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 | */ | |
11f81bec | 119 | cancel_dirty_page(page); |
1da177e4 | 120 | ClearPageMappedToDisk(page); |
5adc7b51 | 121 | delete_from_page_cache(page); |
750b4987 | 122 | return 0; |
1da177e4 LT |
123 | } |
124 | ||
125 | /* | |
fc0ecff6 | 126 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 127 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
128 | * be marked dirty at any time too, so use remove_mapping which safely |
129 | * discards clean, unused pages. | |
1da177e4 LT |
130 | * |
131 | * Returns non-zero if the page was successfully invalidated. | |
132 | */ | |
133 | static int | |
134 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
135 | { | |
0fd0e6b0 NP |
136 | int ret; |
137 | ||
1da177e4 LT |
138 | if (page->mapping != mapping) |
139 | return 0; | |
140 | ||
266cf658 | 141 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
142 | return 0; |
143 | ||
0fd0e6b0 | 144 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
145 | |
146 | return ret; | |
1da177e4 LT |
147 | } |
148 | ||
750b4987 NP |
149 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
150 | { | |
151 | if (page_mapped(page)) { | |
152 | unmap_mapping_range(mapping, | |
153 | (loff_t)page->index << PAGE_CACHE_SHIFT, | |
154 | PAGE_CACHE_SIZE, 0); | |
155 | } | |
156 | return truncate_complete_page(mapping, page); | |
157 | } | |
158 | ||
25718736 AK |
159 | /* |
160 | * Used to get rid of pages on hardware memory corruption. | |
161 | */ | |
162 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
163 | { | |
164 | if (!mapping) | |
165 | return -EINVAL; | |
166 | /* | |
167 | * Only punch for normal data pages for now. | |
168 | * Handling other types like directories would need more auditing. | |
169 | */ | |
170 | if (!S_ISREG(mapping->host->i_mode)) | |
171 | return -EIO; | |
172 | return truncate_inode_page(mapping, page); | |
173 | } | |
174 | EXPORT_SYMBOL(generic_error_remove_page); | |
175 | ||
83f78668 WF |
176 | /* |
177 | * Safely invalidate one page from its pagecache mapping. | |
178 | * It only drops clean, unused pages. The page must be locked. | |
179 | * | |
180 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
181 | */ | |
182 | int invalidate_inode_page(struct page *page) | |
183 | { | |
184 | struct address_space *mapping = page_mapping(page); | |
185 | if (!mapping) | |
186 | return 0; | |
187 | if (PageDirty(page) || PageWriteback(page)) | |
188 | return 0; | |
189 | if (page_mapped(page)) | |
190 | return 0; | |
191 | return invalidate_complete_page(mapping, page); | |
192 | } | |
193 | ||
1da177e4 | 194 | /** |
73c1e204 | 195 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
196 | * @mapping: mapping to truncate |
197 | * @lstart: offset from which to truncate | |
5a720394 | 198 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 199 | * |
d7339071 | 200 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
201 | * specified offsets (and zeroing out partial pages |
202 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
203 | * |
204 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
205 | * block on page locks and it will not block on writeback. The second pass | |
206 | * will wait. This is to prevent as much IO as possible in the affected region. | |
207 | * The first pass will remove most pages, so the search cost of the second pass | |
208 | * is low. | |
209 | * | |
1da177e4 LT |
210 | * We pass down the cache-hot hint to the page freeing code. Even if the |
211 | * mapping is large, it is probably the case that the final pages are the most | |
212 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
213 | * |
214 | * Note that since ->invalidatepage() accepts range to invalidate | |
215 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
216 | * page aligned properly. | |
1da177e4 | 217 | */ |
d7339071 HR |
218 | void truncate_inode_pages_range(struct address_space *mapping, |
219 | loff_t lstart, loff_t lend) | |
1da177e4 | 220 | { |
5a720394 LC |
221 | pgoff_t start; /* inclusive */ |
222 | pgoff_t end; /* exclusive */ | |
223 | unsigned int partial_start; /* inclusive */ | |
224 | unsigned int partial_end; /* exclusive */ | |
225 | struct pagevec pvec; | |
0cd6144a | 226 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
227 | pgoff_t index; |
228 | int i; | |
1da177e4 | 229 | |
3167760f | 230 | cleancache_invalidate_inode(mapping); |
91b0abe3 | 231 | if (mapping->nrpages == 0 && mapping->nrshadows == 0) |
1da177e4 LT |
232 | return; |
233 | ||
5a720394 LC |
234 | /* Offsets within partial pages */ |
235 | partial_start = lstart & (PAGE_CACHE_SIZE - 1); | |
236 | partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); | |
237 | ||
238 | /* | |
239 | * 'start' and 'end' always covers the range of pages to be fully | |
240 | * truncated. Partial pages are covered with 'partial_start' at the | |
241 | * start of the range and 'partial_end' at the end of the range. | |
242 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
243 | */ | |
244 | start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
245 | if (lend == -1) | |
246 | /* | |
247 | * lend == -1 indicates end-of-file so we have to set 'end' | |
248 | * to the highest possible pgoff_t and since the type is | |
249 | * unsigned we're using -1. | |
250 | */ | |
251 | end = -1; | |
252 | else | |
253 | end = (lend + 1) >> PAGE_CACHE_SHIFT; | |
d7339071 | 254 | |
1da177e4 | 255 | pagevec_init(&pvec, 0); |
b85e0eff | 256 | index = start; |
0cd6144a JW |
257 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
258 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
259 | indices)) { | |
1da177e4 LT |
260 | for (i = 0; i < pagevec_count(&pvec); i++) { |
261 | struct page *page = pvec.pages[i]; | |
1da177e4 | 262 | |
b85e0eff | 263 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 264 | index = indices[i]; |
5a720394 | 265 | if (index >= end) |
d7339071 | 266 | break; |
d7339071 | 267 | |
0cd6144a JW |
268 | if (radix_tree_exceptional_entry(page)) { |
269 | clear_exceptional_entry(mapping, index, page); | |
270 | continue; | |
271 | } | |
272 | ||
529ae9aa | 273 | if (!trylock_page(page)) |
1da177e4 | 274 | continue; |
b85e0eff | 275 | WARN_ON(page->index != index); |
1da177e4 LT |
276 | if (PageWriteback(page)) { |
277 | unlock_page(page); | |
278 | continue; | |
279 | } | |
750b4987 | 280 | truncate_inode_page(mapping, page); |
1da177e4 LT |
281 | unlock_page(page); |
282 | } | |
0cd6144a | 283 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
284 | pagevec_release(&pvec); |
285 | cond_resched(); | |
b85e0eff | 286 | index++; |
1da177e4 LT |
287 | } |
288 | ||
5a720394 | 289 | if (partial_start) { |
1da177e4 LT |
290 | struct page *page = find_lock_page(mapping, start - 1); |
291 | if (page) { | |
5a720394 LC |
292 | unsigned int top = PAGE_CACHE_SIZE; |
293 | if (start > end) { | |
294 | /* Truncation within a single page */ | |
295 | top = partial_end; | |
296 | partial_end = 0; | |
297 | } | |
1da177e4 | 298 | wait_on_page_writeback(page); |
5a720394 LC |
299 | zero_user_segment(page, partial_start, top); |
300 | cleancache_invalidate_page(mapping, page); | |
301 | if (page_has_private(page)) | |
302 | do_invalidatepage(page, partial_start, | |
303 | top - partial_start); | |
1da177e4 LT |
304 | unlock_page(page); |
305 | page_cache_release(page); | |
306 | } | |
307 | } | |
5a720394 LC |
308 | if (partial_end) { |
309 | struct page *page = find_lock_page(mapping, end); | |
310 | if (page) { | |
311 | wait_on_page_writeback(page); | |
312 | zero_user_segment(page, 0, partial_end); | |
313 | cleancache_invalidate_page(mapping, page); | |
314 | if (page_has_private(page)) | |
315 | do_invalidatepage(page, 0, | |
316 | partial_end); | |
317 | unlock_page(page); | |
318 | page_cache_release(page); | |
319 | } | |
320 | } | |
321 | /* | |
322 | * If the truncation happened within a single page no pages | |
323 | * will be released, just zeroed, so we can bail out now. | |
324 | */ | |
325 | if (start >= end) | |
326 | return; | |
1da177e4 | 327 | |
b85e0eff | 328 | index = start; |
1da177e4 LT |
329 | for ( ; ; ) { |
330 | cond_resched(); | |
0cd6144a | 331 | if (!pagevec_lookup_entries(&pvec, mapping, index, |
792ceaef HD |
332 | min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { |
333 | /* If all gone from start onwards, we're done */ | |
b85e0eff | 334 | if (index == start) |
1da177e4 | 335 | break; |
792ceaef | 336 | /* Otherwise restart to make sure all gone */ |
b85e0eff | 337 | index = start; |
1da177e4 LT |
338 | continue; |
339 | } | |
0cd6144a | 340 | if (index == start && indices[0] >= end) { |
792ceaef | 341 | /* All gone out of hole to be punched, we're done */ |
0cd6144a | 342 | pagevec_remove_exceptionals(&pvec); |
d7339071 HR |
343 | pagevec_release(&pvec); |
344 | break; | |
345 | } | |
1da177e4 LT |
346 | for (i = 0; i < pagevec_count(&pvec); i++) { |
347 | struct page *page = pvec.pages[i]; | |
348 | ||
b85e0eff | 349 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 350 | index = indices[i]; |
792ceaef HD |
351 | if (index >= end) { |
352 | /* Restart punch to make sure all gone */ | |
353 | index = start - 1; | |
d7339071 | 354 | break; |
792ceaef | 355 | } |
b85e0eff | 356 | |
0cd6144a JW |
357 | if (radix_tree_exceptional_entry(page)) { |
358 | clear_exceptional_entry(mapping, index, page); | |
359 | continue; | |
360 | } | |
361 | ||
1da177e4 | 362 | lock_page(page); |
b85e0eff | 363 | WARN_ON(page->index != index); |
1da177e4 | 364 | wait_on_page_writeback(page); |
750b4987 | 365 | truncate_inode_page(mapping, page); |
1da177e4 LT |
366 | unlock_page(page); |
367 | } | |
0cd6144a | 368 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 369 | pagevec_release(&pvec); |
b85e0eff | 370 | index++; |
1da177e4 | 371 | } |
3167760f | 372 | cleancache_invalidate_inode(mapping); |
1da177e4 | 373 | } |
d7339071 | 374 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 375 | |
d7339071 HR |
376 | /** |
377 | * truncate_inode_pages - truncate *all* the pages from an offset | |
378 | * @mapping: mapping to truncate | |
379 | * @lstart: offset from which to truncate | |
380 | * | |
1b1dcc1b | 381 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
382 | * |
383 | * Note: When this function returns, there can be a page in the process of | |
384 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
385 | * mapping->nrpages can be non-zero when this function returns even after | |
386 | * truncation of the whole mapping. | |
d7339071 HR |
387 | */ |
388 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
389 | { | |
390 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
391 | } | |
1da177e4 LT |
392 | EXPORT_SYMBOL(truncate_inode_pages); |
393 | ||
91b0abe3 JW |
394 | /** |
395 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
396 | * @mapping: mapping to truncate | |
397 | * | |
398 | * Called under (and serialized by) inode->i_mutex. | |
399 | * | |
400 | * Filesystems have to use this in the .evict_inode path to inform the | |
401 | * VM that this is the final truncate and the inode is going away. | |
402 | */ | |
403 | void truncate_inode_pages_final(struct address_space *mapping) | |
404 | { | |
405 | unsigned long nrshadows; | |
406 | unsigned long nrpages; | |
407 | ||
408 | /* | |
409 | * Page reclaim can not participate in regular inode lifetime | |
410 | * management (can't call iput()) and thus can race with the | |
411 | * inode teardown. Tell it when the address space is exiting, | |
412 | * so that it does not install eviction information after the | |
413 | * final truncate has begun. | |
414 | */ | |
415 | mapping_set_exiting(mapping); | |
416 | ||
417 | /* | |
418 | * When reclaim installs eviction entries, it increases | |
419 | * nrshadows first, then decreases nrpages. Make sure we see | |
420 | * this in the right order or we might miss an entry. | |
421 | */ | |
422 | nrpages = mapping->nrpages; | |
423 | smp_rmb(); | |
424 | nrshadows = mapping->nrshadows; | |
425 | ||
426 | if (nrpages || nrshadows) { | |
427 | /* | |
428 | * As truncation uses a lockless tree lookup, cycle | |
429 | * the tree lock to make sure any ongoing tree | |
430 | * modification that does not see AS_EXITING is | |
431 | * completed before starting the final truncate. | |
432 | */ | |
433 | spin_lock_irq(&mapping->tree_lock); | |
434 | spin_unlock_irq(&mapping->tree_lock); | |
435 | ||
436 | truncate_inode_pages(mapping, 0); | |
437 | } | |
438 | } | |
439 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
440 | ||
28697355 MW |
441 | /** |
442 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
443 | * @mapping: the address_space which holds the pages to invalidate | |
444 | * @start: the offset 'from' which to invalidate | |
445 | * @end: the offset 'to' which to invalidate (inclusive) | |
446 | * | |
447 | * This function only removes the unlocked pages, if you want to | |
448 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
449 | * | |
450 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
451 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
452 | * pagetables. | |
453 | */ | |
454 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 455 | pgoff_t start, pgoff_t end) |
1da177e4 | 456 | { |
0cd6144a | 457 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 458 | struct pagevec pvec; |
b85e0eff | 459 | pgoff_t index = start; |
31560180 MK |
460 | unsigned long ret; |
461 | unsigned long count = 0; | |
1da177e4 LT |
462 | int i; |
463 | ||
464 | pagevec_init(&pvec, 0); | |
0cd6144a JW |
465 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
466 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
467 | indices)) { | |
1da177e4 LT |
468 | for (i = 0; i < pagevec_count(&pvec); i++) { |
469 | struct page *page = pvec.pages[i]; | |
e0f23603 | 470 | |
b85e0eff | 471 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 472 | index = indices[i]; |
b85e0eff HD |
473 | if (index > end) |
474 | break; | |
e0f23603 | 475 | |
0cd6144a JW |
476 | if (radix_tree_exceptional_entry(page)) { |
477 | clear_exceptional_entry(mapping, index, page); | |
478 | continue; | |
479 | } | |
480 | ||
b85e0eff HD |
481 | if (!trylock_page(page)) |
482 | continue; | |
483 | WARN_ON(page->index != index); | |
31560180 | 484 | ret = invalidate_inode_page(page); |
1da177e4 | 485 | unlock_page(page); |
31560180 MK |
486 | /* |
487 | * Invalidation is a hint that the page is no longer | |
488 | * of interest and try to speed up its reclaim. | |
489 | */ | |
490 | if (!ret) | |
cc5993bd | 491 | deactivate_file_page(page); |
31560180 | 492 | count += ret; |
1da177e4 | 493 | } |
0cd6144a | 494 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 495 | pagevec_release(&pvec); |
28697355 | 496 | cond_resched(); |
b85e0eff | 497 | index++; |
1da177e4 | 498 | } |
31560180 | 499 | return count; |
1da177e4 | 500 | } |
54bc4855 | 501 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 502 | |
bd4c8ce4 AM |
503 | /* |
504 | * This is like invalidate_complete_page(), except it ignores the page's | |
505 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
506 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
507 | * shrink_page_list() has a temp ref on them, or because they're transiently |
508 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
509 | */ |
510 | static int | |
511 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
512 | { | |
513 | if (page->mapping != mapping) | |
514 | return 0; | |
515 | ||
266cf658 | 516 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
517 | return 0; |
518 | ||
19fd6231 | 519 | spin_lock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
520 | if (PageDirty(page)) |
521 | goto failed; | |
522 | ||
266cf658 | 523 | BUG_ON(page_has_private(page)); |
91b0abe3 | 524 | __delete_from_page_cache(page, NULL); |
19fd6231 | 525 | spin_unlock_irq(&mapping->tree_lock); |
6072d13c LT |
526 | |
527 | if (mapping->a_ops->freepage) | |
528 | mapping->a_ops->freepage(page); | |
529 | ||
bd4c8ce4 AM |
530 | page_cache_release(page); /* pagecache ref */ |
531 | return 1; | |
532 | failed: | |
19fd6231 | 533 | spin_unlock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
534 | return 0; |
535 | } | |
536 | ||
e3db7691 TM |
537 | static int do_launder_page(struct address_space *mapping, struct page *page) |
538 | { | |
539 | if (!PageDirty(page)) | |
540 | return 0; | |
541 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
542 | return 0; | |
543 | return mapping->a_ops->launder_page(page); | |
544 | } | |
545 | ||
1da177e4 LT |
546 | /** |
547 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 548 | * @mapping: the address_space |
1da177e4 LT |
549 | * @start: the page offset 'from' which to invalidate |
550 | * @end: the page offset 'to' which to invalidate (inclusive) | |
551 | * | |
552 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
553 | * invalidation. | |
554 | * | |
6ccfa806 | 555 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
556 | */ |
557 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
558 | pgoff_t start, pgoff_t end) | |
559 | { | |
0cd6144a | 560 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 561 | struct pagevec pvec; |
b85e0eff | 562 | pgoff_t index; |
1da177e4 LT |
563 | int i; |
564 | int ret = 0; | |
0dd1334f | 565 | int ret2 = 0; |
1da177e4 | 566 | int did_range_unmap = 0; |
1da177e4 | 567 | |
3167760f | 568 | cleancache_invalidate_inode(mapping); |
1da177e4 | 569 | pagevec_init(&pvec, 0); |
b85e0eff | 570 | index = start; |
0cd6144a JW |
571 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
572 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
573 | indices)) { | |
7b965e08 | 574 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 575 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
576 | |
577 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 578 | index = indices[i]; |
b85e0eff HD |
579 | if (index > end) |
580 | break; | |
1da177e4 | 581 | |
0cd6144a JW |
582 | if (radix_tree_exceptional_entry(page)) { |
583 | clear_exceptional_entry(mapping, index, page); | |
584 | continue; | |
585 | } | |
586 | ||
1da177e4 | 587 | lock_page(page); |
b85e0eff | 588 | WARN_ON(page->index != index); |
1da177e4 LT |
589 | if (page->mapping != mapping) { |
590 | unlock_page(page); | |
591 | continue; | |
592 | } | |
1da177e4 | 593 | wait_on_page_writeback(page); |
d00806b1 | 594 | if (page_mapped(page)) { |
1da177e4 LT |
595 | if (!did_range_unmap) { |
596 | /* | |
597 | * Zap the rest of the file in one hit. | |
598 | */ | |
599 | unmap_mapping_range(mapping, | |
b85e0eff HD |
600 | (loff_t)index << PAGE_CACHE_SHIFT, |
601 | (loff_t)(1 + end - index) | |
602 | << PAGE_CACHE_SHIFT, | |
1da177e4 LT |
603 | 0); |
604 | did_range_unmap = 1; | |
605 | } else { | |
606 | /* | |
607 | * Just zap this page | |
608 | */ | |
609 | unmap_mapping_range(mapping, | |
b85e0eff HD |
610 | (loff_t)index << PAGE_CACHE_SHIFT, |
611 | PAGE_CACHE_SIZE, 0); | |
1da177e4 LT |
612 | } |
613 | } | |
d00806b1 | 614 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
615 | ret2 = do_launder_page(mapping, page); |
616 | if (ret2 == 0) { | |
617 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 618 | ret2 = -EBUSY; |
0dd1334f HH |
619 | } |
620 | if (ret2 < 0) | |
621 | ret = ret2; | |
1da177e4 LT |
622 | unlock_page(page); |
623 | } | |
0cd6144a | 624 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
625 | pagevec_release(&pvec); |
626 | cond_resched(); | |
b85e0eff | 627 | index++; |
1da177e4 | 628 | } |
3167760f | 629 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
630 | return ret; |
631 | } | |
632 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
633 | ||
634 | /** | |
635 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 636 | * @mapping: the address_space |
1da177e4 LT |
637 | * |
638 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
639 | * invalidation. | |
640 | * | |
e9de25dd | 641 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
642 | */ |
643 | int invalidate_inode_pages2(struct address_space *mapping) | |
644 | { | |
645 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
646 | } | |
647 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 648 | |
649 | /** | |
650 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
651 | * @inode: inode | |
8a549bea | 652 | * @newsize: new file size |
25d9e2d1 | 653 | * |
654 | * inode's new i_size must already be written before truncate_pagecache | |
655 | * is called. | |
656 | * | |
657 | * This function should typically be called before the filesystem | |
658 | * releases resources associated with the freed range (eg. deallocates | |
659 | * blocks). This way, pagecache will always stay logically coherent | |
660 | * with on-disk format, and the filesystem would not have to deal with | |
661 | * situations such as writepage being called for a page that has already | |
662 | * had its underlying blocks deallocated. | |
663 | */ | |
7caef267 | 664 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 665 | { |
cedabed4 | 666 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 667 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
668 | |
669 | /* | |
670 | * unmap_mapping_range is called twice, first simply for | |
671 | * efficiency so that truncate_inode_pages does fewer | |
672 | * single-page unmaps. However after this first call, and | |
673 | * before truncate_inode_pages finishes, it is possible for | |
674 | * private pages to be COWed, which remain after | |
675 | * truncate_inode_pages finishes, hence the second | |
676 | * unmap_mapping_range call must be made for correctness. | |
677 | */ | |
8a549bea HD |
678 | unmap_mapping_range(mapping, holebegin, 0, 1); |
679 | truncate_inode_pages(mapping, newsize); | |
680 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 681 | } |
682 | EXPORT_SYMBOL(truncate_pagecache); | |
683 | ||
2c27c65e CH |
684 | /** |
685 | * truncate_setsize - update inode and pagecache for a new file size | |
686 | * @inode: inode | |
687 | * @newsize: new file size | |
688 | * | |
382e27da JK |
689 | * truncate_setsize updates i_size and performs pagecache truncation (if |
690 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
691 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 692 | * |
77783d06 JK |
693 | * Must be called with a lock serializing truncates and writes (generally |
694 | * i_mutex but e.g. xfs uses a different lock) and before all filesystem | |
695 | * specific block truncation has been performed. | |
2c27c65e CH |
696 | */ |
697 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
698 | { | |
90a80202 JK |
699 | loff_t oldsize = inode->i_size; |
700 | ||
2c27c65e | 701 | i_size_write(inode, newsize); |
90a80202 JK |
702 | if (newsize > oldsize) |
703 | pagecache_isize_extended(inode, oldsize, newsize); | |
7caef267 | 704 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
705 | } |
706 | EXPORT_SYMBOL(truncate_setsize); | |
707 | ||
90a80202 JK |
708 | /** |
709 | * pagecache_isize_extended - update pagecache after extension of i_size | |
710 | * @inode: inode for which i_size was extended | |
711 | * @from: original inode size | |
712 | * @to: new inode size | |
713 | * | |
714 | * Handle extension of inode size either caused by extending truncate or by | |
715 | * write starting after current i_size. We mark the page straddling current | |
716 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
717 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
718 | * the page before user writes to the page via mmap after the i_size has been | |
719 | * changed. | |
720 | * | |
721 | * The function must be called after i_size is updated so that page fault | |
722 | * coming after we unlock the page will already see the new i_size. | |
723 | * The function must be called while we still hold i_mutex - this not only | |
724 | * makes sure i_size is stable but also that userspace cannot observe new | |
725 | * i_size value before we are prepared to store mmap writes at new inode size. | |
726 | */ | |
727 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
728 | { | |
729 | int bsize = 1 << inode->i_blkbits; | |
730 | loff_t rounded_from; | |
731 | struct page *page; | |
732 | pgoff_t index; | |
733 | ||
90a80202 JK |
734 | WARN_ON(to > inode->i_size); |
735 | ||
736 | if (from >= to || bsize == PAGE_CACHE_SIZE) | |
737 | return; | |
738 | /* Page straddling @from will not have any hole block created? */ | |
739 | rounded_from = round_up(from, bsize); | |
740 | if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1))) | |
741 | return; | |
742 | ||
743 | index = from >> PAGE_CACHE_SHIFT; | |
744 | page = find_lock_page(inode->i_mapping, index); | |
745 | /* Page not cached? Nothing to do */ | |
746 | if (!page) | |
747 | return; | |
748 | /* | |
749 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
750 | * is needed. | |
751 | */ | |
752 | if (page_mkclean(page)) | |
753 | set_page_dirty(page); | |
754 | unlock_page(page); | |
755 | page_cache_release(page); | |
756 | } | |
757 | EXPORT_SYMBOL(pagecache_isize_extended); | |
758 | ||
623e3db9 HD |
759 | /** |
760 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
761 | * @inode: inode | |
762 | * @lstart: offset of beginning of hole | |
763 | * @lend: offset of last byte of hole | |
764 | * | |
765 | * This function should typically be called before the filesystem | |
766 | * releases resources associated with the freed range (eg. deallocates | |
767 | * blocks). This way, pagecache will always stay logically coherent | |
768 | * with on-disk format, and the filesystem would not have to deal with | |
769 | * situations such as writepage being called for a page that has already | |
770 | * had its underlying blocks deallocated. | |
771 | */ | |
772 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
773 | { | |
774 | struct address_space *mapping = inode->i_mapping; | |
775 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
776 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
777 | /* | |
778 | * This rounding is currently just for example: unmap_mapping_range | |
779 | * expands its hole outwards, whereas we want it to contract the hole | |
780 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
781 | * doing their own page rounding first. Note that unmap_mapping_range |
782 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
783 | */ |
784 | ||
785 | /* | |
786 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
787 | * once (before truncating pagecache), and without "even_cows" flag: | |
788 | * hole-punching should not remove private COWed pages from the hole. | |
789 | */ | |
790 | if ((u64)unmap_end > (u64)unmap_start) | |
791 | unmap_mapping_range(mapping, unmap_start, | |
792 | 1 + unmap_end - unmap_start, 0); | |
793 | truncate_inode_pages_range(mapping, lstart, lend); | |
794 | } | |
795 | EXPORT_SYMBOL(truncate_pagecache_range); |