]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * mm/truncate.c - code for taking down pages from address_spaces | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds | |
5 | * | |
e1f8e874 | 6 | * 10Sep2002 Andrew Morton |
1da177e4 LT |
7 | * Initial version. |
8 | */ | |
9 | ||
10 | #include <linux/kernel.h> | |
4af3c9cc | 11 | #include <linux/backing-dev.h> |
f9fe48be | 12 | #include <linux/dax.h> |
5a0e3ad6 | 13 | #include <linux/gfp.h> |
1da177e4 | 14 | #include <linux/mm.h> |
0fd0e6b0 | 15 | #include <linux/swap.h> |
b95f1b31 | 16 | #include <linux/export.h> |
1da177e4 | 17 | #include <linux/pagemap.h> |
01f2705d | 18 | #include <linux/highmem.h> |
1da177e4 | 19 | #include <linux/pagevec.h> |
e08748ce | 20 | #include <linux/task_io_accounting_ops.h> |
1da177e4 | 21 | #include <linux/buffer_head.h> /* grr. try_to_release_page, |
aaa4059b | 22 | do_invalidatepage */ |
3a4f8a0b | 23 | #include <linux/shmem_fs.h> |
c515e1fd | 24 | #include <linux/cleancache.h> |
90a80202 | 25 | #include <linux/rmap.h> |
ba470de4 | 26 | #include "internal.h" |
1da177e4 | 27 | |
f2187599 MG |
28 | /* |
29 | * Regular page slots are stabilized by the page lock even without the tree | |
30 | * itself locked. These unlocked entries need verification under the tree | |
31 | * lock. | |
32 | */ | |
33 | static inline void __clear_shadow_entry(struct address_space *mapping, | |
34 | pgoff_t index, void *entry) | |
0cd6144a | 35 | { |
69b6c131 | 36 | XA_STATE(xas, &mapping->i_pages, index); |
449dd698 | 37 | |
69b6c131 MW |
38 | xas_set_update(&xas, workingset_update_node); |
39 | if (xas_load(&xas) != entry) | |
f2187599 | 40 | return; |
69b6c131 | 41 | xas_store(&xas, NULL); |
ac401cc7 | 42 | mapping->nrexceptional--; |
f2187599 MG |
43 | } |
44 | ||
45 | static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, | |
46 | void *entry) | |
47 | { | |
b93b0163 | 48 | xa_lock_irq(&mapping->i_pages); |
f2187599 | 49 | __clear_shadow_entry(mapping, index, entry); |
b93b0163 | 50 | xa_unlock_irq(&mapping->i_pages); |
0cd6144a | 51 | } |
1da177e4 | 52 | |
c6dcf52c | 53 | /* |
f2187599 MG |
54 | * Unconditionally remove exceptional entries. Usually called from truncate |
55 | * path. Note that the pagevec may be altered by this function by removing | |
56 | * exceptional entries similar to what pagevec_remove_exceptionals does. | |
c6dcf52c | 57 | */ |
f2187599 MG |
58 | static void truncate_exceptional_pvec_entries(struct address_space *mapping, |
59 | struct pagevec *pvec, pgoff_t *indices, | |
60 | pgoff_t end) | |
c6dcf52c | 61 | { |
f2187599 MG |
62 | int i, j; |
63 | bool dax, lock; | |
64 | ||
c6dcf52c JK |
65 | /* Handled by shmem itself */ |
66 | if (shmem_mapping(mapping)) | |
67 | return; | |
68 | ||
f2187599 | 69 | for (j = 0; j < pagevec_count(pvec); j++) |
3159f943 | 70 | if (xa_is_value(pvec->pages[j])) |
f2187599 MG |
71 | break; |
72 | ||
73 | if (j == pagevec_count(pvec)) | |
c6dcf52c | 74 | return; |
f2187599 MG |
75 | |
76 | dax = dax_mapping(mapping); | |
77 | lock = !dax && indices[j] < end; | |
78 | if (lock) | |
b93b0163 | 79 | xa_lock_irq(&mapping->i_pages); |
f2187599 MG |
80 | |
81 | for (i = j; i < pagevec_count(pvec); i++) { | |
82 | struct page *page = pvec->pages[i]; | |
83 | pgoff_t index = indices[i]; | |
84 | ||
3159f943 | 85 | if (!xa_is_value(page)) { |
f2187599 MG |
86 | pvec->pages[j++] = page; |
87 | continue; | |
88 | } | |
89 | ||
90 | if (index >= end) | |
91 | continue; | |
92 | ||
93 | if (unlikely(dax)) { | |
94 | dax_delete_mapping_entry(mapping, index); | |
95 | continue; | |
96 | } | |
97 | ||
98 | __clear_shadow_entry(mapping, index, page); | |
c6dcf52c | 99 | } |
f2187599 MG |
100 | |
101 | if (lock) | |
b93b0163 | 102 | xa_unlock_irq(&mapping->i_pages); |
f2187599 | 103 | pvec->nr = j; |
c6dcf52c JK |
104 | } |
105 | ||
106 | /* | |
107 | * Invalidate exceptional entry if easily possible. This handles exceptional | |
4636e70b | 108 | * entries for invalidate_inode_pages(). |
c6dcf52c JK |
109 | */ |
110 | static int invalidate_exceptional_entry(struct address_space *mapping, | |
111 | pgoff_t index, void *entry) | |
112 | { | |
4636e70b RZ |
113 | /* Handled by shmem itself, or for DAX we do nothing. */ |
114 | if (shmem_mapping(mapping) || dax_mapping(mapping)) | |
c6dcf52c | 115 | return 1; |
c6dcf52c JK |
116 | clear_shadow_entry(mapping, index, entry); |
117 | return 1; | |
118 | } | |
119 | ||
120 | /* | |
121 | * Invalidate exceptional entry if clean. This handles exceptional entries for | |
122 | * invalidate_inode_pages2() so for DAX it evicts only clean entries. | |
123 | */ | |
124 | static int invalidate_exceptional_entry2(struct address_space *mapping, | |
125 | pgoff_t index, void *entry) | |
126 | { | |
127 | /* Handled by shmem itself */ | |
128 | if (shmem_mapping(mapping)) | |
129 | return 1; | |
130 | if (dax_mapping(mapping)) | |
131 | return dax_invalidate_mapping_entry_sync(mapping, index); | |
132 | clear_shadow_entry(mapping, index, entry); | |
133 | return 1; | |
134 | } | |
135 | ||
cf9a2ae8 | 136 | /** |
28bc44d7 | 137 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 138 | * @page: the page which is affected |
d47992f8 LC |
139 | * @offset: start of the range to invalidate |
140 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
141 | * |
142 | * do_invalidatepage() is called when all or part of the page has become | |
143 | * invalidated by a truncate operation. | |
144 | * | |
145 | * do_invalidatepage() does not have to release all buffers, but it must | |
146 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
147 | * is underway against any of the blocks which are outside the truncation | |
148 | * point. Because the caller is about to free (and possibly reuse) those | |
149 | * blocks on-disk. | |
150 | */ | |
d47992f8 LC |
151 | void do_invalidatepage(struct page *page, unsigned int offset, |
152 | unsigned int length) | |
cf9a2ae8 | 153 | { |
d47992f8 LC |
154 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
155 | ||
cf9a2ae8 | 156 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 157 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
158 | if (!invalidatepage) |
159 | invalidatepage = block_invalidatepage; | |
9361401e | 160 | #endif |
cf9a2ae8 | 161 | if (invalidatepage) |
d47992f8 | 162 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
163 | } |
164 | ||
1da177e4 LT |
165 | /* |
166 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 167 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 168 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
169 | * |
170 | * We need to bale out if page->mapping is no longer equal to the original | |
171 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
fc0ecff6 | 172 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
173 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
174 | */ | |
9f4e41f4 JK |
175 | static void |
176 | truncate_cleanup_page(struct address_space *mapping, struct page *page) | |
1da177e4 | 177 | { |
9f4e41f4 | 178 | if (page_mapped(page)) { |
977fbdcd MW |
179 | pgoff_t nr = PageTransHuge(page) ? HPAGE_PMD_NR : 1; |
180 | unmap_mapping_pages(mapping, page->index, nr, false); | |
9f4e41f4 | 181 | } |
1da177e4 | 182 | |
266cf658 | 183 | if (page_has_private(page)) |
09cbfeaf | 184 | do_invalidatepage(page, 0, PAGE_SIZE); |
1da177e4 | 185 | |
b9ea2515 KK |
186 | /* |
187 | * Some filesystems seem to re-dirty the page even after | |
188 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
189 | * Hence dirty accounting check is placed after invalidation. | |
190 | */ | |
11f81bec | 191 | cancel_dirty_page(page); |
1da177e4 | 192 | ClearPageMappedToDisk(page); |
1da177e4 LT |
193 | } |
194 | ||
195 | /* | |
fc0ecff6 | 196 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 197 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
198 | * be marked dirty at any time too, so use remove_mapping which safely |
199 | * discards clean, unused pages. | |
1da177e4 LT |
200 | * |
201 | * Returns non-zero if the page was successfully invalidated. | |
202 | */ | |
203 | static int | |
204 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
205 | { | |
0fd0e6b0 NP |
206 | int ret; |
207 | ||
1da177e4 LT |
208 | if (page->mapping != mapping) |
209 | return 0; | |
210 | ||
266cf658 | 211 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
212 | return 0; |
213 | ||
0fd0e6b0 | 214 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
215 | |
216 | return ret; | |
1da177e4 LT |
217 | } |
218 | ||
750b4987 NP |
219 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
220 | { | |
fc127da0 KS |
221 | VM_BUG_ON_PAGE(PageTail(page), page); |
222 | ||
9f4e41f4 JK |
223 | if (page->mapping != mapping) |
224 | return -EIO; | |
225 | ||
226 | truncate_cleanup_page(mapping, page); | |
227 | delete_from_page_cache(page); | |
228 | return 0; | |
750b4987 NP |
229 | } |
230 | ||
25718736 AK |
231 | /* |
232 | * Used to get rid of pages on hardware memory corruption. | |
233 | */ | |
234 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
235 | { | |
236 | if (!mapping) | |
237 | return -EINVAL; | |
238 | /* | |
239 | * Only punch for normal data pages for now. | |
240 | * Handling other types like directories would need more auditing. | |
241 | */ | |
242 | if (!S_ISREG(mapping->host->i_mode)) | |
243 | return -EIO; | |
244 | return truncate_inode_page(mapping, page); | |
245 | } | |
246 | EXPORT_SYMBOL(generic_error_remove_page); | |
247 | ||
83f78668 WF |
248 | /* |
249 | * Safely invalidate one page from its pagecache mapping. | |
250 | * It only drops clean, unused pages. The page must be locked. | |
251 | * | |
252 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
253 | */ | |
254 | int invalidate_inode_page(struct page *page) | |
255 | { | |
256 | struct address_space *mapping = page_mapping(page); | |
257 | if (!mapping) | |
258 | return 0; | |
259 | if (PageDirty(page) || PageWriteback(page)) | |
260 | return 0; | |
261 | if (page_mapped(page)) | |
262 | return 0; | |
263 | return invalidate_complete_page(mapping, page); | |
264 | } | |
265 | ||
1da177e4 | 266 | /** |
73c1e204 | 267 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
268 | * @mapping: mapping to truncate |
269 | * @lstart: offset from which to truncate | |
5a720394 | 270 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 271 | * |
d7339071 | 272 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
273 | * specified offsets (and zeroing out partial pages |
274 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
275 | * |
276 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
277 | * block on page locks and it will not block on writeback. The second pass | |
278 | * will wait. This is to prevent as much IO as possible in the affected region. | |
279 | * The first pass will remove most pages, so the search cost of the second pass | |
280 | * is low. | |
281 | * | |
1da177e4 LT |
282 | * We pass down the cache-hot hint to the page freeing code. Even if the |
283 | * mapping is large, it is probably the case that the final pages are the most | |
284 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
285 | * |
286 | * Note that since ->invalidatepage() accepts range to invalidate | |
287 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
288 | * page aligned properly. | |
1da177e4 | 289 | */ |
d7339071 HR |
290 | void truncate_inode_pages_range(struct address_space *mapping, |
291 | loff_t lstart, loff_t lend) | |
1da177e4 | 292 | { |
5a720394 LC |
293 | pgoff_t start; /* inclusive */ |
294 | pgoff_t end; /* exclusive */ | |
295 | unsigned int partial_start; /* inclusive */ | |
296 | unsigned int partial_end; /* exclusive */ | |
297 | struct pagevec pvec; | |
0cd6144a | 298 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
299 | pgoff_t index; |
300 | int i; | |
1da177e4 | 301 | |
f9fe48be | 302 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
34ccb69e | 303 | goto out; |
1da177e4 | 304 | |
5a720394 | 305 | /* Offsets within partial pages */ |
09cbfeaf KS |
306 | partial_start = lstart & (PAGE_SIZE - 1); |
307 | partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
5a720394 LC |
308 | |
309 | /* | |
310 | * 'start' and 'end' always covers the range of pages to be fully | |
311 | * truncated. Partial pages are covered with 'partial_start' at the | |
312 | * start of the range and 'partial_end' at the end of the range. | |
313 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
314 | */ | |
09cbfeaf | 315 | start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
5a720394 LC |
316 | if (lend == -1) |
317 | /* | |
318 | * lend == -1 indicates end-of-file so we have to set 'end' | |
319 | * to the highest possible pgoff_t and since the type is | |
320 | * unsigned we're using -1. | |
321 | */ | |
322 | end = -1; | |
323 | else | |
09cbfeaf | 324 | end = (lend + 1) >> PAGE_SHIFT; |
d7339071 | 325 | |
86679820 | 326 | pagevec_init(&pvec); |
b85e0eff | 327 | index = start; |
0cd6144a JW |
328 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
329 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
330 | indices)) { | |
aa65c29c JK |
331 | /* |
332 | * Pagevec array has exceptional entries and we may also fail | |
333 | * to lock some pages. So we store pages that can be deleted | |
334 | * in a new pagevec. | |
335 | */ | |
336 | struct pagevec locked_pvec; | |
337 | ||
86679820 | 338 | pagevec_init(&locked_pvec); |
1da177e4 LT |
339 | for (i = 0; i < pagevec_count(&pvec); i++) { |
340 | struct page *page = pvec.pages[i]; | |
1da177e4 | 341 | |
b85e0eff | 342 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 343 | index = indices[i]; |
5a720394 | 344 | if (index >= end) |
d7339071 | 345 | break; |
d7339071 | 346 | |
3159f943 | 347 | if (xa_is_value(page)) |
0cd6144a | 348 | continue; |
0cd6144a | 349 | |
529ae9aa | 350 | if (!trylock_page(page)) |
1da177e4 | 351 | continue; |
5cbc198a | 352 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
353 | if (PageWriteback(page)) { |
354 | unlock_page(page); | |
355 | continue; | |
356 | } | |
aa65c29c JK |
357 | if (page->mapping != mapping) { |
358 | unlock_page(page); | |
359 | continue; | |
360 | } | |
361 | pagevec_add(&locked_pvec, page); | |
1da177e4 | 362 | } |
aa65c29c JK |
363 | for (i = 0; i < pagevec_count(&locked_pvec); i++) |
364 | truncate_cleanup_page(mapping, locked_pvec.pages[i]); | |
365 | delete_from_page_cache_batch(mapping, &locked_pvec); | |
366 | for (i = 0; i < pagevec_count(&locked_pvec); i++) | |
367 | unlock_page(locked_pvec.pages[i]); | |
f2187599 | 368 | truncate_exceptional_pvec_entries(mapping, &pvec, indices, end); |
1da177e4 LT |
369 | pagevec_release(&pvec); |
370 | cond_resched(); | |
b85e0eff | 371 | index++; |
1da177e4 | 372 | } |
5a720394 | 373 | if (partial_start) { |
1da177e4 LT |
374 | struct page *page = find_lock_page(mapping, start - 1); |
375 | if (page) { | |
09cbfeaf | 376 | unsigned int top = PAGE_SIZE; |
5a720394 LC |
377 | if (start > end) { |
378 | /* Truncation within a single page */ | |
379 | top = partial_end; | |
380 | partial_end = 0; | |
381 | } | |
1da177e4 | 382 | wait_on_page_writeback(page); |
5a720394 LC |
383 | zero_user_segment(page, partial_start, top); |
384 | cleancache_invalidate_page(mapping, page); | |
385 | if (page_has_private(page)) | |
386 | do_invalidatepage(page, partial_start, | |
387 | top - partial_start); | |
1da177e4 | 388 | unlock_page(page); |
09cbfeaf | 389 | put_page(page); |
1da177e4 LT |
390 | } |
391 | } | |
5a720394 LC |
392 | if (partial_end) { |
393 | struct page *page = find_lock_page(mapping, end); | |
394 | if (page) { | |
395 | wait_on_page_writeback(page); | |
396 | zero_user_segment(page, 0, partial_end); | |
397 | cleancache_invalidate_page(mapping, page); | |
398 | if (page_has_private(page)) | |
399 | do_invalidatepage(page, 0, | |
400 | partial_end); | |
401 | unlock_page(page); | |
09cbfeaf | 402 | put_page(page); |
5a720394 LC |
403 | } |
404 | } | |
405 | /* | |
406 | * If the truncation happened within a single page no pages | |
407 | * will be released, just zeroed, so we can bail out now. | |
408 | */ | |
409 | if (start >= end) | |
34ccb69e | 410 | goto out; |
1da177e4 | 411 | |
b85e0eff | 412 | index = start; |
1da177e4 LT |
413 | for ( ; ; ) { |
414 | cond_resched(); | |
0cd6144a | 415 | if (!pagevec_lookup_entries(&pvec, mapping, index, |
792ceaef HD |
416 | min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { |
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 | } | |
0cd6144a | 424 | if (index == start && indices[0] >= end) { |
792ceaef | 425 | /* All gone out of hole to be punched, we're done */ |
0cd6144a | 426 | pagevec_remove_exceptionals(&pvec); |
d7339071 HR |
427 | pagevec_release(&pvec); |
428 | break; | |
429 | } | |
f2187599 | 430 | |
1da177e4 LT |
431 | for (i = 0; i < pagevec_count(&pvec); i++) { |
432 | struct page *page = pvec.pages[i]; | |
433 | ||
b85e0eff | 434 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 435 | index = indices[i]; |
792ceaef HD |
436 | if (index >= end) { |
437 | /* Restart punch to make sure all gone */ | |
438 | index = start - 1; | |
d7339071 | 439 | break; |
792ceaef | 440 | } |
b85e0eff | 441 | |
3159f943 | 442 | if (xa_is_value(page)) |
0cd6144a | 443 | continue; |
0cd6144a | 444 | |
1da177e4 | 445 | lock_page(page); |
5cbc198a | 446 | WARN_ON(page_to_index(page) != index); |
1da177e4 | 447 | wait_on_page_writeback(page); |
750b4987 | 448 | truncate_inode_page(mapping, page); |
1da177e4 LT |
449 | unlock_page(page); |
450 | } | |
f2187599 | 451 | truncate_exceptional_pvec_entries(mapping, &pvec, indices, end); |
1da177e4 | 452 | pagevec_release(&pvec); |
b85e0eff | 453 | index++; |
1da177e4 | 454 | } |
34ccb69e AR |
455 | |
456 | out: | |
3167760f | 457 | cleancache_invalidate_inode(mapping); |
1da177e4 | 458 | } |
d7339071 | 459 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 460 | |
d7339071 HR |
461 | /** |
462 | * truncate_inode_pages - truncate *all* the pages from an offset | |
463 | * @mapping: mapping to truncate | |
464 | * @lstart: offset from which to truncate | |
465 | * | |
1b1dcc1b | 466 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
467 | * |
468 | * Note: When this function returns, there can be a page in the process of | |
469 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
470 | * mapping->nrpages can be non-zero when this function returns even after | |
471 | * truncation of the whole mapping. | |
d7339071 HR |
472 | */ |
473 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
474 | { | |
475 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
476 | } | |
1da177e4 LT |
477 | EXPORT_SYMBOL(truncate_inode_pages); |
478 | ||
91b0abe3 JW |
479 | /** |
480 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
481 | * @mapping: mapping to truncate | |
482 | * | |
483 | * Called under (and serialized by) inode->i_mutex. | |
484 | * | |
485 | * Filesystems have to use this in the .evict_inode path to inform the | |
486 | * VM that this is the final truncate and the inode is going away. | |
487 | */ | |
488 | void truncate_inode_pages_final(struct address_space *mapping) | |
489 | { | |
f9fe48be | 490 | unsigned long nrexceptional; |
91b0abe3 JW |
491 | unsigned long nrpages; |
492 | ||
493 | /* | |
494 | * Page reclaim can not participate in regular inode lifetime | |
495 | * management (can't call iput()) and thus can race with the | |
496 | * inode teardown. Tell it when the address space is exiting, | |
497 | * so that it does not install eviction information after the | |
498 | * final truncate has begun. | |
499 | */ | |
500 | mapping_set_exiting(mapping); | |
501 | ||
502 | /* | |
503 | * When reclaim installs eviction entries, it increases | |
f9fe48be | 504 | * nrexceptional first, then decreases nrpages. Make sure we see |
91b0abe3 JW |
505 | * this in the right order or we might miss an entry. |
506 | */ | |
507 | nrpages = mapping->nrpages; | |
508 | smp_rmb(); | |
f9fe48be | 509 | nrexceptional = mapping->nrexceptional; |
91b0abe3 | 510 | |
f9fe48be | 511 | if (nrpages || nrexceptional) { |
91b0abe3 JW |
512 | /* |
513 | * As truncation uses a lockless tree lookup, cycle | |
514 | * the tree lock to make sure any ongoing tree | |
515 | * modification that does not see AS_EXITING is | |
516 | * completed before starting the final truncate. | |
517 | */ | |
b93b0163 MW |
518 | xa_lock_irq(&mapping->i_pages); |
519 | xa_unlock_irq(&mapping->i_pages); | |
91b0abe3 | 520 | } |
6ff38bd4 PT |
521 | |
522 | /* | |
523 | * Cleancache needs notification even if there are no pages or shadow | |
524 | * entries. | |
525 | */ | |
526 | truncate_inode_pages(mapping, 0); | |
91b0abe3 JW |
527 | } |
528 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
529 | ||
28697355 MW |
530 | /** |
531 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
532 | * @mapping: the address_space which holds the pages to invalidate | |
533 | * @start: the offset 'from' which to invalidate | |
534 | * @end: the offset 'to' which to invalidate (inclusive) | |
535 | * | |
536 | * This function only removes the unlocked pages, if you want to | |
537 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
538 | * | |
539 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
540 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
541 | * pagetables. | |
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 | } |
fc127da0 KS |
592 | } |
593 | ||
31560180 | 594 | ret = invalidate_inode_page(page); |
1da177e4 | 595 | unlock_page(page); |
31560180 MK |
596 | /* |
597 | * Invalidation is a hint that the page is no longer | |
598 | * of interest and try to speed up its reclaim. | |
599 | */ | |
600 | if (!ret) | |
cc5993bd | 601 | deactivate_file_page(page); |
31560180 | 602 | count += ret; |
1da177e4 | 603 | } |
0cd6144a | 604 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 605 | pagevec_release(&pvec); |
28697355 | 606 | cond_resched(); |
b85e0eff | 607 | index++; |
1da177e4 | 608 | } |
31560180 | 609 | return count; |
1da177e4 | 610 | } |
54bc4855 | 611 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 612 | |
bd4c8ce4 AM |
613 | /* |
614 | * This is like invalidate_complete_page(), except it ignores the page's | |
615 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
616 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
617 | * shrink_page_list() has a temp ref on them, or because they're transiently |
618 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
619 | */ |
620 | static int | |
621 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
622 | { | |
c4843a75 GT |
623 | unsigned long flags; |
624 | ||
bd4c8ce4 AM |
625 | if (page->mapping != mapping) |
626 | return 0; | |
627 | ||
266cf658 | 628 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
629 | return 0; |
630 | ||
b93b0163 | 631 | xa_lock_irqsave(&mapping->i_pages, flags); |
bd4c8ce4 AM |
632 | if (PageDirty(page)) |
633 | goto failed; | |
634 | ||
266cf658 | 635 | BUG_ON(page_has_private(page)); |
62cccb8c | 636 | __delete_from_page_cache(page, NULL); |
b93b0163 | 637 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
6072d13c LT |
638 | |
639 | if (mapping->a_ops->freepage) | |
640 | mapping->a_ops->freepage(page); | |
641 | ||
09cbfeaf | 642 | put_page(page); /* pagecache ref */ |
bd4c8ce4 AM |
643 | return 1; |
644 | failed: | |
b93b0163 | 645 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
bd4c8ce4 AM |
646 | return 0; |
647 | } | |
648 | ||
e3db7691 TM |
649 | static int do_launder_page(struct address_space *mapping, struct page *page) |
650 | { | |
651 | if (!PageDirty(page)) | |
652 | return 0; | |
653 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
654 | return 0; | |
655 | return mapping->a_ops->launder_page(page); | |
656 | } | |
657 | ||
1da177e4 LT |
658 | /** |
659 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 660 | * @mapping: the address_space |
1da177e4 LT |
661 | * @start: the page offset 'from' which to invalidate |
662 | * @end: the page offset 'to' which to invalidate (inclusive) | |
663 | * | |
664 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
665 | * invalidation. | |
666 | * | |
6ccfa806 | 667 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
668 | */ |
669 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
670 | pgoff_t start, pgoff_t end) | |
671 | { | |
0cd6144a | 672 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 673 | struct pagevec pvec; |
b85e0eff | 674 | pgoff_t index; |
1da177e4 LT |
675 | int i; |
676 | int ret = 0; | |
0dd1334f | 677 | int ret2 = 0; |
1da177e4 | 678 | int did_range_unmap = 0; |
1da177e4 | 679 | |
32691f0f | 680 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
34ccb69e | 681 | goto out; |
32691f0f | 682 | |
86679820 | 683 | pagevec_init(&pvec); |
b85e0eff | 684 | index = start; |
0cd6144a JW |
685 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
686 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
687 | indices)) { | |
7b965e08 | 688 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 689 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
690 | |
691 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 692 | index = indices[i]; |
b85e0eff HD |
693 | if (index > end) |
694 | break; | |
1da177e4 | 695 | |
3159f943 | 696 | if (xa_is_value(page)) { |
c6dcf52c JK |
697 | if (!invalidate_exceptional_entry2(mapping, |
698 | index, page)) | |
699 | ret = -EBUSY; | |
0cd6144a JW |
700 | continue; |
701 | } | |
702 | ||
1da177e4 | 703 | lock_page(page); |
5cbc198a | 704 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
705 | if (page->mapping != mapping) { |
706 | unlock_page(page); | |
707 | continue; | |
708 | } | |
1da177e4 | 709 | wait_on_page_writeback(page); |
d00806b1 | 710 | if (page_mapped(page)) { |
1da177e4 LT |
711 | if (!did_range_unmap) { |
712 | /* | |
713 | * Zap the rest of the file in one hit. | |
714 | */ | |
977fbdcd MW |
715 | unmap_mapping_pages(mapping, index, |
716 | (1 + end - index), false); | |
1da177e4 LT |
717 | did_range_unmap = 1; |
718 | } else { | |
719 | /* | |
720 | * Just zap this page | |
721 | */ | |
977fbdcd MW |
722 | unmap_mapping_pages(mapping, index, |
723 | 1, false); | |
1da177e4 LT |
724 | } |
725 | } | |
d00806b1 | 726 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
727 | ret2 = do_launder_page(mapping, page); |
728 | if (ret2 == 0) { | |
729 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 730 | ret2 = -EBUSY; |
0dd1334f HH |
731 | } |
732 | if (ret2 < 0) | |
733 | ret = ret2; | |
1da177e4 LT |
734 | unlock_page(page); |
735 | } | |
0cd6144a | 736 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
737 | pagevec_release(&pvec); |
738 | cond_resched(); | |
b85e0eff | 739 | index++; |
1da177e4 | 740 | } |
cd656375 | 741 | /* |
69b6c131 | 742 | * For DAX we invalidate page tables after invalidating page cache. We |
cd656375 JK |
743 | * could invalidate page tables while invalidating each entry however |
744 | * that would be expensive. And doing range unmapping before doesn't | |
69b6c131 | 745 | * work as we have no cheap way to find whether page cache entry didn't |
cd656375 JK |
746 | * get remapped later. |
747 | */ | |
748 | if (dax_mapping(mapping)) { | |
977fbdcd | 749 | unmap_mapping_pages(mapping, start, end - start + 1, false); |
cd656375 | 750 | } |
34ccb69e | 751 | out: |
3167760f | 752 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
753 | return ret; |
754 | } | |
755 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
756 | ||
757 | /** | |
758 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 759 | * @mapping: the address_space |
1da177e4 LT |
760 | * |
761 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
762 | * invalidation. | |
763 | * | |
e9de25dd | 764 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
765 | */ |
766 | int invalidate_inode_pages2(struct address_space *mapping) | |
767 | { | |
768 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
769 | } | |
770 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 771 | |
772 | /** | |
773 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
774 | * @inode: inode | |
8a549bea | 775 | * @newsize: new file size |
25d9e2d1 | 776 | * |
777 | * inode's new i_size must already be written before truncate_pagecache | |
778 | * is called. | |
779 | * | |
780 | * This function should typically be called before the filesystem | |
781 | * releases resources associated with the freed range (eg. deallocates | |
782 | * blocks). This way, pagecache will always stay logically coherent | |
783 | * with on-disk format, and the filesystem would not have to deal with | |
784 | * situations such as writepage being called for a page that has already | |
785 | * had its underlying blocks deallocated. | |
786 | */ | |
7caef267 | 787 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 788 | { |
cedabed4 | 789 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 790 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
791 | |
792 | /* | |
793 | * unmap_mapping_range is called twice, first simply for | |
794 | * efficiency so that truncate_inode_pages does fewer | |
795 | * single-page unmaps. However after this first call, and | |
796 | * before truncate_inode_pages finishes, it is possible for | |
797 | * private pages to be COWed, which remain after | |
798 | * truncate_inode_pages finishes, hence the second | |
799 | * unmap_mapping_range call must be made for correctness. | |
800 | */ | |
8a549bea HD |
801 | unmap_mapping_range(mapping, holebegin, 0, 1); |
802 | truncate_inode_pages(mapping, newsize); | |
803 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 804 | } |
805 | EXPORT_SYMBOL(truncate_pagecache); | |
806 | ||
2c27c65e CH |
807 | /** |
808 | * truncate_setsize - update inode and pagecache for a new file size | |
809 | * @inode: inode | |
810 | * @newsize: new file size | |
811 | * | |
382e27da JK |
812 | * truncate_setsize updates i_size and performs pagecache truncation (if |
813 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
814 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 815 | * |
77783d06 JK |
816 | * Must be called with a lock serializing truncates and writes (generally |
817 | * i_mutex but e.g. xfs uses a different lock) and before all filesystem | |
818 | * specific block truncation has been performed. | |
2c27c65e CH |
819 | */ |
820 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
821 | { | |
90a80202 JK |
822 | loff_t oldsize = inode->i_size; |
823 | ||
2c27c65e | 824 | i_size_write(inode, newsize); |
90a80202 JK |
825 | if (newsize > oldsize) |
826 | pagecache_isize_extended(inode, oldsize, newsize); | |
7caef267 | 827 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
828 | } |
829 | EXPORT_SYMBOL(truncate_setsize); | |
830 | ||
90a80202 JK |
831 | /** |
832 | * pagecache_isize_extended - update pagecache after extension of i_size | |
833 | * @inode: inode for which i_size was extended | |
834 | * @from: original inode size | |
835 | * @to: new inode size | |
836 | * | |
837 | * Handle extension of inode size either caused by extending truncate or by | |
838 | * write starting after current i_size. We mark the page straddling current | |
839 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
840 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
841 | * the page before user writes to the page via mmap after the i_size has been | |
842 | * changed. | |
843 | * | |
844 | * The function must be called after i_size is updated so that page fault | |
845 | * coming after we unlock the page will already see the new i_size. | |
846 | * The function must be called while we still hold i_mutex - this not only | |
847 | * makes sure i_size is stable but also that userspace cannot observe new | |
848 | * i_size value before we are prepared to store mmap writes at new inode size. | |
849 | */ | |
850 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
851 | { | |
93407472 | 852 | int bsize = i_blocksize(inode); |
90a80202 JK |
853 | loff_t rounded_from; |
854 | struct page *page; | |
855 | pgoff_t index; | |
856 | ||
90a80202 JK |
857 | WARN_ON(to > inode->i_size); |
858 | ||
09cbfeaf | 859 | if (from >= to || bsize == PAGE_SIZE) |
90a80202 JK |
860 | return; |
861 | /* Page straddling @from will not have any hole block created? */ | |
862 | rounded_from = round_up(from, bsize); | |
09cbfeaf | 863 | if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) |
90a80202 JK |
864 | return; |
865 | ||
09cbfeaf | 866 | index = from >> PAGE_SHIFT; |
90a80202 JK |
867 | page = find_lock_page(inode->i_mapping, index); |
868 | /* Page not cached? Nothing to do */ | |
869 | if (!page) | |
870 | return; | |
871 | /* | |
872 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
873 | * is needed. | |
874 | */ | |
875 | if (page_mkclean(page)) | |
876 | set_page_dirty(page); | |
877 | unlock_page(page); | |
09cbfeaf | 878 | put_page(page); |
90a80202 JK |
879 | } |
880 | EXPORT_SYMBOL(pagecache_isize_extended); | |
881 | ||
623e3db9 HD |
882 | /** |
883 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
884 | * @inode: inode | |
885 | * @lstart: offset of beginning of hole | |
886 | * @lend: offset of last byte of hole | |
887 | * | |
888 | * This function should typically be called before the filesystem | |
889 | * releases resources associated with the freed range (eg. deallocates | |
890 | * blocks). This way, pagecache will always stay logically coherent | |
891 | * with on-disk format, and the filesystem would not have to deal with | |
892 | * situations such as writepage being called for a page that has already | |
893 | * had its underlying blocks deallocated. | |
894 | */ | |
895 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
896 | { | |
897 | struct address_space *mapping = inode->i_mapping; | |
898 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
899 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
900 | /* | |
901 | * This rounding is currently just for example: unmap_mapping_range | |
902 | * expands its hole outwards, whereas we want it to contract the hole | |
903 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
904 | * doing their own page rounding first. Note that unmap_mapping_range |
905 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
906 | */ |
907 | ||
908 | /* | |
909 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
910 | * once (before truncating pagecache), and without "even_cows" flag: | |
911 | * hole-punching should not remove private COWed pages from the hole. | |
912 | */ | |
913 | if ((u64)unmap_end > (u64)unmap_start) | |
914 | unmap_mapping_range(mapping, unmap_start, | |
915 | 1 + unmap_end - unmap_start, 0); | |
916 | truncate_inode_pages_range(mapping, lstart, lend); | |
917 | } | |
918 | EXPORT_SYMBOL(truncate_pagecache_range); |