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> |
1da177e4 LT |
52 | |
53 | #include <asm/tlbflush.h> | |
54 | ||
fcc234f8 | 55 | struct kmem_cache *anon_vma_cachep; |
1da177e4 | 56 | |
1da177e4 LT |
57 | /* This must be called under the mmap_sem. */ |
58 | int 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; | |
0697212a | 84 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
85 | 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 | ||
97 | void __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 | ||
103 | void __anon_vma_link(struct vm_area_struct *vma) | |
104 | { | |
105 | struct anon_vma *anon_vma = vma->anon_vma; | |
106 | ||
30acbaba | 107 | if (anon_vma) |
0697212a | 108 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
109 | } |
110 | ||
111 | void 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); | |
0697212a | 117 | list_add_tail(&vma->anon_vma_node, &anon_vma->head); |
1da177e4 LT |
118 | spin_unlock(&anon_vma->lock); |
119 | } | |
120 | } | |
121 | ||
122 | void 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); | |
1da177e4 LT |
131 | 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 | ||
51cc5068 | 141 | static void anon_vma_ctor(void *data) |
1da177e4 | 142 | { |
a35afb83 | 143 | struct anon_vma *anon_vma = data; |
1da177e4 | 144 | |
a35afb83 CL |
145 | spin_lock_init(&anon_vma->lock); |
146 | INIT_LIST_HEAD(&anon_vma->head); | |
1da177e4 LT |
147 | } |
148 | ||
149 | void __init anon_vma_init(void) | |
150 | { | |
151 | anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma), | |
20c2df83 | 152 | 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor); |
1da177e4 LT |
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 | */ | |
159 | static struct anon_vma *page_lock_anon_vma(struct page *page) | |
160 | { | |
34bbd704 | 161 | struct anon_vma *anon_vma; |
1da177e4 LT |
162 | 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); | |
34bbd704 | 173 | return anon_vma; |
1da177e4 LT |
174 | out: |
175 | rcu_read_unlock(); | |
34bbd704 ON |
176 | return NULL; |
177 | } | |
178 | ||
179 | static void page_unlock_anon_vma(struct anon_vma *anon_vma) | |
180 | { | |
181 | spin_unlock(&anon_vma->lock); | |
182 | rcu_read_unlock(); | |
1da177e4 LT |
183 | } |
184 | ||
185 | /* | |
3ad33b24 LS |
186 | * 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. | |
1da177e4 LT |
189 | */ |
190 | static inline unsigned long | |
191 | vma_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)) { | |
3ad33b24 | 198 | /* page should be within @vma mapping range */ |
1da177e4 LT |
199 | return -EFAULT; |
200 | } | |
201 | return address; | |
202 | } | |
203 | ||
204 | /* | |
205 | * At what user virtual address is page expected in vma? checking that the | |
ee498ed7 | 206 | * page matches the vma: currently only used on anon pages, by unuse_vma; |
1da177e4 LT |
207 | */ |
208 | unsigned 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)) { | |
ee498ed7 HD |
215 | if (!vma->vm_file || |
216 | vma->vm_file->f_mapping != page->mapping) | |
1da177e4 LT |
217 | return -EFAULT; |
218 | } else | |
219 | return -EFAULT; | |
220 | return vma_address(page, vma); | |
221 | } | |
222 | ||
81b4082d ND |
223 | /* |
224 | * Check that @page is mapped at @address into @mm. | |
225 | * | |
b8072f09 | 226 | * On success returns with pte mapped and locked. |
81b4082d | 227 | */ |
ceffc078 | 228 | pte_t *page_check_address(struct page *page, struct mm_struct *mm, |
c0718806 | 229 | unsigned long address, spinlock_t **ptlp) |
81b4082d ND |
230 | { |
231 | pgd_t *pgd; | |
232 | pud_t *pud; | |
233 | pmd_t *pmd; | |
234 | pte_t *pte; | |
c0718806 | 235 | spinlock_t *ptl; |
81b4082d | 236 | |
81b4082d | 237 | pgd = pgd_offset(mm, address); |
c0718806 HD |
238 | 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; | |
254 | } | |
255 | ||
4c21e2f2 | 256 | ptl = pte_lockptr(mm, pmd); |
c0718806 HD |
257 | spin_lock(ptl); |
258 | if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) { | |
259 | *ptlp = ptl; | |
260 | return pte; | |
81b4082d | 261 | } |
c0718806 HD |
262 | pte_unmap_unlock(pte, ptl); |
263 | return NULL; | |
81b4082d ND |
264 | } |
265 | ||
1da177e4 LT |
266 | /* |
267 | * Subfunctions of page_referenced: page_referenced_one called | |
268 | * repeatedly from either page_referenced_anon or page_referenced_file. | |
269 | */ | |
270 | static int page_referenced_one(struct page *page, | |
f7b7fd8f | 271 | struct vm_area_struct *vma, unsigned int *mapcount) |
1da177e4 LT |
272 | { |
273 | struct mm_struct *mm = vma->vm_mm; | |
274 | unsigned long address; | |
1da177e4 | 275 | pte_t *pte; |
c0718806 | 276 | spinlock_t *ptl; |
1da177e4 LT |
277 | int referenced = 0; |
278 | ||
1da177e4 LT |
279 | address = vma_address(page, vma); |
280 | if (address == -EFAULT) | |
281 | goto out; | |
282 | ||
c0718806 HD |
283 | pte = page_check_address(page, mm, address, &ptl); |
284 | if (!pte) | |
285 | goto out; | |
1da177e4 | 286 | |
5a9bbdcd HD |
287 | 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)) | |
c0718806 | 291 | referenced++; |
1da177e4 | 292 | |
c0718806 HD |
293 | /* Pretend the page is referenced if the task has the |
294 | swap token and is in the middle of a page fault. */ | |
f7b7fd8f | 295 | if (mm != current->mm && has_swap_token(mm) && |
c0718806 HD |
296 | rwsem_is_locked(&mm->mmap_sem)) |
297 | referenced++; | |
298 | ||
299 | (*mapcount)--; | |
300 | pte_unmap_unlock(pte, ptl); | |
1da177e4 LT |
301 | out: |
302 | return referenced; | |
303 | } | |
304 | ||
bed7161a BS |
305 | static int page_referenced_anon(struct page *page, |
306 | struct mem_cgroup *mem_cont) | |
1da177e4 LT |
307 | { |
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) { | |
bed7161a BS |
319 | /* |
320 | * If we are reclaiming on behalf of a cgroup, skip | |
321 | * counting on behalf of references from different | |
322 | * cgroups | |
323 | */ | |
bd845e38 | 324 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 325 | continue; |
f7b7fd8f | 326 | referenced += page_referenced_one(page, vma, &mapcount); |
1da177e4 LT |
327 | if (!mapcount) |
328 | break; | |
329 | } | |
34bbd704 ON |
330 | |
331 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
332 | return referenced; |
333 | } | |
334 | ||
335 | /** | |
336 | * page_referenced_file - referenced check for object-based rmap | |
337 | * @page: the page we're checking references on. | |
43d8eac4 | 338 | * @mem_cont: target memory controller |
1da177e4 LT |
339 | * |
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 | */ | |
bed7161a BS |
347 | static int page_referenced_file(struct page *page, |
348 | struct mem_cgroup *mem_cont) | |
1da177e4 LT |
349 | { |
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) { | |
bed7161a BS |
381 | /* |
382 | * If we are reclaiming on behalf of a cgroup, skip | |
383 | * counting on behalf of references from different | |
384 | * cgroups | |
385 | */ | |
bd845e38 | 386 | if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont)) |
bed7161a | 387 | continue; |
1da177e4 LT |
388 | if ((vma->vm_flags & (VM_LOCKED|VM_MAYSHARE)) |
389 | == (VM_LOCKED|VM_MAYSHARE)) { | |
390 | referenced++; | |
391 | break; | |
392 | } | |
f7b7fd8f | 393 | referenced += page_referenced_one(page, vma, &mapcount); |
1da177e4 LT |
394 | 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 | |
43d8eac4 | 406 | * @mem_cont: target memory controller |
1da177e4 LT |
407 | * |
408 | * Quick test_and_clear_referenced for all mappings to a page, | |
409 | * returns the number of ptes which referenced the page. | |
410 | */ | |
bed7161a BS |
411 | int page_referenced(struct page *page, int is_locked, |
412 | struct mem_cgroup *mem_cont) | |
1da177e4 LT |
413 | { |
414 | int referenced = 0; | |
415 | ||
1da177e4 LT |
416 | if (TestClearPageReferenced(page)) |
417 | referenced++; | |
418 | ||
419 | if (page_mapped(page) && page->mapping) { | |
420 | if (PageAnon(page)) | |
bed7161a | 421 | referenced += page_referenced_anon(page, mem_cont); |
1da177e4 | 422 | else if (is_locked) |
bed7161a | 423 | referenced += page_referenced_file(page, mem_cont); |
1da177e4 LT |
424 | else if (TestSetPageLocked(page)) |
425 | referenced++; | |
426 | else { | |
427 | if (page->mapping) | |
bed7161a BS |
428 | referenced += |
429 | page_referenced_file(page, mem_cont); | |
1da177e4 LT |
430 | unlock_page(page); |
431 | } | |
432 | } | |
5b7baf05 CB |
433 | |
434 | if (page_test_and_clear_young(page)) | |
435 | referenced++; | |
436 | ||
1da177e4 LT |
437 | return referenced; |
438 | } | |
439 | ||
d08b3851 PZ |
440 | static 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; | |
c2fda5fe | 444 | pte_t *pte; |
d08b3851 PZ |
445 | 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 | ||
c2fda5fe PZ |
456 | if (pte_dirty(*pte) || pte_write(*pte)) { |
457 | pte_t entry; | |
d08b3851 | 458 | |
c2fda5fe PZ |
459 | 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); | |
d6e88e67 | 463 | set_pte_at(mm, address, pte, entry); |
c2fda5fe PZ |
464 | ret = 1; |
465 | } | |
d08b3851 | 466 | |
d08b3851 PZ |
467 | pte_unmap_unlock(pte, ptl); |
468 | out: | |
469 | return ret; | |
470 | } | |
471 | ||
472 | static 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 | ||
490 | int 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); | |
ce7e9fae | 498 | if (mapping) { |
d08b3851 | 499 | ret = page_mkclean_file(mapping, page); |
ce7e9fae CB |
500 | if (page_test_dirty(page)) { |
501 | page_clear_dirty(page); | |
502 | ret = 1; | |
503 | } | |
6c210482 | 504 | } |
d08b3851 PZ |
505 | } |
506 | ||
507 | return ret; | |
508 | } | |
60b59bea | 509 | EXPORT_SYMBOL_GPL(page_mkclean); |
d08b3851 | 510 | |
9617d95e | 511 | /** |
43d8eac4 | 512 | * __page_set_anon_rmap - setup new anonymous rmap |
9617d95e NP |
513 | * @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 | */ | |
517 | static 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 | ||
a74609fa NP |
528 | /* |
529 | * nr_mapped state can be updated without turning off | |
530 | * interrupts because it is not modified via interrupt. | |
531 | */ | |
f3dbd344 | 532 | __inc_zone_page_state(page, NR_ANON_PAGES); |
9617d95e NP |
533 | } |
534 | ||
c97a9e10 | 535 | /** |
43d8eac4 | 536 | * __page_check_anon_rmap - sanity check anonymous rmap addition |
c97a9e10 NP |
537 | * @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 | */ | |
541 | static 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 | ||
1da177e4 LT |
564 | /** |
565 | * 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 | * | |
c97a9e10 | 570 | * The caller needs to hold the pte lock and the page must be locked. |
1da177e4 LT |
571 | */ |
572 | void page_add_anon_rmap(struct page *page, | |
573 | struct vm_area_struct *vma, unsigned long address) | |
574 | { | |
c97a9e10 NP |
575 | VM_BUG_ON(!PageLocked(page)); |
576 | VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end); | |
9617d95e NP |
577 | if (atomic_inc_and_test(&page->_mapcount)) |
578 | __page_set_anon_rmap(page, vma, address); | |
69029cd5 | 579 | else |
c97a9e10 | 580 | __page_check_anon_rmap(page, vma, address); |
1da177e4 LT |
581 | } |
582 | ||
43d8eac4 | 583 | /** |
9617d95e NP |
584 | * 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. | |
c97a9e10 | 591 | * Page does not have to be locked. |
9617d95e NP |
592 | */ |
593 | void page_add_new_anon_rmap(struct page *page, | |
594 | struct vm_area_struct *vma, unsigned long address) | |
595 | { | |
c97a9e10 | 596 | BUG_ON(address < vma->vm_start || address >= vma->vm_end); |
9617d95e NP |
597 | atomic_set(&page->_mapcount, 0); /* elevate count by 1 (starts at -1) */ |
598 | __page_set_anon_rmap(page, vma, address); | |
599 | } | |
600 | ||
1da177e4 LT |
601 | /** |
602 | * page_add_file_rmap - add pte mapping to a file page | |
603 | * @page: the page to add the mapping to | |
604 | * | |
b8072f09 | 605 | * The caller needs to hold the pte lock. |
1da177e4 LT |
606 | */ |
607 | void page_add_file_rmap(struct page *page) | |
608 | { | |
1da177e4 | 609 | if (atomic_inc_and_test(&page->_mapcount)) |
65ba55f5 | 610 | __inc_zone_page_state(page, NR_FILE_MAPPED); |
1da177e4 LT |
611 | } |
612 | ||
c97a9e10 NP |
613 | #ifdef CONFIG_DEBUG_VM |
614 | /** | |
615 | * page_dup_rmap - duplicate pte mapping to a page | |
616 | * @page: the page to add the mapping to | |
43d8eac4 RD |
617 | * @vma: the vm area being duplicated |
618 | * @address: the user virtual address mapped | |
c97a9e10 NP |
619 | * |
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 | */ | |
626 | void 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 | ||
1da177e4 LT |
635 | /** |
636 | * page_remove_rmap - take down pte mapping from a page | |
637 | * @page: page to remove mapping from | |
43d8eac4 | 638 | * @vma: the vm area in which the mapping is removed |
1da177e4 | 639 | * |
b8072f09 | 640 | * The caller needs to hold the pte lock. |
1da177e4 | 641 | */ |
7de6b805 | 642 | void page_remove_rmap(struct page *page, struct vm_area_struct *vma) |
1da177e4 | 643 | { |
1da177e4 | 644 | if (atomic_add_negative(-1, &page->_mapcount)) { |
b7ab795b | 645 | if (unlikely(page_mapcount(page) < 0)) { |
ef2bf0dc | 646 | printk (KERN_EMERG "Eeek! page_mapcount(page) went negative! (%d)\n", page_mapcount(page)); |
7de6b805 | 647 | printk (KERN_EMERG " page pfn = %lx\n", page_to_pfn(page)); |
ef2bf0dc DJ |
648 | 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); | |
7de6b805 | 651 | print_symbol (KERN_EMERG " vma->vm_ops = %s\n", (unsigned long)vma->vm_ops); |
54cb8821 | 652 | if (vma->vm_ops) { |
54cb8821 NP |
653 | print_symbol (KERN_EMERG " vma->vm_ops->fault = %s\n", (unsigned long)vma->vm_ops->fault); |
654 | } | |
7de6b805 NP |
655 | 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); | |
b16bc64d | 657 | BUG(); |
ef2bf0dc | 658 | } |
b16bc64d | 659 | |
1da177e4 LT |
660 | /* |
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 | */ | |
6c210482 MS |
669 | if (page_test_dirty(page)) { |
670 | page_clear_dirty(page); | |
1da177e4 | 671 | set_page_dirty(page); |
6c210482 | 672 | } |
8a9f3ccd BS |
673 | mem_cgroup_uncharge_page(page); |
674 | ||
f3dbd344 CL |
675 | __dec_zone_page_state(page, |
676 | PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED); | |
1da177e4 LT |
677 | } |
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 | */ | |
a48d07af | 684 | static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma, |
7352349a | 685 | int migration) |
1da177e4 LT |
686 | { |
687 | struct mm_struct *mm = vma->vm_mm; | |
688 | unsigned long address; | |
1da177e4 LT |
689 | pte_t *pte; |
690 | pte_t pteval; | |
c0718806 | 691 | spinlock_t *ptl; |
1da177e4 LT |
692 | int ret = SWAP_AGAIN; |
693 | ||
1da177e4 LT |
694 | address = vma_address(page, vma); |
695 | if (address == -EFAULT) | |
696 | goto out; | |
697 | ||
c0718806 HD |
698 | pte = page_check_address(page, mm, address, &ptl); |
699 | if (!pte) | |
81b4082d | 700 | goto out; |
1da177e4 LT |
701 | |
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 | */ | |
e6a1530d CL |
707 | if (!migration && ((vma->vm_flags & VM_LOCKED) || |
708 | (ptep_clear_flush_young(vma, address, pte)))) { | |
1da177e4 LT |
709 | ret = SWAP_FAIL; |
710 | goto out_unmap; | |
711 | } | |
712 | ||
1da177e4 LT |
713 | /* 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 | ||
365e9c87 HD |
721 | /* Update high watermark before we lower rss */ |
722 | update_hiwater_rss(mm); | |
723 | ||
1da177e4 | 724 | if (PageAnon(page)) { |
4c21e2f2 | 725 | swp_entry_t entry = { .val = page_private(page) }; |
0697212a CL |
726 | |
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 | } | |
442c9137 | 739 | dec_mm_counter(mm, anon_rss); |
04e62a29 | 740 | #ifdef CONFIG_MIGRATION |
0697212a CL |
741 | } 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)); | |
04e62a29 | 749 | #endif |
1da177e4 LT |
750 | } |
751 | set_pte_at(mm, address, pte, swp_entry_to_pte(entry)); | |
752 | BUG_ON(pte_file(*pte)); | |
4294621f | 753 | } else |
04e62a29 CL |
754 | #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 | |
4294621f | 762 | dec_mm_counter(mm, file_rss); |
1da177e4 | 763 | |
04e62a29 | 764 | |
7de6b805 | 765 | page_remove_rmap(page, vma); |
1da177e4 LT |
766 | page_cache_release(page); |
767 | ||
768 | out_unmap: | |
c0718806 | 769 | pte_unmap_unlock(pte, ptl); |
1da177e4 LT |
770 | out: |
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 | ||
796 | static 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; | |
c0718806 | 803 | pte_t *pte; |
1da177e4 | 804 | pte_t pteval; |
c0718806 | 805 | spinlock_t *ptl; |
1da177e4 LT |
806 | struct page *page; |
807 | unsigned long address; | |
808 | unsigned long end; | |
1da177e4 | 809 | |
1da177e4 LT |
810 | 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)) | |
c0718806 | 819 | return; |
1da177e4 LT |
820 | |
821 | pud = pud_offset(pgd, address); | |
822 | if (!pud_present(*pud)) | |
c0718806 | 823 | return; |
1da177e4 LT |
824 | |
825 | pmd = pmd_offset(pud, address); | |
826 | if (!pmd_present(*pmd)) | |
c0718806 HD |
827 | return; |
828 | ||
829 | pte = pte_offset_map_lock(mm, pmd, address, &ptl); | |
1da177e4 | 830 | |
365e9c87 HD |
831 | /* Update high watermark before we lower rss */ |
832 | update_hiwater_rss(mm); | |
833 | ||
c0718806 | 834 | for (; address < end; pte++, address += PAGE_SIZE) { |
1da177e4 LT |
835 | if (!pte_present(*pte)) |
836 | continue; | |
6aab341e LT |
837 | page = vm_normal_page(vma, address, *pte); |
838 | BUG_ON(!page || PageAnon(page)); | |
1da177e4 LT |
839 | |
840 | if (ptep_clear_flush_young(vma, address, pte)) | |
841 | continue; | |
842 | ||
843 | /* Nuke the page table entry. */ | |
eca35133 | 844 | flush_cache_page(vma, address, pte_pfn(*pte)); |
1da177e4 LT |
845 | 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 | ||
7de6b805 | 855 | page_remove_rmap(page, vma); |
1da177e4 | 856 | page_cache_release(page); |
4294621f | 857 | dec_mm_counter(mm, file_rss); |
1da177e4 LT |
858 | (*mapcount)--; |
859 | } | |
c0718806 | 860 | pte_unmap_unlock(pte - 1, ptl); |
1da177e4 LT |
861 | } |
862 | ||
7352349a | 863 | static int try_to_unmap_anon(struct page *page, int migration) |
1da177e4 LT |
864 | { |
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) { | |
7352349a | 874 | ret = try_to_unmap_one(page, vma, migration); |
1da177e4 LT |
875 | if (ret == SWAP_FAIL || !page_mapped(page)) |
876 | break; | |
877 | } | |
34bbd704 ON |
878 | |
879 | page_unlock_anon_vma(anon_vma); | |
1da177e4 LT |
880 | 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 | |
43d8eac4 | 886 | * @migration: migration flag |
1da177e4 LT |
887 | * |
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 | */ | |
7352349a | 893 | static int try_to_unmap_file(struct page *page, int migration) |
1da177e4 LT |
894 | { |
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) { | |
7352349a | 907 | ret = try_to_unmap_one(page, vma, migration); |
1da177e4 LT |
908 | 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) { | |
e6a1530d | 917 | if ((vma->vm_flags & VM_LOCKED) && !migration) |
1da177e4 LT |
918 | 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) { | |
e6a1530d | 951 | if ((vma->vm_flags & VM_LOCKED) && !migration) |
1da177e4 LT |
952 | continue; |
953 | cursor = (unsigned long) vma->vm_private_data; | |
839b9685 | 954 | while ( cursor < max_nl_cursor && |
1da177e4 LT |
955 | 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 | */ | |
101d2be7 HD |
973 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
974 | vma->vm_private_data = NULL; | |
1da177e4 LT |
975 | out: |
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 | |
43d8eac4 | 983 | * @migration: migration flag |
1da177e4 LT |
984 | * |
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 | */ | |
7352349a | 993 | int try_to_unmap(struct page *page, int migration) |
1da177e4 LT |
994 | { |
995 | int ret; | |
996 | ||
1da177e4 LT |
997 | BUG_ON(!PageLocked(page)); |
998 | ||
999 | if (PageAnon(page)) | |
7352349a | 1000 | ret = try_to_unmap_anon(page, migration); |
1da177e4 | 1001 | else |
7352349a | 1002 | ret = try_to_unmap_file(page, migration); |
1da177e4 LT |
1003 | |
1004 | if (!page_mapped(page)) | |
1005 | ret = SWAP_SUCCESS; | |
1006 | return ret; | |
1007 | } | |
81b4082d | 1008 |