]>
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 | */ | |
38 | radix_tree_delete_item(&mapping->page_tree, index, entry); | |
39 | spin_unlock_irq(&mapping->tree_lock); | |
40 | } | |
1da177e4 | 41 | |
cf9a2ae8 | 42 | /** |
28bc44d7 | 43 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 | 44 | * @page: the page which is affected |
d47992f8 LC |
45 | * @offset: start of the range to invalidate |
46 | * @length: length of the range to invalidate | |
cf9a2ae8 DH |
47 | * |
48 | * do_invalidatepage() is called when all or part of the page has become | |
49 | * invalidated by a truncate operation. | |
50 | * | |
51 | * do_invalidatepage() does not have to release all buffers, but it must | |
52 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
53 | * is underway against any of the blocks which are outside the truncation | |
54 | * point. Because the caller is about to free (and possibly reuse) those | |
55 | * blocks on-disk. | |
56 | */ | |
d47992f8 LC |
57 | void do_invalidatepage(struct page *page, unsigned int offset, |
58 | unsigned int length) | |
cf9a2ae8 | 59 | { |
d47992f8 LC |
60 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
61 | ||
cf9a2ae8 | 62 | invalidatepage = page->mapping->a_ops->invalidatepage; |
9361401e | 63 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
64 | if (!invalidatepage) |
65 | invalidatepage = block_invalidatepage; | |
9361401e | 66 | #endif |
cf9a2ae8 | 67 | if (invalidatepage) |
d47992f8 | 68 | (*invalidatepage)(page, offset, length); |
cf9a2ae8 DH |
69 | } |
70 | ||
ecdfc978 LT |
71 | /* |
72 | * This cancels just the dirty bit on the kernel page itself, it | |
73 | * does NOT actually remove dirty bits on any mmap's that may be | |
74 | * around. It also leaves the page tagged dirty, so any sync | |
75 | * activity will still find it on the dirty lists, and in particular, | |
76 | * clear_page_dirty_for_io() will still look at the dirty bits in | |
77 | * the VM. | |
78 | * | |
79 | * Doing this should *normally* only ever be done when a page | |
80 | * is truncated, and is not actually mapped anywhere at all. However, | |
81 | * fs/buffer.c does this when it notices that somebody has cleaned | |
82 | * out all the buffers on a page without actually doing it through | |
83 | * the VM. Can you say "ext3 is horribly ugly"? Tought you could. | |
84 | */ | |
fba2591b LT |
85 | void cancel_dirty_page(struct page *page, unsigned int account_size) |
86 | { | |
8368e328 LT |
87 | if (TestClearPageDirty(page)) { |
88 | struct address_space *mapping = page->mapping; | |
89 | if (mapping && mapping_cap_account_dirty(mapping)) { | |
90 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
c9e51e41 PZ |
91 | dec_bdi_stat(mapping->backing_dev_info, |
92 | BDI_RECLAIMABLE); | |
8368e328 LT |
93 | if (account_size) |
94 | task_io_account_cancelled_write(account_size); | |
95 | } | |
3e67c098 | 96 | } |
fba2591b | 97 | } |
8368e328 | 98 | EXPORT_SYMBOL(cancel_dirty_page); |
fba2591b | 99 | |
1da177e4 LT |
100 | /* |
101 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 102 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 103 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
104 | * |
105 | * We need to bale out if page->mapping is no longer equal to the original | |
106 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
fc0ecff6 | 107 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
108 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
109 | */ | |
750b4987 | 110 | static int |
1da177e4 LT |
111 | truncate_complete_page(struct address_space *mapping, struct page *page) |
112 | { | |
113 | if (page->mapping != mapping) | |
750b4987 | 114 | return -EIO; |
1da177e4 | 115 | |
266cf658 | 116 | if (page_has_private(page)) |
d47992f8 | 117 | do_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
1da177e4 | 118 | |
a2b34564 BS |
119 | cancel_dirty_page(page, PAGE_CACHE_SIZE); |
120 | ||
1da177e4 | 121 | ClearPageMappedToDisk(page); |
5adc7b51 | 122 | delete_from_page_cache(page); |
750b4987 | 123 | return 0; |
1da177e4 LT |
124 | } |
125 | ||
126 | /* | |
fc0ecff6 | 127 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 128 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
129 | * be marked dirty at any time too, so use remove_mapping which safely |
130 | * discards clean, unused pages. | |
1da177e4 LT |
131 | * |
132 | * Returns non-zero if the page was successfully invalidated. | |
133 | */ | |
134 | static int | |
135 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
136 | { | |
0fd0e6b0 NP |
137 | int ret; |
138 | ||
1da177e4 LT |
139 | if (page->mapping != mapping) |
140 | return 0; | |
141 | ||
266cf658 | 142 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
143 | return 0; |
144 | ||
0fd0e6b0 | 145 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
146 | |
147 | return ret; | |
1da177e4 LT |
148 | } |
149 | ||
750b4987 NP |
150 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
151 | { | |
152 | if (page_mapped(page)) { | |
153 | unmap_mapping_range(mapping, | |
154 | (loff_t)page->index << PAGE_CACHE_SHIFT, | |
155 | PAGE_CACHE_SIZE, 0); | |
156 | } | |
157 | return truncate_complete_page(mapping, page); | |
158 | } | |
159 | ||
25718736 AK |
160 | /* |
161 | * Used to get rid of pages on hardware memory corruption. | |
162 | */ | |
163 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
164 | { | |
165 | if (!mapping) | |
166 | return -EINVAL; | |
167 | /* | |
168 | * Only punch for normal data pages for now. | |
169 | * Handling other types like directories would need more auditing. | |
170 | */ | |
171 | if (!S_ISREG(mapping->host->i_mode)) | |
172 | return -EIO; | |
173 | return truncate_inode_page(mapping, page); | |
174 | } | |
175 | EXPORT_SYMBOL(generic_error_remove_page); | |
176 | ||
83f78668 WF |
177 | /* |
178 | * Safely invalidate one page from its pagecache mapping. | |
179 | * It only drops clean, unused pages. The page must be locked. | |
180 | * | |
181 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
182 | */ | |
183 | int invalidate_inode_page(struct page *page) | |
184 | { | |
185 | struct address_space *mapping = page_mapping(page); | |
186 | if (!mapping) | |
187 | return 0; | |
188 | if (PageDirty(page) || PageWriteback(page)) | |
189 | return 0; | |
190 | if (page_mapped(page)) | |
191 | return 0; | |
192 | return invalidate_complete_page(mapping, page); | |
193 | } | |
194 | ||
1da177e4 | 195 | /** |
73c1e204 | 196 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
197 | * @mapping: mapping to truncate |
198 | * @lstart: offset from which to truncate | |
5a720394 | 199 | * @lend: offset to which to truncate (inclusive) |
1da177e4 | 200 | * |
d7339071 | 201 | * Truncate the page cache, removing the pages that are between |
5a720394 LC |
202 | * specified offsets (and zeroing out partial pages |
203 | * if lstart or lend + 1 is not page aligned). | |
1da177e4 LT |
204 | * |
205 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
206 | * block on page locks and it will not block on writeback. The second pass | |
207 | * will wait. This is to prevent as much IO as possible in the affected region. | |
208 | * The first pass will remove most pages, so the search cost of the second pass | |
209 | * is low. | |
210 | * | |
1da177e4 LT |
211 | * We pass down the cache-hot hint to the page freeing code. Even if the |
212 | * mapping is large, it is probably the case that the final pages are the most | |
213 | * recently touched, and freeing happens in ascending file offset order. | |
5a720394 LC |
214 | * |
215 | * Note that since ->invalidatepage() accepts range to invalidate | |
216 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not | |
217 | * page aligned properly. | |
1da177e4 | 218 | */ |
d7339071 HR |
219 | void truncate_inode_pages_range(struct address_space *mapping, |
220 | loff_t lstart, loff_t lend) | |
1da177e4 | 221 | { |
5a720394 LC |
222 | pgoff_t start; /* inclusive */ |
223 | pgoff_t end; /* exclusive */ | |
224 | unsigned int partial_start; /* inclusive */ | |
225 | unsigned int partial_end; /* exclusive */ | |
226 | struct pagevec pvec; | |
0cd6144a | 227 | pgoff_t indices[PAGEVEC_SIZE]; |
5a720394 LC |
228 | pgoff_t index; |
229 | int i; | |
1da177e4 | 230 | |
3167760f | 231 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
232 | if (mapping->nrpages == 0) |
233 | return; | |
234 | ||
5a720394 LC |
235 | /* Offsets within partial pages */ |
236 | partial_start = lstart & (PAGE_CACHE_SIZE - 1); | |
237 | partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); | |
238 | ||
239 | /* | |
240 | * 'start' and 'end' always covers the range of pages to be fully | |
241 | * truncated. Partial pages are covered with 'partial_start' at the | |
242 | * start of the range and 'partial_end' at the end of the range. | |
243 | * Note that 'end' is exclusive while 'lend' is inclusive. | |
244 | */ | |
245 | start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
246 | if (lend == -1) | |
247 | /* | |
248 | * lend == -1 indicates end-of-file so we have to set 'end' | |
249 | * to the highest possible pgoff_t and since the type is | |
250 | * unsigned we're using -1. | |
251 | */ | |
252 | end = -1; | |
253 | else | |
254 | end = (lend + 1) >> PAGE_CACHE_SHIFT; | |
d7339071 | 255 | |
1da177e4 | 256 | pagevec_init(&pvec, 0); |
b85e0eff | 257 | index = start; |
0cd6144a JW |
258 | while (index < end && pagevec_lookup_entries(&pvec, mapping, index, |
259 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
260 | indices)) { | |
e5598f8b | 261 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
262 | for (i = 0; i < pagevec_count(&pvec); i++) { |
263 | struct page *page = pvec.pages[i]; | |
1da177e4 | 264 | |
b85e0eff | 265 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 266 | index = indices[i]; |
5a720394 | 267 | if (index >= end) |
d7339071 | 268 | break; |
d7339071 | 269 | |
0cd6144a JW |
270 | if (radix_tree_exceptional_entry(page)) { |
271 | clear_exceptional_entry(mapping, index, page); | |
272 | continue; | |
273 | } | |
274 | ||
529ae9aa | 275 | if (!trylock_page(page)) |
1da177e4 | 276 | continue; |
b85e0eff | 277 | WARN_ON(page->index != index); |
1da177e4 LT |
278 | if (PageWriteback(page)) { |
279 | unlock_page(page); | |
280 | continue; | |
281 | } | |
750b4987 | 282 | truncate_inode_page(mapping, page); |
1da177e4 LT |
283 | unlock_page(page); |
284 | } | |
0cd6144a | 285 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 286 | pagevec_release(&pvec); |
e5598f8b | 287 | mem_cgroup_uncharge_end(); |
1da177e4 | 288 | cond_resched(); |
b85e0eff | 289 | index++; |
1da177e4 LT |
290 | } |
291 | ||
5a720394 | 292 | if (partial_start) { |
1da177e4 LT |
293 | struct page *page = find_lock_page(mapping, start - 1); |
294 | if (page) { | |
5a720394 LC |
295 | unsigned int top = PAGE_CACHE_SIZE; |
296 | if (start > end) { | |
297 | /* Truncation within a single page */ | |
298 | top = partial_end; | |
299 | partial_end = 0; | |
300 | } | |
1da177e4 | 301 | wait_on_page_writeback(page); |
5a720394 LC |
302 | zero_user_segment(page, partial_start, top); |
303 | cleancache_invalidate_page(mapping, page); | |
304 | if (page_has_private(page)) | |
305 | do_invalidatepage(page, partial_start, | |
306 | top - partial_start); | |
1da177e4 LT |
307 | unlock_page(page); |
308 | page_cache_release(page); | |
309 | } | |
310 | } | |
5a720394 LC |
311 | if (partial_end) { |
312 | struct page *page = find_lock_page(mapping, end); | |
313 | if (page) { | |
314 | wait_on_page_writeback(page); | |
315 | zero_user_segment(page, 0, partial_end); | |
316 | cleancache_invalidate_page(mapping, page); | |
317 | if (page_has_private(page)) | |
318 | do_invalidatepage(page, 0, | |
319 | partial_end); | |
320 | unlock_page(page); | |
321 | page_cache_release(page); | |
322 | } | |
323 | } | |
324 | /* | |
325 | * If the truncation happened within a single page no pages | |
326 | * will be released, just zeroed, so we can bail out now. | |
327 | */ | |
328 | if (start >= end) | |
329 | return; | |
1da177e4 | 330 | |
b85e0eff | 331 | index = start; |
1da177e4 LT |
332 | for ( ; ; ) { |
333 | cond_resched(); | |
0cd6144a JW |
334 | if (!pagevec_lookup_entries(&pvec, mapping, index, |
335 | min(end - index, (pgoff_t)PAGEVEC_SIZE), | |
336 | indices)) { | |
b85e0eff | 337 | if (index == start) |
1da177e4 | 338 | break; |
b85e0eff | 339 | index = start; |
1da177e4 LT |
340 | continue; |
341 | } | |
0cd6144a JW |
342 | if (index == start && indices[0] >= end) { |
343 | pagevec_remove_exceptionals(&pvec); | |
d7339071 HR |
344 | pagevec_release(&pvec); |
345 | break; | |
346 | } | |
569b846d | 347 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
348 | for (i = 0; i < pagevec_count(&pvec); i++) { |
349 | struct page *page = pvec.pages[i]; | |
350 | ||
b85e0eff | 351 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 352 | index = indices[i]; |
5a720394 | 353 | if (index >= end) |
d7339071 | 354 | break; |
b85e0eff | 355 | |
0cd6144a JW |
356 | if (radix_tree_exceptional_entry(page)) { |
357 | clear_exceptional_entry(mapping, index, page); | |
358 | continue; | |
359 | } | |
360 | ||
1da177e4 | 361 | lock_page(page); |
b85e0eff | 362 | WARN_ON(page->index != index); |
1da177e4 | 363 | wait_on_page_writeback(page); |
750b4987 | 364 | truncate_inode_page(mapping, page); |
1da177e4 LT |
365 | unlock_page(page); |
366 | } | |
0cd6144a | 367 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 368 | pagevec_release(&pvec); |
569b846d | 369 | mem_cgroup_uncharge_end(); |
b85e0eff | 370 | index++; |
1da177e4 | 371 | } |
3167760f | 372 | cleancache_invalidate_inode(mapping); |
1da177e4 | 373 | } |
d7339071 | 374 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 375 | |
d7339071 HR |
376 | /** |
377 | * truncate_inode_pages - truncate *all* the pages from an offset | |
378 | * @mapping: mapping to truncate | |
379 | * @lstart: offset from which to truncate | |
380 | * | |
1b1dcc1b | 381 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
382 | * |
383 | * Note: When this function returns, there can be a page in the process of | |
384 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
385 | * mapping->nrpages can be non-zero when this function returns even after | |
386 | * truncation of the whole mapping. | |
d7339071 HR |
387 | */ |
388 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
389 | { | |
390 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
391 | } | |
1da177e4 LT |
392 | EXPORT_SYMBOL(truncate_inode_pages); |
393 | ||
28697355 MW |
394 | /** |
395 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
396 | * @mapping: the address_space which holds the pages to invalidate | |
397 | * @start: the offset 'from' which to invalidate | |
398 | * @end: the offset 'to' which to invalidate (inclusive) | |
399 | * | |
400 | * This function only removes the unlocked pages, if you want to | |
401 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
402 | * | |
403 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
404 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
405 | * pagetables. | |
406 | */ | |
407 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 408 | pgoff_t start, pgoff_t end) |
1da177e4 | 409 | { |
0cd6144a | 410 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 411 | struct pagevec pvec; |
b85e0eff | 412 | pgoff_t index = start; |
31560180 MK |
413 | unsigned long ret; |
414 | unsigned long count = 0; | |
1da177e4 LT |
415 | int i; |
416 | ||
31475dd6 HD |
417 | /* |
418 | * Note: this function may get called on a shmem/tmpfs mapping: | |
419 | * pagevec_lookup() might then return 0 prematurely (because it | |
420 | * got a gangful of swap entries); but it's hardly worth worrying | |
421 | * about - it can rarely have anything to free from such a mapping | |
422 | * (most pages are dirty), and already skips over any difficulties. | |
423 | */ | |
424 | ||
1da177e4 | 425 | pagevec_init(&pvec, 0); |
0cd6144a JW |
426 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
427 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
428 | indices)) { | |
569b846d | 429 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
430 | for (i = 0; i < pagevec_count(&pvec); i++) { |
431 | struct page *page = pvec.pages[i]; | |
e0f23603 | 432 | |
b85e0eff | 433 | /* We rely upon deletion not changing page->index */ |
0cd6144a | 434 | index = indices[i]; |
b85e0eff HD |
435 | if (index > end) |
436 | break; | |
e0f23603 | 437 | |
0cd6144a JW |
438 | if (radix_tree_exceptional_entry(page)) { |
439 | clear_exceptional_entry(mapping, index, page); | |
440 | continue; | |
441 | } | |
442 | ||
b85e0eff HD |
443 | if (!trylock_page(page)) |
444 | continue; | |
445 | WARN_ON(page->index != index); | |
31560180 | 446 | ret = invalidate_inode_page(page); |
1da177e4 | 447 | unlock_page(page); |
31560180 MK |
448 | /* |
449 | * Invalidation is a hint that the page is no longer | |
450 | * of interest and try to speed up its reclaim. | |
451 | */ | |
452 | if (!ret) | |
453 | deactivate_page(page); | |
454 | count += ret; | |
1da177e4 | 455 | } |
0cd6144a | 456 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 457 | pagevec_release(&pvec); |
569b846d | 458 | mem_cgroup_uncharge_end(); |
28697355 | 459 | cond_resched(); |
b85e0eff | 460 | index++; |
1da177e4 | 461 | } |
31560180 | 462 | return count; |
1da177e4 | 463 | } |
54bc4855 | 464 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 465 | |
bd4c8ce4 AM |
466 | /* |
467 | * This is like invalidate_complete_page(), except it ignores the page's | |
468 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
469 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
470 | * shrink_page_list() has a temp ref on them, or because they're transiently |
471 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
472 | */ |
473 | static int | |
474 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
475 | { | |
476 | if (page->mapping != mapping) | |
477 | return 0; | |
478 | ||
266cf658 | 479 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
480 | return 0; |
481 | ||
19fd6231 | 482 | spin_lock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
483 | if (PageDirty(page)) |
484 | goto failed; | |
485 | ||
266cf658 | 486 | BUG_ON(page_has_private(page)); |
e64a782f | 487 | __delete_from_page_cache(page); |
19fd6231 | 488 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 489 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
490 | |
491 | if (mapping->a_ops->freepage) | |
492 | mapping->a_ops->freepage(page); | |
493 | ||
bd4c8ce4 AM |
494 | page_cache_release(page); /* pagecache ref */ |
495 | return 1; | |
496 | failed: | |
19fd6231 | 497 | spin_unlock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
498 | return 0; |
499 | } | |
500 | ||
e3db7691 TM |
501 | static int do_launder_page(struct address_space *mapping, struct page *page) |
502 | { | |
503 | if (!PageDirty(page)) | |
504 | return 0; | |
505 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
506 | return 0; | |
507 | return mapping->a_ops->launder_page(page); | |
508 | } | |
509 | ||
1da177e4 LT |
510 | /** |
511 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 512 | * @mapping: the address_space |
1da177e4 LT |
513 | * @start: the page offset 'from' which to invalidate |
514 | * @end: the page offset 'to' which to invalidate (inclusive) | |
515 | * | |
516 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
517 | * invalidation. | |
518 | * | |
6ccfa806 | 519 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
520 | */ |
521 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
522 | pgoff_t start, pgoff_t end) | |
523 | { | |
0cd6144a | 524 | pgoff_t indices[PAGEVEC_SIZE]; |
1da177e4 | 525 | struct pagevec pvec; |
b85e0eff | 526 | pgoff_t index; |
1da177e4 LT |
527 | int i; |
528 | int ret = 0; | |
0dd1334f | 529 | int ret2 = 0; |
1da177e4 | 530 | int did_range_unmap = 0; |
1da177e4 | 531 | |
3167760f | 532 | cleancache_invalidate_inode(mapping); |
1da177e4 | 533 | pagevec_init(&pvec, 0); |
b85e0eff | 534 | index = start; |
0cd6144a JW |
535 | while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, |
536 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, | |
537 | indices)) { | |
569b846d | 538 | mem_cgroup_uncharge_start(); |
7b965e08 | 539 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 540 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
541 | |
542 | /* We rely upon deletion not changing page->index */ | |
0cd6144a | 543 | index = indices[i]; |
b85e0eff HD |
544 | if (index > end) |
545 | break; | |
1da177e4 | 546 | |
0cd6144a JW |
547 | if (radix_tree_exceptional_entry(page)) { |
548 | clear_exceptional_entry(mapping, index, page); | |
549 | continue; | |
550 | } | |
551 | ||
1da177e4 | 552 | lock_page(page); |
b85e0eff | 553 | WARN_ON(page->index != index); |
1da177e4 LT |
554 | if (page->mapping != mapping) { |
555 | unlock_page(page); | |
556 | continue; | |
557 | } | |
1da177e4 | 558 | wait_on_page_writeback(page); |
d00806b1 | 559 | if (page_mapped(page)) { |
1da177e4 LT |
560 | if (!did_range_unmap) { |
561 | /* | |
562 | * Zap the rest of the file in one hit. | |
563 | */ | |
564 | unmap_mapping_range(mapping, | |
b85e0eff HD |
565 | (loff_t)index << PAGE_CACHE_SHIFT, |
566 | (loff_t)(1 + end - index) | |
567 | << PAGE_CACHE_SHIFT, | |
1da177e4 LT |
568 | 0); |
569 | did_range_unmap = 1; | |
570 | } else { | |
571 | /* | |
572 | * Just zap this page | |
573 | */ | |
574 | unmap_mapping_range(mapping, | |
b85e0eff HD |
575 | (loff_t)index << PAGE_CACHE_SHIFT, |
576 | PAGE_CACHE_SIZE, 0); | |
1da177e4 LT |
577 | } |
578 | } | |
d00806b1 | 579 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
580 | ret2 = do_launder_page(mapping, page); |
581 | if (ret2 == 0) { | |
582 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 583 | ret2 = -EBUSY; |
0dd1334f HH |
584 | } |
585 | if (ret2 < 0) | |
586 | ret = ret2; | |
1da177e4 LT |
587 | unlock_page(page); |
588 | } | |
0cd6144a | 589 | pagevec_remove_exceptionals(&pvec); |
1da177e4 | 590 | pagevec_release(&pvec); |
569b846d | 591 | mem_cgroup_uncharge_end(); |
1da177e4 | 592 | cond_resched(); |
b85e0eff | 593 | index++; |
1da177e4 | 594 | } |
3167760f | 595 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
596 | return ret; |
597 | } | |
598 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
599 | ||
600 | /** | |
601 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 602 | * @mapping: the address_space |
1da177e4 LT |
603 | * |
604 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
605 | * invalidation. | |
606 | * | |
e9de25dd | 607 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
608 | */ |
609 | int invalidate_inode_pages2(struct address_space *mapping) | |
610 | { | |
611 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
612 | } | |
613 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 614 | |
615 | /** | |
616 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
617 | * @inode: inode | |
8a549bea | 618 | * @newsize: new file size |
25d9e2d1 | 619 | * |
620 | * inode's new i_size must already be written before truncate_pagecache | |
621 | * is called. | |
622 | * | |
623 | * This function should typically be called before the filesystem | |
624 | * releases resources associated with the freed range (eg. deallocates | |
625 | * blocks). This way, pagecache will always stay logically coherent | |
626 | * with on-disk format, and the filesystem would not have to deal with | |
627 | * situations such as writepage being called for a page that has already | |
628 | * had its underlying blocks deallocated. | |
629 | */ | |
7caef267 | 630 | void truncate_pagecache(struct inode *inode, loff_t newsize) |
25d9e2d1 | 631 | { |
cedabed4 | 632 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 633 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
634 | |
635 | /* | |
636 | * unmap_mapping_range is called twice, first simply for | |
637 | * efficiency so that truncate_inode_pages does fewer | |
638 | * single-page unmaps. However after this first call, and | |
639 | * before truncate_inode_pages finishes, it is possible for | |
640 | * private pages to be COWed, which remain after | |
641 | * truncate_inode_pages finishes, hence the second | |
642 | * unmap_mapping_range call must be made for correctness. | |
643 | */ | |
8a549bea HD |
644 | unmap_mapping_range(mapping, holebegin, 0, 1); |
645 | truncate_inode_pages(mapping, newsize); | |
646 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 647 | } |
648 | EXPORT_SYMBOL(truncate_pagecache); | |
649 | ||
2c27c65e CH |
650 | /** |
651 | * truncate_setsize - update inode and pagecache for a new file size | |
652 | * @inode: inode | |
653 | * @newsize: new file size | |
654 | * | |
382e27da JK |
655 | * truncate_setsize updates i_size and performs pagecache truncation (if |
656 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
657 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 658 | * |
382e27da JK |
659 | * Must be called with inode_mutex held and before all filesystem specific |
660 | * block truncation has been performed. | |
2c27c65e CH |
661 | */ |
662 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
663 | { | |
2c27c65e | 664 | i_size_write(inode, newsize); |
7caef267 | 665 | truncate_pagecache(inode, newsize); |
2c27c65e CH |
666 | } |
667 | EXPORT_SYMBOL(truncate_setsize); | |
668 | ||
623e3db9 HD |
669 | /** |
670 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
671 | * @inode: inode | |
672 | * @lstart: offset of beginning of hole | |
673 | * @lend: offset of last byte of hole | |
674 | * | |
675 | * This function should typically be called before the filesystem | |
676 | * releases resources associated with the freed range (eg. deallocates | |
677 | * blocks). This way, pagecache will always stay logically coherent | |
678 | * with on-disk format, and the filesystem would not have to deal with | |
679 | * situations such as writepage being called for a page that has already | |
680 | * had its underlying blocks deallocated. | |
681 | */ | |
682 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
683 | { | |
684 | struct address_space *mapping = inode->i_mapping; | |
685 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
686 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
687 | /* | |
688 | * This rounding is currently just for example: unmap_mapping_range | |
689 | * expands its hole outwards, whereas we want it to contract the hole | |
690 | * inwards. However, existing callers of truncate_pagecache_range are | |
5a720394 LC |
691 | * doing their own page rounding first. Note that unmap_mapping_range |
692 | * allows holelen 0 for all, and we allow lend -1 for end of file. | |
623e3db9 HD |
693 | */ |
694 | ||
695 | /* | |
696 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
697 | * once (before truncating pagecache), and without "even_cows" flag: | |
698 | * hole-punching should not remove private COWed pages from the hole. | |
699 | */ | |
700 | if ((u64)unmap_end > (u64)unmap_start) | |
701 | unmap_mapping_range(mapping, unmap_start, | |
702 | 1 + unmap_end - unmap_start, 0); | |
703 | truncate_inode_pages_range(mapping, lstart, lend); | |
704 | } | |
705 | EXPORT_SYMBOL(truncate_pagecache_range); |