Root/
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 | mem_cgroup_uncharge_start(); |
229 | for (i = 0; i < pagevec_count(&pvec); i++) { |
230 | struct page *page = pvec.pages[i]; |
231 | pgoff_t page_index = page->index; |
232 | |
233 | if (page_index > end) { |
234 | next = page_index; |
235 | break; |
236 | } |
237 | |
238 | if (page_index > next) |
239 | next = page_index; |
240 | next++; |
241 | if (!trylock_page(page)) |
242 | continue; |
243 | if (PageWriteback(page)) { |
244 | unlock_page(page); |
245 | continue; |
246 | } |
247 | truncate_inode_page(mapping, page); |
248 | unlock_page(page); |
249 | } |
250 | pagevec_release(&pvec); |
251 | mem_cgroup_uncharge_end(); |
252 | cond_resched(); |
253 | } |
254 | |
255 | if (partial) { |
256 | struct page *page = find_lock_page(mapping, start - 1); |
257 | if (page) { |
258 | wait_on_page_writeback(page); |
259 | truncate_partial_page(page, partial); |
260 | unlock_page(page); |
261 | page_cache_release(page); |
262 | } |
263 | } |
264 | |
265 | next = start; |
266 | for ( ; ; ) { |
267 | cond_resched(); |
268 | if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { |
269 | if (next == start) |
270 | break; |
271 | next = start; |
272 | continue; |
273 | } |
274 | if (pvec.pages[0]->index > end) { |
275 | pagevec_release(&pvec); |
276 | break; |
277 | } |
278 | mem_cgroup_uncharge_start(); |
279 | for (i = 0; i < pagevec_count(&pvec); i++) { |
280 | struct page *page = pvec.pages[i]; |
281 | |
282 | if (page->index > end) |
283 | break; |
284 | lock_page(page); |
285 | wait_on_page_writeback(page); |
286 | truncate_inode_page(mapping, page); |
287 | if (page->index > next) |
288 | next = page->index; |
289 | next++; |
290 | unlock_page(page); |
291 | } |
292 | pagevec_release(&pvec); |
293 | mem_cgroup_uncharge_end(); |
294 | } |
295 | } |
296 | EXPORT_SYMBOL(truncate_inode_pages_range); |
297 | |
298 | /** |
299 | * truncate_inode_pages - truncate *all* the pages from an offset |
300 | * @mapping: mapping to truncate |
301 | * @lstart: offset from which to truncate |
302 | * |
303 | * Called under (and serialised by) inode->i_mutex. |
304 | */ |
305 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) |
306 | { |
307 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); |
308 | } |
309 | EXPORT_SYMBOL(truncate_inode_pages); |
310 | |
311 | /** |
312 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode |
313 | * @mapping: the address_space which holds the pages to invalidate |
314 | * @start: the offset 'from' which to invalidate |
315 | * @end: the offset 'to' which to invalidate (inclusive) |
316 | * |
317 | * This function only removes the unlocked pages, if you want to |
318 | * remove all the pages of one inode, you must call truncate_inode_pages. |
319 | * |
320 | * invalidate_mapping_pages() will not block on IO activity. It will not |
321 | * invalidate pages which are dirty, locked, under writeback or mapped into |
322 | * pagetables. |
323 | */ |
324 | unsigned long invalidate_mapping_pages(struct address_space *mapping, |
325 | pgoff_t start, pgoff_t end) |
326 | { |
327 | struct pagevec pvec; |
328 | pgoff_t next = start; |
329 | unsigned long ret = 0; |
330 | int i; |
331 | |
332 | pagevec_init(&pvec, 0); |
333 | while (next <= end && |
334 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { |
335 | mem_cgroup_uncharge_start(); |
336 | for (i = 0; i < pagevec_count(&pvec); i++) { |
337 | struct page *page = pvec.pages[i]; |
338 | pgoff_t index; |
339 | int lock_failed; |
340 | |
341 | lock_failed = !trylock_page(page); |
342 | |
343 | /* |
344 | * We really shouldn't be looking at the ->index of an |
345 | * unlocked page. But we're not allowed to lock these |
346 | * pages. So we rely upon nobody altering the ->index |
347 | * of this (pinned-by-us) page. |
348 | */ |
349 | index = page->index; |
350 | if (index > next) |
351 | next = index; |
352 | next++; |
353 | if (lock_failed) |
354 | continue; |
355 | |
356 | ret += invalidate_inode_page(page); |
357 | |
358 | unlock_page(page); |
359 | if (next > end) |
360 | break; |
361 | } |
362 | pagevec_release(&pvec); |
363 | mem_cgroup_uncharge_end(); |
364 | cond_resched(); |
365 | } |
366 | return ret; |
367 | } |
368 | EXPORT_SYMBOL(invalidate_mapping_pages); |
369 | |
370 | /* |
371 | * This is like invalidate_complete_page(), except it ignores the page's |
372 | * refcount. We do this because invalidate_inode_pages2() needs stronger |
373 | * invalidation guarantees, and cannot afford to leave pages behind because |
374 | * shrink_page_list() has a temp ref on them, or because they're transiently |
375 | * sitting in the lru_cache_add() pagevecs. |
376 | */ |
377 | static int |
378 | invalidate_complete_page2(struct address_space *mapping, struct page *page) |
379 | { |
380 | if (page->mapping != mapping) |
381 | return 0; |
382 | |
383 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
384 | return 0; |
385 | |
386 | spin_lock_irq(&mapping->tree_lock); |
387 | if (PageDirty(page)) |
388 | goto failed; |
389 | |
390 | clear_page_mlock(page); |
391 | BUG_ON(page_has_private(page)); |
392 | __remove_from_page_cache(page); |
393 | spin_unlock_irq(&mapping->tree_lock); |
394 | mem_cgroup_uncharge_cache_page(page); |
395 | |
396 | if (mapping->a_ops->freepage) |
397 | mapping->a_ops->freepage(page); |
398 | |
399 | page_cache_release(page); /* pagecache ref */ |
400 | return 1; |
401 | failed: |
402 | spin_unlock_irq(&mapping->tree_lock); |
403 | return 0; |
404 | } |
405 | |
406 | static int do_launder_page(struct address_space *mapping, struct page *page) |
407 | { |
408 | if (!PageDirty(page)) |
409 | return 0; |
410 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) |
411 | return 0; |
412 | return mapping->a_ops->launder_page(page); |
413 | } |
414 | |
415 | /** |
416 | * invalidate_inode_pages2_range - remove range of pages from an address_space |
417 | * @mapping: the address_space |
418 | * @start: the page offset 'from' which to invalidate |
419 | * @end: the page offset 'to' which to invalidate (inclusive) |
420 | * |
421 | * Any pages which are found to be mapped into pagetables are unmapped prior to |
422 | * invalidation. |
423 | * |
424 | * Returns -EBUSY if any pages could not be invalidated. |
425 | */ |
426 | int invalidate_inode_pages2_range(struct address_space *mapping, |
427 | pgoff_t start, pgoff_t end) |
428 | { |
429 | struct pagevec pvec; |
430 | pgoff_t next; |
431 | int i; |
432 | int ret = 0; |
433 | int ret2 = 0; |
434 | int did_range_unmap = 0; |
435 | int wrapped = 0; |
436 | |
437 | pagevec_init(&pvec, 0); |
438 | next = start; |
439 | while (next <= end && !wrapped && |
440 | pagevec_lookup(&pvec, mapping, next, |
441 | min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { |
442 | mem_cgroup_uncharge_start(); |
443 | for (i = 0; i < pagevec_count(&pvec); i++) { |
444 | struct page *page = pvec.pages[i]; |
445 | pgoff_t page_index; |
446 | |
447 | lock_page(page); |
448 | if (page->mapping != mapping) { |
449 | unlock_page(page); |
450 | continue; |
451 | } |
452 | page_index = page->index; |
453 | next = page_index + 1; |
454 | if (next == 0) |
455 | wrapped = 1; |
456 | if (page_index > end) { |
457 | unlock_page(page); |
458 | break; |
459 | } |
460 | wait_on_page_writeback(page); |
461 | if (page_mapped(page)) { |
462 | if (!did_range_unmap) { |
463 | /* |
464 | * Zap the rest of the file in one hit. |
465 | */ |
466 | unmap_mapping_range(mapping, |
467 | (loff_t)page_index<<PAGE_CACHE_SHIFT, |
468 | (loff_t)(end - page_index + 1) |
469 | << PAGE_CACHE_SHIFT, |
470 | 0); |
471 | did_range_unmap = 1; |
472 | } else { |
473 | /* |
474 | * Just zap this page |
475 | */ |
476 | unmap_mapping_range(mapping, |
477 | (loff_t)page_index<<PAGE_CACHE_SHIFT, |
478 | PAGE_CACHE_SIZE, 0); |
479 | } |
480 | } |
481 | BUG_ON(page_mapped(page)); |
482 | ret2 = do_launder_page(mapping, page); |
483 | if (ret2 == 0) { |
484 | if (!invalidate_complete_page2(mapping, page)) |
485 | ret2 = -EBUSY; |
486 | } |
487 | if (ret2 < 0) |
488 | ret = ret2; |
489 | unlock_page(page); |
490 | } |
491 | pagevec_release(&pvec); |
492 | mem_cgroup_uncharge_end(); |
493 | cond_resched(); |
494 | } |
495 | return ret; |
496 | } |
497 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); |
498 | |
499 | /** |
500 | * invalidate_inode_pages2 - remove all pages from an address_space |
501 | * @mapping: the address_space |
502 | * |
503 | * Any pages which are found to be mapped into pagetables are unmapped prior to |
504 | * invalidation. |
505 | * |
506 | * Returns -EBUSY if any pages could not be invalidated. |
507 | */ |
508 | int invalidate_inode_pages2(struct address_space *mapping) |
509 | { |
510 | return invalidate_inode_pages2_range(mapping, 0, -1); |
511 | } |
512 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); |
513 | |
514 | /** |
515 | * truncate_pagecache - unmap and remove pagecache that has been truncated |
516 | * @inode: inode |
517 | * @old: old file offset |
518 | * @new: new file offset |
519 | * |
520 | * inode's new i_size must already be written before truncate_pagecache |
521 | * is called. |
522 | * |
523 | * This function should typically be called before the filesystem |
524 | * releases resources associated with the freed range (eg. deallocates |
525 | * blocks). This way, pagecache will always stay logically coherent |
526 | * with on-disk format, and the filesystem would not have to deal with |
527 | * situations such as writepage being called for a page that has already |
528 | * had its underlying blocks deallocated. |
529 | */ |
530 | void truncate_pagecache(struct inode *inode, loff_t old, loff_t new) |
531 | { |
532 | struct address_space *mapping = inode->i_mapping; |
533 | |
534 | /* |
535 | * unmap_mapping_range is called twice, first simply for |
536 | * efficiency so that truncate_inode_pages does fewer |
537 | * single-page unmaps. However after this first call, and |
538 | * before truncate_inode_pages finishes, it is possible for |
539 | * private pages to be COWed, which remain after |
540 | * truncate_inode_pages finishes, hence the second |
541 | * unmap_mapping_range call must be made for correctness. |
542 | */ |
543 | unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1); |
544 | truncate_inode_pages(mapping, new); |
545 | unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1); |
546 | } |
547 | EXPORT_SYMBOL(truncate_pagecache); |
548 | |
549 | /** |
550 | * truncate_setsize - update inode and pagecache for a new file size |
551 | * @inode: inode |
552 | * @newsize: new file size |
553 | * |
554 | * truncate_setsize updates i_size and performs pagecache truncation (if |
555 | * necessary) to @newsize. It will be typically be called from the filesystem's |
556 | * setattr function when ATTR_SIZE is passed in. |
557 | * |
558 | * Must be called with inode_mutex held and before all filesystem specific |
559 | * block truncation has been performed. |
560 | */ |
561 | void truncate_setsize(struct inode *inode, loff_t newsize) |
562 | { |
563 | loff_t oldsize; |
564 | |
565 | oldsize = inode->i_size; |
566 | i_size_write(inode, newsize); |
567 | |
568 | truncate_pagecache(inode, oldsize, newsize); |
569 | } |
570 | EXPORT_SYMBOL(truncate_setsize); |
571 | |
572 | /** |
573 | * vmtruncate - unmap mappings "freed" by truncate() syscall |
574 | * @inode: inode of the file used |
575 | * @offset: file offset to start truncating |
576 | * |
577 | * This function is deprecated and truncate_setsize or truncate_pagecache |
578 | * should be used instead, together with filesystem specific block truncation. |
579 | */ |
580 | int vmtruncate(struct inode *inode, loff_t offset) |
581 | { |
582 | int error; |
583 | |
584 | error = inode_newsize_ok(inode, offset); |
585 | if (error) |
586 | return error; |
587 | |
588 | truncate_setsize(inode, offset); |
589 | if (inode->i_op->truncate) |
590 | inode->i_op->truncate(inode); |
591 | return 0; |
592 | } |
593 | EXPORT_SYMBOL(vmtruncate); |
594 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9