]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * mm/truncate.c - code for taking down pages from address_spaces | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds | |
5 | * | |
6 | * 10Sep2002 Andrew Morton | |
7 | * Initial version. | |
8 | */ | |
9 | ||
10 | #include <linux/kernel.h> | |
11 | #include <linux/backing-dev.h> | |
12 | #include <linux/gfp.h> | |
13 | #include <linux/mm.h> | |
14 | #include <linux/swap.h> | |
15 | #include <linux/module.h> | |
16 | #include <linux/pagemap.h> | |
17 | #include <linux/highmem.h> | |
18 | #include <linux/pagevec.h> | |
19 | #include <linux/task_io_accounting_ops.h> | |
20 | #include <linux/buffer_head.h> /* grr. try_to_release_page, | |
21 | do_invalidatepage */ | |
22 | #include "internal.h" | |
23 | ||
24 | ||
25 | /** | |
26 | * do_invalidatepage - invalidate part or all of a page | |
27 | * @page: the page which is affected | |
28 | * @offset: the index of the truncation point | |
29 | * | |
30 | * do_invalidatepage() is called when all or part of the page has become | |
31 | * invalidated by a truncate operation. | |
32 | * | |
33 | * do_invalidatepage() does not have to release all buffers, but it must | |
34 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
35 | * is underway against any of the blocks which are outside the truncation | |
36 | * point. Because the caller is about to free (and possibly reuse) those | |
37 | * blocks on-disk. | |
38 | */ | |
39 | void do_invalidatepage(struct page *page, unsigned long offset) | |
40 | { | |
41 | void (*invalidatepage)(struct page *, unsigned long); | |
42 | invalidatepage = page->mapping->a_ops->invalidatepage; | |
43 | #ifdef CONFIG_BLOCK | |
44 | if (!invalidatepage) | |
45 | invalidatepage = block_invalidatepage; | |
46 | #endif | |
47 | if (invalidatepage) | |
48 | (*invalidatepage)(page, offset); | |
49 | } | |
50 | ||
51 | static inline void truncate_partial_page(struct page *page, unsigned partial) | |
52 | { | |
53 | zero_user_segment(page, partial, PAGE_CACHE_SIZE); | |
54 | if (page_has_private(page)) | |
55 | do_invalidatepage(page, partial); | |
56 | } | |
57 | ||
58 | /* | |
59 | * This cancels just the dirty bit on the kernel page itself, it | |
60 | * does NOT actually remove dirty bits on any mmap's that may be | |
61 | * around. It also leaves the page tagged dirty, so any sync | |
62 | * activity will still find it on the dirty lists, and in particular, | |
63 | * clear_page_dirty_for_io() will still look at the dirty bits in | |
64 | * the VM. | |
65 | * | |
66 | * Doing this should *normally* only ever be done when a page | |
67 | * is truncated, and is not actually mapped anywhere at all. However, | |
68 | * fs/buffer.c does this when it notices that somebody has cleaned | |
69 | * out all the buffers on a page without actually doing it through | |
70 | * the VM. Can you say "ext3 is horribly ugly"? Tought you could. | |
71 | */ | |
72 | void cancel_dirty_page(struct page *page, unsigned int account_size) | |
73 | { | |
74 | if (TestClearPageDirty(page)) { | |
75 | struct address_space *mapping = page->mapping; | |
76 | if (mapping && mapping_cap_account_dirty(mapping)) { | |
77 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
78 | dec_bdi_stat(mapping->backing_dev_info, | |
79 | BDI_RECLAIMABLE); | |
80 | if (account_size) | |
81 | task_io_account_cancelled_write(account_size); | |
82 | } | |
83 | } | |
84 | } | |
85 | EXPORT_SYMBOL(cancel_dirty_page); | |
86 | ||
87 | /* | |
88 | * If truncate cannot remove the fs-private metadata from the page, the page | |
89 | * becomes orphaned. It will be left on the LRU and may even be mapped into | |
90 | * user pagetables if we're racing with filemap_fault(). | |
91 | * | |
92 | * We need to bale out if page->mapping is no longer equal to the original | |
93 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
94 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and | |
95 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. | |
96 | */ | |
97 | static int | |
98 | truncate_complete_page(struct address_space *mapping, struct page *page) | |
99 | { | |
100 | if (page->mapping != mapping) | |
101 | return -EIO; | |
102 | ||
103 | if (page_has_private(page)) | |
104 | do_invalidatepage(page, 0); | |
105 | ||
106 | cancel_dirty_page(page, PAGE_CACHE_SIZE); | |
107 | ||
108 | clear_page_mlock(page); | |
109 | remove_from_page_cache(page); | |
110 | ClearPageMappedToDisk(page); | |
111 | page_cache_release(page); /* pagecache ref */ | |
112 | return 0; | |
113 | } | |
114 | ||
115 | /* | |
116 | * This is for invalidate_mapping_pages(). That function can be called at | |
117 | * any time, and is not supposed to throw away dirty pages. But pages can | |
118 | * be marked dirty at any time too, so use remove_mapping which safely | |
119 | * discards clean, unused pages. | |
120 | * | |
121 | * Returns non-zero if the page was successfully invalidated. | |
122 | */ | |
123 | static int | |
124 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
125 | { | |
126 | int ret; | |
127 | ||
128 | if (page->mapping != mapping) | |
129 | return 0; | |
130 | ||
131 | if (page_has_private(page) && !try_to_release_page(page, 0)) | |
132 | return 0; | |
133 | ||
134 | clear_page_mlock(page); | |
135 | ret = remove_mapping(mapping, page); | |
136 | ||
137 | return ret; | |
138 | } | |
139 | ||
140 | int truncate_inode_page(struct address_space *mapping, struct page *page) | |
141 | { | |
142 | if (page_mapped(page)) { | |
143 | unmap_mapping_range(mapping, | |
144 | (loff_t)page->index << PAGE_CACHE_SHIFT, | |
145 | PAGE_CACHE_SIZE, 0); | |
146 | } | |
147 | return truncate_complete_page(mapping, page); | |
148 | } | |
149 | ||
150 | /* | |
151 | * Used to get rid of pages on hardware memory corruption. | |
152 | */ | |
153 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
154 | { | |
155 | if (!mapping) | |
156 | return -EINVAL; | |
157 | /* | |
158 | * Only punch for normal data pages for now. | |
159 | * Handling other types like directories would need more auditing. | |
160 | */ | |
161 | if (!S_ISREG(mapping->host->i_mode)) | |
162 | return -EIO; | |
163 | return truncate_inode_page(mapping, page); | |
164 | } | |
165 | EXPORT_SYMBOL(generic_error_remove_page); | |
166 | ||
167 | /* | |
168 | * Safely invalidate one page from its pagecache mapping. | |
169 | * It only drops clean, unused pages. The page must be locked. | |
170 | * | |
171 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
172 | */ | |
173 | int invalidate_inode_page(struct page *page) | |
174 | { | |
175 | struct address_space *mapping = page_mapping(page); | |
176 | if (!mapping) | |
177 | return 0; | |
178 | if (PageDirty(page) || PageWriteback(page)) | |
179 | return 0; | |
180 | if (page_mapped(page)) | |
181 | return 0; | |
182 | return invalidate_complete_page(mapping, page); | |
183 | } | |
184 | ||
185 | /** | |
186 | * truncate_inode_pages - truncate range of pages specified by start & end byte offsets | |
187 | * @mapping: mapping to truncate | |
188 | * @lstart: offset from which to truncate | |
189 | * @lend: offset to which to truncate | |
190 | * | |
191 | * Truncate the page cache, removing the pages that are between | |
192 | * specified offsets (and zeroing out partial page | |
193 | * (if lstart is not page aligned)). | |
194 | * | |
195 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
196 | * block on page locks and it will not block on writeback. The second pass | |
197 | * will wait. This is to prevent as much IO as possible in the affected region. | |
198 | * The first pass will remove most pages, so the search cost of the second pass | |
199 | * is low. | |
200 | * | |
201 | * When looking at page->index outside the page lock we need to be careful to | |
202 | * copy it into a local to avoid races (it could change at any time). | |
203 | * | |
204 | * We pass down the cache-hot hint to the page freeing code. Even if the | |
205 | * mapping is large, it is probably the case that the final pages are the most | |
206 | * recently touched, and freeing happens in ascending file offset order. | |
207 | */ | |
208 | void truncate_inode_pages_range(struct address_space *mapping, | |
209 | loff_t lstart, loff_t lend) | |
210 | { | |
211 | const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; | |
212 | pgoff_t end; | |
213 | const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1); | |
214 | struct pagevec pvec; | |
215 | pgoff_t next; | |
216 | int i; | |
217 | ||
218 | if (mapping->nrpages == 0) | |
219 | return; | |
220 | ||
221 | BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1)); | |
222 | end = (lend >> PAGE_CACHE_SHIFT); | |
223 | ||
224 | pagevec_init(&pvec, 0); | |
225 | next = start; | |
226 | while (next <= end && | |
227 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
228 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
229 | struct page *page = pvec.pages[i]; | |
230 | pgoff_t page_index = page->index; | |
231 | ||
232 | if (page_index > end) { | |
233 | next = page_index; | |
234 | break; | |
235 | } | |
236 | ||
237 | if (page_index > next) | |
238 | next = page_index; | |
239 | next++; | |
240 | if (!trylock_page(page)) | |
241 | continue; | |
242 | if (PageWriteback(page)) { | |
243 | unlock_page(page); | |
244 | continue; | |
245 | } | |
246 | truncate_inode_page(mapping, page); | |
247 | unlock_page(page); | |
248 | } | |
249 | pagevec_release(&pvec); | |
250 | cond_resched(); | |
251 | } | |
252 | ||
253 | if (partial) { | |
254 | struct page *page = find_lock_page(mapping, start - 1); | |
255 | if (page) { | |
256 | wait_on_page_writeback(page); | |
257 | truncate_partial_page(page, partial); | |
258 | unlock_page(page); | |
259 | page_cache_release(page); | |
260 | } | |
261 | } | |
262 | ||
263 | next = start; | |
264 | for ( ; ; ) { | |
265 | cond_resched(); | |
266 | if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
267 | if (next == start) | |
268 | break; | |
269 | next = start; | |
270 | continue; | |
271 | } | |
272 | if (pvec.pages[0]->index > end) { | |
273 | pagevec_release(&pvec); | |
274 | break; | |
275 | } | |
276 | mem_cgroup_uncharge_start(); | |
277 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
278 | struct page *page = pvec.pages[i]; | |
279 | ||
280 | if (page->index > end) | |
281 | break; | |
282 | lock_page(page); | |
283 | wait_on_page_writeback(page); | |
284 | truncate_inode_page(mapping, page); | |
285 | if (page->index > next) | |
286 | next = page->index; | |
287 | next++; | |
288 | unlock_page(page); | |
289 | } | |
290 | pagevec_release(&pvec); | |
291 | mem_cgroup_uncharge_end(); | |
292 | } | |
293 | } | |
294 | EXPORT_SYMBOL(truncate_inode_pages_range); | |
295 | ||
296 | /** | |
297 | * truncate_inode_pages - truncate *all* the pages from an offset | |
298 | * @mapping: mapping to truncate | |
299 | * @lstart: offset from which to truncate | |
300 | * | |
301 | * Called under (and serialised by) inode->i_mutex. | |
302 | */ | |
303 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
304 | { | |
305 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
306 | } | |
307 | EXPORT_SYMBOL(truncate_inode_pages); | |
308 | ||
309 | /** | |
310 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
311 | * @mapping: the address_space which holds the pages to invalidate | |
312 | * @start: the offset 'from' which to invalidate | |
313 | * @end: the offset 'to' which to invalidate (inclusive) | |
314 | * | |
315 | * This function only removes the unlocked pages, if you want to | |
316 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
317 | * | |
318 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
319 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
320 | * pagetables. | |
321 | */ | |
322 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
323 | pgoff_t start, pgoff_t end) | |
324 | { | |
325 | struct pagevec pvec; | |
326 | pgoff_t next = start; | |
327 | unsigned long ret = 0; | |
328 | int i; | |
329 | ||
330 | pagevec_init(&pvec, 0); | |
331 | while (next <= end && | |
332 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
333 | mem_cgroup_uncharge_start(); | |
334 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
335 | struct page *page = pvec.pages[i]; | |
336 | pgoff_t index; | |
337 | int lock_failed; | |
338 | ||
339 | lock_failed = !trylock_page(page); | |
340 | ||
341 | /* | |
342 | * We really shouldn't be looking at the ->index of an | |
343 | * unlocked page. But we're not allowed to lock these | |
344 | * pages. So we rely upon nobody altering the ->index | |
345 | * of this (pinned-by-us) page. | |
346 | */ | |
347 | index = page->index; | |
348 | if (index > next) | |
349 | next = index; | |
350 | next++; | |
351 | if (lock_failed) | |
352 | continue; | |
353 | ||
354 | ret += invalidate_inode_page(page); | |
355 | ||
356 | unlock_page(page); | |
357 | if (next > end) | |
358 | break; | |
359 | } | |
360 | pagevec_release(&pvec); | |
361 | mem_cgroup_uncharge_end(); | |
362 | cond_resched(); | |
363 | } | |
364 | return ret; | |
365 | } | |
366 | EXPORT_SYMBOL(invalidate_mapping_pages); | |
367 | ||
368 | /* | |
369 | * This is like invalidate_complete_page(), except it ignores the page's | |
370 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
371 | * invalidation guarantees, and cannot afford to leave pages behind because | |
372 | * shrink_page_list() has a temp ref on them, or because they're transiently | |
373 | * sitting in the lru_cache_add() pagevecs. | |
374 | */ | |
375 | static int | |
376 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
377 | { | |
378 | if (page->mapping != mapping) | |
379 | return 0; | |
380 | ||
381 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) | |
382 | return 0; | |
383 | ||
384 | spin_lock_irq(&mapping->tree_lock); | |
385 | if (PageDirty(page)) | |
386 | goto failed; | |
387 | ||
388 | clear_page_mlock(page); | |
389 | BUG_ON(page_has_private(page)); | |
390 | __remove_from_page_cache(page); | |
391 | spin_unlock_irq(&mapping->tree_lock); | |
392 | mem_cgroup_uncharge_cache_page(page); | |
393 | page_cache_release(page); /* pagecache ref */ | |
394 | return 1; | |
395 | failed: | |
396 | spin_unlock_irq(&mapping->tree_lock); | |
397 | return 0; | |
398 | } | |
399 | ||
400 | static int do_launder_page(struct address_space *mapping, struct page *page) | |
401 | { | |
402 | if (!PageDirty(page)) | |
403 | return 0; | |
404 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
405 | return 0; | |
406 | return mapping->a_ops->launder_page(page); | |
407 | } | |
408 | ||
409 | /** | |
410 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
411 | * @mapping: the address_space | |
412 | * @start: the page offset 'from' which to invalidate | |
413 | * @end: the page offset 'to' which to invalidate (inclusive) | |
414 | * | |
415 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
416 | * invalidation. | |
417 | * | |
418 | * Returns -EBUSY if any pages could not be invalidated. | |
419 | */ | |
420 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
421 | pgoff_t start, pgoff_t end) | |
422 | { | |
423 | struct pagevec pvec; | |
424 | pgoff_t next; | |
425 | int i; | |
426 | int ret = 0; | |
427 | int ret2 = 0; | |
428 | int did_range_unmap = 0; | |
429 | int wrapped = 0; | |
430 | ||
431 | pagevec_init(&pvec, 0); | |
432 | next = start; | |
433 | while (next <= end && !wrapped && | |
434 | pagevec_lookup(&pvec, mapping, next, | |
435 | min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
436 | mem_cgroup_uncharge_start(); | |
437 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
438 | struct page *page = pvec.pages[i]; | |
439 | pgoff_t page_index; | |
440 | ||
441 | lock_page(page); | |
442 | if (page->mapping != mapping) { | |
443 | unlock_page(page); | |
444 | continue; | |
445 | } | |
446 | page_index = page->index; | |
447 | next = page_index + 1; | |
448 | if (next == 0) | |
449 | wrapped = 1; | |
450 | if (page_index > end) { | |
451 | unlock_page(page); | |
452 | break; | |
453 | } | |
454 | wait_on_page_writeback(page); | |
455 | if (page_mapped(page)) { | |
456 | if (!did_range_unmap) { | |
457 | /* | |
458 | * Zap the rest of the file in one hit. | |
459 | */ | |
460 | unmap_mapping_range(mapping, | |
461 | (loff_t)page_index<<PAGE_CACHE_SHIFT, | |
462 | (loff_t)(end - page_index + 1) | |
463 | << PAGE_CACHE_SHIFT, | |
464 | 0); | |
465 | did_range_unmap = 1; | |
466 | } else { | |
467 | /* | |
468 | * Just zap this page | |
469 | */ | |
470 | unmap_mapping_range(mapping, | |
471 | (loff_t)page_index<<PAGE_CACHE_SHIFT, | |
472 | PAGE_CACHE_SIZE, 0); | |
473 | } | |
474 | } | |
475 | BUG_ON(page_mapped(page)); | |
476 | ret2 = do_launder_page(mapping, page); | |
477 | if (ret2 == 0) { | |
478 | if (!invalidate_complete_page2(mapping, page)) | |
479 | ret2 = -EBUSY; | |
480 | } | |
481 | if (ret2 < 0) | |
482 | ret = ret2; | |
483 | unlock_page(page); | |
484 | } | |
485 | pagevec_release(&pvec); | |
486 | mem_cgroup_uncharge_end(); | |
487 | cond_resched(); | |
488 | } | |
489 | return ret; | |
490 | } | |
491 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
492 | ||
493 | /** | |
494 | * invalidate_inode_pages2 - remove all pages from an address_space | |
495 | * @mapping: the address_space | |
496 | * | |
497 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
498 | * invalidation. | |
499 | * | |
500 | * Returns -EBUSY if any pages could not be invalidated. | |
501 | */ | |
502 | int invalidate_inode_pages2(struct address_space *mapping) | |
503 | { | |
504 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
505 | } | |
506 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
507 | ||
508 | /** | |
509 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
510 | * @inode: inode | |
511 | * @old: old file offset | |
512 | * @new: new file offset | |
513 | * | |
514 | * inode's new i_size must already be written before truncate_pagecache | |
515 | * is called. | |
516 | * | |
517 | * This function should typically be called before the filesystem | |
518 | * releases resources associated with the freed range (eg. deallocates | |
519 | * blocks). This way, pagecache will always stay logically coherent | |
520 | * with on-disk format, and the filesystem would not have to deal with | |
521 | * situations such as writepage being called for a page that has already | |
522 | * had its underlying blocks deallocated. | |
523 | */ | |
524 | void truncate_pagecache(struct inode *inode, loff_t old, loff_t new) | |
525 | { | |
526 | struct address_space *mapping = inode->i_mapping; | |
527 | ||
528 | /* | |
529 | * unmap_mapping_range is called twice, first simply for | |
530 | * efficiency so that truncate_inode_pages does fewer | |
531 | * single-page unmaps. However after this first call, and | |
532 | * before truncate_inode_pages finishes, it is possible for | |
533 | * private pages to be COWed, which remain after | |
534 | * truncate_inode_pages finishes, hence the second | |
535 | * unmap_mapping_range call must be made for correctness. | |
536 | */ | |
537 | unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1); | |
538 | truncate_inode_pages(mapping, new); | |
539 | unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1); | |
540 | } | |
541 | EXPORT_SYMBOL(truncate_pagecache); | |
542 | ||
543 | /** | |
544 | * truncate_setsize - update inode and pagecache for a new file size | |
545 | * @inode: inode | |
546 | * @newsize: new file size | |
547 | * | |
548 | * truncate_setsize updastes i_size update and performs pagecache | |
549 | * truncation (if necessary) for a file size updates. It will be | |
550 | * typically be called from the filesystem's setattr function when | |
551 | * ATTR_SIZE is passed in. | |
552 | * | |
553 | * Must be called with inode_mutex held and after all filesystem | |
554 | * specific block truncation has been performed. | |
555 | */ | |
556 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
557 | { | |
558 | loff_t oldsize; | |
559 | ||
560 | oldsize = inode->i_size; | |
561 | i_size_write(inode, newsize); | |
562 | ||
563 | truncate_pagecache(inode, oldsize, newsize); | |
564 | } | |
565 | EXPORT_SYMBOL(truncate_setsize); | |
566 | ||
567 | /** | |
568 | * vmtruncate - unmap mappings "freed" by truncate() syscall | |
569 | * @inode: inode of the file used | |
570 | * @offset: file offset to start truncating | |
571 | * | |
572 | * This function is deprecated and truncate_setsize or truncate_pagecache | |
573 | * should be used instead, together with filesystem specific block truncation. | |
574 | */ | |
575 | int vmtruncate(struct inode *inode, loff_t offset) | |
576 | { | |
577 | int error; | |
578 | ||
579 | error = inode_newsize_ok(inode, offset); | |
580 | if (error) | |
581 | return error; | |
582 | ||
583 | truncate_setsize(inode, offset); | |
584 | if (inode->i_op->truncate) | |
585 | inode->i_op->truncate(inode); | |
586 | return 0; | |
587 | } | |
588 | EXPORT_SYMBOL(vmtruncate); |