Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/rmap.c - physical to virtual reverse mappings | |
3 | * | |
4 | * Copyright 2001, Rik van Riel <riel@conectiva.com.br> | |
5 | * Released under the General Public License (GPL). | |
6 | * | |
7 | * Simple, low overhead reverse mapping scheme. | |
8 | * Please try to keep this thing as modular as possible. | |
9 | * | |
10 | * Provides methods for unmapping each kind of mapped page: | |
11 | * the anon methods track anonymous pages, and | |
12 | * the file methods track pages belonging to an inode. | |
13 | * | |
14 | * Original design by Rik van Riel <riel@conectiva.com.br> 2001 | |
15 | * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004 | |
16 | * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004 | |
17 | * Contributions by Hugh Dickins <hugh@veritas.com> 2003, 2004 | |
18 | */ | |
19 | ||
20 | /* | |
21 | * Lock ordering in mm: | |
22 | * | |
1b1dcc1b | 23 | * inode->i_mutex (while writing or truncating, not reading or faulting) |
82591e6e NP |
24 | * inode->i_alloc_sem (vmtruncate_range) |
25 | * mm->mmap_sem | |
26 | * page->flags PG_locked (lock_page) | |
27 | * mapping->i_mmap_lock | |
28 | * anon_vma->lock | |
29 | * mm->page_table_lock or pte_lock | |
30 | * zone->lru_lock (in mark_page_accessed, isolate_lru_page) | |
31 | * swap_lock (in swap_duplicate, swap_info_get) | |
32 | * mmlist_lock (in mmput, drain_mmlist and others) | |
33 | * mapping->private_lock (in __set_page_dirty_buffers) | |
34 | * inode_lock (in set_page_dirty's __mark_inode_dirty) | |
35 | * sb_lock (within inode_lock in fs/fs-writeback.c) | |
36 | * mapping->tree_lock (widely used, in set_page_dirty, | |
37 | * in arch-dependent flush_dcache_mmap_lock, | |
38 | * within inode_lock in __sync_single_inode) | |
1da177e4 LT |
39 | */ |
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> | |
a48d07af | 49 | #include <linux/module.h> |
7de6b805 | 50 | #include <linux/kallsyms.h> |
8a9f3ccd | 51 | #include <linux/memcontrol.h> |
cddb8a5c | 52 | #include <linux/mmu_notifier.h> |
1da177e4 LT |
53 | |
54 | #include <asm/tlbflush.h> | |
55 | ||
fcc234f8 | 56 | struct kmem_cache *anon_vma_cachep; |
1da177e4 | 57 | |
d9d332e0 LT |
58 | /** |
59 | * anon_vma_prepare - attach an anon_vma to a memory region | |
60 | * @vma: the memory region in question | |
61 | * | |
62 | * This makes sure the memory mapping described by 'vma' has | |
63 | * an 'anon_vma' attached to it, so that we can associate the | |
64 | * anonymous pages mapped into it with that anon_vma. | |
65 | * | |
66 | * The common case will be that we already have one, but if | |
67 | * if not we either need to find an adjacent mapping that we | |
68 | * can re-use the anon_vma from (very common when the only | |
69 | * reason for splitting a vma has been mprotect()), or we | |
70 | * allocate a new one. | |
71 | * | |
72 | * Anon-vma allocations are very subtle, because we may have | |
73 | * optimistically looked up an anon_vma in page_lock_anon_vma() | |
74 | * and that may actually touch the spinlock even in the newly | |
75 | * allocated vma (it depends on RCU to make sure that the | |
76 | * anon_vma isn't actually destroyed). | |
77 | * | |
78 | * As a result, we need to do proper anon_vma locking even | |
79 | * for the new allocation. At the same time, we do not want | |
80 | * to do any locking for the common case of already having | |
81 | * an anon_vma. | |
82 | * | |
83 | * This must be called with the mmap_sem held for reading. | |
84 | */ | |
1da177e4 LT |
85 | int anon_vma_prepare(struct vm_area_struct *vma) |
86 | { | |
87 | struct anon_vma *anon_vma = vma->anon_vma; | |
88 | ||
89 | might_sleep(); | |
90 | if (unlikely(!anon_vma)) { | |
91 | struct mm_struct *mm = vma->vm_mm; | |
d9d332e0 | 92 | struct anon_vma *allocated; |
1da177e4 LT |
93 | |
94 | anon_vma = find_mergeable_anon_vma(vma); | |
d9d332e0 LT |
95 | allocated = NULL; |
96 | if (!anon_vma) { | |
1da177e4 LT |
97 | anon_vma = anon_vma_alloc(); |
98 | if (unlikely(!anon_vma)) | |
99 | return -ENOMEM; | |
100 | allocated = anon_vma; | |
1da177e4 | 101 | } |
d9d332e0 | 102 | spin_lock(&anon_vma->lock); |
1da177e4 LT |
103 | |
104 | /* page_table_lock to protect against threads */ | |
105 | spin_lock(&mm->page_table_lock); | |
106 | if (likely(!vma->anon_vma)) { | |
107 | vma->anon_vma = anon_vma; | |
0697212a | 108 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
109 | allocated = NULL; |
110 | } | |
111 | spin_unlock(&mm->page_table_lock); | |
112 | ||
d9d332e0 | 113 | spin_unlock(&anon_vma->lock); |
1da177e4 LT |
114 | if (unlikely(allocated)) |
115 | anon_vma_free(allocated); | |
116 | } | |
117 | return 0; | |
118 | } | |
119 | ||
120 | void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next) | |
121 | { | |
122 | BUG_ON(vma->anon_vma != next->anon_vma); | |
123 | list_del(&next->anon_vma_node); | |
124 | } | |
125 | ||
126 | void __anon_vma_link(struct vm_area_struct *vma) | |
127 | { | |
128 | struct anon_vma *anon_vma = vma->anon_vma; | |
129 | ||
30acbaba | 130 | if (anon_vma) |
0697212a | 131 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
132 | } |
133 | ||
134 | void anon_vma_link(struct vm_area_struct *vma) | |
135 | { | |
136 | struct anon_vma *anon_vma = vma->anon_vma; | |
137 | ||
138 | if (anon_vma) { | |
139 | spin_lock(&anon_vma->lock); | |
0697212a | 140 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
141 | spin_unlock(&anon_vma->lock); |
142 | } | |
143 | } | |
144 | ||
145 | void anon_vma_unlink(struct vm_area_struct *vma) | |
146 | { | |
147 | struct anon_vma *anon_vma = vma->anon_vma; | |
148 | int empty; | |
149 | ||
150 | if (!anon_vma) | |
151 | return; | |
152 | ||
153 | spin_lock(&anon_vma->lock); | |
1da177e4 LT |
154 | list_del(&vma->anon_vma_node); |
155 | ||
156 | /* We must garbage collect the anon_vma if it's empty */ | |
157 | empty = list_empty(&anon_vma->head); | |
158 | spin_unlock(&anon_vma->lock); | |
159 | ||
160 | if (empty) | |
161 | anon_vma_free(anon_vma); | |
162 | } | |
163 | ||
51cc5068 | 164 | static void anon_vma_ctor(void *data) |
1da177e4 | 165 | { |
a35afb83 | 166 | struct anon_vma *anon_vma = data; |
1da177e4 | 167 | |
a35afb83 CL |
168 | spin_lock_init(&anon_vma->lock); |
169 | INIT_LIST_HEAD(&anon_vma->head); | |
1da177e4 LT |
170 | } |
171 | ||
172 | void __init anon_vma_init(void) | |
173 | { | |
174 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
20c2df83 | 175 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
1da177e4 LT |
176 | } |
177 | ||
178 | /* | |
179 | * Getting a lock on a stable anon_vma from a page off the LRU is | |
180 | * tricky: page_lock_anon_vma rely on RCU to guard against the races. | |
181 | */ | |
182 | static struct anon_vma *page_lock_anon_vma(struct page *page) | |
183 | { | |
34bbd704 | 184 | struct anon_vma *anon_vma; |
1da177e4 LT |
185 | unsigned long anon_mapping; |
186 | ||
187 | rcu_read_lock(); | |
188 | anon_mapping = (unsigned long) page->mapping; | |
189 | if (!(anon_mapping & PAGE_MAPPING_ANON)) | |
190 | goto out; | |
191 | if (!page_mapped(page)) | |
192 | goto out; | |
193 | ||
194 | anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON); | |
195 | spin_lock(&anon_vma->lock); | |
34bbd704 | 196 | return anon_vma; |
1da177e4 LT |
197 | out: |
198 | rcu_read_unlock(); | |
34bbd704 ON |
199 | return NULL; |
200 | } | |
201 | ||
202 | static void page_unlock_anon_vma(struct anon_vma *anon_vma) | |
203 | { | |
204 | spin_unlock(&anon_vma->lock); | |
205 | rcu_read_unlock(); | |
1da177e4 LT |
206 | } |
207 | ||
208 | /* | |
3ad33b24 LS |
209 | * At what user virtual address is page expected in @vma? |
210 | * Returns virtual address or -EFAULT if page's index/offset is not | |
211 | * within the range mapped the @vma. | |
1da177e4 LT |
212 | */ |
213 | static inline unsigned long | |
214 | vma_address(struct page *page, struct vm_area_struct *vma) | |
215 | { | |
216 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
217 | unsigned long address; | |
218 | ||
219 | address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); | |
220 | if (unlikely(address < vma->vm_start || address >= vma->vm_end)) { | |
3ad33b24 | 221 | /* page should be within @vma mapping range */ |
1da177e4 LT |
222 | return -EFAULT; |
223 | } | |
224 | return address; | |
225 | } | |
226 | ||
227 | /* | |
228 | * At what user virtual address is page expected in vma? checking that the | |
ee498ed7 | 229 | * page matches the vma: currently only used on anon pages, by unuse_vma; |
1da177e4 LT |
230 | */ |
231 | unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma) | |
232 | { | |
233 | if (PageAnon(page)) { | |
234 | if ((void *)vma->anon_vma != | |
235 | (void *)page->mapping - PAGE_MAPPING_ANON) | |
236 | return -EFAULT; | |
237 | } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) { | |
ee498ed7 HD |
238 | if (!vma->vm_file || |
239 | vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
240 | return -EFAULT; |
241 | } else | |
242 | return -EFAULT; | |
243 | return vma_address(page, vma); | |
244 | } | |
245 | ||
81b4082d ND |
246 | /* |
247 | * Check that @page is mapped at @address into @mm. | |
248 | * | |
479db0bf NP |
249 | * If @sync is false, page_check_address may perform a racy check to avoid |
250 | * the page table lock when the pte is not present (helpful when reclaiming | |
251 | * highly shared pages). | |
252 | * | |
b8072f09 | 253 | * On success returns with pte mapped and locked. |
81b4082d | 254 | */ |
ceffc078 | 255 | pte_t *page_check_address(struct page *page, struct mm_struct *mm, |
479db0bf | 256 | unsigned long address, spinlock_t **ptlp, int sync) |
81b4082d ND |
257 | { |
258 | pgd_t *pgd; | |
259 | pud_t *pud; | |
260 | pmd_t *pmd; | |
261 | pte_t *pte; | |
c0718806 | 262 | spinlock_t *ptl; |
81b4082d | 263 | |
81b4082d | 264 | pgd = pgd_offset(mm, address); |
c0718806 HD |
265 | if (!pgd_present(*pgd)) |
266 | return NULL; | |
267 | ||
268 | pud = pud_offset(pgd, address); | |
269 | if (!pud_present(*pud)) | |
270 | return NULL; | |
271 | ||
272 | pmd = pmd_offset(pud, address); | |
273 | if (!pmd_present(*pmd)) | |
274 | return NULL; | |
275 | ||
276 | pte = pte_offset_map(pmd, address); | |
277 | /* Make a quick check before getting the lock */ | |
479db0bf | 278 | if (!sync && !pte_present(*pte)) { |
c0718806 HD |
279 | pte_unmap(pte); |
280 | return NULL; | |
281 | } | |
282 | ||
4c21e2f2 | 283 | ptl = pte_lockptr(mm, pmd); |
c0718806 HD |
284 | spin_lock(ptl); |
285 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
286 | *ptlp = ptl; | |
287 | return pte; | |
81b4082d | 288 | } |
c0718806 HD |
289 | pte_unmap_unlock(pte, ptl); |
290 | return NULL; | |
81b4082d ND |
291 | } |
292 | ||
1da177e4 LT |
293 | /* |
294 | * Subfunctions of page_referenced: page_referenced_one called | |
295 | * repeatedly from either page_referenced_anon or page_referenced_file. | |
296 | */ | |
297 | static int page_referenced_one(struct page *page, | |
f7b7fd8f | 298 | struct vm_area_struct *vma, unsigned int *mapcount) |
1da177e4 LT |
299 | { |
300 | struct mm_struct *mm = vma->vm_mm; | |
301 | unsigned long address; | |
1da177e4 | 302 | pte_t *pte; |
c0718806 | 303 | spinlock_t *ptl; |
1da177e4 LT |
304 | int referenced = 0; |
305 | ||
1da177e4 LT |
306 | address = vma_address(page, vma); |
307 | if (address == -EFAULT) | |
308 | goto out; | |
309 | ||
479db0bf | 310 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 HD |
311 | if (!pte) |
312 | goto out; | |
1da177e4 | 313 | |
5a9bbdcd HD |
314 | if (vma->vm_flags & VM_LOCKED) { |
315 | referenced++; | |
316 | *mapcount = 1; /* break early from loop */ | |
cddb8a5c | 317 | } else if (ptep_clear_flush_young_notify(vma, address, pte)) |
c0718806 | 318 | referenced++; |
1da177e4 | 319 | |
c0718806 HD |
320 | /* Pretend the page is referenced if the task has the |
321 | swap token and is in the middle of a page fault. */ | |
f7b7fd8f | 322 | if (mm != current->mm && has_swap_token(mm) && |
c0718806 HD |
323 | rwsem_is_locked(&mm->mmap_sem)) |
324 | referenced++; | |
325 | ||
326 | (*mapcount)--; | |
327 | pte_unmap_unlock(pte, ptl); | |
1da177e4 LT |
328 | out: |
329 | return referenced; | |
330 | } | |
331 | ||
bed7161a BS |
332 | static int page_referenced_anon(struct page *page, |
333 | struct mem_cgroup *mem_cont) | |
1da177e4 LT |
334 | { |
335 | unsigned int mapcount; | |
336 | struct anon_vma *anon_vma; | |
337 | struct vm_area_struct *vma; | |
338 | int referenced = 0; | |
339 | ||
340 | anon_vma = page_lock_anon_vma(page); | |
341 | if (!anon_vma) | |
342 | return referenced; | |
343 | ||
344 | mapcount = page_mapcount(page); | |
345 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { | |
bed7161a BS |
346 | /* |
347 | * If we are reclaiming on behalf of a cgroup, skip | |
348 | * counting on behalf of references from different | |
349 | * cgroups | |
350 | */ | |
bd845e38 | 351 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 352 | continue; |
f7b7fd8f | 353 | referenced += page_referenced_one(page, vma, &mapcount); |
1da177e4 LT |
354 | if (!mapcount) |
355 | break; | |
356 | } | |
34bbd704 ON |
357 | |
358 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
359 | return referenced; |
360 | } | |
361 | ||
362 | /** | |
363 | * page_referenced_file - referenced check for object-based rmap | |
364 | * @page: the page we're checking references on. | |
43d8eac4 | 365 | * @mem_cont: target memory controller |
1da177e4 LT |
366 | * |
367 | * For an object-based mapped page, find all the places it is mapped and | |
368 | * check/clear the referenced flag. This is done by following the page->mapping | |
369 | * pointer, then walking the chain of vmas it holds. It returns the number | |
370 | * of references it found. | |
371 | * | |
372 | * This function is only called from page_referenced for object-based pages. | |
373 | */ | |
bed7161a BS |
374 | static int page_referenced_file(struct page *page, |
375 | struct mem_cgroup *mem_cont) | |
1da177e4 LT |
376 | { |
377 | unsigned int mapcount; | |
378 | struct address_space *mapping = page->mapping; | |
379 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
380 | struct vm_area_struct *vma; | |
381 | struct prio_tree_iter iter; | |
382 | int referenced = 0; | |
383 | ||
384 | /* | |
385 | * The caller's checks on page->mapping and !PageAnon have made | |
386 | * sure that this is a file page: the check for page->mapping | |
387 | * excludes the case just before it gets set on an anon page. | |
388 | */ | |
389 | BUG_ON(PageAnon(page)); | |
390 | ||
391 | /* | |
392 | * The page lock not only makes sure that page->mapping cannot | |
393 | * suddenly be NULLified by truncation, it makes sure that the | |
394 | * structure at mapping cannot be freed and reused yet, | |
395 | * so we can safely take mapping->i_mmap_lock. | |
396 | */ | |
397 | BUG_ON(!PageLocked(page)); | |
398 | ||
399 | spin_lock(&mapping->i_mmap_lock); | |
400 | ||
401 | /* | |
402 | * i_mmap_lock does not stabilize mapcount at all, but mapcount | |
403 | * is more likely to be accurate if we note it after spinning. | |
404 | */ | |
405 | mapcount = page_mapcount(page); | |
406 | ||
407 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
bed7161a BS |
408 | /* |
409 | * If we are reclaiming on behalf of a cgroup, skip | |
410 | * counting on behalf of references from different | |
411 | * cgroups | |
412 | */ | |
bd845e38 | 413 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 414 | continue; |
1da177e4 LT |
415 | if ((vma->vm_flags & (VM_LOCKED|VM_MAYSHARE)) |
416 | == (VM_LOCKED|VM_MAYSHARE)) { | |
417 | referenced++; | |
418 | break; | |
419 | } | |
f7b7fd8f | 420 | referenced += page_referenced_one(page, vma, &mapcount); |
1da177e4 LT |
421 | if (!mapcount) |
422 | break; | |
423 | } | |
424 | ||
425 | spin_unlock(&mapping->i_mmap_lock); | |
426 | return referenced; | |
427 | } | |
428 | ||
429 | /** | |
430 | * page_referenced - test if the page was referenced | |
431 | * @page: the page to test | |
432 | * @is_locked: caller holds lock on the page | |
43d8eac4 | 433 | * @mem_cont: target memory controller |
1da177e4 LT |
434 | * |
435 | * Quick test_and_clear_referenced for all mappings to a page, | |
436 | * returns the number of ptes which referenced the page. | |
437 | */ | |
bed7161a BS |
438 | int page_referenced(struct page *page, int is_locked, |
439 | struct mem_cgroup *mem_cont) | |
1da177e4 LT |
440 | { |
441 | int referenced = 0; | |
442 | ||
1da177e4 LT |
443 | if (TestClearPageReferenced(page)) |
444 | referenced++; | |
445 | ||
446 | if (page_mapped(page) && page->mapping) { | |
447 | if (PageAnon(page)) | |
bed7161a | 448 | referenced += page_referenced_anon(page, mem_cont); |
1da177e4 | 449 | else if (is_locked) |
bed7161a | 450 | referenced += page_referenced_file(page, mem_cont); |
529ae9aa | 451 | else if (!trylock_page(page)) |
1da177e4 LT |
452 | referenced++; |
453 | else { | |
454 | if (page->mapping) | |
bed7161a BS |
455 | referenced += |
456 | page_referenced_file(page, mem_cont); | |
1da177e4 LT |
457 | unlock_page(page); |
458 | } | |
459 | } | |
5b7baf05 CB |
460 | |
461 | if (page_test_and_clear_young(page)) | |
462 | referenced++; | |
463 | ||
1da177e4 LT |
464 | return referenced; |
465 | } | |
466 | ||
d08b3851 PZ |
467 | static int page_mkclean_one(struct page *page, struct vm_area_struct *vma) |
468 | { | |
469 | struct mm_struct *mm = vma->vm_mm; | |
470 | unsigned long address; | |
c2fda5fe | 471 | pte_t *pte; |
d08b3851 PZ |
472 | spinlock_t *ptl; |
473 | int ret = 0; | |
474 | ||
475 | address = vma_address(page, vma); | |
476 | if (address == -EFAULT) | |
477 | goto out; | |
478 | ||
479db0bf | 479 | pte = page_check_address(page, mm, address, &ptl, 1); |
d08b3851 PZ |
480 | if (!pte) |
481 | goto out; | |
482 | ||
c2fda5fe PZ |
483 | if (pte_dirty(*pte) || pte_write(*pte)) { |
484 | pte_t entry; | |
d08b3851 | 485 | |
c2fda5fe | 486 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 487 | entry = ptep_clear_flush_notify(vma, address, pte); |
c2fda5fe PZ |
488 | entry = pte_wrprotect(entry); |
489 | entry = pte_mkclean(entry); | |
d6e88e67 | 490 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
491 | ret = 1; |
492 | } | |
d08b3851 | 493 | |
d08b3851 PZ |
494 | pte_unmap_unlock(pte, ptl); |
495 | out: | |
496 | return ret; | |
497 | } | |
498 | ||
499 | static int page_mkclean_file(struct address_space *mapping, struct page *page) | |
500 | { | |
501 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
502 | struct vm_area_struct *vma; | |
503 | struct prio_tree_iter iter; | |
504 | int ret = 0; | |
505 | ||
506 | BUG_ON(PageAnon(page)); | |
507 | ||
508 | spin_lock(&mapping->i_mmap_lock); | |
509 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
510 | if (vma->vm_flags & VM_SHARED) | |
511 | ret += page_mkclean_one(page, vma); | |
512 | } | |
513 | spin_unlock(&mapping->i_mmap_lock); | |
514 | return ret; | |
515 | } | |
516 | ||
517 | int page_mkclean(struct page *page) | |
518 | { | |
519 | int ret = 0; | |
520 | ||
521 | BUG_ON(!PageLocked(page)); | |
522 | ||
523 | if (page_mapped(page)) { | |
524 | struct address_space *mapping = page_mapping(page); | |
ce7e9fae | 525 | if (mapping) { |
d08b3851 | 526 | ret = page_mkclean_file(mapping, page); |
ce7e9fae CB |
527 | if (page_test_dirty(page)) { |
528 | page_clear_dirty(page); | |
529 | ret = 1; | |
530 | } | |
6c210482 | 531 | } |
d08b3851 PZ |
532 | } |
533 | ||
534 | return ret; | |
535 | } | |
60b59bea | 536 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 537 | |
9617d95e | 538 | /** |
43d8eac4 | 539 | * __page_set_anon_rmap - setup new anonymous rmap |
9617d95e NP |
540 | * @page: the page to add the mapping to |
541 | * @vma: the vm area in which the mapping is added | |
542 | * @address: the user virtual address mapped | |
543 | */ | |
544 | static void __page_set_anon_rmap(struct page *page, | |
545 | struct vm_area_struct *vma, unsigned long address) | |
546 | { | |
547 | struct anon_vma *anon_vma = vma->anon_vma; | |
548 | ||
549 | BUG_ON(!anon_vma); | |
550 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
551 | page->mapping = (struct address_space *) anon_vma; | |
552 | ||
553 | page->index = linear_page_index(vma, address); | |
554 | ||
a74609fa NP |
555 | /* |
556 | * nr_mapped state can be updated without turning off | |
557 | * interrupts because it is not modified via interrupt. | |
558 | */ | |
f3dbd344 | 559 | __inc_zone_page_state(page, NR_ANON_PAGES); |
9617d95e NP |
560 | } |
561 | ||
c97a9e10 | 562 | /** |
43d8eac4 | 563 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
564 | * @page: the page to add the mapping to |
565 | * @vma: the vm area in which the mapping is added | |
566 | * @address: the user virtual address mapped | |
567 | */ | |
568 | static void __page_check_anon_rmap(struct page *page, | |
569 | struct vm_area_struct *vma, unsigned long address) | |
570 | { | |
571 | #ifdef CONFIG_DEBUG_VM | |
572 | /* | |
573 | * The page's anon-rmap details (mapping and index) are guaranteed to | |
574 | * be set up correctly at this point. | |
575 | * | |
576 | * We have exclusion against page_add_anon_rmap because the caller | |
577 | * always holds the page locked, except if called from page_dup_rmap, | |
578 | * in which case the page is already known to be setup. | |
579 | * | |
580 | * We have exclusion against page_add_new_anon_rmap because those pages | |
581 | * are initially only visible via the pagetables, and the pte is locked | |
582 | * over the call to page_add_new_anon_rmap. | |
583 | */ | |
584 | struct anon_vma *anon_vma = vma->anon_vma; | |
585 | anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON; | |
586 | BUG_ON(page->mapping != (struct address_space *)anon_vma); | |
587 | BUG_ON(page->index != linear_page_index(vma, address)); | |
588 | #endif | |
589 | } | |
590 | ||
1da177e4 LT |
591 | /** |
592 | * page_add_anon_rmap - add pte mapping to an anonymous page | |
593 | * @page: the page to add the mapping to | |
594 | * @vma: the vm area in which the mapping is added | |
595 | * @address: the user virtual address mapped | |
596 | * | |
c97a9e10 | 597 | * The caller needs to hold the pte lock and the page must be locked. |
1da177e4 LT |
598 | */ |
599 | void page_add_anon_rmap(struct page *page, | |
600 | struct vm_area_struct *vma, unsigned long address) | |
601 | { | |
c97a9e10 NP |
602 | VM_BUG_ON(!PageLocked(page)); |
603 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
9617d95e NP |
604 | if (atomic_inc_and_test(&page->_mapcount)) |
605 | __page_set_anon_rmap(page, vma, address); | |
69029cd5 | 606 | else |
c97a9e10 | 607 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
608 | } |
609 | ||
43d8eac4 | 610 | /** |
9617d95e NP |
611 | * page_add_new_anon_rmap - add pte mapping to a new anonymous page |
612 | * @page: the page to add the mapping to | |
613 | * @vma: the vm area in which the mapping is added | |
614 | * @address: the user virtual address mapped | |
615 | * | |
616 | * Same as page_add_anon_rmap but must only be called on *new* pages. | |
617 | * This means the inc-and-test can be bypassed. | |
c97a9e10 | 618 | * Page does not have to be locked. |
9617d95e NP |
619 | */ |
620 | void page_add_new_anon_rmap(struct page *page, | |
621 | struct vm_area_struct *vma, unsigned long address) | |
622 | { | |
c97a9e10 | 623 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
9617d95e NP |
624 | atomic_set(&page->_mapcount, 0); /* elevate count by 1 (starts at -1) */ |
625 | __page_set_anon_rmap(page, vma, address); | |
626 | } | |
627 | ||
1da177e4 LT |
628 | /** |
629 | * page_add_file_rmap - add pte mapping to a file page | |
630 | * @page: the page to add the mapping to | |
631 | * | |
b8072f09 | 632 | * The caller needs to hold the pte lock. |
1da177e4 LT |
633 | */ |
634 | void page_add_file_rmap(struct page *page) | |
635 | { | |
1da177e4 | 636 | if (atomic_inc_and_test(&page->_mapcount)) |
65ba55f5 | 637 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
1da177e4 LT |
638 | } |
639 | ||
c97a9e10 NP |
640 | #ifdef CONFIG_DEBUG_VM |
641 | /** | |
642 | * page_dup_rmap - duplicate pte mapping to a page | |
643 | * @page: the page to add the mapping to | |
43d8eac4 RD |
644 | * @vma: the vm area being duplicated |
645 | * @address: the user virtual address mapped | |
c97a9e10 NP |
646 | * |
647 | * For copy_page_range only: minimal extract from page_add_file_rmap / | |
648 | * page_add_anon_rmap, avoiding unnecessary tests (already checked) so it's | |
649 | * quicker. | |
650 | * | |
651 | * The caller needs to hold the pte lock. | |
652 | */ | |
653 | void page_dup_rmap(struct page *page, struct vm_area_struct *vma, unsigned long address) | |
654 | { | |
655 | BUG_ON(page_mapcount(page) == 0); | |
656 | if (PageAnon(page)) | |
657 | __page_check_anon_rmap(page, vma, address); | |
658 | atomic_inc(&page->_mapcount); | |
659 | } | |
660 | #endif | |
661 | ||
1da177e4 LT |
662 | /** |
663 | * page_remove_rmap - take down pte mapping from a page | |
664 | * @page: page to remove mapping from | |
43d8eac4 | 665 | * @vma: the vm area in which the mapping is removed |
1da177e4 | 666 | * |
b8072f09 | 667 | * The caller needs to hold the pte lock. |
1da177e4 | 668 | */ |
7de6b805 | 669 | void page_remove_rmap(struct page *page, struct vm_area_struct *vma) |
1da177e4 | 670 | { |
1da177e4 | 671 | if (atomic_add_negative(-1, &page->_mapcount)) { |
b7ab795b | 672 | if (unlikely(page_mapcount(page) < 0)) { |
ef2bf0dc | 673 | printk (KERN_EMERG "Eeek! page_mapcount(page) went negative! (%d)\n", page_mapcount(page)); |
7de6b805 | 674 | printk (KERN_EMERG " page pfn = %lx\n", page_to_pfn(page)); |
ef2bf0dc DJ |
675 | printk (KERN_EMERG " page->flags = %lx\n", page->flags); |
676 | printk (KERN_EMERG " page->count = %x\n", page_count(page)); | |
677 | printk (KERN_EMERG " page->mapping = %p\n", page->mapping); | |
7de6b805 | 678 | print_symbol (KERN_EMERG " vma->vm_ops = %s\n", (unsigned long)vma->vm_ops); |
54cb8821 | 679 | if (vma->vm_ops) { |
54cb8821 NP |
680 | print_symbol (KERN_EMERG " vma->vm_ops->fault = %s\n", (unsigned long)vma->vm_ops->fault); |
681 | } | |
7de6b805 NP |
682 | if (vma->vm_file && vma->vm_file->f_op) |
683 | print_symbol (KERN_EMERG " vma->vm_file->f_op->mmap = %s\n", (unsigned long)vma->vm_file->f_op->mmap); | |
b16bc64d | 684 | BUG(); |
ef2bf0dc | 685 | } |
b16bc64d | 686 | |
1da177e4 | 687 | /* |
16f8c5b2 HD |
688 | * Now that the last pte has gone, s390 must transfer dirty |
689 | * flag from storage key to struct page. We can usually skip | |
690 | * this if the page is anon, so about to be freed; but perhaps | |
691 | * not if it's in swapcache - there might be another pte slot | |
692 | * containing the swap entry, but page not yet written to swap. | |
1da177e4 | 693 | */ |
a4b526b3 MS |
694 | if ((!PageAnon(page) || PageSwapCache(page)) && |
695 | page_test_dirty(page)) { | |
6c210482 | 696 | page_clear_dirty(page); |
1da177e4 | 697 | set_page_dirty(page); |
6c210482 | 698 | } |
8a9f3ccd | 699 | |
16f8c5b2 | 700 | mem_cgroup_uncharge_page(page); |
f3dbd344 | 701 | __dec_zone_page_state(page, |
16f8c5b2 HD |
702 | PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED); |
703 | /* | |
704 | * It would be tidy to reset the PageAnon mapping here, | |
705 | * but that might overwrite a racing page_add_anon_rmap | |
706 | * which increments mapcount after us but sets mapping | |
707 | * before us: so leave the reset to free_hot_cold_page, | |
708 | * and remember that it's only reliable while mapped. | |
709 | * Leaving it set also helps swapoff to reinstate ptes | |
710 | * faster for those pages still in swapcache. | |
711 | */ | |
1da177e4 LT |
712 | } |
713 | } | |
714 | ||
715 | /* | |
716 | * Subfunctions of try_to_unmap: try_to_unmap_one called | |
717 | * repeatedly from either try_to_unmap_anon or try_to_unmap_file. | |
718 | */ | |
a48d07af | 719 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
7352349a | 720 | int migration) |
1da177e4 LT |
721 | { |
722 | struct mm_struct *mm = vma->vm_mm; | |
723 | unsigned long address; | |
1da177e4 LT |
724 | pte_t *pte; |
725 | pte_t pteval; | |
c0718806 | 726 | spinlock_t *ptl; |
1da177e4 LT |
727 | int ret = SWAP_AGAIN; |
728 | ||
1da177e4 LT |
729 | address = vma_address(page, vma); |
730 | if (address == -EFAULT) | |
731 | goto out; | |
732 | ||
479db0bf | 733 | pte = page_check_address(page, mm, address, &ptl, 0); |
c0718806 | 734 | if (!pte) |
81b4082d | 735 | goto out; |
1da177e4 LT |
736 | |
737 | /* | |
738 | * If the page is mlock()d, we cannot swap it out. | |
739 | * If it's recently referenced (perhaps page_referenced | |
740 | * skipped over this mm) then we should reactivate it. | |
741 | */ | |
e6a1530d | 742 | if (!migration && ((vma->vm_flags & VM_LOCKED) || |
cddb8a5c | 743 | (ptep_clear_flush_young_notify(vma, address, pte)))) { |
1da177e4 LT |
744 | ret = SWAP_FAIL; |
745 | goto out_unmap; | |
746 | } | |
747 | ||
1da177e4 LT |
748 | /* Nuke the page table entry. */ |
749 | flush_cache_page(vma, address, page_to_pfn(page)); | |
cddb8a5c | 750 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
751 | |
752 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
753 | if (pte_dirty(pteval)) | |
754 | set_page_dirty(page); | |
755 | ||
365e9c87 HD |
756 | /* Update high watermark before we lower rss */ |
757 | update_hiwater_rss(mm); | |
758 | ||
1da177e4 | 759 | if (PageAnon(page)) { |
4c21e2f2 | 760 | swp_entry_t entry = { .val = page_private(page) }; |
0697212a CL |
761 | |
762 | if (PageSwapCache(page)) { | |
763 | /* | |
764 | * Store the swap location in the pte. | |
765 | * See handle_pte_fault() ... | |
766 | */ | |
767 | swap_duplicate(entry); | |
768 | if (list_empty(&mm->mmlist)) { | |
769 | spin_lock(&mmlist_lock); | |
770 | if (list_empty(&mm->mmlist)) | |
771 | list_add(&mm->mmlist, &init_mm.mmlist); | |
772 | spin_unlock(&mmlist_lock); | |
773 | } | |
442c9137 | 774 | dec_mm_counter(mm, anon_rss); |
04e62a29 | 775 | #ifdef CONFIG_MIGRATION |
0697212a CL |
776 | } else { |
777 | /* | |
778 | * Store the pfn of the page in a special migration | |
779 | * pte. do_swap_page() will wait until the migration | |
780 | * pte is removed and then restart fault handling. | |
781 | */ | |
782 | BUG_ON(!migration); | |
783 | entry = make_migration_entry(page, pte_write(pteval)); | |
04e62a29 | 784 | #endif |
1da177e4 LT |
785 | } |
786 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
787 | BUG_ON(pte_file(*pte)); | |
4294621f | 788 | } else |
04e62a29 CL |
789 | #ifdef CONFIG_MIGRATION |
790 | if (migration) { | |
791 | /* Establish migration entry for a file page */ | |
792 | swp_entry_t entry; | |
793 | entry = make_migration_entry(page, pte_write(pteval)); | |
794 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
795 | } else | |
796 | #endif | |
4294621f | 797 | dec_mm_counter(mm, file_rss); |
1da177e4 | 798 | |
04e62a29 | 799 | |
7de6b805 | 800 | page_remove_rmap(page, vma); |
1da177e4 LT |
801 | page_cache_release(page); |
802 | ||
803 | out_unmap: | |
c0718806 | 804 | pte_unmap_unlock(pte, ptl); |
1da177e4 LT |
805 | out: |
806 | return ret; | |
807 | } | |
808 | ||
809 | /* | |
810 | * objrmap doesn't work for nonlinear VMAs because the assumption that | |
811 | * offset-into-file correlates with offset-into-virtual-addresses does not hold. | |
812 | * Consequently, given a particular page and its ->index, we cannot locate the | |
813 | * ptes which are mapping that page without an exhaustive linear search. | |
814 | * | |
815 | * So what this code does is a mini "virtual scan" of each nonlinear VMA which | |
816 | * maps the file to which the target page belongs. The ->vm_private_data field | |
817 | * holds the current cursor into that scan. Successive searches will circulate | |
818 | * around the vma's virtual address space. | |
819 | * | |
820 | * So as more replacement pressure is applied to the pages in a nonlinear VMA, | |
821 | * more scanning pressure is placed against them as well. Eventually pages | |
822 | * will become fully unmapped and are eligible for eviction. | |
823 | * | |
824 | * For very sparsely populated VMAs this is a little inefficient - chances are | |
825 | * there there won't be many ptes located within the scan cluster. In this case | |
826 | * maybe we could scan further - to the end of the pte page, perhaps. | |
827 | */ | |
828 | #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE) | |
829 | #define CLUSTER_MASK (~(CLUSTER_SIZE - 1)) | |
830 | ||
831 | static void try_to_unmap_cluster(unsigned long cursor, | |
832 | unsigned int *mapcount, struct vm_area_struct *vma) | |
833 | { | |
834 | struct mm_struct *mm = vma->vm_mm; | |
835 | pgd_t *pgd; | |
836 | pud_t *pud; | |
837 | pmd_t *pmd; | |
c0718806 | 838 | pte_t *pte; |
1da177e4 | 839 | pte_t pteval; |
c0718806 | 840 | spinlock_t *ptl; |
1da177e4 LT |
841 | struct page *page; |
842 | unsigned long address; | |
843 | unsigned long end; | |
1da177e4 | 844 | |
1da177e4 LT |
845 | address = (vma->vm_start + cursor) & CLUSTER_MASK; |
846 | end = address + CLUSTER_SIZE; | |
847 | if (address < vma->vm_start) | |
848 | address = vma->vm_start; | |
849 | if (end > vma->vm_end) | |
850 | end = vma->vm_end; | |
851 | ||
852 | pgd = pgd_offset(mm, address); | |
853 | if (!pgd_present(*pgd)) | |
c0718806 | 854 | return; |
1da177e4 LT |
855 | |
856 | pud = pud_offset(pgd, address); | |
857 | if (!pud_present(*pud)) | |
c0718806 | 858 | return; |
1da177e4 LT |
859 | |
860 | pmd = pmd_offset(pud, address); | |
861 | if (!pmd_present(*pmd)) | |
c0718806 HD |
862 | return; |
863 | ||
864 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
1da177e4 | 865 | |
365e9c87 HD |
866 | /* Update high watermark before we lower rss */ |
867 | update_hiwater_rss(mm); | |
868 | ||
c0718806 | 869 | for (; address < end; pte++, address += PAGE_SIZE) { |
1da177e4 LT |
870 | if (!pte_present(*pte)) |
871 | continue; | |
6aab341e LT |
872 | page = vm_normal_page(vma, address, *pte); |
873 | BUG_ON(!page || PageAnon(page)); | |
1da177e4 | 874 | |
cddb8a5c | 875 | if (ptep_clear_flush_young_notify(vma, address, pte)) |
1da177e4 LT |
876 | continue; |
877 | ||
878 | /* Nuke the page table entry. */ | |
eca35133 | 879 | flush_cache_page(vma, address, pte_pfn(*pte)); |
cddb8a5c | 880 | pteval = ptep_clear_flush_notify(vma, address, pte); |
1da177e4 LT |
881 | |
882 | /* If nonlinear, store the file page offset in the pte. */ | |
883 | if (page->index != linear_page_index(vma, address)) | |
884 | set_pte_at(mm, address, pte, pgoff_to_pte(page->index)); | |
885 | ||
886 | /* Move the dirty bit to the physical page now the pte is gone. */ | |
887 | if (pte_dirty(pteval)) | |
888 | set_page_dirty(page); | |
889 | ||
7de6b805 | 890 | page_remove_rmap(page, vma); |
1da177e4 | 891 | page_cache_release(page); |
4294621f | 892 | dec_mm_counter(mm, file_rss); |
1da177e4 LT |
893 | (*mapcount)--; |
894 | } | |
c0718806 | 895 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
896 | } |
897 | ||
7352349a | 898 | static int try_to_unmap_anon(struct page *page, int migration) |
1da177e4 LT |
899 | { |
900 | struct anon_vma *anon_vma; | |
901 | struct vm_area_struct *vma; | |
902 | int ret = SWAP_AGAIN; | |
903 | ||
904 | anon_vma = page_lock_anon_vma(page); | |
905 | if (!anon_vma) | |
906 | return ret; | |
907 | ||
908 | list_for_each_entry(vma, &anon_vma->head, anon_vma_node) { | |
7352349a | 909 | ret = try_to_unmap_one(page, vma, migration); |
1da177e4 LT |
910 | if (ret == SWAP_FAIL || !page_mapped(page)) |
911 | break; | |
912 | } | |
34bbd704 ON |
913 | |
914 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
915 | return ret; |
916 | } | |
917 | ||
918 | /** | |
919 | * try_to_unmap_file - unmap file page using the object-based rmap method | |
920 | * @page: the page to unmap | |
43d8eac4 | 921 | * @migration: migration flag |
1da177e4 LT |
922 | * |
923 | * Find all the mappings of a page using the mapping pointer and the vma chains | |
924 | * contained in the address_space struct it points to. | |
925 | * | |
926 | * This function is only called from try_to_unmap for object-based pages. | |
927 | */ | |
7352349a | 928 | static int try_to_unmap_file(struct page *page, int migration) |
1da177e4 LT |
929 | { |
930 | struct address_space *mapping = page->mapping; | |
931 | pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT); | |
932 | struct vm_area_struct *vma; | |
933 | struct prio_tree_iter iter; | |
934 | int ret = SWAP_AGAIN; | |
935 | unsigned long cursor; | |
936 | unsigned long max_nl_cursor = 0; | |
937 | unsigned long max_nl_size = 0; | |
938 | unsigned int mapcount; | |
939 | ||
940 | spin_lock(&mapping->i_mmap_lock); | |
941 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { | |
7352349a | 942 | ret = try_to_unmap_one(page, vma, migration); |
1da177e4 LT |
943 | if (ret == SWAP_FAIL || !page_mapped(page)) |
944 | goto out; | |
945 | } | |
946 | ||
947 | if (list_empty(&mapping->i_mmap_nonlinear)) | |
948 | goto out; | |
949 | ||
950 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
951 | shared.vm_set.list) { | |
e6a1530d | 952 | if ((vma->vm_flags & VM_LOCKED) && !migration) |
1da177e4 LT |
953 | continue; |
954 | cursor = (unsigned long) vma->vm_private_data; | |
955 | if (cursor > max_nl_cursor) | |
956 | max_nl_cursor = cursor; | |
957 | cursor = vma->vm_end - vma->vm_start; | |
958 | if (cursor > max_nl_size) | |
959 | max_nl_size = cursor; | |
960 | } | |
961 | ||
962 | if (max_nl_size == 0) { /* any nonlinears locked or reserved */ | |
963 | ret = SWAP_FAIL; | |
964 | goto out; | |
965 | } | |
966 | ||
967 | /* | |
968 | * We don't try to search for this page in the nonlinear vmas, | |
969 | * and page_referenced wouldn't have found it anyway. Instead | |
970 | * just walk the nonlinear vmas trying to age and unmap some. | |
971 | * The mapcount of the page we came in with is irrelevant, | |
972 | * but even so use it as a guide to how hard we should try? | |
973 | */ | |
974 | mapcount = page_mapcount(page); | |
975 | if (!mapcount) | |
976 | goto out; | |
977 | cond_resched_lock(&mapping->i_mmap_lock); | |
978 | ||
979 | max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK; | |
980 | if (max_nl_cursor == 0) | |
981 | max_nl_cursor = CLUSTER_SIZE; | |
982 | ||
983 | do { | |
984 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, | |
985 | shared.vm_set.list) { | |
e6a1530d | 986 | if ((vma->vm_flags & VM_LOCKED) && !migration) |
1da177e4 LT |
987 | continue; |
988 | cursor = (unsigned long) vma->vm_private_data; | |
839b9685 | 989 | while ( cursor < max_nl_cursor && |
1da177e4 LT |
990 | cursor < vma->vm_end - vma->vm_start) { |
991 | try_to_unmap_cluster(cursor, &mapcount, vma); | |
992 | cursor += CLUSTER_SIZE; | |
993 | vma->vm_private_data = (void *) cursor; | |
994 | if ((int)mapcount <= 0) | |
995 | goto out; | |
996 | } | |
997 | vma->vm_private_data = (void *) max_nl_cursor; | |
998 | } | |
999 | cond_resched_lock(&mapping->i_mmap_lock); | |
1000 | max_nl_cursor += CLUSTER_SIZE; | |
1001 | } while (max_nl_cursor <= max_nl_size); | |
1002 | ||
1003 | /* | |
1004 | * Don't loop forever (perhaps all the remaining pages are | |
1005 | * in locked vmas). Reset cursor on all unreserved nonlinear | |
1006 | * vmas, now forgetting on which ones it had fallen behind. | |
1007 | */ | |
101d2be7 HD |
1008 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
1009 | vma->vm_private_data = NULL; | |
1da177e4 LT |
1010 | out: |
1011 | spin_unlock(&mapping->i_mmap_lock); | |
1012 | return ret; | |
1013 | } | |
1014 | ||
1015 | /** | |
1016 | * try_to_unmap - try to remove all page table mappings to a page | |
1017 | * @page: the page to get unmapped | |
43d8eac4 | 1018 | * @migration: migration flag |
1da177e4 LT |
1019 | * |
1020 | * Tries to remove all the page table entries which are mapping this | |
1021 | * page, used in the pageout path. Caller must hold the page lock. | |
1022 | * Return values are: | |
1023 | * | |
1024 | * SWAP_SUCCESS - we succeeded in removing all mappings | |
1025 | * SWAP_AGAIN - we missed a mapping, try again later | |
1026 | * SWAP_FAIL - the page is unswappable | |
1027 | */ | |
7352349a | 1028 | int try_to_unmap(struct page *page, int migration) |
1da177e4 LT |
1029 | { |
1030 | int ret; | |
1031 | ||
1da177e4 LT |
1032 | BUG_ON(!PageLocked(page)); |
1033 | ||
1034 | if (PageAnon(page)) | |
7352349a | 1035 | ret = try_to_unmap_anon(page, migration); |
1da177e4 | 1036 | else |
7352349a | 1037 | ret = try_to_unmap_file(page, migration); |
1da177e4 LT |
1038 | |
1039 | if (!page_mapped(page)) | |
1040 | ret = SWAP_SUCCESS; | |
1041 | return ret; | |
1042 | } | |
81b4082d | 1043 |