Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * mm/truncate.c - code for taking down pages from address_spaces | |
4 | * | |
5 | * Copyright (C) 2002, Linus Torvalds | |
6 | * | |
e1f8e874 | 7 | * 10Sep2002 Andrew Morton |
1da177e4 LT |
8 | * Initial version. |
9 | */ | |
10 | ||
11 | #include <linux/kernel.h> | |
4af3c9cc | 12 | #include <linux/backing-dev.h> |
f9fe48be | 13 | #include <linux/dax.h> |
5a0e3ad6 | 14 | #include <linux/gfp.h> |
1da177e4 | 15 | #include <linux/mm.h> |
0fd0e6b0 | 16 | #include <linux/swap.h> |
b95f1b31 | 17 | #include <linux/export.h> |
1da177e4 | 18 | #include <linux/pagemap.h> |
01f2705d | 19 | #include <linux/highmem.h> |
1da177e4 | 20 | #include <linux/pagevec.h> |
e08748ce | 21 | #include <linux/task_io_accounting_ops.h> |
1da177e4 | 22 | #include <linux/buffer_head.h> /* grr. try_to_release_page, |
aaa4059b | 23 | do_invalidatepage */ |
3a4f8a0b | 24 | #include <linux/shmem_fs.h> |
c515e1fd | 25 | #include <linux/cleancache.h> |
90a80202 | 26 | #include <linux/rmap.h> |
ba470de4 | 27 | #include "internal.h" |
1da177e4 | 28 | |
f2187599 MG |
29 | /* |
30 | * Regular page slots are stabilized by the page lock even without the tree | |
31 | * itself locked. These unlocked entries need verification under the tree | |
32 | * lock. | |
33 | */ | |
34 | static inline void __clear_shadow_entry(struct address_space *mapping, | |
35 | pgoff_t index, void *entry) | |
0cd6144a | 36 | { |
69b6c131 | 37 | XA_STATE(xas, &mapping->i_pages, index); |
449dd698 | 38 | |
69b6c131 MW |
39 | xas_set_update(&xas, workingset_update_node); |
40 | if (xas_load(&xas) != entry) | |
f2187599 | 41 | return; |
69b6c131 | 42 | xas_store(&xas, NULL); |
f2187599 MG |
43 | } |
44 | ||
45 | static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, | |
46 | void *entry) | |
47 | { | |
51b8c1fe | 48 | spin_lock(&mapping->host->i_lock); |
b93b0163 | 49 | xa_lock_irq(&mapping->i_pages); |
f2187599 | 50 | __clear_shadow_entry(mapping, index, entry); |
b93b0163 | 51 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
52 | if (mapping_shrinkable(mapping)) |
53 | inode_add_lru(mapping->host); | |
54 | spin_unlock(&mapping->host->i_lock); | |
0cd6144a | 55 | } |
1da177e4 | 56 | |
c6dcf52c | 57 | /* |
f2187599 MG |
58 | * Unconditionally remove exceptional entries. Usually called from truncate |
59 | * path. Note that the pagevec may be altered by this function by removing | |
60 | * exceptional entries similar to what pagevec_remove_exceptionals does. | |
c6dcf52c | 61 | */ |
f2187599 | 62 | static void truncate_exceptional_pvec_entries(struct address_space *mapping, |
31d270fd | 63 | struct pagevec *pvec, pgoff_t *indices) |
c6dcf52c | 64 | { |
f2187599 | 65 | int i, j; |
31d270fd | 66 | bool dax; |
f2187599 | 67 | |
c6dcf52c JK |
68 | /* Handled by shmem itself */ |
69 | if (shmem_mapping(mapping)) | |
70 | return; | |
71 | ||
f2187599 | 72 | for (j = 0; j < pagevec_count(pvec); j++) |
3159f943 | 73 | if (xa_is_value(pvec->pages[j])) |
f2187599 MG |
74 | break; |
75 | ||
76 | if (j == pagevec_count(pvec)) | |
c6dcf52c | 77 | return; |
f2187599 MG |
78 | |
79 | dax = dax_mapping(mapping); | |
51b8c1fe JW |
80 | if (!dax) { |
81 | spin_lock(&mapping->host->i_lock); | |
b93b0163 | 82 | xa_lock_irq(&mapping->i_pages); |
51b8c1fe | 83 | } |
f2187599 MG |
84 | |
85 | for (i = j; i < pagevec_count(pvec); i++) { | |
86 | struct page *page = pvec->pages[i]; | |
87 | pgoff_t index = indices[i]; | |
88 | ||
3159f943 | 89 | if (!xa_is_value(page)) { |
f2187599 MG |
90 | pvec->pages[j++] = page; |
91 | continue; | |
92 | } | |
93 | ||
f2187599 MG |
94 | if (unlikely(dax)) { |
95 | dax_delete_mapping_entry(mapping, index); | |
96 | continue; | |
97 | } | |
98 | ||
99 | __clear_shadow_entry(mapping, index, page); | |
c6dcf52c | 100 | } |
f2187599 | 101 | |
51b8c1fe | 102 | if (!dax) { |
b93b0163 | 103 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
104 | if (mapping_shrinkable(mapping)) |
105 | inode_add_lru(mapping->host); | |
106 | spin_unlock(&mapping->host->i_lock); | |
107 | } | |
f2187599 | 108 | pvec->nr = j; |
c6dcf52c JK |
109 | } |
110 | ||
111 | /* | |
112 | * Invalidate exceptional entry if easily possible. This handles exceptional | |
4636e70b | 113 | * entries for invalidate_inode_pages(). |
c6dcf52c JK |
114 | */ |
115 | static int invalidate_exceptional_entry(struct address_space *mapping, | |
116 | pgoff_t index, void *entry) | |
117 | { | |
4636e70b RZ |
118 | /* Handled by shmem itself, or for DAX we do nothing. */ |
119 | if (shmem_mapping(mapping) || dax_mapping(mapping)) | |
c6dcf52c | 120 | return 1; |
c6dcf52c JK |
121 | clear_shadow_entry(mapping, index, entry); |
122 | return 1; | |
123 | } | |
124 | ||
125 | /* | |
126 | * Invalidate exceptional entry if clean. This handles exceptional entries for | |
127 | * invalidate_inode_pages2() so for DAX it evicts only clean entries. | |
128 | */ | |
129 | static int invalidate_exceptional_entry2(struct address_space *mapping, | |
130 | pgoff_t index, void *entry) | |
131 | { | |
132 | /* Handled by shmem itself */ | |
133 | if (shmem_mapping(mapping)) | |
134 | return 1; | |
135 | if (dax_mapping(mapping)) | |
136 | return dax_invalidate_mapping_entry_sync(mapping, index); | |
137 | clear_shadow_entry(mapping, index, entry); | |
138 | return 1; | |
139 | } | |
140 | ||
cf9a2ae8 | 141 | /** |
28bc44d7 | 142 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 143 | * @page: the page which is affected |
d47992f8 LC |
144 | * @offset: start of the range to invalidate |
145 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
146 | * |
147 | * do_invalidatepage() is called when all or part of the page has become | |
148 | * invalidated by a truncate operation. | |
149 | * | |
150 | * do_invalidatepage() does not have to release all buffers, but it must | |
151 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
152 | * is underway against any of the blocks which are outside the truncation | |
153 | * point. Because the caller is about to free (and possibly reuse) those | |
154 | * blocks on-disk. | |
155 | */ | |
d47992f8 LC |
156 | void do_invalidatepage(struct page *page, unsigned int offset, |
157 | unsigned int length) | |
cf9a2ae8 | 158 | { |
d47992f8 LC |
159 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
160 | ||
cf9a2ae8 | 161 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 162 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
163 | if (!invalidatepage) |
164 | invalidatepage = block_invalidatepage; | |
9361401e | 165 | #endif |
cf9a2ae8 | 166 | if (invalidatepage) |
d47992f8 | 167 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
168 | } |
169 | ||
1da177e4 LT |
170 | /* |
171 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 172 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 173 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 | 174 | * |
fc3a5ac5 | 175 | * We need to bail out if page->mapping is no longer equal to the original |
1da177e4 | 176 | * mapping. This happens a) when the VM reclaimed the page while we waited on |
fc0ecff6 | 177 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
178 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
179 | */ | |
efe99bba | 180 | static void truncate_cleanup_folio(struct folio *folio) |
1da177e4 | 181 | { |
efe99bba MWO |
182 | if (folio_mapped(folio)) |
183 | unmap_mapping_page(&folio->page); | |
1da177e4 | 184 | |
efe99bba MWO |
185 | if (folio_has_private(folio)) |
186 | do_invalidatepage(&folio->page, 0, folio_size(folio)); | |
1da177e4 | 187 | |
b9ea2515 KK |
188 | /* |
189 | * Some filesystems seem to re-dirty the page even after | |
190 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
191 | * Hence dirty accounting check is placed after invalidation. | |
192 | */ | |
efe99bba MWO |
193 | folio_cancel_dirty(folio); |
194 | folio_clear_mappedtodisk(folio); | |
1da177e4 LT |
195 | } |
196 | ||
197 | /* | |
fc0ecff6 | 198 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 199 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
200 | * be marked dirty at any time too, so use remove_mapping which safely |
201 | * discards clean, unused pages. | |
1da177e4 LT |
202 | * |
203 | * Returns non-zero if the page was successfully invalidated. | |
204 | */ | |
205 | static int | |
206 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
207 | { | |
0fd0e6b0 NP |
208 | int ret; |
209 | ||
1da177e4 LT |
210 | if (page->mapping != mapping) |
211 | return 0; | |
212 | ||
266cf658 | 213 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
214 | return 0; |
215 | ||
0fd0e6b0 | 216 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
217 | |
218 | return ret; | |
1da177e4 LT |
219 | } |
220 | ||
750b4987 NP |
221 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
222 | { | |
efe99bba | 223 | struct folio *folio = page_folio(page); |
fc127da0 KS |
224 | VM_BUG_ON_PAGE(PageTail(page), page); |
225 | ||
9f4e41f4 JK |
226 | if (page->mapping != mapping) |
227 | return -EIO; | |
228 | ||
efe99bba MWO |
229 | truncate_cleanup_folio(folio); |
230 | filemap_remove_folio(folio); | |
9f4e41f4 | 231 | return 0; |
750b4987 NP |
232 | } |
233 | ||
25718736 AK |
234 | /* |
235 | * Used to get rid of pages on hardware memory corruption. | |
236 | */ | |
237 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
238 | { | |
239 | if (!mapping) | |
240 | return -EINVAL; | |
241 | /* | |
242 | * Only punch for normal data pages for now. | |
243 | * Handling other types like directories would need more auditing. | |
244 | */ | |
245 | if (!S_ISREG(mapping->host->i_mode)) | |
246 | return -EIO; | |
247 | return truncate_inode_page(mapping, page); | |
248 | } | |
249 | EXPORT_SYMBOL(generic_error_remove_page); | |
250 | ||
83f78668 WF |
251 | /* |
252 | * Safely invalidate one page from its pagecache mapping. | |
253 | * It only drops clean, unused pages. The page must be locked. | |
254 | * | |
255 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
256 | */ | |
257 | int invalidate_inode_page(struct page *page) | |
258 | { | |
259 | struct address_space *mapping = page_mapping(page); | |
260 | if (!mapping) | |
261 | return 0; | |
262 | if (PageDirty(page) || PageWriteback(page)) | |
263 | return 0; | |
264 | if (page_mapped(page)) | |
265 | return 0; | |
266 | return invalidate_complete_page(mapping, page); | |
267 | } | |
268 | ||
1da177e4 | 269 | /** |
73c1e204 | 270 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
271 | * @mapping: mapping to truncate |
272 | * @lstart: offset from which to truncate | |
5a720394 | 273 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 274 | * |
d7339071 | 275 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
276 | * specified offsets (and zeroing out partial pages |
277 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
278 | * |
279 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
280 | * block on page locks and it will not block on writeback. The second pass | |
281 | * will wait. This is to prevent as much IO as possible in the affected region. | |
282 | * The first pass will remove most pages, so the search cost of the second pass | |
283 | * is low. | |
284 | * | |
1da177e4 LT |
285 | * We pass down the cache-hot hint to the page freeing code. Even if the |
286 | * mapping is large, it is probably the case that the final pages are the most | |
287 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
288 | * |
289 | * Note that since ->invalidatepage() accepts range to invalidate | |
290 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
291 | * page aligned properly. | |
1da177e4 | 292 | */ |
d7339071 HR |
293 | void truncate_inode_pages_range(struct address_space *mapping, |
294 | loff_t lstart, loff_t lend) | |
1da177e4 | 295 | { |
5a720394 LC |
296 | pgoff_t start; /* inclusive */ |
297 | pgoff_t end; /* exclusive */ | |
298 | unsigned int partial_start; /* inclusive */ | |
299 | unsigned int partial_end; /* exclusive */ | |
300 | struct pagevec pvec; | |
0cd6144a | 301 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
302 | pgoff_t index; |
303 | int i; | |
1da177e4 | 304 | |
7716506a | 305 | if (mapping_empty(mapping)) |
34ccb69e | 306 | goto out; |
1da177e4 | 307 | |
5a720394 | 308 | /* Offsets within partial pages */ |
09cbfeaf KS |
309 | partial_start = lstart & (PAGE_SIZE - 1); |
310 | partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
5a720394 LC |
311 | |
312 | /* | |
313 | * 'start' and 'end' always covers the range of pages to be fully | |
314 | * truncated. Partial pages are covered with 'partial_start' at the | |
315 | * start of the range and 'partial_end' at the end of the range. | |
316 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
317 | */ | |
09cbfeaf | 318 | start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
5a720394 LC |
319 | if (lend == -1) |
320 | /* | |
321 | * lend == -1 indicates end-of-file so we have to set 'end' | |
322 | * to the highest possible pgoff_t and since the type is | |
323 | * unsigned we're using -1. | |
324 | */ | |
325 | end = -1; | |
326 | else | |
09cbfeaf | 327 | end = (lend + 1) >> PAGE_SHIFT; |
d7339071 | 328 | |
86679820 | 329 | pagevec_init(&pvec); |
b85e0eff | 330 | index = start; |
5c211ba2 MWO |
331 | while (index < end && find_lock_entries(mapping, index, end - 1, |
332 | &pvec, indices)) { | |
333 | index = indices[pagevec_count(&pvec) - 1] + 1; | |
31d270fd | 334 | truncate_exceptional_pvec_entries(mapping, &pvec, indices); |
5c211ba2 | 335 | for (i = 0; i < pagevec_count(&pvec); i++) |
efe99bba | 336 | truncate_cleanup_folio(page_folio(pvec.pages[i])); |
5c211ba2 MWO |
337 | delete_from_page_cache_batch(mapping, &pvec); |
338 | for (i = 0; i < pagevec_count(&pvec); i++) | |
339 | unlock_page(pvec.pages[i]); | |
1da177e4 LT |
340 | pagevec_release(&pvec); |
341 | cond_resched(); | |
342 | } | |
5c211ba2 | 343 | |
5a720394 | 344 | if (partial_start) { |
1da177e4 LT |
345 | struct page *page = find_lock_page(mapping, start - 1); |
346 | if (page) { | |
09cbfeaf | 347 | unsigned int top = PAGE_SIZE; |
5a720394 LC |
348 | if (start > end) { |
349 | /* Truncation within a single page */ | |
350 | top = partial_end; | |
351 | partial_end = 0; | |
352 | } | |
1da177e4 | 353 | wait_on_page_writeback(page); |
5a720394 LC |
354 | zero_user_segment(page, partial_start, top); |
355 | cleancache_invalidate_page(mapping, page); | |
356 | if (page_has_private(page)) | |
357 | do_invalidatepage(page, partial_start, | |
358 | top - partial_start); | |
1da177e4 | 359 | unlock_page(page); |
09cbfeaf | 360 | put_page(page); |
1da177e4 LT |
361 | } |
362 | } | |
5a720394 LC |
363 | if (partial_end) { |
364 | struct page *page = find_lock_page(mapping, end); | |
365 | if (page) { | |
366 | wait_on_page_writeback(page); | |
367 | zero_user_segment(page, 0, partial_end); | |
368 | cleancache_invalidate_page(mapping, page); | |
369 | if (page_has_private(page)) | |
370 | do_invalidatepage(page, 0, | |
371 | partial_end); | |
372 | unlock_page(page); | |
09cbfeaf | 373 | put_page(page); |
5a720394 LC |
374 | } |
375 | } | |
376 | /* | |
377 | * If the truncation happened within a single page no pages | |
378 | * will be released, just zeroed, so we can bail out now. | |
379 | */ | |
380 | if (start >= end) | |
34ccb69e | 381 | goto out; |
1da177e4 | 382 | |
b85e0eff | 383 | index = start; |
1da177e4 LT |
384 | for ( ; ; ) { |
385 | cond_resched(); | |
a656a202 | 386 | if (!find_get_entries(mapping, index, end - 1, &pvec, |
38cefeb3 | 387 | indices)) { |
792ceaef | 388 | /* If all gone from start onwards, we're done */ |
b85e0eff | 389 | if (index == start) |
1da177e4 | 390 | break; |
792ceaef | 391 | /* Otherwise restart to make sure all gone */ |
b85e0eff | 392 | index = start; |
1da177e4 LT |
393 | continue; |
394 | } | |
f2187599 | 395 | |
1da177e4 LT |
396 | for (i = 0; i < pagevec_count(&pvec); i++) { |
397 | struct page *page = pvec.pages[i]; | |
398 | ||
b85e0eff | 399 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 400 | index = indices[i]; |
b85e0eff | 401 | |
3159f943 | 402 | if (xa_is_value(page)) |
0cd6144a | 403 | continue; |
0cd6144a | 404 | |
1da177e4 | 405 | lock_page(page); |
5cbc198a | 406 | WARN_ON(page_to_index(page) != index); |
1da177e4 | 407 | wait_on_page_writeback(page); |
750b4987 | 408 | truncate_inode_page(mapping, page); |
1da177e4 LT |
409 | unlock_page(page); |
410 | } | |
31d270fd | 411 | truncate_exceptional_pvec_entries(mapping, &pvec, indices); |
1da177e4 | 412 | pagevec_release(&pvec); |
b85e0eff | 413 | index++; |
1da177e4 | 414 | } |
34ccb69e AR |
415 | |
416 | out: | |
3167760f | 417 | cleancache_invalidate_inode(mapping); |
1da177e4 | 418 | } |
d7339071 | 419 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 420 | |
d7339071 HR |
421 | /** |
422 | * truncate_inode_pages - truncate *all* the pages from an offset | |
423 | * @mapping: mapping to truncate | |
424 | * @lstart: offset from which to truncate | |
425 | * | |
730633f0 JK |
426 | * Called under (and serialised by) inode->i_rwsem and |
427 | * mapping->invalidate_lock. | |
08142579 JK |
428 | * |
429 | * Note: When this function returns, there can be a page in the process of | |
430 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
431 | * mapping->nrpages can be non-zero when this function returns even after | |
432 | * truncation of the whole mapping. | |
d7339071 HR |
433 | */ |
434 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
435 | { | |
436 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
437 | } | |
1da177e4 LT |
438 | EXPORT_SYMBOL(truncate_inode_pages); |
439 | ||
91b0abe3 JW |
440 | /** |
441 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
442 | * @mapping: mapping to truncate | |
443 | * | |
9608703e | 444 | * Called under (and serialized by) inode->i_rwsem. |
91b0abe3 JW |
445 | * |
446 | * Filesystems have to use this in the .evict_inode path to inform the | |
447 | * VM that this is the final truncate and the inode is going away. | |
448 | */ | |
449 | void truncate_inode_pages_final(struct address_space *mapping) | |
450 | { | |
91b0abe3 JW |
451 | /* |
452 | * Page reclaim can not participate in regular inode lifetime | |
453 | * management (can't call iput()) and thus can race with the | |
454 | * inode teardown. Tell it when the address space is exiting, | |
455 | * so that it does not install eviction information after the | |
456 | * final truncate has begun. | |
457 | */ | |
458 | mapping_set_exiting(mapping); | |
459 | ||
7716506a | 460 | if (!mapping_empty(mapping)) { |
91b0abe3 JW |
461 | /* |
462 | * As truncation uses a lockless tree lookup, cycle | |
463 | * the tree lock to make sure any ongoing tree | |
464 | * modification that does not see AS_EXITING is | |
465 | * completed before starting the final truncate. | |
466 | */ | |
b93b0163 MW |
467 | xa_lock_irq(&mapping->i_pages); |
468 | xa_unlock_irq(&mapping->i_pages); | |
91b0abe3 | 469 | } |
6ff38bd4 PT |
470 | |
471 | /* | |
472 | * Cleancache needs notification even if there are no pages or shadow | |
473 | * entries. | |
474 | */ | |
475 | truncate_inode_pages(mapping, 0); | |
91b0abe3 JW |
476 | } |
477 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
478 | ||
a77eedbc | 479 | static unsigned long __invalidate_mapping_pages(struct address_space *mapping, |
eb1d7a65 | 480 | pgoff_t start, pgoff_t end, unsigned long *nr_pagevec) |
1da177e4 | 481 | { |
0cd6144a | 482 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 483 | struct pagevec pvec; |
b85e0eff | 484 | pgoff_t index = start; |
31560180 MK |
485 | unsigned long ret; |
486 | unsigned long count = 0; | |
1da177e4 LT |
487 | int i; |
488 | ||
86679820 | 489 | pagevec_init(&pvec); |
5c211ba2 | 490 | while (find_lock_entries(mapping, index, end, &pvec, indices)) { |
1da177e4 LT |
491 | for (i = 0; i < pagevec_count(&pvec); i++) { |
492 | struct page *page = pvec.pages[i]; | |
e0f23603 | 493 | |
b85e0eff | 494 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 495 | index = indices[i]; |
e0f23603 | 496 | |
3159f943 | 497 | if (xa_is_value(page)) { |
7ae12c80 JW |
498 | count += invalidate_exceptional_entry(mapping, |
499 | index, | |
500 | page); | |
0cd6144a JW |
501 | continue; |
502 | } | |
5c211ba2 | 503 | index += thp_nr_pages(page) - 1; |
fc127da0 | 504 | |
31560180 | 505 | ret = invalidate_inode_page(page); |
1da177e4 | 506 | unlock_page(page); |
31560180 MK |
507 | /* |
508 | * Invalidation is a hint that the page is no longer | |
509 | * of interest and try to speed up its reclaim. | |
510 | */ | |
eb1d7a65 | 511 | if (!ret) { |
cc5993bd | 512 | deactivate_file_page(page); |
eb1d7a65 YS |
513 | /* It is likely on the pagevec of a remote CPU */ |
514 | if (nr_pagevec) | |
515 | (*nr_pagevec)++; | |
516 | } | |
31560180 | 517 | count += ret; |
1da177e4 | 518 | } |
0cd6144a | 519 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 520 | pagevec_release(&pvec); |
28697355 | 521 | cond_resched(); |
b85e0eff | 522 | index++; |
1da177e4 | 523 | } |
31560180 | 524 | return count; |
1da177e4 | 525 | } |
eb1d7a65 YS |
526 | |
527 | /** | |
7ae12c80 JW |
528 | * invalidate_mapping_pages - Invalidate all clean, unlocked cache of one inode |
529 | * @mapping: the address_space which holds the cache to invalidate | |
eb1d7a65 YS |
530 | * @start: the offset 'from' which to invalidate |
531 | * @end: the offset 'to' which to invalidate (inclusive) | |
532 | * | |
7ae12c80 JW |
533 | * This function removes pages that are clean, unmapped and unlocked, |
534 | * as well as shadow entries. It will not block on IO activity. | |
eb1d7a65 | 535 | * |
7ae12c80 JW |
536 | * If you want to remove all the pages of one inode, regardless of |
537 | * their use and writeback state, use truncate_inode_pages(). | |
eb1d7a65 | 538 | * |
7ae12c80 | 539 | * Return: the number of the cache entries that were invalidated |
eb1d7a65 YS |
540 | */ |
541 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
542 | pgoff_t start, pgoff_t end) | |
543 | { | |
544 | return __invalidate_mapping_pages(mapping, start, end, NULL); | |
545 | } | |
54bc4855 | 546 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 547 | |
eb1d7a65 | 548 | /** |
649c6dfe AS |
549 | * invalidate_mapping_pagevec - Invalidate all the unlocked pages of one inode |
550 | * @mapping: the address_space which holds the pages to invalidate | |
551 | * @start: the offset 'from' which to invalidate | |
552 | * @end: the offset 'to' which to invalidate (inclusive) | |
553 | * @nr_pagevec: invalidate failed page number for caller | |
554 | * | |
a00cda3f MCC |
555 | * This helper is similar to invalidate_mapping_pages(), except that it accounts |
556 | * for pages that are likely on a pagevec and counts them in @nr_pagevec, which | |
557 | * will be used by the caller. | |
eb1d7a65 YS |
558 | */ |
559 | void invalidate_mapping_pagevec(struct address_space *mapping, | |
560 | pgoff_t start, pgoff_t end, unsigned long *nr_pagevec) | |
561 | { | |
562 | __invalidate_mapping_pages(mapping, start, end, nr_pagevec); | |
563 | } | |
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 | { | |
575 | if (page->mapping != mapping) | |
576 | return 0; | |
577 | ||
266cf658 | 578 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
579 | return 0; |
580 | ||
51b8c1fe | 581 | spin_lock(&mapping->host->i_lock); |
30472509 | 582 | xa_lock_irq(&mapping->i_pages); |
bd4c8ce4 AM |
583 | if (PageDirty(page)) |
584 | goto failed; | |
585 | ||
266cf658 | 586 | BUG_ON(page_has_private(page)); |
62cccb8c | 587 | __delete_from_page_cache(page, NULL); |
30472509 | 588 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
589 | if (mapping_shrinkable(mapping)) |
590 | inode_add_lru(mapping->host); | |
591 | spin_unlock(&mapping->host->i_lock); | |
6072d13c LT |
592 | |
593 | if (mapping->a_ops->freepage) | |
594 | mapping->a_ops->freepage(page); | |
595 | ||
09cbfeaf | 596 | put_page(page); /* pagecache ref */ |
bd4c8ce4 AM |
597 | return 1; |
598 | failed: | |
30472509 | 599 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe | 600 | spin_unlock(&mapping->host->i_lock); |
bd4c8ce4 AM |
601 | return 0; |
602 | } | |
603 | ||
e3db7691 TM |
604 | static int do_launder_page(struct address_space *mapping, struct page *page) |
605 | { | |
606 | if (!PageDirty(page)) | |
607 | return 0; | |
608 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
609 | return 0; | |
610 | return mapping->a_ops->launder_page(page); | |
611 | } | |
612 | ||
1da177e4 LT |
613 | /** |
614 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 615 | * @mapping: the address_space |
1da177e4 LT |
616 | * @start: the page offset 'from' which to invalidate |
617 | * @end: the page offset 'to' which to invalidate (inclusive) | |
618 | * | |
619 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
620 | * invalidation. | |
621 | * | |
a862f68a | 622 | * Return: -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
623 | */ |
624 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
625 | pgoff_t start, pgoff_t end) | |
626 | { | |
0cd6144a | 627 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 628 | struct pagevec pvec; |
b85e0eff | 629 | pgoff_t index; |
1da177e4 LT |
630 | int i; |
631 | int ret = 0; | |
0dd1334f | 632 | int ret2 = 0; |
1da177e4 | 633 | int did_range_unmap = 0; |
1da177e4 | 634 | |
7716506a | 635 | if (mapping_empty(mapping)) |
34ccb69e | 636 | goto out; |
32691f0f | 637 | |
86679820 | 638 | pagevec_init(&pvec); |
b85e0eff | 639 | index = start; |
a656a202 | 640 | while (find_get_entries(mapping, index, end, &pvec, indices)) { |
7b965e08 | 641 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 642 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
643 | |
644 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 645 | index = indices[i]; |
1da177e4 | 646 | |
3159f943 | 647 | if (xa_is_value(page)) { |
c6dcf52c JK |
648 | if (!invalidate_exceptional_entry2(mapping, |
649 | index, page)) | |
650 | ret = -EBUSY; | |
0cd6144a JW |
651 | continue; |
652 | } | |
653 | ||
22061a1f HD |
654 | if (!did_range_unmap && page_mapped(page)) { |
655 | /* | |
656 | * If page is mapped, before taking its lock, | |
657 | * zap the rest of the file in one hit. | |
658 | */ | |
659 | unmap_mapping_pages(mapping, index, | |
660 | (1 + end - index), false); | |
661 | did_range_unmap = 1; | |
662 | } | |
663 | ||
1da177e4 | 664 | lock_page(page); |
5cbc198a | 665 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
666 | if (page->mapping != mapping) { |
667 | unlock_page(page); | |
668 | continue; | |
669 | } | |
1da177e4 | 670 | wait_on_page_writeback(page); |
22061a1f HD |
671 | |
672 | if (page_mapped(page)) | |
673 | unmap_mapping_page(page); | |
d00806b1 | 674 | BUG_ON(page_mapped(page)); |
22061a1f | 675 | |
0dd1334f HH |
676 | ret2 = do_launder_page(mapping, page); |
677 | if (ret2 == 0) { | |
678 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 679 | ret2 = -EBUSY; |
0dd1334f HH |
680 | } |
681 | if (ret2 < 0) | |
682 | ret = ret2; | |
1da177e4 LT |
683 | unlock_page(page); |
684 | } | |
0cd6144a | 685 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
686 | pagevec_release(&pvec); |
687 | cond_resched(); | |
b85e0eff | 688 | index++; |
1da177e4 | 689 | } |
cd656375 | 690 | /* |
69b6c131 | 691 | * For DAX we invalidate page tables after invalidating page cache. We |
cd656375 JK |
692 | * could invalidate page tables while invalidating each entry however |
693 | * that would be expensive. And doing range unmapping before doesn't | |
69b6c131 | 694 | * work as we have no cheap way to find whether page cache entry didn't |
cd656375 JK |
695 | * get remapped later. |
696 | */ | |
697 | if (dax_mapping(mapping)) { | |
977fbdcd | 698 | unmap_mapping_pages(mapping, start, end - start + 1, false); |
cd656375 | 699 | } |
34ccb69e | 700 | out: |
3167760f | 701 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
702 | return ret; |
703 | } | |
704 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
705 | ||
706 | /** | |
707 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 708 | * @mapping: the address_space |
1da177e4 LT |
709 | * |
710 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
711 | * invalidation. | |
712 | * | |
a862f68a | 713 | * Return: -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
714 | */ |
715 | int invalidate_inode_pages2(struct address_space *mapping) | |
716 | { | |
717 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
718 | } | |
719 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 720 | |
721 | /** | |
722 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
723 | * @inode: inode | |
8a549bea | 724 | * @newsize: new file size |
25d9e2d1 | 725 | * |
726 | * inode's new i_size must already be written before truncate_pagecache | |
727 | * is called. | |
728 | * | |
729 | * This function should typically be called before the filesystem | |
730 | * releases resources associated with the freed range (eg. deallocates | |
731 | * blocks). This way, pagecache will always stay logically coherent | |
732 | * with on-disk format, and the filesystem would not have to deal with | |
733 | * situations such as writepage being called for a page that has already | |
734 | * had its underlying blocks deallocated. | |
735 | */ | |
7caef267 | 736 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 737 | { |
cedabed4 | 738 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 739 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
740 | |
741 | /* | |
742 | * unmap_mapping_range is called twice, first simply for | |
743 | * efficiency so that truncate_inode_pages does fewer | |
744 | * single-page unmaps. However after this first call, and | |
745 | * before truncate_inode_pages finishes, it is possible for | |
746 | * private pages to be COWed, which remain after | |
747 | * truncate_inode_pages finishes, hence the second | |
748 | * unmap_mapping_range call must be made for correctness. | |
749 | */ | |
8a549bea HD |
750 | unmap_mapping_range(mapping, holebegin, 0, 1); |
751 | truncate_inode_pages(mapping, newsize); | |
752 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 753 | } |
754 | EXPORT_SYMBOL(truncate_pagecache); | |
755 | ||
2c27c65e CH |
756 | /** |
757 | * truncate_setsize - update inode and pagecache for a new file size | |
758 | * @inode: inode | |
759 | * @newsize: new file size | |
760 | * | |
382e27da JK |
761 | * truncate_setsize updates i_size and performs pagecache truncation (if |
762 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
763 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 764 | * |
77783d06 | 765 | * Must be called with a lock serializing truncates and writes (generally |
9608703e | 766 | * i_rwsem but e.g. xfs uses a different lock) and before all filesystem |
77783d06 | 767 | * specific block truncation has been performed. |
2c27c65e CH |
768 | */ |
769 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
770 | { | |
90a80202 JK |
771 | loff_t oldsize = inode->i_size; |
772 | ||
2c27c65e | 773 | i_size_write(inode, newsize); |
90a80202 JK |
774 | if (newsize > oldsize) |
775 | pagecache_isize_extended(inode, oldsize, newsize); | |
7caef267 | 776 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
777 | } |
778 | EXPORT_SYMBOL(truncate_setsize); | |
779 | ||
90a80202 JK |
780 | /** |
781 | * pagecache_isize_extended - update pagecache after extension of i_size | |
782 | * @inode: inode for which i_size was extended | |
783 | * @from: original inode size | |
784 | * @to: new inode size | |
785 | * | |
786 | * Handle extension of inode size either caused by extending truncate or by | |
787 | * write starting after current i_size. We mark the page straddling current | |
788 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
789 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
790 | * the page before user writes to the page via mmap after the i_size has been | |
791 | * changed. | |
792 | * | |
793 | * The function must be called after i_size is updated so that page fault | |
794 | * coming after we unlock the page will already see the new i_size. | |
9608703e | 795 | * The function must be called while we still hold i_rwsem - this not only |
90a80202 JK |
796 | * makes sure i_size is stable but also that userspace cannot observe new |
797 | * i_size value before we are prepared to store mmap writes at new inode size. | |
798 | */ | |
799 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
800 | { | |
93407472 | 801 | int bsize = i_blocksize(inode); |
90a80202 JK |
802 | loff_t rounded_from; |
803 | struct page *page; | |
804 | pgoff_t index; | |
805 | ||
90a80202 JK |
806 | WARN_ON(to > inode->i_size); |
807 | ||
09cbfeaf | 808 | if (from >= to || bsize == PAGE_SIZE) |
90a80202 JK |
809 | return; |
810 | /* Page straddling @from will not have any hole block created? */ | |
811 | rounded_from = round_up(from, bsize); | |
09cbfeaf | 812 | if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) |
90a80202 JK |
813 | return; |
814 | ||
09cbfeaf | 815 | index = from >> PAGE_SHIFT; |
90a80202 JK |
816 | page = find_lock_page(inode->i_mapping, index); |
817 | /* Page not cached? Nothing to do */ | |
818 | if (!page) | |
819 | return; | |
820 | /* | |
821 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
822 | * is needed. | |
823 | */ | |
824 | if (page_mkclean(page)) | |
825 | set_page_dirty(page); | |
826 | unlock_page(page); | |
09cbfeaf | 827 | put_page(page); |
90a80202 JK |
828 | } |
829 | EXPORT_SYMBOL(pagecache_isize_extended); | |
830 | ||
623e3db9 HD |
831 | /** |
832 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
833 | * @inode: inode | |
834 | * @lstart: offset of beginning of hole | |
835 | * @lend: offset of last byte of hole | |
836 | * | |
837 | * This function should typically be called before the filesystem | |
838 | * releases resources associated with the freed range (eg. deallocates | |
839 | * blocks). This way, pagecache will always stay logically coherent | |
840 | * with on-disk format, and the filesystem would not have to deal with | |
841 | * situations such as writepage being called for a page that has already | |
842 | * had its underlying blocks deallocated. | |
843 | */ | |
844 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
845 | { | |
846 | struct address_space *mapping = inode->i_mapping; | |
847 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
848 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
849 | /* | |
850 | * This rounding is currently just for example: unmap_mapping_range | |
851 | * expands its hole outwards, whereas we want it to contract the hole | |
852 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
853 | * doing their own page rounding first. Note that unmap_mapping_range |
854 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
855 | */ |
856 | ||
857 | /* | |
858 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
859 | * once (before truncating pagecache), and without "even_cows" flag: | |
860 | * hole-punching should not remove private COWed pages from the hole. | |
861 | */ | |
862 | if ((u64)unmap_end > (u64)unmap_start) | |
863 | unmap_mapping_range(mapping, unmap_start, | |
864 | 1 + unmap_end - unmap_start, 0); | |
865 | truncate_inode_pages_range(mapping, lstart, lend); | |
866 | } | |
867 | EXPORT_SYMBOL(truncate_pagecache_range); |