Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/truncate.c - code for taking down pages from address_spaces | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds | |
5 | * | |
e1f8e874 | 6 | * 10Sep2002 Andrew Morton |
1da177e4 LT |
7 | * Initial version. |
8 | */ | |
9 | ||
10 | #include <linux/kernel.h> | |
4af3c9cc | 11 | #include <linux/backing-dev.h> |
5a0e3ad6 | 12 | #include <linux/gfp.h> |
1da177e4 | 13 | #include <linux/mm.h> |
0fd0e6b0 | 14 | #include <linux/swap.h> |
1da177e4 LT |
15 | #include <linux/module.h> |
16 | #include <linux/pagemap.h> | |
01f2705d | 17 | #include <linux/highmem.h> |
1da177e4 | 18 | #include <linux/pagevec.h> |
e08748ce | 19 | #include <linux/task_io_accounting_ops.h> |
1da177e4 | 20 | #include <linux/buffer_head.h> /* grr. try_to_release_page, |
aaa4059b | 21 | do_invalidatepage */ |
ba470de4 | 22 | #include "internal.h" |
1da177e4 LT |
23 | |
24 | ||
cf9a2ae8 | 25 | /** |
28bc44d7 | 26 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 DH |
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; | |
9361401e | 43 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
44 | if (!invalidatepage) |
45 | invalidatepage = block_invalidatepage; | |
9361401e | 46 | #endif |
cf9a2ae8 DH |
47 | if (invalidatepage) |
48 | (*invalidatepage)(page, offset); | |
49 | } | |
50 | ||
1da177e4 LT |
51 | static inline void truncate_partial_page(struct page *page, unsigned partial) |
52 | { | |
eebd2aa3 | 53 | zero_user_segment(page, partial, PAGE_CACHE_SIZE); |
266cf658 | 54 | if (page_has_private(page)) |
1da177e4 LT |
55 | do_invalidatepage(page, partial); |
56 | } | |
57 | ||
ecdfc978 LT |
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 | */ | |
fba2591b LT |
72 | void cancel_dirty_page(struct page *page, unsigned int account_size) |
73 | { | |
8368e328 LT |
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); | |
c9e51e41 PZ |
78 | dec_bdi_stat(mapping->backing_dev_info, |
79 | BDI_RECLAIMABLE); | |
8368e328 LT |
80 | if (account_size) |
81 | task_io_account_cancelled_write(account_size); | |
82 | } | |
3e67c098 | 83 | } |
fba2591b | 84 | } |
8368e328 | 85 | EXPORT_SYMBOL(cancel_dirty_page); |
fba2591b | 86 | |
1da177e4 LT |
87 | /* |
88 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 89 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 90 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
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 | |
fc0ecff6 | 94 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
95 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
96 | */ | |
750b4987 | 97 | static int |
1da177e4 LT |
98 | truncate_complete_page(struct address_space *mapping, struct page *page) |
99 | { | |
100 | if (page->mapping != mapping) | |
750b4987 | 101 | return -EIO; |
1da177e4 | 102 | |
266cf658 | 103 | if (page_has_private(page)) |
1da177e4 LT |
104 | do_invalidatepage(page, 0); |
105 | ||
a2b34564 BS |
106 | cancel_dirty_page(page, PAGE_CACHE_SIZE); |
107 | ||
ba470de4 | 108 | clear_page_mlock(page); |
1da177e4 | 109 | ClearPageMappedToDisk(page); |
5adc7b51 | 110 | delete_from_page_cache(page); |
750b4987 | 111 | return 0; |
1da177e4 LT |
112 | } |
113 | ||
114 | /* | |
fc0ecff6 | 115 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 116 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
117 | * be marked dirty at any time too, so use remove_mapping which safely |
118 | * discards clean, unused pages. | |
1da177e4 LT |
119 | * |
120 | * Returns non-zero if the page was successfully invalidated. | |
121 | */ | |
122 | static int | |
123 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
124 | { | |
0fd0e6b0 NP |
125 | int ret; |
126 | ||
1da177e4 LT |
127 | if (page->mapping != mapping) |
128 | return 0; | |
129 | ||
266cf658 | 130 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
131 | return 0; |
132 | ||
ba470de4 | 133 | clear_page_mlock(page); |
0fd0e6b0 | 134 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
135 | |
136 | return ret; | |
1da177e4 LT |
137 | } |
138 | ||
750b4987 NP |
139 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
140 | { | |
141 | if (page_mapped(page)) { | |
142 | unmap_mapping_range(mapping, | |
143 | (loff_t)page->index << PAGE_CACHE_SHIFT, | |
144 | PAGE_CACHE_SIZE, 0); | |
145 | } | |
146 | return truncate_complete_page(mapping, page); | |
147 | } | |
148 | ||
25718736 AK |
149 | /* |
150 | * Used to get rid of pages on hardware memory corruption. | |
151 | */ | |
152 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
153 | { | |
154 | if (!mapping) | |
155 | return -EINVAL; | |
156 | /* | |
157 | * Only punch for normal data pages for now. | |
158 | * Handling other types like directories would need more auditing. | |
159 | */ | |
160 | if (!S_ISREG(mapping->host->i_mode)) | |
161 | return -EIO; | |
162 | return truncate_inode_page(mapping, page); | |
163 | } | |
164 | EXPORT_SYMBOL(generic_error_remove_page); | |
165 | ||
83f78668 WF |
166 | /* |
167 | * Safely invalidate one page from its pagecache mapping. | |
168 | * It only drops clean, unused pages. The page must be locked. | |
169 | * | |
170 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
171 | */ | |
172 | int invalidate_inode_page(struct page *page) | |
173 | { | |
174 | struct address_space *mapping = page_mapping(page); | |
175 | if (!mapping) | |
176 | return 0; | |
177 | if (PageDirty(page) || PageWriteback(page)) | |
178 | return 0; | |
179 | if (page_mapped(page)) | |
180 | return 0; | |
181 | return invalidate_complete_page(mapping, page); | |
182 | } | |
183 | ||
1da177e4 | 184 | /** |
0643245f | 185 | * truncate_inode_pages - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
186 | * @mapping: mapping to truncate |
187 | * @lstart: offset from which to truncate | |
d7339071 | 188 | * @lend: offset to which to truncate |
1da177e4 | 189 | * |
d7339071 HR |
190 | * Truncate the page cache, removing the pages that are between |
191 | * specified offsets (and zeroing out partial page | |
192 | * (if lstart is not page aligned)). | |
1da177e4 LT |
193 | * |
194 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
195 | * block on page locks and it will not block on writeback. The second pass | |
196 | * will wait. This is to prevent as much IO as possible in the affected region. | |
197 | * The first pass will remove most pages, so the search cost of the second pass | |
198 | * is low. | |
199 | * | |
200 | * When looking at page->index outside the page lock we need to be careful to | |
201 | * copy it into a local to avoid races (it could change at any time). | |
202 | * | |
203 | * We pass down the cache-hot hint to the page freeing code. Even if the | |
204 | * mapping is large, it is probably the case that the final pages are the most | |
205 | * recently touched, and freeing happens in ascending file offset order. | |
1da177e4 | 206 | */ |
d7339071 HR |
207 | void truncate_inode_pages_range(struct address_space *mapping, |
208 | loff_t lstart, loff_t lend) | |
1da177e4 LT |
209 | { |
210 | const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; | |
d7339071 | 211 | pgoff_t end; |
1da177e4 LT |
212 | const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1); |
213 | struct pagevec pvec; | |
214 | pgoff_t next; | |
215 | int i; | |
216 | ||
217 | if (mapping->nrpages == 0) | |
218 | return; | |
219 | ||
d7339071 HR |
220 | BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1)); |
221 | end = (lend >> PAGE_CACHE_SHIFT); | |
222 | ||
1da177e4 LT |
223 | pagevec_init(&pvec, 0); |
224 | next = start; | |
d7339071 HR |
225 | while (next <= end && |
226 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
e5598f8b | 227 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
228 | for (i = 0; i < pagevec_count(&pvec); i++) { |
229 | struct page *page = pvec.pages[i]; | |
230 | pgoff_t page_index = page->index; | |
231 | ||
d7339071 HR |
232 | if (page_index > end) { |
233 | next = page_index; | |
234 | break; | |
235 | } | |
236 | ||
1da177e4 LT |
237 | if (page_index > next) |
238 | next = page_index; | |
239 | next++; | |
529ae9aa | 240 | if (!trylock_page(page)) |
1da177e4 LT |
241 | continue; |
242 | if (PageWriteback(page)) { | |
243 | unlock_page(page); | |
244 | continue; | |
245 | } | |
750b4987 | 246 | truncate_inode_page(mapping, page); |
1da177e4 LT |
247 | unlock_page(page); |
248 | } | |
249 | pagevec_release(&pvec); | |
e5598f8b | 250 | mem_cgroup_uncharge_end(); |
1da177e4 LT |
251 | cond_resched(); |
252 | } | |
253 | ||
254 | if (partial) { | |
255 | struct page *page = find_lock_page(mapping, start - 1); | |
256 | if (page) { | |
257 | wait_on_page_writeback(page); | |
258 | truncate_partial_page(page, partial); | |
259 | unlock_page(page); | |
260 | page_cache_release(page); | |
261 | } | |
262 | } | |
263 | ||
264 | next = start; | |
265 | for ( ; ; ) { | |
266 | cond_resched(); | |
267 | if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
268 | if (next == start) | |
269 | break; | |
270 | next = start; | |
271 | continue; | |
272 | } | |
d7339071 HR |
273 | if (pvec.pages[0]->index > end) { |
274 | pagevec_release(&pvec); | |
275 | break; | |
276 | } | |
569b846d | 277 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
278 | for (i = 0; i < pagevec_count(&pvec); i++) { |
279 | struct page *page = pvec.pages[i]; | |
280 | ||
d7339071 HR |
281 | if (page->index > end) |
282 | break; | |
1da177e4 LT |
283 | lock_page(page); |
284 | wait_on_page_writeback(page); | |
750b4987 | 285 | truncate_inode_page(mapping, page); |
1da177e4 LT |
286 | if (page->index > next) |
287 | next = page->index; | |
288 | next++; | |
1da177e4 LT |
289 | unlock_page(page); |
290 | } | |
291 | pagevec_release(&pvec); | |
569b846d | 292 | mem_cgroup_uncharge_end(); |
1da177e4 LT |
293 | } |
294 | } | |
d7339071 | 295 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 296 | |
d7339071 HR |
297 | /** |
298 | * truncate_inode_pages - truncate *all* the pages from an offset | |
299 | * @mapping: mapping to truncate | |
300 | * @lstart: offset from which to truncate | |
301 | * | |
1b1dcc1b | 302 | * Called under (and serialised by) inode->i_mutex. |
d7339071 HR |
303 | */ |
304 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
305 | { | |
306 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
307 | } | |
1da177e4 LT |
308 | EXPORT_SYMBOL(truncate_inode_pages); |
309 | ||
28697355 MW |
310 | /** |
311 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
312 | * @mapping: the address_space which holds the pages to invalidate | |
313 | * @start: the offset 'from' which to invalidate | |
314 | * @end: the offset 'to' which to invalidate (inclusive) | |
315 | * | |
316 | * This function only removes the unlocked pages, if you want to | |
317 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
318 | * | |
319 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
320 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
321 | * pagetables. | |
322 | */ | |
323 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
324 | pgoff_t start, pgoff_t end) | |
1da177e4 LT |
325 | { |
326 | struct pagevec pvec; | |
327 | pgoff_t next = start; | |
328 | unsigned long ret = 0; | |
329 | int i; | |
330 | ||
331 | pagevec_init(&pvec, 0); | |
332 | while (next <= end && | |
333 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
569b846d | 334 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
335 | for (i = 0; i < pagevec_count(&pvec); i++) { |
336 | struct page *page = pvec.pages[i]; | |
e0f23603 N |
337 | pgoff_t index; |
338 | int lock_failed; | |
1da177e4 | 339 | |
529ae9aa | 340 | lock_failed = !trylock_page(page); |
e0f23603 N |
341 | |
342 | /* | |
343 | * We really shouldn't be looking at the ->index of an | |
344 | * unlocked page. But we're not allowed to lock these | |
345 | * pages. So we rely upon nobody altering the ->index | |
346 | * of this (pinned-by-us) page. | |
347 | */ | |
348 | index = page->index; | |
349 | if (index > next) | |
350 | next = index; | |
1da177e4 | 351 | next++; |
e0f23603 N |
352 | if (lock_failed) |
353 | continue; | |
354 | ||
83f78668 WF |
355 | ret += invalidate_inode_page(page); |
356 | ||
1da177e4 LT |
357 | unlock_page(page); |
358 | if (next > end) | |
359 | break; | |
360 | } | |
361 | pagevec_release(&pvec); | |
569b846d | 362 | mem_cgroup_uncharge_end(); |
28697355 | 363 | cond_resched(); |
1da177e4 LT |
364 | } |
365 | return ret; | |
366 | } | |
54bc4855 | 367 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 368 | |
bd4c8ce4 AM |
369 | /* |
370 | * This is like invalidate_complete_page(), except it ignores the page's | |
371 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
372 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
373 | * shrink_page_list() has a temp ref on them, or because they're transiently |
374 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
375 | */ |
376 | static int | |
377 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
378 | { | |
379 | if (page->mapping != mapping) | |
380 | return 0; | |
381 | ||
266cf658 | 382 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
383 | return 0; |
384 | ||
19fd6231 | 385 | spin_lock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
386 | if (PageDirty(page)) |
387 | goto failed; | |
388 | ||
ba470de4 | 389 | clear_page_mlock(page); |
266cf658 | 390 | BUG_ON(page_has_private(page)); |
e64a782f | 391 | __delete_from_page_cache(page); |
19fd6231 | 392 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 393 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
394 | |
395 | if (mapping->a_ops->freepage) | |
396 | mapping->a_ops->freepage(page); | |
397 | ||
bd4c8ce4 AM |
398 | page_cache_release(page); /* pagecache ref */ |
399 | return 1; | |
400 | failed: | |
19fd6231 | 401 | spin_unlock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
402 | return 0; |
403 | } | |
404 | ||
e3db7691 TM |
405 | static int do_launder_page(struct address_space *mapping, struct page *page) |
406 | { | |
407 | if (!PageDirty(page)) | |
408 | return 0; | |
409 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
410 | return 0; | |
411 | return mapping->a_ops->launder_page(page); | |
412 | } | |
413 | ||
1da177e4 LT |
414 | /** |
415 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 416 | * @mapping: the address_space |
1da177e4 LT |
417 | * @start: the page offset 'from' which to invalidate |
418 | * @end: the page offset 'to' which to invalidate (inclusive) | |
419 | * | |
420 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
421 | * invalidation. | |
422 | * | |
6ccfa806 | 423 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
424 | */ |
425 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
426 | pgoff_t start, pgoff_t end) | |
427 | { | |
428 | struct pagevec pvec; | |
429 | pgoff_t next; | |
430 | int i; | |
431 | int ret = 0; | |
0dd1334f | 432 | int ret2 = 0; |
1da177e4 LT |
433 | int did_range_unmap = 0; |
434 | int wrapped = 0; | |
435 | ||
436 | pagevec_init(&pvec, 0); | |
437 | next = start; | |
7b965e08 | 438 | while (next <= end && !wrapped && |
1da177e4 LT |
439 | pagevec_lookup(&pvec, mapping, next, |
440 | min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
569b846d | 441 | mem_cgroup_uncharge_start(); |
7b965e08 | 442 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 LT |
443 | struct page *page = pvec.pages[i]; |
444 | pgoff_t page_index; | |
1da177e4 LT |
445 | |
446 | lock_page(page); | |
447 | if (page->mapping != mapping) { | |
448 | unlock_page(page); | |
449 | continue; | |
450 | } | |
451 | page_index = page->index; | |
452 | next = page_index + 1; | |
453 | if (next == 0) | |
454 | wrapped = 1; | |
455 | if (page_index > end) { | |
456 | unlock_page(page); | |
457 | break; | |
458 | } | |
459 | wait_on_page_writeback(page); | |
d00806b1 | 460 | if (page_mapped(page)) { |
1da177e4 LT |
461 | if (!did_range_unmap) { |
462 | /* | |
463 | * Zap the rest of the file in one hit. | |
464 | */ | |
465 | unmap_mapping_range(mapping, | |
479ef592 OD |
466 | (loff_t)page_index<<PAGE_CACHE_SHIFT, |
467 | (loff_t)(end - page_index + 1) | |
1da177e4 LT |
468 | << PAGE_CACHE_SHIFT, |
469 | 0); | |
470 | did_range_unmap = 1; | |
471 | } else { | |
472 | /* | |
473 | * Just zap this page | |
474 | */ | |
475 | unmap_mapping_range(mapping, | |
479ef592 | 476 | (loff_t)page_index<<PAGE_CACHE_SHIFT, |
1da177e4 LT |
477 | PAGE_CACHE_SIZE, 0); |
478 | } | |
479 | } | |
d00806b1 | 480 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
481 | ret2 = do_launder_page(mapping, page); |
482 | if (ret2 == 0) { | |
483 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 484 | ret2 = -EBUSY; |
0dd1334f HH |
485 | } |
486 | if (ret2 < 0) | |
487 | ret = ret2; | |
1da177e4 LT |
488 | unlock_page(page); |
489 | } | |
490 | pagevec_release(&pvec); | |
569b846d | 491 | mem_cgroup_uncharge_end(); |
1da177e4 LT |
492 | cond_resched(); |
493 | } | |
494 | return ret; | |
495 | } | |
496 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
497 | ||
498 | /** | |
499 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 500 | * @mapping: the address_space |
1da177e4 LT |
501 | * |
502 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
503 | * invalidation. | |
504 | * | |
e9de25dd | 505 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
506 | */ |
507 | int invalidate_inode_pages2(struct address_space *mapping) | |
508 | { | |
509 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
510 | } | |
511 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 512 | |
513 | /** | |
514 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
515 | * @inode: inode | |
516 | * @old: old file offset | |
517 | * @new: new file offset | |
518 | * | |
519 | * inode's new i_size must already be written before truncate_pagecache | |
520 | * is called. | |
521 | * | |
522 | * This function should typically be called before the filesystem | |
523 | * releases resources associated with the freed range (eg. deallocates | |
524 | * blocks). This way, pagecache will always stay logically coherent | |
525 | * with on-disk format, and the filesystem would not have to deal with | |
526 | * situations such as writepage being called for a page that has already | |
527 | * had its underlying blocks deallocated. | |
528 | */ | |
529 | void truncate_pagecache(struct inode *inode, loff_t old, loff_t new) | |
530 | { | |
cedabed4 OH |
531 | struct address_space *mapping = inode->i_mapping; |
532 | ||
533 | /* | |
534 | * unmap_mapping_range is called twice, first simply for | |
535 | * efficiency so that truncate_inode_pages does fewer | |
536 | * single-page unmaps. However after this first call, and | |
537 | * before truncate_inode_pages finishes, it is possible for | |
538 | * private pages to be COWed, which remain after | |
539 | * truncate_inode_pages finishes, hence the second | |
540 | * unmap_mapping_range call must be made for correctness. | |
541 | */ | |
542 | unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1); | |
543 | truncate_inode_pages(mapping, new); | |
544 | unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1); | |
25d9e2d1 | 545 | } |
546 | EXPORT_SYMBOL(truncate_pagecache); | |
547 | ||
2c27c65e CH |
548 | /** |
549 | * truncate_setsize - update inode and pagecache for a new file size | |
550 | * @inode: inode | |
551 | * @newsize: new file size | |
552 | * | |
382e27da JK |
553 | * truncate_setsize updates i_size and performs pagecache truncation (if |
554 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
555 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 556 | * |
382e27da JK |
557 | * Must be called with inode_mutex held and before all filesystem specific |
558 | * block truncation has been performed. | |
2c27c65e CH |
559 | */ |
560 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
561 | { | |
562 | loff_t oldsize; | |
563 | ||
564 | oldsize = inode->i_size; | |
565 | i_size_write(inode, newsize); | |
566 | ||
567 | truncate_pagecache(inode, oldsize, newsize); | |
568 | } | |
569 | EXPORT_SYMBOL(truncate_setsize); | |
570 | ||
25d9e2d1 | 571 | /** |
572 | * vmtruncate - unmap mappings "freed" by truncate() syscall | |
573 | * @inode: inode of the file used | |
574 | * @offset: file offset to start truncating | |
575 | * | |
2c27c65e CH |
576 | * This function is deprecated and truncate_setsize or truncate_pagecache |
577 | * should be used instead, together with filesystem specific block truncation. | |
25d9e2d1 | 578 | */ |
579 | int vmtruncate(struct inode *inode, loff_t offset) | |
580 | { | |
25d9e2d1 | 581 | int error; |
582 | ||
2c27c65e | 583 | error = inode_newsize_ok(inode, offset); |
25d9e2d1 | 584 | if (error) |
585 | return error; | |
7bb46a67 | 586 | |
2c27c65e | 587 | truncate_setsize(inode, offset); |
25d9e2d1 | 588 | if (inode->i_op->truncate) |
589 | inode->i_op->truncate(inode); | |
2c27c65e | 590 | return 0; |
25d9e2d1 | 591 | } |
592 | EXPORT_SYMBOL(vmtruncate); |