]>
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 | ||
ecdfc978 LT |
54 | /* |
55 | * This cancels just the dirty bit on the kernel page itself, it | |
56 | * does NOT actually remove dirty bits on any mmap's that may be | |
57 | * around. It also leaves the page tagged dirty, so any sync | |
58 | * activity will still find it on the dirty lists, and in particular, | |
59 | * clear_page_dirty_for_io() will still look at the dirty bits in | |
60 | * the VM. | |
61 | * | |
62 | * Doing this should *normally* only ever be done when a page | |
63 | * is truncated, and is not actually mapped anywhere at all. However, | |
64 | * fs/buffer.c does this when it notices that somebody has cleaned | |
65 | * out all the buffers on a page without actually doing it through | |
66 | * the VM. Can you say "ext3 is horribly ugly"? Tought you could. | |
67 | */ | |
fba2591b LT |
68 | void cancel_dirty_page(struct page *page, unsigned int account_size) |
69 | { | |
8368e328 LT |
70 | if (TestClearPageDirty(page)) { |
71 | struct address_space *mapping = page->mapping; | |
72 | if (mapping && mapping_cap_account_dirty(mapping)) { | |
73 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
74 | if (account_size) | |
75 | task_io_account_cancelled_write(account_size); | |
76 | } | |
3e67c098 | 77 | } |
fba2591b | 78 | } |
8368e328 | 79 | EXPORT_SYMBOL(cancel_dirty_page); |
fba2591b | 80 | |
1da177e4 LT |
81 | /* |
82 | * If truncate cannot remove the fs-private metadata from the page, the page | |
83 | * becomes anonymous. It will be left on the LRU and may even be mapped into | |
84 | * user pagetables if we're racing with filemap_nopage(). | |
85 | * | |
86 | * We need to bale out if page->mapping is no longer equal to the original | |
87 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
88 | * its lock, b) when a concurrent invalidate_inode_pages got there first and | |
89 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. | |
90 | */ | |
91 | static void | |
92 | truncate_complete_page(struct address_space *mapping, struct page *page) | |
93 | { | |
94 | if (page->mapping != mapping) | |
95 | return; | |
96 | ||
3e67c098 AM |
97 | cancel_dirty_page(page, PAGE_CACHE_SIZE); |
98 | ||
1da177e4 LT |
99 | if (PagePrivate(page)) |
100 | do_invalidatepage(page, 0); | |
101 | ||
1da177e4 LT |
102 | ClearPageUptodate(page); |
103 | ClearPageMappedToDisk(page); | |
104 | remove_from_page_cache(page); | |
105 | page_cache_release(page); /* pagecache ref */ | |
106 | } | |
107 | ||
108 | /* | |
109 | * This is for invalidate_inode_pages(). That function can be called at | |
110 | * any time, and is not supposed to throw away dirty pages. But pages can | |
0fd0e6b0 NP |
111 | * be marked dirty at any time too, so use remove_mapping which safely |
112 | * discards clean, unused pages. | |
1da177e4 LT |
113 | * |
114 | * Returns non-zero if the page was successfully invalidated. | |
115 | */ | |
116 | static int | |
117 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
118 | { | |
0fd0e6b0 NP |
119 | int ret; |
120 | ||
1da177e4 LT |
121 | if (page->mapping != mapping) |
122 | return 0; | |
123 | ||
124 | if (PagePrivate(page) && !try_to_release_page(page, 0)) | |
125 | return 0; | |
126 | ||
0fd0e6b0 | 127 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
128 | |
129 | return ret; | |
1da177e4 LT |
130 | } |
131 | ||
132 | /** | |
d7339071 HR |
133 | * truncate_inode_pages - truncate range of pages specified by start and |
134 | * end byte offsets | |
1da177e4 LT |
135 | * @mapping: mapping to truncate |
136 | * @lstart: offset from which to truncate | |
d7339071 | 137 | * @lend: offset to which to truncate |
1da177e4 | 138 | * |
d7339071 HR |
139 | * Truncate the page cache, removing the pages that are between |
140 | * specified offsets (and zeroing out partial page | |
141 | * (if lstart is not page aligned)). | |
1da177e4 LT |
142 | * |
143 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
144 | * block on page locks and it will not block on writeback. The second pass | |
145 | * will wait. This is to prevent as much IO as possible in the affected region. | |
146 | * The first pass will remove most pages, so the search cost of the second pass | |
147 | * is low. | |
148 | * | |
149 | * When looking at page->index outside the page lock we need to be careful to | |
150 | * copy it into a local to avoid races (it could change at any time). | |
151 | * | |
152 | * We pass down the cache-hot hint to the page freeing code. Even if the | |
153 | * mapping is large, it is probably the case that the final pages are the most | |
154 | * recently touched, and freeing happens in ascending file offset order. | |
1da177e4 | 155 | */ |
d7339071 HR |
156 | void truncate_inode_pages_range(struct address_space *mapping, |
157 | loff_t lstart, loff_t lend) | |
1da177e4 LT |
158 | { |
159 | const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; | |
d7339071 | 160 | pgoff_t end; |
1da177e4 LT |
161 | const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1); |
162 | struct pagevec pvec; | |
163 | pgoff_t next; | |
164 | int i; | |
165 | ||
166 | if (mapping->nrpages == 0) | |
167 | return; | |
168 | ||
d7339071 HR |
169 | BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1)); |
170 | end = (lend >> PAGE_CACHE_SHIFT); | |
171 | ||
1da177e4 LT |
172 | pagevec_init(&pvec, 0); |
173 | next = start; | |
d7339071 HR |
174 | while (next <= end && |
175 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
1da177e4 LT |
176 | for (i = 0; i < pagevec_count(&pvec); i++) { |
177 | struct page *page = pvec.pages[i]; | |
178 | pgoff_t page_index = page->index; | |
179 | ||
d7339071 HR |
180 | if (page_index > end) { |
181 | next = page_index; | |
182 | break; | |
183 | } | |
184 | ||
1da177e4 LT |
185 | if (page_index > next) |
186 | next = page_index; | |
187 | next++; | |
188 | if (TestSetPageLocked(page)) | |
189 | continue; | |
190 | if (PageWriteback(page)) { | |
191 | unlock_page(page); | |
192 | continue; | |
193 | } | |
194 | truncate_complete_page(mapping, page); | |
195 | unlock_page(page); | |
196 | } | |
197 | pagevec_release(&pvec); | |
198 | cond_resched(); | |
199 | } | |
200 | ||
201 | if (partial) { | |
202 | struct page *page = find_lock_page(mapping, start - 1); | |
203 | if (page) { | |
204 | wait_on_page_writeback(page); | |
205 | truncate_partial_page(page, partial); | |
206 | unlock_page(page); | |
207 | page_cache_release(page); | |
208 | } | |
209 | } | |
210 | ||
211 | next = start; | |
212 | for ( ; ; ) { | |
213 | cond_resched(); | |
214 | if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
215 | if (next == start) | |
216 | break; | |
217 | next = start; | |
218 | continue; | |
219 | } | |
d7339071 HR |
220 | if (pvec.pages[0]->index > end) { |
221 | pagevec_release(&pvec); | |
222 | break; | |
223 | } | |
1da177e4 LT |
224 | for (i = 0; i < pagevec_count(&pvec); i++) { |
225 | struct page *page = pvec.pages[i]; | |
226 | ||
d7339071 HR |
227 | if (page->index > end) |
228 | break; | |
1da177e4 LT |
229 | lock_page(page); |
230 | wait_on_page_writeback(page); | |
231 | if (page->index > next) | |
232 | next = page->index; | |
233 | next++; | |
234 | truncate_complete_page(mapping, page); | |
235 | unlock_page(page); | |
236 | } | |
237 | pagevec_release(&pvec); | |
238 | } | |
239 | } | |
d7339071 | 240 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 241 | |
d7339071 HR |
242 | /** |
243 | * truncate_inode_pages - truncate *all* the pages from an offset | |
244 | * @mapping: mapping to truncate | |
245 | * @lstart: offset from which to truncate | |
246 | * | |
1b1dcc1b | 247 | * Called under (and serialised by) inode->i_mutex. |
d7339071 HR |
248 | */ |
249 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
250 | { | |
251 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
252 | } | |
1da177e4 LT |
253 | EXPORT_SYMBOL(truncate_inode_pages); |
254 | ||
255 | /** | |
256 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
257 | * @mapping: the address_space which holds the pages to invalidate | |
258 | * @start: the offset 'from' which to invalidate | |
259 | * @end: the offset 'to' which to invalidate (inclusive) | |
260 | * | |
261 | * This function only removes the unlocked pages, if you want to | |
262 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
263 | * | |
264 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
265 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
266 | * pagetables. | |
267 | */ | |
268 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
269 | pgoff_t start, pgoff_t end) | |
270 | { | |
271 | struct pagevec pvec; | |
272 | pgoff_t next = start; | |
273 | unsigned long ret = 0; | |
274 | int i; | |
275 | ||
276 | pagevec_init(&pvec, 0); | |
277 | while (next <= end && | |
278 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
279 | for (i = 0; i < pagevec_count(&pvec); i++) { | |
280 | struct page *page = pvec.pages[i]; | |
e0f23603 N |
281 | pgoff_t index; |
282 | int lock_failed; | |
1da177e4 | 283 | |
e0f23603 N |
284 | lock_failed = TestSetPageLocked(page); |
285 | ||
286 | /* | |
287 | * We really shouldn't be looking at the ->index of an | |
288 | * unlocked page. But we're not allowed to lock these | |
289 | * pages. So we rely upon nobody altering the ->index | |
290 | * of this (pinned-by-us) page. | |
291 | */ | |
292 | index = page->index; | |
293 | if (index > next) | |
294 | next = index; | |
1da177e4 | 295 | next++; |
e0f23603 N |
296 | if (lock_failed) |
297 | continue; | |
298 | ||
1da177e4 LT |
299 | if (PageDirty(page) || PageWriteback(page)) |
300 | goto unlock; | |
301 | if (page_mapped(page)) | |
302 | goto unlock; | |
303 | ret += invalidate_complete_page(mapping, page); | |
304 | unlock: | |
305 | unlock_page(page); | |
306 | if (next > end) | |
307 | break; | |
308 | } | |
309 | pagevec_release(&pvec); | |
1da177e4 LT |
310 | } |
311 | return ret; | |
312 | } | |
54bc4855 | 313 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 314 | |
bd4c8ce4 AM |
315 | /* |
316 | * This is like invalidate_complete_page(), except it ignores the page's | |
317 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
318 | * invalidation guarantees, and cannot afford to leave pages behind because | |
319 | * shrink_list() has a temp ref on them, or because they're transiently sitting | |
320 | * in the lru_cache_add() pagevecs. | |
321 | */ | |
322 | static int | |
323 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
324 | { | |
325 | if (page->mapping != mapping) | |
326 | return 0; | |
327 | ||
887ed2f3 | 328 | if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
329 | return 0; |
330 | ||
331 | write_lock_irq(&mapping->tree_lock); | |
332 | if (PageDirty(page)) | |
333 | goto failed; | |
334 | ||
335 | BUG_ON(PagePrivate(page)); | |
336 | __remove_from_page_cache(page); | |
337 | write_unlock_irq(&mapping->tree_lock); | |
338 | ClearPageUptodate(page); | |
339 | page_cache_release(page); /* pagecache ref */ | |
340 | return 1; | |
341 | failed: | |
342 | write_unlock_irq(&mapping->tree_lock); | |
343 | return 0; | |
344 | } | |
345 | ||
e3db7691 TM |
346 | static int do_launder_page(struct address_space *mapping, struct page *page) |
347 | { | |
348 | if (!PageDirty(page)) | |
349 | return 0; | |
350 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
351 | return 0; | |
352 | return mapping->a_ops->launder_page(page); | |
353 | } | |
354 | ||
1da177e4 LT |
355 | /** |
356 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 357 | * @mapping: the address_space |
1da177e4 LT |
358 | * @start: the page offset 'from' which to invalidate |
359 | * @end: the page offset 'to' which to invalidate (inclusive) | |
360 | * | |
361 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
362 | * invalidation. | |
363 | * | |
364 | * Returns -EIO if any pages could not be invalidated. | |
365 | */ | |
366 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
367 | pgoff_t start, pgoff_t end) | |
368 | { | |
369 | struct pagevec pvec; | |
370 | pgoff_t next; | |
371 | int i; | |
372 | int ret = 0; | |
373 | int did_range_unmap = 0; | |
374 | int wrapped = 0; | |
375 | ||
376 | pagevec_init(&pvec, 0); | |
377 | next = start; | |
378 | while (next <= end && !ret && !wrapped && | |
379 | pagevec_lookup(&pvec, mapping, next, | |
380 | min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
381 | for (i = 0; !ret && i < pagevec_count(&pvec); i++) { | |
382 | struct page *page = pvec.pages[i]; | |
383 | pgoff_t page_index; | |
1da177e4 LT |
384 | |
385 | lock_page(page); | |
386 | if (page->mapping != mapping) { | |
387 | unlock_page(page); | |
388 | continue; | |
389 | } | |
390 | page_index = page->index; | |
391 | next = page_index + 1; | |
392 | if (next == 0) | |
393 | wrapped = 1; | |
394 | if (page_index > end) { | |
395 | unlock_page(page); | |
396 | break; | |
397 | } | |
398 | wait_on_page_writeback(page); | |
399 | while (page_mapped(page)) { | |
400 | if (!did_range_unmap) { | |
401 | /* | |
402 | * Zap the rest of the file in one hit. | |
403 | */ | |
404 | unmap_mapping_range(mapping, | |
479ef592 OD |
405 | (loff_t)page_index<<PAGE_CACHE_SHIFT, |
406 | (loff_t)(end - page_index + 1) | |
1da177e4 LT |
407 | << PAGE_CACHE_SHIFT, |
408 | 0); | |
409 | did_range_unmap = 1; | |
410 | } else { | |
411 | /* | |
412 | * Just zap this page | |
413 | */ | |
414 | unmap_mapping_range(mapping, | |
479ef592 | 415 | (loff_t)page_index<<PAGE_CACHE_SHIFT, |
1da177e4 LT |
416 | PAGE_CACHE_SIZE, 0); |
417 | } | |
418 | } | |
e3db7691 TM |
419 | ret = do_launder_page(mapping, page); |
420 | if (ret == 0 && !invalidate_complete_page2(mapping, page)) | |
1da177e4 | 421 | ret = -EIO; |
1da177e4 LT |
422 | unlock_page(page); |
423 | } | |
424 | pagevec_release(&pvec); | |
425 | cond_resched(); | |
426 | } | |
427 | return ret; | |
428 | } | |
429 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
430 | ||
431 | /** | |
432 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 433 | * @mapping: the address_space |
1da177e4 LT |
434 | * |
435 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
436 | * invalidation. | |
437 | * | |
438 | * Returns -EIO if any pages could not be invalidated. | |
439 | */ | |
440 | int invalidate_inode_pages2(struct address_space *mapping) | |
441 | { | |
442 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
443 | } | |
444 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); |