]>
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> |
3a4f8a0b | 22 | #include <linux/shmem_fs.h> |
90a80202 | 23 | #include <linux/rmap.h> |
ba470de4 | 24 | #include "internal.h" |
1da177e4 | 25 | |
f2187599 MG |
26 | /* |
27 | * Regular page slots are stabilized by the page lock even without the tree | |
28 | * itself locked. These unlocked entries need verification under the tree | |
29 | * lock. | |
30 | */ | |
31 | static inline void __clear_shadow_entry(struct address_space *mapping, | |
32 | pgoff_t index, void *entry) | |
0cd6144a | 33 | { |
69b6c131 | 34 | XA_STATE(xas, &mapping->i_pages, index); |
449dd698 | 35 | |
69b6c131 MW |
36 | xas_set_update(&xas, workingset_update_node); |
37 | if (xas_load(&xas) != entry) | |
f2187599 | 38 | return; |
69b6c131 | 39 | xas_store(&xas, NULL); |
f2187599 MG |
40 | } |
41 | ||
42 | static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, | |
43 | void *entry) | |
44 | { | |
51b8c1fe | 45 | spin_lock(&mapping->host->i_lock); |
b93b0163 | 46 | xa_lock_irq(&mapping->i_pages); |
f2187599 | 47 | __clear_shadow_entry(mapping, index, entry); |
b93b0163 | 48 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
49 | if (mapping_shrinkable(mapping)) |
50 | inode_add_lru(mapping->host); | |
51 | spin_unlock(&mapping->host->i_lock); | |
0cd6144a | 52 | } |
1da177e4 | 53 | |
c6dcf52c | 54 | /* |
f2187599 | 55 | * Unconditionally remove exceptional entries. Usually called from truncate |
51dcbdac | 56 | * path. Note that the folio_batch may be altered by this function by removing |
1613fac9 | 57 | * exceptional entries similar to what folio_batch_remove_exceptionals() does. |
c6dcf52c | 58 | */ |
51dcbdac MWO |
59 | static void truncate_folio_batch_exceptionals(struct address_space *mapping, |
60 | struct folio_batch *fbatch, pgoff_t *indices) | |
c6dcf52c | 61 | { |
f2187599 | 62 | int i, j; |
31d270fd | 63 | bool dax; |
f2187599 | 64 | |
c6dcf52c JK |
65 | /* Handled by shmem itself */ |
66 | if (shmem_mapping(mapping)) | |
67 | return; | |
68 | ||
51dcbdac MWO |
69 | for (j = 0; j < folio_batch_count(fbatch); j++) |
70 | if (xa_is_value(fbatch->folios[j])) | |
f2187599 MG |
71 | break; |
72 | ||
51dcbdac | 73 | if (j == folio_batch_count(fbatch)) |
c6dcf52c | 74 | return; |
f2187599 MG |
75 | |
76 | dax = dax_mapping(mapping); | |
51b8c1fe JW |
77 | if (!dax) { |
78 | spin_lock(&mapping->host->i_lock); | |
b93b0163 | 79 | xa_lock_irq(&mapping->i_pages); |
51b8c1fe | 80 | } |
f2187599 | 81 | |
51dcbdac MWO |
82 | for (i = j; i < folio_batch_count(fbatch); i++) { |
83 | struct folio *folio = fbatch->folios[i]; | |
f2187599 MG |
84 | pgoff_t index = indices[i]; |
85 | ||
51dcbdac MWO |
86 | if (!xa_is_value(folio)) { |
87 | fbatch->folios[j++] = folio; | |
f2187599 MG |
88 | continue; |
89 | } | |
90 | ||
f2187599 MG |
91 | if (unlikely(dax)) { |
92 | dax_delete_mapping_entry(mapping, index); | |
93 | continue; | |
94 | } | |
95 | ||
51dcbdac | 96 | __clear_shadow_entry(mapping, index, folio); |
c6dcf52c | 97 | } |
f2187599 | 98 | |
51b8c1fe | 99 | if (!dax) { |
b93b0163 | 100 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
101 | if (mapping_shrinkable(mapping)) |
102 | inode_add_lru(mapping->host); | |
103 | spin_unlock(&mapping->host->i_lock); | |
104 | } | |
51dcbdac | 105 | fbatch->nr = j; |
0e499ed3 MWO |
106 | } |
107 | ||
c6dcf52c JK |
108 | /* |
109 | * Invalidate exceptional entry if easily possible. This handles exceptional | |
4636e70b | 110 | * entries for invalidate_inode_pages(). |
c6dcf52c JK |
111 | */ |
112 | static int invalidate_exceptional_entry(struct address_space *mapping, | |
113 | pgoff_t index, void *entry) | |
114 | { | |
4636e70b RZ |
115 | /* Handled by shmem itself, or for DAX we do nothing. */ |
116 | if (shmem_mapping(mapping) || dax_mapping(mapping)) | |
c6dcf52c | 117 | return 1; |
c6dcf52c JK |
118 | clear_shadow_entry(mapping, index, entry); |
119 | return 1; | |
120 | } | |
121 | ||
122 | /* | |
123 | * Invalidate exceptional entry if clean. This handles exceptional entries for | |
124 | * invalidate_inode_pages2() so for DAX it evicts only clean entries. | |
125 | */ | |
126 | static int invalidate_exceptional_entry2(struct address_space *mapping, | |
127 | pgoff_t index, void *entry) | |
128 | { | |
129 | /* Handled by shmem itself */ | |
130 | if (shmem_mapping(mapping)) | |
131 | return 1; | |
132 | if (dax_mapping(mapping)) | |
133 | return dax_invalidate_mapping_entry_sync(mapping, index); | |
134 | clear_shadow_entry(mapping, index, entry); | |
135 | return 1; | |
136 | } | |
137 | ||
cf9a2ae8 | 138 | /** |
5ad6b2bd MWO |
139 | * folio_invalidate - Invalidate part or all of a folio. |
140 | * @folio: The folio which is affected. | |
d47992f8 LC |
141 | * @offset: start of the range to invalidate |
142 | * @length: length of the range to invalidate | |
cf9a2ae8 | 143 | * |
5ad6b2bd | 144 | * folio_invalidate() is called when all or part of the folio has become |
cf9a2ae8 DH |
145 | * invalidated by a truncate operation. |
146 | * | |
5ad6b2bd | 147 | * folio_invalidate() does not have to release all buffers, but it must |
cf9a2ae8 DH |
148 | * ensure that no dirty buffer is left outside @offset and that no I/O |
149 | * is underway against any of the blocks which are outside the truncation | |
150 | * point. Because the caller is about to free (and possibly reuse) those | |
151 | * blocks on-disk. | |
152 | */ | |
5ad6b2bd | 153 | void folio_invalidate(struct folio *folio, size_t offset, size_t length) |
cf9a2ae8 | 154 | { |
128d1f82 | 155 | const struct address_space_operations *aops = folio->mapping->a_ops; |
d47992f8 | 156 | |
f50015a5 | 157 | if (aops->invalidate_folio) |
128d1f82 | 158 | aops->invalidate_folio(folio, offset, length); |
cf9a2ae8 | 159 | } |
5ad6b2bd | 160 | EXPORT_SYMBOL_GPL(folio_invalidate); |
cf9a2ae8 | 161 | |
1da177e4 LT |
162 | /* |
163 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 164 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 165 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 | 166 | * |
fc3a5ac5 | 167 | * We need to bail out if page->mapping is no longer equal to the original |
1da177e4 | 168 | * mapping. This happens a) when the VM reclaimed the page while we waited on |
fc0ecff6 | 169 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
170 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
171 | */ | |
efe99bba | 172 | static void truncate_cleanup_folio(struct folio *folio) |
1da177e4 | 173 | { |
efe99bba | 174 | if (folio_mapped(folio)) |
3506659e | 175 | unmap_mapping_folio(folio); |
1da177e4 | 176 | |
efe99bba | 177 | if (folio_has_private(folio)) |
5ad6b2bd | 178 | folio_invalidate(folio, 0, folio_size(folio)); |
1da177e4 | 179 | |
b9ea2515 KK |
180 | /* |
181 | * Some filesystems seem to re-dirty the page even after | |
182 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
183 | * Hence dirty accounting check is placed after invalidation. | |
184 | */ | |
efe99bba MWO |
185 | folio_cancel_dirty(folio); |
186 | folio_clear_mappedtodisk(folio); | |
1da177e4 LT |
187 | } |
188 | ||
1e84a3d9 | 189 | int truncate_inode_folio(struct address_space *mapping, struct folio *folio) |
750b4987 | 190 | { |
1e84a3d9 | 191 | if (folio->mapping != mapping) |
9f4e41f4 JK |
192 | return -EIO; |
193 | ||
efe99bba MWO |
194 | truncate_cleanup_folio(folio); |
195 | filemap_remove_folio(folio); | |
9f4e41f4 | 196 | return 0; |
750b4987 NP |
197 | } |
198 | ||
b9a8a419 MWO |
199 | /* |
200 | * Handle partial folios. The folio may be entirely within the | |
201 | * range if a split has raced with us. If not, we zero the part of the | |
202 | * folio that's within the [start, end] range, and then split the folio if | |
203 | * it's large. split_page_range() will discard pages which now lie beyond | |
204 | * i_size, and we rely on the caller to discard pages which lie within a | |
205 | * newly created hole. | |
206 | * | |
207 | * Returns false if splitting failed so the caller can avoid | |
208 | * discarding the entire folio which is stubbornly unsplit. | |
209 | */ | |
210 | bool truncate_inode_partial_folio(struct folio *folio, loff_t start, loff_t end) | |
211 | { | |
212 | loff_t pos = folio_pos(folio); | |
213 | unsigned int offset, length; | |
214 | ||
215 | if (pos < start) | |
216 | offset = start - pos; | |
217 | else | |
218 | offset = 0; | |
219 | length = folio_size(folio); | |
220 | if (pos + length <= (u64)end) | |
221 | length = length - offset; | |
222 | else | |
223 | length = end + 1 - pos - offset; | |
224 | ||
225 | folio_wait_writeback(folio); | |
226 | if (length == folio_size(folio)) { | |
227 | truncate_inode_folio(folio->mapping, folio); | |
228 | return true; | |
229 | } | |
230 | ||
231 | /* | |
232 | * We may be zeroing pages we're about to discard, but it avoids | |
233 | * doing a complex calculation here, and then doing the zeroing | |
234 | * anyway if the page split fails. | |
235 | */ | |
236 | folio_zero_range(folio, offset, length); | |
237 | ||
b9a8a419 | 238 | if (folio_has_private(folio)) |
5ad6b2bd | 239 | folio_invalidate(folio, offset, length); |
b9a8a419 MWO |
240 | if (!folio_test_large(folio)) |
241 | return true; | |
d788f5b3 | 242 | if (split_folio(folio) == 0) |
b9a8a419 MWO |
243 | return true; |
244 | if (folio_test_dirty(folio)) | |
245 | return false; | |
246 | truncate_inode_folio(folio->mapping, folio); | |
247 | return true; | |
248 | } | |
249 | ||
25718736 AK |
250 | /* |
251 | * Used to get rid of pages on hardware memory corruption. | |
252 | */ | |
af7628d6 MWO |
253 | int generic_error_remove_folio(struct address_space *mapping, |
254 | struct folio *folio) | |
25718736 AK |
255 | { |
256 | if (!mapping) | |
257 | return -EINVAL; | |
258 | /* | |
259 | * Only punch for normal data pages for now. | |
260 | * Handling other types like directories would need more auditing. | |
261 | */ | |
262 | if (!S_ISREG(mapping->host->i_mode)) | |
263 | return -EIO; | |
af7628d6 | 264 | return truncate_inode_folio(mapping, folio); |
25718736 | 265 | } |
af7628d6 | 266 | EXPORT_SYMBOL(generic_error_remove_folio); |
25718736 | 267 | |
1e12cbb9 MWO |
268 | /** |
269 | * mapping_evict_folio() - Remove an unused folio from the page-cache. | |
270 | * @mapping: The mapping this folio belongs to. | |
271 | * @folio: The folio to remove. | |
272 | * | |
273 | * Safely remove one folio from the page cache. | |
274 | * It only drops clean, unused folios. | |
275 | * | |
276 | * Context: Folio must be locked. | |
277 | * Return: The number of pages successfully removed. | |
278 | */ | |
279 | long mapping_evict_folio(struct address_space *mapping, struct folio *folio) | |
83f78668 | 280 | { |
1e12cbb9 MWO |
281 | /* The page may have been truncated before it was locked */ |
282 | if (!mapping) | |
283 | return 0; | |
44184813 | 284 | if (folio_test_dirty(folio) || folio_test_writeback(folio)) |
83f78668 | 285 | return 0; |
e41c81d0 MWO |
286 | /* The refcount will be elevated if any page in the folio is mapped */ |
287 | if (folio_ref_count(folio) > | |
288 | folio_nr_pages(folio) + folio_has_private(folio) + 1) | |
83f78668 | 289 | return 0; |
0201ebf2 | 290 | if (!filemap_release_folio(folio, 0)) |
1b8ddbee MWO |
291 | return 0; |
292 | ||
5100da38 | 293 | return remove_mapping(mapping, folio); |
83f78668 WF |
294 | } |
295 | ||
1da177e4 | 296 | /** |
73c1e204 | 297 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
298 | * @mapping: mapping to truncate |
299 | * @lstart: offset from which to truncate | |
5a720394 | 300 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 301 | * |
d7339071 | 302 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
303 | * specified offsets (and zeroing out partial pages |
304 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
305 | * |
306 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
307 | * block on page locks and it will not block on writeback. The second pass | |
308 | * will wait. This is to prevent as much IO as possible in the affected region. | |
309 | * The first pass will remove most pages, so the search cost of the second pass | |
310 | * is low. | |
311 | * | |
1da177e4 LT |
312 | * We pass down the cache-hot hint to the page freeing code. Even if the |
313 | * mapping is large, it is probably the case that the final pages are the most | |
314 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 | 315 | * |
f50015a5 | 316 | * Note that since ->invalidate_folio() accepts range to invalidate |
5a720394 LC |
317 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not |
318 | * page aligned properly. | |
1da177e4 | 319 | */ |
d7339071 HR |
320 | void truncate_inode_pages_range(struct address_space *mapping, |
321 | loff_t lstart, loff_t lend) | |
1da177e4 | 322 | { |
5a720394 LC |
323 | pgoff_t start; /* inclusive */ |
324 | pgoff_t end; /* exclusive */ | |
0e499ed3 | 325 | struct folio_batch fbatch; |
0cd6144a | 326 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
327 | pgoff_t index; |
328 | int i; | |
b9a8a419 MWO |
329 | struct folio *folio; |
330 | bool same_folio; | |
1da177e4 | 331 | |
7716506a | 332 | if (mapping_empty(mapping)) |
0a4ee518 | 333 | return; |
1da177e4 | 334 | |
5a720394 LC |
335 | /* |
336 | * 'start' and 'end' always covers the range of pages to be fully | |
337 | * truncated. Partial pages are covered with 'partial_start' at the | |
338 | * start of the range and 'partial_end' at the end of the range. | |
339 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
340 | */ | |
09cbfeaf | 341 | start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
5a720394 LC |
342 | if (lend == -1) |
343 | /* | |
344 | * lend == -1 indicates end-of-file so we have to set 'end' | |
345 | * to the highest possible pgoff_t and since the type is | |
346 | * unsigned we're using -1. | |
347 | */ | |
348 | end = -1; | |
349 | else | |
09cbfeaf | 350 | end = (lend + 1) >> PAGE_SHIFT; |
d7339071 | 351 | |
51dcbdac | 352 | folio_batch_init(&fbatch); |
b85e0eff | 353 | index = start; |
3392ca12 | 354 | while (index < end && find_lock_entries(mapping, &index, end - 1, |
51dcbdac | 355 | &fbatch, indices)) { |
51dcbdac MWO |
356 | truncate_folio_batch_exceptionals(mapping, &fbatch, indices); |
357 | for (i = 0; i < folio_batch_count(&fbatch); i++) | |
358 | truncate_cleanup_folio(fbatch.folios[i]); | |
359 | delete_from_page_cache_batch(mapping, &fbatch); | |
360 | for (i = 0; i < folio_batch_count(&fbatch); i++) | |
361 | folio_unlock(fbatch.folios[i]); | |
362 | folio_batch_release(&fbatch); | |
1da177e4 LT |
363 | cond_resched(); |
364 | } | |
5c211ba2 | 365 | |
b9a8a419 MWO |
366 | same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT); |
367 | folio = __filemap_get_folio(mapping, lstart >> PAGE_SHIFT, FGP_LOCK, 0); | |
66dabbb6 | 368 | if (!IS_ERR(folio)) { |
b9a8a419 MWO |
369 | same_folio = lend < folio_pos(folio) + folio_size(folio); |
370 | if (!truncate_inode_partial_folio(folio, lstart, lend)) { | |
87b11f86 | 371 | start = folio_next_index(folio); |
b9a8a419 MWO |
372 | if (same_folio) |
373 | end = folio->index; | |
1da177e4 | 374 | } |
b9a8a419 MWO |
375 | folio_unlock(folio); |
376 | folio_put(folio); | |
377 | folio = NULL; | |
1da177e4 | 378 | } |
b9a8a419 | 379 | |
66dabbb6 | 380 | if (!same_folio) { |
b9a8a419 MWO |
381 | folio = __filemap_get_folio(mapping, lend >> PAGE_SHIFT, |
382 | FGP_LOCK, 0); | |
66dabbb6 CH |
383 | if (!IS_ERR(folio)) { |
384 | if (!truncate_inode_partial_folio(folio, lstart, lend)) | |
385 | end = folio->index; | |
386 | folio_unlock(folio); | |
387 | folio_put(folio); | |
388 | } | |
5a720394 | 389 | } |
1da177e4 | 390 | |
b85e0eff | 391 | index = start; |
b9a8a419 | 392 | while (index < end) { |
1da177e4 | 393 | cond_resched(); |
9fb6beea | 394 | if (!find_get_entries(mapping, &index, end - 1, &fbatch, |
38cefeb3 | 395 | indices)) { |
792ceaef | 396 | /* If all gone from start onwards, we're done */ |
b85e0eff | 397 | if (index == start) |
1da177e4 | 398 | break; |
792ceaef | 399 | /* Otherwise restart to make sure all gone */ |
b85e0eff | 400 | index = start; |
1da177e4 LT |
401 | continue; |
402 | } | |
f2187599 | 403 | |
0e499ed3 MWO |
404 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
405 | struct folio *folio = fbatch.folios[i]; | |
1da177e4 | 406 | |
b85e0eff | 407 | /* We rely upon deletion not changing page->index */ |
b85e0eff | 408 | |
0e499ed3 | 409 | if (xa_is_value(folio)) |
0cd6144a | 410 | continue; |
0cd6144a | 411 | |
1e84a3d9 | 412 | folio_lock(folio); |
9fb6beea | 413 | VM_BUG_ON_FOLIO(!folio_contains(folio, indices[i]), folio); |
1e84a3d9 MWO |
414 | folio_wait_writeback(folio); |
415 | truncate_inode_folio(mapping, folio); | |
416 | folio_unlock(folio); | |
1da177e4 | 417 | } |
0e499ed3 MWO |
418 | truncate_folio_batch_exceptionals(mapping, &fbatch, indices); |
419 | folio_batch_release(&fbatch); | |
1da177e4 LT |
420 | } |
421 | } | |
d7339071 | 422 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 423 | |
d7339071 HR |
424 | /** |
425 | * truncate_inode_pages - truncate *all* the pages from an offset | |
426 | * @mapping: mapping to truncate | |
427 | * @lstart: offset from which to truncate | |
428 | * | |
730633f0 JK |
429 | * Called under (and serialised by) inode->i_rwsem and |
430 | * mapping->invalidate_lock. | |
08142579 JK |
431 | * |
432 | * Note: When this function returns, there can be a page in the process of | |
6ffcd825 | 433 | * deletion (inside __filemap_remove_folio()) in the specified range. Thus |
08142579 JK |
434 | * mapping->nrpages can be non-zero when this function returns even after |
435 | * truncation of the whole mapping. | |
d7339071 HR |
436 | */ |
437 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
438 | { | |
439 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
440 | } | |
1da177e4 LT |
441 | EXPORT_SYMBOL(truncate_inode_pages); |
442 | ||
91b0abe3 JW |
443 | /** |
444 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
445 | * @mapping: mapping to truncate | |
446 | * | |
9608703e | 447 | * Called under (and serialized by) inode->i_rwsem. |
91b0abe3 JW |
448 | * |
449 | * Filesystems have to use this in the .evict_inode path to inform the | |
450 | * VM that this is the final truncate and the inode is going away. | |
451 | */ | |
452 | void truncate_inode_pages_final(struct address_space *mapping) | |
453 | { | |
91b0abe3 JW |
454 | /* |
455 | * Page reclaim can not participate in regular inode lifetime | |
456 | * management (can't call iput()) and thus can race with the | |
457 | * inode teardown. Tell it when the address space is exiting, | |
458 | * so that it does not install eviction information after the | |
459 | * final truncate has begun. | |
460 | */ | |
461 | mapping_set_exiting(mapping); | |
462 | ||
7716506a | 463 | if (!mapping_empty(mapping)) { |
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 | */ | |
b93b0163 MW |
470 | xa_lock_irq(&mapping->i_pages); |
471 | xa_unlock_irq(&mapping->i_pages); | |
91b0abe3 | 472 | } |
6ff38bd4 | 473 | |
6ff38bd4 | 474 | truncate_inode_pages(mapping, 0); |
91b0abe3 JW |
475 | } |
476 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
477 | ||
c56109dd | 478 | /** |
1a0fc811 MWO |
479 | * mapping_try_invalidate - Invalidate all the evictable folios of one inode |
480 | * @mapping: the address_space which holds the folios to invalidate | |
c56109dd MWO |
481 | * @start: the offset 'from' which to invalidate |
482 | * @end: the offset 'to' which to invalidate (inclusive) | |
1a0fc811 | 483 | * @nr_failed: How many folio invalidations failed |
c56109dd | 484 | * |
1a0fc811 MWO |
485 | * This function is similar to invalidate_mapping_pages(), except that it |
486 | * returns the number of folios which could not be evicted in @nr_failed. | |
c56109dd | 487 | */ |
1a0fc811 MWO |
488 | unsigned long mapping_try_invalidate(struct address_space *mapping, |
489 | pgoff_t start, pgoff_t end, unsigned long *nr_failed) | |
1da177e4 | 490 | { |
0cd6144a | 491 | pgoff_t indices[PAGEVEC_SIZE]; |
51dcbdac | 492 | struct folio_batch fbatch; |
b85e0eff | 493 | pgoff_t index = start; |
31560180 MK |
494 | unsigned long ret; |
495 | unsigned long count = 0; | |
1da177e4 LT |
496 | int i; |
497 | ||
51dcbdac | 498 | folio_batch_init(&fbatch); |
3392ca12 | 499 | while (find_lock_entries(mapping, &index, end, &fbatch, indices)) { |
51dcbdac | 500 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
b4545f46 | 501 | struct folio *folio = fbatch.folios[i]; |
e0f23603 | 502 | |
b4545f46 | 503 | /* We rely upon deletion not changing folio->index */ |
e0f23603 | 504 | |
b4545f46 | 505 | if (xa_is_value(folio)) { |
7ae12c80 | 506 | count += invalidate_exceptional_entry(mapping, |
3392ca12 | 507 | indices[i], folio); |
0cd6144a JW |
508 | continue; |
509 | } | |
fc127da0 | 510 | |
b4545f46 MWO |
511 | ret = mapping_evict_folio(mapping, folio); |
512 | folio_unlock(folio); | |
31560180 | 513 | /* |
b4545f46 | 514 | * Invalidation is a hint that the folio is no longer |
31560180 MK |
515 | * of interest and try to speed up its reclaim. |
516 | */ | |
eb1d7a65 | 517 | if (!ret) { |
261b6840 | 518 | deactivate_file_folio(folio); |
1a0fc811 MWO |
519 | /* Likely in the lru cache of a remote CPU */ |
520 | if (nr_failed) | |
521 | (*nr_failed)++; | |
eb1d7a65 | 522 | } |
31560180 | 523 | count += ret; |
1da177e4 | 524 | } |
51dcbdac MWO |
525 | folio_batch_remove_exceptionals(&fbatch); |
526 | folio_batch_release(&fbatch); | |
28697355 | 527 | cond_resched(); |
1da177e4 | 528 | } |
31560180 | 529 | return count; |
1da177e4 | 530 | } |
eb1d7a65 YS |
531 | |
532 | /** | |
7ae12c80 JW |
533 | * invalidate_mapping_pages - Invalidate all clean, unlocked cache of one inode |
534 | * @mapping: the address_space which holds the cache to invalidate | |
eb1d7a65 YS |
535 | * @start: the offset 'from' which to invalidate |
536 | * @end: the offset 'to' which to invalidate (inclusive) | |
537 | * | |
7ae12c80 JW |
538 | * This function removes pages that are clean, unmapped and unlocked, |
539 | * as well as shadow entries. It will not block on IO activity. | |
eb1d7a65 | 540 | * |
7ae12c80 JW |
541 | * If you want to remove all the pages of one inode, regardless of |
542 | * their use and writeback state, use truncate_inode_pages(). | |
eb1d7a65 | 543 | * |
1a0fc811 | 544 | * Return: The number of indices that had their contents invalidated |
eb1d7a65 YS |
545 | */ |
546 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
547 | pgoff_t start, pgoff_t end) | |
548 | { | |
1a0fc811 | 549 | return mapping_try_invalidate(mapping, start, end, NULL); |
eb1d7a65 | 550 | } |
54bc4855 | 551 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 552 | |
bd4c8ce4 | 553 | /* |
2033c98c | 554 | * This is like mapping_evict_folio(), except it ignores the folio's |
bd4c8ce4 | 555 | * refcount. We do this because invalidate_inode_pages2() needs stronger |
2033c98c | 556 | * invalidation guarantees, and cannot afford to leave folios behind because |
2706a1b8 | 557 | * shrink_page_list() has a temp ref on them, or because they're transiently |
1fec6890 | 558 | * sitting in the folio_add_lru() caches. |
bd4c8ce4 | 559 | */ |
78f42660 MWO |
560 | static int invalidate_complete_folio2(struct address_space *mapping, |
561 | struct folio *folio) | |
bd4c8ce4 | 562 | { |
78f42660 | 563 | if (folio->mapping != mapping) |
bd4c8ce4 AM |
564 | return 0; |
565 | ||
0201ebf2 | 566 | if (!filemap_release_folio(folio, GFP_KERNEL)) |
bd4c8ce4 AM |
567 | return 0; |
568 | ||
51b8c1fe | 569 | spin_lock(&mapping->host->i_lock); |
30472509 | 570 | xa_lock_irq(&mapping->i_pages); |
78f42660 | 571 | if (folio_test_dirty(folio)) |
bd4c8ce4 AM |
572 | goto failed; |
573 | ||
78f42660 MWO |
574 | BUG_ON(folio_has_private(folio)); |
575 | __filemap_remove_folio(folio, NULL); | |
30472509 | 576 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
577 | if (mapping_shrinkable(mapping)) |
578 | inode_add_lru(mapping->host); | |
579 | spin_unlock(&mapping->host->i_lock); | |
6072d13c | 580 | |
78f42660 | 581 | filemap_free_folio(mapping, folio); |
bd4c8ce4 AM |
582 | return 1; |
583 | failed: | |
30472509 | 584 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe | 585 | spin_unlock(&mapping->host->i_lock); |
bd4c8ce4 AM |
586 | return 0; |
587 | } | |
588 | ||
affa80e8 | 589 | static int folio_launder(struct address_space *mapping, struct folio *folio) |
e3db7691 | 590 | { |
f6357c3a | 591 | if (!folio_test_dirty(folio)) |
e3db7691 | 592 | return 0; |
affa80e8 | 593 | if (folio->mapping != mapping || mapping->a_ops->launder_folio == NULL) |
e3db7691 | 594 | return 0; |
affa80e8 | 595 | return mapping->a_ops->launder_folio(folio); |
e3db7691 TM |
596 | } |
597 | ||
1da177e4 LT |
598 | /** |
599 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 600 | * @mapping: the address_space |
1da177e4 LT |
601 | * @start: the page offset 'from' which to invalidate |
602 | * @end: the page offset 'to' which to invalidate (inclusive) | |
603 | * | |
604 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
605 | * invalidation. | |
606 | * | |
a862f68a | 607 | * Return: -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
608 | */ |
609 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
610 | pgoff_t start, pgoff_t end) | |
611 | { | |
0cd6144a | 612 | pgoff_t indices[PAGEVEC_SIZE]; |
0e499ed3 | 613 | struct folio_batch fbatch; |
b85e0eff | 614 | pgoff_t index; |
1da177e4 LT |
615 | int i; |
616 | int ret = 0; | |
0dd1334f | 617 | int ret2 = 0; |
1da177e4 | 618 | int did_range_unmap = 0; |
1da177e4 | 619 | |
7716506a | 620 | if (mapping_empty(mapping)) |
0a4ee518 | 621 | return 0; |
32691f0f | 622 | |
0e499ed3 | 623 | folio_batch_init(&fbatch); |
b85e0eff | 624 | index = start; |
9fb6beea | 625 | while (find_get_entries(mapping, &index, end, &fbatch, indices)) { |
0e499ed3 MWO |
626 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
627 | struct folio *folio = fbatch.folios[i]; | |
b85e0eff | 628 | |
fae9bc4a | 629 | /* We rely upon deletion not changing folio->index */ |
1da177e4 | 630 | |
0e499ed3 | 631 | if (xa_is_value(folio)) { |
c6dcf52c | 632 | if (!invalidate_exceptional_entry2(mapping, |
9fb6beea | 633 | indices[i], folio)) |
c6dcf52c | 634 | ret = -EBUSY; |
0cd6144a JW |
635 | continue; |
636 | } | |
637 | ||
fae9bc4a | 638 | if (!did_range_unmap && folio_mapped(folio)) { |
22061a1f | 639 | /* |
fae9bc4a | 640 | * If folio is mapped, before taking its lock, |
22061a1f HD |
641 | * zap the rest of the file in one hit. |
642 | */ | |
9fb6beea VMO |
643 | unmap_mapping_pages(mapping, indices[i], |
644 | (1 + end - indices[i]), false); | |
22061a1f HD |
645 | did_range_unmap = 1; |
646 | } | |
647 | ||
fae9bc4a | 648 | folio_lock(folio); |
aa5b9178 | 649 | if (unlikely(folio->mapping != mapping)) { |
fae9bc4a | 650 | folio_unlock(folio); |
1da177e4 LT |
651 | continue; |
652 | } | |
aa5b9178 | 653 | VM_BUG_ON_FOLIO(!folio_contains(folio, indices[i]), folio); |
fae9bc4a | 654 | folio_wait_writeback(folio); |
22061a1f | 655 | |
fae9bc4a MWO |
656 | if (folio_mapped(folio)) |
657 | unmap_mapping_folio(folio); | |
658 | BUG_ON(folio_mapped(folio)); | |
22061a1f | 659 | |
affa80e8 | 660 | ret2 = folio_launder(mapping, folio); |
0dd1334f | 661 | if (ret2 == 0) { |
78f42660 | 662 | if (!invalidate_complete_folio2(mapping, folio)) |
6ccfa806 | 663 | ret2 = -EBUSY; |
0dd1334f HH |
664 | } |
665 | if (ret2 < 0) | |
666 | ret = ret2; | |
fae9bc4a | 667 | folio_unlock(folio); |
1da177e4 | 668 | } |
0e499ed3 MWO |
669 | folio_batch_remove_exceptionals(&fbatch); |
670 | folio_batch_release(&fbatch); | |
1da177e4 LT |
671 | cond_resched(); |
672 | } | |
cd656375 | 673 | /* |
69b6c131 | 674 | * For DAX we invalidate page tables after invalidating page cache. We |
cd656375 JK |
675 | * could invalidate page tables while invalidating each entry however |
676 | * that would be expensive. And doing range unmapping before doesn't | |
69b6c131 | 677 | * work as we have no cheap way to find whether page cache entry didn't |
cd656375 JK |
678 | * get remapped later. |
679 | */ | |
680 | if (dax_mapping(mapping)) { | |
977fbdcd | 681 | unmap_mapping_pages(mapping, start, end - start + 1, false); |
cd656375 | 682 | } |
1da177e4 LT |
683 | return ret; |
684 | } | |
685 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
686 | ||
687 | /** | |
688 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 689 | * @mapping: the address_space |
1da177e4 LT |
690 | * |
691 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
692 | * invalidation. | |
693 | * | |
a862f68a | 694 | * Return: -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
695 | */ |
696 | int invalidate_inode_pages2(struct address_space *mapping) | |
697 | { | |
698 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
699 | } | |
700 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 NP |
701 | |
702 | /** | |
703 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
704 | * @inode: inode | |
8a549bea | 705 | * @newsize: new file size |
25d9e2d1 NP |
706 | * |
707 | * inode's new i_size must already be written before truncate_pagecache | |
708 | * is called. | |
709 | * | |
710 | * This function should typically be called before the filesystem | |
711 | * releases resources associated with the freed range (eg. deallocates | |
712 | * blocks). This way, pagecache will always stay logically coherent | |
713 | * with on-disk format, and the filesystem would not have to deal with | |
714 | * situations such as writepage being called for a page that has already | |
715 | * had its underlying blocks deallocated. | |
716 | */ | |
7caef267 | 717 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 718 | { |
cedabed4 | 719 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 720 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
721 | |
722 | /* | |
723 | * unmap_mapping_range is called twice, first simply for | |
724 | * efficiency so that truncate_inode_pages does fewer | |
725 | * single-page unmaps. However after this first call, and | |
726 | * before truncate_inode_pages finishes, it is possible for | |
727 | * private pages to be COWed, which remain after | |
728 | * truncate_inode_pages finishes, hence the second | |
729 | * unmap_mapping_range call must be made for correctness. | |
730 | */ | |
8a549bea HD |
731 | unmap_mapping_range(mapping, holebegin, 0, 1); |
732 | truncate_inode_pages(mapping, newsize); | |
733 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 NP |
734 | } |
735 | EXPORT_SYMBOL(truncate_pagecache); | |
736 | ||
2c27c65e CH |
737 | /** |
738 | * truncate_setsize - update inode and pagecache for a new file size | |
739 | * @inode: inode | |
740 | * @newsize: new file size | |
741 | * | |
382e27da JK |
742 | * truncate_setsize updates i_size and performs pagecache truncation (if |
743 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
744 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 745 | * |
77783d06 | 746 | * Must be called with a lock serializing truncates and writes (generally |
9608703e | 747 | * i_rwsem but e.g. xfs uses a different lock) and before all filesystem |
77783d06 | 748 | * specific block truncation has been performed. |
2c27c65e CH |
749 | */ |
750 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
751 | { | |
90a80202 JK |
752 | loff_t oldsize = inode->i_size; |
753 | ||
2c27c65e | 754 | i_size_write(inode, newsize); |
90a80202 JK |
755 | if (newsize > oldsize) |
756 | pagecache_isize_extended(inode, oldsize, newsize); | |
7caef267 | 757 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
758 | } |
759 | EXPORT_SYMBOL(truncate_setsize); | |
760 | ||
90a80202 JK |
761 | /** |
762 | * pagecache_isize_extended - update pagecache after extension of i_size | |
763 | * @inode: inode for which i_size was extended | |
764 | * @from: original inode size | |
765 | * @to: new inode size | |
766 | * | |
767 | * Handle extension of inode size either caused by extending truncate or by | |
768 | * write starting after current i_size. We mark the page straddling current | |
769 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
770 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
771 | * the page before user writes to the page via mmap after the i_size has been | |
772 | * changed. | |
773 | * | |
774 | * The function must be called after i_size is updated so that page fault | |
775 | * coming after we unlock the page will already see the new i_size. | |
9608703e | 776 | * The function must be called while we still hold i_rwsem - this not only |
90a80202 JK |
777 | * makes sure i_size is stable but also that userspace cannot observe new |
778 | * i_size value before we are prepared to store mmap writes at new inode size. | |
779 | */ | |
780 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
781 | { | |
93407472 | 782 | int bsize = i_blocksize(inode); |
90a80202 JK |
783 | loff_t rounded_from; |
784 | struct page *page; | |
785 | pgoff_t index; | |
786 | ||
90a80202 JK |
787 | WARN_ON(to > inode->i_size); |
788 | ||
09cbfeaf | 789 | if (from >= to || bsize == PAGE_SIZE) |
90a80202 JK |
790 | return; |
791 | /* Page straddling @from will not have any hole block created? */ | |
792 | rounded_from = round_up(from, bsize); | |
09cbfeaf | 793 | if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) |
90a80202 JK |
794 | return; |
795 | ||
09cbfeaf | 796 | index = from >> PAGE_SHIFT; |
90a80202 JK |
797 | page = find_lock_page(inode->i_mapping, index); |
798 | /* Page not cached? Nothing to do */ | |
799 | if (!page) | |
800 | return; | |
801 | /* | |
802 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
803 | * is needed. | |
804 | */ | |
805 | if (page_mkclean(page)) | |
806 | set_page_dirty(page); | |
807 | unlock_page(page); | |
09cbfeaf | 808 | put_page(page); |
90a80202 JK |
809 | } |
810 | EXPORT_SYMBOL(pagecache_isize_extended); | |
811 | ||
623e3db9 HD |
812 | /** |
813 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
814 | * @inode: inode | |
815 | * @lstart: offset of beginning of hole | |
816 | * @lend: offset of last byte of hole | |
817 | * | |
818 | * This function should typically be called before the filesystem | |
819 | * releases resources associated with the freed range (eg. deallocates | |
820 | * blocks). This way, pagecache will always stay logically coherent | |
821 | * with on-disk format, and the filesystem would not have to deal with | |
822 | * situations such as writepage being called for a page that has already | |
823 | * had its underlying blocks deallocated. | |
824 | */ | |
825 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
826 | { | |
827 | struct address_space *mapping = inode->i_mapping; | |
828 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
829 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
830 | /* | |
831 | * This rounding is currently just for example: unmap_mapping_range | |
832 | * expands its hole outwards, whereas we want it to contract the hole | |
833 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
834 | * doing their own page rounding first. Note that unmap_mapping_range |
835 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
836 | */ |
837 | ||
838 | /* | |
839 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
840 | * once (before truncating pagecache), and without "even_cows" flag: | |
841 | * hole-punching should not remove private COWed pages from the hole. | |
842 | */ | |
843 | if ((u64)unmap_end > (u64)unmap_start) | |
844 | unmap_mapping_range(mapping, unmap_start, | |
845 | 1 + unmap_end - unmap_start, 0); | |
846 | truncate_inode_pages_range(mapping, lstart, lend); | |
847 | } | |
848 | EXPORT_SYMBOL(truncate_pagecache_range); |