Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/truncate.c - code for taking down pages from address_spaces | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds | |
5 | * | |
e1f8e874 | 6 | * 10Sep2002 Andrew Morton |
1da177e4 LT |
7 | * Initial version. |
8 | */ | |
9 | ||
10 | #include <linux/kernel.h> | |
4af3c9cc | 11 | #include <linux/backing-dev.h> |
5a0e3ad6 | 12 | #include <linux/gfp.h> |
1da177e4 | 13 | #include <linux/mm.h> |
0fd0e6b0 | 14 | #include <linux/swap.h> |
b95f1b31 | 15 | #include <linux/export.h> |
1da177e4 | 16 | #include <linux/pagemap.h> |
01f2705d | 17 | #include <linux/highmem.h> |
1da177e4 | 18 | #include <linux/pagevec.h> |
e08748ce | 19 | #include <linux/task_io_accounting_ops.h> |
1da177e4 | 20 | #include <linux/buffer_head.h> /* grr. try_to_release_page, |
aaa4059b | 21 | do_invalidatepage */ |
c515e1fd | 22 | #include <linux/cleancache.h> |
ba470de4 | 23 | #include "internal.h" |
1da177e4 | 24 | |
0cd6144a JW |
25 | static void clear_exceptional_entry(struct address_space *mapping, |
26 | pgoff_t index, void *entry) | |
27 | { | |
449dd698 JW |
28 | struct radix_tree_node *node; |
29 | void **slot; | |
30 | ||
0cd6144a JW |
31 | /* Handled by shmem itself */ |
32 | if (shmem_mapping(mapping)) | |
33 | return; | |
34 | ||
35 | spin_lock_irq(&mapping->tree_lock); | |
36 | /* | |
37 | * Regular page slots are stabilized by the page lock even | |
38 | * without the tree itself locked. These unlocked entries | |
39 | * need verification under the tree lock. | |
40 | */ | |
449dd698 JW |
41 | if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot)) |
42 | goto unlock; | |
43 | if (*slot != entry) | |
44 | goto unlock; | |
45 | radix_tree_replace_slot(slot, NULL); | |
46 | mapping->nrshadows--; | |
47 | if (!node) | |
48 | goto unlock; | |
49 | workingset_node_shadows_dec(node); | |
50 | /* | |
51 | * Don't track node without shadow entries. | |
52 | * | |
53 | * Avoid acquiring the list_lru lock if already untracked. | |
54 | * The list_empty() test is safe as node->private_list is | |
55 | * protected by mapping->tree_lock. | |
56 | */ | |
57 | if (!workingset_node_shadows(node) && | |
58 | !list_empty(&node->private_list)) | |
59 | list_lru_del(&workingset_shadow_nodes, &node->private_list); | |
60 | __radix_tree_delete_node(&mapping->page_tree, node); | |
61 | unlock: | |
0cd6144a JW |
62 | spin_unlock_irq(&mapping->tree_lock); |
63 | } | |
1da177e4 | 64 | |
cf9a2ae8 | 65 | /** |
28bc44d7 | 66 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 67 | * @page: the page which is affected |
d47992f8 LC |
68 | * @offset: start of the range to invalidate |
69 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
70 | * |
71 | * do_invalidatepage() is called when all or part of the page has become | |
72 | * invalidated by a truncate operation. | |
73 | * | |
74 | * do_invalidatepage() does not have to release all buffers, but it must | |
75 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
76 | * is underway against any of the blocks which are outside the truncation | |
77 | * point. Because the caller is about to free (and possibly reuse) those | |
78 | * blocks on-disk. | |
79 | */ | |
d47992f8 LC |
80 | void do_invalidatepage(struct page *page, unsigned int offset, |
81 | unsigned int length) | |
cf9a2ae8 | 82 | { |
d47992f8 LC |
83 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
84 | ||
cf9a2ae8 | 85 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 86 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
87 | if (!invalidatepage) |
88 | invalidatepage = block_invalidatepage; | |
9361401e | 89 | #endif |
cf9a2ae8 | 90 | if (invalidatepage) |
d47992f8 | 91 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
92 | } |
93 | ||
ecdfc978 LT |
94 | /* |
95 | * This cancels just the dirty bit on the kernel page itself, it | |
96 | * does NOT actually remove dirty bits on any mmap's that may be | |
97 | * around. It also leaves the page tagged dirty, so any sync | |
98 | * activity will still find it on the dirty lists, and in particular, | |
99 | * clear_page_dirty_for_io() will still look at the dirty bits in | |
100 | * the VM. | |
101 | * | |
102 | * Doing this should *normally* only ever be done when a page | |
103 | * is truncated, and is not actually mapped anywhere at all. However, | |
104 | * fs/buffer.c does this when it notices that somebody has cleaned | |
105 | * out all the buffers on a page without actually doing it through | |
106 | * the VM. Can you say "ext3 is horribly ugly"? Tought you could. | |
107 | */ | |
fba2591b LT |
108 | void cancel_dirty_page(struct page *page, unsigned int account_size) |
109 | { | |
8368e328 LT |
110 | if (TestClearPageDirty(page)) { |
111 | struct address_space *mapping = page->mapping; | |
112 | if (mapping && mapping_cap_account_dirty(mapping)) { | |
113 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
c9e51e41 PZ |
114 | dec_bdi_stat(mapping->backing_dev_info, |
115 | BDI_RECLAIMABLE); | |
8368e328 LT |
116 | if (account_size) |
117 | task_io_account_cancelled_write(account_size); | |
118 | } | |
3e67c098 | 119 | } |
fba2591b | 120 | } |
8368e328 | 121 | EXPORT_SYMBOL(cancel_dirty_page); |
fba2591b | 122 | |
1da177e4 LT |
123 | /* |
124 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 125 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 126 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
127 | * |
128 | * We need to bale out if page->mapping is no longer equal to the original | |
129 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
fc0ecff6 | 130 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
131 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
132 | */ | |
750b4987 | 133 | static int |
1da177e4 LT |
134 | truncate_complete_page(struct address_space *mapping, struct page *page) |
135 | { | |
136 | if (page->mapping != mapping) | |
750b4987 | 137 | return -EIO; |
1da177e4 | 138 | |
266cf658 | 139 | if (page_has_private(page)) |
d47992f8 | 140 | do_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
1da177e4 | 141 | |
a2b34564 BS |
142 | cancel_dirty_page(page, PAGE_CACHE_SIZE); |
143 | ||
1da177e4 | 144 | ClearPageMappedToDisk(page); |
5adc7b51 | 145 | delete_from_page_cache(page); |
750b4987 | 146 | return 0; |
1da177e4 LT |
147 | } |
148 | ||
149 | /* | |
fc0ecff6 | 150 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 151 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
152 | * be marked dirty at any time too, so use remove_mapping which safely |
153 | * discards clean, unused pages. | |
1da177e4 LT |
154 | * |
155 | * Returns non-zero if the page was successfully invalidated. | |
156 | */ | |
157 | static int | |
158 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
159 | { | |
0fd0e6b0 NP |
160 | int ret; |
161 | ||
1da177e4 LT |
162 | if (page->mapping != mapping) |
163 | return 0; | |
164 | ||
266cf658 | 165 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
166 | return 0; |
167 | ||
0fd0e6b0 | 168 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
169 | |
170 | return ret; | |
1da177e4 LT |
171 | } |
172 | ||
750b4987 NP |
173 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
174 | { | |
175 | if (page_mapped(page)) { | |
176 | unmap_mapping_range(mapping, | |
177 | (loff_t)page->index << PAGE_CACHE_SHIFT, | |
178 | PAGE_CACHE_SIZE, 0); | |
179 | } | |
180 | return truncate_complete_page(mapping, page); | |
181 | } | |
182 | ||
25718736 AK |
183 | /* |
184 | * Used to get rid of pages on hardware memory corruption. | |
185 | */ | |
186 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
187 | { | |
188 | if (!mapping) | |
189 | return -EINVAL; | |
190 | /* | |
191 | * Only punch for normal data pages for now. | |
192 | * Handling other types like directories would need more auditing. | |
193 | */ | |
194 | if (!S_ISREG(mapping->host->i_mode)) | |
195 | return -EIO; | |
196 | return truncate_inode_page(mapping, page); | |
197 | } | |
198 | EXPORT_SYMBOL(generic_error_remove_page); | |
199 | ||
83f78668 WF |
200 | /* |
201 | * Safely invalidate one page from its pagecache mapping. | |
202 | * It only drops clean, unused pages. The page must be locked. | |
203 | * | |
204 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
205 | */ | |
206 | int invalidate_inode_page(struct page *page) | |
207 | { | |
208 | struct address_space *mapping = page_mapping(page); | |
209 | if (!mapping) | |
210 | return 0; | |
211 | if (PageDirty(page) || PageWriteback(page)) | |
212 | return 0; | |
213 | if (page_mapped(page)) | |
214 | return 0; | |
215 | return invalidate_complete_page(mapping, page); | |
216 | } | |
217 | ||
1da177e4 | 218 | /** |
73c1e204 | 219 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
220 | * @mapping: mapping to truncate |
221 | * @lstart: offset from which to truncate | |
5a720394 | 222 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 223 | * |
d7339071 | 224 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
225 | * specified offsets (and zeroing out partial pages |
226 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
227 | * |
228 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
229 | * block on page locks and it will not block on writeback. The second pass | |
230 | * will wait. This is to prevent as much IO as possible in the affected region. | |
231 | * The first pass will remove most pages, so the search cost of the second pass | |
232 | * is low. | |
233 | * | |
1da177e4 LT |
234 | * We pass down the cache-hot hint to the page freeing code. Even if the |
235 | * mapping is large, it is probably the case that the final pages are the most | |
236 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
237 | * |
238 | * Note that since ->invalidatepage() accepts range to invalidate | |
239 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
240 | * page aligned properly. | |
1da177e4 | 241 | */ |
d7339071 HR |
242 | void truncate_inode_pages_range(struct address_space *mapping, |
243 | loff_t lstart, loff_t lend) | |
1da177e4 | 244 | { |
5a720394 LC |
245 | pgoff_t start; /* inclusive */ |
246 | pgoff_t end; /* exclusive */ | |
247 | unsigned int partial_start; /* inclusive */ | |
248 | unsigned int partial_end; /* exclusive */ | |
249 | struct pagevec pvec; | |
0cd6144a | 250 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
251 | pgoff_t index; |
252 | int i; | |
1da177e4 | 253 | |
3167760f | 254 | cleancache_invalidate_inode(mapping); |
91b0abe3 | 255 | if (mapping->nrpages == 0 && mapping->nrshadows == 0) |
1da177e4 LT |
256 | return; |
257 | ||
5a720394 LC |
258 | /* Offsets within partial pages */ |
259 | partial_start = lstart & (PAGE_CACHE_SIZE - 1); | |
260 | partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); | |
261 | ||
262 | /* | |
263 | * 'start' and 'end' always covers the range of pages to be fully | |
264 | * truncated. Partial pages are covered with 'partial_start' at the | |
265 | * start of the range and 'partial_end' at the end of the range. | |
266 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
267 | */ | |
268 | start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
269 | if (lend == -1) | |
270 | /* | |
271 | * lend == -1 indicates end-of-file so we have to set 'end' | |
272 | * to the highest possible pgoff_t and since the type is | |
273 | * unsigned we're using -1. | |
274 | */ | |
275 | end = -1; | |
276 | else | |
277 | end = (lend + 1) >> PAGE_CACHE_SHIFT; | |
d7339071 | 278 | |
1da177e4 | 279 | pagevec_init(&pvec, 0); |
b85e0eff | 280 | index = start; |
0cd6144a JW |
281 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
282 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
283 | indices)) { | |
e5598f8b | 284 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
285 | for (i = 0; i < pagevec_count(&pvec); i++) { |
286 | struct page *page = pvec.pages[i]; | |
1da177e4 | 287 | |
b85e0eff | 288 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 289 | index = indices[i]; |
5a720394 | 290 | if (index >= end) |
d7339071 | 291 | break; |
d7339071 | 292 | |
0cd6144a JW |
293 | if (radix_tree_exceptional_entry(page)) { |
294 | clear_exceptional_entry(mapping, index, page); | |
295 | continue; | |
296 | } | |
297 | ||
529ae9aa | 298 | if (!trylock_page(page)) |
1da177e4 | 299 | continue; |
b85e0eff | 300 | WARN_ON(page->index != index); |
1da177e4 LT |
301 | if (PageWriteback(page)) { |
302 | unlock_page(page); | |
303 | continue; | |
304 | } | |
750b4987 | 305 | truncate_inode_page(mapping, page); |
1da177e4 LT |
306 | unlock_page(page); |
307 | } | |
0cd6144a | 308 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 309 | pagevec_release(&pvec); |
e5598f8b | 310 | mem_cgroup_uncharge_end(); |
1da177e4 | 311 | cond_resched(); |
b85e0eff | 312 | index++; |
1da177e4 LT |
313 | } |
314 | ||
5a720394 | 315 | if (partial_start) { |
1da177e4 LT |
316 | struct page *page = find_lock_page(mapping, start - 1); |
317 | if (page) { | |
5a720394 LC |
318 | unsigned int top = PAGE_CACHE_SIZE; |
319 | if (start > end) { | |
320 | /* Truncation within a single page */ | |
321 | top = partial_end; | |
322 | partial_end = 0; | |
323 | } | |
1da177e4 | 324 | wait_on_page_writeback(page); |
5a720394 LC |
325 | zero_user_segment(page, partial_start, top); |
326 | cleancache_invalidate_page(mapping, page); | |
327 | if (page_has_private(page)) | |
328 | do_invalidatepage(page, partial_start, | |
329 | top - partial_start); | |
1da177e4 LT |
330 | unlock_page(page); |
331 | page_cache_release(page); | |
332 | } | |
333 | } | |
5a720394 LC |
334 | if (partial_end) { |
335 | struct page *page = find_lock_page(mapping, end); | |
336 | if (page) { | |
337 | wait_on_page_writeback(page); | |
338 | zero_user_segment(page, 0, partial_end); | |
339 | cleancache_invalidate_page(mapping, page); | |
340 | if (page_has_private(page)) | |
341 | do_invalidatepage(page, 0, | |
342 | partial_end); | |
343 | unlock_page(page); | |
344 | page_cache_release(page); | |
345 | } | |
346 | } | |
347 | /* | |
348 | * If the truncation happened within a single page no pages | |
349 | * will be released, just zeroed, so we can bail out now. | |
350 | */ | |
351 | if (start >= end) | |
352 | return; | |
1da177e4 | 353 | |
b85e0eff | 354 | index = start; |
1da177e4 LT |
355 | for ( ; ; ) { |
356 | cond_resched(); | |
0cd6144a | 357 | if (!pagevec_lookup_entries(&pvec, mapping, index, |
792ceaef HD |
358 | min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { |
359 | /* If all gone from start onwards, we're done */ | |
b85e0eff | 360 | if (index == start) |
1da177e4 | 361 | break; |
792ceaef | 362 | /* Otherwise restart to make sure all gone */ |
b85e0eff | 363 | index = start; |
1da177e4 LT |
364 | continue; |
365 | } | |
0cd6144a | 366 | if (index == start && indices[0] >= end) { |
792ceaef | 367 | /* All gone out of hole to be punched, we're done */ |
0cd6144a | 368 | pagevec_remove_exceptionals(&pvec); |
d7339071 HR |
369 | pagevec_release(&pvec); |
370 | break; | |
371 | } | |
569b846d | 372 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
373 | for (i = 0; i < pagevec_count(&pvec); i++) { |
374 | struct page *page = pvec.pages[i]; | |
375 | ||
b85e0eff | 376 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 377 | index = indices[i]; |
792ceaef HD |
378 | if (index >= end) { |
379 | /* Restart punch to make sure all gone */ | |
380 | index = start - 1; | |
d7339071 | 381 | break; |
792ceaef | 382 | } |
b85e0eff | 383 | |
0cd6144a JW |
384 | if (radix_tree_exceptional_entry(page)) { |
385 | clear_exceptional_entry(mapping, index, page); | |
386 | continue; | |
387 | } | |
388 | ||
1da177e4 | 389 | lock_page(page); |
b85e0eff | 390 | WARN_ON(page->index != index); |
1da177e4 | 391 | wait_on_page_writeback(page); |
750b4987 | 392 | truncate_inode_page(mapping, page); |
1da177e4 LT |
393 | unlock_page(page); |
394 | } | |
0cd6144a | 395 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 396 | pagevec_release(&pvec); |
569b846d | 397 | mem_cgroup_uncharge_end(); |
b85e0eff | 398 | index++; |
1da177e4 | 399 | } |
3167760f | 400 | cleancache_invalidate_inode(mapping); |
1da177e4 | 401 | } |
d7339071 | 402 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 403 | |
d7339071 HR |
404 | /** |
405 | * truncate_inode_pages - truncate *all* the pages from an offset | |
406 | * @mapping: mapping to truncate | |
407 | * @lstart: offset from which to truncate | |
408 | * | |
1b1dcc1b | 409 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
410 | * |
411 | * Note: When this function returns, there can be a page in the process of | |
412 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
413 | * mapping->nrpages can be non-zero when this function returns even after | |
414 | * truncation of the whole mapping. | |
d7339071 HR |
415 | */ |
416 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
417 | { | |
418 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
419 | } | |
1da177e4 LT |
420 | EXPORT_SYMBOL(truncate_inode_pages); |
421 | ||
91b0abe3 JW |
422 | /** |
423 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
424 | * @mapping: mapping to truncate | |
425 | * | |
426 | * Called under (and serialized by) inode->i_mutex. | |
427 | * | |
428 | * Filesystems have to use this in the .evict_inode path to inform the | |
429 | * VM that this is the final truncate and the inode is going away. | |
430 | */ | |
431 | void truncate_inode_pages_final(struct address_space *mapping) | |
432 | { | |
433 | unsigned long nrshadows; | |
434 | unsigned long nrpages; | |
435 | ||
436 | /* | |
437 | * Page reclaim can not participate in regular inode lifetime | |
438 | * management (can't call iput()) and thus can race with the | |
439 | * inode teardown. Tell it when the address space is exiting, | |
440 | * so that it does not install eviction information after the | |
441 | * final truncate has begun. | |
442 | */ | |
443 | mapping_set_exiting(mapping); | |
444 | ||
445 | /* | |
446 | * When reclaim installs eviction entries, it increases | |
447 | * nrshadows first, then decreases nrpages. Make sure we see | |
448 | * this in the right order or we might miss an entry. | |
449 | */ | |
450 | nrpages = mapping->nrpages; | |
451 | smp_rmb(); | |
452 | nrshadows = mapping->nrshadows; | |
453 | ||
454 | if (nrpages || nrshadows) { | |
455 | /* | |
456 | * As truncation uses a lockless tree lookup, cycle | |
457 | * the tree lock to make sure any ongoing tree | |
458 | * modification that does not see AS_EXITING is | |
459 | * completed before starting the final truncate. | |
460 | */ | |
461 | spin_lock_irq(&mapping->tree_lock); | |
462 | spin_unlock_irq(&mapping->tree_lock); | |
463 | ||
464 | truncate_inode_pages(mapping, 0); | |
465 | } | |
466 | } | |
467 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
468 | ||
28697355 MW |
469 | /** |
470 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
471 | * @mapping: the address_space which holds the pages to invalidate | |
472 | * @start: the offset 'from' which to invalidate | |
473 | * @end: the offset 'to' which to invalidate (inclusive) | |
474 | * | |
475 | * This function only removes the unlocked pages, if you want to | |
476 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
477 | * | |
478 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
479 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
480 | * pagetables. | |
481 | */ | |
482 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 483 | pgoff_t start, pgoff_t end) |
1da177e4 | 484 | { |
0cd6144a | 485 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 486 | struct pagevec pvec; |
b85e0eff | 487 | pgoff_t index = start; |
31560180 MK |
488 | unsigned long ret; |
489 | unsigned long count = 0; | |
1da177e4 LT |
490 | int i; |
491 | ||
492 | pagevec_init(&pvec, 0); | |
0cd6144a JW |
493 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
494 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
495 | indices)) { | |
569b846d | 496 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
497 | for (i = 0; i < pagevec_count(&pvec); i++) { |
498 | struct page *page = pvec.pages[i]; | |
e0f23603 | 499 | |
b85e0eff | 500 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 501 | index = indices[i]; |
b85e0eff HD |
502 | if (index > end) |
503 | break; | |
e0f23603 | 504 | |
0cd6144a JW |
505 | if (radix_tree_exceptional_entry(page)) { |
506 | clear_exceptional_entry(mapping, index, page); | |
507 | continue; | |
508 | } | |
509 | ||
b85e0eff HD |
510 | if (!trylock_page(page)) |
511 | continue; | |
512 | WARN_ON(page->index != index); | |
31560180 | 513 | ret = invalidate_inode_page(page); |
1da177e4 | 514 | unlock_page(page); |
31560180 MK |
515 | /* |
516 | * Invalidation is a hint that the page is no longer | |
517 | * of interest and try to speed up its reclaim. | |
518 | */ | |
519 | if (!ret) | |
520 | deactivate_page(page); | |
521 | count += ret; | |
1da177e4 | 522 | } |
0cd6144a | 523 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 524 | pagevec_release(&pvec); |
569b846d | 525 | mem_cgroup_uncharge_end(); |
28697355 | 526 | cond_resched(); |
b85e0eff | 527 | index++; |
1da177e4 | 528 | } |
31560180 | 529 | return count; |
1da177e4 | 530 | } |
54bc4855 | 531 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 532 | |
bd4c8ce4 AM |
533 | /* |
534 | * This is like invalidate_complete_page(), except it ignores the page's | |
535 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
536 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
537 | * shrink_page_list() has a temp ref on them, or because they're transiently |
538 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
539 | */ |
540 | static int | |
541 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
542 | { | |
543 | if (page->mapping != mapping) | |
544 | return 0; | |
545 | ||
266cf658 | 546 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
547 | return 0; |
548 | ||
19fd6231 | 549 | spin_lock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
550 | if (PageDirty(page)) |
551 | goto failed; | |
552 | ||
266cf658 | 553 | BUG_ON(page_has_private(page)); |
91b0abe3 | 554 | __delete_from_page_cache(page, NULL); |
19fd6231 | 555 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 556 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
557 | |
558 | if (mapping->a_ops->freepage) | |
559 | mapping->a_ops->freepage(page); | |
560 | ||
bd4c8ce4 AM |
561 | page_cache_release(page); /* pagecache ref */ |
562 | return 1; | |
563 | failed: | |
19fd6231 | 564 | spin_unlock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
565 | return 0; |
566 | } | |
567 | ||
e3db7691 TM |
568 | static int do_launder_page(struct address_space *mapping, struct page *page) |
569 | { | |
570 | if (!PageDirty(page)) | |
571 | return 0; | |
572 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
573 | return 0; | |
574 | return mapping->a_ops->launder_page(page); | |
575 | } | |
576 | ||
1da177e4 LT |
577 | /** |
578 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 579 | * @mapping: the address_space |
1da177e4 LT |
580 | * @start: the page offset 'from' which to invalidate |
581 | * @end: the page offset 'to' which to invalidate (inclusive) | |
582 | * | |
583 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
584 | * invalidation. | |
585 | * | |
6ccfa806 | 586 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
587 | */ |
588 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
589 | pgoff_t start, pgoff_t end) | |
590 | { | |
0cd6144a | 591 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 592 | struct pagevec pvec; |
b85e0eff | 593 | pgoff_t index; |
1da177e4 LT |
594 | int i; |
595 | int ret = 0; | |
0dd1334f | 596 | int ret2 = 0; |
1da177e4 | 597 | int did_range_unmap = 0; |
1da177e4 | 598 | |
3167760f | 599 | cleancache_invalidate_inode(mapping); |
1da177e4 | 600 | pagevec_init(&pvec, 0); |
b85e0eff | 601 | index = start; |
0cd6144a JW |
602 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
603 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
604 | indices)) { | |
569b846d | 605 | mem_cgroup_uncharge_start(); |
7b965e08 | 606 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 607 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
608 | |
609 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 610 | index = indices[i]; |
b85e0eff HD |
611 | if (index > end) |
612 | break; | |
1da177e4 | 613 | |
0cd6144a JW |
614 | if (radix_tree_exceptional_entry(page)) { |
615 | clear_exceptional_entry(mapping, index, page); | |
616 | continue; | |
617 | } | |
618 | ||
1da177e4 | 619 | lock_page(page); |
b85e0eff | 620 | WARN_ON(page->index != index); |
1da177e4 LT |
621 | if (page->mapping != mapping) { |
622 | unlock_page(page); | |
623 | continue; | |
624 | } | |
1da177e4 | 625 | wait_on_page_writeback(page); |
d00806b1 | 626 | if (page_mapped(page)) { |
1da177e4 LT |
627 | if (!did_range_unmap) { |
628 | /* | |
629 | * Zap the rest of the file in one hit. | |
630 | */ | |
631 | unmap_mapping_range(mapping, | |
b85e0eff HD |
632 | (loff_t)index << PAGE_CACHE_SHIFT, |
633 | (loff_t)(1 + end - index) | |
634 | << PAGE_CACHE_SHIFT, | |
1da177e4 LT |
635 | 0); |
636 | did_range_unmap = 1; | |
637 | } else { | |
638 | /* | |
639 | * Just zap this page | |
640 | */ | |
641 | unmap_mapping_range(mapping, | |
b85e0eff HD |
642 | (loff_t)index << PAGE_CACHE_SHIFT, |
643 | PAGE_CACHE_SIZE, 0); | |
1da177e4 LT |
644 | } |
645 | } | |
d00806b1 | 646 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
647 | ret2 = do_launder_page(mapping, page); |
648 | if (ret2 == 0) { | |
649 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 650 | ret2 = -EBUSY; |
0dd1334f HH |
651 | } |
652 | if (ret2 < 0) | |
653 | ret = ret2; | |
1da177e4 LT |
654 | unlock_page(page); |
655 | } | |
0cd6144a | 656 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 657 | pagevec_release(&pvec); |
569b846d | 658 | mem_cgroup_uncharge_end(); |
1da177e4 | 659 | cond_resched(); |
b85e0eff | 660 | index++; |
1da177e4 | 661 | } |
3167760f | 662 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
663 | return ret; |
664 | } | |
665 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
666 | ||
667 | /** | |
668 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 669 | * @mapping: the address_space |
1da177e4 LT |
670 | * |
671 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
672 | * invalidation. | |
673 | * | |
e9de25dd | 674 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
675 | */ |
676 | int invalidate_inode_pages2(struct address_space *mapping) | |
677 | { | |
678 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
679 | } | |
680 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 681 | |
682 | /** | |
683 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
684 | * @inode: inode | |
8a549bea | 685 | * @newsize: new file size |
25d9e2d1 | 686 | * |
687 | * inode's new i_size must already be written before truncate_pagecache | |
688 | * is called. | |
689 | * | |
690 | * This function should typically be called before the filesystem | |
691 | * releases resources associated with the freed range (eg. deallocates | |
692 | * blocks). This way, pagecache will always stay logically coherent | |
693 | * with on-disk format, and the filesystem would not have to deal with | |
694 | * situations such as writepage being called for a page that has already | |
695 | * had its underlying blocks deallocated. | |
696 | */ | |
7caef267 | 697 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 698 | { |
cedabed4 | 699 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 700 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
701 | |
702 | /* | |
703 | * unmap_mapping_range is called twice, first simply for | |
704 | * efficiency so that truncate_inode_pages does fewer | |
705 | * single-page unmaps. However after this first call, and | |
706 | * before truncate_inode_pages finishes, it is possible for | |
707 | * private pages to be COWed, which remain after | |
708 | * truncate_inode_pages finishes, hence the second | |
709 | * unmap_mapping_range call must be made for correctness. | |
710 | */ | |
8a549bea HD |
711 | unmap_mapping_range(mapping, holebegin, 0, 1); |
712 | truncate_inode_pages(mapping, newsize); | |
713 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 714 | } |
715 | EXPORT_SYMBOL(truncate_pagecache); | |
716 | ||
2c27c65e CH |
717 | /** |
718 | * truncate_setsize - update inode and pagecache for a new file size | |
719 | * @inode: inode | |
720 | * @newsize: new file size | |
721 | * | |
382e27da JK |
722 | * truncate_setsize updates i_size and performs pagecache truncation (if |
723 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
724 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 725 | * |
382e27da JK |
726 | * Must be called with inode_mutex held and before all filesystem specific |
727 | * block truncation has been performed. | |
2c27c65e CH |
728 | */ |
729 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
730 | { | |
2c27c65e | 731 | i_size_write(inode, newsize); |
7caef267 | 732 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
733 | } |
734 | EXPORT_SYMBOL(truncate_setsize); | |
735 | ||
623e3db9 HD |
736 | /** |
737 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
738 | * @inode: inode | |
739 | * @lstart: offset of beginning of hole | |
740 | * @lend: offset of last byte of hole | |
741 | * | |
742 | * This function should typically be called before the filesystem | |
743 | * releases resources associated with the freed range (eg. deallocates | |
744 | * blocks). This way, pagecache will always stay logically coherent | |
745 | * with on-disk format, and the filesystem would not have to deal with | |
746 | * situations such as writepage being called for a page that has already | |
747 | * had its underlying blocks deallocated. | |
748 | */ | |
749 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
750 | { | |
751 | struct address_space *mapping = inode->i_mapping; | |
752 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
753 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
754 | /* | |
755 | * This rounding is currently just for example: unmap_mapping_range | |
756 | * expands its hole outwards, whereas we want it to contract the hole | |
757 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
758 | * doing their own page rounding first. Note that unmap_mapping_range |
759 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
760 | */ |
761 | ||
762 | /* | |
763 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
764 | * once (before truncating pagecache), and without "even_cows" flag: | |
765 | * hole-punching should not remove private COWed pages from the hole. | |
766 | */ | |
767 | if ((u64)unmap_end > (u64)unmap_start) | |
768 | unmap_mapping_range(mapping, unmap_start, | |
769 | 1 + unmap_end - unmap_start, 0); | |
770 | truncate_inode_pages_range(mapping, lstart, lend); | |
771 | } | |
772 | EXPORT_SYMBOL(truncate_pagecache_range); |