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> |
b95f1b31 | 15 | #include <linux/export.h> |
1da177e4 | 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 */ |
c515e1fd | 22 | #include <linux/cleancache.h> |
ba470de4 | 23 | #include "internal.h" |
1da177e4 LT |
24 | |
25 | ||
cf9a2ae8 | 26 | /** |
28bc44d7 | 27 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 28 | * @page: the page which is affected |
d47992f8 LC |
29 | * @offset: start of the range to invalidate |
30 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
31 | * |
32 | * do_invalidatepage() is called when all or part of the page has become | |
33 | * invalidated by a truncate operation. | |
34 | * | |
35 | * do_invalidatepage() does not have to release all buffers, but it must | |
36 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
37 | * is underway against any of the blocks which are outside the truncation | |
38 | * point. Because the caller is about to free (and possibly reuse) those | |
39 | * blocks on-disk. | |
40 | */ | |
d47992f8 LC |
41 | void do_invalidatepage(struct page *page, unsigned int offset, |
42 | unsigned int length) | |
cf9a2ae8 | 43 | { |
d47992f8 LC |
44 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
45 | ||
cf9a2ae8 | 46 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 47 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
48 | if (!invalidatepage) |
49 | invalidatepage = block_invalidatepage; | |
9361401e | 50 | #endif |
cf9a2ae8 | 51 | if (invalidatepage) |
d47992f8 | 52 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
53 | } |
54 | ||
1da177e4 LT |
55 | static inline void truncate_partial_page(struct page *page, unsigned partial) |
56 | { | |
eebd2aa3 | 57 | zero_user_segment(page, partial, PAGE_CACHE_SIZE); |
3167760f | 58 | cleancache_invalidate_page(page->mapping, page); |
266cf658 | 59 | if (page_has_private(page)) |
d47992f8 | 60 | do_invalidatepage(page, partial, PAGE_CACHE_SIZE - partial); |
1da177e4 LT |
61 | } |
62 | ||
ecdfc978 LT |
63 | /* |
64 | * This cancels just the dirty bit on the kernel page itself, it | |
65 | * does NOT actually remove dirty bits on any mmap's that may be | |
66 | * around. It also leaves the page tagged dirty, so any sync | |
67 | * activity will still find it on the dirty lists, and in particular, | |
68 | * clear_page_dirty_for_io() will still look at the dirty bits in | |
69 | * the VM. | |
70 | * | |
71 | * Doing this should *normally* only ever be done when a page | |
72 | * is truncated, and is not actually mapped anywhere at all. However, | |
73 | * fs/buffer.c does this when it notices that somebody has cleaned | |
74 | * out all the buffers on a page without actually doing it through | |
75 | * the VM. Can you say "ext3 is horribly ugly"? Tought you could. | |
76 | */ | |
fba2591b LT |
77 | void cancel_dirty_page(struct page *page, unsigned int account_size) |
78 | { | |
8368e328 LT |
79 | if (TestClearPageDirty(page)) { |
80 | struct address_space *mapping = page->mapping; | |
81 | if (mapping && mapping_cap_account_dirty(mapping)) { | |
82 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
c9e51e41 PZ |
83 | dec_bdi_stat(mapping->backing_dev_info, |
84 | BDI_RECLAIMABLE); | |
8368e328 LT |
85 | if (account_size) |
86 | task_io_account_cancelled_write(account_size); | |
87 | } | |
3e67c098 | 88 | } |
fba2591b | 89 | } |
8368e328 | 90 | EXPORT_SYMBOL(cancel_dirty_page); |
fba2591b | 91 | |
1da177e4 LT |
92 | /* |
93 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 94 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 95 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
96 | * |
97 | * We need to bale out if page->mapping is no longer equal to the original | |
98 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
fc0ecff6 | 99 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
100 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
101 | */ | |
750b4987 | 102 | static int |
1da177e4 LT |
103 | truncate_complete_page(struct address_space *mapping, struct page *page) |
104 | { | |
105 | if (page->mapping != mapping) | |
750b4987 | 106 | return -EIO; |
1da177e4 | 107 | |
266cf658 | 108 | if (page_has_private(page)) |
d47992f8 | 109 | do_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
1da177e4 | 110 | |
a2b34564 BS |
111 | cancel_dirty_page(page, PAGE_CACHE_SIZE); |
112 | ||
1da177e4 | 113 | ClearPageMappedToDisk(page); |
5adc7b51 | 114 | delete_from_page_cache(page); |
750b4987 | 115 | return 0; |
1da177e4 LT |
116 | } |
117 | ||
118 | /* | |
fc0ecff6 | 119 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 120 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
121 | * be marked dirty at any time too, so use remove_mapping which safely |
122 | * discards clean, unused pages. | |
1da177e4 LT |
123 | * |
124 | * Returns non-zero if the page was successfully invalidated. | |
125 | */ | |
126 | static int | |
127 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
128 | { | |
0fd0e6b0 NP |
129 | int ret; |
130 | ||
1da177e4 LT |
131 | if (page->mapping != mapping) |
132 | return 0; | |
133 | ||
266cf658 | 134 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
135 | return 0; |
136 | ||
0fd0e6b0 | 137 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
138 | |
139 | return ret; | |
1da177e4 LT |
140 | } |
141 | ||
750b4987 NP |
142 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
143 | { | |
144 | if (page_mapped(page)) { | |
145 | unmap_mapping_range(mapping, | |
146 | (loff_t)page->index << PAGE_CACHE_SHIFT, | |
147 | PAGE_CACHE_SIZE, 0); | |
148 | } | |
149 | return truncate_complete_page(mapping, page); | |
150 | } | |
151 | ||
25718736 AK |
152 | /* |
153 | * Used to get rid of pages on hardware memory corruption. | |
154 | */ | |
155 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
156 | { | |
157 | if (!mapping) | |
158 | return -EINVAL; | |
159 | /* | |
160 | * Only punch for normal data pages for now. | |
161 | * Handling other types like directories would need more auditing. | |
162 | */ | |
163 | if (!S_ISREG(mapping->host->i_mode)) | |
164 | return -EIO; | |
165 | return truncate_inode_page(mapping, page); | |
166 | } | |
167 | EXPORT_SYMBOL(generic_error_remove_page); | |
168 | ||
83f78668 WF |
169 | /* |
170 | * Safely invalidate one page from its pagecache mapping. | |
171 | * It only drops clean, unused pages. The page must be locked. | |
172 | * | |
173 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
174 | */ | |
175 | int invalidate_inode_page(struct page *page) | |
176 | { | |
177 | struct address_space *mapping = page_mapping(page); | |
178 | if (!mapping) | |
179 | return 0; | |
180 | if (PageDirty(page) || PageWriteback(page)) | |
181 | return 0; | |
182 | if (page_mapped(page)) | |
183 | return 0; | |
184 | return invalidate_complete_page(mapping, page); | |
185 | } | |
186 | ||
1da177e4 | 187 | /** |
73c1e204 | 188 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
189 | * @mapping: mapping to truncate |
190 | * @lstart: offset from which to truncate | |
d7339071 | 191 | * @lend: offset to which to truncate |
1da177e4 | 192 | * |
d7339071 HR |
193 | * Truncate the page cache, removing the pages that are between |
194 | * specified offsets (and zeroing out partial page | |
195 | * (if lstart is not page aligned)). | |
1da177e4 LT |
196 | * |
197 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
198 | * block on page locks and it will not block on writeback. The second pass | |
199 | * will wait. This is to prevent as much IO as possible in the affected region. | |
200 | * The first pass will remove most pages, so the search cost of the second pass | |
201 | * is low. | |
202 | * | |
1da177e4 LT |
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; | |
211 | const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1); | |
212 | struct pagevec pvec; | |
b85e0eff HD |
213 | pgoff_t index; |
214 | pgoff_t end; | |
1da177e4 LT |
215 | int i; |
216 | ||
3167760f | 217 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
218 | if (mapping->nrpages == 0) |
219 | return; | |
220 | ||
d7339071 HR |
221 | BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1)); |
222 | end = (lend >> PAGE_CACHE_SHIFT); | |
223 | ||
1da177e4 | 224 | pagevec_init(&pvec, 0); |
b85e0eff HD |
225 | index = start; |
226 | while (index <= end && pagevec_lookup(&pvec, mapping, index, | |
227 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
e5598f8b | 228 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
229 | for (i = 0; i < pagevec_count(&pvec); i++) { |
230 | struct page *page = pvec.pages[i]; | |
1da177e4 | 231 | |
b85e0eff HD |
232 | /* We rely upon deletion not changing page->index */ |
233 | index = page->index; | |
234 | if (index > end) | |
d7339071 | 235 | break; |
d7339071 | 236 | |
529ae9aa | 237 | if (!trylock_page(page)) |
1da177e4 | 238 | continue; |
b85e0eff | 239 | WARN_ON(page->index != index); |
1da177e4 LT |
240 | if (PageWriteback(page)) { |
241 | unlock_page(page); | |
242 | continue; | |
243 | } | |
750b4987 | 244 | truncate_inode_page(mapping, page); |
1da177e4 LT |
245 | unlock_page(page); |
246 | } | |
247 | pagevec_release(&pvec); | |
e5598f8b | 248 | mem_cgroup_uncharge_end(); |
1da177e4 | 249 | cond_resched(); |
b85e0eff | 250 | index++; |
1da177e4 LT |
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 | ||
b85e0eff | 263 | index = start; |
1da177e4 LT |
264 | for ( ; ; ) { |
265 | cond_resched(); | |
b85e0eff HD |
266 | if (!pagevec_lookup(&pvec, mapping, index, |
267 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
268 | if (index == start) | |
1da177e4 | 269 | break; |
b85e0eff | 270 | index = start; |
1da177e4 LT |
271 | continue; |
272 | } | |
d0823576 | 273 | if (index == start && pvec.pages[0]->index > end) { |
d7339071 HR |
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 | ||
b85e0eff HD |
281 | /* We rely upon deletion not changing page->index */ |
282 | index = page->index; | |
283 | if (index > end) | |
d7339071 | 284 | break; |
b85e0eff | 285 | |
1da177e4 | 286 | lock_page(page); |
b85e0eff | 287 | WARN_ON(page->index != index); |
1da177e4 | 288 | wait_on_page_writeback(page); |
750b4987 | 289 | truncate_inode_page(mapping, page); |
1da177e4 LT |
290 | unlock_page(page); |
291 | } | |
292 | pagevec_release(&pvec); | |
569b846d | 293 | mem_cgroup_uncharge_end(); |
b85e0eff | 294 | index++; |
1da177e4 | 295 | } |
3167760f | 296 | cleancache_invalidate_inode(mapping); |
1da177e4 | 297 | } |
d7339071 | 298 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 299 | |
d7339071 HR |
300 | /** |
301 | * truncate_inode_pages - truncate *all* the pages from an offset | |
302 | * @mapping: mapping to truncate | |
303 | * @lstart: offset from which to truncate | |
304 | * | |
1b1dcc1b | 305 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
306 | * |
307 | * Note: When this function returns, there can be a page in the process of | |
308 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
309 | * mapping->nrpages can be non-zero when this function returns even after | |
310 | * truncation of the whole mapping. | |
d7339071 HR |
311 | */ |
312 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
313 | { | |
314 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
315 | } | |
1da177e4 LT |
316 | EXPORT_SYMBOL(truncate_inode_pages); |
317 | ||
28697355 MW |
318 | /** |
319 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
320 | * @mapping: the address_space which holds the pages to invalidate | |
321 | * @start: the offset 'from' which to invalidate | |
322 | * @end: the offset 'to' which to invalidate (inclusive) | |
323 | * | |
324 | * This function only removes the unlocked pages, if you want to | |
325 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
326 | * | |
327 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
328 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
329 | * pagetables. | |
330 | */ | |
331 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 332 | pgoff_t start, pgoff_t end) |
1da177e4 LT |
333 | { |
334 | struct pagevec pvec; | |
b85e0eff | 335 | pgoff_t index = start; |
31560180 MK |
336 | unsigned long ret; |
337 | unsigned long count = 0; | |
1da177e4 LT |
338 | int i; |
339 | ||
31475dd6 HD |
340 | /* |
341 | * Note: this function may get called on a shmem/tmpfs mapping: | |
342 | * pagevec_lookup() might then return 0 prematurely (because it | |
343 | * got a gangful of swap entries); but it's hardly worth worrying | |
344 | * about - it can rarely have anything to free from such a mapping | |
345 | * (most pages are dirty), and already skips over any difficulties. | |
346 | */ | |
347 | ||
1da177e4 | 348 | pagevec_init(&pvec, 0); |
b85e0eff HD |
349 | while (index <= end && pagevec_lookup(&pvec, mapping, index, |
350 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
569b846d | 351 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
352 | for (i = 0; i < pagevec_count(&pvec); i++) { |
353 | struct page *page = pvec.pages[i]; | |
e0f23603 | 354 | |
b85e0eff | 355 | /* We rely upon deletion not changing page->index */ |
e0f23603 | 356 | index = page->index; |
b85e0eff HD |
357 | if (index > end) |
358 | break; | |
e0f23603 | 359 | |
b85e0eff HD |
360 | if (!trylock_page(page)) |
361 | continue; | |
362 | WARN_ON(page->index != index); | |
31560180 | 363 | ret = invalidate_inode_page(page); |
1da177e4 | 364 | unlock_page(page); |
31560180 MK |
365 | /* |
366 | * Invalidation is a hint that the page is no longer | |
367 | * of interest and try to speed up its reclaim. | |
368 | */ | |
369 | if (!ret) | |
370 | deactivate_page(page); | |
371 | count += ret; | |
1da177e4 LT |
372 | } |
373 | pagevec_release(&pvec); | |
569b846d | 374 | mem_cgroup_uncharge_end(); |
28697355 | 375 | cond_resched(); |
b85e0eff | 376 | index++; |
1da177e4 | 377 | } |
31560180 | 378 | return count; |
1da177e4 | 379 | } |
54bc4855 | 380 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 381 | |
bd4c8ce4 AM |
382 | /* |
383 | * This is like invalidate_complete_page(), except it ignores the page's | |
384 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
385 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
386 | * shrink_page_list() has a temp ref on them, or because they're transiently |
387 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
388 | */ |
389 | static int | |
390 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
391 | { | |
392 | if (page->mapping != mapping) | |
393 | return 0; | |
394 | ||
266cf658 | 395 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
396 | return 0; |
397 | ||
19fd6231 | 398 | spin_lock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
399 | if (PageDirty(page)) |
400 | goto failed; | |
401 | ||
266cf658 | 402 | BUG_ON(page_has_private(page)); |
e64a782f | 403 | __delete_from_page_cache(page); |
19fd6231 | 404 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 405 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
406 | |
407 | if (mapping->a_ops->freepage) | |
408 | mapping->a_ops->freepage(page); | |
409 | ||
bd4c8ce4 AM |
410 | page_cache_release(page); /* pagecache ref */ |
411 | return 1; | |
412 | failed: | |
19fd6231 | 413 | spin_unlock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
414 | return 0; |
415 | } | |
416 | ||
e3db7691 TM |
417 | static int do_launder_page(struct address_space *mapping, struct page *page) |
418 | { | |
419 | if (!PageDirty(page)) | |
420 | return 0; | |
421 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
422 | return 0; | |
423 | return mapping->a_ops->launder_page(page); | |
424 | } | |
425 | ||
1da177e4 LT |
426 | /** |
427 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 428 | * @mapping: the address_space |
1da177e4 LT |
429 | * @start: the page offset 'from' which to invalidate |
430 | * @end: the page offset 'to' which to invalidate (inclusive) | |
431 | * | |
432 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
433 | * invalidation. | |
434 | * | |
6ccfa806 | 435 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
436 | */ |
437 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
438 | pgoff_t start, pgoff_t end) | |
439 | { | |
440 | struct pagevec pvec; | |
b85e0eff | 441 | pgoff_t index; |
1da177e4 LT |
442 | int i; |
443 | int ret = 0; | |
0dd1334f | 444 | int ret2 = 0; |
1da177e4 | 445 | int did_range_unmap = 0; |
1da177e4 | 446 | |
3167760f | 447 | cleancache_invalidate_inode(mapping); |
1da177e4 | 448 | pagevec_init(&pvec, 0); |
b85e0eff HD |
449 | index = start; |
450 | while (index <= end && pagevec_lookup(&pvec, mapping, index, | |
451 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
569b846d | 452 | mem_cgroup_uncharge_start(); |
7b965e08 | 453 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 454 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
455 | |
456 | /* We rely upon deletion not changing page->index */ | |
457 | index = page->index; | |
458 | if (index > end) | |
459 | break; | |
1da177e4 LT |
460 | |
461 | lock_page(page); | |
b85e0eff | 462 | WARN_ON(page->index != index); |
1da177e4 LT |
463 | if (page->mapping != mapping) { |
464 | unlock_page(page); | |
465 | continue; | |
466 | } | |
1da177e4 | 467 | wait_on_page_writeback(page); |
d00806b1 | 468 | if (page_mapped(page)) { |
1da177e4 LT |
469 | if (!did_range_unmap) { |
470 | /* | |
471 | * Zap the rest of the file in one hit. | |
472 | */ | |
473 | unmap_mapping_range(mapping, | |
b85e0eff HD |
474 | (loff_t)index << PAGE_CACHE_SHIFT, |
475 | (loff_t)(1 + end - index) | |
476 | << PAGE_CACHE_SHIFT, | |
1da177e4 LT |
477 | 0); |
478 | did_range_unmap = 1; | |
479 | } else { | |
480 | /* | |
481 | * Just zap this page | |
482 | */ | |
483 | unmap_mapping_range(mapping, | |
b85e0eff HD |
484 | (loff_t)index << PAGE_CACHE_SHIFT, |
485 | PAGE_CACHE_SIZE, 0); | |
1da177e4 LT |
486 | } |
487 | } | |
d00806b1 | 488 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
489 | ret2 = do_launder_page(mapping, page); |
490 | if (ret2 == 0) { | |
491 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 492 | ret2 = -EBUSY; |
0dd1334f HH |
493 | } |
494 | if (ret2 < 0) | |
495 | ret = ret2; | |
1da177e4 LT |
496 | unlock_page(page); |
497 | } | |
498 | pagevec_release(&pvec); | |
569b846d | 499 | mem_cgroup_uncharge_end(); |
1da177e4 | 500 | cond_resched(); |
b85e0eff | 501 | index++; |
1da177e4 | 502 | } |
3167760f | 503 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
504 | return ret; |
505 | } | |
506 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
507 | ||
508 | /** | |
509 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 510 | * @mapping: the address_space |
1da177e4 LT |
511 | * |
512 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
513 | * invalidation. | |
514 | * | |
e9de25dd | 515 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
516 | */ |
517 | int invalidate_inode_pages2(struct address_space *mapping) | |
518 | { | |
519 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
520 | } | |
521 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 522 | |
523 | /** | |
524 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
525 | * @inode: inode | |
8a549bea HD |
526 | * @oldsize: old file size |
527 | * @newsize: new file size | |
25d9e2d1 | 528 | * |
529 | * inode's new i_size must already be written before truncate_pagecache | |
530 | * is called. | |
531 | * | |
532 | * This function should typically be called before the filesystem | |
533 | * releases resources associated with the freed range (eg. deallocates | |
534 | * blocks). This way, pagecache will always stay logically coherent | |
535 | * with on-disk format, and the filesystem would not have to deal with | |
536 | * situations such as writepage being called for a page that has already | |
537 | * had its underlying blocks deallocated. | |
538 | */ | |
8a549bea | 539 | void truncate_pagecache(struct inode *inode, loff_t oldsize, loff_t newsize) |
25d9e2d1 | 540 | { |
cedabed4 | 541 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 542 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
543 | |
544 | /* | |
545 | * unmap_mapping_range is called twice, first simply for | |
546 | * efficiency so that truncate_inode_pages does fewer | |
547 | * single-page unmaps. However after this first call, and | |
548 | * before truncate_inode_pages finishes, it is possible for | |
549 | * private pages to be COWed, which remain after | |
550 | * truncate_inode_pages finishes, hence the second | |
551 | * unmap_mapping_range call must be made for correctness. | |
552 | */ | |
8a549bea HD |
553 | unmap_mapping_range(mapping, holebegin, 0, 1); |
554 | truncate_inode_pages(mapping, newsize); | |
555 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 556 | } |
557 | EXPORT_SYMBOL(truncate_pagecache); | |
558 | ||
2c27c65e CH |
559 | /** |
560 | * truncate_setsize - update inode and pagecache for a new file size | |
561 | * @inode: inode | |
562 | * @newsize: new file size | |
563 | * | |
382e27da JK |
564 | * truncate_setsize updates i_size and performs pagecache truncation (if |
565 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
566 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 567 | * |
382e27da JK |
568 | * Must be called with inode_mutex held and before all filesystem specific |
569 | * block truncation has been performed. | |
2c27c65e CH |
570 | */ |
571 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
572 | { | |
573 | loff_t oldsize; | |
574 | ||
575 | oldsize = inode->i_size; | |
576 | i_size_write(inode, newsize); | |
577 | ||
578 | truncate_pagecache(inode, oldsize, newsize); | |
579 | } | |
580 | EXPORT_SYMBOL(truncate_setsize); | |
581 | ||
623e3db9 HD |
582 | /** |
583 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
584 | * @inode: inode | |
585 | * @lstart: offset of beginning of hole | |
586 | * @lend: offset of last byte of hole | |
587 | * | |
588 | * This function should typically be called before the filesystem | |
589 | * releases resources associated with the freed range (eg. deallocates | |
590 | * blocks). This way, pagecache will always stay logically coherent | |
591 | * with on-disk format, and the filesystem would not have to deal with | |
592 | * situations such as writepage being called for a page that has already | |
593 | * had its underlying blocks deallocated. | |
594 | */ | |
595 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
596 | { | |
597 | struct address_space *mapping = inode->i_mapping; | |
598 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
599 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
600 | /* | |
601 | * This rounding is currently just for example: unmap_mapping_range | |
602 | * expands its hole outwards, whereas we want it to contract the hole | |
603 | * inwards. However, existing callers of truncate_pagecache_range are | |
604 | * doing their own page rounding first; and truncate_inode_pages_range | |
605 | * currently BUGs if lend is not pagealigned-1 (it handles partial | |
606 | * page at start of hole, but not partial page at end of hole). Note | |
607 | * unmap_mapping_range allows holelen 0 for all, and we allow lend -1. | |
608 | */ | |
609 | ||
610 | /* | |
611 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
612 | * once (before truncating pagecache), and without "even_cows" flag: | |
613 | * hole-punching should not remove private COWed pages from the hole. | |
614 | */ | |
615 | if ((u64)unmap_end > (u64)unmap_start) | |
616 | unmap_mapping_range(mapping, unmap_start, | |
617 | 1 + unmap_end - unmap_start, 0); | |
618 | truncate_inode_pages_range(mapping, lstart, lend); | |
619 | } | |
620 | EXPORT_SYMBOL(truncate_pagecache_range); |