]>
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> |
f9fe48be | 12 | #include <linux/dax.h> |
5a0e3ad6 | 13 | #include <linux/gfp.h> |
1da177e4 | 14 | #include <linux/mm.h> |
0fd0e6b0 | 15 | #include <linux/swap.h> |
b95f1b31 | 16 | #include <linux/export.h> |
1da177e4 | 17 | #include <linux/pagemap.h> |
01f2705d | 18 | #include <linux/highmem.h> |
1da177e4 | 19 | #include <linux/pagevec.h> |
e08748ce | 20 | #include <linux/task_io_accounting_ops.h> |
1da177e4 | 21 | #include <linux/buffer_head.h> /* grr. try_to_release_page, |
aaa4059b | 22 | do_invalidatepage */ |
c515e1fd | 23 | #include <linux/cleancache.h> |
90a80202 | 24 | #include <linux/rmap.h> |
ba470de4 | 25 | #include "internal.h" |
1da177e4 | 26 | |
0cd6144a JW |
27 | static void clear_exceptional_entry(struct address_space *mapping, |
28 | pgoff_t index, void *entry) | |
29 | { | |
449dd698 JW |
30 | struct radix_tree_node *node; |
31 | void **slot; | |
32 | ||
0cd6144a JW |
33 | /* Handled by shmem itself */ |
34 | if (shmem_mapping(mapping)) | |
35 | return; | |
36 | ||
f9fe48be | 37 | if (dax_mapping(mapping)) { |
ac401cc7 JK |
38 | dax_delete_mapping_entry(mapping, index); |
39 | return; | |
f9fe48be | 40 | } |
ac401cc7 JK |
41 | spin_lock_irq(&mapping->tree_lock); |
42 | /* | |
43 | * Regular page slots are stabilized by the page lock even | |
44 | * without the tree itself locked. These unlocked entries | |
45 | * need verification under the tree lock. | |
46 | */ | |
14b46879 | 47 | if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot)) |
ac401cc7 JK |
48 | goto unlock; |
49 | if (*slot != entry) | |
50 | goto unlock; | |
14b46879 JW |
51 | __radix_tree_replace(&mapping->page_tree, node, slot, NULL, |
52 | workingset_update_node, mapping); | |
ac401cc7 | 53 | mapping->nrexceptional--; |
449dd698 | 54 | unlock: |
0cd6144a JW |
55 | spin_unlock_irq(&mapping->tree_lock); |
56 | } | |
1da177e4 | 57 | |
cf9a2ae8 | 58 | /** |
28bc44d7 | 59 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 60 | * @page: the page which is affected |
d47992f8 LC |
61 | * @offset: start of the range to invalidate |
62 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
63 | * |
64 | * do_invalidatepage() is called when all or part of the page has become | |
65 | * invalidated by a truncate operation. | |
66 | * | |
67 | * do_invalidatepage() does not have to release all buffers, but it must | |
68 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
69 | * is underway against any of the blocks which are outside the truncation | |
70 | * point. Because the caller is about to free (and possibly reuse) those | |
71 | * blocks on-disk. | |
72 | */ | |
d47992f8 LC |
73 | void do_invalidatepage(struct page *page, unsigned int offset, |
74 | unsigned int length) | |
cf9a2ae8 | 75 | { |
d47992f8 LC |
76 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
77 | ||
cf9a2ae8 | 78 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 79 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
80 | if (!invalidatepage) |
81 | invalidatepage = block_invalidatepage; | |
9361401e | 82 | #endif |
cf9a2ae8 | 83 | if (invalidatepage) |
d47992f8 | 84 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
85 | } |
86 | ||
1da177e4 LT |
87 | /* |
88 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 89 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 90 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
91 | * |
92 | * We need to bale out if page->mapping is no longer equal to the original | |
93 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
fc0ecff6 | 94 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
95 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
96 | */ | |
750b4987 | 97 | static int |
1da177e4 LT |
98 | truncate_complete_page(struct address_space *mapping, struct page *page) |
99 | { | |
100 | if (page->mapping != mapping) | |
750b4987 | 101 | return -EIO; |
1da177e4 | 102 | |
266cf658 | 103 | if (page_has_private(page)) |
09cbfeaf | 104 | do_invalidatepage(page, 0, PAGE_SIZE); |
1da177e4 | 105 | |
b9ea2515 KK |
106 | /* |
107 | * Some filesystems seem to re-dirty the page even after | |
108 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
109 | * Hence dirty accounting check is placed after invalidation. | |
110 | */ | |
11f81bec | 111 | cancel_dirty_page(page); |
1da177e4 | 112 | ClearPageMappedToDisk(page); |
5adc7b51 | 113 | delete_from_page_cache(page); |
750b4987 | 114 | return 0; |
1da177e4 LT |
115 | } |
116 | ||
117 | /* | |
fc0ecff6 | 118 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 119 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
120 | * be marked dirty at any time too, so use remove_mapping which safely |
121 | * discards clean, unused pages. | |
1da177e4 LT |
122 | * |
123 | * Returns non-zero if the page was successfully invalidated. | |
124 | */ | |
125 | static int | |
126 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
127 | { | |
0fd0e6b0 NP |
128 | int ret; |
129 | ||
1da177e4 LT |
130 | if (page->mapping != mapping) |
131 | return 0; | |
132 | ||
266cf658 | 133 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
134 | return 0; |
135 | ||
0fd0e6b0 | 136 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
137 | |
138 | return ret; | |
1da177e4 LT |
139 | } |
140 | ||
750b4987 NP |
141 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
142 | { | |
fc127da0 KS |
143 | loff_t holelen; |
144 | VM_BUG_ON_PAGE(PageTail(page), page); | |
145 | ||
146 | holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE; | |
750b4987 NP |
147 | if (page_mapped(page)) { |
148 | unmap_mapping_range(mapping, | |
09cbfeaf | 149 | (loff_t)page->index << PAGE_SHIFT, |
fc127da0 | 150 | holelen, 0); |
750b4987 NP |
151 | } |
152 | return truncate_complete_page(mapping, page); | |
153 | } | |
154 | ||
25718736 AK |
155 | /* |
156 | * Used to get rid of pages on hardware memory corruption. | |
157 | */ | |
158 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
159 | { | |
160 | if (!mapping) | |
161 | return -EINVAL; | |
162 | /* | |
163 | * Only punch for normal data pages for now. | |
164 | * Handling other types like directories would need more auditing. | |
165 | */ | |
166 | if (!S_ISREG(mapping->host->i_mode)) | |
167 | return -EIO; | |
168 | return truncate_inode_page(mapping, page); | |
169 | } | |
170 | EXPORT_SYMBOL(generic_error_remove_page); | |
171 | ||
83f78668 WF |
172 | /* |
173 | * Safely invalidate one page from its pagecache mapping. | |
174 | * It only drops clean, unused pages. The page must be locked. | |
175 | * | |
176 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
177 | */ | |
178 | int invalidate_inode_page(struct page *page) | |
179 | { | |
180 | struct address_space *mapping = page_mapping(page); | |
181 | if (!mapping) | |
182 | return 0; | |
183 | if (PageDirty(page) || PageWriteback(page)) | |
184 | return 0; | |
185 | if (page_mapped(page)) | |
186 | return 0; | |
187 | return invalidate_complete_page(mapping, page); | |
188 | } | |
189 | ||
1da177e4 | 190 | /** |
73c1e204 | 191 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
192 | * @mapping: mapping to truncate |
193 | * @lstart: offset from which to truncate | |
5a720394 | 194 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 195 | * |
d7339071 | 196 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
197 | * specified offsets (and zeroing out partial pages |
198 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
199 | * |
200 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
201 | * block on page locks and it will not block on writeback. The second pass | |
202 | * will wait. This is to prevent as much IO as possible in the affected region. | |
203 | * The first pass will remove most pages, so the search cost of the second pass | |
204 | * is low. | |
205 | * | |
1da177e4 LT |
206 | * We pass down the cache-hot hint to the page freeing code. Even if the |
207 | * mapping is large, it is probably the case that the final pages are the most | |
208 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
209 | * |
210 | * Note that since ->invalidatepage() accepts range to invalidate | |
211 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
212 | * page aligned properly. | |
1da177e4 | 213 | */ |
d7339071 HR |
214 | void truncate_inode_pages_range(struct address_space *mapping, |
215 | loff_t lstart, loff_t lend) | |
1da177e4 | 216 | { |
5a720394 LC |
217 | pgoff_t start; /* inclusive */ |
218 | pgoff_t end; /* exclusive */ | |
219 | unsigned int partial_start; /* inclusive */ | |
220 | unsigned int partial_end; /* exclusive */ | |
221 | struct pagevec pvec; | |
0cd6144a | 222 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
223 | pgoff_t index; |
224 | int i; | |
1da177e4 | 225 | |
3167760f | 226 | cleancache_invalidate_inode(mapping); |
f9fe48be | 227 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
1da177e4 LT |
228 | return; |
229 | ||
5a720394 | 230 | /* Offsets within partial pages */ |
09cbfeaf KS |
231 | partial_start = lstart & (PAGE_SIZE - 1); |
232 | partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
5a720394 LC |
233 | |
234 | /* | |
235 | * 'start' and 'end' always covers the range of pages to be fully | |
236 | * truncated. Partial pages are covered with 'partial_start' at the | |
237 | * start of the range and 'partial_end' at the end of the range. | |
238 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
239 | */ | |
09cbfeaf | 240 | start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
5a720394 LC |
241 | if (lend == -1) |
242 | /* | |
243 | * lend == -1 indicates end-of-file so we have to set 'end' | |
244 | * to the highest possible pgoff_t and since the type is | |
245 | * unsigned we're using -1. | |
246 | */ | |
247 | end = -1; | |
248 | else | |
09cbfeaf | 249 | end = (lend + 1) >> PAGE_SHIFT; |
d7339071 | 250 | |
1da177e4 | 251 | pagevec_init(&pvec, 0); |
b85e0eff | 252 | index = start; |
0cd6144a JW |
253 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
254 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
255 | indices)) { | |
1da177e4 LT |
256 | for (i = 0; i < pagevec_count(&pvec); i++) { |
257 | struct page *page = pvec.pages[i]; | |
1da177e4 | 258 | |
b85e0eff | 259 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 260 | index = indices[i]; |
5a720394 | 261 | if (index >= end) |
d7339071 | 262 | break; |
d7339071 | 263 | |
0cd6144a JW |
264 | if (radix_tree_exceptional_entry(page)) { |
265 | clear_exceptional_entry(mapping, index, page); | |
266 | continue; | |
267 | } | |
268 | ||
529ae9aa | 269 | if (!trylock_page(page)) |
1da177e4 | 270 | continue; |
5cbc198a | 271 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
272 | if (PageWriteback(page)) { |
273 | unlock_page(page); | |
274 | continue; | |
275 | } | |
750b4987 | 276 | truncate_inode_page(mapping, page); |
1da177e4 LT |
277 | unlock_page(page); |
278 | } | |
0cd6144a | 279 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
280 | pagevec_release(&pvec); |
281 | cond_resched(); | |
b85e0eff | 282 | index++; |
1da177e4 LT |
283 | } |
284 | ||
5a720394 | 285 | if (partial_start) { |
1da177e4 LT |
286 | struct page *page = find_lock_page(mapping, start - 1); |
287 | if (page) { | |
09cbfeaf | 288 | unsigned int top = PAGE_SIZE; |
5a720394 LC |
289 | if (start > end) { |
290 | /* Truncation within a single page */ | |
291 | top = partial_end; | |
292 | partial_end = 0; | |
293 | } | |
1da177e4 | 294 | wait_on_page_writeback(page); |
5a720394 LC |
295 | zero_user_segment(page, partial_start, top); |
296 | cleancache_invalidate_page(mapping, page); | |
297 | if (page_has_private(page)) | |
298 | do_invalidatepage(page, partial_start, | |
299 | top - partial_start); | |
1da177e4 | 300 | unlock_page(page); |
09cbfeaf | 301 | put_page(page); |
1da177e4 LT |
302 | } |
303 | } | |
5a720394 LC |
304 | if (partial_end) { |
305 | struct page *page = find_lock_page(mapping, end); | |
306 | if (page) { | |
307 | wait_on_page_writeback(page); | |
308 | zero_user_segment(page, 0, partial_end); | |
309 | cleancache_invalidate_page(mapping, page); | |
310 | if (page_has_private(page)) | |
311 | do_invalidatepage(page, 0, | |
312 | partial_end); | |
313 | unlock_page(page); | |
09cbfeaf | 314 | put_page(page); |
5a720394 LC |
315 | } |
316 | } | |
317 | /* | |
318 | * If the truncation happened within a single page no pages | |
319 | * will be released, just zeroed, so we can bail out now. | |
320 | */ | |
321 | if (start >= end) | |
322 | return; | |
1da177e4 | 323 | |
b85e0eff | 324 | index = start; |
1da177e4 LT |
325 | for ( ; ; ) { |
326 | cond_resched(); | |
0cd6144a | 327 | if (!pagevec_lookup_entries(&pvec, mapping, index, |
792ceaef HD |
328 | min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { |
329 | /* If all gone from start onwards, we're done */ | |
b85e0eff | 330 | if (index == start) |
1da177e4 | 331 | break; |
792ceaef | 332 | /* Otherwise restart to make sure all gone */ |
b85e0eff | 333 | index = start; |
1da177e4 LT |
334 | continue; |
335 | } | |
0cd6144a | 336 | if (index == start && indices[0] >= end) { |
792ceaef | 337 | /* All gone out of hole to be punched, we're done */ |
0cd6144a | 338 | pagevec_remove_exceptionals(&pvec); |
d7339071 HR |
339 | pagevec_release(&pvec); |
340 | break; | |
341 | } | |
1da177e4 LT |
342 | for (i = 0; i < pagevec_count(&pvec); i++) { |
343 | struct page *page = pvec.pages[i]; | |
344 | ||
b85e0eff | 345 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 346 | index = indices[i]; |
792ceaef HD |
347 | if (index >= end) { |
348 | /* Restart punch to make sure all gone */ | |
349 | index = start - 1; | |
d7339071 | 350 | break; |
792ceaef | 351 | } |
b85e0eff | 352 | |
0cd6144a JW |
353 | if (radix_tree_exceptional_entry(page)) { |
354 | clear_exceptional_entry(mapping, index, page); | |
355 | continue; | |
356 | } | |
357 | ||
1da177e4 | 358 | lock_page(page); |
5cbc198a | 359 | WARN_ON(page_to_index(page) != index); |
1da177e4 | 360 | wait_on_page_writeback(page); |
750b4987 | 361 | truncate_inode_page(mapping, page); |
1da177e4 LT |
362 | unlock_page(page); |
363 | } | |
0cd6144a | 364 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 365 | pagevec_release(&pvec); |
b85e0eff | 366 | index++; |
1da177e4 | 367 | } |
3167760f | 368 | cleancache_invalidate_inode(mapping); |
1da177e4 | 369 | } |
d7339071 | 370 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 371 | |
d7339071 HR |
372 | /** |
373 | * truncate_inode_pages - truncate *all* the pages from an offset | |
374 | * @mapping: mapping to truncate | |
375 | * @lstart: offset from which to truncate | |
376 | * | |
1b1dcc1b | 377 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
378 | * |
379 | * Note: When this function returns, there can be a page in the process of | |
380 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
381 | * mapping->nrpages can be non-zero when this function returns even after | |
382 | * truncation of the whole mapping. | |
d7339071 HR |
383 | */ |
384 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
385 | { | |
386 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
387 | } | |
1da177e4 LT |
388 | EXPORT_SYMBOL(truncate_inode_pages); |
389 | ||
91b0abe3 JW |
390 | /** |
391 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
392 | * @mapping: mapping to truncate | |
393 | * | |
394 | * Called under (and serialized by) inode->i_mutex. | |
395 | * | |
396 | * Filesystems have to use this in the .evict_inode path to inform the | |
397 | * VM that this is the final truncate and the inode is going away. | |
398 | */ | |
399 | void truncate_inode_pages_final(struct address_space *mapping) | |
400 | { | |
f9fe48be | 401 | unsigned long nrexceptional; |
91b0abe3 JW |
402 | unsigned long nrpages; |
403 | ||
404 | /* | |
405 | * Page reclaim can not participate in regular inode lifetime | |
406 | * management (can't call iput()) and thus can race with the | |
407 | * inode teardown. Tell it when the address space is exiting, | |
408 | * so that it does not install eviction information after the | |
409 | * final truncate has begun. | |
410 | */ | |
411 | mapping_set_exiting(mapping); | |
412 | ||
413 | /* | |
414 | * When reclaim installs eviction entries, it increases | |
f9fe48be | 415 | * nrexceptional first, then decreases nrpages. Make sure we see |
91b0abe3 JW |
416 | * this in the right order or we might miss an entry. |
417 | */ | |
418 | nrpages = mapping->nrpages; | |
419 | smp_rmb(); | |
f9fe48be | 420 | nrexceptional = mapping->nrexceptional; |
91b0abe3 | 421 | |
f9fe48be | 422 | if (nrpages || nrexceptional) { |
91b0abe3 JW |
423 | /* |
424 | * As truncation uses a lockless tree lookup, cycle | |
425 | * the tree lock to make sure any ongoing tree | |
426 | * modification that does not see AS_EXITING is | |
427 | * completed before starting the final truncate. | |
428 | */ | |
429 | spin_lock_irq(&mapping->tree_lock); | |
430 | spin_unlock_irq(&mapping->tree_lock); | |
431 | ||
432 | truncate_inode_pages(mapping, 0); | |
433 | } | |
434 | } | |
435 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
436 | ||
28697355 MW |
437 | /** |
438 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
439 | * @mapping: the address_space which holds the pages to invalidate | |
440 | * @start: the offset 'from' which to invalidate | |
441 | * @end: the offset 'to' which to invalidate (inclusive) | |
442 | * | |
443 | * This function only removes the unlocked pages, if you want to | |
444 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
445 | * | |
446 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
447 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
448 | * pagetables. | |
449 | */ | |
450 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 451 | pgoff_t start, pgoff_t end) |
1da177e4 | 452 | { |
0cd6144a | 453 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 454 | struct pagevec pvec; |
b85e0eff | 455 | pgoff_t index = start; |
31560180 MK |
456 | unsigned long ret; |
457 | unsigned long count = 0; | |
1da177e4 LT |
458 | int i; |
459 | ||
460 | pagevec_init(&pvec, 0); | |
0cd6144a JW |
461 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
462 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
463 | indices)) { | |
1da177e4 LT |
464 | for (i = 0; i < pagevec_count(&pvec); i++) { |
465 | struct page *page = pvec.pages[i]; | |
e0f23603 | 466 | |
b85e0eff | 467 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 468 | index = indices[i]; |
b85e0eff HD |
469 | if (index > end) |
470 | break; | |
e0f23603 | 471 | |
0cd6144a JW |
472 | if (radix_tree_exceptional_entry(page)) { |
473 | clear_exceptional_entry(mapping, index, page); | |
474 | continue; | |
475 | } | |
476 | ||
b85e0eff HD |
477 | if (!trylock_page(page)) |
478 | continue; | |
fc127da0 | 479 | |
5cbc198a | 480 | WARN_ON(page_to_index(page) != index); |
fc127da0 KS |
481 | |
482 | /* Middle of THP: skip */ | |
483 | if (PageTransTail(page)) { | |
484 | unlock_page(page); | |
485 | continue; | |
486 | } else if (PageTransHuge(page)) { | |
487 | index += HPAGE_PMD_NR - 1; | |
488 | i += HPAGE_PMD_NR - 1; | |
489 | /* 'end' is in the middle of THP */ | |
490 | if (index == round_down(end, HPAGE_PMD_NR)) | |
491 | continue; | |
492 | } | |
493 | ||
31560180 | 494 | ret = invalidate_inode_page(page); |
1da177e4 | 495 | unlock_page(page); |
31560180 MK |
496 | /* |
497 | * Invalidation is a hint that the page is no longer | |
498 | * of interest and try to speed up its reclaim. | |
499 | */ | |
500 | if (!ret) | |
cc5993bd | 501 | deactivate_file_page(page); |
31560180 | 502 | count += ret; |
1da177e4 | 503 | } |
0cd6144a | 504 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 505 | pagevec_release(&pvec); |
28697355 | 506 | cond_resched(); |
b85e0eff | 507 | index++; |
1da177e4 | 508 | } |
31560180 | 509 | return count; |
1da177e4 | 510 | } |
54bc4855 | 511 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 512 | |
bd4c8ce4 AM |
513 | /* |
514 | * This is like invalidate_complete_page(), except it ignores the page's | |
515 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
516 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
517 | * shrink_page_list() has a temp ref on them, or because they're transiently |
518 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
519 | */ |
520 | static int | |
521 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
522 | { | |
c4843a75 GT |
523 | unsigned long flags; |
524 | ||
bd4c8ce4 AM |
525 | if (page->mapping != mapping) |
526 | return 0; | |
527 | ||
266cf658 | 528 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
529 | return 0; |
530 | ||
c4843a75 | 531 | spin_lock_irqsave(&mapping->tree_lock, flags); |
bd4c8ce4 AM |
532 | if (PageDirty(page)) |
533 | goto failed; | |
534 | ||
266cf658 | 535 | BUG_ON(page_has_private(page)); |
62cccb8c | 536 | __delete_from_page_cache(page, NULL); |
c4843a75 | 537 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6072d13c LT |
538 | |
539 | if (mapping->a_ops->freepage) | |
540 | mapping->a_ops->freepage(page); | |
541 | ||
09cbfeaf | 542 | put_page(page); /* pagecache ref */ |
bd4c8ce4 AM |
543 | return 1; |
544 | failed: | |
c4843a75 | 545 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
bd4c8ce4 AM |
546 | return 0; |
547 | } | |
548 | ||
e3db7691 TM |
549 | static int do_launder_page(struct address_space *mapping, struct page *page) |
550 | { | |
551 | if (!PageDirty(page)) | |
552 | return 0; | |
553 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
554 | return 0; | |
555 | return mapping->a_ops->launder_page(page); | |
556 | } | |
557 | ||
1da177e4 LT |
558 | /** |
559 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 560 | * @mapping: the address_space |
1da177e4 LT |
561 | * @start: the page offset 'from' which to invalidate |
562 | * @end: the page offset 'to' which to invalidate (inclusive) | |
563 | * | |
564 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
565 | * invalidation. | |
566 | * | |
6ccfa806 | 567 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
568 | */ |
569 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
570 | pgoff_t start, pgoff_t end) | |
571 | { | |
0cd6144a | 572 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 573 | struct pagevec pvec; |
b85e0eff | 574 | pgoff_t index; |
1da177e4 LT |
575 | int i; |
576 | int ret = 0; | |
0dd1334f | 577 | int ret2 = 0; |
1da177e4 | 578 | int did_range_unmap = 0; |
1da177e4 | 579 | |
3167760f | 580 | cleancache_invalidate_inode(mapping); |
1da177e4 | 581 | pagevec_init(&pvec, 0); |
b85e0eff | 582 | index = start; |
0cd6144a JW |
583 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
584 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
585 | indices)) { | |
7b965e08 | 586 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 587 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
588 | |
589 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 590 | index = indices[i]; |
b85e0eff HD |
591 | if (index > end) |
592 | break; | |
1da177e4 | 593 | |
0cd6144a JW |
594 | if (radix_tree_exceptional_entry(page)) { |
595 | clear_exceptional_entry(mapping, index, page); | |
596 | continue; | |
597 | } | |
598 | ||
1da177e4 | 599 | lock_page(page); |
5cbc198a | 600 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
601 | if (page->mapping != mapping) { |
602 | unlock_page(page); | |
603 | continue; | |
604 | } | |
1da177e4 | 605 | wait_on_page_writeback(page); |
d00806b1 | 606 | if (page_mapped(page)) { |
1da177e4 LT |
607 | if (!did_range_unmap) { |
608 | /* | |
609 | * Zap the rest of the file in one hit. | |
610 | */ | |
611 | unmap_mapping_range(mapping, | |
09cbfeaf | 612 | (loff_t)index << PAGE_SHIFT, |
b85e0eff | 613 | (loff_t)(1 + end - index) |
09cbfeaf KS |
614 | << PAGE_SHIFT, |
615 | 0); | |
1da177e4 LT |
616 | did_range_unmap = 1; |
617 | } else { | |
618 | /* | |
619 | * Just zap this page | |
620 | */ | |
621 | unmap_mapping_range(mapping, | |
09cbfeaf KS |
622 | (loff_t)index << PAGE_SHIFT, |
623 | PAGE_SIZE, 0); | |
1da177e4 LT |
624 | } |
625 | } | |
d00806b1 | 626 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
627 | ret2 = do_launder_page(mapping, page); |
628 | if (ret2 == 0) { | |
629 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 630 | ret2 = -EBUSY; |
0dd1334f HH |
631 | } |
632 | if (ret2 < 0) | |
633 | ret = ret2; | |
1da177e4 LT |
634 | unlock_page(page); |
635 | } | |
0cd6144a | 636 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
637 | pagevec_release(&pvec); |
638 | cond_resched(); | |
b85e0eff | 639 | index++; |
1da177e4 | 640 | } |
3167760f | 641 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
642 | return ret; |
643 | } | |
644 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
645 | ||
646 | /** | |
647 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 648 | * @mapping: the address_space |
1da177e4 LT |
649 | * |
650 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
651 | * invalidation. | |
652 | * | |
e9de25dd | 653 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
654 | */ |
655 | int invalidate_inode_pages2(struct address_space *mapping) | |
656 | { | |
657 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
658 | } | |
659 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 660 | |
661 | /** | |
662 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
663 | * @inode: inode | |
8a549bea | 664 | * @newsize: new file size |
25d9e2d1 | 665 | * |
666 | * inode's new i_size must already be written before truncate_pagecache | |
667 | * is called. | |
668 | * | |
669 | * This function should typically be called before the filesystem | |
670 | * releases resources associated with the freed range (eg. deallocates | |
671 | * blocks). This way, pagecache will always stay logically coherent | |
672 | * with on-disk format, and the filesystem would not have to deal with | |
673 | * situations such as writepage being called for a page that has already | |
674 | * had its underlying blocks deallocated. | |
675 | */ | |
7caef267 | 676 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 677 | { |
cedabed4 | 678 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 679 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
680 | |
681 | /* | |
682 | * unmap_mapping_range is called twice, first simply for | |
683 | * efficiency so that truncate_inode_pages does fewer | |
684 | * single-page unmaps. However after this first call, and | |
685 | * before truncate_inode_pages finishes, it is possible for | |
686 | * private pages to be COWed, which remain after | |
687 | * truncate_inode_pages finishes, hence the second | |
688 | * unmap_mapping_range call must be made for correctness. | |
689 | */ | |
8a549bea HD |
690 | unmap_mapping_range(mapping, holebegin, 0, 1); |
691 | truncate_inode_pages(mapping, newsize); | |
692 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 693 | } |
694 | EXPORT_SYMBOL(truncate_pagecache); | |
695 | ||
2c27c65e CH |
696 | /** |
697 | * truncate_setsize - update inode and pagecache for a new file size | |
698 | * @inode: inode | |
699 | * @newsize: new file size | |
700 | * | |
382e27da JK |
701 | * truncate_setsize updates i_size and performs pagecache truncation (if |
702 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
703 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 704 | * |
77783d06 JK |
705 | * Must be called with a lock serializing truncates and writes (generally |
706 | * i_mutex but e.g. xfs uses a different lock) and before all filesystem | |
707 | * specific block truncation has been performed. | |
2c27c65e CH |
708 | */ |
709 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
710 | { | |
90a80202 JK |
711 | loff_t oldsize = inode->i_size; |
712 | ||
2c27c65e | 713 | i_size_write(inode, newsize); |
90a80202 JK |
714 | if (newsize > oldsize) |
715 | pagecache_isize_extended(inode, oldsize, newsize); | |
7caef267 | 716 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
717 | } |
718 | EXPORT_SYMBOL(truncate_setsize); | |
719 | ||
90a80202 JK |
720 | /** |
721 | * pagecache_isize_extended - update pagecache after extension of i_size | |
722 | * @inode: inode for which i_size was extended | |
723 | * @from: original inode size | |
724 | * @to: new inode size | |
725 | * | |
726 | * Handle extension of inode size either caused by extending truncate or by | |
727 | * write starting after current i_size. We mark the page straddling current | |
728 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
729 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
730 | * the page before user writes to the page via mmap after the i_size has been | |
731 | * changed. | |
732 | * | |
733 | * The function must be called after i_size is updated so that page fault | |
734 | * coming after we unlock the page will already see the new i_size. | |
735 | * The function must be called while we still hold i_mutex - this not only | |
736 | * makes sure i_size is stable but also that userspace cannot observe new | |
737 | * i_size value before we are prepared to store mmap writes at new inode size. | |
738 | */ | |
739 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
740 | { | |
741 | int bsize = 1 << inode->i_blkbits; | |
742 | loff_t rounded_from; | |
743 | struct page *page; | |
744 | pgoff_t index; | |
745 | ||
90a80202 JK |
746 | WARN_ON(to > inode->i_size); |
747 | ||
09cbfeaf | 748 | if (from >= to || bsize == PAGE_SIZE) |
90a80202 JK |
749 | return; |
750 | /* Page straddling @from will not have any hole block created? */ | |
751 | rounded_from = round_up(from, bsize); | |
09cbfeaf | 752 | if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) |
90a80202 JK |
753 | return; |
754 | ||
09cbfeaf | 755 | index = from >> PAGE_SHIFT; |
90a80202 JK |
756 | page = find_lock_page(inode->i_mapping, index); |
757 | /* Page not cached? Nothing to do */ | |
758 | if (!page) | |
759 | return; | |
760 | /* | |
761 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
762 | * is needed. | |
763 | */ | |
764 | if (page_mkclean(page)) | |
765 | set_page_dirty(page); | |
766 | unlock_page(page); | |
09cbfeaf | 767 | put_page(page); |
90a80202 JK |
768 | } |
769 | EXPORT_SYMBOL(pagecache_isize_extended); | |
770 | ||
623e3db9 HD |
771 | /** |
772 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
773 | * @inode: inode | |
774 | * @lstart: offset of beginning of hole | |
775 | * @lend: offset of last byte of hole | |
776 | * | |
777 | * This function should typically be called before the filesystem | |
778 | * releases resources associated with the freed range (eg. deallocates | |
779 | * blocks). This way, pagecache will always stay logically coherent | |
780 | * with on-disk format, and the filesystem would not have to deal with | |
781 | * situations such as writepage being called for a page that has already | |
782 | * had its underlying blocks deallocated. | |
783 | */ | |
784 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
785 | { | |
786 | struct address_space *mapping = inode->i_mapping; | |
787 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
788 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
789 | /* | |
790 | * This rounding is currently just for example: unmap_mapping_range | |
791 | * expands its hole outwards, whereas we want it to contract the hole | |
792 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
793 | * doing their own page rounding first. Note that unmap_mapping_range |
794 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
795 | */ |
796 | ||
797 | /* | |
798 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
799 | * once (before truncating pagecache), and without "even_cows" flag: | |
800 | * hole-punching should not remove private COWed pages from the hole. | |
801 | */ | |
802 | if ((u64)unmap_end > (u64)unmap_start) | |
803 | unmap_mapping_range(mapping, unmap_start, | |
804 | 1 + unmap_end - unmap_start, 0); | |
805 | truncate_inode_pages_range(mapping, lstart, lend); | |
806 | } | |
807 | EXPORT_SYMBOL(truncate_pagecache_range); |