]>
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> |
90a80202 | 23 | #include <linux/rmap.h> |
ba470de4 | 24 | #include "internal.h" |
1da177e4 | 25 | |
0cd6144a JW |
26 | static void clear_exceptional_entry(struct address_space *mapping, |
27 | pgoff_t index, void *entry) | |
28 | { | |
449dd698 JW |
29 | struct radix_tree_node *node; |
30 | void **slot; | |
31 | ||
0cd6144a JW |
32 | /* Handled by shmem itself */ |
33 | if (shmem_mapping(mapping)) | |
34 | return; | |
35 | ||
36 | spin_lock_irq(&mapping->tree_lock); | |
37 | /* | |
38 | * Regular page slots are stabilized by the page lock even | |
39 | * without the tree itself locked. These unlocked entries | |
40 | * need verification under the tree lock. | |
41 | */ | |
449dd698 JW |
42 | if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot)) |
43 | goto unlock; | |
44 | if (*slot != entry) | |
45 | goto unlock; | |
46 | radix_tree_replace_slot(slot, NULL); | |
47 | mapping->nrshadows--; | |
48 | if (!node) | |
49 | goto unlock; | |
50 | workingset_node_shadows_dec(node); | |
51 | /* | |
52 | * Don't track node without shadow entries. | |
53 | * | |
54 | * Avoid acquiring the list_lru lock if already untracked. | |
55 | * The list_empty() test is safe as node->private_list is | |
56 | * protected by mapping->tree_lock. | |
57 | */ | |
58 | if (!workingset_node_shadows(node) && | |
59 | !list_empty(&node->private_list)) | |
60 | list_lru_del(&workingset_shadow_nodes, &node->private_list); | |
61 | __radix_tree_delete_node(&mapping->page_tree, node); | |
62 | unlock: | |
0cd6144a JW |
63 | spin_unlock_irq(&mapping->tree_lock); |
64 | } | |
1da177e4 | 65 | |
cf9a2ae8 | 66 | /** |
28bc44d7 | 67 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 68 | * @page: the page which is affected |
d47992f8 LC |
69 | * @offset: start of the range to invalidate |
70 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
71 | * |
72 | * do_invalidatepage() is called when all or part of the page has become | |
73 | * invalidated by a truncate operation. | |
74 | * | |
75 | * do_invalidatepage() does not have to release all buffers, but it must | |
76 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
77 | * is underway against any of the blocks which are outside the truncation | |
78 | * point. Because the caller is about to free (and possibly reuse) those | |
79 | * blocks on-disk. | |
80 | */ | |
d47992f8 LC |
81 | void do_invalidatepage(struct page *page, unsigned int offset, |
82 | unsigned int length) | |
cf9a2ae8 | 83 | { |
d47992f8 LC |
84 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
85 | ||
cf9a2ae8 | 86 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 87 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
88 | if (!invalidatepage) |
89 | invalidatepage = block_invalidatepage; | |
9361401e | 90 | #endif |
cf9a2ae8 | 91 | if (invalidatepage) |
d47992f8 | 92 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
93 | } |
94 | ||
1da177e4 LT |
95 | /* |
96 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 97 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 98 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
99 | * |
100 | * We need to bale out if page->mapping is no longer equal to the original | |
101 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
fc0ecff6 | 102 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
103 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
104 | */ | |
750b4987 | 105 | static int |
1da177e4 LT |
106 | truncate_complete_page(struct address_space *mapping, struct page *page) |
107 | { | |
108 | if (page->mapping != mapping) | |
750b4987 | 109 | return -EIO; |
1da177e4 | 110 | |
266cf658 | 111 | if (page_has_private(page)) |
d47992f8 | 112 | do_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
1da177e4 | 113 | |
b9ea2515 KK |
114 | /* |
115 | * Some filesystems seem to re-dirty the page even after | |
116 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
117 | * Hence dirty accounting check is placed after invalidation. | |
118 | */ | |
119 | if (TestClearPageDirty(page)) | |
120 | account_page_cleaned(page, mapping); | |
a2b34564 | 121 | |
1da177e4 | 122 | ClearPageMappedToDisk(page); |
5adc7b51 | 123 | delete_from_page_cache(page); |
750b4987 | 124 | return 0; |
1da177e4 LT |
125 | } |
126 | ||
127 | /* | |
fc0ecff6 | 128 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 129 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
130 | * be marked dirty at any time too, so use remove_mapping which safely |
131 | * discards clean, unused pages. | |
1da177e4 LT |
132 | * |
133 | * Returns non-zero if the page was successfully invalidated. | |
134 | */ | |
135 | static int | |
136 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
137 | { | |
0fd0e6b0 NP |
138 | int ret; |
139 | ||
1da177e4 LT |
140 | if (page->mapping != mapping) |
141 | return 0; | |
142 | ||
266cf658 | 143 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
144 | return 0; |
145 | ||
0fd0e6b0 | 146 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
147 | |
148 | return ret; | |
1da177e4 LT |
149 | } |
150 | ||
750b4987 NP |
151 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
152 | { | |
153 | if (page_mapped(page)) { | |
154 | unmap_mapping_range(mapping, | |
155 | (loff_t)page->index << PAGE_CACHE_SHIFT, | |
156 | PAGE_CACHE_SIZE, 0); | |
157 | } | |
158 | return truncate_complete_page(mapping, page); | |
159 | } | |
160 | ||
25718736 AK |
161 | /* |
162 | * Used to get rid of pages on hardware memory corruption. | |
163 | */ | |
164 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
165 | { | |
166 | if (!mapping) | |
167 | return -EINVAL; | |
168 | /* | |
169 | * Only punch for normal data pages for now. | |
170 | * Handling other types like directories would need more auditing. | |
171 | */ | |
172 | if (!S_ISREG(mapping->host->i_mode)) | |
173 | return -EIO; | |
174 | return truncate_inode_page(mapping, page); | |
175 | } | |
176 | EXPORT_SYMBOL(generic_error_remove_page); | |
177 | ||
83f78668 WF |
178 | /* |
179 | * Safely invalidate one page from its pagecache mapping. | |
180 | * It only drops clean, unused pages. The page must be locked. | |
181 | * | |
182 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
183 | */ | |
184 | int invalidate_inode_page(struct page *page) | |
185 | { | |
186 | struct address_space *mapping = page_mapping(page); | |
187 | if (!mapping) | |
188 | return 0; | |
189 | if (PageDirty(page) || PageWriteback(page)) | |
190 | return 0; | |
191 | if (page_mapped(page)) | |
192 | return 0; | |
193 | return invalidate_complete_page(mapping, page); | |
194 | } | |
195 | ||
1da177e4 | 196 | /** |
73c1e204 | 197 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
198 | * @mapping: mapping to truncate |
199 | * @lstart: offset from which to truncate | |
5a720394 | 200 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 201 | * |
d7339071 | 202 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
203 | * specified offsets (and zeroing out partial pages |
204 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
205 | * |
206 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
207 | * block on page locks and it will not block on writeback. The second pass | |
208 | * will wait. This is to prevent as much IO as possible in the affected region. | |
209 | * The first pass will remove most pages, so the search cost of the second pass | |
210 | * is low. | |
211 | * | |
1da177e4 LT |
212 | * We pass down the cache-hot hint to the page freeing code. Even if the |
213 | * mapping is large, it is probably the case that the final pages are the most | |
214 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
215 | * |
216 | * Note that since ->invalidatepage() accepts range to invalidate | |
217 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
218 | * page aligned properly. | |
1da177e4 | 219 | */ |
d7339071 HR |
220 | void truncate_inode_pages_range(struct address_space *mapping, |
221 | loff_t lstart, loff_t lend) | |
1da177e4 | 222 | { |
5a720394 LC |
223 | pgoff_t start; /* inclusive */ |
224 | pgoff_t end; /* exclusive */ | |
225 | unsigned int partial_start; /* inclusive */ | |
226 | unsigned int partial_end; /* exclusive */ | |
227 | struct pagevec pvec; | |
0cd6144a | 228 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
229 | pgoff_t index; |
230 | int i; | |
1da177e4 | 231 | |
3167760f | 232 | cleancache_invalidate_inode(mapping); |
91b0abe3 | 233 | if (mapping->nrpages == 0 && mapping->nrshadows == 0) |
1da177e4 LT |
234 | return; |
235 | ||
5a720394 LC |
236 | /* Offsets within partial pages */ |
237 | partial_start = lstart & (PAGE_CACHE_SIZE - 1); | |
238 | partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); | |
239 | ||
240 | /* | |
241 | * 'start' and 'end' always covers the range of pages to be fully | |
242 | * truncated. Partial pages are covered with 'partial_start' at the | |
243 | * start of the range and 'partial_end' at the end of the range. | |
244 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
245 | */ | |
246 | start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
247 | if (lend == -1) | |
248 | /* | |
249 | * lend == -1 indicates end-of-file so we have to set 'end' | |
250 | * to the highest possible pgoff_t and since the type is | |
251 | * unsigned we're using -1. | |
252 | */ | |
253 | end = -1; | |
254 | else | |
255 | end = (lend + 1) >> PAGE_CACHE_SHIFT; | |
d7339071 | 256 | |
1da177e4 | 257 | pagevec_init(&pvec, 0); |
b85e0eff | 258 | index = start; |
0cd6144a JW |
259 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
260 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
261 | indices)) { | |
1da177e4 LT |
262 | for (i = 0; i < pagevec_count(&pvec); i++) { |
263 | struct page *page = pvec.pages[i]; | |
1da177e4 | 264 | |
b85e0eff | 265 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 266 | index = indices[i]; |
5a720394 | 267 | if (index >= end) |
d7339071 | 268 | break; |
d7339071 | 269 | |
0cd6144a JW |
270 | if (radix_tree_exceptional_entry(page)) { |
271 | clear_exceptional_entry(mapping, index, page); | |
272 | continue; | |
273 | } | |
274 | ||
529ae9aa | 275 | if (!trylock_page(page)) |
1da177e4 | 276 | continue; |
b85e0eff | 277 | WARN_ON(page->index != index); |
1da177e4 LT |
278 | if (PageWriteback(page)) { |
279 | unlock_page(page); | |
280 | continue; | |
281 | } | |
750b4987 | 282 | truncate_inode_page(mapping, page); |
1da177e4 LT |
283 | unlock_page(page); |
284 | } | |
0cd6144a | 285 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
286 | pagevec_release(&pvec); |
287 | cond_resched(); | |
b85e0eff | 288 | index++; |
1da177e4 LT |
289 | } |
290 | ||
5a720394 | 291 | if (partial_start) { |
1da177e4 LT |
292 | struct page *page = find_lock_page(mapping, start - 1); |
293 | if (page) { | |
5a720394 LC |
294 | unsigned int top = PAGE_CACHE_SIZE; |
295 | if (start > end) { | |
296 | /* Truncation within a single page */ | |
297 | top = partial_end; | |
298 | partial_end = 0; | |
299 | } | |
1da177e4 | 300 | wait_on_page_writeback(page); |
5a720394 LC |
301 | zero_user_segment(page, partial_start, top); |
302 | cleancache_invalidate_page(mapping, page); | |
303 | if (page_has_private(page)) | |
304 | do_invalidatepage(page, partial_start, | |
305 | top - partial_start); | |
1da177e4 LT |
306 | unlock_page(page); |
307 | page_cache_release(page); | |
308 | } | |
309 | } | |
5a720394 LC |
310 | if (partial_end) { |
311 | struct page *page = find_lock_page(mapping, end); | |
312 | if (page) { | |
313 | wait_on_page_writeback(page); | |
314 | zero_user_segment(page, 0, partial_end); | |
315 | cleancache_invalidate_page(mapping, page); | |
316 | if (page_has_private(page)) | |
317 | do_invalidatepage(page, 0, | |
318 | partial_end); | |
319 | unlock_page(page); | |
320 | page_cache_release(page); | |
321 | } | |
322 | } | |
323 | /* | |
324 | * If the truncation happened within a single page no pages | |
325 | * will be released, just zeroed, so we can bail out now. | |
326 | */ | |
327 | if (start >= end) | |
328 | return; | |
1da177e4 | 329 | |
b85e0eff | 330 | index = start; |
1da177e4 LT |
331 | for ( ; ; ) { |
332 | cond_resched(); | |
0cd6144a | 333 | if (!pagevec_lookup_entries(&pvec, mapping, index, |
792ceaef HD |
334 | min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { |
335 | /* If all gone from start onwards, we're done */ | |
b85e0eff | 336 | if (index == start) |
1da177e4 | 337 | break; |
792ceaef | 338 | /* Otherwise restart to make sure all gone */ |
b85e0eff | 339 | index = start; |
1da177e4 LT |
340 | continue; |
341 | } | |
0cd6144a | 342 | if (index == start && indices[0] >= end) { |
792ceaef | 343 | /* All gone out of hole to be punched, we're done */ |
0cd6144a | 344 | pagevec_remove_exceptionals(&pvec); |
d7339071 HR |
345 | pagevec_release(&pvec); |
346 | break; | |
347 | } | |
1da177e4 LT |
348 | for (i = 0; i < pagevec_count(&pvec); i++) { |
349 | struct page *page = pvec.pages[i]; | |
350 | ||
b85e0eff | 351 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 352 | index = indices[i]; |
792ceaef HD |
353 | if (index >= end) { |
354 | /* Restart punch to make sure all gone */ | |
355 | index = start - 1; | |
d7339071 | 356 | break; |
792ceaef | 357 | } |
b85e0eff | 358 | |
0cd6144a JW |
359 | if (radix_tree_exceptional_entry(page)) { |
360 | clear_exceptional_entry(mapping, index, page); | |
361 | continue; | |
362 | } | |
363 | ||
1da177e4 | 364 | lock_page(page); |
b85e0eff | 365 | WARN_ON(page->index != index); |
1da177e4 | 366 | wait_on_page_writeback(page); |
750b4987 | 367 | truncate_inode_page(mapping, page); |
1da177e4 LT |
368 | unlock_page(page); |
369 | } | |
0cd6144a | 370 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 371 | pagevec_release(&pvec); |
b85e0eff | 372 | index++; |
1da177e4 | 373 | } |
3167760f | 374 | cleancache_invalidate_inode(mapping); |
1da177e4 | 375 | } |
d7339071 | 376 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 377 | |
d7339071 HR |
378 | /** |
379 | * truncate_inode_pages - truncate *all* the pages from an offset | |
380 | * @mapping: mapping to truncate | |
381 | * @lstart: offset from which to truncate | |
382 | * | |
1b1dcc1b | 383 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
384 | * |
385 | * Note: When this function returns, there can be a page in the process of | |
386 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
387 | * mapping->nrpages can be non-zero when this function returns even after | |
388 | * truncation of the whole mapping. | |
d7339071 HR |
389 | */ |
390 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
391 | { | |
392 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
393 | } | |
1da177e4 LT |
394 | EXPORT_SYMBOL(truncate_inode_pages); |
395 | ||
91b0abe3 JW |
396 | /** |
397 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
398 | * @mapping: mapping to truncate | |
399 | * | |
400 | * Called under (and serialized by) inode->i_mutex. | |
401 | * | |
402 | * Filesystems have to use this in the .evict_inode path to inform the | |
403 | * VM that this is the final truncate and the inode is going away. | |
404 | */ | |
405 | void truncate_inode_pages_final(struct address_space *mapping) | |
406 | { | |
407 | unsigned long nrshadows; | |
408 | unsigned long nrpages; | |
409 | ||
410 | /* | |
411 | * Page reclaim can not participate in regular inode lifetime | |
412 | * management (can't call iput()) and thus can race with the | |
413 | * inode teardown. Tell it when the address space is exiting, | |
414 | * so that it does not install eviction information after the | |
415 | * final truncate has begun. | |
416 | */ | |
417 | mapping_set_exiting(mapping); | |
418 | ||
419 | /* | |
420 | * When reclaim installs eviction entries, it increases | |
421 | * nrshadows first, then decreases nrpages. Make sure we see | |
422 | * this in the right order or we might miss an entry. | |
423 | */ | |
424 | nrpages = mapping->nrpages; | |
425 | smp_rmb(); | |
426 | nrshadows = mapping->nrshadows; | |
427 | ||
428 | if (nrpages || nrshadows) { | |
429 | /* | |
430 | * As truncation uses a lockless tree lookup, cycle | |
431 | * the tree lock to make sure any ongoing tree | |
432 | * modification that does not see AS_EXITING is | |
433 | * completed before starting the final truncate. | |
434 | */ | |
435 | spin_lock_irq(&mapping->tree_lock); | |
436 | spin_unlock_irq(&mapping->tree_lock); | |
437 | ||
438 | truncate_inode_pages(mapping, 0); | |
439 | } | |
440 | } | |
441 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
442 | ||
28697355 MW |
443 | /** |
444 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
445 | * @mapping: the address_space which holds the pages to invalidate | |
446 | * @start: the offset 'from' which to invalidate | |
447 | * @end: the offset 'to' which to invalidate (inclusive) | |
448 | * | |
449 | * This function only removes the unlocked pages, if you want to | |
450 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
451 | * | |
452 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
453 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
454 | * pagetables. | |
455 | */ | |
456 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 457 | pgoff_t start, pgoff_t end) |
1da177e4 | 458 | { |
0cd6144a | 459 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 460 | struct pagevec pvec; |
b85e0eff | 461 | pgoff_t index = start; |
31560180 MK |
462 | unsigned long ret; |
463 | unsigned long count = 0; | |
1da177e4 LT |
464 | int i; |
465 | ||
466 | pagevec_init(&pvec, 0); | |
0cd6144a JW |
467 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
468 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
469 | indices)) { | |
1da177e4 LT |
470 | for (i = 0; i < pagevec_count(&pvec); i++) { |
471 | struct page *page = pvec.pages[i]; | |
e0f23603 | 472 | |
b85e0eff | 473 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 474 | index = indices[i]; |
b85e0eff HD |
475 | if (index > end) |
476 | break; | |
e0f23603 | 477 | |
0cd6144a JW |
478 | if (radix_tree_exceptional_entry(page)) { |
479 | clear_exceptional_entry(mapping, index, page); | |
480 | continue; | |
481 | } | |
482 | ||
b85e0eff HD |
483 | if (!trylock_page(page)) |
484 | continue; | |
485 | WARN_ON(page->index != index); | |
31560180 | 486 | ret = invalidate_inode_page(page); |
1da177e4 | 487 | unlock_page(page); |
31560180 MK |
488 | /* |
489 | * Invalidation is a hint that the page is no longer | |
490 | * of interest and try to speed up its reclaim. | |
491 | */ | |
492 | if (!ret) | |
cc5993bd | 493 | deactivate_file_page(page); |
31560180 | 494 | count += ret; |
1da177e4 | 495 | } |
0cd6144a | 496 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 497 | pagevec_release(&pvec); |
28697355 | 498 | cond_resched(); |
b85e0eff | 499 | index++; |
1da177e4 | 500 | } |
31560180 | 501 | return count; |
1da177e4 | 502 | } |
54bc4855 | 503 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 504 | |
bd4c8ce4 AM |
505 | /* |
506 | * This is like invalidate_complete_page(), except it ignores the page's | |
507 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
508 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
509 | * shrink_page_list() has a temp ref on them, or because they're transiently |
510 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
511 | */ |
512 | static int | |
513 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
514 | { | |
515 | if (page->mapping != mapping) | |
516 | return 0; | |
517 | ||
266cf658 | 518 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
519 | return 0; |
520 | ||
19fd6231 | 521 | spin_lock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
522 | if (PageDirty(page)) |
523 | goto failed; | |
524 | ||
266cf658 | 525 | BUG_ON(page_has_private(page)); |
91b0abe3 | 526 | __delete_from_page_cache(page, NULL); |
19fd6231 | 527 | spin_unlock_irq(&mapping->tree_lock); |
6072d13c LT |
528 | |
529 | if (mapping->a_ops->freepage) | |
530 | mapping->a_ops->freepage(page); | |
531 | ||
bd4c8ce4 AM |
532 | page_cache_release(page); /* pagecache ref */ |
533 | return 1; | |
534 | failed: | |
19fd6231 | 535 | spin_unlock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
536 | return 0; |
537 | } | |
538 | ||
e3db7691 TM |
539 | static int do_launder_page(struct address_space *mapping, struct page *page) |
540 | { | |
541 | if (!PageDirty(page)) | |
542 | return 0; | |
543 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
544 | return 0; | |
545 | return mapping->a_ops->launder_page(page); | |
546 | } | |
547 | ||
1da177e4 LT |
548 | /** |
549 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 550 | * @mapping: the address_space |
1da177e4 LT |
551 | * @start: the page offset 'from' which to invalidate |
552 | * @end: the page offset 'to' which to invalidate (inclusive) | |
553 | * | |
554 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
555 | * invalidation. | |
556 | * | |
6ccfa806 | 557 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
558 | */ |
559 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
560 | pgoff_t start, pgoff_t end) | |
561 | { | |
0cd6144a | 562 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 563 | struct pagevec pvec; |
b85e0eff | 564 | pgoff_t index; |
1da177e4 LT |
565 | int i; |
566 | int ret = 0; | |
0dd1334f | 567 | int ret2 = 0; |
1da177e4 | 568 | int did_range_unmap = 0; |
1da177e4 | 569 | |
3167760f | 570 | cleancache_invalidate_inode(mapping); |
1da177e4 | 571 | pagevec_init(&pvec, 0); |
b85e0eff | 572 | index = start; |
0cd6144a JW |
573 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
574 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
575 | indices)) { | |
7b965e08 | 576 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 577 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
578 | |
579 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 580 | index = indices[i]; |
b85e0eff HD |
581 | if (index > end) |
582 | break; | |
1da177e4 | 583 | |
0cd6144a JW |
584 | if (radix_tree_exceptional_entry(page)) { |
585 | clear_exceptional_entry(mapping, index, page); | |
586 | continue; | |
587 | } | |
588 | ||
1da177e4 | 589 | lock_page(page); |
b85e0eff | 590 | WARN_ON(page->index != index); |
1da177e4 LT |
591 | if (page->mapping != mapping) { |
592 | unlock_page(page); | |
593 | continue; | |
594 | } | |
1da177e4 | 595 | wait_on_page_writeback(page); |
d00806b1 | 596 | if (page_mapped(page)) { |
1da177e4 LT |
597 | if (!did_range_unmap) { |
598 | /* | |
599 | * Zap the rest of the file in one hit. | |
600 | */ | |
601 | unmap_mapping_range(mapping, | |
b85e0eff HD |
602 | (loff_t)index << PAGE_CACHE_SHIFT, |
603 | (loff_t)(1 + end - index) | |
604 | << PAGE_CACHE_SHIFT, | |
1da177e4 LT |
605 | 0); |
606 | did_range_unmap = 1; | |
607 | } else { | |
608 | /* | |
609 | * Just zap this page | |
610 | */ | |
611 | unmap_mapping_range(mapping, | |
b85e0eff HD |
612 | (loff_t)index << PAGE_CACHE_SHIFT, |
613 | PAGE_CACHE_SIZE, 0); | |
1da177e4 LT |
614 | } |
615 | } | |
d00806b1 | 616 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
617 | ret2 = do_launder_page(mapping, page); |
618 | if (ret2 == 0) { | |
619 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 620 | ret2 = -EBUSY; |
0dd1334f HH |
621 | } |
622 | if (ret2 < 0) | |
623 | ret = ret2; | |
1da177e4 LT |
624 | unlock_page(page); |
625 | } | |
0cd6144a | 626 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
627 | pagevec_release(&pvec); |
628 | cond_resched(); | |
b85e0eff | 629 | index++; |
1da177e4 | 630 | } |
3167760f | 631 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
632 | return ret; |
633 | } | |
634 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
635 | ||
636 | /** | |
637 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 638 | * @mapping: the address_space |
1da177e4 LT |
639 | * |
640 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
641 | * invalidation. | |
642 | * | |
e9de25dd | 643 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
644 | */ |
645 | int invalidate_inode_pages2(struct address_space *mapping) | |
646 | { | |
647 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
648 | } | |
649 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 650 | |
651 | /** | |
652 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
653 | * @inode: inode | |
8a549bea | 654 | * @newsize: new file size |
25d9e2d1 | 655 | * |
656 | * inode's new i_size must already be written before truncate_pagecache | |
657 | * is called. | |
658 | * | |
659 | * This function should typically be called before the filesystem | |
660 | * releases resources associated with the freed range (eg. deallocates | |
661 | * blocks). This way, pagecache will always stay logically coherent | |
662 | * with on-disk format, and the filesystem would not have to deal with | |
663 | * situations such as writepage being called for a page that has already | |
664 | * had its underlying blocks deallocated. | |
665 | */ | |
7caef267 | 666 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 667 | { |
cedabed4 | 668 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 669 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
670 | |
671 | /* | |
672 | * unmap_mapping_range is called twice, first simply for | |
673 | * efficiency so that truncate_inode_pages does fewer | |
674 | * single-page unmaps. However after this first call, and | |
675 | * before truncate_inode_pages finishes, it is possible for | |
676 | * private pages to be COWed, which remain after | |
677 | * truncate_inode_pages finishes, hence the second | |
678 | * unmap_mapping_range call must be made for correctness. | |
679 | */ | |
8a549bea HD |
680 | unmap_mapping_range(mapping, holebegin, 0, 1); |
681 | truncate_inode_pages(mapping, newsize); | |
682 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 683 | } |
684 | EXPORT_SYMBOL(truncate_pagecache); | |
685 | ||
2c27c65e CH |
686 | /** |
687 | * truncate_setsize - update inode and pagecache for a new file size | |
688 | * @inode: inode | |
689 | * @newsize: new file size | |
690 | * | |
382e27da JK |
691 | * truncate_setsize updates i_size and performs pagecache truncation (if |
692 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
693 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 694 | * |
77783d06 JK |
695 | * Must be called with a lock serializing truncates and writes (generally |
696 | * i_mutex but e.g. xfs uses a different lock) and before all filesystem | |
697 | * specific block truncation has been performed. | |
2c27c65e CH |
698 | */ |
699 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
700 | { | |
90a80202 JK |
701 | loff_t oldsize = inode->i_size; |
702 | ||
2c27c65e | 703 | i_size_write(inode, newsize); |
90a80202 JK |
704 | if (newsize > oldsize) |
705 | pagecache_isize_extended(inode, oldsize, newsize); | |
7caef267 | 706 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
707 | } |
708 | EXPORT_SYMBOL(truncate_setsize); | |
709 | ||
90a80202 JK |
710 | /** |
711 | * pagecache_isize_extended - update pagecache after extension of i_size | |
712 | * @inode: inode for which i_size was extended | |
713 | * @from: original inode size | |
714 | * @to: new inode size | |
715 | * | |
716 | * Handle extension of inode size either caused by extending truncate or by | |
717 | * write starting after current i_size. We mark the page straddling current | |
718 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
719 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
720 | * the page before user writes to the page via mmap after the i_size has been | |
721 | * changed. | |
722 | * | |
723 | * The function must be called after i_size is updated so that page fault | |
724 | * coming after we unlock the page will already see the new i_size. | |
725 | * The function must be called while we still hold i_mutex - this not only | |
726 | * makes sure i_size is stable but also that userspace cannot observe new | |
727 | * i_size value before we are prepared to store mmap writes at new inode size. | |
728 | */ | |
729 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
730 | { | |
731 | int bsize = 1 << inode->i_blkbits; | |
732 | loff_t rounded_from; | |
733 | struct page *page; | |
734 | pgoff_t index; | |
735 | ||
90a80202 JK |
736 | WARN_ON(to > inode->i_size); |
737 | ||
738 | if (from >= to || bsize == PAGE_CACHE_SIZE) | |
739 | return; | |
740 | /* Page straddling @from will not have any hole block created? */ | |
741 | rounded_from = round_up(from, bsize); | |
742 | if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1))) | |
743 | return; | |
744 | ||
745 | index = from >> PAGE_CACHE_SHIFT; | |
746 | page = find_lock_page(inode->i_mapping, index); | |
747 | /* Page not cached? Nothing to do */ | |
748 | if (!page) | |
749 | return; | |
750 | /* | |
751 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
752 | * is needed. | |
753 | */ | |
754 | if (page_mkclean(page)) | |
755 | set_page_dirty(page); | |
756 | unlock_page(page); | |
757 | page_cache_release(page); | |
758 | } | |
759 | EXPORT_SYMBOL(pagecache_isize_extended); | |
760 | ||
623e3db9 HD |
761 | /** |
762 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
763 | * @inode: inode | |
764 | * @lstart: offset of beginning of hole | |
765 | * @lend: offset of last byte of hole | |
766 | * | |
767 | * This function should typically be called before the filesystem | |
768 | * releases resources associated with the freed range (eg. deallocates | |
769 | * blocks). This way, pagecache will always stay logically coherent | |
770 | * with on-disk format, and the filesystem would not have to deal with | |
771 | * situations such as writepage being called for a page that has already | |
772 | * had its underlying blocks deallocated. | |
773 | */ | |
774 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
775 | { | |
776 | struct address_space *mapping = inode->i_mapping; | |
777 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
778 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
779 | /* | |
780 | * This rounding is currently just for example: unmap_mapping_range | |
781 | * expands its hole outwards, whereas we want it to contract the hole | |
782 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
783 | * doing their own page rounding first. Note that unmap_mapping_range |
784 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
785 | */ |
786 | ||
787 | /* | |
788 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
789 | * once (before truncating pagecache), and without "even_cows" flag: | |
790 | * hole-punching should not remove private COWed pages from the hole. | |
791 | */ | |
792 | if ((u64)unmap_end > (u64)unmap_start) | |
793 | unmap_mapping_range(mapping, unmap_start, | |
794 | 1 + unmap_end - unmap_start, 0); | |
795 | truncate_inode_pages_range(mapping, lstart, lend); | |
796 | } | |
797 | EXPORT_SYMBOL(truncate_pagecache_range); |