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 | * | |
6 | * 10Sep2002 akpm@zip.com.au | |
7 | * Initial version. | |
8 | */ | |
9 | ||
10 | #include <linux/kernel.h> | |
11 | #include <linux/mm.h> | |
0fd0e6b0 | 12 | #include <linux/swap.h> |
1da177e4 LT |
13 | #include <linux/module.h> |
14 | #include <linux/pagemap.h> | |
15 | #include <linux/pagevec.h> | |
e08748ce | 16 | #include <linux/task_io_accounting_ops.h> |
1da177e4 | 17 | #include <linux/buffer_head.h> /* grr. try_to_release_page, |
aaa4059b | 18 | do_invalidatepage */ |
1da177e4 LT |
19 | |
20 | ||
cf9a2ae8 DH |
21 | /** |
22 | * do_invalidatepage - invalidate part of all of a page | |
23 | * @page: the page which is affected | |
24 | * @offset: the index of the truncation point | |
25 | * | |
26 | * do_invalidatepage() is called when all or part of the page has become | |
27 | * invalidated by a truncate operation. | |
28 | * | |
29 | * do_invalidatepage() does not have to release all buffers, but it must | |
30 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
31 | * is underway against any of the blocks which are outside the truncation | |
32 | * point. Because the caller is about to free (and possibly reuse) those | |
33 | * blocks on-disk. | |
34 | */ | |
35 | void do_invalidatepage(struct page *page, unsigned long offset) | |
36 | { | |
37 | void (*invalidatepage)(struct page *, unsigned long); | |
38 | invalidatepage = page->mapping->a_ops->invalidatepage; | |
9361401e | 39 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
40 | if (!invalidatepage) |
41 | invalidatepage = block_invalidatepage; | |
9361401e | 42 | #endif |
cf9a2ae8 DH |
43 | if (invalidatepage) |
44 | (*invalidatepage)(page, offset); | |
45 | } | |
46 | ||
1da177e4 LT |
47 | static inline void truncate_partial_page(struct page *page, unsigned partial) |
48 | { | |
49 | memclear_highpage_flush(page, partial, PAGE_CACHE_SIZE-partial); | |
50 | if (PagePrivate(page)) | |
51 | do_invalidatepage(page, partial); | |
52 | } | |
53 | ||
fba2591b LT |
54 | void cancel_dirty_page(struct page *page, unsigned int account_size) |
55 | { | |
56 | /* If we're cancelling the page, it had better not be mapped any more */ | |
57 | if (page_mapped(page)) { | |
58 | static unsigned int warncount; | |
59 | ||
60 | WARN_ON(++warncount < 5); | |
61 | } | |
62 | ||
63 | if (TestClearPageDirty(page) && account_size) | |
64 | task_io_account_cancelled_write(account_size); | |
65 | } | |
66 | ||
67 | ||
1da177e4 LT |
68 | /* |
69 | * If truncate cannot remove the fs-private metadata from the page, the page | |
70 | * becomes anonymous. It will be left on the LRU and may even be mapped into | |
71 | * user pagetables if we're racing with filemap_nopage(). | |
72 | * | |
73 | * We need to bale out if page->mapping is no longer equal to the original | |
74 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
75 | * its lock, b) when a concurrent invalidate_inode_pages got there first and | |
76 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. | |
77 | */ | |
78 | static void | |
79 | truncate_complete_page(struct address_space *mapping, struct page *page) | |
80 | { | |
81 | if (page->mapping != mapping) | |
82 | return; | |
83 | ||
84 | if (PagePrivate(page)) | |
85 | do_invalidatepage(page, 0); | |
86 | ||
fba2591b LT |
87 | cancel_dirty_page(page, PAGE_CACHE_SIZE); |
88 | ||
1da177e4 LT |
89 | ClearPageUptodate(page); |
90 | ClearPageMappedToDisk(page); | |
91 | remove_from_page_cache(page); | |
92 | page_cache_release(page); /* pagecache ref */ | |
93 | } | |
94 | ||
95 | /* | |
96 | * This is for invalidate_inode_pages(). That function can be called at | |
97 | * any time, and is not supposed to throw away dirty pages. But pages can | |
0fd0e6b0 NP |
98 | * be marked dirty at any time too, so use remove_mapping which safely |
99 | * discards clean, unused pages. | |
1da177e4 LT |
100 | * |
101 | * Returns non-zero if the page was successfully invalidated. | |
102 | */ | |
103 | static int | |
104 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
105 | { | |
0fd0e6b0 NP |
106 | int ret; |
107 | ||
1da177e4 LT |
108 | if (page->mapping != mapping) |
109 | return 0; | |
110 | ||
111 | if (PagePrivate(page) && !try_to_release_page(page, 0)) | |
112 | return 0; | |
113 | ||
0fd0e6b0 | 114 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
115 | |
116 | return ret; | |
1da177e4 LT |
117 | } |
118 | ||
119 | /** | |
d7339071 HR |
120 | * truncate_inode_pages - truncate range of pages specified by start and |
121 | * end byte offsets | |
1da177e4 LT |
122 | * @mapping: mapping to truncate |
123 | * @lstart: offset from which to truncate | |
d7339071 | 124 | * @lend: offset to which to truncate |
1da177e4 | 125 | * |
d7339071 HR |
126 | * Truncate the page cache, removing the pages that are between |
127 | * specified offsets (and zeroing out partial page | |
128 | * (if lstart is not page aligned)). | |
1da177e4 LT |
129 | * |
130 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
131 | * block on page locks and it will not block on writeback. The second pass | |
132 | * will wait. This is to prevent as much IO as possible in the affected region. | |
133 | * The first pass will remove most pages, so the search cost of the second pass | |
134 | * is low. | |
135 | * | |
136 | * When looking at page->index outside the page lock we need to be careful to | |
137 | * copy it into a local to avoid races (it could change at any time). | |
138 | * | |
139 | * We pass down the cache-hot hint to the page freeing code. Even if the | |
140 | * mapping is large, it is probably the case that the final pages are the most | |
141 | * recently touched, and freeing happens in ascending file offset order. | |
1da177e4 | 142 | */ |
d7339071 HR |
143 | void truncate_inode_pages_range(struct address_space *mapping, |
144 | loff_t lstart, loff_t lend) | |
1da177e4 LT |
145 | { |
146 | const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; | |
d7339071 | 147 | pgoff_t end; |
1da177e4 LT |
148 | const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1); |
149 | struct pagevec pvec; | |
150 | pgoff_t next; | |
151 | int i; | |
152 | ||
153 | if (mapping->nrpages == 0) | |
154 | return; | |
155 | ||
d7339071 HR |
156 | BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1)); |
157 | end = (lend >> PAGE_CACHE_SHIFT); | |
158 | ||
1da177e4 LT |
159 | pagevec_init(&pvec, 0); |
160 | next = start; | |
d7339071 HR |
161 | while (next <= end && |
162 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
1da177e4 LT |
163 | for (i = 0; i < pagevec_count(&pvec); i++) { |
164 | struct page *page = pvec.pages[i]; | |
165 | pgoff_t page_index = page->index; | |
166 | ||
d7339071 HR |
167 | if (page_index > end) { |
168 | next = page_index; | |
169 | break; | |
170 | } | |
171 | ||
1da177e4 LT |
172 | if (page_index > next) |
173 | next = page_index; | |
174 | next++; | |
175 | if (TestSetPageLocked(page)) | |
176 | continue; | |
177 | if (PageWriteback(page)) { | |
178 | unlock_page(page); | |
179 | continue; | |
180 | } | |
181 | truncate_complete_page(mapping, page); | |
182 | unlock_page(page); | |
183 | } | |
184 | pagevec_release(&pvec); | |
185 | cond_resched(); | |
186 | } | |
187 | ||
188 | if (partial) { | |
189 | struct page *page = find_lock_page(mapping, start - 1); | |
190 | if (page) { | |
191 | wait_on_page_writeback(page); | |
192 | truncate_partial_page(page, partial); | |
193 | unlock_page(page); | |
194 | page_cache_release(page); | |
195 | } | |
196 | } | |
197 | ||
198 | next = start; | |
199 | for ( ; ; ) { | |
200 | cond_resched(); | |
201 | if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
202 | if (next == start) | |
203 | break; | |
204 | next = start; | |
205 | continue; | |
206 | } | |
d7339071 HR |
207 | if (pvec.pages[0]->index > end) { |
208 | pagevec_release(&pvec); | |
209 | break; | |
210 | } | |
1da177e4 LT |
211 | for (i = 0; i < pagevec_count(&pvec); i++) { |
212 | struct page *page = pvec.pages[i]; | |
213 | ||
d7339071 HR |
214 | if (page->index > end) |
215 | break; | |
1da177e4 LT |
216 | lock_page(page); |
217 | wait_on_page_writeback(page); | |
218 | if (page->index > next) | |
219 | next = page->index; | |
220 | next++; | |
221 | truncate_complete_page(mapping, page); | |
222 | unlock_page(page); | |
223 | } | |
224 | pagevec_release(&pvec); | |
225 | } | |
226 | } | |
d7339071 | 227 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 228 | |
d7339071 HR |
229 | /** |
230 | * truncate_inode_pages - truncate *all* the pages from an offset | |
231 | * @mapping: mapping to truncate | |
232 | * @lstart: offset from which to truncate | |
233 | * | |
1b1dcc1b | 234 | * Called under (and serialised by) inode->i_mutex. |
d7339071 HR |
235 | */ |
236 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
237 | { | |
238 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
239 | } | |
1da177e4 LT |
240 | EXPORT_SYMBOL(truncate_inode_pages); |
241 | ||
242 | /** | |
243 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
244 | * @mapping: the address_space which holds the pages to invalidate | |
245 | * @start: the offset 'from' which to invalidate | |
246 | * @end: the offset 'to' which to invalidate (inclusive) | |
247 | * | |
248 | * This function only removes the unlocked pages, if you want to | |
249 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
250 | * | |
251 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
252 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
253 | * pagetables. | |
254 | */ | |
255 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
256 | pgoff_t start, pgoff_t end) | |
257 | { | |
258 | struct pagevec pvec; | |
259 | pgoff_t next = start; | |
260 | unsigned long ret = 0; | |
261 | int i; | |
262 | ||
263 | pagevec_init(&pvec, 0); | |
264 | while (next <= end && | |
265 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
266 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
267 | struct page *page = pvec.pages[i]; | |
e0f23603 N |
268 | pgoff_t index; |
269 | int lock_failed; | |
1da177e4 | 270 | |
e0f23603 N |
271 | lock_failed = TestSetPageLocked(page); |
272 | ||
273 | /* | |
274 | * We really shouldn't be looking at the ->index of an | |
275 | * unlocked page. But we're not allowed to lock these | |
276 | * pages. So we rely upon nobody altering the ->index | |
277 | * of this (pinned-by-us) page. | |
278 | */ | |
279 | index = page->index; | |
280 | if (index > next) | |
281 | next = index; | |
1da177e4 | 282 | next++; |
e0f23603 N |
283 | if (lock_failed) |
284 | continue; | |
285 | ||
1da177e4 LT |
286 | if (PageDirty(page) || PageWriteback(page)) |
287 | goto unlock; | |
288 | if (page_mapped(page)) | |
289 | goto unlock; | |
290 | ret += invalidate_complete_page(mapping, page); | |
291 | unlock: | |
292 | unlock_page(page); | |
293 | if (next > end) | |
294 | break; | |
295 | } | |
296 | pagevec_release(&pvec); | |
1da177e4 LT |
297 | } |
298 | return ret; | |
299 | } | |
300 | ||
301 | unsigned long invalidate_inode_pages(struct address_space *mapping) | |
302 | { | |
303 | return invalidate_mapping_pages(mapping, 0, ~0UL); | |
304 | } | |
1da177e4 LT |
305 | EXPORT_SYMBOL(invalidate_inode_pages); |
306 | ||
bd4c8ce4 AM |
307 | /* |
308 | * This is like invalidate_complete_page(), except it ignores the page's | |
309 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
310 | * invalidation guarantees, and cannot afford to leave pages behind because | |
311 | * shrink_list() has a temp ref on them, or because they're transiently sitting | |
312 | * in the lru_cache_add() pagevecs. | |
313 | */ | |
314 | static int | |
315 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
316 | { | |
317 | if (page->mapping != mapping) | |
318 | return 0; | |
319 | ||
887ed2f3 | 320 | if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
321 | return 0; |
322 | ||
323 | write_lock_irq(&mapping->tree_lock); | |
324 | if (PageDirty(page)) | |
325 | goto failed; | |
326 | ||
327 | BUG_ON(PagePrivate(page)); | |
328 | __remove_from_page_cache(page); | |
329 | write_unlock_irq(&mapping->tree_lock); | |
330 | ClearPageUptodate(page); | |
331 | page_cache_release(page); /* pagecache ref */ | |
332 | return 1; | |
333 | failed: | |
334 | write_unlock_irq(&mapping->tree_lock); | |
335 | return 0; | |
336 | } | |
337 | ||
1da177e4 LT |
338 | /** |
339 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 340 | * @mapping: the address_space |
1da177e4 LT |
341 | * @start: the page offset 'from' which to invalidate |
342 | * @end: the page offset 'to' which to invalidate (inclusive) | |
343 | * | |
344 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
345 | * invalidation. | |
346 | * | |
347 | * Returns -EIO if any pages could not be invalidated. | |
348 | */ | |
349 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
350 | pgoff_t start, pgoff_t end) | |
351 | { | |
352 | struct pagevec pvec; | |
353 | pgoff_t next; | |
354 | int i; | |
355 | int ret = 0; | |
356 | int did_range_unmap = 0; | |
357 | int wrapped = 0; | |
358 | ||
359 | pagevec_init(&pvec, 0); | |
360 | next = start; | |
361 | while (next <= end && !ret && !wrapped && | |
362 | pagevec_lookup(&pvec, mapping, next, | |
363 | min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
364 | for (i = 0; !ret && i < pagevec_count(&pvec); i++) { | |
365 | struct page *page = pvec.pages[i]; | |
366 | pgoff_t page_index; | |
1da177e4 LT |
367 | |
368 | lock_page(page); | |
369 | if (page->mapping != mapping) { | |
370 | unlock_page(page); | |
371 | continue; | |
372 | } | |
373 | page_index = page->index; | |
374 | next = page_index + 1; | |
375 | if (next == 0) | |
376 | wrapped = 1; | |
377 | if (page_index > end) { | |
378 | unlock_page(page); | |
379 | break; | |
380 | } | |
381 | wait_on_page_writeback(page); | |
382 | while (page_mapped(page)) { | |
383 | if (!did_range_unmap) { | |
384 | /* | |
385 | * Zap the rest of the file in one hit. | |
386 | */ | |
387 | unmap_mapping_range(mapping, | |
479ef592 OD |
388 | (loff_t)page_index<<PAGE_CACHE_SHIFT, |
389 | (loff_t)(end - page_index + 1) | |
1da177e4 LT |
390 | << PAGE_CACHE_SHIFT, |
391 | 0); | |
392 | did_range_unmap = 1; | |
393 | } else { | |
394 | /* | |
395 | * Just zap this page | |
396 | */ | |
397 | unmap_mapping_range(mapping, | |
479ef592 | 398 | (loff_t)page_index<<PAGE_CACHE_SHIFT, |
1da177e4 LT |
399 | PAGE_CACHE_SIZE, 0); |
400 | } | |
401 | } | |
fba2591b | 402 | if (!invalidate_complete_page2(mapping, page)) |
1da177e4 | 403 | ret = -EIO; |
1da177e4 LT |
404 | unlock_page(page); |
405 | } | |
406 | pagevec_release(&pvec); | |
407 | cond_resched(); | |
408 | } | |
8258d4a5 | 409 | WARN_ON_ONCE(ret); |
1da177e4 LT |
410 | return ret; |
411 | } | |
412 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
413 | ||
414 | /** | |
415 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 416 | * @mapping: the address_space |
1da177e4 LT |
417 | * |
418 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
419 | * invalidation. | |
420 | * | |
421 | * Returns -EIO if any pages could not be invalidated. | |
422 | */ | |
423 | int invalidate_inode_pages2(struct address_space *mapping) | |
424 | { | |
425 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
426 | } | |
427 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); |