]>
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 | |
c6dcf52c JK |
27 | static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, |
28 | void *entry) | |
0cd6144a | 29 | { |
449dd698 JW |
30 | struct radix_tree_node *node; |
31 | void **slot; | |
32 | ||
ac401cc7 JK |
33 | spin_lock_irq(&mapping->tree_lock); |
34 | /* | |
35 | * Regular page slots are stabilized by the page lock even | |
36 | * without the tree itself locked. These unlocked entries | |
37 | * need verification under the tree lock. | |
38 | */ | |
14b46879 | 39 | if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot)) |
ac401cc7 JK |
40 | goto unlock; |
41 | if (*slot != entry) | |
42 | goto unlock; | |
14b46879 JW |
43 | __radix_tree_replace(&mapping->page_tree, node, slot, NULL, |
44 | workingset_update_node, mapping); | |
ac401cc7 | 45 | mapping->nrexceptional--; |
449dd698 | 46 | unlock: |
0cd6144a JW |
47 | spin_unlock_irq(&mapping->tree_lock); |
48 | } | |
1da177e4 | 49 | |
c6dcf52c JK |
50 | /* |
51 | * Unconditionally remove exceptional entry. Usually called from truncate path. | |
52 | */ | |
53 | static void truncate_exceptional_entry(struct address_space *mapping, | |
54 | pgoff_t index, void *entry) | |
55 | { | |
56 | /* Handled by shmem itself */ | |
57 | if (shmem_mapping(mapping)) | |
58 | return; | |
59 | ||
60 | if (dax_mapping(mapping)) { | |
61 | dax_delete_mapping_entry(mapping, index); | |
62 | return; | |
63 | } | |
64 | clear_shadow_entry(mapping, index, entry); | |
65 | } | |
66 | ||
67 | /* | |
68 | * Invalidate exceptional entry if easily possible. This handles exceptional | |
8b13a5f4 | 69 | * entries for invalidate_inode_pages(). |
c6dcf52c JK |
70 | */ |
71 | static int invalidate_exceptional_entry(struct address_space *mapping, | |
72 | pgoff_t index, void *entry) | |
73 | { | |
8b13a5f4 RZ |
74 | /* Handled by shmem itself, or for DAX we do nothing. */ |
75 | if (shmem_mapping(mapping) || dax_mapping(mapping)) | |
c6dcf52c | 76 | return 1; |
c6dcf52c JK |
77 | clear_shadow_entry(mapping, index, entry); |
78 | return 1; | |
79 | } | |
80 | ||
81 | /* | |
82 | * Invalidate exceptional entry if clean. This handles exceptional entries for | |
83 | * invalidate_inode_pages2() so for DAX it evicts only clean entries. | |
84 | */ | |
85 | static int invalidate_exceptional_entry2(struct address_space *mapping, | |
86 | pgoff_t index, void *entry) | |
87 | { | |
88 | /* Handled by shmem itself */ | |
89 | if (shmem_mapping(mapping)) | |
90 | return 1; | |
91 | if (dax_mapping(mapping)) | |
92 | return dax_invalidate_mapping_entry_sync(mapping, index); | |
93 | clear_shadow_entry(mapping, index, entry); | |
94 | return 1; | |
95 | } | |
96 | ||
cf9a2ae8 | 97 | /** |
28bc44d7 | 98 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 99 | * @page: the page which is affected |
d47992f8 LC |
100 | * @offset: start of the range to invalidate |
101 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
102 | * |
103 | * do_invalidatepage() is called when all or part of the page has become | |
104 | * invalidated by a truncate operation. | |
105 | * | |
106 | * do_invalidatepage() does not have to release all buffers, but it must | |
107 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
108 | * is underway against any of the blocks which are outside the truncation | |
109 | * point. Because the caller is about to free (and possibly reuse) those | |
110 | * blocks on-disk. | |
111 | */ | |
d47992f8 LC |
112 | void do_invalidatepage(struct page *page, unsigned int offset, |
113 | unsigned int length) | |
cf9a2ae8 | 114 | { |
d47992f8 LC |
115 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
116 | ||
cf9a2ae8 | 117 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 118 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
119 | if (!invalidatepage) |
120 | invalidatepage = block_invalidatepage; | |
9361401e | 121 | #endif |
cf9a2ae8 | 122 | if (invalidatepage) |
d47992f8 | 123 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
124 | } |
125 | ||
1da177e4 LT |
126 | /* |
127 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 128 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 129 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
130 | * |
131 | * We need to bale out if page->mapping is no longer equal to the original | |
132 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
fc0ecff6 | 133 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
134 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
135 | */ | |
750b4987 | 136 | static int |
1da177e4 LT |
137 | truncate_complete_page(struct address_space *mapping, struct page *page) |
138 | { | |
139 | if (page->mapping != mapping) | |
750b4987 | 140 | return -EIO; |
1da177e4 | 141 | |
266cf658 | 142 | if (page_has_private(page)) |
09cbfeaf | 143 | do_invalidatepage(page, 0, PAGE_SIZE); |
1da177e4 | 144 | |
b9ea2515 KK |
145 | /* |
146 | * Some filesystems seem to re-dirty the page even after | |
147 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
148 | * Hence dirty accounting check is placed after invalidation. | |
149 | */ | |
11f81bec | 150 | cancel_dirty_page(page); |
1da177e4 | 151 | ClearPageMappedToDisk(page); |
5adc7b51 | 152 | delete_from_page_cache(page); |
750b4987 | 153 | return 0; |
1da177e4 LT |
154 | } |
155 | ||
156 | /* | |
fc0ecff6 | 157 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 158 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
159 | * be marked dirty at any time too, so use remove_mapping which safely |
160 | * discards clean, unused pages. | |
1da177e4 LT |
161 | * |
162 | * Returns non-zero if the page was successfully invalidated. | |
163 | */ | |
164 | static int | |
165 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
166 | { | |
0fd0e6b0 NP |
167 | int ret; |
168 | ||
1da177e4 LT |
169 | if (page->mapping != mapping) |
170 | return 0; | |
171 | ||
266cf658 | 172 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
173 | return 0; |
174 | ||
0fd0e6b0 | 175 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
176 | |
177 | return ret; | |
1da177e4 LT |
178 | } |
179 | ||
750b4987 NP |
180 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
181 | { | |
fc127da0 KS |
182 | loff_t holelen; |
183 | VM_BUG_ON_PAGE(PageTail(page), page); | |
184 | ||
185 | holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE; | |
750b4987 NP |
186 | if (page_mapped(page)) { |
187 | unmap_mapping_range(mapping, | |
09cbfeaf | 188 | (loff_t)page->index << PAGE_SHIFT, |
fc127da0 | 189 | holelen, 0); |
750b4987 NP |
190 | } |
191 | return truncate_complete_page(mapping, page); | |
192 | } | |
193 | ||
25718736 AK |
194 | /* |
195 | * Used to get rid of pages on hardware memory corruption. | |
196 | */ | |
197 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
198 | { | |
199 | if (!mapping) | |
200 | return -EINVAL; | |
201 | /* | |
202 | * Only punch for normal data pages for now. | |
203 | * Handling other types like directories would need more auditing. | |
204 | */ | |
205 | if (!S_ISREG(mapping->host->i_mode)) | |
206 | return -EIO; | |
207 | return truncate_inode_page(mapping, page); | |
208 | } | |
209 | EXPORT_SYMBOL(generic_error_remove_page); | |
210 | ||
83f78668 WF |
211 | /* |
212 | * Safely invalidate one page from its pagecache mapping. | |
213 | * It only drops clean, unused pages. The page must be locked. | |
214 | * | |
215 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
216 | */ | |
217 | int invalidate_inode_page(struct page *page) | |
218 | { | |
219 | struct address_space *mapping = page_mapping(page); | |
220 | if (!mapping) | |
221 | return 0; | |
222 | if (PageDirty(page) || PageWriteback(page)) | |
223 | return 0; | |
224 | if (page_mapped(page)) | |
225 | return 0; | |
226 | return invalidate_complete_page(mapping, page); | |
227 | } | |
228 | ||
1da177e4 | 229 | /** |
73c1e204 | 230 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
231 | * @mapping: mapping to truncate |
232 | * @lstart: offset from which to truncate | |
5a720394 | 233 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 234 | * |
d7339071 | 235 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
236 | * specified offsets (and zeroing out partial pages |
237 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
238 | * |
239 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
240 | * block on page locks and it will not block on writeback. The second pass | |
241 | * will wait. This is to prevent as much IO as possible in the affected region. | |
242 | * The first pass will remove most pages, so the search cost of the second pass | |
243 | * is low. | |
244 | * | |
1da177e4 LT |
245 | * We pass down the cache-hot hint to the page freeing code. Even if the |
246 | * mapping is large, it is probably the case that the final pages are the most | |
247 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
248 | * |
249 | * Note that since ->invalidatepage() accepts range to invalidate | |
250 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
251 | * page aligned properly. | |
1da177e4 | 252 | */ |
d7339071 HR |
253 | void truncate_inode_pages_range(struct address_space *mapping, |
254 | loff_t lstart, loff_t lend) | |
1da177e4 | 255 | { |
5a720394 LC |
256 | pgoff_t start; /* inclusive */ |
257 | pgoff_t end; /* exclusive */ | |
258 | unsigned int partial_start; /* inclusive */ | |
259 | unsigned int partial_end; /* exclusive */ | |
260 | struct pagevec pvec; | |
0cd6144a | 261 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
262 | pgoff_t index; |
263 | int i; | |
1da177e4 | 264 | |
3167760f | 265 | cleancache_invalidate_inode(mapping); |
f9fe48be | 266 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
1da177e4 LT |
267 | return; |
268 | ||
5a720394 | 269 | /* Offsets within partial pages */ |
09cbfeaf KS |
270 | partial_start = lstart & (PAGE_SIZE - 1); |
271 | partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
5a720394 LC |
272 | |
273 | /* | |
274 | * 'start' and 'end' always covers the range of pages to be fully | |
275 | * truncated. Partial pages are covered with 'partial_start' at the | |
276 | * start of the range and 'partial_end' at the end of the range. | |
277 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
278 | */ | |
09cbfeaf | 279 | start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
5a720394 LC |
280 | if (lend == -1) |
281 | /* | |
282 | * lend == -1 indicates end-of-file so we have to set 'end' | |
283 | * to the highest possible pgoff_t and since the type is | |
284 | * unsigned we're using -1. | |
285 | */ | |
286 | end = -1; | |
287 | else | |
09cbfeaf | 288 | end = (lend + 1) >> PAGE_SHIFT; |
d7339071 | 289 | |
1da177e4 | 290 | pagevec_init(&pvec, 0); |
b85e0eff | 291 | index = start; |
0cd6144a JW |
292 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
293 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
294 | indices)) { | |
1da177e4 LT |
295 | for (i = 0; i < pagevec_count(&pvec); i++) { |
296 | struct page *page = pvec.pages[i]; | |
1da177e4 | 297 | |
b85e0eff | 298 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 299 | index = indices[i]; |
5a720394 | 300 | if (index >= end) |
d7339071 | 301 | break; |
d7339071 | 302 | |
0cd6144a | 303 | if (radix_tree_exceptional_entry(page)) { |
c6dcf52c JK |
304 | truncate_exceptional_entry(mapping, index, |
305 | page); | |
0cd6144a JW |
306 | continue; |
307 | } | |
308 | ||
529ae9aa | 309 | if (!trylock_page(page)) |
1da177e4 | 310 | continue; |
5cbc198a | 311 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
312 | if (PageWriteback(page)) { |
313 | unlock_page(page); | |
314 | continue; | |
315 | } | |
750b4987 | 316 | truncate_inode_page(mapping, page); |
1da177e4 LT |
317 | unlock_page(page); |
318 | } | |
0cd6144a | 319 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
320 | pagevec_release(&pvec); |
321 | cond_resched(); | |
b85e0eff | 322 | index++; |
1da177e4 LT |
323 | } |
324 | ||
5a720394 | 325 | if (partial_start) { |
1da177e4 LT |
326 | struct page *page = find_lock_page(mapping, start - 1); |
327 | if (page) { | |
09cbfeaf | 328 | unsigned int top = PAGE_SIZE; |
5a720394 LC |
329 | if (start > end) { |
330 | /* Truncation within a single page */ | |
331 | top = partial_end; | |
332 | partial_end = 0; | |
333 | } | |
1da177e4 | 334 | wait_on_page_writeback(page); |
5a720394 LC |
335 | zero_user_segment(page, partial_start, top); |
336 | cleancache_invalidate_page(mapping, page); | |
337 | if (page_has_private(page)) | |
338 | do_invalidatepage(page, partial_start, | |
339 | top - partial_start); | |
1da177e4 | 340 | unlock_page(page); |
09cbfeaf | 341 | put_page(page); |
1da177e4 LT |
342 | } |
343 | } | |
5a720394 LC |
344 | if (partial_end) { |
345 | struct page *page = find_lock_page(mapping, end); | |
346 | if (page) { | |
347 | wait_on_page_writeback(page); | |
348 | zero_user_segment(page, 0, partial_end); | |
349 | cleancache_invalidate_page(mapping, page); | |
350 | if (page_has_private(page)) | |
351 | do_invalidatepage(page, 0, | |
352 | partial_end); | |
353 | unlock_page(page); | |
09cbfeaf | 354 | put_page(page); |
5a720394 LC |
355 | } |
356 | } | |
357 | /* | |
358 | * If the truncation happened within a single page no pages | |
359 | * will be released, just zeroed, so we can bail out now. | |
360 | */ | |
361 | if (start >= end) | |
362 | return; | |
1da177e4 | 363 | |
b85e0eff | 364 | index = start; |
1da177e4 LT |
365 | for ( ; ; ) { |
366 | cond_resched(); | |
0cd6144a | 367 | if (!pagevec_lookup_entries(&pvec, mapping, index, |
792ceaef HD |
368 | min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { |
369 | /* If all gone from start onwards, we're done */ | |
b85e0eff | 370 | if (index == start) |
1da177e4 | 371 | break; |
792ceaef | 372 | /* Otherwise restart to make sure all gone */ |
b85e0eff | 373 | index = start; |
1da177e4 LT |
374 | continue; |
375 | } | |
0cd6144a | 376 | if (index == start && indices[0] >= end) { |
792ceaef | 377 | /* All gone out of hole to be punched, we're done */ |
0cd6144a | 378 | pagevec_remove_exceptionals(&pvec); |
d7339071 HR |
379 | pagevec_release(&pvec); |
380 | break; | |
381 | } | |
1da177e4 LT |
382 | for (i = 0; i < pagevec_count(&pvec); i++) { |
383 | struct page *page = pvec.pages[i]; | |
384 | ||
b85e0eff | 385 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 386 | index = indices[i]; |
792ceaef HD |
387 | if (index >= end) { |
388 | /* Restart punch to make sure all gone */ | |
389 | index = start - 1; | |
d7339071 | 390 | break; |
792ceaef | 391 | } |
b85e0eff | 392 | |
0cd6144a | 393 | if (radix_tree_exceptional_entry(page)) { |
c6dcf52c JK |
394 | truncate_exceptional_entry(mapping, index, |
395 | page); | |
0cd6144a JW |
396 | continue; |
397 | } | |
398 | ||
1da177e4 | 399 | lock_page(page); |
5cbc198a | 400 | WARN_ON(page_to_index(page) != index); |
1da177e4 | 401 | wait_on_page_writeback(page); |
750b4987 | 402 | truncate_inode_page(mapping, page); |
1da177e4 LT |
403 | unlock_page(page); |
404 | } | |
0cd6144a | 405 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 406 | pagevec_release(&pvec); |
b85e0eff | 407 | index++; |
1da177e4 | 408 | } |
3167760f | 409 | cleancache_invalidate_inode(mapping); |
1da177e4 | 410 | } |
d7339071 | 411 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 412 | |
d7339071 HR |
413 | /** |
414 | * truncate_inode_pages - truncate *all* the pages from an offset | |
415 | * @mapping: mapping to truncate | |
416 | * @lstart: offset from which to truncate | |
417 | * | |
1b1dcc1b | 418 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
419 | * |
420 | * Note: When this function returns, there can be a page in the process of | |
421 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
422 | * mapping->nrpages can be non-zero when this function returns even after | |
423 | * truncation of the whole mapping. | |
d7339071 HR |
424 | */ |
425 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
426 | { | |
427 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
428 | } | |
1da177e4 LT |
429 | EXPORT_SYMBOL(truncate_inode_pages); |
430 | ||
91b0abe3 JW |
431 | /** |
432 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
433 | * @mapping: mapping to truncate | |
434 | * | |
435 | * Called under (and serialized by) inode->i_mutex. | |
436 | * | |
437 | * Filesystems have to use this in the .evict_inode path to inform the | |
438 | * VM that this is the final truncate and the inode is going away. | |
439 | */ | |
440 | void truncate_inode_pages_final(struct address_space *mapping) | |
441 | { | |
f9fe48be | 442 | unsigned long nrexceptional; |
91b0abe3 JW |
443 | unsigned long nrpages; |
444 | ||
445 | /* | |
446 | * Page reclaim can not participate in regular inode lifetime | |
447 | * management (can't call iput()) and thus can race with the | |
448 | * inode teardown. Tell it when the address space is exiting, | |
449 | * so that it does not install eviction information after the | |
450 | * final truncate has begun. | |
451 | */ | |
452 | mapping_set_exiting(mapping); | |
453 | ||
454 | /* | |
455 | * When reclaim installs eviction entries, it increases | |
f9fe48be | 456 | * nrexceptional first, then decreases nrpages. Make sure we see |
91b0abe3 JW |
457 | * this in the right order or we might miss an entry. |
458 | */ | |
459 | nrpages = mapping->nrpages; | |
460 | smp_rmb(); | |
f9fe48be | 461 | nrexceptional = mapping->nrexceptional; |
91b0abe3 | 462 | |
f9fe48be | 463 | if (nrpages || nrexceptional) { |
91b0abe3 JW |
464 | /* |
465 | * As truncation uses a lockless tree lookup, cycle | |
466 | * the tree lock to make sure any ongoing tree | |
467 | * modification that does not see AS_EXITING is | |
468 | * completed before starting the final truncate. | |
469 | */ | |
470 | spin_lock_irq(&mapping->tree_lock); | |
471 | spin_unlock_irq(&mapping->tree_lock); | |
472 | ||
473 | truncate_inode_pages(mapping, 0); | |
474 | } | |
475 | } | |
476 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
477 | ||
28697355 MW |
478 | /** |
479 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
480 | * @mapping: the address_space which holds the pages to invalidate | |
481 | * @start: the offset 'from' which to invalidate | |
482 | * @end: the offset 'to' which to invalidate (inclusive) | |
483 | * | |
484 | * This function only removes the unlocked pages, if you want to | |
485 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
486 | * | |
487 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
488 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
489 | * pagetables. | |
490 | */ | |
491 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 492 | pgoff_t start, pgoff_t end) |
1da177e4 | 493 | { |
0cd6144a | 494 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 495 | struct pagevec pvec; |
b85e0eff | 496 | pgoff_t index = start; |
31560180 MK |
497 | unsigned long ret; |
498 | unsigned long count = 0; | |
1da177e4 LT |
499 | int i; |
500 | ||
501 | pagevec_init(&pvec, 0); | |
0cd6144a JW |
502 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
503 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
504 | indices)) { | |
1da177e4 LT |
505 | for (i = 0; i < pagevec_count(&pvec); i++) { |
506 | struct page *page = pvec.pages[i]; | |
e0f23603 | 507 | |
b85e0eff | 508 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 509 | index = indices[i]; |
b85e0eff HD |
510 | if (index > end) |
511 | break; | |
e0f23603 | 512 | |
0cd6144a | 513 | if (radix_tree_exceptional_entry(page)) { |
c6dcf52c JK |
514 | invalidate_exceptional_entry(mapping, index, |
515 | page); | |
0cd6144a JW |
516 | continue; |
517 | } | |
518 | ||
b85e0eff HD |
519 | if (!trylock_page(page)) |
520 | continue; | |
fc127da0 | 521 | |
5cbc198a | 522 | WARN_ON(page_to_index(page) != index); |
fc127da0 KS |
523 | |
524 | /* Middle of THP: skip */ | |
525 | if (PageTransTail(page)) { | |
526 | unlock_page(page); | |
527 | continue; | |
528 | } else if (PageTransHuge(page)) { | |
529 | index += HPAGE_PMD_NR - 1; | |
530 | i += HPAGE_PMD_NR - 1; | |
531 | /* 'end' is in the middle of THP */ | |
532 | if (index == round_down(end, HPAGE_PMD_NR)) | |
533 | continue; | |
534 | } | |
535 | ||
31560180 | 536 | ret = invalidate_inode_page(page); |
1da177e4 | 537 | unlock_page(page); |
31560180 MK |
538 | /* |
539 | * Invalidation is a hint that the page is no longer | |
540 | * of interest and try to speed up its reclaim. | |
541 | */ | |
542 | if (!ret) | |
cc5993bd | 543 | deactivate_file_page(page); |
31560180 | 544 | count += ret; |
1da177e4 | 545 | } |
0cd6144a | 546 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 547 | pagevec_release(&pvec); |
28697355 | 548 | cond_resched(); |
b85e0eff | 549 | index++; |
1da177e4 | 550 | } |
31560180 | 551 | return count; |
1da177e4 | 552 | } |
54bc4855 | 553 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 554 | |
bd4c8ce4 AM |
555 | /* |
556 | * This is like invalidate_complete_page(), except it ignores the page's | |
557 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
558 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
559 | * shrink_page_list() has a temp ref on them, or because they're transiently |
560 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
561 | */ |
562 | static int | |
563 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
564 | { | |
c4843a75 GT |
565 | unsigned long flags; |
566 | ||
bd4c8ce4 AM |
567 | if (page->mapping != mapping) |
568 | return 0; | |
569 | ||
266cf658 | 570 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
571 | return 0; |
572 | ||
c4843a75 | 573 | spin_lock_irqsave(&mapping->tree_lock, flags); |
bd4c8ce4 AM |
574 | if (PageDirty(page)) |
575 | goto failed; | |
576 | ||
266cf658 | 577 | BUG_ON(page_has_private(page)); |
62cccb8c | 578 | __delete_from_page_cache(page, NULL); |
c4843a75 | 579 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6072d13c LT |
580 | |
581 | if (mapping->a_ops->freepage) | |
582 | mapping->a_ops->freepage(page); | |
583 | ||
09cbfeaf | 584 | put_page(page); /* pagecache ref */ |
bd4c8ce4 AM |
585 | return 1; |
586 | failed: | |
c4843a75 | 587 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
bd4c8ce4 AM |
588 | return 0; |
589 | } | |
590 | ||
e3db7691 TM |
591 | static int do_launder_page(struct address_space *mapping, struct page *page) |
592 | { | |
593 | if (!PageDirty(page)) | |
594 | return 0; | |
595 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
596 | return 0; | |
597 | return mapping->a_ops->launder_page(page); | |
598 | } | |
599 | ||
1da177e4 LT |
600 | /** |
601 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 602 | * @mapping: the address_space |
1da177e4 LT |
603 | * @start: the page offset 'from' which to invalidate |
604 | * @end: the page offset 'to' which to invalidate (inclusive) | |
605 | * | |
606 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
607 | * invalidation. | |
608 | * | |
6ccfa806 | 609 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
610 | */ |
611 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
612 | pgoff_t start, pgoff_t end) | |
613 | { | |
0cd6144a | 614 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 615 | struct pagevec pvec; |
b85e0eff | 616 | pgoff_t index; |
1da177e4 LT |
617 | int i; |
618 | int ret = 0; | |
0dd1334f | 619 | int ret2 = 0; |
1da177e4 | 620 | int did_range_unmap = 0; |
1da177e4 | 621 | |
3167760f | 622 | cleancache_invalidate_inode(mapping); |
1da177e4 | 623 | pagevec_init(&pvec, 0); |
b85e0eff | 624 | index = start; |
0cd6144a JW |
625 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
626 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
627 | indices)) { | |
7b965e08 | 628 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 629 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
630 | |
631 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 632 | index = indices[i]; |
b85e0eff HD |
633 | if (index > end) |
634 | break; | |
1da177e4 | 635 | |
0cd6144a | 636 | if (radix_tree_exceptional_entry(page)) { |
c6dcf52c JK |
637 | if (!invalidate_exceptional_entry2(mapping, |
638 | index, page)) | |
639 | ret = -EBUSY; | |
0cd6144a JW |
640 | continue; |
641 | } | |
642 | ||
1da177e4 | 643 | lock_page(page); |
5cbc198a | 644 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
645 | if (page->mapping != mapping) { |
646 | unlock_page(page); | |
647 | continue; | |
648 | } | |
1da177e4 | 649 | wait_on_page_writeback(page); |
d00806b1 | 650 | if (page_mapped(page)) { |
1da177e4 LT |
651 | if (!did_range_unmap) { |
652 | /* | |
653 | * Zap the rest of the file in one hit. | |
654 | */ | |
655 | unmap_mapping_range(mapping, | |
09cbfeaf | 656 | (loff_t)index << PAGE_SHIFT, |
b85e0eff | 657 | (loff_t)(1 + end - index) |
09cbfeaf KS |
658 | << PAGE_SHIFT, |
659 | 0); | |
1da177e4 LT |
660 | did_range_unmap = 1; |
661 | } else { | |
662 | /* | |
663 | * Just zap this page | |
664 | */ | |
665 | unmap_mapping_range(mapping, | |
09cbfeaf KS |
666 | (loff_t)index << PAGE_SHIFT, |
667 | PAGE_SIZE, 0); | |
1da177e4 LT |
668 | } |
669 | } | |
d00806b1 | 670 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
671 | ret2 = do_launder_page(mapping, page); |
672 | if (ret2 == 0) { | |
673 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 674 | ret2 = -EBUSY; |
0dd1334f HH |
675 | } |
676 | if (ret2 < 0) | |
677 | ret = ret2; | |
1da177e4 LT |
678 | unlock_page(page); |
679 | } | |
0cd6144a | 680 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
681 | pagevec_release(&pvec); |
682 | cond_resched(); | |
b85e0eff | 683 | index++; |
1da177e4 | 684 | } |
32250644 JK |
685 | /* |
686 | * For DAX we invalidate page tables after invalidating radix tree. We | |
687 | * could invalidate page tables while invalidating each entry however | |
688 | * that would be expensive. And doing range unmapping before doesn't | |
689 | * work as we have no cheap way to find whether radix tree entry didn't | |
690 | * get remapped later. | |
691 | */ | |
692 | if (dax_mapping(mapping)) { | |
693 | unmap_mapping_range(mapping, (loff_t)start << PAGE_SHIFT, | |
694 | (loff_t)(end - start + 1) << PAGE_SHIFT, 0); | |
695 | } | |
3167760f | 696 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
697 | return ret; |
698 | } | |
699 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
700 | ||
701 | /** | |
702 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 703 | * @mapping: the address_space |
1da177e4 LT |
704 | * |
705 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
706 | * invalidation. | |
707 | * | |
e9de25dd | 708 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
709 | */ |
710 | int invalidate_inode_pages2(struct address_space *mapping) | |
711 | { | |
712 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
713 | } | |
714 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 715 | |
716 | /** | |
717 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
718 | * @inode: inode | |
8a549bea | 719 | * @newsize: new file size |
25d9e2d1 | 720 | * |
721 | * inode's new i_size must already be written before truncate_pagecache | |
722 | * is called. | |
723 | * | |
724 | * This function should typically be called before the filesystem | |
725 | * releases resources associated with the freed range (eg. deallocates | |
726 | * blocks). This way, pagecache will always stay logically coherent | |
727 | * with on-disk format, and the filesystem would not have to deal with | |
728 | * situations such as writepage being called for a page that has already | |
729 | * had its underlying blocks deallocated. | |
730 | */ | |
7caef267 | 731 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 732 | { |
cedabed4 | 733 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 734 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
735 | |
736 | /* | |
737 | * unmap_mapping_range is called twice, first simply for | |
738 | * efficiency so that truncate_inode_pages does fewer | |
739 | * single-page unmaps. However after this first call, and | |
740 | * before truncate_inode_pages finishes, it is possible for | |
741 | * private pages to be COWed, which remain after | |
742 | * truncate_inode_pages finishes, hence the second | |
743 | * unmap_mapping_range call must be made for correctness. | |
744 | */ | |
8a549bea HD |
745 | unmap_mapping_range(mapping, holebegin, 0, 1); |
746 | truncate_inode_pages(mapping, newsize); | |
747 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 748 | } |
749 | EXPORT_SYMBOL(truncate_pagecache); | |
750 | ||
2c27c65e CH |
751 | /** |
752 | * truncate_setsize - update inode and pagecache for a new file size | |
753 | * @inode: inode | |
754 | * @newsize: new file size | |
755 | * | |
382e27da JK |
756 | * truncate_setsize updates i_size and performs pagecache truncation (if |
757 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
758 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 759 | * |
77783d06 JK |
760 | * Must be called with a lock serializing truncates and writes (generally |
761 | * i_mutex but e.g. xfs uses a different lock) and before all filesystem | |
762 | * specific block truncation has been performed. | |
2c27c65e CH |
763 | */ |
764 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
765 | { | |
90a80202 JK |
766 | loff_t oldsize = inode->i_size; |
767 | ||
2c27c65e | 768 | i_size_write(inode, newsize); |
90a80202 JK |
769 | if (newsize > oldsize) |
770 | pagecache_isize_extended(inode, oldsize, newsize); | |
7caef267 | 771 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
772 | } |
773 | EXPORT_SYMBOL(truncate_setsize); | |
774 | ||
90a80202 JK |
775 | /** |
776 | * pagecache_isize_extended - update pagecache after extension of i_size | |
777 | * @inode: inode for which i_size was extended | |
778 | * @from: original inode size | |
779 | * @to: new inode size | |
780 | * | |
781 | * Handle extension of inode size either caused by extending truncate or by | |
782 | * write starting after current i_size. We mark the page straddling current | |
783 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
784 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
785 | * the page before user writes to the page via mmap after the i_size has been | |
786 | * changed. | |
787 | * | |
788 | * The function must be called after i_size is updated so that page fault | |
789 | * coming after we unlock the page will already see the new i_size. | |
790 | * The function must be called while we still hold i_mutex - this not only | |
791 | * makes sure i_size is stable but also that userspace cannot observe new | |
792 | * i_size value before we are prepared to store mmap writes at new inode size. | |
793 | */ | |
794 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
795 | { | |
796 | int bsize = 1 << inode->i_blkbits; | |
797 | loff_t rounded_from; | |
798 | struct page *page; | |
799 | pgoff_t index; | |
800 | ||
90a80202 JK |
801 | WARN_ON(to > inode->i_size); |
802 | ||
09cbfeaf | 803 | if (from >= to || bsize == PAGE_SIZE) |
90a80202 JK |
804 | return; |
805 | /* Page straddling @from will not have any hole block created? */ | |
806 | rounded_from = round_up(from, bsize); | |
09cbfeaf | 807 | if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) |
90a80202 JK |
808 | return; |
809 | ||
09cbfeaf | 810 | index = from >> PAGE_SHIFT; |
90a80202 JK |
811 | page = find_lock_page(inode->i_mapping, index); |
812 | /* Page not cached? Nothing to do */ | |
813 | if (!page) | |
814 | return; | |
815 | /* | |
816 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
817 | * is needed. | |
818 | */ | |
819 | if (page_mkclean(page)) | |
820 | set_page_dirty(page); | |
821 | unlock_page(page); | |
09cbfeaf | 822 | put_page(page); |
90a80202 JK |
823 | } |
824 | EXPORT_SYMBOL(pagecache_isize_extended); | |
825 | ||
623e3db9 HD |
826 | /** |
827 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
828 | * @inode: inode | |
829 | * @lstart: offset of beginning of hole | |
830 | * @lend: offset of last byte of hole | |
831 | * | |
832 | * This function should typically be called before the filesystem | |
833 | * releases resources associated with the freed range (eg. deallocates | |
834 | * blocks). This way, pagecache will always stay logically coherent | |
835 | * with on-disk format, and the filesystem would not have to deal with | |
836 | * situations such as writepage being called for a page that has already | |
837 | * had its underlying blocks deallocated. | |
838 | */ | |
839 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
840 | { | |
841 | struct address_space *mapping = inode->i_mapping; | |
842 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
843 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
844 | /* | |
845 | * This rounding is currently just for example: unmap_mapping_range | |
846 | * expands its hole outwards, whereas we want it to contract the hole | |
847 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
848 | * doing their own page rounding first. Note that unmap_mapping_range |
849 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
850 | */ |
851 | ||
852 | /* | |
853 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
854 | * once (before truncating pagecache), and without "even_cows" flag: | |
855 | * hole-punching should not remove private COWed pages from the hole. | |
856 | */ | |
857 | if ((u64)unmap_end > (u64)unmap_start) | |
858 | unmap_mapping_range(mapping, unmap_start, | |
859 | 1 + unmap_end - unmap_start, 0); | |
860 | truncate_inode_pages_range(mapping, lstart, lend); | |
861 | } | |
862 | EXPORT_SYMBOL(truncate_pagecache_range); |