]>
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> |
5a0e3ad6 | 12 | #include <linux/gfp.h> |
1da177e4 | 13 | #include <linux/mm.h> |
0fd0e6b0 | 14 | #include <linux/swap.h> |
b95f1b31 | 15 | #include <linux/export.h> |
1da177e4 | 16 | #include <linux/pagemap.h> |
01f2705d | 17 | #include <linux/highmem.h> |
1da177e4 | 18 | #include <linux/pagevec.h> |
e08748ce | 19 | #include <linux/task_io_accounting_ops.h> |
1da177e4 | 20 | #include <linux/buffer_head.h> /* grr. try_to_release_page, |
aaa4059b | 21 | do_invalidatepage */ |
c515e1fd | 22 | #include <linux/cleancache.h> |
ba470de4 | 23 | #include "internal.h" |
1da177e4 | 24 | |
0cd6144a JW |
25 | static void clear_exceptional_entry(struct address_space *mapping, |
26 | pgoff_t index, void *entry) | |
27 | { | |
28 | /* Handled by shmem itself */ | |
29 | if (shmem_mapping(mapping)) | |
30 | return; | |
31 | ||
32 | spin_lock_irq(&mapping->tree_lock); | |
33 | /* | |
34 | * Regular page slots are stabilized by the page lock even | |
35 | * without the tree itself locked. These unlocked entries | |
36 | * need verification under the tree lock. | |
37 | */ | |
91b0abe3 JW |
38 | if (radix_tree_delete_item(&mapping->page_tree, index, entry) == entry) |
39 | mapping->nrshadows--; | |
0cd6144a JW |
40 | spin_unlock_irq(&mapping->tree_lock); |
41 | } | |
1da177e4 | 42 | |
cf9a2ae8 | 43 | /** |
28bc44d7 | 44 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 45 | * @page: the page which is affected |
d47992f8 LC |
46 | * @offset: start of the range to invalidate |
47 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
48 | * |
49 | * do_invalidatepage() is called when all or part of the page has become | |
50 | * invalidated by a truncate operation. | |
51 | * | |
52 | * do_invalidatepage() does not have to release all buffers, but it must | |
53 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
54 | * is underway against any of the blocks which are outside the truncation | |
55 | * point. Because the caller is about to free (and possibly reuse) those | |
56 | * blocks on-disk. | |
57 | */ | |
d47992f8 LC |
58 | void do_invalidatepage(struct page *page, unsigned int offset, |
59 | unsigned int length) | |
cf9a2ae8 | 60 | { |
d47992f8 LC |
61 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
62 | ||
cf9a2ae8 | 63 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 64 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
65 | if (!invalidatepage) |
66 | invalidatepage = block_invalidatepage; | |
9361401e | 67 | #endif |
cf9a2ae8 | 68 | if (invalidatepage) |
d47992f8 | 69 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
70 | } |
71 | ||
ecdfc978 LT |
72 | /* |
73 | * This cancels just the dirty bit on the kernel page itself, it | |
74 | * does NOT actually remove dirty bits on any mmap's that may be | |
75 | * around. It also leaves the page tagged dirty, so any sync | |
76 | * activity will still find it on the dirty lists, and in particular, | |
77 | * clear_page_dirty_for_io() will still look at the dirty bits in | |
78 | * the VM. | |
79 | * | |
80 | * Doing this should *normally* only ever be done when a page | |
81 | * is truncated, and is not actually mapped anywhere at all. However, | |
82 | * fs/buffer.c does this when it notices that somebody has cleaned | |
83 | * out all the buffers on a page without actually doing it through | |
84 | * the VM. Can you say "ext3 is horribly ugly"? Tought you could. | |
85 | */ | |
fba2591b LT |
86 | void cancel_dirty_page(struct page *page, unsigned int account_size) |
87 | { | |
8368e328 LT |
88 | if (TestClearPageDirty(page)) { |
89 | struct address_space *mapping = page->mapping; | |
90 | if (mapping && mapping_cap_account_dirty(mapping)) { | |
91 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
c9e51e41 PZ |
92 | dec_bdi_stat(mapping->backing_dev_info, |
93 | BDI_RECLAIMABLE); | |
8368e328 LT |
94 | if (account_size) |
95 | task_io_account_cancelled_write(account_size); | |
96 | } | |
3e67c098 | 97 | } |
fba2591b | 98 | } |
8368e328 | 99 | EXPORT_SYMBOL(cancel_dirty_page); |
fba2591b | 100 | |
1da177e4 LT |
101 | /* |
102 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 103 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 104 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
105 | * |
106 | * We need to bale out if page->mapping is no longer equal to the original | |
107 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
fc0ecff6 | 108 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
109 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
110 | */ | |
750b4987 | 111 | static int |
1da177e4 LT |
112 | truncate_complete_page(struct address_space *mapping, struct page *page) |
113 | { | |
114 | if (page->mapping != mapping) | |
750b4987 | 115 | return -EIO; |
1da177e4 | 116 | |
266cf658 | 117 | if (page_has_private(page)) |
d47992f8 | 118 | do_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
1da177e4 | 119 | |
a2b34564 BS |
120 | cancel_dirty_page(page, PAGE_CACHE_SIZE); |
121 | ||
1da177e4 | 122 | ClearPageMappedToDisk(page); |
5adc7b51 | 123 | delete_from_page_cache(page); |
750b4987 | 124 | return 0; |
1da177e4 LT |
125 | } |
126 | ||
127 | /* | |
fc0ecff6 | 128 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 129 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
130 | * be marked dirty at any time too, so use remove_mapping which safely |
131 | * discards clean, unused pages. | |
1da177e4 LT |
132 | * |
133 | * Returns non-zero if the page was successfully invalidated. | |
134 | */ | |
135 | static int | |
136 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
137 | { | |
0fd0e6b0 NP |
138 | int ret; |
139 | ||
1da177e4 LT |
140 | if (page->mapping != mapping) |
141 | return 0; | |
142 | ||
266cf658 | 143 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
144 | return 0; |
145 | ||
0fd0e6b0 | 146 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
147 | |
148 | return ret; | |
1da177e4 LT |
149 | } |
150 | ||
750b4987 NP |
151 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
152 | { | |
153 | if (page_mapped(page)) { | |
154 | unmap_mapping_range(mapping, | |
155 | (loff_t)page->index << PAGE_CACHE_SHIFT, | |
156 | PAGE_CACHE_SIZE, 0); | |
157 | } | |
158 | return truncate_complete_page(mapping, page); | |
159 | } | |
160 | ||
25718736 AK |
161 | /* |
162 | * Used to get rid of pages on hardware memory corruption. | |
163 | */ | |
164 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
165 | { | |
166 | if (!mapping) | |
167 | return -EINVAL; | |
168 | /* | |
169 | * Only punch for normal data pages for now. | |
170 | * Handling other types like directories would need more auditing. | |
171 | */ | |
172 | if (!S_ISREG(mapping->host->i_mode)) | |
173 | return -EIO; | |
174 | return truncate_inode_page(mapping, page); | |
175 | } | |
176 | EXPORT_SYMBOL(generic_error_remove_page); | |
177 | ||
83f78668 WF |
178 | /* |
179 | * Safely invalidate one page from its pagecache mapping. | |
180 | * It only drops clean, unused pages. The page must be locked. | |
181 | * | |
182 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
183 | */ | |
184 | int invalidate_inode_page(struct page *page) | |
185 | { | |
186 | struct address_space *mapping = page_mapping(page); | |
187 | if (!mapping) | |
188 | return 0; | |
189 | if (PageDirty(page) || PageWriteback(page)) | |
190 | return 0; | |
191 | if (page_mapped(page)) | |
192 | return 0; | |
193 | return invalidate_complete_page(mapping, page); | |
194 | } | |
195 | ||
1da177e4 | 196 | /** |
73c1e204 | 197 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
198 | * @mapping: mapping to truncate |
199 | * @lstart: offset from which to truncate | |
5a720394 | 200 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 201 | * |
d7339071 | 202 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
203 | * specified offsets (and zeroing out partial pages |
204 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
205 | * |
206 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
207 | * block on page locks and it will not block on writeback. The second pass | |
208 | * will wait. This is to prevent as much IO as possible in the affected region. | |
209 | * The first pass will remove most pages, so the search cost of the second pass | |
210 | * is low. | |
211 | * | |
1da177e4 LT |
212 | * We pass down the cache-hot hint to the page freeing code. Even if the |
213 | * mapping is large, it is probably the case that the final pages are the most | |
214 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
215 | * |
216 | * Note that since ->invalidatepage() accepts range to invalidate | |
217 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
218 | * page aligned properly. | |
1da177e4 | 219 | */ |
d7339071 HR |
220 | void truncate_inode_pages_range(struct address_space *mapping, |
221 | loff_t lstart, loff_t lend) | |
1da177e4 | 222 | { |
5a720394 LC |
223 | pgoff_t start; /* inclusive */ |
224 | pgoff_t end; /* exclusive */ | |
225 | unsigned int partial_start; /* inclusive */ | |
226 | unsigned int partial_end; /* exclusive */ | |
227 | struct pagevec pvec; | |
0cd6144a | 228 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
229 | pgoff_t index; |
230 | int i; | |
1da177e4 | 231 | |
3167760f | 232 | cleancache_invalidate_inode(mapping); |
91b0abe3 | 233 | if (mapping->nrpages == 0 && mapping->nrshadows == 0) |
1da177e4 LT |
234 | return; |
235 | ||
5a720394 LC |
236 | /* Offsets within partial pages */ |
237 | partial_start = lstart & (PAGE_CACHE_SIZE - 1); | |
238 | partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); | |
239 | ||
240 | /* | |
241 | * 'start' and 'end' always covers the range of pages to be fully | |
242 | * truncated. Partial pages are covered with 'partial_start' at the | |
243 | * start of the range and 'partial_end' at the end of the range. | |
244 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
245 | */ | |
246 | start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
247 | if (lend == -1) | |
248 | /* | |
249 | * lend == -1 indicates end-of-file so we have to set 'end' | |
250 | * to the highest possible pgoff_t and since the type is | |
251 | * unsigned we're using -1. | |
252 | */ | |
253 | end = -1; | |
254 | else | |
255 | end = (lend + 1) >> PAGE_CACHE_SHIFT; | |
d7339071 | 256 | |
1da177e4 | 257 | pagevec_init(&pvec, 0); |
b85e0eff | 258 | index = start; |
0cd6144a JW |
259 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
260 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
261 | indices)) { | |
e5598f8b | 262 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
263 | for (i = 0; i < pagevec_count(&pvec); i++) { |
264 | struct page *page = pvec.pages[i]; | |
1da177e4 | 265 | |
b85e0eff | 266 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 267 | index = indices[i]; |
5a720394 | 268 | if (index >= end) |
d7339071 | 269 | break; |
d7339071 | 270 | |
0cd6144a JW |
271 | if (radix_tree_exceptional_entry(page)) { |
272 | clear_exceptional_entry(mapping, index, page); | |
273 | continue; | |
274 | } | |
275 | ||
529ae9aa | 276 | if (!trylock_page(page)) |
1da177e4 | 277 | continue; |
b85e0eff | 278 | WARN_ON(page->index != index); |
1da177e4 LT |
279 | if (PageWriteback(page)) { |
280 | unlock_page(page); | |
281 | continue; | |
282 | } | |
750b4987 | 283 | truncate_inode_page(mapping, page); |
1da177e4 LT |
284 | unlock_page(page); |
285 | } | |
0cd6144a | 286 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 287 | pagevec_release(&pvec); |
e5598f8b | 288 | mem_cgroup_uncharge_end(); |
1da177e4 | 289 | cond_resched(); |
b85e0eff | 290 | index++; |
1da177e4 LT |
291 | } |
292 | ||
5a720394 | 293 | if (partial_start) { |
1da177e4 LT |
294 | struct page *page = find_lock_page(mapping, start - 1); |
295 | if (page) { | |
5a720394 LC |
296 | unsigned int top = PAGE_CACHE_SIZE; |
297 | if (start > end) { | |
298 | /* Truncation within a single page */ | |
299 | top = partial_end; | |
300 | partial_end = 0; | |
301 | } | |
1da177e4 | 302 | wait_on_page_writeback(page); |
5a720394 LC |
303 | zero_user_segment(page, partial_start, top); |
304 | cleancache_invalidate_page(mapping, page); | |
305 | if (page_has_private(page)) | |
306 | do_invalidatepage(page, partial_start, | |
307 | top - partial_start); | |
1da177e4 LT |
308 | unlock_page(page); |
309 | page_cache_release(page); | |
310 | } | |
311 | } | |
5a720394 LC |
312 | if (partial_end) { |
313 | struct page *page = find_lock_page(mapping, end); | |
314 | if (page) { | |
315 | wait_on_page_writeback(page); | |
316 | zero_user_segment(page, 0, partial_end); | |
317 | cleancache_invalidate_page(mapping, page); | |
318 | if (page_has_private(page)) | |
319 | do_invalidatepage(page, 0, | |
320 | partial_end); | |
321 | unlock_page(page); | |
322 | page_cache_release(page); | |
323 | } | |
324 | } | |
325 | /* | |
326 | * If the truncation happened within a single page no pages | |
327 | * will be released, just zeroed, so we can bail out now. | |
328 | */ | |
329 | if (start >= end) | |
330 | return; | |
1da177e4 | 331 | |
b85e0eff | 332 | index = start; |
1da177e4 LT |
333 | for ( ; ; ) { |
334 | cond_resched(); | |
0cd6144a JW |
335 | if (!pagevec_lookup_entries(&pvec, mapping, index, |
336 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
337 | indices)) { | |
b85e0eff | 338 | if (index == start) |
1da177e4 | 339 | break; |
b85e0eff | 340 | index = start; |
1da177e4 LT |
341 | continue; |
342 | } | |
0cd6144a JW |
343 | if (index == start && indices[0] >= end) { |
344 | pagevec_remove_exceptionals(&pvec); | |
d7339071 HR |
345 | pagevec_release(&pvec); |
346 | break; | |
347 | } | |
569b846d | 348 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
349 | for (i = 0; i < pagevec_count(&pvec); i++) { |
350 | struct page *page = pvec.pages[i]; | |
351 | ||
b85e0eff | 352 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 353 | index = indices[i]; |
5a720394 | 354 | if (index >= end) |
d7339071 | 355 | break; |
b85e0eff | 356 | |
0cd6144a JW |
357 | if (radix_tree_exceptional_entry(page)) { |
358 | clear_exceptional_entry(mapping, index, page); | |
359 | continue; | |
360 | } | |
361 | ||
1da177e4 | 362 | lock_page(page); |
b85e0eff | 363 | WARN_ON(page->index != index); |
1da177e4 | 364 | wait_on_page_writeback(page); |
750b4987 | 365 | truncate_inode_page(mapping, page); |
1da177e4 LT |
366 | unlock_page(page); |
367 | } | |
0cd6144a | 368 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 369 | pagevec_release(&pvec); |
569b846d | 370 | mem_cgroup_uncharge_end(); |
b85e0eff | 371 | index++; |
1da177e4 | 372 | } |
3167760f | 373 | cleancache_invalidate_inode(mapping); |
1da177e4 | 374 | } |
d7339071 | 375 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 376 | |
d7339071 HR |
377 | /** |
378 | * truncate_inode_pages - truncate *all* the pages from an offset | |
379 | * @mapping: mapping to truncate | |
380 | * @lstart: offset from which to truncate | |
381 | * | |
1b1dcc1b | 382 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
383 | * |
384 | * Note: When this function returns, there can be a page in the process of | |
385 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
386 | * mapping->nrpages can be non-zero when this function returns even after | |
387 | * truncation of the whole mapping. | |
d7339071 HR |
388 | */ |
389 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
390 | { | |
391 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
392 | } | |
1da177e4 LT |
393 | EXPORT_SYMBOL(truncate_inode_pages); |
394 | ||
91b0abe3 JW |
395 | /** |
396 | * truncate_inode_pages_final - truncate *all* pages before inode dies | |
397 | * @mapping: mapping to truncate | |
398 | * | |
399 | * Called under (and serialized by) inode->i_mutex. | |
400 | * | |
401 | * Filesystems have to use this in the .evict_inode path to inform the | |
402 | * VM that this is the final truncate and the inode is going away. | |
403 | */ | |
404 | void truncate_inode_pages_final(struct address_space *mapping) | |
405 | { | |
406 | unsigned long nrshadows; | |
407 | unsigned long nrpages; | |
408 | ||
409 | /* | |
410 | * Page reclaim can not participate in regular inode lifetime | |
411 | * management (can't call iput()) and thus can race with the | |
412 | * inode teardown. Tell it when the address space is exiting, | |
413 | * so that it does not install eviction information after the | |
414 | * final truncate has begun. | |
415 | */ | |
416 | mapping_set_exiting(mapping); | |
417 | ||
418 | /* | |
419 | * When reclaim installs eviction entries, it increases | |
420 | * nrshadows first, then decreases nrpages. Make sure we see | |
421 | * this in the right order or we might miss an entry. | |
422 | */ | |
423 | nrpages = mapping->nrpages; | |
424 | smp_rmb(); | |
425 | nrshadows = mapping->nrshadows; | |
426 | ||
427 | if (nrpages || nrshadows) { | |
428 | /* | |
429 | * As truncation uses a lockless tree lookup, cycle | |
430 | * the tree lock to make sure any ongoing tree | |
431 | * modification that does not see AS_EXITING is | |
432 | * completed before starting the final truncate. | |
433 | */ | |
434 | spin_lock_irq(&mapping->tree_lock); | |
435 | spin_unlock_irq(&mapping->tree_lock); | |
436 | ||
437 | truncate_inode_pages(mapping, 0); | |
438 | } | |
439 | } | |
440 | EXPORT_SYMBOL(truncate_inode_pages_final); | |
441 | ||
28697355 MW |
442 | /** |
443 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
444 | * @mapping: the address_space which holds the pages to invalidate | |
445 | * @start: the offset 'from' which to invalidate | |
446 | * @end: the offset 'to' which to invalidate (inclusive) | |
447 | * | |
448 | * This function only removes the unlocked pages, if you want to | |
449 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
450 | * | |
451 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
452 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
453 | * pagetables. | |
454 | */ | |
455 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 456 | pgoff_t start, pgoff_t end) |
1da177e4 | 457 | { |
0cd6144a | 458 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 459 | struct pagevec pvec; |
b85e0eff | 460 | pgoff_t index = start; |
31560180 MK |
461 | unsigned long ret; |
462 | unsigned long count = 0; | |
1da177e4 LT |
463 | int i; |
464 | ||
31475dd6 HD |
465 | /* |
466 | * Note: this function may get called on a shmem/tmpfs mapping: | |
467 | * pagevec_lookup() might then return 0 prematurely (because it | |
468 | * got a gangful of swap entries); but it's hardly worth worrying | |
469 | * about - it can rarely have anything to free from such a mapping | |
470 | * (most pages are dirty), and already skips over any difficulties. | |
471 | */ | |
472 | ||
1da177e4 | 473 | pagevec_init(&pvec, 0); |
0cd6144a JW |
474 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
475 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
476 | indices)) { | |
569b846d | 477 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
478 | for (i = 0; i < pagevec_count(&pvec); i++) { |
479 | struct page *page = pvec.pages[i]; | |
e0f23603 | 480 | |
b85e0eff | 481 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 482 | index = indices[i]; |
b85e0eff HD |
483 | if (index > end) |
484 | break; | |
e0f23603 | 485 | |
0cd6144a JW |
486 | if (radix_tree_exceptional_entry(page)) { |
487 | clear_exceptional_entry(mapping, index, page); | |
488 | continue; | |
489 | } | |
490 | ||
b85e0eff HD |
491 | if (!trylock_page(page)) |
492 | continue; | |
493 | WARN_ON(page->index != index); | |
31560180 | 494 | ret = invalidate_inode_page(page); |
1da177e4 | 495 | unlock_page(page); |
31560180 MK |
496 | /* |
497 | * Invalidation is a hint that the page is no longer | |
498 | * of interest and try to speed up its reclaim. | |
499 | */ | |
500 | if (!ret) | |
501 | deactivate_page(page); | |
502 | count += ret; | |
1da177e4 | 503 | } |
0cd6144a | 504 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 505 | pagevec_release(&pvec); |
569b846d | 506 | mem_cgroup_uncharge_end(); |
28697355 | 507 | cond_resched(); |
b85e0eff | 508 | index++; |
1da177e4 | 509 | } |
31560180 | 510 | return count; |
1da177e4 | 511 | } |
54bc4855 | 512 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 513 | |
bd4c8ce4 AM |
514 | /* |
515 | * This is like invalidate_complete_page(), except it ignores the page's | |
516 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
517 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
518 | * shrink_page_list() has a temp ref on them, or because they're transiently |
519 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
520 | */ |
521 | static int | |
522 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
523 | { | |
524 | if (page->mapping != mapping) | |
525 | return 0; | |
526 | ||
266cf658 | 527 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
528 | return 0; |
529 | ||
19fd6231 | 530 | spin_lock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
531 | if (PageDirty(page)) |
532 | goto failed; | |
533 | ||
266cf658 | 534 | BUG_ON(page_has_private(page)); |
91b0abe3 | 535 | __delete_from_page_cache(page, NULL); |
19fd6231 | 536 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 537 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
538 | |
539 | if (mapping->a_ops->freepage) | |
540 | mapping->a_ops->freepage(page); | |
541 | ||
bd4c8ce4 AM |
542 | page_cache_release(page); /* pagecache ref */ |
543 | return 1; | |
544 | failed: | |
19fd6231 | 545 | spin_unlock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
546 | return 0; |
547 | } | |
548 | ||
e3db7691 TM |
549 | static int do_launder_page(struct address_space *mapping, struct page *page) |
550 | { | |
551 | if (!PageDirty(page)) | |
552 | return 0; | |
553 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
554 | return 0; | |
555 | return mapping->a_ops->launder_page(page); | |
556 | } | |
557 | ||
1da177e4 LT |
558 | /** |
559 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 560 | * @mapping: the address_space |
1da177e4 LT |
561 | * @start: the page offset 'from' which to invalidate |
562 | * @end: the page offset 'to' which to invalidate (inclusive) | |
563 | * | |
564 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
565 | * invalidation. | |
566 | * | |
6ccfa806 | 567 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
568 | */ |
569 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
570 | pgoff_t start, pgoff_t end) | |
571 | { | |
0cd6144a | 572 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 573 | struct pagevec pvec; |
b85e0eff | 574 | pgoff_t index; |
1da177e4 LT |
575 | int i; |
576 | int ret = 0; | |
0dd1334f | 577 | int ret2 = 0; |
1da177e4 | 578 | int did_range_unmap = 0; |
1da177e4 | 579 | |
3167760f | 580 | cleancache_invalidate_inode(mapping); |
1da177e4 | 581 | pagevec_init(&pvec, 0); |
b85e0eff | 582 | index = start; |
0cd6144a JW |
583 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
584 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
585 | indices)) { | |
569b846d | 586 | mem_cgroup_uncharge_start(); |
7b965e08 | 587 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 588 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
589 | |
590 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 591 | index = indices[i]; |
b85e0eff HD |
592 | if (index > end) |
593 | break; | |
1da177e4 | 594 | |
0cd6144a JW |
595 | if (radix_tree_exceptional_entry(page)) { |
596 | clear_exceptional_entry(mapping, index, page); | |
597 | continue; | |
598 | } | |
599 | ||
1da177e4 | 600 | lock_page(page); |
b85e0eff | 601 | WARN_ON(page->index != index); |
1da177e4 LT |
602 | if (page->mapping != mapping) { |
603 | unlock_page(page); | |
604 | continue; | |
605 | } | |
1da177e4 | 606 | wait_on_page_writeback(page); |
d00806b1 | 607 | if (page_mapped(page)) { |
1da177e4 LT |
608 | if (!did_range_unmap) { |
609 | /* | |
610 | * Zap the rest of the file in one hit. | |
611 | */ | |
612 | unmap_mapping_range(mapping, | |
b85e0eff HD |
613 | (loff_t)index << PAGE_CACHE_SHIFT, |
614 | (loff_t)(1 + end - index) | |
615 | << PAGE_CACHE_SHIFT, | |
1da177e4 LT |
616 | 0); |
617 | did_range_unmap = 1; | |
618 | } else { | |
619 | /* | |
620 | * Just zap this page | |
621 | */ | |
622 | unmap_mapping_range(mapping, | |
b85e0eff HD |
623 | (loff_t)index << PAGE_CACHE_SHIFT, |
624 | PAGE_CACHE_SIZE, 0); | |
1da177e4 LT |
625 | } |
626 | } | |
d00806b1 | 627 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
628 | ret2 = do_launder_page(mapping, page); |
629 | if (ret2 == 0) { | |
630 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 631 | ret2 = -EBUSY; |
0dd1334f HH |
632 | } |
633 | if (ret2 < 0) | |
634 | ret = ret2; | |
1da177e4 LT |
635 | unlock_page(page); |
636 | } | |
0cd6144a | 637 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 638 | pagevec_release(&pvec); |
569b846d | 639 | mem_cgroup_uncharge_end(); |
1da177e4 | 640 | cond_resched(); |
b85e0eff | 641 | index++; |
1da177e4 | 642 | } |
3167760f | 643 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
644 | return ret; |
645 | } | |
646 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
647 | ||
648 | /** | |
649 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 650 | * @mapping: the address_space |
1da177e4 LT |
651 | * |
652 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
653 | * invalidation. | |
654 | * | |
e9de25dd | 655 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
656 | */ |
657 | int invalidate_inode_pages2(struct address_space *mapping) | |
658 | { | |
659 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
660 | } | |
661 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 NP |
662 | |
663 | /** | |
664 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
665 | * @inode: inode | |
8a549bea | 666 | * @newsize: new file size |
25d9e2d1 NP |
667 | * |
668 | * inode's new i_size must already be written before truncate_pagecache | |
669 | * is called. | |
670 | * | |
671 | * This function should typically be called before the filesystem | |
672 | * releases resources associated with the freed range (eg. deallocates | |
673 | * blocks). This way, pagecache will always stay logically coherent | |
674 | * with on-disk format, and the filesystem would not have to deal with | |
675 | * situations such as writepage being called for a page that has already | |
676 | * had its underlying blocks deallocated. | |
677 | */ | |
7caef267 | 678 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 679 | { |
cedabed4 | 680 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 681 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
682 | |
683 | /* | |
684 | * unmap_mapping_range is called twice, first simply for | |
685 | * efficiency so that truncate_inode_pages does fewer | |
686 | * single-page unmaps. However after this first call, and | |
687 | * before truncate_inode_pages finishes, it is possible for | |
688 | * private pages to be COWed, which remain after | |
689 | * truncate_inode_pages finishes, hence the second | |
690 | * unmap_mapping_range call must be made for correctness. | |
691 | */ | |
8a549bea HD |
692 | unmap_mapping_range(mapping, holebegin, 0, 1); |
693 | truncate_inode_pages(mapping, newsize); | |
694 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 NP |
695 | } |
696 | EXPORT_SYMBOL(truncate_pagecache); | |
697 | ||
2c27c65e CH |
698 | /** |
699 | * truncate_setsize - update inode and pagecache for a new file size | |
700 | * @inode: inode | |
701 | * @newsize: new file size | |
702 | * | |
382e27da JK |
703 | * truncate_setsize updates i_size and performs pagecache truncation (if |
704 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
705 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 706 | * |
382e27da JK |
707 | * Must be called with inode_mutex held and before all filesystem specific |
708 | * block truncation has been performed. | |
2c27c65e CH |
709 | */ |
710 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
711 | { | |
2c27c65e | 712 | i_size_write(inode, newsize); |
7caef267 | 713 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
714 | } |
715 | EXPORT_SYMBOL(truncate_setsize); | |
716 | ||
623e3db9 HD |
717 | /** |
718 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
719 | * @inode: inode | |
720 | * @lstart: offset of beginning of hole | |
721 | * @lend: offset of last byte of hole | |
722 | * | |
723 | * This function should typically be called before the filesystem | |
724 | * releases resources associated with the freed range (eg. deallocates | |
725 | * blocks). This way, pagecache will always stay logically coherent | |
726 | * with on-disk format, and the filesystem would not have to deal with | |
727 | * situations such as writepage being called for a page that has already | |
728 | * had its underlying blocks deallocated. | |
729 | */ | |
730 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
731 | { | |
732 | struct address_space *mapping = inode->i_mapping; | |
733 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
734 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
735 | /* | |
736 | * This rounding is currently just for example: unmap_mapping_range | |
737 | * expands its hole outwards, whereas we want it to contract the hole | |
738 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
739 | * doing their own page rounding first. Note that unmap_mapping_range |
740 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
741 | */ |
742 | ||
743 | /* | |
744 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
745 | * once (before truncating pagecache), and without "even_cows" flag: | |
746 | * hole-punching should not remove private COWed pages from the hole. | |
747 | */ | |
748 | if ((u64)unmap_end > (u64)unmap_start) | |
749 | unmap_mapping_range(mapping, unmap_start, | |
750 | 1 + unmap_end - unmap_start, 0); | |
751 | truncate_inode_pages_range(mapping, lstart, lend); | |
752 | } | |
753 | EXPORT_SYMBOL(truncate_pagecache_range); |