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 | |
61c663e0 | 26 | static void clear_shadow_entries(struct address_space *mapping, |
d3db2c04 | 27 | unsigned long start, unsigned long max) |
f2187599 | 28 | { |
d3db2c04 SB |
29 | XA_STATE(xas, &mapping->i_pages, start); |
30 | struct folio *folio; | |
61c663e0 YZ |
31 | |
32 | /* Handled by shmem itself, or for DAX we do nothing. */ | |
33 | if (shmem_mapping(mapping) || dax_mapping(mapping)) | |
34 | return; | |
35 | ||
d3db2c04 | 36 | xas_set_update(&xas, workingset_update_node); |
61c663e0 | 37 | |
d3db2c04 SB |
38 | spin_lock(&mapping->host->i_lock); |
39 | xas_lock_irq(&xas); | |
61c663e0 | 40 | |
d3db2c04 SB |
41 | /* Clear all shadow entries from start to max */ |
42 | xas_for_each(&xas, folio, max) { | |
61c663e0 | 43 | if (xa_is_value(folio)) |
d3db2c04 | 44 | xas_store(&xas, NULL); |
61c663e0 YZ |
45 | } |
46 | ||
d3db2c04 | 47 | xas_unlock_irq(&xas); |
51b8c1fe JW |
48 | if (mapping_shrinkable(mapping)) |
49 | inode_add_lru(mapping->host); | |
50 | spin_unlock(&mapping->host->i_lock); | |
0cd6144a | 51 | } |
1da177e4 | 52 | |
c6dcf52c | 53 | /* |
f2187599 | 54 | * Unconditionally remove exceptional entries. Usually called from truncate |
51dcbdac | 55 | * path. Note that the folio_batch may be altered by this function by removing |
1613fac9 | 56 | * exceptional entries similar to what folio_batch_remove_exceptionals() does. |
cb8e64be SB |
57 | * Please note that indices[] has entries in ascending order as guaranteed by |
58 | * either find_get_entries() or find_lock_entries(). | |
c6dcf52c | 59 | */ |
51dcbdac MWO |
60 | static void truncate_folio_batch_exceptionals(struct address_space *mapping, |
61 | struct folio_batch *fbatch, pgoff_t *indices) | |
c6dcf52c | 62 | { |
cb8e64be SB |
63 | XA_STATE(xas, &mapping->i_pages, indices[0]); |
64 | int nr = folio_batch_count(fbatch); | |
65 | struct folio *folio; | |
f2187599 | 66 | int i, j; |
f2187599 | 67 | |
c6dcf52c JK |
68 | /* Handled by shmem itself */ |
69 | if (shmem_mapping(mapping)) | |
70 | return; | |
71 | ||
cb8e64be | 72 | for (j = 0; j < nr; j++) |
51dcbdac | 73 | if (xa_is_value(fbatch->folios[j])) |
f2187599 MG |
74 | break; |
75 | ||
cb8e64be | 76 | if (j == nr) |
c6dcf52c | 77 | return; |
f2187599 | 78 | |
cb8e64be SB |
79 | if (dax_mapping(mapping)) { |
80 | for (i = j; i < nr; i++) { | |
bde708f1 AP |
81 | if (xa_is_value(fbatch->folios[i])) { |
82 | /* | |
83 | * File systems should already have called | |
84 | * dax_break_layout_entry() to remove all DAX | |
85 | * entries while holding a lock to prevent | |
86 | * establishing new entries. Therefore we | |
87 | * shouldn't find any here. | |
88 | */ | |
89 | WARN_ON_ONCE(1); | |
90 | ||
91 | /* | |
92 | * Delete the mapping so truncate_pagecache() | |
93 | * doesn't loop forever. | |
94 | */ | |
cb8e64be | 95 | dax_delete_mapping_entry(mapping, indices[i]); |
bde708f1 | 96 | } |
cb8e64be SB |
97 | } |
98 | goto out; | |
51b8c1fe | 99 | } |
f2187599 | 100 | |
cb8e64be SB |
101 | xas_set(&xas, indices[j]); |
102 | xas_set_update(&xas, workingset_update_node); | |
f2187599 | 103 | |
cb8e64be SB |
104 | spin_lock(&mapping->host->i_lock); |
105 | xas_lock_irq(&xas); | |
f2187599 | 106 | |
cb8e64be SB |
107 | xas_for_each(&xas, folio, indices[nr-1]) { |
108 | if (xa_is_value(folio)) | |
109 | xas_store(&xas, NULL); | |
c6dcf52c | 110 | } |
f2187599 | 111 | |
cb8e64be SB |
112 | xas_unlock_irq(&xas); |
113 | if (mapping_shrinkable(mapping)) | |
114 | inode_add_lru(mapping->host); | |
115 | spin_unlock(&mapping->host->i_lock); | |
116 | out: | |
117 | folio_batch_remove_exceptionals(fbatch); | |
0e499ed3 MWO |
118 | } |
119 | ||
cf9a2ae8 | 120 | /** |
5ad6b2bd MWO |
121 | * folio_invalidate - Invalidate part or all of a folio. |
122 | * @folio: The folio which is affected. | |
d47992f8 LC |
123 | * @offset: start of the range to invalidate |
124 | * @length: length of the range to invalidate | |
cf9a2ae8 | 125 | * |
5ad6b2bd | 126 | * folio_invalidate() is called when all or part of the folio has become |
cf9a2ae8 DH |
127 | * invalidated by a truncate operation. |
128 | * | |
5ad6b2bd | 129 | * folio_invalidate() does not have to release all buffers, but it must |
cf9a2ae8 DH |
130 | * ensure that no dirty buffer is left outside @offset and that no I/O |
131 | * is underway against any of the blocks which are outside the truncation | |
132 | * point. Because the caller is about to free (and possibly reuse) those | |
133 | * blocks on-disk. | |
134 | */ | |
5ad6b2bd | 135 | void folio_invalidate(struct folio *folio, size_t offset, size_t length) |
cf9a2ae8 | 136 | { |
128d1f82 | 137 | const struct address_space_operations *aops = folio->mapping->a_ops; |
d47992f8 | 138 | |
f50015a5 | 139 | if (aops->invalidate_folio) |
128d1f82 | 140 | aops->invalidate_folio(folio, offset, length); |
cf9a2ae8 | 141 | } |
5ad6b2bd | 142 | EXPORT_SYMBOL_GPL(folio_invalidate); |
cf9a2ae8 | 143 | |
1da177e4 LT |
144 | /* |
145 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 146 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 147 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 | 148 | * |
fc3a5ac5 | 149 | * We need to bail out if page->mapping is no longer equal to the original |
1da177e4 | 150 | * mapping. This happens a) when the VM reclaimed the page while we waited on |
fc0ecff6 | 151 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
152 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
153 | */ | |
efe99bba | 154 | static void truncate_cleanup_folio(struct folio *folio) |
1da177e4 | 155 | { |
efe99bba | 156 | if (folio_mapped(folio)) |
3506659e | 157 | unmap_mapping_folio(folio); |
1da177e4 | 158 | |
0aa2e1b2 | 159 | if (folio_needs_release(folio)) |
5ad6b2bd | 160 | folio_invalidate(folio, 0, folio_size(folio)); |
1da177e4 | 161 | |
b9ea2515 KK |
162 | /* |
163 | * Some filesystems seem to re-dirty the page even after | |
164 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
165 | * Hence dirty accounting check is placed after invalidation. | |
166 | */ | |
efe99bba | 167 | folio_cancel_dirty(folio); |
1da177e4 LT |
168 | } |
169 | ||
1e84a3d9 | 170 | int truncate_inode_folio(struct address_space *mapping, struct folio *folio) |
750b4987 | 171 | { |
1e84a3d9 | 172 | if (folio->mapping != mapping) |
9f4e41f4 JK |
173 | return -EIO; |
174 | ||
efe99bba MWO |
175 | truncate_cleanup_folio(folio); |
176 | filemap_remove_folio(folio); | |
9f4e41f4 | 177 | return 0; |
750b4987 NP |
178 | } |
179 | ||
b9a8a419 MWO |
180 | /* |
181 | * Handle partial folios. The folio may be entirely within the | |
182 | * range if a split has raced with us. If not, we zero the part of the | |
183 | * folio that's within the [start, end] range, and then split the folio if | |
184 | * it's large. split_page_range() will discard pages which now lie beyond | |
185 | * i_size, and we rely on the caller to discard pages which lie within a | |
186 | * newly created hole. | |
187 | * | |
188 | * Returns false if splitting failed so the caller can avoid | |
189 | * discarding the entire folio which is stubbornly unsplit. | |
190 | */ | |
191 | bool truncate_inode_partial_folio(struct folio *folio, loff_t start, loff_t end) | |
192 | { | |
193 | loff_t pos = folio_pos(folio); | |
66f28ffb | 194 | size_t size = folio_size(folio); |
b9a8a419 | 195 | unsigned int offset, length; |
7460b470 | 196 | struct page *split_at, *split_at2; |
b9a8a419 MWO |
197 | |
198 | if (pos < start) | |
199 | offset = start - pos; | |
200 | else | |
201 | offset = 0; | |
66f28ffb ZY |
202 | if (pos + size <= (u64)end) |
203 | length = size - offset; | |
b9a8a419 MWO |
204 | else |
205 | length = end + 1 - pos - offset; | |
206 | ||
207 | folio_wait_writeback(folio); | |
66f28ffb | 208 | if (length == size) { |
b9a8a419 MWO |
209 | truncate_inode_folio(folio->mapping, folio); |
210 | return true; | |
211 | } | |
212 | ||
213 | /* | |
214 | * We may be zeroing pages we're about to discard, but it avoids | |
215 | * doing a complex calculation here, and then doing the zeroing | |
216 | * anyway if the page split fails. | |
217 | */ | |
27e6a24a | 218 | if (!mapping_inaccessible(folio->mapping)) |
c72ceafb | 219 | folio_zero_range(folio, offset, length); |
b9a8a419 | 220 | |
0aa2e1b2 | 221 | if (folio_needs_release(folio)) |
5ad6b2bd | 222 | folio_invalidate(folio, offset, length); |
b9a8a419 MWO |
223 | if (!folio_test_large(folio)) |
224 | return true; | |
7460b470 ZY |
225 | |
226 | split_at = folio_page(folio, PAGE_ALIGN_DOWN(offset) / PAGE_SIZE); | |
7460b470 ZY |
227 | if (!try_folio_split(folio, split_at, NULL)) { |
228 | /* | |
229 | * try to split at offset + length to make sure folios within | |
230 | * the range can be dropped, especially to avoid memory waste | |
231 | * for shmem truncate | |
232 | */ | |
66f28ffb ZY |
233 | struct folio *folio2; |
234 | ||
235 | if (offset + length == size) | |
236 | goto no_split; | |
237 | ||
238 | split_at2 = folio_page(folio, | |
239 | PAGE_ALIGN_DOWN(offset + length) / PAGE_SIZE); | |
240 | folio2 = page_folio(split_at2); | |
7460b470 ZY |
241 | |
242 | if (!folio_try_get(folio2)) | |
243 | goto no_split; | |
244 | ||
245 | if (!folio_test_large(folio2)) | |
246 | goto out; | |
247 | ||
248 | if (!folio_trylock(folio2)) | |
249 | goto out; | |
250 | ||
251 | /* | |
252 | * make sure folio2 is large and does not change its mapping. | |
253 | * Its split result does not matter here. | |
254 | */ | |
255 | if (folio_test_large(folio2) && | |
256 | folio2->mapping == folio->mapping) | |
257 | try_folio_split(folio2, split_at2, NULL); | |
258 | ||
259 | folio_unlock(folio2); | |
260 | out: | |
261 | folio_put(folio2); | |
262 | no_split: | |
b9a8a419 | 263 | return true; |
7460b470 | 264 | } |
b9a8a419 MWO |
265 | if (folio_test_dirty(folio)) |
266 | return false; | |
267 | truncate_inode_folio(folio->mapping, folio); | |
268 | return true; | |
269 | } | |
270 | ||
25718736 AK |
271 | /* |
272 | * Used to get rid of pages on hardware memory corruption. | |
273 | */ | |
af7628d6 MWO |
274 | int generic_error_remove_folio(struct address_space *mapping, |
275 | struct folio *folio) | |
25718736 AK |
276 | { |
277 | if (!mapping) | |
278 | return -EINVAL; | |
279 | /* | |
280 | * Only punch for normal data pages for now. | |
281 | * Handling other types like directories would need more auditing. | |
282 | */ | |
283 | if (!S_ISREG(mapping->host->i_mode)) | |
284 | return -EIO; | |
af7628d6 | 285 | return truncate_inode_folio(mapping, folio); |
25718736 | 286 | } |
af7628d6 | 287 | EXPORT_SYMBOL(generic_error_remove_folio); |
25718736 | 288 | |
1e12cbb9 MWO |
289 | /** |
290 | * mapping_evict_folio() - Remove an unused folio from the page-cache. | |
291 | * @mapping: The mapping this folio belongs to. | |
292 | * @folio: The folio to remove. | |
293 | * | |
294 | * Safely remove one folio from the page cache. | |
295 | * It only drops clean, unused folios. | |
296 | * | |
297 | * Context: Folio must be locked. | |
298 | * Return: The number of pages successfully removed. | |
299 | */ | |
300 | long mapping_evict_folio(struct address_space *mapping, struct folio *folio) | |
83f78668 | 301 | { |
1e12cbb9 MWO |
302 | /* The page may have been truncated before it was locked */ |
303 | if (!mapping) | |
304 | return 0; | |
44184813 | 305 | if (folio_test_dirty(folio) || folio_test_writeback(folio)) |
83f78668 | 306 | return 0; |
e41c81d0 MWO |
307 | /* The refcount will be elevated if any page in the folio is mapped */ |
308 | if (folio_ref_count(folio) > | |
309 | folio_nr_pages(folio) + folio_has_private(folio) + 1) | |
83f78668 | 310 | return 0; |
0201ebf2 | 311 | if (!filemap_release_folio(folio, 0)) |
1b8ddbee MWO |
312 | return 0; |
313 | ||
5100da38 | 314 | return remove_mapping(mapping, folio); |
83f78668 WF |
315 | } |
316 | ||
1da177e4 | 317 | /** |
73c1e204 | 318 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
319 | * @mapping: mapping to truncate |
320 | * @lstart: offset from which to truncate | |
5a720394 | 321 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 322 | * |
d7339071 | 323 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
324 | * specified offsets (and zeroing out partial pages |
325 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
326 | * |
327 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
328 | * block on page locks and it will not block on writeback. The second pass | |
329 | * will wait. This is to prevent as much IO as possible in the affected region. | |
330 | * The first pass will remove most pages, so the search cost of the second pass | |
331 | * is low. | |
332 | * | |
1da177e4 LT |
333 | * We pass down the cache-hot hint to the page freeing code. Even if the |
334 | * mapping is large, it is probably the case that the final pages are the most | |
335 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 | 336 | * |
f50015a5 | 337 | * Note that since ->invalidate_folio() accepts range to invalidate |
5a720394 LC |
338 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not |
339 | * page aligned properly. | |
1da177e4 | 340 | */ |
d7339071 HR |
341 | void truncate_inode_pages_range(struct address_space *mapping, |
342 | loff_t lstart, loff_t lend) | |
1da177e4 | 343 | { |
5a720394 LC |
344 | pgoff_t start; /* inclusive */ |
345 | pgoff_t end; /* exclusive */ | |
0e499ed3 | 346 | struct folio_batch fbatch; |
0cd6144a | 347 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
348 | pgoff_t index; |
349 | int i; | |
b9a8a419 MWO |
350 | struct folio *folio; |
351 | bool same_folio; | |
1da177e4 | 352 | |
7716506a | 353 | if (mapping_empty(mapping)) |
0a4ee518 | 354 | return; |
1da177e4 | 355 | |
5a720394 LC |
356 | /* |
357 | * 'start' and 'end' always covers the range of pages to be fully | |
358 | * truncated. Partial pages are covered with 'partial_start' at the | |
359 | * start of the range and 'partial_end' at the end of the range. | |
360 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
361 | */ | |
09cbfeaf | 362 | start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
5a720394 LC |
363 | if (lend == -1) |
364 | /* | |
365 | * lend == -1 indicates end-of-file so we have to set 'end' | |
366 | * to the highest possible pgoff_t and since the type is | |
367 | * unsigned we're using -1. | |
368 | */ | |
369 | end = -1; | |
370 | else | |
09cbfeaf | 371 | end = (lend + 1) >> PAGE_SHIFT; |
d7339071 | 372 | |
51dcbdac | 373 | folio_batch_init(&fbatch); |
b85e0eff | 374 | index = start; |
3392ca12 | 375 | while (index < end && find_lock_entries(mapping, &index, end - 1, |
51dcbdac | 376 | &fbatch, indices)) { |
51dcbdac MWO |
377 | truncate_folio_batch_exceptionals(mapping, &fbatch, indices); |
378 | for (i = 0; i < folio_batch_count(&fbatch); i++) | |
379 | truncate_cleanup_folio(fbatch.folios[i]); | |
380 | delete_from_page_cache_batch(mapping, &fbatch); | |
381 | for (i = 0; i < folio_batch_count(&fbatch); i++) | |
382 | folio_unlock(fbatch.folios[i]); | |
383 | folio_batch_release(&fbatch); | |
1da177e4 LT |
384 | cond_resched(); |
385 | } | |
5c211ba2 | 386 | |
b9a8a419 MWO |
387 | same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT); |
388 | folio = __filemap_get_folio(mapping, lstart >> PAGE_SHIFT, FGP_LOCK, 0); | |
66dabbb6 | 389 | if (!IS_ERR(folio)) { |
b9a8a419 MWO |
390 | same_folio = lend < folio_pos(folio) + folio_size(folio); |
391 | if (!truncate_inode_partial_folio(folio, lstart, lend)) { | |
87b11f86 | 392 | start = folio_next_index(folio); |
b9a8a419 MWO |
393 | if (same_folio) |
394 | end = folio->index; | |
1da177e4 | 395 | } |
b9a8a419 MWO |
396 | folio_unlock(folio); |
397 | folio_put(folio); | |
398 | folio = NULL; | |
1da177e4 | 399 | } |
b9a8a419 | 400 | |
66dabbb6 | 401 | if (!same_folio) { |
b9a8a419 MWO |
402 | folio = __filemap_get_folio(mapping, lend >> PAGE_SHIFT, |
403 | FGP_LOCK, 0); | |
66dabbb6 CH |
404 | if (!IS_ERR(folio)) { |
405 | if (!truncate_inode_partial_folio(folio, lstart, lend)) | |
406 | end = folio->index; | |
407 | folio_unlock(folio); | |
408 | folio_put(folio); | |
409 | } | |
5a720394 | 410 | } |
1da177e4 | 411 | |
b85e0eff | 412 | index = start; |
b9a8a419 | 413 | while (index < end) { |
1da177e4 | 414 | cond_resched(); |
9fb6beea | 415 | if (!find_get_entries(mapping, &index, end - 1, &fbatch, |
38cefeb3 | 416 | indices)) { |
792ceaef | 417 | /* If all gone from start onwards, we're done */ |
b85e0eff | 418 | if (index == start) |
1da177e4 | 419 | break; |
792ceaef | 420 | /* Otherwise restart to make sure all gone */ |
b85e0eff | 421 | index = start; |
1da177e4 LT |
422 | continue; |
423 | } | |
f2187599 | 424 | |
0e499ed3 MWO |
425 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
426 | struct folio *folio = fbatch.folios[i]; | |
1da177e4 | 427 | |
acc53a0b | 428 | /* We rely upon deletion not changing folio->index */ |
b85e0eff | 429 | |
0e499ed3 | 430 | if (xa_is_value(folio)) |
0cd6144a | 431 | continue; |
0cd6144a | 432 | |
1e84a3d9 | 433 | folio_lock(folio); |
9fb6beea | 434 | VM_BUG_ON_FOLIO(!folio_contains(folio, indices[i]), folio); |
1e84a3d9 MWO |
435 | folio_wait_writeback(folio); |
436 | truncate_inode_folio(mapping, folio); | |
437 | folio_unlock(folio); | |
1da177e4 | 438 | } |
0e499ed3 MWO |
439 | truncate_folio_batch_exceptionals(mapping, &fbatch, indices); |
440 | folio_batch_release(&fbatch); | |
1da177e4 LT |
441 | } |
442 | } | |
d7339071 | 443 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 444 | |
d7339071 HR |
445 | /** |
446 | * truncate_inode_pages - truncate *all* the pages from an offset | |
447 | * @mapping: mapping to truncate | |
448 | * @lstart: offset from which to truncate | |
449 | * | |
730633f0 JK |
450 | * Called under (and serialised by) inode->i_rwsem and |
451 | * mapping->invalidate_lock. | |
08142579 JK |
452 | * |
453 | * Note: When this function returns, there can be a page in the process of | |
6ffcd825 | 454 | * deletion (inside __filemap_remove_folio()) in the specified range. Thus |
08142579 JK |
455 | * mapping->nrpages can be non-zero when this function returns even after |
456 | * truncation of the whole mapping. | |
d7339071 HR |
457 | */ |
458 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
459 | { | |
460 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
461 | } | |
1da177e4 LT |
462 | EXPORT_SYMBOL(truncate_inode_pages); |
463 | ||
91b0abe3 JW |
464 | /** |
465 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
466 | * @mapping: mapping to truncate | |
467 | * | |
9608703e | 468 | * Called under (and serialized by) inode->i_rwsem. |
91b0abe3 JW |
469 | * |
470 | * Filesystems have to use this in the .evict_inode path to inform the | |
471 | * VM that this is the final truncate and the inode is going away. | |
472 | */ | |
473 | void truncate_inode_pages_final(struct address_space *mapping) | |
474 | { | |
91b0abe3 JW |
475 | /* |
476 | * Page reclaim can not participate in regular inode lifetime | |
477 | * management (can't call iput()) and thus can race with the | |
478 | * inode teardown. Tell it when the address space is exiting, | |
479 | * so that it does not install eviction information after the | |
480 | * final truncate has begun. | |
481 | */ | |
482 | mapping_set_exiting(mapping); | |
483 | ||
7716506a | 484 | if (!mapping_empty(mapping)) { |
91b0abe3 JW |
485 | /* |
486 | * As truncation uses a lockless tree lookup, cycle | |
487 | * the tree lock to make sure any ongoing tree | |
488 | * modification that does not see AS_EXITING is | |
489 | * completed before starting the final truncate. | |
490 | */ | |
b93b0163 MW |
491 | xa_lock_irq(&mapping->i_pages); |
492 | xa_unlock_irq(&mapping->i_pages); | |
91b0abe3 | 493 | } |
6ff38bd4 | 494 | |
6ff38bd4 | 495 | truncate_inode_pages(mapping, 0); |
91b0abe3 JW |
496 | } |
497 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
498 | ||
c56109dd | 499 | /** |
1a0fc811 MWO |
500 | * mapping_try_invalidate - Invalidate all the evictable folios of one inode |
501 | * @mapping: the address_space which holds the folios to invalidate | |
c56109dd MWO |
502 | * @start: the offset 'from' which to invalidate |
503 | * @end: the offset 'to' which to invalidate (inclusive) | |
1a0fc811 | 504 | * @nr_failed: How many folio invalidations failed |
c56109dd | 505 | * |
1a0fc811 MWO |
506 | * This function is similar to invalidate_mapping_pages(), except that it |
507 | * returns the number of folios which could not be evicted in @nr_failed. | |
c56109dd | 508 | */ |
1a0fc811 MWO |
509 | unsigned long mapping_try_invalidate(struct address_space *mapping, |
510 | pgoff_t start, pgoff_t end, unsigned long *nr_failed) | |
1da177e4 | 511 | { |
0cd6144a | 512 | pgoff_t indices[PAGEVEC_SIZE]; |
51dcbdac | 513 | struct folio_batch fbatch; |
b85e0eff | 514 | pgoff_t index = start; |
31560180 MK |
515 | unsigned long ret; |
516 | unsigned long count = 0; | |
1da177e4 LT |
517 | int i; |
518 | ||
51dcbdac | 519 | folio_batch_init(&fbatch); |
3392ca12 | 520 | while (find_lock_entries(mapping, &index, end, &fbatch, indices)) { |
5f5a3e95 | 521 | bool xa_has_values = false; |
d3db2c04 SB |
522 | int nr = folio_batch_count(&fbatch); |
523 | ||
524 | for (i = 0; i < nr; i++) { | |
b4545f46 | 525 | struct folio *folio = fbatch.folios[i]; |
e0f23603 | 526 | |
b4545f46 | 527 | /* We rely upon deletion not changing folio->index */ |
e0f23603 | 528 | |
b4545f46 | 529 | if (xa_is_value(folio)) { |
61c663e0 YZ |
530 | xa_has_values = true; |
531 | count++; | |
0cd6144a JW |
532 | continue; |
533 | } | |
fc127da0 | 534 | |
b4545f46 MWO |
535 | ret = mapping_evict_folio(mapping, folio); |
536 | folio_unlock(folio); | |
31560180 | 537 | /* |
b4545f46 | 538 | * Invalidation is a hint that the folio is no longer |
31560180 MK |
539 | * of interest and try to speed up its reclaim. |
540 | */ | |
eb1d7a65 | 541 | if (!ret) { |
261b6840 | 542 | deactivate_file_folio(folio); |
1a0fc811 MWO |
543 | /* Likely in the lru cache of a remote CPU */ |
544 | if (nr_failed) | |
545 | (*nr_failed)++; | |
eb1d7a65 | 546 | } |
31560180 | 547 | count += ret; |
1da177e4 | 548 | } |
61c663e0 YZ |
549 | |
550 | if (xa_has_values) | |
d3db2c04 | 551 | clear_shadow_entries(mapping, indices[0], indices[nr-1]); |
61c663e0 | 552 | |
51dcbdac MWO |
553 | folio_batch_remove_exceptionals(&fbatch); |
554 | folio_batch_release(&fbatch); | |
28697355 | 555 | cond_resched(); |
1da177e4 | 556 | } |
31560180 | 557 | return count; |
1da177e4 | 558 | } |
eb1d7a65 YS |
559 | |
560 | /** | |
7ae12c80 JW |
561 | * invalidate_mapping_pages - Invalidate all clean, unlocked cache of one inode |
562 | * @mapping: the address_space which holds the cache to invalidate | |
eb1d7a65 YS |
563 | * @start: the offset 'from' which to invalidate |
564 | * @end: the offset 'to' which to invalidate (inclusive) | |
565 | * | |
7ae12c80 JW |
566 | * This function removes pages that are clean, unmapped and unlocked, |
567 | * as well as shadow entries. It will not block on IO activity. | |
eb1d7a65 | 568 | * |
7ae12c80 JW |
569 | * If you want to remove all the pages of one inode, regardless of |
570 | * their use and writeback state, use truncate_inode_pages(). | |
eb1d7a65 | 571 | * |
1a0fc811 | 572 | * Return: The number of indices that had their contents invalidated |
eb1d7a65 YS |
573 | */ |
574 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
575 | pgoff_t start, pgoff_t end) | |
576 | { | |
1a0fc811 | 577 | return mapping_try_invalidate(mapping, start, end, NULL); |
eb1d7a65 | 578 | } |
54bc4855 | 579 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 580 | |
4a9e2315 JA |
581 | static int folio_launder(struct address_space *mapping, struct folio *folio) |
582 | { | |
583 | if (!folio_test_dirty(folio)) | |
584 | return 0; | |
585 | if (folio->mapping != mapping || mapping->a_ops->launder_folio == NULL) | |
586 | return 0; | |
587 | return mapping->a_ops->launder_folio(folio); | |
588 | } | |
589 | ||
bd4c8ce4 | 590 | /* |
2033c98c | 591 | * This is like mapping_evict_folio(), except it ignores the folio's |
bd4c8ce4 | 592 | * refcount. We do this because invalidate_inode_pages2() needs stronger |
2033c98c | 593 | * invalidation guarantees, and cannot afford to leave folios behind because |
0ba5e806 | 594 | * shrink_folio_list() has a temp ref on them, or because they're transiently |
1fec6890 | 595 | * sitting in the folio_add_lru() caches. |
bd4c8ce4 | 596 | */ |
4a9e2315 JA |
597 | int folio_unmap_invalidate(struct address_space *mapping, struct folio *folio, |
598 | gfp_t gfp) | |
bd4c8ce4 | 599 | { |
4a9e2315 JA |
600 | int ret; |
601 | ||
602 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); | |
bd4c8ce4 | 603 | |
4a9e2315 JA |
604 | if (folio_mapped(folio)) |
605 | unmap_mapping_folio(folio); | |
606 | BUG_ON(folio_mapped(folio)); | |
607 | ||
608 | ret = folio_launder(mapping, folio); | |
609 | if (ret) | |
610 | return ret; | |
611 | if (folio->mapping != mapping) | |
612 | return -EBUSY; | |
613 | if (!filemap_release_folio(folio, gfp)) | |
614 | return -EBUSY; | |
bd4c8ce4 | 615 | |
51b8c1fe | 616 | spin_lock(&mapping->host->i_lock); |
30472509 | 617 | xa_lock_irq(&mapping->i_pages); |
78f42660 | 618 | if (folio_test_dirty(folio)) |
bd4c8ce4 AM |
619 | goto failed; |
620 | ||
78f42660 MWO |
621 | BUG_ON(folio_has_private(folio)); |
622 | __filemap_remove_folio(folio, NULL); | |
30472509 | 623 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
624 | if (mapping_shrinkable(mapping)) |
625 | inode_add_lru(mapping->host); | |
626 | spin_unlock(&mapping->host->i_lock); | |
6072d13c | 627 | |
78f42660 | 628 | filemap_free_folio(mapping, folio); |
bd4c8ce4 AM |
629 | return 1; |
630 | failed: | |
30472509 | 631 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe | 632 | spin_unlock(&mapping->host->i_lock); |
4a9e2315 | 633 | return -EBUSY; |
e3db7691 TM |
634 | } |
635 | ||
1da177e4 LT |
636 | /** |
637 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 638 | * @mapping: the address_space |
1da177e4 LT |
639 | * @start: the page offset 'from' which to invalidate |
640 | * @end: the page offset 'to' which to invalidate (inclusive) | |
641 | * | |
642 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
643 | * invalidation. | |
644 | * | |
a862f68a | 645 | * Return: -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
646 | */ |
647 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
648 | pgoff_t start, pgoff_t end) | |
649 | { | |
0cd6144a | 650 | pgoff_t indices[PAGEVEC_SIZE]; |
0e499ed3 | 651 | struct folio_batch fbatch; |
b85e0eff | 652 | pgoff_t index; |
1da177e4 LT |
653 | int i; |
654 | int ret = 0; | |
0dd1334f | 655 | int ret2 = 0; |
1da177e4 | 656 | int did_range_unmap = 0; |
1da177e4 | 657 | |
7716506a | 658 | if (mapping_empty(mapping)) |
0a4ee518 | 659 | return 0; |
32691f0f | 660 | |
0e499ed3 | 661 | folio_batch_init(&fbatch); |
b85e0eff | 662 | index = start; |
9fb6beea | 663 | while (find_get_entries(mapping, &index, end, &fbatch, indices)) { |
5f5a3e95 | 664 | bool xa_has_values = false; |
d3db2c04 SB |
665 | int nr = folio_batch_count(&fbatch); |
666 | ||
667 | for (i = 0; i < nr; i++) { | |
0e499ed3 | 668 | struct folio *folio = fbatch.folios[i]; |
b85e0eff | 669 | |
fae9bc4a | 670 | /* We rely upon deletion not changing folio->index */ |
1da177e4 | 671 | |
0e499ed3 | 672 | if (xa_is_value(folio)) { |
61c663e0 YZ |
673 | xa_has_values = true; |
674 | if (dax_mapping(mapping) && | |
675 | !dax_invalidate_mapping_entry_sync(mapping, indices[i])) | |
c6dcf52c | 676 | ret = -EBUSY; |
0cd6144a JW |
677 | continue; |
678 | } | |
679 | ||
fae9bc4a | 680 | if (!did_range_unmap && folio_mapped(folio)) { |
22061a1f | 681 | /* |
fae9bc4a | 682 | * If folio is mapped, before taking its lock, |
22061a1f HD |
683 | * zap the rest of the file in one hit. |
684 | */ | |
9fb6beea VMO |
685 | unmap_mapping_pages(mapping, indices[i], |
686 | (1 + end - indices[i]), false); | |
22061a1f HD |
687 | did_range_unmap = 1; |
688 | } | |
689 | ||
fae9bc4a | 690 | folio_lock(folio); |
aa5b9178 | 691 | if (unlikely(folio->mapping != mapping)) { |
fae9bc4a | 692 | folio_unlock(folio); |
1da177e4 LT |
693 | continue; |
694 | } | |
aa5b9178 | 695 | VM_BUG_ON_FOLIO(!folio_contains(folio, indices[i]), folio); |
fae9bc4a | 696 | folio_wait_writeback(folio); |
4a9e2315 | 697 | ret2 = folio_unmap_invalidate(mapping, folio, GFP_KERNEL); |
0dd1334f HH |
698 | if (ret2 < 0) |
699 | ret = ret2; | |
fae9bc4a | 700 | folio_unlock(folio); |
1da177e4 | 701 | } |
61c663e0 YZ |
702 | |
703 | if (xa_has_values) | |
d3db2c04 | 704 | clear_shadow_entries(mapping, indices[0], indices[nr-1]); |
61c663e0 | 705 | |
0e499ed3 MWO |
706 | folio_batch_remove_exceptionals(&fbatch); |
707 | folio_batch_release(&fbatch); | |
1da177e4 LT |
708 | cond_resched(); |
709 | } | |
cd656375 | 710 | /* |
69b6c131 | 711 | * For DAX we invalidate page tables after invalidating page cache. We |
cd656375 JK |
712 | * could invalidate page tables while invalidating each entry however |
713 | * that would be expensive. And doing range unmapping before doesn't | |
69b6c131 | 714 | * work as we have no cheap way to find whether page cache entry didn't |
cd656375 JK |
715 | * get remapped later. |
716 | */ | |
717 | if (dax_mapping(mapping)) { | |
977fbdcd | 718 | unmap_mapping_pages(mapping, start, end - start + 1, false); |
cd656375 | 719 | } |
1da177e4 LT |
720 | return ret; |
721 | } | |
722 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
723 | ||
724 | /** | |
725 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 726 | * @mapping: the address_space |
1da177e4 LT |
727 | * |
728 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
729 | * invalidation. | |
730 | * | |
a862f68a | 731 | * Return: -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
732 | */ |
733 | int invalidate_inode_pages2(struct address_space *mapping) | |
734 | { | |
735 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
736 | } | |
737 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 738 | |
739 | /** | |
740 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
741 | * @inode: inode | |
8a549bea | 742 | * @newsize: new file size |
25d9e2d1 | 743 | * |
744 | * inode's new i_size must already be written before truncate_pagecache | |
745 | * is called. | |
746 | * | |
747 | * This function should typically be called before the filesystem | |
748 | * releases resources associated with the freed range (eg. deallocates | |
749 | * blocks). This way, pagecache will always stay logically coherent | |
750 | * with on-disk format, and the filesystem would not have to deal with | |
751 | * situations such as writepage being called for a page that has already | |
752 | * had its underlying blocks deallocated. | |
753 | */ | |
7caef267 | 754 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 755 | { |
cedabed4 | 756 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 757 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
758 | |
759 | /* | |
760 | * unmap_mapping_range is called twice, first simply for | |
761 | * efficiency so that truncate_inode_pages does fewer | |
762 | * single-page unmaps. However after this first call, and | |
763 | * before truncate_inode_pages finishes, it is possible for | |
764 | * private pages to be COWed, which remain after | |
765 | * truncate_inode_pages finishes, hence the second | |
766 | * unmap_mapping_range call must be made for correctness. | |
767 | */ | |
8a549bea HD |
768 | unmap_mapping_range(mapping, holebegin, 0, 1); |
769 | truncate_inode_pages(mapping, newsize); | |
770 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 771 | } |
772 | EXPORT_SYMBOL(truncate_pagecache); | |
773 | ||
2c27c65e CH |
774 | /** |
775 | * truncate_setsize - update inode and pagecache for a new file size | |
776 | * @inode: inode | |
777 | * @newsize: new file size | |
778 | * | |
382e27da JK |
779 | * truncate_setsize updates i_size and performs pagecache truncation (if |
780 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
781 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 782 | * |
77783d06 | 783 | * Must be called with a lock serializing truncates and writes (generally |
9608703e | 784 | * i_rwsem but e.g. xfs uses a different lock) and before all filesystem |
77783d06 | 785 | * specific block truncation has been performed. |
2c27c65e CH |
786 | */ |
787 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
788 | { | |
90a80202 JK |
789 | loff_t oldsize = inode->i_size; |
790 | ||
2c27c65e | 791 | i_size_write(inode, newsize); |
90a80202 JK |
792 | if (newsize > oldsize) |
793 | pagecache_isize_extended(inode, oldsize, newsize); | |
7caef267 | 794 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
795 | } |
796 | EXPORT_SYMBOL(truncate_setsize); | |
797 | ||
90a80202 JK |
798 | /** |
799 | * pagecache_isize_extended - update pagecache after extension of i_size | |
800 | * @inode: inode for which i_size was extended | |
801 | * @from: original inode size | |
802 | * @to: new inode size | |
803 | * | |
2ebe90da MWO |
804 | * Handle extension of inode size either caused by extending truncate or |
805 | * by write starting after current i_size. We mark the page straddling | |
806 | * current i_size RO so that page_mkwrite() is called on the first | |
807 | * write access to the page. The filesystem will update its per-block | |
808 | * information before user writes to the page via mmap after the i_size | |
809 | * has been changed. | |
90a80202 JK |
810 | * |
811 | * The function must be called after i_size is updated so that page fault | |
2ebe90da | 812 | * coming after we unlock the folio will already see the new i_size. |
9608703e | 813 | * The function must be called while we still hold i_rwsem - this not only |
90a80202 JK |
814 | * makes sure i_size is stable but also that userspace cannot observe new |
815 | * i_size value before we are prepared to store mmap writes at new inode size. | |
816 | */ | |
817 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
818 | { | |
93407472 | 819 | int bsize = i_blocksize(inode); |
90a80202 | 820 | loff_t rounded_from; |
2ebe90da | 821 | struct folio *folio; |
90a80202 | 822 | |
90a80202 JK |
823 | WARN_ON(to > inode->i_size); |
824 | ||
2ebe90da | 825 | if (from >= to || bsize >= PAGE_SIZE) |
90a80202 JK |
826 | return; |
827 | /* Page straddling @from will not have any hole block created? */ | |
828 | rounded_from = round_up(from, bsize); | |
09cbfeaf | 829 | if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) |
90a80202 JK |
830 | return; |
831 | ||
2ebe90da MWO |
832 | folio = filemap_lock_folio(inode->i_mapping, from / PAGE_SIZE); |
833 | /* Folio not cached? Nothing to do */ | |
834 | if (IS_ERR(folio)) | |
90a80202 JK |
835 | return; |
836 | /* | |
2ebe90da | 837 | * See folio_clear_dirty_for_io() for details why folio_mark_dirty() |
90a80202 JK |
838 | * is needed. |
839 | */ | |
2ebe90da MWO |
840 | if (folio_mkclean(folio)) |
841 | folio_mark_dirty(folio); | |
52aecaee BF |
842 | |
843 | /* | |
844 | * The post-eof range of the folio must be zeroed before it is exposed | |
845 | * to the file. Writeback normally does this, but since i_size has been | |
846 | * increased we handle it here. | |
847 | */ | |
848 | if (folio_test_dirty(folio)) { | |
849 | unsigned int offset, end; | |
850 | ||
851 | offset = from - folio_pos(folio); | |
852 | end = min_t(unsigned int, to - folio_pos(folio), | |
853 | folio_size(folio)); | |
854 | folio_zero_segment(folio, offset, end); | |
855 | } | |
856 | ||
2ebe90da MWO |
857 | folio_unlock(folio); |
858 | folio_put(folio); | |
90a80202 JK |
859 | } |
860 | EXPORT_SYMBOL(pagecache_isize_extended); | |
861 | ||
623e3db9 HD |
862 | /** |
863 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
864 | * @inode: inode | |
865 | * @lstart: offset of beginning of hole | |
866 | * @lend: offset of last byte of hole | |
867 | * | |
868 | * This function should typically be called before the filesystem | |
869 | * releases resources associated with the freed range (eg. deallocates | |
870 | * blocks). This way, pagecache will always stay logically coherent | |
871 | * with on-disk format, and the filesystem would not have to deal with | |
872 | * situations such as writepage being called for a page that has already | |
873 | * had its underlying blocks deallocated. | |
874 | */ | |
875 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
876 | { | |
877 | struct address_space *mapping = inode->i_mapping; | |
878 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
879 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
880 | /* | |
881 | * This rounding is currently just for example: unmap_mapping_range | |
882 | * expands its hole outwards, whereas we want it to contract the hole | |
883 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
884 | * doing their own page rounding first. Note that unmap_mapping_range |
885 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
886 | */ |
887 | ||
888 | /* | |
889 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
890 | * once (before truncating pagecache), and without "even_cows" flag: | |
891 | * hole-punching should not remove private COWed pages from the hole. | |
892 | */ | |
893 | if ((u64)unmap_end > (u64)unmap_start) | |
894 | unmap_mapping_range(mapping, unmap_start, | |
895 | 1 + unmap_end - unmap_start, 0); | |
896 | truncate_inode_pages_range(mapping, lstart, lend); | |
897 | } | |
898 | EXPORT_SYMBOL(truncate_pagecache_range); |