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); |
ac401cc7 | 43 | mapping->nrexceptional--; |
f2187599 MG |
44 | } |
45 | ||
46 | static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, | |
47 | void *entry) | |
48 | { | |
b93b0163 | 49 | xa_lock_irq(&mapping->i_pages); |
f2187599 | 50 | __clear_shadow_entry(mapping, index, entry); |
b93b0163 | 51 | xa_unlock_irq(&mapping->i_pages); |
0cd6144a | 52 | } |
1da177e4 | 53 | |
c6dcf52c | 54 | /* |
f2187599 MG |
55 | * Unconditionally remove exceptional entries. Usually called from truncate |
56 | * path. Note that the pagevec may be altered by this function by removing | |
57 | * exceptional entries similar to what pagevec_remove_exceptionals does. | |
c6dcf52c | 58 | */ |
f2187599 MG |
59 | static void truncate_exceptional_pvec_entries(struct address_space *mapping, |
60 | struct pagevec *pvec, pgoff_t *indices, | |
61 | pgoff_t end) | |
c6dcf52c | 62 | { |
f2187599 MG |
63 | int i, j; |
64 | bool dax, lock; | |
65 | ||
c6dcf52c JK |
66 | /* Handled by shmem itself */ |
67 | if (shmem_mapping(mapping)) | |
68 | return; | |
69 | ||
f2187599 | 70 | for (j = 0; j < pagevec_count(pvec); j++) |
3159f943 | 71 | if (xa_is_value(pvec->pages[j])) |
f2187599 MG |
72 | break; |
73 | ||
74 | if (j == pagevec_count(pvec)) | |
c6dcf52c | 75 | return; |
f2187599 MG |
76 | |
77 | dax = dax_mapping(mapping); | |
78 | lock = !dax && indices[j] < end; | |
79 | if (lock) | |
b93b0163 | 80 | xa_lock_irq(&mapping->i_pages); |
f2187599 MG |
81 | |
82 | for (i = j; i < pagevec_count(pvec); i++) { | |
83 | struct page *page = pvec->pages[i]; | |
84 | pgoff_t index = indices[i]; | |
85 | ||
3159f943 | 86 | if (!xa_is_value(page)) { |
f2187599 MG |
87 | pvec->pages[j++] = page; |
88 | continue; | |
89 | } | |
90 | ||
91 | if (index >= end) | |
92 | continue; | |
93 | ||
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 MG |
101 | |
102 | if (lock) | |
b93b0163 | 103 | xa_unlock_irq(&mapping->i_pages); |
f2187599 | 104 | pvec->nr = j; |
c6dcf52c JK |
105 | } |
106 | ||
107 | /* | |
108 | * Invalidate exceptional entry if easily possible. This handles exceptional | |
4636e70b | 109 | * entries for invalidate_inode_pages(). |
c6dcf52c JK |
110 | */ |
111 | static int invalidate_exceptional_entry(struct address_space *mapping, | |
112 | pgoff_t index, void *entry) | |
113 | { | |
4636e70b RZ |
114 | /* Handled by shmem itself, or for DAX we do nothing. */ |
115 | if (shmem_mapping(mapping) || dax_mapping(mapping)) | |
c6dcf52c | 116 | return 1; |
c6dcf52c JK |
117 | clear_shadow_entry(mapping, index, entry); |
118 | return 1; | |
119 | } | |
120 | ||
121 | /* | |
122 | * Invalidate exceptional entry if clean. This handles exceptional entries for | |
123 | * invalidate_inode_pages2() so for DAX it evicts only clean entries. | |
124 | */ | |
125 | static int invalidate_exceptional_entry2(struct address_space *mapping, | |
126 | pgoff_t index, void *entry) | |
127 | { | |
128 | /* Handled by shmem itself */ | |
129 | if (shmem_mapping(mapping)) | |
130 | return 1; | |
131 | if (dax_mapping(mapping)) | |
132 | return dax_invalidate_mapping_entry_sync(mapping, index); | |
133 | clear_shadow_entry(mapping, index, entry); | |
134 | return 1; | |
135 | } | |
136 | ||
cf9a2ae8 | 137 | /** |
28bc44d7 | 138 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 139 | * @page: the page which is affected |
d47992f8 LC |
140 | * @offset: start of the range to invalidate |
141 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
142 | * |
143 | * do_invalidatepage() is called when all or part of the page has become | |
144 | * invalidated by a truncate operation. | |
145 | * | |
146 | * do_invalidatepage() does not have to release all buffers, but it must | |
147 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
148 | * is underway against any of the blocks which are outside the truncation | |
149 | * point. Because the caller is about to free (and possibly reuse) those | |
150 | * blocks on-disk. | |
151 | */ | |
d47992f8 LC |
152 | void do_invalidatepage(struct page *page, unsigned int offset, |
153 | unsigned int length) | |
cf9a2ae8 | 154 | { |
d47992f8 LC |
155 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
156 | ||
cf9a2ae8 | 157 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 158 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
159 | if (!invalidatepage) |
160 | invalidatepage = block_invalidatepage; | |
9361401e | 161 | #endif |
cf9a2ae8 | 162 | if (invalidatepage) |
d47992f8 | 163 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
164 | } |
165 | ||
1da177e4 LT |
166 | /* |
167 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 168 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 169 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 | 170 | * |
fc3a5ac5 | 171 | * We need to bail out if page->mapping is no longer equal to the original |
1da177e4 | 172 | * mapping. This happens a) when the VM reclaimed the page while we waited on |
fc0ecff6 | 173 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
174 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
175 | */ | |
9f4e41f4 JK |
176 | static void |
177 | truncate_cleanup_page(struct address_space *mapping, struct page *page) | |
1da177e4 | 178 | { |
9f4e41f4 | 179 | if (page_mapped(page)) { |
fc3a5ac5 | 180 | unsigned int nr = thp_nr_pages(page); |
977fbdcd | 181 | unmap_mapping_pages(mapping, page->index, nr, false); |
9f4e41f4 | 182 | } |
1da177e4 | 183 | |
266cf658 | 184 | if (page_has_private(page)) |
fc3a5ac5 | 185 | do_invalidatepage(page, 0, thp_size(page)); |
1da177e4 | 186 | |
b9ea2515 KK |
187 | /* |
188 | * Some filesystems seem to re-dirty the page even after | |
189 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
190 | * Hence dirty accounting check is placed after invalidation. | |
191 | */ | |
11f81bec | 192 | cancel_dirty_page(page); |
1da177e4 | 193 | ClearPageMappedToDisk(page); |
1da177e4 LT |
194 | } |
195 | ||
196 | /* | |
fc0ecff6 | 197 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 198 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
199 | * be marked dirty at any time too, so use remove_mapping which safely |
200 | * discards clean, unused pages. | |
1da177e4 LT |
201 | * |
202 | * Returns non-zero if the page was successfully invalidated. | |
203 | */ | |
204 | static int | |
205 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
206 | { | |
0fd0e6b0 NP |
207 | int ret; |
208 | ||
1da177e4 LT |
209 | if (page->mapping != mapping) |
210 | return 0; | |
211 | ||
266cf658 | 212 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
213 | return 0; |
214 | ||
0fd0e6b0 | 215 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
216 | |
217 | return ret; | |
1da177e4 LT |
218 | } |
219 | ||
750b4987 NP |
220 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
221 | { | |
fc127da0 KS |
222 | VM_BUG_ON_PAGE(PageTail(page), page); |
223 | ||
9f4e41f4 JK |
224 | if (page->mapping != mapping) |
225 | return -EIO; | |
226 | ||
227 | truncate_cleanup_page(mapping, page); | |
228 | delete_from_page_cache(page); | |
229 | return 0; | |
750b4987 NP |
230 | } |
231 | ||
25718736 AK |
232 | /* |
233 | * Used to get rid of pages on hardware memory corruption. | |
234 | */ | |
235 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
236 | { | |
237 | if (!mapping) | |
238 | return -EINVAL; | |
239 | /* | |
240 | * Only punch for normal data pages for now. | |
241 | * Handling other types like directories would need more auditing. | |
242 | */ | |
243 | if (!S_ISREG(mapping->host->i_mode)) | |
244 | return -EIO; | |
245 | return truncate_inode_page(mapping, page); | |
246 | } | |
247 | EXPORT_SYMBOL(generic_error_remove_page); | |
248 | ||
83f78668 WF |
249 | /* |
250 | * Safely invalidate one page from its pagecache mapping. | |
251 | * It only drops clean, unused pages. The page must be locked. | |
252 | * | |
253 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
254 | */ | |
255 | int invalidate_inode_page(struct page *page) | |
256 | { | |
257 | struct address_space *mapping = page_mapping(page); | |
258 | if (!mapping) | |
259 | return 0; | |
260 | if (PageDirty(page) || PageWriteback(page)) | |
261 | return 0; | |
262 | if (page_mapped(page)) | |
263 | return 0; | |
264 | return invalidate_complete_page(mapping, page); | |
265 | } | |
266 | ||
1da177e4 | 267 | /** |
73c1e204 | 268 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
269 | * @mapping: mapping to truncate |
270 | * @lstart: offset from which to truncate | |
5a720394 | 271 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 272 | * |
d7339071 | 273 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
274 | * specified offsets (and zeroing out partial pages |
275 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
276 | * |
277 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
278 | * block on page locks and it will not block on writeback. The second pass | |
279 | * will wait. This is to prevent as much IO as possible in the affected region. | |
280 | * The first pass will remove most pages, so the search cost of the second pass | |
281 | * is low. | |
282 | * | |
1da177e4 LT |
283 | * We pass down the cache-hot hint to the page freeing code. Even if the |
284 | * mapping is large, it is probably the case that the final pages are the most | |
285 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
286 | * |
287 | * Note that since ->invalidatepage() accepts range to invalidate | |
288 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
289 | * page aligned properly. | |
1da177e4 | 290 | */ |
d7339071 HR |
291 | void truncate_inode_pages_range(struct address_space *mapping, |
292 | loff_t lstart, loff_t lend) | |
1da177e4 | 293 | { |
5a720394 LC |
294 | pgoff_t start; /* inclusive */ |
295 | pgoff_t end; /* exclusive */ | |
296 | unsigned int partial_start; /* inclusive */ | |
297 | unsigned int partial_end; /* exclusive */ | |
298 | struct pagevec pvec; | |
0cd6144a | 299 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
300 | pgoff_t index; |
301 | int i; | |
1da177e4 | 302 | |
f9fe48be | 303 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
34ccb69e | 304 | goto out; |
1da177e4 | 305 | |
5a720394 | 306 | /* Offsets within partial pages */ |
09cbfeaf KS |
307 | partial_start = lstart & (PAGE_SIZE - 1); |
308 | partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
5a720394 LC |
309 | |
310 | /* | |
311 | * 'start' and 'end' always covers the range of pages to be fully | |
312 | * truncated. Partial pages are covered with 'partial_start' at the | |
313 | * start of the range and 'partial_end' at the end of the range. | |
314 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
315 | */ | |
09cbfeaf | 316 | start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
5a720394 LC |
317 | if (lend == -1) |
318 | /* | |
319 | * lend == -1 indicates end-of-file so we have to set 'end' | |
320 | * to the highest possible pgoff_t and since the type is | |
321 | * unsigned we're using -1. | |
322 | */ | |
323 | end = -1; | |
324 | else | |
09cbfeaf | 325 | end = (lend + 1) >> PAGE_SHIFT; |
d7339071 | 326 | |
86679820 | 327 | pagevec_init(&pvec); |
b85e0eff | 328 | index = start; |
0cd6144a JW |
329 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
330 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
331 | indices)) { | |
aa65c29c JK |
332 | /* |
333 | * Pagevec array has exceptional entries and we may also fail | |
334 | * to lock some pages. So we store pages that can be deleted | |
335 | * in a new pagevec. | |
336 | */ | |
337 | struct pagevec locked_pvec; | |
338 | ||
86679820 | 339 | pagevec_init(&locked_pvec); |
1da177e4 LT |
340 | for (i = 0; i < pagevec_count(&pvec); i++) { |
341 | struct page *page = pvec.pages[i]; | |
1da177e4 | 342 | |
b85e0eff | 343 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 344 | index = indices[i]; |
5a720394 | 345 | if (index >= end) |
d7339071 | 346 | break; |
d7339071 | 347 | |
3159f943 | 348 | if (xa_is_value(page)) |
0cd6144a | 349 | continue; |
0cd6144a | 350 | |
529ae9aa | 351 | if (!trylock_page(page)) |
1da177e4 | 352 | continue; |
5cbc198a | 353 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
354 | if (PageWriteback(page)) { |
355 | unlock_page(page); | |
356 | continue; | |
357 | } | |
aa65c29c JK |
358 | if (page->mapping != mapping) { |
359 | unlock_page(page); | |
360 | continue; | |
361 | } | |
362 | pagevec_add(&locked_pvec, page); | |
1da177e4 | 363 | } |
aa65c29c JK |
364 | for (i = 0; i < pagevec_count(&locked_pvec); i++) |
365 | truncate_cleanup_page(mapping, locked_pvec.pages[i]); | |
366 | delete_from_page_cache_batch(mapping, &locked_pvec); | |
367 | for (i = 0; i < pagevec_count(&locked_pvec); i++) | |
368 | unlock_page(locked_pvec.pages[i]); | |
f2187599 | 369 | truncate_exceptional_pvec_entries(mapping, &pvec, indices, end); |
1da177e4 LT |
370 | pagevec_release(&pvec); |
371 | cond_resched(); | |
b85e0eff | 372 | index++; |
1da177e4 | 373 | } |
5a720394 | 374 | if (partial_start) { |
1da177e4 LT |
375 | struct page *page = find_lock_page(mapping, start - 1); |
376 | if (page) { | |
09cbfeaf | 377 | unsigned int top = PAGE_SIZE; |
5a720394 LC |
378 | if (start > end) { |
379 | /* Truncation within a single page */ | |
380 | top = partial_end; | |
381 | partial_end = 0; | |
382 | } | |
1da177e4 | 383 | wait_on_page_writeback(page); |
5a720394 LC |
384 | zero_user_segment(page, partial_start, top); |
385 | cleancache_invalidate_page(mapping, page); | |
386 | if (page_has_private(page)) | |
387 | do_invalidatepage(page, partial_start, | |
388 | top - partial_start); | |
1da177e4 | 389 | unlock_page(page); |
09cbfeaf | 390 | put_page(page); |
1da177e4 LT |
391 | } |
392 | } | |
5a720394 LC |
393 | if (partial_end) { |
394 | struct page *page = find_lock_page(mapping, end); | |
395 | if (page) { | |
396 | wait_on_page_writeback(page); | |
397 | zero_user_segment(page, 0, partial_end); | |
398 | cleancache_invalidate_page(mapping, page); | |
399 | if (page_has_private(page)) | |
400 | do_invalidatepage(page, 0, | |
401 | partial_end); | |
402 | unlock_page(page); | |
09cbfeaf | 403 | put_page(page); |
5a720394 LC |
404 | } |
405 | } | |
406 | /* | |
407 | * If the truncation happened within a single page no pages | |
408 | * will be released, just zeroed, so we can bail out now. | |
409 | */ | |
410 | if (start >= end) | |
34ccb69e | 411 | goto out; |
1da177e4 | 412 | |
b85e0eff | 413 | index = start; |
1da177e4 LT |
414 | for ( ; ; ) { |
415 | cond_resched(); | |
0cd6144a | 416 | if (!pagevec_lookup_entries(&pvec, mapping, index, |
792ceaef HD |
417 | min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { |
418 | /* If all gone from start onwards, we're done */ | |
b85e0eff | 419 | if (index == start) |
1da177e4 | 420 | break; |
792ceaef | 421 | /* Otherwise restart to make sure all gone */ |
b85e0eff | 422 | index = start; |
1da177e4 LT |
423 | continue; |
424 | } | |
0cd6144a | 425 | if (index == start && indices[0] >= end) { |
792ceaef | 426 | /* All gone out of hole to be punched, we're done */ |
0cd6144a | 427 | pagevec_remove_exceptionals(&pvec); |
d7339071 HR |
428 | pagevec_release(&pvec); |
429 | break; | |
430 | } | |
f2187599 | 431 | |
1da177e4 LT |
432 | for (i = 0; i < pagevec_count(&pvec); i++) { |
433 | struct page *page = pvec.pages[i]; | |
434 | ||
b85e0eff | 435 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 436 | index = indices[i]; |
792ceaef HD |
437 | if (index >= end) { |
438 | /* Restart punch to make sure all gone */ | |
439 | index = start - 1; | |
d7339071 | 440 | break; |
792ceaef | 441 | } |
b85e0eff | 442 | |
3159f943 | 443 | if (xa_is_value(page)) |
0cd6144a | 444 | continue; |
0cd6144a | 445 | |
1da177e4 | 446 | lock_page(page); |
5cbc198a | 447 | WARN_ON(page_to_index(page) != index); |
1da177e4 | 448 | wait_on_page_writeback(page); |
750b4987 | 449 | truncate_inode_page(mapping, page); |
1da177e4 LT |
450 | unlock_page(page); |
451 | } | |
f2187599 | 452 | truncate_exceptional_pvec_entries(mapping, &pvec, indices, end); |
1da177e4 | 453 | pagevec_release(&pvec); |
b85e0eff | 454 | index++; |
1da177e4 | 455 | } |
34ccb69e AR |
456 | |
457 | out: | |
3167760f | 458 | cleancache_invalidate_inode(mapping); |
1da177e4 | 459 | } |
d7339071 | 460 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 461 | |
d7339071 HR |
462 | /** |
463 | * truncate_inode_pages - truncate *all* the pages from an offset | |
464 | * @mapping: mapping to truncate | |
465 | * @lstart: offset from which to truncate | |
466 | * | |
1b1dcc1b | 467 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
468 | * |
469 | * Note: When this function returns, there can be a page in the process of | |
470 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
471 | * mapping->nrpages can be non-zero when this function returns even after | |
472 | * truncation of the whole mapping. | |
d7339071 HR |
473 | */ |
474 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
475 | { | |
476 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
477 | } | |
1da177e4 LT |
478 | EXPORT_SYMBOL(truncate_inode_pages); |
479 | ||
91b0abe3 JW |
480 | /** |
481 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
482 | * @mapping: mapping to truncate | |
483 | * | |
484 | * Called under (and serialized by) inode->i_mutex. | |
485 | * | |
486 | * Filesystems have to use this in the .evict_inode path to inform the | |
487 | * VM that this is the final truncate and the inode is going away. | |
488 | */ | |
489 | void truncate_inode_pages_final(struct address_space *mapping) | |
490 | { | |
f9fe48be | 491 | unsigned long nrexceptional; |
91b0abe3 JW |
492 | unsigned long nrpages; |
493 | ||
494 | /* | |
495 | * Page reclaim can not participate in regular inode lifetime | |
496 | * management (can't call iput()) and thus can race with the | |
497 | * inode teardown. Tell it when the address space is exiting, | |
498 | * so that it does not install eviction information after the | |
499 | * final truncate has begun. | |
500 | */ | |
501 | mapping_set_exiting(mapping); | |
502 | ||
503 | /* | |
504 | * When reclaim installs eviction entries, it increases | |
f9fe48be | 505 | * nrexceptional first, then decreases nrpages. Make sure we see |
91b0abe3 JW |
506 | * this in the right order or we might miss an entry. |
507 | */ | |
508 | nrpages = mapping->nrpages; | |
509 | smp_rmb(); | |
f9fe48be | 510 | nrexceptional = mapping->nrexceptional; |
91b0abe3 | 511 | |
f9fe48be | 512 | if (nrpages || nrexceptional) { |
91b0abe3 JW |
513 | /* |
514 | * As truncation uses a lockless tree lookup, cycle | |
515 | * the tree lock to make sure any ongoing tree | |
516 | * modification that does not see AS_EXITING is | |
517 | * completed before starting the final truncate. | |
518 | */ | |
b93b0163 MW |
519 | xa_lock_irq(&mapping->i_pages); |
520 | xa_unlock_irq(&mapping->i_pages); | |
91b0abe3 | 521 | } |
6ff38bd4 PT |
522 | |
523 | /* | |
524 | * Cleancache needs notification even if there are no pages or shadow | |
525 | * entries. | |
526 | */ | |
527 | truncate_inode_pages(mapping, 0); | |
91b0abe3 JW |
528 | } |
529 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
530 | ||
a77eedbc | 531 | static unsigned long __invalidate_mapping_pages(struct address_space *mapping, |
eb1d7a65 | 532 | pgoff_t start, pgoff_t end, unsigned long *nr_pagevec) |
1da177e4 | 533 | { |
0cd6144a | 534 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 535 | struct pagevec pvec; |
b85e0eff | 536 | pgoff_t index = start; |
31560180 MK |
537 | unsigned long ret; |
538 | unsigned long count = 0; | |
1da177e4 LT |
539 | int i; |
540 | ||
86679820 | 541 | pagevec_init(&pvec); |
0cd6144a JW |
542 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
543 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
544 | indices)) { | |
1da177e4 LT |
545 | for (i = 0; i < pagevec_count(&pvec); i++) { |
546 | struct page *page = pvec.pages[i]; | |
e0f23603 | 547 | |
b85e0eff | 548 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 549 | index = indices[i]; |
b85e0eff HD |
550 | if (index > end) |
551 | break; | |
e0f23603 | 552 | |
3159f943 | 553 | if (xa_is_value(page)) { |
c6dcf52c JK |
554 | invalidate_exceptional_entry(mapping, index, |
555 | page); | |
0cd6144a JW |
556 | continue; |
557 | } | |
558 | ||
b85e0eff HD |
559 | if (!trylock_page(page)) |
560 | continue; | |
fc127da0 | 561 | |
5cbc198a | 562 | WARN_ON(page_to_index(page) != index); |
fc127da0 KS |
563 | |
564 | /* Middle of THP: skip */ | |
565 | if (PageTransTail(page)) { | |
566 | unlock_page(page); | |
567 | continue; | |
568 | } else if (PageTransHuge(page)) { | |
569 | index += HPAGE_PMD_NR - 1; | |
570 | i += HPAGE_PMD_NR - 1; | |
76b6f9b7 JK |
571 | /* |
572 | * 'end' is in the middle of THP. Don't | |
573 | * invalidate the page as the part outside of | |
574 | * 'end' could be still useful. | |
575 | */ | |
576 | if (index > end) { | |
577 | unlock_page(page); | |
fc127da0 | 578 | continue; |
76b6f9b7 | 579 | } |
ef18a1ca KS |
580 | |
581 | /* Take a pin outside pagevec */ | |
582 | get_page(page); | |
583 | ||
584 | /* | |
585 | * Drop extra pins before trying to invalidate | |
586 | * the huge page. | |
587 | */ | |
588 | pagevec_remove_exceptionals(&pvec); | |
589 | pagevec_release(&pvec); | |
fc127da0 KS |
590 | } |
591 | ||
31560180 | 592 | ret = invalidate_inode_page(page); |
1da177e4 | 593 | unlock_page(page); |
31560180 MK |
594 | /* |
595 | * Invalidation is a hint that the page is no longer | |
596 | * of interest and try to speed up its reclaim. | |
597 | */ | |
eb1d7a65 | 598 | if (!ret) { |
cc5993bd | 599 | deactivate_file_page(page); |
eb1d7a65 YS |
600 | /* It is likely on the pagevec of a remote CPU */ |
601 | if (nr_pagevec) | |
602 | (*nr_pagevec)++; | |
603 | } | |
604 | ||
ef18a1ca KS |
605 | if (PageTransHuge(page)) |
606 | put_page(page); | |
31560180 | 607 | count += ret; |
1da177e4 | 608 | } |
0cd6144a | 609 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 610 | pagevec_release(&pvec); |
28697355 | 611 | cond_resched(); |
b85e0eff | 612 | index++; |
1da177e4 | 613 | } |
31560180 | 614 | return count; |
1da177e4 | 615 | } |
eb1d7a65 YS |
616 | |
617 | /** | |
618 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
619 | * @mapping: the address_space which holds the pages to invalidate | |
620 | * @start: the offset 'from' which to invalidate | |
621 | * @end: the offset 'to' which to invalidate (inclusive) | |
622 | * | |
623 | * This function only removes the unlocked pages, if you want to | |
624 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
625 | * | |
626 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
627 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
628 | * pagetables. | |
629 | * | |
630 | * Return: the number of the pages that were invalidated | |
631 | */ | |
632 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
633 | pgoff_t start, pgoff_t end) | |
634 | { | |
635 | return __invalidate_mapping_pages(mapping, start, end, NULL); | |
636 | } | |
54bc4855 | 637 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 638 | |
eb1d7a65 | 639 | /** |
649c6dfe AS |
640 | * invalidate_mapping_pagevec - Invalidate all the unlocked pages of one inode |
641 | * @mapping: the address_space which holds the pages to invalidate | |
642 | * @start: the offset 'from' which to invalidate | |
643 | * @end: the offset 'to' which to invalidate (inclusive) | |
644 | * @nr_pagevec: invalidate failed page number for caller | |
645 | * | |
a00cda3f MCC |
646 | * This helper is similar to invalidate_mapping_pages(), except that it accounts |
647 | * for pages that are likely on a pagevec and counts them in @nr_pagevec, which | |
648 | * will be used by the caller. | |
eb1d7a65 YS |
649 | */ |
650 | void invalidate_mapping_pagevec(struct address_space *mapping, | |
651 | pgoff_t start, pgoff_t end, unsigned long *nr_pagevec) | |
652 | { | |
653 | __invalidate_mapping_pages(mapping, start, end, nr_pagevec); | |
654 | } | |
655 | ||
bd4c8ce4 AM |
656 | /* |
657 | * This is like invalidate_complete_page(), except it ignores the page's | |
658 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
659 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
660 | * shrink_page_list() has a temp ref on them, or because they're transiently |
661 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
662 | */ |
663 | static int | |
664 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
665 | { | |
c4843a75 GT |
666 | unsigned long flags; |
667 | ||
bd4c8ce4 AM |
668 | if (page->mapping != mapping) |
669 | return 0; | |
670 | ||
266cf658 | 671 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
672 | return 0; |
673 | ||
b93b0163 | 674 | xa_lock_irqsave(&mapping->i_pages, flags); |
bd4c8ce4 AM |
675 | if (PageDirty(page)) |
676 | goto failed; | |
677 | ||
266cf658 | 678 | BUG_ON(page_has_private(page)); |
62cccb8c | 679 | __delete_from_page_cache(page, NULL); |
b93b0163 | 680 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
6072d13c LT |
681 | |
682 | if (mapping->a_ops->freepage) | |
683 | mapping->a_ops->freepage(page); | |
684 | ||
09cbfeaf | 685 | put_page(page); /* pagecache ref */ |
bd4c8ce4 AM |
686 | return 1; |
687 | failed: | |
b93b0163 | 688 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
bd4c8ce4 AM |
689 | return 0; |
690 | } | |
691 | ||
e3db7691 TM |
692 | static int do_launder_page(struct address_space *mapping, struct page *page) |
693 | { | |
694 | if (!PageDirty(page)) | |
695 | return 0; | |
696 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
697 | return 0; | |
698 | return mapping->a_ops->launder_page(page); | |
699 | } | |
700 | ||
1da177e4 LT |
701 | /** |
702 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 703 | * @mapping: the address_space |
1da177e4 LT |
704 | * @start: the page offset 'from' which to invalidate |
705 | * @end: the page offset 'to' which to invalidate (inclusive) | |
706 | * | |
707 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
708 | * invalidation. | |
709 | * | |
a862f68a | 710 | * Return: -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
711 | */ |
712 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
713 | pgoff_t start, pgoff_t end) | |
714 | { | |
0cd6144a | 715 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 716 | struct pagevec pvec; |
b85e0eff | 717 | pgoff_t index; |
1da177e4 LT |
718 | int i; |
719 | int ret = 0; | |
0dd1334f | 720 | int ret2 = 0; |
1da177e4 | 721 | int did_range_unmap = 0; |
1da177e4 | 722 | |
32691f0f | 723 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
34ccb69e | 724 | goto out; |
32691f0f | 725 | |
86679820 | 726 | pagevec_init(&pvec); |
b85e0eff | 727 | index = start; |
0cd6144a JW |
728 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
729 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
730 | indices)) { | |
7b965e08 | 731 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 732 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
733 | |
734 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 735 | index = indices[i]; |
b85e0eff HD |
736 | if (index > end) |
737 | break; | |
1da177e4 | 738 | |
3159f943 | 739 | if (xa_is_value(page)) { |
c6dcf52c JK |
740 | if (!invalidate_exceptional_entry2(mapping, |
741 | index, page)) | |
742 | ret = -EBUSY; | |
0cd6144a JW |
743 | continue; |
744 | } | |
745 | ||
1da177e4 | 746 | lock_page(page); |
5cbc198a | 747 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
748 | if (page->mapping != mapping) { |
749 | unlock_page(page); | |
750 | continue; | |
751 | } | |
1da177e4 | 752 | wait_on_page_writeback(page); |
d00806b1 | 753 | if (page_mapped(page)) { |
1da177e4 LT |
754 | if (!did_range_unmap) { |
755 | /* | |
756 | * Zap the rest of the file in one hit. | |
757 | */ | |
977fbdcd MW |
758 | unmap_mapping_pages(mapping, index, |
759 | (1 + end - index), false); | |
1da177e4 LT |
760 | did_range_unmap = 1; |
761 | } else { | |
762 | /* | |
763 | * Just zap this page | |
764 | */ | |
977fbdcd MW |
765 | unmap_mapping_pages(mapping, index, |
766 | 1, false); | |
1da177e4 LT |
767 | } |
768 | } | |
d00806b1 | 769 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
770 | ret2 = do_launder_page(mapping, page); |
771 | if (ret2 == 0) { | |
772 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 773 | ret2 = -EBUSY; |
0dd1334f HH |
774 | } |
775 | if (ret2 < 0) | |
776 | ret = ret2; | |
1da177e4 LT |
777 | unlock_page(page); |
778 | } | |
0cd6144a | 779 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
780 | pagevec_release(&pvec); |
781 | cond_resched(); | |
b85e0eff | 782 | index++; |
1da177e4 | 783 | } |
cd656375 | 784 | /* |
69b6c131 | 785 | * For DAX we invalidate page tables after invalidating page cache. We |
cd656375 JK |
786 | * could invalidate page tables while invalidating each entry however |
787 | * that would be expensive. And doing range unmapping before doesn't | |
69b6c131 | 788 | * work as we have no cheap way to find whether page cache entry didn't |
cd656375 JK |
789 | * get remapped later. |
790 | */ | |
791 | if (dax_mapping(mapping)) { | |
977fbdcd | 792 | unmap_mapping_pages(mapping, start, end - start + 1, false); |
cd656375 | 793 | } |
34ccb69e | 794 | out: |
3167760f | 795 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
796 | return ret; |
797 | } | |
798 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
799 | ||
800 | /** | |
801 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 802 | * @mapping: the address_space |
1da177e4 LT |
803 | * |
804 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
805 | * invalidation. | |
806 | * | |
a862f68a | 807 | * Return: -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
808 | */ |
809 | int invalidate_inode_pages2(struct address_space *mapping) | |
810 | { | |
811 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
812 | } | |
813 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 814 | |
815 | /** | |
816 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
817 | * @inode: inode | |
8a549bea | 818 | * @newsize: new file size |
25d9e2d1 | 819 | * |
820 | * inode's new i_size must already be written before truncate_pagecache | |
821 | * is called. | |
822 | * | |
823 | * This function should typically be called before the filesystem | |
824 | * releases resources associated with the freed range (eg. deallocates | |
825 | * blocks). This way, pagecache will always stay logically coherent | |
826 | * with on-disk format, and the filesystem would not have to deal with | |
827 | * situations such as writepage being called for a page that has already | |
828 | * had its underlying blocks deallocated. | |
829 | */ | |
7caef267 | 830 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 831 | { |
cedabed4 | 832 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 833 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
834 | |
835 | /* | |
836 | * unmap_mapping_range is called twice, first simply for | |
837 | * efficiency so that truncate_inode_pages does fewer | |
838 | * single-page unmaps. However after this first call, and | |
839 | * before truncate_inode_pages finishes, it is possible for | |
840 | * private pages to be COWed, which remain after | |
841 | * truncate_inode_pages finishes, hence the second | |
842 | * unmap_mapping_range call must be made for correctness. | |
843 | */ | |
8a549bea HD |
844 | unmap_mapping_range(mapping, holebegin, 0, 1); |
845 | truncate_inode_pages(mapping, newsize); | |
846 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 847 | } |
848 | EXPORT_SYMBOL(truncate_pagecache); | |
849 | ||
2c27c65e CH |
850 | /** |
851 | * truncate_setsize - update inode and pagecache for a new file size | |
852 | * @inode: inode | |
853 | * @newsize: new file size | |
854 | * | |
382e27da JK |
855 | * truncate_setsize updates i_size and performs pagecache truncation (if |
856 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
857 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 858 | * |
77783d06 JK |
859 | * Must be called with a lock serializing truncates and writes (generally |
860 | * i_mutex but e.g. xfs uses a different lock) and before all filesystem | |
861 | * specific block truncation has been performed. | |
2c27c65e CH |
862 | */ |
863 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
864 | { | |
90a80202 JK |
865 | loff_t oldsize = inode->i_size; |
866 | ||
2c27c65e | 867 | i_size_write(inode, newsize); |
90a80202 JK |
868 | if (newsize > oldsize) |
869 | pagecache_isize_extended(inode, oldsize, newsize); | |
7caef267 | 870 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
871 | } |
872 | EXPORT_SYMBOL(truncate_setsize); | |
873 | ||
90a80202 JK |
874 | /** |
875 | * pagecache_isize_extended - update pagecache after extension of i_size | |
876 | * @inode: inode for which i_size was extended | |
877 | * @from: original inode size | |
878 | * @to: new inode size | |
879 | * | |
880 | * Handle extension of inode size either caused by extending truncate or by | |
881 | * write starting after current i_size. We mark the page straddling current | |
882 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
883 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
884 | * the page before user writes to the page via mmap after the i_size has been | |
885 | * changed. | |
886 | * | |
887 | * The function must be called after i_size is updated so that page fault | |
888 | * coming after we unlock the page will already see the new i_size. | |
889 | * The function must be called while we still hold i_mutex - this not only | |
890 | * makes sure i_size is stable but also that userspace cannot observe new | |
891 | * i_size value before we are prepared to store mmap writes at new inode size. | |
892 | */ | |
893 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
894 | { | |
93407472 | 895 | int bsize = i_blocksize(inode); |
90a80202 JK |
896 | loff_t rounded_from; |
897 | struct page *page; | |
898 | pgoff_t index; | |
899 | ||
90a80202 JK |
900 | WARN_ON(to > inode->i_size); |
901 | ||
09cbfeaf | 902 | if (from >= to || bsize == PAGE_SIZE) |
90a80202 JK |
903 | return; |
904 | /* Page straddling @from will not have any hole block created? */ | |
905 | rounded_from = round_up(from, bsize); | |
09cbfeaf | 906 | if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) |
90a80202 JK |
907 | return; |
908 | ||
09cbfeaf | 909 | index = from >> PAGE_SHIFT; |
90a80202 JK |
910 | page = find_lock_page(inode->i_mapping, index); |
911 | /* Page not cached? Nothing to do */ | |
912 | if (!page) | |
913 | return; | |
914 | /* | |
915 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
916 | * is needed. | |
917 | */ | |
918 | if (page_mkclean(page)) | |
919 | set_page_dirty(page); | |
920 | unlock_page(page); | |
09cbfeaf | 921 | put_page(page); |
90a80202 JK |
922 | } |
923 | EXPORT_SYMBOL(pagecache_isize_extended); | |
924 | ||
623e3db9 HD |
925 | /** |
926 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
927 | * @inode: inode | |
928 | * @lstart: offset of beginning of hole | |
929 | * @lend: offset of last byte of hole | |
930 | * | |
931 | * This function should typically be called before the filesystem | |
932 | * releases resources associated with the freed range (eg. deallocates | |
933 | * blocks). This way, pagecache will always stay logically coherent | |
934 | * with on-disk format, and the filesystem would not have to deal with | |
935 | * situations such as writepage being called for a page that has already | |
936 | * had its underlying blocks deallocated. | |
937 | */ | |
938 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
939 | { | |
940 | struct address_space *mapping = inode->i_mapping; | |
941 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
942 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
943 | /* | |
944 | * This rounding is currently just for example: unmap_mapping_range | |
945 | * expands its hole outwards, whereas we want it to contract the hole | |
946 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
947 | * doing their own page rounding first. Note that unmap_mapping_range |
948 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
949 | */ |
950 | ||
951 | /* | |
952 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
953 | * once (before truncating pagecache), and without "even_cows" flag: | |
954 | * hole-punching should not remove private COWed pages from the hole. | |
955 | */ | |
956 | if ((u64)unmap_end > (u64)unmap_start) | |
957 | unmap_mapping_range(mapping, unmap_start, | |
958 | 1 + unmap_end - unmap_start, 0); | |
959 | truncate_inode_pages_range(mapping, lstart, lend); | |
960 | } | |
961 | EXPORT_SYMBOL(truncate_pagecache_range); |