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 LT |
24 | |
25 | ||
cf9a2ae8 | 26 | /** |
28bc44d7 | 27 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 28 | * @page: the page which is affected |
d47992f8 LC |
29 | * @offset: start of the range to invalidate |
30 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
31 | * |
32 | * do_invalidatepage() is called when all or part of the page has become | |
33 | * invalidated by a truncate operation. | |
34 | * | |
35 | * do_invalidatepage() does not have to release all buffers, but it must | |
36 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
37 | * is underway against any of the blocks which are outside the truncation | |
38 | * point. Because the caller is about to free (and possibly reuse) those | |
39 | * blocks on-disk. | |
40 | */ | |
d47992f8 LC |
41 | void do_invalidatepage(struct page *page, unsigned int offset, |
42 | unsigned int length) | |
cf9a2ae8 | 43 | { |
d47992f8 LC |
44 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
45 | ||
cf9a2ae8 | 46 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 47 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
48 | if (!invalidatepage) |
49 | invalidatepage = block_invalidatepage; | |
9361401e | 50 | #endif |
cf9a2ae8 | 51 | if (invalidatepage) |
d47992f8 | 52 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
53 | } |
54 | ||
ecdfc978 LT |
55 | /* |
56 | * This cancels just the dirty bit on the kernel page itself, it | |
57 | * does NOT actually remove dirty bits on any mmap's that may be | |
58 | * around. It also leaves the page tagged dirty, so any sync | |
59 | * activity will still find it on the dirty lists, and in particular, | |
60 | * clear_page_dirty_for_io() will still look at the dirty bits in | |
61 | * the VM. | |
62 | * | |
63 | * Doing this should *normally* only ever be done when a page | |
64 | * is truncated, and is not actually mapped anywhere at all. However, | |
65 | * fs/buffer.c does this when it notices that somebody has cleaned | |
66 | * out all the buffers on a page without actually doing it through | |
67 | * the VM. Can you say "ext3 is horribly ugly"? Tought you could. | |
68 | */ | |
fba2591b LT |
69 | void cancel_dirty_page(struct page *page, unsigned int account_size) |
70 | { | |
8368e328 LT |
71 | if (TestClearPageDirty(page)) { |
72 | struct address_space *mapping = page->mapping; | |
73 | if (mapping && mapping_cap_account_dirty(mapping)) { | |
74 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
c9e51e41 PZ |
75 | dec_bdi_stat(mapping->backing_dev_info, |
76 | BDI_RECLAIMABLE); | |
8368e328 LT |
77 | if (account_size) |
78 | task_io_account_cancelled_write(account_size); | |
79 | } | |
3e67c098 | 80 | } |
fba2591b | 81 | } |
8368e328 | 82 | EXPORT_SYMBOL(cancel_dirty_page); |
fba2591b | 83 | |
1da177e4 LT |
84 | /* |
85 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 86 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 87 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
88 | * |
89 | * We need to bale out if page->mapping is no longer equal to the original | |
90 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
fc0ecff6 | 91 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
92 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
93 | */ | |
750b4987 | 94 | static int |
1da177e4 LT |
95 | truncate_complete_page(struct address_space *mapping, struct page *page) |
96 | { | |
97 | if (page->mapping != mapping) | |
750b4987 | 98 | return -EIO; |
1da177e4 | 99 | |
266cf658 | 100 | if (page_has_private(page)) |
d47992f8 | 101 | do_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
1da177e4 | 102 | |
a2b34564 BS |
103 | cancel_dirty_page(page, PAGE_CACHE_SIZE); |
104 | ||
1da177e4 | 105 | ClearPageMappedToDisk(page); |
5adc7b51 | 106 | delete_from_page_cache(page); |
750b4987 | 107 | return 0; |
1da177e4 LT |
108 | } |
109 | ||
110 | /* | |
fc0ecff6 | 111 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 112 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
113 | * be marked dirty at any time too, so use remove_mapping which safely |
114 | * discards clean, unused pages. | |
1da177e4 LT |
115 | * |
116 | * Returns non-zero if the page was successfully invalidated. | |
117 | */ | |
118 | static int | |
119 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
120 | { | |
0fd0e6b0 NP |
121 | int ret; |
122 | ||
1da177e4 LT |
123 | if (page->mapping != mapping) |
124 | return 0; | |
125 | ||
266cf658 | 126 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
127 | return 0; |
128 | ||
0fd0e6b0 | 129 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
130 | |
131 | return ret; | |
1da177e4 LT |
132 | } |
133 | ||
750b4987 NP |
134 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
135 | { | |
136 | if (page_mapped(page)) { | |
137 | unmap_mapping_range(mapping, | |
138 | (loff_t)page->index << PAGE_CACHE_SHIFT, | |
139 | PAGE_CACHE_SIZE, 0); | |
140 | } | |
141 | return truncate_complete_page(mapping, page); | |
142 | } | |
143 | ||
25718736 AK |
144 | /* |
145 | * Used to get rid of pages on hardware memory corruption. | |
146 | */ | |
147 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
148 | { | |
149 | if (!mapping) | |
150 | return -EINVAL; | |
151 | /* | |
152 | * Only punch for normal data pages for now. | |
153 | * Handling other types like directories would need more auditing. | |
154 | */ | |
155 | if (!S_ISREG(mapping->host->i_mode)) | |
156 | return -EIO; | |
157 | return truncate_inode_page(mapping, page); | |
158 | } | |
159 | EXPORT_SYMBOL(generic_error_remove_page); | |
160 | ||
83f78668 WF |
161 | /* |
162 | * Safely invalidate one page from its pagecache mapping. | |
163 | * It only drops clean, unused pages. The page must be locked. | |
164 | * | |
165 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
166 | */ | |
167 | int invalidate_inode_page(struct page *page) | |
168 | { | |
169 | struct address_space *mapping = page_mapping(page); | |
170 | if (!mapping) | |
171 | return 0; | |
172 | if (PageDirty(page) || PageWriteback(page)) | |
173 | return 0; | |
174 | if (page_mapped(page)) | |
175 | return 0; | |
176 | return invalidate_complete_page(mapping, page); | |
177 | } | |
178 | ||
1da177e4 | 179 | /** |
73c1e204 | 180 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
181 | * @mapping: mapping to truncate |
182 | * @lstart: offset from which to truncate | |
5a720394 | 183 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 184 | * |
d7339071 | 185 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
186 | * specified offsets (and zeroing out partial pages |
187 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
188 | * |
189 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
190 | * block on page locks and it will not block on writeback. The second pass | |
191 | * will wait. This is to prevent as much IO as possible in the affected region. | |
192 | * The first pass will remove most pages, so the search cost of the second pass | |
193 | * is low. | |
194 | * | |
1da177e4 LT |
195 | * We pass down the cache-hot hint to the page freeing code. Even if the |
196 | * mapping is large, it is probably the case that the final pages are the most | |
197 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
198 | * |
199 | * Note that since ->invalidatepage() accepts range to invalidate | |
200 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
201 | * page aligned properly. | |
1da177e4 | 202 | */ |
d7339071 HR |
203 | void truncate_inode_pages_range(struct address_space *mapping, |
204 | loff_t lstart, loff_t lend) | |
1da177e4 | 205 | { |
5a720394 LC |
206 | pgoff_t start; /* inclusive */ |
207 | pgoff_t end; /* exclusive */ | |
208 | unsigned int partial_start; /* inclusive */ | |
209 | unsigned int partial_end; /* exclusive */ | |
210 | struct pagevec pvec; | |
211 | pgoff_t index; | |
212 | int i; | |
1da177e4 | 213 | |
3167760f | 214 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
215 | if (mapping->nrpages == 0) |
216 | return; | |
217 | ||
5a720394 LC |
218 | /* Offsets within partial pages */ |
219 | partial_start = lstart & (PAGE_CACHE_SIZE - 1); | |
220 | partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); | |
221 | ||
222 | /* | |
223 | * 'start' and 'end' always covers the range of pages to be fully | |
224 | * truncated. Partial pages are covered with 'partial_start' at the | |
225 | * start of the range and 'partial_end' at the end of the range. | |
226 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
227 | */ | |
228 | start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
229 | if (lend == -1) | |
230 | /* | |
231 | * lend == -1 indicates end-of-file so we have to set 'end' | |
232 | * to the highest possible pgoff_t and since the type is | |
233 | * unsigned we're using -1. | |
234 | */ | |
235 | end = -1; | |
236 | else | |
237 | end = (lend + 1) >> PAGE_CACHE_SHIFT; | |
d7339071 | 238 | |
1da177e4 | 239 | pagevec_init(&pvec, 0); |
b85e0eff | 240 | index = start; |
5a720394 LC |
241 | while (index < end && pagevec_lookup(&pvec, mapping, index, |
242 | min(end - index, (pgoff_t)PAGEVEC_SIZE))) { | |
e5598f8b | 243 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
244 | for (i = 0; i < pagevec_count(&pvec); i++) { |
245 | struct page *page = pvec.pages[i]; | |
1da177e4 | 246 | |
b85e0eff HD |
247 | /* We rely upon deletion not changing page->index */ |
248 | index = page->index; | |
5a720394 | 249 | if (index >= end) |
d7339071 | 250 | break; |
d7339071 | 251 | |
529ae9aa | 252 | if (!trylock_page(page)) |
1da177e4 | 253 | continue; |
b85e0eff | 254 | WARN_ON(page->index != index); |
1da177e4 LT |
255 | if (PageWriteback(page)) { |
256 | unlock_page(page); | |
257 | continue; | |
258 | } | |
750b4987 | 259 | truncate_inode_page(mapping, page); |
1da177e4 LT |
260 | unlock_page(page); |
261 | } | |
262 | pagevec_release(&pvec); | |
e5598f8b | 263 | mem_cgroup_uncharge_end(); |
1da177e4 | 264 | cond_resched(); |
b85e0eff | 265 | index++; |
1da177e4 LT |
266 | } |
267 | ||
5a720394 | 268 | if (partial_start) { |
1da177e4 LT |
269 | struct page *page = find_lock_page(mapping, start - 1); |
270 | if (page) { | |
5a720394 LC |
271 | unsigned int top = PAGE_CACHE_SIZE; |
272 | if (start > end) { | |
273 | /* Truncation within a single page */ | |
274 | top = partial_end; | |
275 | partial_end = 0; | |
276 | } | |
1da177e4 | 277 | wait_on_page_writeback(page); |
5a720394 LC |
278 | zero_user_segment(page, partial_start, top); |
279 | cleancache_invalidate_page(mapping, page); | |
280 | if (page_has_private(page)) | |
281 | do_invalidatepage(page, partial_start, | |
282 | top - partial_start); | |
1da177e4 LT |
283 | unlock_page(page); |
284 | page_cache_release(page); | |
285 | } | |
286 | } | |
5a720394 LC |
287 | if (partial_end) { |
288 | struct page *page = find_lock_page(mapping, end); | |
289 | if (page) { | |
290 | wait_on_page_writeback(page); | |
291 | zero_user_segment(page, 0, partial_end); | |
292 | cleancache_invalidate_page(mapping, page); | |
293 | if (page_has_private(page)) | |
294 | do_invalidatepage(page, 0, | |
295 | partial_end); | |
296 | unlock_page(page); | |
297 | page_cache_release(page); | |
298 | } | |
299 | } | |
300 | /* | |
301 | * If the truncation happened within a single page no pages | |
302 | * will be released, just zeroed, so we can bail out now. | |
303 | */ | |
304 | if (start >= end) | |
305 | return; | |
1da177e4 | 306 | |
b85e0eff | 307 | index = start; |
1da177e4 LT |
308 | for ( ; ; ) { |
309 | cond_resched(); | |
b85e0eff | 310 | if (!pagevec_lookup(&pvec, mapping, index, |
5a720394 | 311 | min(end - index, (pgoff_t)PAGEVEC_SIZE))) { |
b85e0eff | 312 | if (index == start) |
1da177e4 | 313 | break; |
b85e0eff | 314 | index = start; |
1da177e4 LT |
315 | continue; |
316 | } | |
5a720394 | 317 | if (index == start && pvec.pages[0]->index >= end) { |
d7339071 HR |
318 | pagevec_release(&pvec); |
319 | break; | |
320 | } | |
569b846d | 321 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
322 | for (i = 0; i < pagevec_count(&pvec); i++) { |
323 | struct page *page = pvec.pages[i]; | |
324 | ||
b85e0eff HD |
325 | /* We rely upon deletion not changing page->index */ |
326 | index = page->index; | |
5a720394 | 327 | if (index >= end) |
d7339071 | 328 | break; |
b85e0eff | 329 | |
1da177e4 | 330 | lock_page(page); |
b85e0eff | 331 | WARN_ON(page->index != index); |
1da177e4 | 332 | wait_on_page_writeback(page); |
750b4987 | 333 | truncate_inode_page(mapping, page); |
1da177e4 LT |
334 | unlock_page(page); |
335 | } | |
336 | pagevec_release(&pvec); | |
569b846d | 337 | mem_cgroup_uncharge_end(); |
b85e0eff | 338 | index++; |
1da177e4 | 339 | } |
3167760f | 340 | cleancache_invalidate_inode(mapping); |
1da177e4 | 341 | } |
d7339071 | 342 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 343 | |
d7339071 HR |
344 | /** |
345 | * truncate_inode_pages - truncate *all* the pages from an offset | |
346 | * @mapping: mapping to truncate | |
347 | * @lstart: offset from which to truncate | |
348 | * | |
1b1dcc1b | 349 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
350 | * |
351 | * Note: When this function returns, there can be a page in the process of | |
352 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
353 | * mapping->nrpages can be non-zero when this function returns even after | |
354 | * truncation of the whole mapping. | |
d7339071 HR |
355 | */ |
356 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
357 | { | |
358 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
359 | } | |
1da177e4 LT |
360 | EXPORT_SYMBOL(truncate_inode_pages); |
361 | ||
28697355 MW |
362 | /** |
363 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
364 | * @mapping: the address_space which holds the pages to invalidate | |
365 | * @start: the offset 'from' which to invalidate | |
366 | * @end: the offset 'to' which to invalidate (inclusive) | |
367 | * | |
368 | * This function only removes the unlocked pages, if you want to | |
369 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
370 | * | |
371 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
372 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
373 | * pagetables. | |
374 | */ | |
375 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 376 | pgoff_t start, pgoff_t end) |
1da177e4 LT |
377 | { |
378 | struct pagevec pvec; | |
b85e0eff | 379 | pgoff_t index = start; |
31560180 MK |
380 | unsigned long ret; |
381 | unsigned long count = 0; | |
1da177e4 LT |
382 | int i; |
383 | ||
31475dd6 HD |
384 | /* |
385 | * Note: this function may get called on a shmem/tmpfs mapping: | |
386 | * pagevec_lookup() might then return 0 prematurely (because it | |
387 | * got a gangful of swap entries); but it's hardly worth worrying | |
388 | * about - it can rarely have anything to free from such a mapping | |
389 | * (most pages are dirty), and already skips over any difficulties. | |
390 | */ | |
391 | ||
1da177e4 | 392 | pagevec_init(&pvec, 0); |
b85e0eff HD |
393 | while (index <= end && pagevec_lookup(&pvec, mapping, index, |
394 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
569b846d | 395 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
396 | for (i = 0; i < pagevec_count(&pvec); i++) { |
397 | struct page *page = pvec.pages[i]; | |
e0f23603 | 398 | |
b85e0eff | 399 | /* We rely upon deletion not changing page->index */ |
e0f23603 | 400 | index = page->index; |
b85e0eff HD |
401 | if (index > end) |
402 | break; | |
e0f23603 | 403 | |
b85e0eff HD |
404 | if (!trylock_page(page)) |
405 | continue; | |
406 | WARN_ON(page->index != index); | |
31560180 | 407 | ret = invalidate_inode_page(page); |
1da177e4 | 408 | unlock_page(page); |
31560180 MK |
409 | /* |
410 | * Invalidation is a hint that the page is no longer | |
411 | * of interest and try to speed up its reclaim. | |
412 | */ | |
413 | if (!ret) | |
414 | deactivate_page(page); | |
415 | count += ret; | |
1da177e4 LT |
416 | } |
417 | pagevec_release(&pvec); | |
569b846d | 418 | mem_cgroup_uncharge_end(); |
28697355 | 419 | cond_resched(); |
b85e0eff | 420 | index++; |
1da177e4 | 421 | } |
31560180 | 422 | return count; |
1da177e4 | 423 | } |
54bc4855 | 424 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 425 | |
bd4c8ce4 AM |
426 | /* |
427 | * This is like invalidate_complete_page(), except it ignores the page's | |
428 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
429 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
430 | * shrink_page_list() has a temp ref on them, or because they're transiently |
431 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
432 | */ |
433 | static int | |
434 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
435 | { | |
436 | if (page->mapping != mapping) | |
437 | return 0; | |
438 | ||
266cf658 | 439 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
440 | return 0; |
441 | ||
19fd6231 | 442 | spin_lock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
443 | if (PageDirty(page)) |
444 | goto failed; | |
445 | ||
266cf658 | 446 | BUG_ON(page_has_private(page)); |
e64a782f | 447 | __delete_from_page_cache(page); |
19fd6231 | 448 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 449 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
450 | |
451 | if (mapping->a_ops->freepage) | |
452 | mapping->a_ops->freepage(page); | |
453 | ||
bd4c8ce4 AM |
454 | page_cache_release(page); /* pagecache ref */ |
455 | return 1; | |
456 | failed: | |
19fd6231 | 457 | spin_unlock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
458 | return 0; |
459 | } | |
460 | ||
e3db7691 TM |
461 | static int do_launder_page(struct address_space *mapping, struct page *page) |
462 | { | |
463 | if (!PageDirty(page)) | |
464 | return 0; | |
465 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
466 | return 0; | |
467 | return mapping->a_ops->launder_page(page); | |
468 | } | |
469 | ||
1da177e4 LT |
470 | /** |
471 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 472 | * @mapping: the address_space |
1da177e4 LT |
473 | * @start: the page offset 'from' which to invalidate |
474 | * @end: the page offset 'to' which to invalidate (inclusive) | |
475 | * | |
476 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
477 | * invalidation. | |
478 | * | |
6ccfa806 | 479 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
480 | */ |
481 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
482 | pgoff_t start, pgoff_t end) | |
483 | { | |
484 | struct pagevec pvec; | |
b85e0eff | 485 | pgoff_t index; |
1da177e4 LT |
486 | int i; |
487 | int ret = 0; | |
0dd1334f | 488 | int ret2 = 0; |
1da177e4 | 489 | int did_range_unmap = 0; |
1da177e4 | 490 | |
3167760f | 491 | cleancache_invalidate_inode(mapping); |
1da177e4 | 492 | pagevec_init(&pvec, 0); |
b85e0eff HD |
493 | index = start; |
494 | while (index <= end && pagevec_lookup(&pvec, mapping, index, | |
495 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
569b846d | 496 | mem_cgroup_uncharge_start(); |
7b965e08 | 497 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 498 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
499 | |
500 | /* We rely upon deletion not changing page->index */ | |
501 | index = page->index; | |
502 | if (index > end) | |
503 | break; | |
1da177e4 LT |
504 | |
505 | lock_page(page); | |
b85e0eff | 506 | WARN_ON(page->index != index); |
1da177e4 LT |
507 | if (page->mapping != mapping) { |
508 | unlock_page(page); | |
509 | continue; | |
510 | } | |
1da177e4 | 511 | wait_on_page_writeback(page); |
d00806b1 | 512 | if (page_mapped(page)) { |
1da177e4 LT |
513 | if (!did_range_unmap) { |
514 | /* | |
515 | * Zap the rest of the file in one hit. | |
516 | */ | |
517 | unmap_mapping_range(mapping, | |
b85e0eff HD |
518 | (loff_t)index << PAGE_CACHE_SHIFT, |
519 | (loff_t)(1 + end - index) | |
520 | << PAGE_CACHE_SHIFT, | |
1da177e4 LT |
521 | 0); |
522 | did_range_unmap = 1; | |
523 | } else { | |
524 | /* | |
525 | * Just zap this page | |
526 | */ | |
527 | unmap_mapping_range(mapping, | |
b85e0eff HD |
528 | (loff_t)index << PAGE_CACHE_SHIFT, |
529 | PAGE_CACHE_SIZE, 0); | |
1da177e4 LT |
530 | } |
531 | } | |
d00806b1 | 532 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
533 | ret2 = do_launder_page(mapping, page); |
534 | if (ret2 == 0) { | |
535 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 536 | ret2 = -EBUSY; |
0dd1334f HH |
537 | } |
538 | if (ret2 < 0) | |
539 | ret = ret2; | |
1da177e4 LT |
540 | unlock_page(page); |
541 | } | |
542 | pagevec_release(&pvec); | |
569b846d | 543 | mem_cgroup_uncharge_end(); |
1da177e4 | 544 | cond_resched(); |
b85e0eff | 545 | index++; |
1da177e4 | 546 | } |
3167760f | 547 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
548 | return ret; |
549 | } | |
550 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
551 | ||
552 | /** | |
553 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 554 | * @mapping: the address_space |
1da177e4 LT |
555 | * |
556 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
557 | * invalidation. | |
558 | * | |
e9de25dd | 559 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
560 | */ |
561 | int invalidate_inode_pages2(struct address_space *mapping) | |
562 | { | |
563 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
564 | } | |
565 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 566 | |
567 | /** | |
568 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
569 | * @inode: inode | |
8a549bea | 570 | * @newsize: new file size |
25d9e2d1 | 571 | * |
572 | * inode's new i_size must already be written before truncate_pagecache | |
573 | * is called. | |
574 | * | |
575 | * This function should typically be called before the filesystem | |
576 | * releases resources associated with the freed range (eg. deallocates | |
577 | * blocks). This way, pagecache will always stay logically coherent | |
578 | * with on-disk format, and the filesystem would not have to deal with | |
579 | * situations such as writepage being called for a page that has already | |
580 | * had its underlying blocks deallocated. | |
581 | */ | |
7caef267 | 582 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 583 | { |
cedabed4 | 584 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 585 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
586 | |
587 | /* | |
588 | * unmap_mapping_range is called twice, first simply for | |
589 | * efficiency so that truncate_inode_pages does fewer | |
590 | * single-page unmaps. However after this first call, and | |
591 | * before truncate_inode_pages finishes, it is possible for | |
592 | * private pages to be COWed, which remain after | |
593 | * truncate_inode_pages finishes, hence the second | |
594 | * unmap_mapping_range call must be made for correctness. | |
595 | */ | |
8a549bea HD |
596 | unmap_mapping_range(mapping, holebegin, 0, 1); |
597 | truncate_inode_pages(mapping, newsize); | |
598 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 599 | } |
600 | EXPORT_SYMBOL(truncate_pagecache); | |
601 | ||
2c27c65e CH |
602 | /** |
603 | * truncate_setsize - update inode and pagecache for a new file size | |
604 | * @inode: inode | |
605 | * @newsize: new file size | |
606 | * | |
382e27da JK |
607 | * truncate_setsize updates i_size and performs pagecache truncation (if |
608 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
609 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 610 | * |
382e27da JK |
611 | * Must be called with inode_mutex held and before all filesystem specific |
612 | * block truncation has been performed. | |
2c27c65e CH |
613 | */ |
614 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
615 | { | |
2c27c65e | 616 | i_size_write(inode, newsize); |
7caef267 | 617 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
618 | } |
619 | EXPORT_SYMBOL(truncate_setsize); | |
620 | ||
623e3db9 HD |
621 | /** |
622 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
623 | * @inode: inode | |
624 | * @lstart: offset of beginning of hole | |
625 | * @lend: offset of last byte of hole | |
626 | * | |
627 | * This function should typically be called before the filesystem | |
628 | * releases resources associated with the freed range (eg. deallocates | |
629 | * blocks). This way, pagecache will always stay logically coherent | |
630 | * with on-disk format, and the filesystem would not have to deal with | |
631 | * situations such as writepage being called for a page that has already | |
632 | * had its underlying blocks deallocated. | |
633 | */ | |
634 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
635 | { | |
636 | struct address_space *mapping = inode->i_mapping; | |
637 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
638 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
639 | /* | |
640 | * This rounding is currently just for example: unmap_mapping_range | |
641 | * expands its hole outwards, whereas we want it to contract the hole | |
642 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
643 | * doing their own page rounding first. Note that unmap_mapping_range |
644 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
645 | */ |
646 | ||
647 | /* | |
648 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
649 | * once (before truncating pagecache), and without "even_cows" flag: | |
650 | * hole-punching should not remove private COWed pages from the hole. | |
651 | */ | |
652 | if ((u64)unmap_end > (u64)unmap_start) | |
653 | unmap_mapping_range(mapping, unmap_start, | |
654 | 1 + unmap_end - unmap_start, 0); | |
655 | truncate_inode_pages_range(mapping, lstart, lend); | |
656 | } | |
657 | EXPORT_SYMBOL(truncate_pagecache_range); |