]>
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 LT |
170 | * |
171 | * We need to bale out if page->mapping is no longer equal to the original | |
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 | */ | |
69f21c4e | 176 | static void truncate_cleanup_page(struct page *page) |
1da177e4 | 177 | { |
69f21c4e HD |
178 | if (page_mapped(page)) |
179 | unmap_mapping_page(page); | |
1da177e4 | 180 | |
266cf658 | 181 | if (page_has_private(page)) |
09cbfeaf | 182 | do_invalidatepage(page, 0, PAGE_SIZE); |
1da177e4 | 183 | |
b9ea2515 KK |
184 | /* |
185 | * Some filesystems seem to re-dirty the page even after | |
186 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
187 | * Hence dirty accounting check is placed after invalidation. | |
188 | */ | |
11f81bec | 189 | cancel_dirty_page(page); |
1da177e4 | 190 | ClearPageMappedToDisk(page); |
1da177e4 LT |
191 | } |
192 | ||
193 | /* | |
fc0ecff6 | 194 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 195 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
196 | * be marked dirty at any time too, so use remove_mapping which safely |
197 | * discards clean, unused pages. | |
1da177e4 LT |
198 | * |
199 | * Returns non-zero if the page was successfully invalidated. | |
200 | */ | |
201 | static int | |
202 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
203 | { | |
0fd0e6b0 NP |
204 | int ret; |
205 | ||
1da177e4 LT |
206 | if (page->mapping != mapping) |
207 | return 0; | |
208 | ||
266cf658 | 209 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
210 | return 0; |
211 | ||
0fd0e6b0 | 212 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
213 | |
214 | return ret; | |
1da177e4 LT |
215 | } |
216 | ||
750b4987 NP |
217 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
218 | { | |
fc127da0 KS |
219 | VM_BUG_ON_PAGE(PageTail(page), page); |
220 | ||
9f4e41f4 JK |
221 | if (page->mapping != mapping) |
222 | return -EIO; | |
223 | ||
69f21c4e | 224 | truncate_cleanup_page(page); |
9f4e41f4 JK |
225 | delete_from_page_cache(page); |
226 | return 0; | |
750b4987 NP |
227 | } |
228 | ||
25718736 AK |
229 | /* |
230 | * Used to get rid of pages on hardware memory corruption. | |
231 | */ | |
232 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
233 | { | |
234 | if (!mapping) | |
235 | return -EINVAL; | |
236 | /* | |
237 | * Only punch for normal data pages for now. | |
238 | * Handling other types like directories would need more auditing. | |
239 | */ | |
240 | if (!S_ISREG(mapping->host->i_mode)) | |
241 | return -EIO; | |
242 | return truncate_inode_page(mapping, page); | |
243 | } | |
244 | EXPORT_SYMBOL(generic_error_remove_page); | |
245 | ||
83f78668 WF |
246 | /* |
247 | * Safely invalidate one page from its pagecache mapping. | |
248 | * It only drops clean, unused pages. The page must be locked. | |
249 | * | |
250 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
251 | */ | |
252 | int invalidate_inode_page(struct page *page) | |
253 | { | |
254 | struct address_space *mapping = page_mapping(page); | |
255 | if (!mapping) | |
256 | return 0; | |
257 | if (PageDirty(page) || PageWriteback(page)) | |
258 | return 0; | |
259 | if (page_mapped(page)) | |
260 | return 0; | |
261 | return invalidate_complete_page(mapping, page); | |
262 | } | |
263 | ||
1da177e4 | 264 | /** |
73c1e204 | 265 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
266 | * @mapping: mapping to truncate |
267 | * @lstart: offset from which to truncate | |
5a720394 | 268 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 269 | * |
d7339071 | 270 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
271 | * specified offsets (and zeroing out partial pages |
272 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
273 | * |
274 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
275 | * block on page locks and it will not block on writeback. The second pass | |
276 | * will wait. This is to prevent as much IO as possible in the affected region. | |
277 | * The first pass will remove most pages, so the search cost of the second pass | |
278 | * is low. | |
279 | * | |
1da177e4 LT |
280 | * We pass down the cache-hot hint to the page freeing code. Even if the |
281 | * mapping is large, it is probably the case that the final pages are the most | |
282 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
283 | * |
284 | * Note that since ->invalidatepage() accepts range to invalidate | |
285 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
286 | * page aligned properly. | |
1da177e4 | 287 | */ |
d7339071 HR |
288 | void truncate_inode_pages_range(struct address_space *mapping, |
289 | loff_t lstart, loff_t lend) | |
1da177e4 | 290 | { |
5a720394 LC |
291 | pgoff_t start; /* inclusive */ |
292 | pgoff_t end; /* exclusive */ | |
293 | unsigned int partial_start; /* inclusive */ | |
294 | unsigned int partial_end; /* exclusive */ | |
295 | struct pagevec pvec; | |
0cd6144a | 296 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
297 | pgoff_t index; |
298 | int i; | |
1da177e4 | 299 | |
f9fe48be | 300 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
34ccb69e | 301 | goto out; |
1da177e4 | 302 | |
5a720394 | 303 | /* Offsets within partial pages */ |
09cbfeaf KS |
304 | partial_start = lstart & (PAGE_SIZE - 1); |
305 | partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
5a720394 LC |
306 | |
307 | /* | |
308 | * 'start' and 'end' always covers the range of pages to be fully | |
309 | * truncated. Partial pages are covered with 'partial_start' at the | |
310 | * start of the range and 'partial_end' at the end of the range. | |
311 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
312 | */ | |
09cbfeaf | 313 | start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
5a720394 LC |
314 | if (lend == -1) |
315 | /* | |
316 | * lend == -1 indicates end-of-file so we have to set 'end' | |
317 | * to the highest possible pgoff_t and since the type is | |
318 | * unsigned we're using -1. | |
319 | */ | |
320 | end = -1; | |
321 | else | |
09cbfeaf | 322 | end = (lend + 1) >> PAGE_SHIFT; |
d7339071 | 323 | |
86679820 | 324 | pagevec_init(&pvec); |
b85e0eff | 325 | index = start; |
0cd6144a JW |
326 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
327 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
328 | indices)) { | |
aa65c29c JK |
329 | /* |
330 | * Pagevec array has exceptional entries and we may also fail | |
331 | * to lock some pages. So we store pages that can be deleted | |
332 | * in a new pagevec. | |
333 | */ | |
334 | struct pagevec locked_pvec; | |
335 | ||
86679820 | 336 | pagevec_init(&locked_pvec); |
1da177e4 LT |
337 | for (i = 0; i < pagevec_count(&pvec); i++) { |
338 | struct page *page = pvec.pages[i]; | |
1da177e4 | 339 | |
b85e0eff | 340 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 341 | index = indices[i]; |
5a720394 | 342 | if (index >= end) |
d7339071 | 343 | break; |
d7339071 | 344 | |
3159f943 | 345 | if (xa_is_value(page)) |
0cd6144a | 346 | continue; |
0cd6144a | 347 | |
529ae9aa | 348 | if (!trylock_page(page)) |
1da177e4 | 349 | continue; |
5cbc198a | 350 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
351 | if (PageWriteback(page)) { |
352 | unlock_page(page); | |
353 | continue; | |
354 | } | |
aa65c29c JK |
355 | if (page->mapping != mapping) { |
356 | unlock_page(page); | |
357 | continue; | |
358 | } | |
359 | pagevec_add(&locked_pvec, page); | |
1da177e4 | 360 | } |
aa65c29c | 361 | for (i = 0; i < pagevec_count(&locked_pvec); i++) |
69f21c4e | 362 | truncate_cleanup_page(locked_pvec.pages[i]); |
aa65c29c JK |
363 | delete_from_page_cache_batch(mapping, &locked_pvec); |
364 | for (i = 0; i < pagevec_count(&locked_pvec); i++) | |
365 | unlock_page(locked_pvec.pages[i]); | |
f2187599 | 366 | truncate_exceptional_pvec_entries(mapping, &pvec, indices, end); |
1da177e4 LT |
367 | pagevec_release(&pvec); |
368 | cond_resched(); | |
b85e0eff | 369 | index++; |
1da177e4 | 370 | } |
5a720394 | 371 | if (partial_start) { |
1da177e4 LT |
372 | struct page *page = find_lock_page(mapping, start - 1); |
373 | if (page) { | |
09cbfeaf | 374 | unsigned int top = PAGE_SIZE; |
5a720394 LC |
375 | if (start > end) { |
376 | /* Truncation within a single page */ | |
377 | top = partial_end; | |
378 | partial_end = 0; | |
379 | } | |
1da177e4 | 380 | wait_on_page_writeback(page); |
5a720394 LC |
381 | zero_user_segment(page, partial_start, top); |
382 | cleancache_invalidate_page(mapping, page); | |
383 | if (page_has_private(page)) | |
384 | do_invalidatepage(page, partial_start, | |
385 | top - partial_start); | |
1da177e4 | 386 | unlock_page(page); |
09cbfeaf | 387 | put_page(page); |
1da177e4 LT |
388 | } |
389 | } | |
5a720394 LC |
390 | if (partial_end) { |
391 | struct page *page = find_lock_page(mapping, end); | |
392 | if (page) { | |
393 | wait_on_page_writeback(page); | |
394 | zero_user_segment(page, 0, partial_end); | |
395 | cleancache_invalidate_page(mapping, page); | |
396 | if (page_has_private(page)) | |
397 | do_invalidatepage(page, 0, | |
398 | partial_end); | |
399 | unlock_page(page); | |
09cbfeaf | 400 | put_page(page); |
5a720394 LC |
401 | } |
402 | } | |
403 | /* | |
404 | * If the truncation happened within a single page no pages | |
405 | * will be released, just zeroed, so we can bail out now. | |
406 | */ | |
407 | if (start >= end) | |
34ccb69e | 408 | goto out; |
1da177e4 | 409 | |
b85e0eff | 410 | index = start; |
1da177e4 LT |
411 | for ( ; ; ) { |
412 | cond_resched(); | |
0cd6144a | 413 | if (!pagevec_lookup_entries(&pvec, mapping, index, |
792ceaef HD |
414 | min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { |
415 | /* If all gone from start onwards, we're done */ | |
b85e0eff | 416 | if (index == start) |
1da177e4 | 417 | break; |
792ceaef | 418 | /* Otherwise restart to make sure all gone */ |
b85e0eff | 419 | index = start; |
1da177e4 LT |
420 | continue; |
421 | } | |
0cd6144a | 422 | if (index == start && indices[0] >= end) { |
792ceaef | 423 | /* All gone out of hole to be punched, we're done */ |
0cd6144a | 424 | pagevec_remove_exceptionals(&pvec); |
d7339071 HR |
425 | pagevec_release(&pvec); |
426 | break; | |
427 | } | |
f2187599 | 428 | |
1da177e4 LT |
429 | for (i = 0; i < pagevec_count(&pvec); i++) { |
430 | struct page *page = pvec.pages[i]; | |
431 | ||
b85e0eff | 432 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 433 | index = indices[i]; |
792ceaef HD |
434 | if (index >= end) { |
435 | /* Restart punch to make sure all gone */ | |
436 | index = start - 1; | |
d7339071 | 437 | break; |
792ceaef | 438 | } |
b85e0eff | 439 | |
3159f943 | 440 | if (xa_is_value(page)) |
0cd6144a | 441 | continue; |
0cd6144a | 442 | |
1da177e4 | 443 | lock_page(page); |
5cbc198a | 444 | WARN_ON(page_to_index(page) != index); |
1da177e4 | 445 | wait_on_page_writeback(page); |
750b4987 | 446 | truncate_inode_page(mapping, page); |
1da177e4 LT |
447 | unlock_page(page); |
448 | } | |
f2187599 | 449 | truncate_exceptional_pvec_entries(mapping, &pvec, indices, end); |
1da177e4 | 450 | pagevec_release(&pvec); |
b85e0eff | 451 | index++; |
1da177e4 | 452 | } |
34ccb69e AR |
453 | |
454 | out: | |
3167760f | 455 | cleancache_invalidate_inode(mapping); |
1da177e4 | 456 | } |
d7339071 | 457 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 458 | |
d7339071 HR |
459 | /** |
460 | * truncate_inode_pages - truncate *all* the pages from an offset | |
461 | * @mapping: mapping to truncate | |
462 | * @lstart: offset from which to truncate | |
463 | * | |
1b1dcc1b | 464 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
465 | * |
466 | * Note: When this function returns, there can be a page in the process of | |
467 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
468 | * mapping->nrpages can be non-zero when this function returns even after | |
469 | * truncation of the whole mapping. | |
d7339071 HR |
470 | */ |
471 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
472 | { | |
473 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
474 | } | |
1da177e4 LT |
475 | EXPORT_SYMBOL(truncate_inode_pages); |
476 | ||
91b0abe3 JW |
477 | /** |
478 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
479 | * @mapping: mapping to truncate | |
480 | * | |
481 | * Called under (and serialized by) inode->i_mutex. | |
482 | * | |
483 | * Filesystems have to use this in the .evict_inode path to inform the | |
484 | * VM that this is the final truncate and the inode is going away. | |
485 | */ | |
486 | void truncate_inode_pages_final(struct address_space *mapping) | |
487 | { | |
f9fe48be | 488 | unsigned long nrexceptional; |
91b0abe3 JW |
489 | unsigned long nrpages; |
490 | ||
491 | /* | |
492 | * Page reclaim can not participate in regular inode lifetime | |
493 | * management (can't call iput()) and thus can race with the | |
494 | * inode teardown. Tell it when the address space is exiting, | |
495 | * so that it does not install eviction information after the | |
496 | * final truncate has begun. | |
497 | */ | |
498 | mapping_set_exiting(mapping); | |
499 | ||
500 | /* | |
501 | * When reclaim installs eviction entries, it increases | |
f9fe48be | 502 | * nrexceptional first, then decreases nrpages. Make sure we see |
91b0abe3 JW |
503 | * this in the right order or we might miss an entry. |
504 | */ | |
505 | nrpages = mapping->nrpages; | |
506 | smp_rmb(); | |
f9fe48be | 507 | nrexceptional = mapping->nrexceptional; |
91b0abe3 | 508 | |
f9fe48be | 509 | if (nrpages || nrexceptional) { |
91b0abe3 JW |
510 | /* |
511 | * As truncation uses a lockless tree lookup, cycle | |
512 | * the tree lock to make sure any ongoing tree | |
513 | * modification that does not see AS_EXITING is | |
514 | * completed before starting the final truncate. | |
515 | */ | |
b93b0163 MW |
516 | xa_lock_irq(&mapping->i_pages); |
517 | xa_unlock_irq(&mapping->i_pages); | |
91b0abe3 | 518 | } |
6ff38bd4 PT |
519 | |
520 | /* | |
521 | * Cleancache needs notification even if there are no pages or shadow | |
522 | * entries. | |
523 | */ | |
524 | truncate_inode_pages(mapping, 0); | |
91b0abe3 JW |
525 | } |
526 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
527 | ||
28697355 MW |
528 | /** |
529 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
530 | * @mapping: the address_space which holds the pages to invalidate | |
531 | * @start: the offset 'from' which to invalidate | |
532 | * @end: the offset 'to' which to invalidate (inclusive) | |
533 | * | |
534 | * This function only removes the unlocked pages, if you want to | |
535 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
536 | * | |
537 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
538 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
539 | * pagetables. | |
a862f68a MR |
540 | * |
541 | * Return: the number of the pages that were invalidated | |
28697355 MW |
542 | */ |
543 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 544 | pgoff_t start, pgoff_t end) |
1da177e4 | 545 | { |
0cd6144a | 546 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 547 | struct pagevec pvec; |
b85e0eff | 548 | pgoff_t index = start; |
31560180 MK |
549 | unsigned long ret; |
550 | unsigned long count = 0; | |
1da177e4 LT |
551 | int i; |
552 | ||
86679820 | 553 | pagevec_init(&pvec); |
0cd6144a JW |
554 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
555 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
556 | indices)) { | |
1da177e4 LT |
557 | for (i = 0; i < pagevec_count(&pvec); i++) { |
558 | struct page *page = pvec.pages[i]; | |
e0f23603 | 559 | |
b85e0eff | 560 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 561 | index = indices[i]; |
b85e0eff HD |
562 | if (index > end) |
563 | break; | |
e0f23603 | 564 | |
3159f943 | 565 | if (xa_is_value(page)) { |
c6dcf52c JK |
566 | invalidate_exceptional_entry(mapping, index, |
567 | page); | |
0cd6144a JW |
568 | continue; |
569 | } | |
570 | ||
b85e0eff HD |
571 | if (!trylock_page(page)) |
572 | continue; | |
fc127da0 | 573 | |
5cbc198a | 574 | WARN_ON(page_to_index(page) != index); |
fc127da0 KS |
575 | |
576 | /* Middle of THP: skip */ | |
577 | if (PageTransTail(page)) { | |
578 | unlock_page(page); | |
579 | continue; | |
580 | } else if (PageTransHuge(page)) { | |
581 | index += HPAGE_PMD_NR - 1; | |
582 | i += HPAGE_PMD_NR - 1; | |
76b6f9b7 JK |
583 | /* |
584 | * 'end' is in the middle of THP. Don't | |
585 | * invalidate the page as the part outside of | |
586 | * 'end' could be still useful. | |
587 | */ | |
588 | if (index > end) { | |
589 | unlock_page(page); | |
fc127da0 | 590 | continue; |
76b6f9b7 | 591 | } |
ef18a1ca KS |
592 | |
593 | /* Take a pin outside pagevec */ | |
594 | get_page(page); | |
595 | ||
596 | /* | |
597 | * Drop extra pins before trying to invalidate | |
598 | * the huge page. | |
599 | */ | |
600 | pagevec_remove_exceptionals(&pvec); | |
601 | pagevec_release(&pvec); | |
fc127da0 KS |
602 | } |
603 | ||
31560180 | 604 | ret = invalidate_inode_page(page); |
1da177e4 | 605 | unlock_page(page); |
31560180 MK |
606 | /* |
607 | * Invalidation is a hint that the page is no longer | |
608 | * of interest and try to speed up its reclaim. | |
609 | */ | |
610 | if (!ret) | |
cc5993bd | 611 | deactivate_file_page(page); |
ef18a1ca KS |
612 | if (PageTransHuge(page)) |
613 | put_page(page); | |
31560180 | 614 | count += ret; |
1da177e4 | 615 | } |
0cd6144a | 616 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 617 | pagevec_release(&pvec); |
28697355 | 618 | cond_resched(); |
b85e0eff | 619 | index++; |
1da177e4 | 620 | } |
31560180 | 621 | return count; |
1da177e4 | 622 | } |
54bc4855 | 623 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 624 | |
bd4c8ce4 AM |
625 | /* |
626 | * This is like invalidate_complete_page(), except it ignores the page's | |
627 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
628 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
629 | * shrink_page_list() has a temp ref on them, or because they're transiently |
630 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
631 | */ |
632 | static int | |
633 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
634 | { | |
c4843a75 GT |
635 | unsigned long flags; |
636 | ||
bd4c8ce4 AM |
637 | if (page->mapping != mapping) |
638 | return 0; | |
639 | ||
266cf658 | 640 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
641 | return 0; |
642 | ||
b93b0163 | 643 | xa_lock_irqsave(&mapping->i_pages, flags); |
bd4c8ce4 AM |
644 | if (PageDirty(page)) |
645 | goto failed; | |
646 | ||
266cf658 | 647 | BUG_ON(page_has_private(page)); |
62cccb8c | 648 | __delete_from_page_cache(page, NULL); |
b93b0163 | 649 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
6072d13c LT |
650 | |
651 | if (mapping->a_ops->freepage) | |
652 | mapping->a_ops->freepage(page); | |
653 | ||
09cbfeaf | 654 | put_page(page); /* pagecache ref */ |
bd4c8ce4 AM |
655 | return 1; |
656 | failed: | |
b93b0163 | 657 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
bd4c8ce4 AM |
658 | return 0; |
659 | } | |
660 | ||
e3db7691 TM |
661 | static int do_launder_page(struct address_space *mapping, struct page *page) |
662 | { | |
663 | if (!PageDirty(page)) | |
664 | return 0; | |
665 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
666 | return 0; | |
667 | return mapping->a_ops->launder_page(page); | |
668 | } | |
669 | ||
1da177e4 LT |
670 | /** |
671 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 672 | * @mapping: the address_space |
1da177e4 LT |
673 | * @start: the page offset 'from' which to invalidate |
674 | * @end: the page offset 'to' which to invalidate (inclusive) | |
675 | * | |
676 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
677 | * invalidation. | |
678 | * | |
a862f68a | 679 | * Return: -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
680 | */ |
681 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
682 | pgoff_t start, pgoff_t end) | |
683 | { | |
0cd6144a | 684 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 685 | struct pagevec pvec; |
b85e0eff | 686 | pgoff_t index; |
1da177e4 LT |
687 | int i; |
688 | int ret = 0; | |
0dd1334f | 689 | int ret2 = 0; |
1da177e4 | 690 | int did_range_unmap = 0; |
1da177e4 | 691 | |
32691f0f | 692 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
34ccb69e | 693 | goto out; |
32691f0f | 694 | |
86679820 | 695 | pagevec_init(&pvec); |
b85e0eff | 696 | index = start; |
0cd6144a JW |
697 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
698 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
699 | indices)) { | |
7b965e08 | 700 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 701 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
702 | |
703 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 704 | index = indices[i]; |
b85e0eff HD |
705 | if (index > end) |
706 | break; | |
1da177e4 | 707 | |
3159f943 | 708 | if (xa_is_value(page)) { |
c6dcf52c JK |
709 | if (!invalidate_exceptional_entry2(mapping, |
710 | index, page)) | |
711 | ret = -EBUSY; | |
0cd6144a JW |
712 | continue; |
713 | } | |
714 | ||
69f21c4e HD |
715 | if (!did_range_unmap && page_mapped(page)) { |
716 | /* | |
717 | * If page is mapped, before taking its lock, | |
718 | * zap the rest of the file in one hit. | |
719 | */ | |
720 | unmap_mapping_pages(mapping, index, | |
721 | (1 + end - index), false); | |
722 | did_range_unmap = 1; | |
723 | } | |
724 | ||
1da177e4 | 725 | lock_page(page); |
5cbc198a | 726 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
727 | if (page->mapping != mapping) { |
728 | unlock_page(page); | |
729 | continue; | |
730 | } | |
1da177e4 | 731 | wait_on_page_writeback(page); |
69f21c4e HD |
732 | |
733 | if (page_mapped(page)) | |
734 | unmap_mapping_page(page); | |
d00806b1 | 735 | BUG_ON(page_mapped(page)); |
69f21c4e | 736 | |
0dd1334f HH |
737 | ret2 = do_launder_page(mapping, page); |
738 | if (ret2 == 0) { | |
739 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 740 | ret2 = -EBUSY; |
0dd1334f HH |
741 | } |
742 | if (ret2 < 0) | |
743 | ret = ret2; | |
1da177e4 LT |
744 | unlock_page(page); |
745 | } | |
0cd6144a | 746 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
747 | pagevec_release(&pvec); |
748 | cond_resched(); | |
b85e0eff | 749 | index++; |
1da177e4 | 750 | } |
cd656375 | 751 | /* |
69b6c131 | 752 | * For DAX we invalidate page tables after invalidating page cache. We |
cd656375 JK |
753 | * could invalidate page tables while invalidating each entry however |
754 | * that would be expensive. And doing range unmapping before doesn't | |
69b6c131 | 755 | * work as we have no cheap way to find whether page cache entry didn't |
cd656375 JK |
756 | * get remapped later. |
757 | */ | |
758 | if (dax_mapping(mapping)) { | |
977fbdcd | 759 | unmap_mapping_pages(mapping, start, end - start + 1, false); |
cd656375 | 760 | } |
34ccb69e | 761 | out: |
3167760f | 762 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
763 | return ret; |
764 | } | |
765 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
766 | ||
767 | /** | |
768 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 769 | * @mapping: the address_space |
1da177e4 LT |
770 | * |
771 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
772 | * invalidation. | |
773 | * | |
a862f68a | 774 | * Return: -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
775 | */ |
776 | int invalidate_inode_pages2(struct address_space *mapping) | |
777 | { | |
778 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
779 | } | |
780 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 781 | |
782 | /** | |
783 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
784 | * @inode: inode | |
8a549bea | 785 | * @newsize: new file size |
25d9e2d1 | 786 | * |
787 | * inode's new i_size must already be written before truncate_pagecache | |
788 | * is called. | |
789 | * | |
790 | * This function should typically be called before the filesystem | |
791 | * releases resources associated with the freed range (eg. deallocates | |
792 | * blocks). This way, pagecache will always stay logically coherent | |
793 | * with on-disk format, and the filesystem would not have to deal with | |
794 | * situations such as writepage being called for a page that has already | |
795 | * had its underlying blocks deallocated. | |
796 | */ | |
7caef267 | 797 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 798 | { |
cedabed4 | 799 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 800 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
801 | |
802 | /* | |
803 | * unmap_mapping_range is called twice, first simply for | |
804 | * efficiency so that truncate_inode_pages does fewer | |
805 | * single-page unmaps. However after this first call, and | |
806 | * before truncate_inode_pages finishes, it is possible for | |
807 | * private pages to be COWed, which remain after | |
808 | * truncate_inode_pages finishes, hence the second | |
809 | * unmap_mapping_range call must be made for correctness. | |
810 | */ | |
8a549bea HD |
811 | unmap_mapping_range(mapping, holebegin, 0, 1); |
812 | truncate_inode_pages(mapping, newsize); | |
813 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 814 | } |
815 | EXPORT_SYMBOL(truncate_pagecache); | |
816 | ||
2c27c65e CH |
817 | /** |
818 | * truncate_setsize - update inode and pagecache for a new file size | |
819 | * @inode: inode | |
820 | * @newsize: new file size | |
821 | * | |
382e27da JK |
822 | * truncate_setsize updates i_size and performs pagecache truncation (if |
823 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
824 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 825 | * |
77783d06 JK |
826 | * Must be called with a lock serializing truncates and writes (generally |
827 | * i_mutex but e.g. xfs uses a different lock) and before all filesystem | |
828 | * specific block truncation has been performed. | |
2c27c65e CH |
829 | */ |
830 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
831 | { | |
90a80202 JK |
832 | loff_t oldsize = inode->i_size; |
833 | ||
2c27c65e | 834 | i_size_write(inode, newsize); |
90a80202 JK |
835 | if (newsize > oldsize) |
836 | pagecache_isize_extended(inode, oldsize, newsize); | |
7caef267 | 837 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
838 | } |
839 | EXPORT_SYMBOL(truncate_setsize); | |
840 | ||
90a80202 JK |
841 | /** |
842 | * pagecache_isize_extended - update pagecache after extension of i_size | |
843 | * @inode: inode for which i_size was extended | |
844 | * @from: original inode size | |
845 | * @to: new inode size | |
846 | * | |
847 | * Handle extension of inode size either caused by extending truncate or by | |
848 | * write starting after current i_size. We mark the page straddling current | |
849 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
850 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
851 | * the page before user writes to the page via mmap after the i_size has been | |
852 | * changed. | |
853 | * | |
854 | * The function must be called after i_size is updated so that page fault | |
855 | * coming after we unlock the page will already see the new i_size. | |
856 | * The function must be called while we still hold i_mutex - this not only | |
857 | * makes sure i_size is stable but also that userspace cannot observe new | |
858 | * i_size value before we are prepared to store mmap writes at new inode size. | |
859 | */ | |
860 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
861 | { | |
93407472 | 862 | int bsize = i_blocksize(inode); |
90a80202 JK |
863 | loff_t rounded_from; |
864 | struct page *page; | |
865 | pgoff_t index; | |
866 | ||
90a80202 JK |
867 | WARN_ON(to > inode->i_size); |
868 | ||
09cbfeaf | 869 | if (from >= to || bsize == PAGE_SIZE) |
90a80202 JK |
870 | return; |
871 | /* Page straddling @from will not have any hole block created? */ | |
872 | rounded_from = round_up(from, bsize); | |
09cbfeaf | 873 | if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) |
90a80202 JK |
874 | return; |
875 | ||
09cbfeaf | 876 | index = from >> PAGE_SHIFT; |
90a80202 JK |
877 | page = find_lock_page(inode->i_mapping, index); |
878 | /* Page not cached? Nothing to do */ | |
879 | if (!page) | |
880 | return; | |
881 | /* | |
882 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
883 | * is needed. | |
884 | */ | |
885 | if (page_mkclean(page)) | |
886 | set_page_dirty(page); | |
887 | unlock_page(page); | |
09cbfeaf | 888 | put_page(page); |
90a80202 JK |
889 | } |
890 | EXPORT_SYMBOL(pagecache_isize_extended); | |
891 | ||
623e3db9 HD |
892 | /** |
893 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
894 | * @inode: inode | |
895 | * @lstart: offset of beginning of hole | |
896 | * @lend: offset of last byte of hole | |
897 | * | |
898 | * This function should typically be called before the filesystem | |
899 | * releases resources associated with the freed range (eg. deallocates | |
900 | * blocks). This way, pagecache will always stay logically coherent | |
901 | * with on-disk format, and the filesystem would not have to deal with | |
902 | * situations such as writepage being called for a page that has already | |
903 | * had its underlying blocks deallocated. | |
904 | */ | |
905 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
906 | { | |
907 | struct address_space *mapping = inode->i_mapping; | |
908 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
909 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
910 | /* | |
911 | * This rounding is currently just for example: unmap_mapping_range | |
912 | * expands its hole outwards, whereas we want it to contract the hole | |
913 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
914 | * doing their own page rounding first. Note that unmap_mapping_range |
915 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
916 | */ |
917 | ||
918 | /* | |
919 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
920 | * once (before truncating pagecache), and without "even_cows" flag: | |
921 | * hole-punching should not remove private COWed pages from the hole. | |
922 | */ | |
923 | if ((u64)unmap_end > (u64)unmap_start) | |
924 | unmap_mapping_range(mapping, unmap_start, | |
925 | 1 + unmap_end - unmap_start, 0); | |
926 | truncate_inode_pages_range(mapping, lstart, lend); | |
927 | } | |
928 | EXPORT_SYMBOL(truncate_pagecache_range); |