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 | { |
449dd698 JW |
36 | struct radix_tree_node *node; |
37 | void **slot; | |
38 | ||
14b46879 | 39 | if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot)) |
f2187599 | 40 | return; |
ac401cc7 | 41 | if (*slot != entry) |
f2187599 | 42 | return; |
14b46879 | 43 | __radix_tree_replace(&mapping->page_tree, node, slot, NULL, |
c7df8ad2 | 44 | workingset_update_node); |
ac401cc7 | 45 | mapping->nrexceptional--; |
f2187599 MG |
46 | } |
47 | ||
48 | static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, | |
49 | void *entry) | |
50 | { | |
51 | spin_lock_irq(&mapping->tree_lock); | |
52 | __clear_shadow_entry(mapping, index, entry); | |
0cd6144a JW |
53 | spin_unlock_irq(&mapping->tree_lock); |
54 | } | |
1da177e4 | 55 | |
c6dcf52c | 56 | /* |
f2187599 MG |
57 | * Unconditionally remove exceptional entries. Usually called from truncate |
58 | * path. Note that the pagevec may be altered by this function by removing | |
59 | * exceptional entries similar to what pagevec_remove_exceptionals does. | |
c6dcf52c | 60 | */ |
f2187599 MG |
61 | static void truncate_exceptional_pvec_entries(struct address_space *mapping, |
62 | struct pagevec *pvec, pgoff_t *indices, | |
63 | pgoff_t end) | |
c6dcf52c | 64 | { |
f2187599 MG |
65 | int i, j; |
66 | bool dax, lock; | |
67 | ||
c6dcf52c JK |
68 | /* Handled by shmem itself */ |
69 | if (shmem_mapping(mapping)) | |
70 | return; | |
71 | ||
f2187599 MG |
72 | for (j = 0; j < pagevec_count(pvec); j++) |
73 | if (radix_tree_exceptional_entry(pvec->pages[j])) | |
74 | break; | |
75 | ||
76 | if (j == pagevec_count(pvec)) | |
c6dcf52c | 77 | return; |
f2187599 MG |
78 | |
79 | dax = dax_mapping(mapping); | |
80 | lock = !dax && indices[j] < end; | |
81 | if (lock) | |
82 | spin_lock_irq(&mapping->tree_lock); | |
83 | ||
84 | for (i = j; i < pagevec_count(pvec); i++) { | |
85 | struct page *page = pvec->pages[i]; | |
86 | pgoff_t index = indices[i]; | |
87 | ||
88 | if (!radix_tree_exceptional_entry(page)) { | |
89 | pvec->pages[j++] = page; | |
90 | continue; | |
91 | } | |
92 | ||
93 | if (index >= end) | |
94 | continue; | |
95 | ||
96 | if (unlikely(dax)) { | |
97 | dax_delete_mapping_entry(mapping, index); | |
98 | continue; | |
99 | } | |
100 | ||
101 | __clear_shadow_entry(mapping, index, page); | |
c6dcf52c | 102 | } |
f2187599 MG |
103 | |
104 | if (lock) | |
105 | spin_unlock_irq(&mapping->tree_lock); | |
106 | pvec->nr = j; | |
c6dcf52c JK |
107 | } |
108 | ||
109 | /* | |
110 | * Invalidate exceptional entry if easily possible. This handles exceptional | |
4636e70b | 111 | * entries for invalidate_inode_pages(). |
c6dcf52c JK |
112 | */ |
113 | static int invalidate_exceptional_entry(struct address_space *mapping, | |
114 | pgoff_t index, void *entry) | |
115 | { | |
4636e70b RZ |
116 | /* Handled by shmem itself, or for DAX we do nothing. */ |
117 | if (shmem_mapping(mapping) || dax_mapping(mapping)) | |
c6dcf52c | 118 | return 1; |
c6dcf52c JK |
119 | clear_shadow_entry(mapping, index, entry); |
120 | return 1; | |
121 | } | |
122 | ||
123 | /* | |
124 | * Invalidate exceptional entry if clean. This handles exceptional entries for | |
125 | * invalidate_inode_pages2() so for DAX it evicts only clean entries. | |
126 | */ | |
127 | static int invalidate_exceptional_entry2(struct address_space *mapping, | |
128 | pgoff_t index, void *entry) | |
129 | { | |
130 | /* Handled by shmem itself */ | |
131 | if (shmem_mapping(mapping)) | |
132 | return 1; | |
133 | if (dax_mapping(mapping)) | |
134 | return dax_invalidate_mapping_entry_sync(mapping, index); | |
135 | clear_shadow_entry(mapping, index, entry); | |
136 | return 1; | |
137 | } | |
138 | ||
cf9a2ae8 | 139 | /** |
28bc44d7 | 140 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 141 | * @page: the page which is affected |
d47992f8 LC |
142 | * @offset: start of the range to invalidate |
143 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
144 | * |
145 | * do_invalidatepage() is called when all or part of the page has become | |
146 | * invalidated by a truncate operation. | |
147 | * | |
148 | * do_invalidatepage() does not have to release all buffers, but it must | |
149 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
150 | * is underway against any of the blocks which are outside the truncation | |
151 | * point. Because the caller is about to free (and possibly reuse) those | |
152 | * blocks on-disk. | |
153 | */ | |
d47992f8 LC |
154 | void do_invalidatepage(struct page *page, unsigned int offset, |
155 | unsigned int length) | |
cf9a2ae8 | 156 | { |
d47992f8 LC |
157 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
158 | ||
cf9a2ae8 | 159 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 160 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
161 | if (!invalidatepage) |
162 | invalidatepage = block_invalidatepage; | |
9361401e | 163 | #endif |
cf9a2ae8 | 164 | if (invalidatepage) |
d47992f8 | 165 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
166 | } |
167 | ||
1da177e4 LT |
168 | /* |
169 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 170 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 171 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
172 | * |
173 | * We need to bale out if page->mapping is no longer equal to the original | |
174 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
fc0ecff6 | 175 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
176 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
177 | */ | |
9f4e41f4 JK |
178 | static void |
179 | truncate_cleanup_page(struct address_space *mapping, struct page *page) | |
1da177e4 | 180 | { |
9f4e41f4 JK |
181 | if (page_mapped(page)) { |
182 | loff_t holelen; | |
183 | ||
184 | holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE; | |
185 | unmap_mapping_range(mapping, | |
186 | (loff_t)page->index << PAGE_SHIFT, | |
187 | holelen, 0); | |
188 | } | |
1da177e4 | 189 | |
266cf658 | 190 | if (page_has_private(page)) |
09cbfeaf | 191 | do_invalidatepage(page, 0, PAGE_SIZE); |
1da177e4 | 192 | |
b9ea2515 KK |
193 | /* |
194 | * Some filesystems seem to re-dirty the page even after | |
195 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
196 | * Hence dirty accounting check is placed after invalidation. | |
197 | */ | |
11f81bec | 198 | cancel_dirty_page(page); |
1da177e4 | 199 | ClearPageMappedToDisk(page); |
1da177e4 LT |
200 | } |
201 | ||
202 | /* | |
fc0ecff6 | 203 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 204 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
205 | * be marked dirty at any time too, so use remove_mapping which safely |
206 | * discards clean, unused pages. | |
1da177e4 LT |
207 | * |
208 | * Returns non-zero if the page was successfully invalidated. | |
209 | */ | |
210 | static int | |
211 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
212 | { | |
0fd0e6b0 NP |
213 | int ret; |
214 | ||
1da177e4 LT |
215 | if (page->mapping != mapping) |
216 | return 0; | |
217 | ||
266cf658 | 218 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
219 | return 0; |
220 | ||
0fd0e6b0 | 221 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
222 | |
223 | return ret; | |
1da177e4 LT |
224 | } |
225 | ||
750b4987 NP |
226 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
227 | { | |
fc127da0 KS |
228 | VM_BUG_ON_PAGE(PageTail(page), page); |
229 | ||
9f4e41f4 JK |
230 | if (page->mapping != mapping) |
231 | return -EIO; | |
232 | ||
233 | truncate_cleanup_page(mapping, page); | |
234 | delete_from_page_cache(page); | |
235 | return 0; | |
750b4987 NP |
236 | } |
237 | ||
25718736 AK |
238 | /* |
239 | * Used to get rid of pages on hardware memory corruption. | |
240 | */ | |
241 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
242 | { | |
243 | if (!mapping) | |
244 | return -EINVAL; | |
245 | /* | |
246 | * Only punch for normal data pages for now. | |
247 | * Handling other types like directories would need more auditing. | |
248 | */ | |
249 | if (!S_ISREG(mapping->host->i_mode)) | |
250 | return -EIO; | |
251 | return truncate_inode_page(mapping, page); | |
252 | } | |
253 | EXPORT_SYMBOL(generic_error_remove_page); | |
254 | ||
83f78668 WF |
255 | /* |
256 | * Safely invalidate one page from its pagecache mapping. | |
257 | * It only drops clean, unused pages. The page must be locked. | |
258 | * | |
259 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
260 | */ | |
261 | int invalidate_inode_page(struct page *page) | |
262 | { | |
263 | struct address_space *mapping = page_mapping(page); | |
264 | if (!mapping) | |
265 | return 0; | |
266 | if (PageDirty(page) || PageWriteback(page)) | |
267 | return 0; | |
268 | if (page_mapped(page)) | |
269 | return 0; | |
270 | return invalidate_complete_page(mapping, page); | |
271 | } | |
272 | ||
1da177e4 | 273 | /** |
73c1e204 | 274 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
275 | * @mapping: mapping to truncate |
276 | * @lstart: offset from which to truncate | |
5a720394 | 277 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 278 | * |
d7339071 | 279 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
280 | * specified offsets (and zeroing out partial pages |
281 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
282 | * |
283 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
284 | * block on page locks and it will not block on writeback. The second pass | |
285 | * will wait. This is to prevent as much IO as possible in the affected region. | |
286 | * The first pass will remove most pages, so the search cost of the second pass | |
287 | * is low. | |
288 | * | |
1da177e4 LT |
289 | * We pass down the cache-hot hint to the page freeing code. Even if the |
290 | * mapping is large, it is probably the case that the final pages are the most | |
291 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
292 | * |
293 | * Note that since ->invalidatepage() accepts range to invalidate | |
294 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
295 | * page aligned properly. | |
1da177e4 | 296 | */ |
d7339071 HR |
297 | void truncate_inode_pages_range(struct address_space *mapping, |
298 | loff_t lstart, loff_t lend) | |
1da177e4 | 299 | { |
5a720394 LC |
300 | pgoff_t start; /* inclusive */ |
301 | pgoff_t end; /* exclusive */ | |
302 | unsigned int partial_start; /* inclusive */ | |
303 | unsigned int partial_end; /* exclusive */ | |
304 | struct pagevec pvec; | |
0cd6144a | 305 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
306 | pgoff_t index; |
307 | int i; | |
1da177e4 | 308 | |
f9fe48be | 309 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
34ccb69e | 310 | goto out; |
1da177e4 | 311 | |
5a720394 | 312 | /* Offsets within partial pages */ |
09cbfeaf KS |
313 | partial_start = lstart & (PAGE_SIZE - 1); |
314 | partial_end = (lend + 1) & (PAGE_SIZE - 1); | |
5a720394 LC |
315 | |
316 | /* | |
317 | * 'start' and 'end' always covers the range of pages to be fully | |
318 | * truncated. Partial pages are covered with 'partial_start' at the | |
319 | * start of the range and 'partial_end' at the end of the range. | |
320 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
321 | */ | |
09cbfeaf | 322 | start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; |
5a720394 LC |
323 | if (lend == -1) |
324 | /* | |
325 | * lend == -1 indicates end-of-file so we have to set 'end' | |
326 | * to the highest possible pgoff_t and since the type is | |
327 | * unsigned we're using -1. | |
328 | */ | |
329 | end = -1; | |
330 | else | |
09cbfeaf | 331 | end = (lend + 1) >> PAGE_SHIFT; |
d7339071 | 332 | |
86679820 | 333 | pagevec_init(&pvec); |
b85e0eff | 334 | index = start; |
0cd6144a JW |
335 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
336 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
337 | indices)) { | |
aa65c29c JK |
338 | /* |
339 | * Pagevec array has exceptional entries and we may also fail | |
340 | * to lock some pages. So we store pages that can be deleted | |
341 | * in a new pagevec. | |
342 | */ | |
343 | struct pagevec locked_pvec; | |
344 | ||
86679820 | 345 | pagevec_init(&locked_pvec); |
1da177e4 LT |
346 | for (i = 0; i < pagevec_count(&pvec); i++) { |
347 | struct page *page = pvec.pages[i]; | |
1da177e4 | 348 | |
b85e0eff | 349 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 350 | index = indices[i]; |
5a720394 | 351 | if (index >= end) |
d7339071 | 352 | break; |
d7339071 | 353 | |
f2187599 | 354 | if (radix_tree_exceptional_entry(page)) |
0cd6144a | 355 | continue; |
0cd6144a | 356 | |
529ae9aa | 357 | if (!trylock_page(page)) |
1da177e4 | 358 | continue; |
5cbc198a | 359 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
360 | if (PageWriteback(page)) { |
361 | unlock_page(page); | |
362 | continue; | |
363 | } | |
aa65c29c JK |
364 | if (page->mapping != mapping) { |
365 | unlock_page(page); | |
366 | continue; | |
367 | } | |
368 | pagevec_add(&locked_pvec, page); | |
1da177e4 | 369 | } |
aa65c29c JK |
370 | for (i = 0; i < pagevec_count(&locked_pvec); i++) |
371 | truncate_cleanup_page(mapping, locked_pvec.pages[i]); | |
372 | delete_from_page_cache_batch(mapping, &locked_pvec); | |
373 | for (i = 0; i < pagevec_count(&locked_pvec); i++) | |
374 | unlock_page(locked_pvec.pages[i]); | |
f2187599 | 375 | truncate_exceptional_pvec_entries(mapping, &pvec, indices, end); |
1da177e4 LT |
376 | pagevec_release(&pvec); |
377 | cond_resched(); | |
b85e0eff | 378 | index++; |
1da177e4 | 379 | } |
5a720394 | 380 | if (partial_start) { |
1da177e4 LT |
381 | struct page *page = find_lock_page(mapping, start - 1); |
382 | if (page) { | |
09cbfeaf | 383 | unsigned int top = PAGE_SIZE; |
5a720394 LC |
384 | if (start > end) { |
385 | /* Truncation within a single page */ | |
386 | top = partial_end; | |
387 | partial_end = 0; | |
388 | } | |
1da177e4 | 389 | wait_on_page_writeback(page); |
5a720394 LC |
390 | zero_user_segment(page, partial_start, top); |
391 | cleancache_invalidate_page(mapping, page); | |
392 | if (page_has_private(page)) | |
393 | do_invalidatepage(page, partial_start, | |
394 | top - partial_start); | |
1da177e4 | 395 | unlock_page(page); |
09cbfeaf | 396 | put_page(page); |
1da177e4 LT |
397 | } |
398 | } | |
5a720394 LC |
399 | if (partial_end) { |
400 | struct page *page = find_lock_page(mapping, end); | |
401 | if (page) { | |
402 | wait_on_page_writeback(page); | |
403 | zero_user_segment(page, 0, partial_end); | |
404 | cleancache_invalidate_page(mapping, page); | |
405 | if (page_has_private(page)) | |
406 | do_invalidatepage(page, 0, | |
407 | partial_end); | |
408 | unlock_page(page); | |
09cbfeaf | 409 | put_page(page); |
5a720394 LC |
410 | } |
411 | } | |
412 | /* | |
413 | * If the truncation happened within a single page no pages | |
414 | * will be released, just zeroed, so we can bail out now. | |
415 | */ | |
416 | if (start >= end) | |
34ccb69e | 417 | goto out; |
1da177e4 | 418 | |
b85e0eff | 419 | index = start; |
1da177e4 LT |
420 | for ( ; ; ) { |
421 | cond_resched(); | |
0cd6144a | 422 | if (!pagevec_lookup_entries(&pvec, mapping, index, |
792ceaef HD |
423 | min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { |
424 | /* If all gone from start onwards, we're done */ | |
b85e0eff | 425 | if (index == start) |
1da177e4 | 426 | break; |
792ceaef | 427 | /* Otherwise restart to make sure all gone */ |
b85e0eff | 428 | index = start; |
1da177e4 LT |
429 | continue; |
430 | } | |
0cd6144a | 431 | if (index == start && indices[0] >= end) { |
792ceaef | 432 | /* All gone out of hole to be punched, we're done */ |
0cd6144a | 433 | pagevec_remove_exceptionals(&pvec); |
d7339071 HR |
434 | pagevec_release(&pvec); |
435 | break; | |
436 | } | |
f2187599 | 437 | |
1da177e4 LT |
438 | for (i = 0; i < pagevec_count(&pvec); i++) { |
439 | struct page *page = pvec.pages[i]; | |
440 | ||
b85e0eff | 441 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 442 | index = indices[i]; |
792ceaef HD |
443 | if (index >= end) { |
444 | /* Restart punch to make sure all gone */ | |
445 | index = start - 1; | |
d7339071 | 446 | break; |
792ceaef | 447 | } |
b85e0eff | 448 | |
f2187599 | 449 | if (radix_tree_exceptional_entry(page)) |
0cd6144a | 450 | continue; |
0cd6144a | 451 | |
1da177e4 | 452 | lock_page(page); |
5cbc198a | 453 | WARN_ON(page_to_index(page) != index); |
1da177e4 | 454 | wait_on_page_writeback(page); |
750b4987 | 455 | truncate_inode_page(mapping, page); |
1da177e4 LT |
456 | unlock_page(page); |
457 | } | |
f2187599 | 458 | truncate_exceptional_pvec_entries(mapping, &pvec, indices, end); |
1da177e4 | 459 | pagevec_release(&pvec); |
b85e0eff | 460 | index++; |
1da177e4 | 461 | } |
34ccb69e AR |
462 | |
463 | out: | |
3167760f | 464 | cleancache_invalidate_inode(mapping); |
1da177e4 | 465 | } |
d7339071 | 466 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 467 | |
d7339071 HR |
468 | /** |
469 | * truncate_inode_pages - truncate *all* the pages from an offset | |
470 | * @mapping: mapping to truncate | |
471 | * @lstart: offset from which to truncate | |
472 | * | |
1b1dcc1b | 473 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
474 | * |
475 | * Note: When this function returns, there can be a page in the process of | |
476 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
477 | * mapping->nrpages can be non-zero when this function returns even after | |
478 | * truncation of the whole mapping. | |
d7339071 HR |
479 | */ |
480 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
481 | { | |
482 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
483 | } | |
1da177e4 LT |
484 | EXPORT_SYMBOL(truncate_inode_pages); |
485 | ||
91b0abe3 JW |
486 | /** |
487 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
488 | * @mapping: mapping to truncate | |
489 | * | |
490 | * Called under (and serialized by) inode->i_mutex. | |
491 | * | |
492 | * Filesystems have to use this in the .evict_inode path to inform the | |
493 | * VM that this is the final truncate and the inode is going away. | |
494 | */ | |
495 | void truncate_inode_pages_final(struct address_space *mapping) | |
496 | { | |
f9fe48be | 497 | unsigned long nrexceptional; |
91b0abe3 JW |
498 | unsigned long nrpages; |
499 | ||
500 | /* | |
501 | * Page reclaim can not participate in regular inode lifetime | |
502 | * management (can't call iput()) and thus can race with the | |
503 | * inode teardown. Tell it when the address space is exiting, | |
504 | * so that it does not install eviction information after the | |
505 | * final truncate has begun. | |
506 | */ | |
507 | mapping_set_exiting(mapping); | |
508 | ||
509 | /* | |
510 | * When reclaim installs eviction entries, it increases | |
f9fe48be | 511 | * nrexceptional first, then decreases nrpages. Make sure we see |
91b0abe3 JW |
512 | * this in the right order or we might miss an entry. |
513 | */ | |
514 | nrpages = mapping->nrpages; | |
515 | smp_rmb(); | |
f9fe48be | 516 | nrexceptional = mapping->nrexceptional; |
91b0abe3 | 517 | |
f9fe48be | 518 | if (nrpages || nrexceptional) { |
91b0abe3 JW |
519 | /* |
520 | * As truncation uses a lockless tree lookup, cycle | |
521 | * the tree lock to make sure any ongoing tree | |
522 | * modification that does not see AS_EXITING is | |
523 | * completed before starting the final truncate. | |
524 | */ | |
525 | spin_lock_irq(&mapping->tree_lock); | |
526 | spin_unlock_irq(&mapping->tree_lock); | |
527 | ||
528 | truncate_inode_pages(mapping, 0); | |
529 | } | |
530 | } | |
531 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
532 | ||
28697355 MW |
533 | /** |
534 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
535 | * @mapping: the address_space which holds the pages to invalidate | |
536 | * @start: the offset 'from' which to invalidate | |
537 | * @end: the offset 'to' which to invalidate (inclusive) | |
538 | * | |
539 | * This function only removes the unlocked pages, if you want to | |
540 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
541 | * | |
542 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
543 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
544 | * pagetables. | |
545 | */ | |
546 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 547 | pgoff_t start, pgoff_t end) |
1da177e4 | 548 | { |
0cd6144a | 549 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 550 | struct pagevec pvec; |
b85e0eff | 551 | pgoff_t index = start; |
31560180 MK |
552 | unsigned long ret; |
553 | unsigned long count = 0; | |
1da177e4 LT |
554 | int i; |
555 | ||
86679820 | 556 | pagevec_init(&pvec); |
0cd6144a JW |
557 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
558 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
559 | indices)) { | |
1da177e4 LT |
560 | for (i = 0; i < pagevec_count(&pvec); i++) { |
561 | struct page *page = pvec.pages[i]; | |
e0f23603 | 562 | |
b85e0eff | 563 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 564 | index = indices[i]; |
b85e0eff HD |
565 | if (index > end) |
566 | break; | |
e0f23603 | 567 | |
0cd6144a | 568 | if (radix_tree_exceptional_entry(page)) { |
c6dcf52c JK |
569 | invalidate_exceptional_entry(mapping, index, |
570 | page); | |
0cd6144a JW |
571 | continue; |
572 | } | |
573 | ||
b85e0eff HD |
574 | if (!trylock_page(page)) |
575 | continue; | |
fc127da0 | 576 | |
5cbc198a | 577 | WARN_ON(page_to_index(page) != index); |
fc127da0 KS |
578 | |
579 | /* Middle of THP: skip */ | |
580 | if (PageTransTail(page)) { | |
581 | unlock_page(page); | |
582 | continue; | |
583 | } else if (PageTransHuge(page)) { | |
584 | index += HPAGE_PMD_NR - 1; | |
585 | i += HPAGE_PMD_NR - 1; | |
76b6f9b7 JK |
586 | /* |
587 | * 'end' is in the middle of THP. Don't | |
588 | * invalidate the page as the part outside of | |
589 | * 'end' could be still useful. | |
590 | */ | |
591 | if (index > end) { | |
592 | unlock_page(page); | |
fc127da0 | 593 | continue; |
76b6f9b7 | 594 | } |
fc127da0 KS |
595 | } |
596 | ||
31560180 | 597 | ret = invalidate_inode_page(page); |
1da177e4 | 598 | unlock_page(page); |
31560180 MK |
599 | /* |
600 | * Invalidation is a hint that the page is no longer | |
601 | * of interest and try to speed up its reclaim. | |
602 | */ | |
603 | if (!ret) | |
cc5993bd | 604 | deactivate_file_page(page); |
31560180 | 605 | count += ret; |
1da177e4 | 606 | } |
0cd6144a | 607 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 608 | pagevec_release(&pvec); |
28697355 | 609 | cond_resched(); |
b85e0eff | 610 | index++; |
1da177e4 | 611 | } |
31560180 | 612 | return count; |
1da177e4 | 613 | } |
54bc4855 | 614 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 615 | |
bd4c8ce4 AM |
616 | /* |
617 | * This is like invalidate_complete_page(), except it ignores the page's | |
618 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
619 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
620 | * shrink_page_list() has a temp ref on them, or because they're transiently |
621 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
622 | */ |
623 | static int | |
624 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
625 | { | |
c4843a75 GT |
626 | unsigned long flags; |
627 | ||
bd4c8ce4 AM |
628 | if (page->mapping != mapping) |
629 | return 0; | |
630 | ||
266cf658 | 631 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
632 | return 0; |
633 | ||
c4843a75 | 634 | spin_lock_irqsave(&mapping->tree_lock, flags); |
bd4c8ce4 AM |
635 | if (PageDirty(page)) |
636 | goto failed; | |
637 | ||
266cf658 | 638 | BUG_ON(page_has_private(page)); |
62cccb8c | 639 | __delete_from_page_cache(page, NULL); |
c4843a75 | 640 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6072d13c LT |
641 | |
642 | if (mapping->a_ops->freepage) | |
643 | mapping->a_ops->freepage(page); | |
644 | ||
09cbfeaf | 645 | put_page(page); /* pagecache ref */ |
bd4c8ce4 AM |
646 | return 1; |
647 | failed: | |
c4843a75 | 648 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
bd4c8ce4 AM |
649 | return 0; |
650 | } | |
651 | ||
e3db7691 TM |
652 | static int do_launder_page(struct address_space *mapping, struct page *page) |
653 | { | |
654 | if (!PageDirty(page)) | |
655 | return 0; | |
656 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
657 | return 0; | |
658 | return mapping->a_ops->launder_page(page); | |
659 | } | |
660 | ||
1da177e4 LT |
661 | /** |
662 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 663 | * @mapping: the address_space |
1da177e4 LT |
664 | * @start: the page offset 'from' which to invalidate |
665 | * @end: the page offset 'to' which to invalidate (inclusive) | |
666 | * | |
667 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
668 | * invalidation. | |
669 | * | |
6ccfa806 | 670 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
671 | */ |
672 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
673 | pgoff_t start, pgoff_t end) | |
674 | { | |
0cd6144a | 675 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 676 | struct pagevec pvec; |
b85e0eff | 677 | pgoff_t index; |
1da177e4 LT |
678 | int i; |
679 | int ret = 0; | |
0dd1334f | 680 | int ret2 = 0; |
1da177e4 | 681 | int did_range_unmap = 0; |
1da177e4 | 682 | |
32691f0f | 683 | if (mapping->nrpages == 0 && mapping->nrexceptional == 0) |
34ccb69e | 684 | goto out; |
32691f0f | 685 | |
86679820 | 686 | pagevec_init(&pvec); |
b85e0eff | 687 | index = start; |
0cd6144a JW |
688 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
689 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
690 | indices)) { | |
7b965e08 | 691 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 692 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
693 | |
694 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 695 | index = indices[i]; |
b85e0eff HD |
696 | if (index > end) |
697 | break; | |
1da177e4 | 698 | |
0cd6144a | 699 | if (radix_tree_exceptional_entry(page)) { |
c6dcf52c JK |
700 | if (!invalidate_exceptional_entry2(mapping, |
701 | index, page)) | |
702 | ret = -EBUSY; | |
0cd6144a JW |
703 | continue; |
704 | } | |
705 | ||
1da177e4 | 706 | lock_page(page); |
5cbc198a | 707 | WARN_ON(page_to_index(page) != index); |
1da177e4 LT |
708 | if (page->mapping != mapping) { |
709 | unlock_page(page); | |
710 | continue; | |
711 | } | |
1da177e4 | 712 | wait_on_page_writeback(page); |
d00806b1 | 713 | if (page_mapped(page)) { |
1da177e4 LT |
714 | if (!did_range_unmap) { |
715 | /* | |
716 | * Zap the rest of the file in one hit. | |
717 | */ | |
718 | unmap_mapping_range(mapping, | |
09cbfeaf | 719 | (loff_t)index << PAGE_SHIFT, |
b85e0eff | 720 | (loff_t)(1 + end - index) |
09cbfeaf KS |
721 | << PAGE_SHIFT, |
722 | 0); | |
1da177e4 LT |
723 | did_range_unmap = 1; |
724 | } else { | |
725 | /* | |
726 | * Just zap this page | |
727 | */ | |
728 | unmap_mapping_range(mapping, | |
09cbfeaf KS |
729 | (loff_t)index << PAGE_SHIFT, |
730 | PAGE_SIZE, 0); | |
1da177e4 LT |
731 | } |
732 | } | |
d00806b1 | 733 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
734 | ret2 = do_launder_page(mapping, page); |
735 | if (ret2 == 0) { | |
736 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 737 | ret2 = -EBUSY; |
0dd1334f HH |
738 | } |
739 | if (ret2 < 0) | |
740 | ret = ret2; | |
1da177e4 LT |
741 | unlock_page(page); |
742 | } | |
0cd6144a | 743 | pagevec_remove_exceptionals(&pvec); |
1da177e4 LT |
744 | pagevec_release(&pvec); |
745 | cond_resched(); | |
b85e0eff | 746 | index++; |
1da177e4 | 747 | } |
cd656375 JK |
748 | /* |
749 | * For DAX we invalidate page tables after invalidating radix tree. We | |
750 | * could invalidate page tables while invalidating each entry however | |
751 | * that would be expensive. And doing range unmapping before doesn't | |
752 | * work as we have no cheap way to find whether radix tree entry didn't | |
753 | * get remapped later. | |
754 | */ | |
755 | if (dax_mapping(mapping)) { | |
756 | unmap_mapping_range(mapping, (loff_t)start << PAGE_SHIFT, | |
757 | (loff_t)(end - start + 1) << PAGE_SHIFT, 0); | |
758 | } | |
34ccb69e | 759 | out: |
3167760f | 760 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
761 | return ret; |
762 | } | |
763 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
764 | ||
765 | /** | |
766 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 767 | * @mapping: the address_space |
1da177e4 LT |
768 | * |
769 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
770 | * invalidation. | |
771 | * | |
e9de25dd | 772 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
773 | */ |
774 | int invalidate_inode_pages2(struct address_space *mapping) | |
775 | { | |
776 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
777 | } | |
778 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 779 | |
780 | /** | |
781 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
782 | * @inode: inode | |
8a549bea | 783 | * @newsize: new file size |
25d9e2d1 | 784 | * |
785 | * inode's new i_size must already be written before truncate_pagecache | |
786 | * is called. | |
787 | * | |
788 | * This function should typically be called before the filesystem | |
789 | * releases resources associated with the freed range (eg. deallocates | |
790 | * blocks). This way, pagecache will always stay logically coherent | |
791 | * with on-disk format, and the filesystem would not have to deal with | |
792 | * situations such as writepage being called for a page that has already | |
793 | * had its underlying blocks deallocated. | |
794 | */ | |
7caef267 | 795 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 796 | { |
cedabed4 | 797 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 798 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
799 | |
800 | /* | |
801 | * unmap_mapping_range is called twice, first simply for | |
802 | * efficiency so that truncate_inode_pages does fewer | |
803 | * single-page unmaps. However after this first call, and | |
804 | * before truncate_inode_pages finishes, it is possible for | |
805 | * private pages to be COWed, which remain after | |
806 | * truncate_inode_pages finishes, hence the second | |
807 | * unmap_mapping_range call must be made for correctness. | |
808 | */ | |
8a549bea HD |
809 | unmap_mapping_range(mapping, holebegin, 0, 1); |
810 | truncate_inode_pages(mapping, newsize); | |
811 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 812 | } |
813 | EXPORT_SYMBOL(truncate_pagecache); | |
814 | ||
2c27c65e CH |
815 | /** |
816 | * truncate_setsize - update inode and pagecache for a new file size | |
817 | * @inode: inode | |
818 | * @newsize: new file size | |
819 | * | |
382e27da JK |
820 | * truncate_setsize updates i_size and performs pagecache truncation (if |
821 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
822 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 823 | * |
77783d06 JK |
824 | * Must be called with a lock serializing truncates and writes (generally |
825 | * i_mutex but e.g. xfs uses a different lock) and before all filesystem | |
826 | * specific block truncation has been performed. | |
2c27c65e CH |
827 | */ |
828 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
829 | { | |
90a80202 JK |
830 | loff_t oldsize = inode->i_size; |
831 | ||
2c27c65e | 832 | i_size_write(inode, newsize); |
90a80202 JK |
833 | if (newsize > oldsize) |
834 | pagecache_isize_extended(inode, oldsize, newsize); | |
7caef267 | 835 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
836 | } |
837 | EXPORT_SYMBOL(truncate_setsize); | |
838 | ||
90a80202 JK |
839 | /** |
840 | * pagecache_isize_extended - update pagecache after extension of i_size | |
841 | * @inode: inode for which i_size was extended | |
842 | * @from: original inode size | |
843 | * @to: new inode size | |
844 | * | |
845 | * Handle extension of inode size either caused by extending truncate or by | |
846 | * write starting after current i_size. We mark the page straddling current | |
847 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
848 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
849 | * the page before user writes to the page via mmap after the i_size has been | |
850 | * changed. | |
851 | * | |
852 | * The function must be called after i_size is updated so that page fault | |
853 | * coming after we unlock the page will already see the new i_size. | |
854 | * The function must be called while we still hold i_mutex - this not only | |
855 | * makes sure i_size is stable but also that userspace cannot observe new | |
856 | * i_size value before we are prepared to store mmap writes at new inode size. | |
857 | */ | |
858 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
859 | { | |
93407472 | 860 | int bsize = i_blocksize(inode); |
90a80202 JK |
861 | loff_t rounded_from; |
862 | struct page *page; | |
863 | pgoff_t index; | |
864 | ||
90a80202 JK |
865 | WARN_ON(to > inode->i_size); |
866 | ||
09cbfeaf | 867 | if (from >= to || bsize == PAGE_SIZE) |
90a80202 JK |
868 | return; |
869 | /* Page straddling @from will not have any hole block created? */ | |
870 | rounded_from = round_up(from, bsize); | |
09cbfeaf | 871 | if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) |
90a80202 JK |
872 | return; |
873 | ||
09cbfeaf | 874 | index = from >> PAGE_SHIFT; |
90a80202 JK |
875 | page = find_lock_page(inode->i_mapping, index); |
876 | /* Page not cached? Nothing to do */ | |
877 | if (!page) | |
878 | return; | |
879 | /* | |
880 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
881 | * is needed. | |
882 | */ | |
883 | if (page_mkclean(page)) | |
884 | set_page_dirty(page); | |
885 | unlock_page(page); | |
09cbfeaf | 886 | put_page(page); |
90a80202 JK |
887 | } |
888 | EXPORT_SYMBOL(pagecache_isize_extended); | |
889 | ||
623e3db9 HD |
890 | /** |
891 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
892 | * @inode: inode | |
893 | * @lstart: offset of beginning of hole | |
894 | * @lend: offset of last byte of hole | |
895 | * | |
896 | * This function should typically be called before the filesystem | |
897 | * releases resources associated with the freed range (eg. deallocates | |
898 | * blocks). This way, pagecache will always stay logically coherent | |
899 | * with on-disk format, and the filesystem would not have to deal with | |
900 | * situations such as writepage being called for a page that has already | |
901 | * had its underlying blocks deallocated. | |
902 | */ | |
903 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
904 | { | |
905 | struct address_space *mapping = inode->i_mapping; | |
906 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
907 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
908 | /* | |
909 | * This rounding is currently just for example: unmap_mapping_range | |
910 | * expands its hole outwards, whereas we want it to contract the hole | |
911 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
912 | * doing their own page rounding first. Note that unmap_mapping_range |
913 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
914 | */ |
915 | ||
916 | /* | |
917 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
918 | * once (before truncating pagecache), and without "even_cows" flag: | |
919 | * hole-punching should not remove private COWed pages from the hole. | |
920 | */ | |
921 | if ((u64)unmap_end > (u64)unmap_start) | |
922 | unmap_mapping_range(mapping, unmap_start, | |
923 | 1 + unmap_end - unmap_start, 0); | |
924 | truncate_inode_pages_range(mapping, lstart, lend); | |
925 | } | |
926 | EXPORT_SYMBOL(truncate_pagecache_range); |