mm = vma->vm_mm;
769 pgd_t *pgd;
770 pud_t *pud;
771 pmd_t *pmd;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700772 pte_t *pte;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 pte_t pteval;
Hugh Dickinsc0718802005-10-29 18:16:31 -0700774 spinlock_t *ptl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 struct page *page;
776 unsigned long address;
777 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 address = (vma->vm_start + cursor) & CLUSTER_MASK;
780 end = address + CLUSTER_SIZE;
781 if (address < vma->vm_start)
782 address = vma->vm_start;
783 if (end > vma->vm_end)
784 end = vma->vm_end;
785
786 pgd = pgd_offset(mm, address);
787 if (!pgd_present(*pgd))
Hugh Dickinsc0718802005-10-29 18:16:31 -0700788 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
790 pud = pud_offset(pgd, address);
791 if (!pud_present(*pud))
Hugh Dickinsc0718802005-10-29 18:16:31 -0700792 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793
794 pmd = pmd_offset(pud, address);
795 if (!pmd_present(*pmd))
Hugh Dickinsc0718802005-10-29 18:16:31 -0700796 return;
797
798 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Hugh Dickins365e9c872005-10-29 18:16:18 -0700800 /* Update high watermark before we lower rss */
801 update_hiwater_rss(mm);
802
Hugh Dickinsc0718802005-10-29 18:16:31 -0700803 for (; address < end; pte++, address += PAGE_SIZE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 if (!pte_present(*pte))
805 continue;
Linus Torvalds6aab3412005-11-28 14:34:23 -0800806 page = vm_normal_page(vma, address, *pte);
807 BUG_ON(!page || PageAnon(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
809 if (ptep_clear_flush_young(vma, address, pte))
810 continue;
811
812 /* Nuke the page table entry. */
Ben Collinseca35132005-11-29 11:45:26 -0800813 flush_cache_page(vma, address, pte_pfn(*pte));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 pteval = ptep_clear_flush(vma, address, pte);
815
816 /* If nonlinear, store the file page offset in the pte. */
817 if (page->index != linear_page_index(vma, address))
818 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
819
820 /* Move the dirty bit to the physical page now the pte is gone. */
821 if (pte_dirty(pteval))
822 set_page_dirty(page);
823
Nick Piggin7de6b802006-12-22 01:09:33 -0800824 page_remove_rmap(page, vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 page_cache_release(page);
Hugh Dickins42946212005-10-29 18:16:05 -0700826 dec_mm_counter(mm, file_rss);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 (*mapcount)--;
828 }
Hugh Dickinsc0718802005-10-29 18:16:31 -0700829 pte_unmap_unlock(pte - 1, ptl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830}
831
Christoph Lameter73523492006-06-23 02:03:27 -0700832static int try_to_unmap_anon(struct page *page, int migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833{
834 struct anon_vma *anon_vma;
835 struct vm_area_struct *vma;
836 int ret = SWAP_AGAIN;
837
838 anon_vma = page_lock_anon_vma(page);
839 if (!anon_vma)
840 return ret;
841
842 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
Christoph Lameter73523492006-06-23 02:03:27 -0700843 ret = try_to_unmap_one(page, vma, migration);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (ret == SWAP_FAIL || !page_mapped(page))
845 break;
846 }
Oleg Nesterov34bbd702007-02-28 20:13:49 -0800847
848 page_unlock_anon_vma(anon_vma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 return ret;
850}
851
852/**
853 * try_to_unmap_file - unmap file page using the object-based rmap method
854 * @page: the page to unmap
855 *
856 * Find all the mappings of a page using the mapping pointer and the vma chains
857 * contained in the address_space struct it points to.
858 *
859 * This function is only called from try_to_unmap for object-based pages.
860 */
Christoph Lameter73523492006-06-23 02:03:27 -0700861static int try_to_unmap_file(struct page *page, int migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862{
863 struct address_space *mapping = page->mapping;
864 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
865 struct vm_area_struct *vma;
866 struct prio_tree_iter iter;
867 int ret = SWAP_AGAIN;
868 unsigned long cursor;
869 unsigned long max_nl_cursor = 0;
870 unsigned long max_nl_size = 0;
871 unsigned int mapcount;
872
873 spin_lock(&mapping->i_mmap_lock);
874 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
Christoph Lameter73523492006-06-23 02:03:27 -0700875 ret = try_to_unmap_one(page, vma, migration);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 if (ret == SWAP_FAIL || !page_mapped(page))
877 goto out;
878 }
879
880 if (list_empty(&mapping->i_mmap_nonlinear))
881 goto out;
882
883 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
884 shared.vm_set.list) {
Christoph Lametere6a15302006-06-25 05:46:49 -0700885 if ((vma->vm_flags & VM_LOCKED) && !migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 continue;
887 cursor = (unsigned long) vma->vm_private_data;
888 if (cursor > max_nl_cursor)
889 max_nl_cursor = cursor;
890 cursor = vma->vm_end - vma->vm_start;
891 if (cursor > max_nl_size)
892 max_nl_size = cursor;
893 }
894
895 if (max_nl_size == 0) { /* any nonlinears locked or reserved */
896 ret = SWAP_FAIL;
897 goto out;
898 }
899
900 /*
901 * We don't try to search for this page in the nonlinear vmas,
902 * and page_referenced wouldn't have found it anyway. Instead
903 * just walk the nonlinear vmas trying to age and unmap some.
904 * The mapcount of the page we came in with is irrelevant,
905 * but even so use it as a guide to how hard we should try?
906 */
907 mapcount = page_mapcount(page);
908 if (!mapcount)
909 goto out;
910 cond_resched_lock(&mapping->i_mmap_lock);
911
912 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
913 if (max_nl_cursor == 0)
914 max_nl_cursor = CLUSTER_SIZE;
915
916 do {
917 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
918 shared.vm_set.list) {
Christoph Lametere6a15302006-06-25 05:46:49 -0700919 if ((vma->vm_flags & VM_LOCKED) && !migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 continue;
921 cursor = (unsigned long) vma->vm_private_data;
Hugh Dickins839b9682005-09-03 15:54:43 -0700922 while ( cursor < max_nl_cursor &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 cursor < vma->vm_end - vma->vm_start) {
924 try_to_unmap_cluster(cursor, &mapcount, vma);
925 cursor += CLUSTER_SIZE;
926 vma->vm_private_data = (void *) cursor;
927 if ((int)mapcount <= 0)
928 goto out;
929 }
930 vma->vm_private_data = (void *) max_nl_cursor;
931 }
932 cond_resched_lock(&mapping->i_mmap_lock);
933 max_nl_cursor += CLUSTER_SIZE;
934 } while (max_nl_cursor <= max_nl_size);
935
936 /*
937 * Don't loop forever (perhaps all the remaining pages are
938 * in locked vmas). Reset cursor on all unreserved nonlinear
939 * vmas, now forgetting on which ones it had fallen behind.
940 */
Hugh Dickins101d2be2005-11-21 21:32:16 -0800941 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
942 vma->vm_private_data = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943out:
944 spin_unlock(&mapping->i_mmap_lock);
945 return ret;
946}
947
948/**
949 * try_to_unmap - try to remove all page table mappings to a page
950 * @page: the page to get unmapped
951 *
952 * Tries to remove all the page table entries which are mapping this
953 * page, used in the pageout path. Caller must hold the page lock.
954 * Return values are:
955 *
956 * SWAP_SUCCESS - we succeeded in removing all mappings
957 * SWAP_AGAIN - we missed a mapping, try again later
958 * SWAP_FAIL - the page is unswappable
959 */
Christoph Lameter73523492006-06-23 02:03:27 -0700960int try_to_unmap(struct page *page, int migration)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961{
962 int ret;
963
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 BUG_ON(!PageLocked(page));
965
966 if (PageAnon(page))
Christoph Lameter73523492006-06-23 02:03:27 -0700967 ret = try_to_unmap_anon(page, migration);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 else
Christoph Lameter73523492006-06-23 02:03:27 -0700969 ret = try_to_unmap_file(page, migration);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970
971 if (!page_mapped(page))
972 ret = SWAP_SUCCESS;
973 return ret;
974}
Nikita Danilov81b40822005-05-01 08:58:36 -0700975