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 */ |
3a4f8a0b | 23 | #include <linux/shmem_fs.h> |
c515e1fd | 24 | #include <linux/cleancache.h> |
90a80202 | 25 | #include <linux/rmap.h> |
ba470de4 | 26 | #include "internal.h" |
1da177e4 | 27 | |
c6dcf52c JK |
28 | static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, |
29 | void *entry) | |
0cd6144a | 30 | { |
449dd698 JW |
31 | struct radix_tree_node *node; |
32 | void **slot; | |
33 | ||
ac401cc7 JK |
34 | spin_lock_irq(&mapping->tree_lock); |
35 | /* | |
36 | * Regular page slots are stabilized by the page lock even | |
37 | * without the tree itself locked. These unlocked entries | |
38 | * need verification under the tree lock. | |
39 | */ | |
14b46879 | 40 | if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot)) |
ac401cc7 JK |
41 | goto unlock; |
42 | if (*slot != entry) | |
43 | goto unlock; | |
14b46879 JW |
44 | __radix_tree_replace(&mapping->page_tree, node, slot, NULL, |
45 | workingset_update_node, mapping); | |
ac401cc7 | 46 | mapping->nrexceptional--; |
449dd698 | 47 | unlock: |
0cd6144a JW |
48 | spin_unlock_irq(&mapping->tree_lock); |
49 | } | |
1da177e4 | 50 | |
c6dcf52c JK |
51 | /* |
52 | * Unconditionally remove exceptional entry. Usually called from truncate path. | |
53 | */ | |
54 | static void truncate_exceptional_entry(struct address_space *mapping, | |
55 | pgoff_t index, void *entry) | |
56 | { | |
57 | /* Handled by shmem itself */ | |
58 | if (shmem_mapping(mapping)) | |
59 | return; | |
60 | ||
61 | if (dax_mapping(mapping)) { | |
62 | dax_delete_mapping_entry(mapping, index); | |
63 | return; | |
64 | } | |
65 | clear_shadow_entry(mapping, index, entry); | |
66 | } | |
67 | ||
68 | /* | |
69 | * Invalidate exceptional entry if easily possible. This handles exceptional | |
4636e70b | 70 | * entries for invalidate_inode_pages(). |
c6dcf52c JK |
71 | */ |
72 | static int invalidate_exceptional_entry(struct address_space *mapping, | |
73 | pgoff_t index, void *entry) | |
74 | { | |
4636e70b RZ |
75 | /* Handled by shmem itself, or for DAX we do nothing. */ |
76 | if (shmem_mapping(mapping) || dax_mapping(mapping)) | |
c6dcf52c | 77 | return 1; |
c6dcf52c JK |
78 | clear_shadow_entry(mapping, index, entry); |
79 | return 1; | |
80 | } | |
81 | ||
82 | /* | |
83 | * Invalidate exceptional entry if clean. This handles exceptional entries for | |
84 | * invalidate_inode_pages2() so for DAX it evicts only clean entries. | |
85 | */ | |
86 | static int invalidate_exceptional_entry2(struct address_space *mapping, | |
87 | pgoff_t index, void *entry) | |
88 | { | |
89 | /* Handled by shmem itself */ | |
90 | if (shmem_mapping(mapping)) | |
91 | return 1; | |
92 | if (dax_mapping(mapping)) | |
93 | return dax_invalidate_mapping_entry_sync(mapping, index); | |
94 | clear_shadow_entry(mapping, index, entry); | |
95 | return 1; | |
96 | } | |
97 | ||
cf9a2ae8 | 98 | /** |
28bc44d7 | 99 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 100 | * @page: the page which is affected |
d47992f8 LC |
101 | * @offset: start of the range to invalidate |
102 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
103 | * |
104 | * do_invalidatepage() is called when all or part of the page has become | |
105 | * invalidated by a truncate operation. | |
106 | * | |
107 | * do_invalidatepage() does not have to release all buffers, but it must | |
108 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
109 | * is underway against any of the blocks which are outside the truncation | |
110 | * point. Because the caller is about to free (and possibly reuse) those | |
111 | * blocks on-disk. | |
112 | */ | |
d47992f8 LC |
113 | void do_invalidatepage(struct page *page, unsigned int offset, |
114 | unsigned int length) | |
cf9a2ae8 | 115 | { |
d47992f8 LC |
116 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
117 | ||
cf9a2ae8 | 118 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 119 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
120 | if (!invalidatepage) |
121 | invalidatepage = block_invalidatepage; | |
9361401e | 122 | #endif |
cf9a2ae8 | 123 | if (invalidatepage) |
d47992f8 | 124 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
125 | } |
126 | ||
1da177e4 LT |
127 | /* |
128 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 129 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 130 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
131 | * |
132 | * We need to bale out if page->mapping is no longer equal to the original | |
133 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
fc0ecff6 | 134 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
135 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
136 | */ | |
9f4e41f4 JK |
137 | static void |
138 | truncate_cleanup_page(struct address_space *mapping, struct page *page) | |
1da177e4 | 139 | { |
9f4e41f4 JK |
140 | if (page_mapped(page)) { |
141 | loff_t holelen; | |
142 | ||
143 | holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE; | |
144 | unmap_mapping_range(mapping, | |
145 | (loff_t)page->index << PAGE_SHIFT, | |
146 | holelen, 0); | |
147 | } | |
1da177e4 | 148 | |
266cf658 | 149 | if (page_has_private(page)) |
09cbfeaf | 150 | do_invalidatepage(page, 0, PAGE_SIZE); |
1da177e4 | 151 | |
b9ea2515 KK |
152 | /* |
153 | * Some filesystems seem to re-dirty the page even after | |
154 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
155 | * Hence dirty accounting check is placed after invalidation. | |
156 | */ | |
11f81bec | 157 | cancel_dirty_page(page); |
1da177e4 | 158 | ClearPageMappedToDisk(page); |
1da177e4 LT |
159 | } |
160 | ||
161 | /* | |
fc0ecff6 | 162 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 163 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
164 | * be marked dirty at any time too, so use remove_mapping which safely |
165 | * discards clean, unused pages. | |
1da177e4 LT |
166 | * |
167 | * Returns non-zero if the page was successfully invalidated. | |
168 | */ | |
169 | static int | |
170 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
171 | { | |
0fd0e6b0 NP |
172 | int ret; |
173 | ||
1da177e4 LT |
174 | if (page->mapping != mapping) |
175 | return 0; | |
176 | ||
266cf658 | 177 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
178 | return 0; |
179 | ||
0fd0e6b0 | 180 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
181 | |
182 | return ret; | |
1da177e4 LT |
183 | } |
184 | ||
750b4987 NP |
185 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
186 | { | |
fc127da0 KS |
187 | VM_BUG_ON_PAGE(PageTail(page), page); |
188 | ||
9f4e41f4 JK |
189 | if (page->mapping != mapping) |
190 | return -EIO; | |
191 | ||
192 | truncate_cleanup_page(mapping, page); | |
193 | delete_from_page_cache(page); | |
194 | return 0; | |
750b4987 NP |
195 | } |
196 | ||
25718736 AK |
197 | /* |
198 | * Used to get rid of pages on hardware memory corruption. | |
199 | */ | |
200 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
201 | { | |
202 | if (!mapping) | |
203 | return -EINVAL; | |
204 | /* | |
205 | * Only punch for normal data pages for now. | |
206 | * Handling other types like directories would need more auditing. | |
207 | */ | |
208 | if (!S_ISREG(mapping->host->i_mode)) | |
209 | return -EIO; | |
210 | return truncate_inode_page(mapping, page); | |
211 | } | |
212 | EXPORT_SYMBOL(generic_error_remove_page); | |
213 | ||
83f78668 WF |
214 | /* |
215 | * Safely invalidate one page from its pagecache mapping. | |
216 | * It only drops clean, unused pages. The page must be locked. | |
217 | * | |
218 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
219 | */ | |
220 | int invalidate_inode_page(struct page *page) | |
221 | { | |
222 | struct address_space *mapping = page_mapping(page); | |
223 | if (!mapping) | |
224 | return 0; | |
225 | if (PageDirty(page) || PageWriteback(page)) | |
226 | return 0; | |
227 | if (page_mapped(page)) | |
228 | return 0; | |
229 | return invalidate_complete_page(mapping, page); | |
230 | } | |
231 | ||
1da177e4 | 232 | /** |
73c1e204 | 233 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
234 | * @mapping: mapping to truncate |
235 | * @lstart: offset from which to truncate | |
5a720394 | 236 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 237 | * |
d7339071 | 238 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
239 | * specified offsets (and zeroing out partial pages |
240 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
241 | * |
242 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
243 | * block on page locks and it will not block on writeback. The second pass | |
244 | * will wait. This is to prevent as much IO as possible in the affected region. | |
245 | * The first pass will remove most pages, so the search cost of the second pass | |
246 | * is low. | |
247 | * | |
1da177e4 LT |
248 | * We pass down the cache-hot hint to the page freeing code. Even if the |
249 | * mapping is large, it is probably the case that the final pages are the most | |
250 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
251 | * |
252 | * Note that since ->invalidatepage() accepts range to invalidate | |
253 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
254 | * page aligned properly. | |
1da177e4 | 255 | */ |
d7339071 HR |
256 | void truncate_inode_pages_range(struct address_space *mapping, |
257 | loff_t lstart, loff_t lend) | |
1da177e4 | 258 | { |
5a720394 LC |
259 | pgoff_t start; /* inclusive */ |
260 | pgoff_t end; /* exclusive */ | |
261 | unsigned int partial_start; /* inclusive */ | |
262 | unsigned int partial_end; /* exclusive */ | |
263 | struct pagevec pvec; | |
0cd6144a | 264 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
265 | pgoff_t index; |
266 | int i; | |
1da177e4 | 267 | |
f9fe48be | 268 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
34ccb69e | 269 | goto out; |
1da177e4 | 270 | |
5a720394 | 271 | /* Offsets within partial pages */ |
09cbfeaf KS |
272 | partial_start = lstart & (PAGE_SIZE - 1); |
273 | partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
5a720394 LC |
274 | |
275 | /* | |
276 | * 'start' and 'end' always covers the range of pages to be fully | |
277 | * truncated. Partial pages are covered with 'partial_start' at the | |
278 | * start of the range and 'partial_end' at the end of the range. | |
279 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
280 | */ | |
09cbfeaf | 281 | start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
5a720394 LC |
282 | if (lend == -1) |
283 | /* | |
284 | * lend == -1 indicates end-of-file so we have to set 'end' | |
285 | * to the highest possible pgoff_t and since the type is | |
286 | * unsigned we're using -1. | |
287 | */ | |
288 | end = -1; | |
289 | else | |
09cbfeaf | 290 | end = (lend + 1) >> PAGE_SHIFT; |
d7339071 | 291 | |
1da177e4 | 292 | pagevec_init(&pvec, 0); |
b85e0eff | 293 | index = start; |
0cd6144a JW |
294 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
295 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
296 | indices)) { | |
1da177e4 LT |
297 | for (i = 0; i < pagevec_count(&pvec); i++) { |
298 | struct page *page = pvec.pages[i]; | |
1da177e4 | 299 | |
b85e0eff | 300 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 301 | index = indices[i]; |
5a720394 | 302 | if (index >= end) |
d7339071 | 303 | break; |
d7339071 | 304 | |
0cd6144a | 305 | if (radix_tree_exceptional_entry(page)) { |
c6dcf52c JK |
306 | truncate_exceptional_entry(mapping, index, |
307 | page); | |
0cd6144a JW |
308 | continue; |
309 | } | |
310 | ||
529ae9aa | 311 | if (!trylock_page(page)) |
1da177e4 | 312 | continue; |
5cbc198a | 313 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
314 | if (PageWriteback(page)) { |
315 | unlock_page(page); | |
316 | continue; | |
317 | } | |
750b4987 | 318 | truncate_inode_page(mapping, page); |
1da177e4 LT |
319 | unlock_page(page); |
320 | } | |
0cd6144a | 321 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
322 | pagevec_release(&pvec); |
323 | cond_resched(); | |
b85e0eff | 324 | index++; |
1da177e4 LT |
325 | } |
326 | ||
5a720394 | 327 | if (partial_start) { |
1da177e4 LT |
328 | struct page *page = find_lock_page(mapping, start - 1); |
329 | if (page) { | |
09cbfeaf | 330 | unsigned int top = PAGE_SIZE; |
5a720394 LC |
331 | if (start > end) { |
332 | /* Truncation within a single page */ | |
333 | top = partial_end; | |
334 | partial_end = 0; | |
335 | } | |
1da177e4 | 336 | wait_on_page_writeback(page); |
5a720394 LC |
337 | zero_user_segment(page, partial_start, top); |
338 | cleancache_invalidate_page(mapping, page); | |
339 | if (page_has_private(page)) | |
340 | do_invalidatepage(page, partial_start, | |
341 | top - partial_start); | |
1da177e4 | 342 | unlock_page(page); |
09cbfeaf | 343 | put_page(page); |
1da177e4 LT |
344 | } |
345 | } | |
5a720394 LC |
346 | if (partial_end) { |
347 | struct page *page = find_lock_page(mapping, end); | |
348 | if (page) { | |
349 | wait_on_page_writeback(page); | |
350 | zero_user_segment(page, 0, partial_end); | |
351 | cleancache_invalidate_page(mapping, page); | |
352 | if (page_has_private(page)) | |
353 | do_invalidatepage(page, 0, | |
354 | partial_end); | |
355 | unlock_page(page); | |
09cbfeaf | 356 | put_page(page); |
5a720394 LC |
357 | } |
358 | } | |
359 | /* | |
360 | * If the truncation happened within a single page no pages | |
361 | * will be released, just zeroed, so we can bail out now. | |
362 | */ | |
363 | if (start >= end) | |
34ccb69e | 364 | goto out; |
1da177e4 | 365 | |
b85e0eff | 366 | index = start; |
1da177e4 LT |
367 | for ( ; ; ) { |
368 | cond_resched(); | |
0cd6144a | 369 | if (!pagevec_lookup_entries(&pvec, mapping, index, |
792ceaef HD |
370 | min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { |
371 | /* If all gone from start onwards, we're done */ | |
b85e0eff | 372 | if (index == start) |
1da177e4 | 373 | break; |
792ceaef | 374 | /* Otherwise restart to make sure all gone */ |
b85e0eff | 375 | index = start; |
1da177e4 LT |
376 | continue; |
377 | } | |
0cd6144a | 378 | if (index == start && indices[0] >= end) { |
792ceaef | 379 | /* All gone out of hole to be punched, we're done */ |
0cd6144a | 380 | pagevec_remove_exceptionals(&pvec); |
d7339071 HR |
381 | pagevec_release(&pvec); |
382 | break; | |
383 | } | |
1da177e4 LT |
384 | for (i = 0; i < pagevec_count(&pvec); i++) { |
385 | struct page *page = pvec.pages[i]; | |
386 | ||
b85e0eff | 387 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 388 | index = indices[i]; |
792ceaef HD |
389 | if (index >= end) { |
390 | /* Restart punch to make sure all gone */ | |
391 | index = start - 1; | |
d7339071 | 392 | break; |
792ceaef | 393 | } |
b85e0eff | 394 | |
0cd6144a | 395 | if (radix_tree_exceptional_entry(page)) { |
c6dcf52c JK |
396 | truncate_exceptional_entry(mapping, index, |
397 | page); | |
0cd6144a JW |
398 | continue; |
399 | } | |
400 | ||
1da177e4 | 401 | lock_page(page); |
5cbc198a | 402 | WARN_ON(page_to_index(page) != index); |
1da177e4 | 403 | wait_on_page_writeback(page); |
750b4987 | 404 | truncate_inode_page(mapping, page); |
1da177e4 LT |
405 | unlock_page(page); |
406 | } | |
0cd6144a | 407 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 408 | pagevec_release(&pvec); |
b85e0eff | 409 | index++; |
1da177e4 | 410 | } |
34ccb69e AR |
411 | |
412 | out: | |
3167760f | 413 | cleancache_invalidate_inode(mapping); |
1da177e4 | 414 | } |
d7339071 | 415 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 416 | |
d7339071 HR |
417 | /** |
418 | * truncate_inode_pages - truncate *all* the pages from an offset | |
419 | * @mapping: mapping to truncate | |
420 | * @lstart: offset from which to truncate | |
421 | * | |
1b1dcc1b | 422 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
423 | * |
424 | * Note: When this function returns, there can be a page in the process of | |
425 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
426 | * mapping->nrpages can be non-zero when this function returns even after | |
427 | * truncation of the whole mapping. | |
d7339071 HR |
428 | */ |
429 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
430 | { | |
431 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
432 | } | |
1da177e4 LT |
433 | EXPORT_SYMBOL(truncate_inode_pages); |
434 | ||
91b0abe3 JW |
435 | /** |
436 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
437 | * @mapping: mapping to truncate | |
438 | * | |
439 | * Called under (and serialized by) inode->i_mutex. | |
440 | * | |
441 | * Filesystems have to use this in the .evict_inode path to inform the | |
442 | * VM that this is the final truncate and the inode is going away. | |
443 | */ | |
444 | void truncate_inode_pages_final(struct address_space *mapping) | |
445 | { | |
f9fe48be | 446 | unsigned long nrexceptional; |
91b0abe3 JW |
447 | unsigned long nrpages; |
448 | ||
449 | /* | |
450 | * Page reclaim can not participate in regular inode lifetime | |
451 | * management (can't call iput()) and thus can race with the | |
452 | * inode teardown. Tell it when the address space is exiting, | |
453 | * so that it does not install eviction information after the | |
454 | * final truncate has begun. | |
455 | */ | |
456 | mapping_set_exiting(mapping); | |
457 | ||
458 | /* | |
459 | * When reclaim installs eviction entries, it increases | |
f9fe48be | 460 | * nrexceptional first, then decreases nrpages. Make sure we see |
91b0abe3 JW |
461 | * this in the right order or we might miss an entry. |
462 | */ | |
463 | nrpages = mapping->nrpages; | |
464 | smp_rmb(); | |
f9fe48be | 465 | nrexceptional = mapping->nrexceptional; |
91b0abe3 | 466 | |
f9fe48be | 467 | if (nrpages || nrexceptional) { |
91b0abe3 JW |
468 | /* |
469 | * As truncation uses a lockless tree lookup, cycle | |
470 | * the tree lock to make sure any ongoing tree | |
471 | * modification that does not see AS_EXITING is | |
472 | * completed before starting the final truncate. | |
473 | */ | |
474 | spin_lock_irq(&mapping->tree_lock); | |
475 | spin_unlock_irq(&mapping->tree_lock); | |
476 | ||
477 | truncate_inode_pages(mapping, 0); | |
478 | } | |
479 | } | |
480 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
481 | ||
28697355 MW |
482 | /** |
483 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
484 | * @mapping: the address_space which holds the pages to invalidate | |
485 | * @start: the offset 'from' which to invalidate | |
486 | * @end: the offset 'to' which to invalidate (inclusive) | |
487 | * | |
488 | * This function only removes the unlocked pages, if you want to | |
489 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
490 | * | |
491 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
492 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
493 | * pagetables. | |
494 | */ | |
495 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 496 | pgoff_t start, pgoff_t end) |
1da177e4 | 497 | { |
0cd6144a | 498 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 499 | struct pagevec pvec; |
b85e0eff | 500 | pgoff_t index = start; |
31560180 MK |
501 | unsigned long ret; |
502 | unsigned long count = 0; | |
1da177e4 LT |
503 | int i; |
504 | ||
505 | pagevec_init(&pvec, 0); | |
0cd6144a JW |
506 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
507 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
508 | indices)) { | |
1da177e4 LT |
509 | for (i = 0; i < pagevec_count(&pvec); i++) { |
510 | struct page *page = pvec.pages[i]; | |
e0f23603 | 511 | |
b85e0eff | 512 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 513 | index = indices[i]; |
b85e0eff HD |
514 | if (index > end) |
515 | break; | |
e0f23603 | 516 | |
0cd6144a | 517 | if (radix_tree_exceptional_entry(page)) { |
c6dcf52c JK |
518 | invalidate_exceptional_entry(mapping, index, |
519 | page); | |
0cd6144a JW |
520 | continue; |
521 | } | |
522 | ||
b85e0eff HD |
523 | if (!trylock_page(page)) |
524 | continue; | |
fc127da0 | 525 | |
5cbc198a | 526 | WARN_ON(page_to_index(page) != index); |
fc127da0 KS |
527 | |
528 | /* Middle of THP: skip */ | |
529 | if (PageTransTail(page)) { | |
530 | unlock_page(page); | |
531 | continue; | |
532 | } else if (PageTransHuge(page)) { | |
533 | index += HPAGE_PMD_NR - 1; | |
534 | i += HPAGE_PMD_NR - 1; | |
76b6f9b7 JK |
535 | /* |
536 | * 'end' is in the middle of THP. Don't | |
537 | * invalidate the page as the part outside of | |
538 | * 'end' could be still useful. | |
539 | */ | |
540 | if (index > end) { | |
541 | unlock_page(page); | |
fc127da0 | 542 | continue; |
76b6f9b7 | 543 | } |
fc127da0 KS |
544 | } |
545 | ||
31560180 | 546 | ret = invalidate_inode_page(page); |
1da177e4 | 547 | unlock_page(page); |
31560180 MK |
548 | /* |
549 | * Invalidation is a hint that the page is no longer | |
550 | * of interest and try to speed up its reclaim. | |
551 | */ | |
552 | if (!ret) | |
cc5993bd | 553 | deactivate_file_page(page); |
31560180 | 554 | count += ret; |
1da177e4 | 555 | } |
0cd6144a | 556 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 557 | pagevec_release(&pvec); |
28697355 | 558 | cond_resched(); |
b85e0eff | 559 | index++; |
1da177e4 | 560 | } |
31560180 | 561 | return count; |
1da177e4 | 562 | } |
54bc4855 | 563 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 564 | |
bd4c8ce4 AM |
565 | /* |
566 | * This is like invalidate_complete_page(), except it ignores the page's | |
567 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
568 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
569 | * shrink_page_list() has a temp ref on them, or because they're transiently |
570 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
571 | */ |
572 | static int | |
573 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
574 | { | |
c4843a75 GT |
575 | unsigned long flags; |
576 | ||
bd4c8ce4 AM |
577 | if (page->mapping != mapping) |
578 | return 0; | |
579 | ||
266cf658 | 580 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
581 | return 0; |
582 | ||
c4843a75 | 583 | spin_lock_irqsave(&mapping->tree_lock, flags); |
bd4c8ce4 AM |
584 | if (PageDirty(page)) |
585 | goto failed; | |
586 | ||
266cf658 | 587 | BUG_ON(page_has_private(page)); |
62cccb8c | 588 | __delete_from_page_cache(page, NULL); |
c4843a75 | 589 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6072d13c LT |
590 | |
591 | if (mapping->a_ops->freepage) | |
592 | mapping->a_ops->freepage(page); | |
593 | ||
09cbfeaf | 594 | put_page(page); /* pagecache ref */ |
bd4c8ce4 AM |
595 | return 1; |
596 | failed: | |
c4843a75 | 597 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
bd4c8ce4 AM |
598 | return 0; |
599 | } | |
600 | ||
e3db7691 TM |
601 | static int do_launder_page(struct address_space *mapping, struct page *page) |
602 | { | |
603 | if (!PageDirty(page)) | |
604 | return 0; | |
605 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
606 | return 0; | |
607 | return mapping->a_ops->launder_page(page); | |
608 | } | |
609 | ||
1da177e4 LT |
610 | /** |
611 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 612 | * @mapping: the address_space |
1da177e4 LT |
613 | * @start: the page offset 'from' which to invalidate |
614 | * @end: the page offset 'to' which to invalidate (inclusive) | |
615 | * | |
616 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
617 | * invalidation. | |
618 | * | |
6ccfa806 | 619 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
620 | */ |
621 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
622 | pgoff_t start, pgoff_t end) | |
623 | { | |
0cd6144a | 624 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 625 | struct pagevec pvec; |
b85e0eff | 626 | pgoff_t index; |
1da177e4 LT |
627 | int i; |
628 | int ret = 0; | |
0dd1334f | 629 | int ret2 = 0; |
1da177e4 | 630 | int did_range_unmap = 0; |
1da177e4 | 631 | |
32691f0f | 632 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
34ccb69e | 633 | goto out; |
32691f0f | 634 | |
1da177e4 | 635 | pagevec_init(&pvec, 0); |
b85e0eff | 636 | index = start; |
0cd6144a JW |
637 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
638 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
639 | indices)) { | |
7b965e08 | 640 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 641 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
642 | |
643 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 644 | index = indices[i]; |
b85e0eff HD |
645 | if (index > end) |
646 | break; | |
1da177e4 | 647 | |
0cd6144a | 648 | if (radix_tree_exceptional_entry(page)) { |
c6dcf52c JK |
649 | if (!invalidate_exceptional_entry2(mapping, |
650 | index, page)) | |
651 | ret = -EBUSY; | |
0cd6144a JW |
652 | continue; |
653 | } | |
654 | ||
1da177e4 | 655 | lock_page(page); |
5cbc198a | 656 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
657 | if (page->mapping != mapping) { |
658 | unlock_page(page); | |
659 | continue; | |
660 | } | |
1da177e4 | 661 | wait_on_page_writeback(page); |
d00806b1 | 662 | if (page_mapped(page)) { |
1da177e4 LT |
663 | if (!did_range_unmap) { |
664 | /* | |
665 | * Zap the rest of the file in one hit. | |
666 | */ | |
667 | unmap_mapping_range(mapping, | |
09cbfeaf | 668 | (loff_t)index << PAGE_SHIFT, |
b85e0eff | 669 | (loff_t)(1 + end - index) |
09cbfeaf KS |
670 | << PAGE_SHIFT, |
671 | 0); | |
1da177e4 LT |
672 | did_range_unmap = 1; |
673 | } else { | |
674 | /* | |
675 | * Just zap this page | |
676 | */ | |
677 | unmap_mapping_range(mapping, | |
09cbfeaf KS |
678 | (loff_t)index << PAGE_SHIFT, |
679 | PAGE_SIZE, 0); | |
1da177e4 LT |
680 | } |
681 | } | |
d00806b1 | 682 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
683 | ret2 = do_launder_page(mapping, page); |
684 | if (ret2 == 0) { | |
685 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 686 | ret2 = -EBUSY; |
0dd1334f HH |
687 | } |
688 | if (ret2 < 0) | |
689 | ret = ret2; | |
1da177e4 LT |
690 | unlock_page(page); |
691 | } | |
0cd6144a | 692 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
693 | pagevec_release(&pvec); |
694 | cond_resched(); | |
b85e0eff | 695 | index++; |
1da177e4 | 696 | } |
cd656375 JK |
697 | /* |
698 | * For DAX we invalidate page tables after invalidating radix tree. We | |
699 | * could invalidate page tables while invalidating each entry however | |
700 | * that would be expensive. And doing range unmapping before doesn't | |
701 | * work as we have no cheap way to find whether radix tree entry didn't | |
702 | * get remapped later. | |
703 | */ | |
704 | if (dax_mapping(mapping)) { | |
705 | unmap_mapping_range(mapping, (loff_t)start << PAGE_SHIFT, | |
706 | (loff_t)(end - start + 1) << PAGE_SHIFT, 0); | |
707 | } | |
34ccb69e | 708 | out: |
3167760f | 709 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
710 | return ret; |
711 | } | |
712 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
713 | ||
714 | /** | |
715 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 716 | * @mapping: the address_space |
1da177e4 LT |
717 | * |
718 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
719 | * invalidation. | |
720 | * | |
e9de25dd | 721 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
722 | */ |
723 | int invalidate_inode_pages2(struct address_space *mapping) | |
724 | { | |
725 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
726 | } | |
727 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 728 | |
729 | /** | |
730 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
731 | * @inode: inode | |
8a549bea | 732 | * @newsize: new file size |
25d9e2d1 | 733 | * |
734 | * inode's new i_size must already be written before truncate_pagecache | |
735 | * is called. | |
736 | * | |
737 | * This function should typically be called before the filesystem | |
738 | * releases resources associated with the freed range (eg. deallocates | |
739 | * blocks). This way, pagecache will always stay logically coherent | |
740 | * with on-disk format, and the filesystem would not have to deal with | |
741 | * situations such as writepage being called for a page that has already | |
742 | * had its underlying blocks deallocated. | |
743 | */ | |
7caef267 | 744 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 745 | { |
cedabed4 | 746 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 747 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
748 | |
749 | /* | |
750 | * unmap_mapping_range is called twice, first simply for | |
751 | * efficiency so that truncate_inode_pages does fewer | |
752 | * single-page unmaps. However after this first call, and | |
753 | * before truncate_inode_pages finishes, it is possible for | |
754 | * private pages to be COWed, which remain after | |
755 | * truncate_inode_pages finishes, hence the second | |
756 | * unmap_mapping_range call must be made for correctness. | |
757 | */ | |
8a549bea HD |
758 | unmap_mapping_range(mapping, holebegin, 0, 1); |
759 | truncate_inode_pages(mapping, newsize); | |
760 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 761 | } |
762 | EXPORT_SYMBOL(truncate_pagecache); | |
763 | ||
2c27c65e CH |
764 | /** |
765 | * truncate_setsize - update inode and pagecache for a new file size | |
766 | * @inode: inode | |
767 | * @newsize: new file size | |
768 | * | |
382e27da JK |
769 | * truncate_setsize updates i_size and performs pagecache truncation (if |
770 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
771 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 772 | * |
77783d06 JK |
773 | * Must be called with a lock serializing truncates and writes (generally |
774 | * i_mutex but e.g. xfs uses a different lock) and before all filesystem | |
775 | * specific block truncation has been performed. | |
2c27c65e CH |
776 | */ |
777 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
778 | { | |
90a80202 JK |
779 | loff_t oldsize = inode->i_size; |
780 | ||
2c27c65e | 781 | i_size_write(inode, newsize); |
90a80202 JK |
782 | if (newsize > oldsize) |
783 | pagecache_isize_extended(inode, oldsize, newsize); | |
7caef267 | 784 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
785 | } |
786 | EXPORT_SYMBOL(truncate_setsize); | |
787 | ||
90a80202 JK |
788 | /** |
789 | * pagecache_isize_extended - update pagecache after extension of i_size | |
790 | * @inode: inode for which i_size was extended | |
791 | * @from: original inode size | |
792 | * @to: new inode size | |
793 | * | |
794 | * Handle extension of inode size either caused by extending truncate or by | |
795 | * write starting after current i_size. We mark the page straddling current | |
796 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
797 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
798 | * the page before user writes to the page via mmap after the i_size has been | |
799 | * changed. | |
800 | * | |
801 | * The function must be called after i_size is updated so that page fault | |
802 | * coming after we unlock the page will already see the new i_size. | |
803 | * The function must be called while we still hold i_mutex - this not only | |
804 | * makes sure i_size is stable but also that userspace cannot observe new | |
805 | * i_size value before we are prepared to store mmap writes at new inode size. | |
806 | */ | |
807 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
808 | { | |
93407472 | 809 | int bsize = i_blocksize(inode); |
90a80202 JK |
810 | loff_t rounded_from; |
811 | struct page *page; | |
812 | pgoff_t index; | |
813 | ||
90a80202 JK |
814 | WARN_ON(to > inode->i_size); |
815 | ||
09cbfeaf | 816 | if (from >= to || bsize == PAGE_SIZE) |
90a80202 JK |
817 | return; |
818 | /* Page straddling @from will not have any hole block created? */ | |
819 | rounded_from = round_up(from, bsize); | |
09cbfeaf | 820 | if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) |
90a80202 JK |
821 | return; |
822 | ||
09cbfeaf | 823 | index = from >> PAGE_SHIFT; |
90a80202 JK |
824 | page = find_lock_page(inode->i_mapping, index); |
825 | /* Page not cached? Nothing to do */ | |
826 | if (!page) | |
827 | return; | |
828 | /* | |
829 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
830 | * is needed. | |
831 | */ | |
832 | if (page_mkclean(page)) | |
833 | set_page_dirty(page); | |
834 | unlock_page(page); | |
09cbfeaf | 835 | put_page(page); |
90a80202 JK |
836 | } |
837 | EXPORT_SYMBOL(pagecache_isize_extended); | |
838 | ||
623e3db9 HD |
839 | /** |
840 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
841 | * @inode: inode | |
842 | * @lstart: offset of beginning of hole | |
843 | * @lend: offset of last byte of hole | |
844 | * | |
845 | * This function should typically be called before the filesystem | |
846 | * releases resources associated with the freed range (eg. deallocates | |
847 | * blocks). This way, pagecache will always stay logically coherent | |
848 | * with on-disk format, and the filesystem would not have to deal with | |
849 | * situations such as writepage being called for a page that has already | |
850 | * had its underlying blocks deallocated. | |
851 | */ | |
852 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
853 | { | |
854 | struct address_space *mapping = inode->i_mapping; | |
855 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
856 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
857 | /* | |
858 | * This rounding is currently just for example: unmap_mapping_range | |
859 | * expands its hole outwards, whereas we want it to contract the hole | |
860 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
861 | * doing their own page rounding first. Note that unmap_mapping_range |
862 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
863 | */ |
864 | ||
865 | /* | |
866 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
867 | * once (before truncating pagecache), and without "even_cows" flag: | |
868 | * hole-punching should not remove private COWed pages from the hole. | |
869 | */ | |
870 | if ((u64)unmap_end > (u64)unmap_start) | |
871 | unmap_mapping_range(mapping, unmap_start, | |
872 | 1 + unmap_end - unmap_start, 0); | |
873 | truncate_inode_pages_range(mapping, lstart, lend); | |
874 | } | |
875 | EXPORT_SYMBOL(truncate_pagecache_range); |