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> |
6cbdf115 | 23 | #include <linux/rmap.h> |
ba470de4 | 24 | #include "internal.h" |
1da177e4 LT |
25 | |
26 | ||
cf9a2ae8 | 27 | /** |
28bc44d7 | 28 | * do_invalidatepage - invalidate part or all of a page |
cf9a2ae8 DH |
29 | * @page: the page which is affected |
30 | * @offset: the index of the truncation point | |
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 | */ | |
41 | void do_invalidatepage(struct page *page, unsigned long offset) | |
42 | { | |
43 | void (*invalidatepage)(struct page *, unsigned long); | |
44 | invalidatepage = page->mapping->a_ops->invalidatepage; | |
9361401e | 45 | #ifdef CONFIG_BLOCK |
cf9a2ae8 DH |
46 | if (!invalidatepage) |
47 | invalidatepage = block_invalidatepage; | |
9361401e | 48 | #endif |
cf9a2ae8 DH |
49 | if (invalidatepage) |
50 | (*invalidatepage)(page, offset); | |
51 | } | |
52 | ||
1da177e4 LT |
53 | static inline void truncate_partial_page(struct page *page, unsigned partial) |
54 | { | |
eebd2aa3 | 55 | zero_user_segment(page, partial, PAGE_CACHE_SIZE); |
3167760f | 56 | cleancache_invalidate_page(page->mapping, page); |
266cf658 | 57 | if (page_has_private(page)) |
1da177e4 LT |
58 | do_invalidatepage(page, partial); |
59 | } | |
60 | ||
ecdfc978 LT |
61 | /* |
62 | * This cancels just the dirty bit on the kernel page itself, it | |
63 | * does NOT actually remove dirty bits on any mmap's that may be | |
64 | * around. It also leaves the page tagged dirty, so any sync | |
65 | * activity will still find it on the dirty lists, and in particular, | |
66 | * clear_page_dirty_for_io() will still look at the dirty bits in | |
67 | * the VM. | |
68 | * | |
69 | * Doing this should *normally* only ever be done when a page | |
70 | * is truncated, and is not actually mapped anywhere at all. However, | |
71 | * fs/buffer.c does this when it notices that somebody has cleaned | |
72 | * out all the buffers on a page without actually doing it through | |
73 | * the VM. Can you say "ext3 is horribly ugly"? Tought you could. | |
74 | */ | |
fba2591b LT |
75 | void cancel_dirty_page(struct page *page, unsigned int account_size) |
76 | { | |
8368e328 LT |
77 | if (TestClearPageDirty(page)) { |
78 | struct address_space *mapping = page->mapping; | |
79 | if (mapping && mapping_cap_account_dirty(mapping)) { | |
80 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
c9e51e41 PZ |
81 | dec_bdi_stat(mapping->backing_dev_info, |
82 | BDI_RECLAIMABLE); | |
8368e328 LT |
83 | if (account_size) |
84 | task_io_account_cancelled_write(account_size); | |
85 | } | |
3e67c098 | 86 | } |
fba2591b | 87 | } |
8368e328 | 88 | EXPORT_SYMBOL(cancel_dirty_page); |
fba2591b | 89 | |
1da177e4 LT |
90 | /* |
91 | * If truncate cannot remove the fs-private metadata from the page, the page | |
62e1c553 | 92 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
54cb8821 | 93 | * user pagetables if we're racing with filemap_fault(). |
1da177e4 LT |
94 | * |
95 | * We need to bale out if page->mapping is no longer equal to the original | |
96 | * mapping. This happens a) when the VM reclaimed the page while we waited on | |
fc0ecff6 | 97 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
1da177e4 LT |
98 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
99 | */ | |
750b4987 | 100 | static int |
1da177e4 LT |
101 | truncate_complete_page(struct address_space *mapping, struct page *page) |
102 | { | |
103 | if (page->mapping != mapping) | |
750b4987 | 104 | return -EIO; |
1da177e4 | 105 | |
266cf658 | 106 | if (page_has_private(page)) |
1da177e4 LT |
107 | do_invalidatepage(page, 0); |
108 | ||
a2b34564 BS |
109 | cancel_dirty_page(page, PAGE_CACHE_SIZE); |
110 | ||
1da177e4 | 111 | ClearPageMappedToDisk(page); |
5adc7b51 | 112 | delete_from_page_cache(page); |
750b4987 | 113 | return 0; |
1da177e4 LT |
114 | } |
115 | ||
116 | /* | |
fc0ecff6 | 117 | * This is for invalidate_mapping_pages(). That function can be called at |
1da177e4 | 118 | * any time, and is not supposed to throw away dirty pages. But pages can |
0fd0e6b0 NP |
119 | * be marked dirty at any time too, so use remove_mapping which safely |
120 | * discards clean, unused pages. | |
1da177e4 LT |
121 | * |
122 | * Returns non-zero if the page was successfully invalidated. | |
123 | */ | |
124 | static int | |
125 | invalidate_complete_page(struct address_space *mapping, struct page *page) | |
126 | { | |
0fd0e6b0 NP |
127 | int ret; |
128 | ||
1da177e4 LT |
129 | if (page->mapping != mapping) |
130 | return 0; | |
131 | ||
266cf658 | 132 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
1da177e4 LT |
133 | return 0; |
134 | ||
0fd0e6b0 | 135 | ret = remove_mapping(mapping, page); |
0fd0e6b0 NP |
136 | |
137 | return ret; | |
1da177e4 LT |
138 | } |
139 | ||
750b4987 NP |
140 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
141 | { | |
142 | if (page_mapped(page)) { | |
143 | unmap_mapping_range(mapping, | |
144 | (loff_t)page->index << PAGE_CACHE_SHIFT, | |
145 | PAGE_CACHE_SIZE, 0); | |
146 | } | |
147 | return truncate_complete_page(mapping, page); | |
148 | } | |
149 | ||
25718736 AK |
150 | /* |
151 | * Used to get rid of pages on hardware memory corruption. | |
152 | */ | |
153 | int generic_error_remove_page(struct address_space *mapping, struct page *page) | |
154 | { | |
155 | if (!mapping) | |
156 | return -EINVAL; | |
157 | /* | |
158 | * Only punch for normal data pages for now. | |
159 | * Handling other types like directories would need more auditing. | |
160 | */ | |
161 | if (!S_ISREG(mapping->host->i_mode)) | |
162 | return -EIO; | |
163 | return truncate_inode_page(mapping, page); | |
164 | } | |
165 | EXPORT_SYMBOL(generic_error_remove_page); | |
166 | ||
83f78668 WF |
167 | /* |
168 | * Safely invalidate one page from its pagecache mapping. | |
169 | * It only drops clean, unused pages. The page must be locked. | |
170 | * | |
171 | * Returns 1 if the page is successfully invalidated, otherwise 0. | |
172 | */ | |
173 | int invalidate_inode_page(struct page *page) | |
174 | { | |
175 | struct address_space *mapping = page_mapping(page); | |
176 | if (!mapping) | |
177 | return 0; | |
178 | if (PageDirty(page) || PageWriteback(page)) | |
179 | return 0; | |
180 | if (page_mapped(page)) | |
181 | return 0; | |
182 | return invalidate_complete_page(mapping, page); | |
183 | } | |
184 | ||
1da177e4 | 185 | /** |
73c1e204 | 186 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
1da177e4 LT |
187 | * @mapping: mapping to truncate |
188 | * @lstart: offset from which to truncate | |
d7339071 | 189 | * @lend: offset to which to truncate |
1da177e4 | 190 | * |
d7339071 HR |
191 | * Truncate the page cache, removing the pages that are between |
192 | * specified offsets (and zeroing out partial page | |
193 | * (if lstart is not page aligned)). | |
1da177e4 LT |
194 | * |
195 | * Truncate takes two passes - the first pass is nonblocking. It will not | |
196 | * block on page locks and it will not block on writeback. The second pass | |
197 | * will wait. This is to prevent as much IO as possible in the affected region. | |
198 | * The first pass will remove most pages, so the search cost of the second pass | |
199 | * is low. | |
200 | * | |
1da177e4 LT |
201 | * We pass down the cache-hot hint to the page freeing code. Even if the |
202 | * mapping is large, it is probably the case that the final pages are the most | |
203 | * recently touched, and freeing happens in ascending file offset order. | |
1da177e4 | 204 | */ |
d7339071 HR |
205 | void truncate_inode_pages_range(struct address_space *mapping, |
206 | loff_t lstart, loff_t lend) | |
1da177e4 LT |
207 | { |
208 | const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; | |
209 | const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1); | |
210 | struct pagevec pvec; | |
b85e0eff HD |
211 | pgoff_t index; |
212 | pgoff_t end; | |
1da177e4 LT |
213 | int i; |
214 | ||
3167760f | 215 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
216 | if (mapping->nrpages == 0) |
217 | return; | |
218 | ||
d7339071 HR |
219 | BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1)); |
220 | end = (lend >> PAGE_CACHE_SHIFT); | |
221 | ||
1da177e4 | 222 | pagevec_init(&pvec, 0); |
b85e0eff HD |
223 | index = start; |
224 | while (index <= end && pagevec_lookup(&pvec, mapping, index, | |
225 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
e5598f8b | 226 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
227 | for (i = 0; i < pagevec_count(&pvec); i++) { |
228 | struct page *page = pvec.pages[i]; | |
1da177e4 | 229 | |
b85e0eff HD |
230 | /* We rely upon deletion not changing page->index */ |
231 | index = page->index; | |
232 | if (index > end) | |
d7339071 | 233 | break; |
d7339071 | 234 | |
529ae9aa | 235 | if (!trylock_page(page)) |
1da177e4 | 236 | continue; |
b85e0eff | 237 | WARN_ON(page->index != index); |
1da177e4 LT |
238 | if (PageWriteback(page)) { |
239 | unlock_page(page); | |
240 | continue; | |
241 | } | |
750b4987 | 242 | truncate_inode_page(mapping, page); |
1da177e4 LT |
243 | unlock_page(page); |
244 | } | |
245 | pagevec_release(&pvec); | |
e5598f8b | 246 | mem_cgroup_uncharge_end(); |
1da177e4 | 247 | cond_resched(); |
b85e0eff | 248 | index++; |
1da177e4 LT |
249 | } |
250 | ||
251 | if (partial) { | |
252 | struct page *page = find_lock_page(mapping, start - 1); | |
253 | if (page) { | |
254 | wait_on_page_writeback(page); | |
255 | truncate_partial_page(page, partial); | |
256 | unlock_page(page); | |
257 | page_cache_release(page); | |
258 | } | |
259 | } | |
260 | ||
b85e0eff | 261 | index = start; |
1da177e4 LT |
262 | for ( ; ; ) { |
263 | cond_resched(); | |
b85e0eff HD |
264 | if (!pagevec_lookup(&pvec, mapping, index, |
265 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
266 | if (index == start) | |
1da177e4 | 267 | break; |
b85e0eff | 268 | index = start; |
1da177e4 LT |
269 | continue; |
270 | } | |
d0823576 | 271 | if (index == start && pvec.pages[0]->index > end) { |
d7339071 HR |
272 | pagevec_release(&pvec); |
273 | break; | |
274 | } | |
569b846d | 275 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
276 | for (i = 0; i < pagevec_count(&pvec); i++) { |
277 | struct page *page = pvec.pages[i]; | |
278 | ||
b85e0eff HD |
279 | /* We rely upon deletion not changing page->index */ |
280 | index = page->index; | |
281 | if (index > end) | |
d7339071 | 282 | break; |
b85e0eff | 283 | |
1da177e4 | 284 | lock_page(page); |
b85e0eff | 285 | WARN_ON(page->index != index); |
1da177e4 | 286 | wait_on_page_writeback(page); |
750b4987 | 287 | truncate_inode_page(mapping, page); |
1da177e4 LT |
288 | unlock_page(page); |
289 | } | |
290 | pagevec_release(&pvec); | |
569b846d | 291 | mem_cgroup_uncharge_end(); |
b85e0eff | 292 | index++; |
1da177e4 | 293 | } |
3167760f | 294 | cleancache_invalidate_inode(mapping); |
1da177e4 | 295 | } |
d7339071 | 296 | EXPORT_SYMBOL(truncate_inode_pages_range); |
1da177e4 | 297 | |
d7339071 HR |
298 | /** |
299 | * truncate_inode_pages - truncate *all* the pages from an offset | |
300 | * @mapping: mapping to truncate | |
301 | * @lstart: offset from which to truncate | |
302 | * | |
1b1dcc1b | 303 | * Called under (and serialised by) inode->i_mutex. |
08142579 JK |
304 | * |
305 | * Note: When this function returns, there can be a page in the process of | |
306 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus | |
307 | * mapping->nrpages can be non-zero when this function returns even after | |
308 | * truncation of the whole mapping. | |
d7339071 HR |
309 | */ |
310 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) | |
311 | { | |
312 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); | |
313 | } | |
1da177e4 LT |
314 | EXPORT_SYMBOL(truncate_inode_pages); |
315 | ||
28697355 MW |
316 | /** |
317 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode | |
318 | * @mapping: the address_space which holds the pages to invalidate | |
319 | * @start: the offset 'from' which to invalidate | |
320 | * @end: the offset 'to' which to invalidate (inclusive) | |
321 | * | |
322 | * This function only removes the unlocked pages, if you want to | |
323 | * remove all the pages of one inode, you must call truncate_inode_pages. | |
324 | * | |
325 | * invalidate_mapping_pages() will not block on IO activity. It will not | |
326 | * invalidate pages which are dirty, locked, under writeback or mapped into | |
327 | * pagetables. | |
328 | */ | |
329 | unsigned long invalidate_mapping_pages(struct address_space *mapping, | |
31560180 | 330 | pgoff_t start, pgoff_t end) |
1da177e4 LT |
331 | { |
332 | struct pagevec pvec; | |
b85e0eff | 333 | pgoff_t index = start; |
31560180 MK |
334 | unsigned long ret; |
335 | unsigned long count = 0; | |
1da177e4 LT |
336 | int i; |
337 | ||
31475dd6 HD |
338 | /* |
339 | * Note: this function may get called on a shmem/tmpfs mapping: | |
340 | * pagevec_lookup() might then return 0 prematurely (because it | |
341 | * got a gangful of swap entries); but it's hardly worth worrying | |
342 | * about - it can rarely have anything to free from such a mapping | |
343 | * (most pages are dirty), and already skips over any difficulties. | |
344 | */ | |
345 | ||
1da177e4 | 346 | pagevec_init(&pvec, 0); |
b85e0eff HD |
347 | while (index <= end && pagevec_lookup(&pvec, mapping, index, |
348 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
569b846d | 349 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
350 | for (i = 0; i < pagevec_count(&pvec); i++) { |
351 | struct page *page = pvec.pages[i]; | |
e0f23603 | 352 | |
b85e0eff | 353 | /* We rely upon deletion not changing page->index */ |
e0f23603 | 354 | index = page->index; |
b85e0eff HD |
355 | if (index > end) |
356 | break; | |
e0f23603 | 357 | |
b85e0eff HD |
358 | if (!trylock_page(page)) |
359 | continue; | |
360 | WARN_ON(page->index != index); | |
31560180 | 361 | ret = invalidate_inode_page(page); |
1da177e4 | 362 | unlock_page(page); |
31560180 MK |
363 | /* |
364 | * Invalidation is a hint that the page is no longer | |
365 | * of interest and try to speed up its reclaim. | |
366 | */ | |
367 | if (!ret) | |
368 | deactivate_page(page); | |
369 | count += ret; | |
1da177e4 LT |
370 | } |
371 | pagevec_release(&pvec); | |
569b846d | 372 | mem_cgroup_uncharge_end(); |
28697355 | 373 | cond_resched(); |
b85e0eff | 374 | index++; |
1da177e4 | 375 | } |
31560180 | 376 | return count; |
1da177e4 | 377 | } |
54bc4855 | 378 | EXPORT_SYMBOL(invalidate_mapping_pages); |
1da177e4 | 379 | |
bd4c8ce4 AM |
380 | /* |
381 | * This is like invalidate_complete_page(), except it ignores the page's | |
382 | * refcount. We do this because invalidate_inode_pages2() needs stronger | |
383 | * invalidation guarantees, and cannot afford to leave pages behind because | |
2706a1b8 AB |
384 | * shrink_page_list() has a temp ref on them, or because they're transiently |
385 | * sitting in the lru_cache_add() pagevecs. | |
bd4c8ce4 AM |
386 | */ |
387 | static int | |
388 | invalidate_complete_page2(struct address_space *mapping, struct page *page) | |
389 | { | |
390 | if (page->mapping != mapping) | |
391 | return 0; | |
392 | ||
266cf658 | 393 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
bd4c8ce4 AM |
394 | return 0; |
395 | ||
19fd6231 | 396 | spin_lock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
397 | if (PageDirty(page)) |
398 | goto failed; | |
399 | ||
266cf658 | 400 | BUG_ON(page_has_private(page)); |
e64a782f | 401 | __delete_from_page_cache(page); |
19fd6231 | 402 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 403 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
404 | |
405 | if (mapping->a_ops->freepage) | |
406 | mapping->a_ops->freepage(page); | |
407 | ||
bd4c8ce4 AM |
408 | page_cache_release(page); /* pagecache ref */ |
409 | return 1; | |
410 | failed: | |
19fd6231 | 411 | spin_unlock_irq(&mapping->tree_lock); |
bd4c8ce4 AM |
412 | return 0; |
413 | } | |
414 | ||
e3db7691 TM |
415 | static int do_launder_page(struct address_space *mapping, struct page *page) |
416 | { | |
417 | if (!PageDirty(page)) | |
418 | return 0; | |
419 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) | |
420 | return 0; | |
421 | return mapping->a_ops->launder_page(page); | |
422 | } | |
423 | ||
1da177e4 LT |
424 | /** |
425 | * invalidate_inode_pages2_range - remove range of pages from an address_space | |
67be2dd1 | 426 | * @mapping: the address_space |
1da177e4 LT |
427 | * @start: the page offset 'from' which to invalidate |
428 | * @end: the page offset 'to' which to invalidate (inclusive) | |
429 | * | |
430 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
431 | * invalidation. | |
432 | * | |
6ccfa806 | 433 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
434 | */ |
435 | int invalidate_inode_pages2_range(struct address_space *mapping, | |
436 | pgoff_t start, pgoff_t end) | |
437 | { | |
438 | struct pagevec pvec; | |
b85e0eff | 439 | pgoff_t index; |
1da177e4 LT |
440 | int i; |
441 | int ret = 0; | |
0dd1334f | 442 | int ret2 = 0; |
1da177e4 | 443 | int did_range_unmap = 0; |
1da177e4 | 444 | |
3167760f | 445 | cleancache_invalidate_inode(mapping); |
1da177e4 | 446 | pagevec_init(&pvec, 0); |
b85e0eff HD |
447 | index = start; |
448 | while (index <= end && pagevec_lookup(&pvec, mapping, index, | |
449 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { | |
569b846d | 450 | mem_cgroup_uncharge_start(); |
7b965e08 | 451 | for (i = 0; i < pagevec_count(&pvec); i++) { |
1da177e4 | 452 | struct page *page = pvec.pages[i]; |
b85e0eff HD |
453 | |
454 | /* We rely upon deletion not changing page->index */ | |
455 | index = page->index; | |
456 | if (index > end) | |
457 | break; | |
1da177e4 LT |
458 | |
459 | lock_page(page); | |
b85e0eff | 460 | WARN_ON(page->index != index); |
1da177e4 LT |
461 | if (page->mapping != mapping) { |
462 | unlock_page(page); | |
463 | continue; | |
464 | } | |
1da177e4 | 465 | wait_on_page_writeback(page); |
d00806b1 | 466 | if (page_mapped(page)) { |
1da177e4 LT |
467 | if (!did_range_unmap) { |
468 | /* | |
469 | * Zap the rest of the file in one hit. | |
470 | */ | |
471 | unmap_mapping_range(mapping, | |
b85e0eff HD |
472 | (loff_t)index << PAGE_CACHE_SHIFT, |
473 | (loff_t)(1 + end - index) | |
474 | << PAGE_CACHE_SHIFT, | |
1da177e4 LT |
475 | 0); |
476 | did_range_unmap = 1; | |
477 | } else { | |
478 | /* | |
479 | * Just zap this page | |
480 | */ | |
481 | unmap_mapping_range(mapping, | |
b85e0eff HD |
482 | (loff_t)index << PAGE_CACHE_SHIFT, |
483 | PAGE_CACHE_SIZE, 0); | |
1da177e4 LT |
484 | } |
485 | } | |
d00806b1 | 486 | BUG_ON(page_mapped(page)); |
0dd1334f HH |
487 | ret2 = do_launder_page(mapping, page); |
488 | if (ret2 == 0) { | |
489 | if (!invalidate_complete_page2(mapping, page)) | |
6ccfa806 | 490 | ret2 = -EBUSY; |
0dd1334f HH |
491 | } |
492 | if (ret2 < 0) | |
493 | ret = ret2; | |
1da177e4 LT |
494 | unlock_page(page); |
495 | } | |
496 | pagevec_release(&pvec); | |
569b846d | 497 | mem_cgroup_uncharge_end(); |
1da177e4 | 498 | cond_resched(); |
b85e0eff | 499 | index++; |
1da177e4 | 500 | } |
3167760f | 501 | cleancache_invalidate_inode(mapping); |
1da177e4 LT |
502 | return ret; |
503 | } | |
504 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); | |
505 | ||
506 | /** | |
507 | * invalidate_inode_pages2 - remove all pages from an address_space | |
67be2dd1 | 508 | * @mapping: the address_space |
1da177e4 LT |
509 | * |
510 | * Any pages which are found to be mapped into pagetables are unmapped prior to | |
511 | * invalidation. | |
512 | * | |
e9de25dd | 513 | * Returns -EBUSY if any pages could not be invalidated. |
1da177e4 LT |
514 | */ |
515 | int invalidate_inode_pages2(struct address_space *mapping) | |
516 | { | |
517 | return invalidate_inode_pages2_range(mapping, 0, -1); | |
518 | } | |
519 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); | |
25d9e2d1 | 520 | |
521 | /** | |
522 | * truncate_pagecache - unmap and remove pagecache that has been truncated | |
523 | * @inode: inode | |
8a549bea HD |
524 | * @oldsize: old file size |
525 | * @newsize: new file size | |
25d9e2d1 | 526 | * |
527 | * inode's new i_size must already be written before truncate_pagecache | |
528 | * is called. | |
529 | * | |
530 | * This function should typically be called before the filesystem | |
531 | * releases resources associated with the freed range (eg. deallocates | |
532 | * blocks). This way, pagecache will always stay logically coherent | |
533 | * with on-disk format, and the filesystem would not have to deal with | |
534 | * situations such as writepage being called for a page that has already | |
535 | * had its underlying blocks deallocated. | |
536 | */ | |
8a549bea | 537 | void truncate_pagecache(struct inode *inode, loff_t oldsize, loff_t newsize) |
25d9e2d1 | 538 | { |
cedabed4 | 539 | struct address_space *mapping = inode->i_mapping; |
8a549bea | 540 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
cedabed4 OH |
541 | |
542 | /* | |
543 | * unmap_mapping_range is called twice, first simply for | |
544 | * efficiency so that truncate_inode_pages does fewer | |
545 | * single-page unmaps. However after this first call, and | |
546 | * before truncate_inode_pages finishes, it is possible for | |
547 | * private pages to be COWed, which remain after | |
548 | * truncate_inode_pages finishes, hence the second | |
549 | * unmap_mapping_range call must be made for correctness. | |
550 | */ | |
8a549bea HD |
551 | unmap_mapping_range(mapping, holebegin, 0, 1); |
552 | truncate_inode_pages(mapping, newsize); | |
553 | unmap_mapping_range(mapping, holebegin, 0, 1); | |
25d9e2d1 | 554 | } |
555 | EXPORT_SYMBOL(truncate_pagecache); | |
556 | ||
2c27c65e CH |
557 | /** |
558 | * truncate_setsize - update inode and pagecache for a new file size | |
559 | * @inode: inode | |
560 | * @newsize: new file size | |
561 | * | |
382e27da JK |
562 | * truncate_setsize updates i_size and performs pagecache truncation (if |
563 | * necessary) to @newsize. It will be typically be called from the filesystem's | |
564 | * setattr function when ATTR_SIZE is passed in. | |
2c27c65e | 565 | * |
382e27da JK |
566 | * Must be called with inode_mutex held and before all filesystem specific |
567 | * block truncation has been performed. | |
2c27c65e CH |
568 | */ |
569 | void truncate_setsize(struct inode *inode, loff_t newsize) | |
570 | { | |
6cbdf115 | 571 | loff_t oldsize = inode->i_size; |
2c27c65e | 572 | |
2c27c65e | 573 | i_size_write(inode, newsize); |
6cbdf115 JK |
574 | if (newsize > oldsize) |
575 | pagecache_isize_extended(inode, oldsize, newsize); | |
2c27c65e CH |
576 | truncate_pagecache(inode, oldsize, newsize); |
577 | } | |
578 | EXPORT_SYMBOL(truncate_setsize); | |
579 | ||
6cbdf115 JK |
580 | /** |
581 | * pagecache_isize_extended - update pagecache after extension of i_size | |
582 | * @inode: inode for which i_size was extended | |
583 | * @from: original inode size | |
584 | * @to: new inode size | |
585 | * | |
586 | * Handle extension of inode size either caused by extending truncate or by | |
587 | * write starting after current i_size. We mark the page straddling current | |
588 | * i_size RO so that page_mkwrite() is called on the nearest write access to | |
589 | * the page. This way filesystem can be sure that page_mkwrite() is called on | |
590 | * the page before user writes to the page via mmap after the i_size has been | |
591 | * changed. | |
592 | * | |
593 | * The function must be called after i_size is updated so that page fault | |
594 | * coming after we unlock the page will already see the new i_size. | |
595 | * The function must be called while we still hold i_mutex - this not only | |
596 | * makes sure i_size is stable but also that userspace cannot observe new | |
597 | * i_size value before we are prepared to store mmap writes at new inode size. | |
598 | */ | |
599 | void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) | |
600 | { | |
601 | int bsize = 1 << inode->i_blkbits; | |
602 | loff_t rounded_from; | |
603 | struct page *page; | |
604 | pgoff_t index; | |
605 | ||
6cbdf115 JK |
606 | WARN_ON(to > inode->i_size); |
607 | ||
608 | if (from >= to || bsize == PAGE_CACHE_SIZE) | |
609 | return; | |
610 | /* Page straddling @from will not have any hole block created? */ | |
611 | rounded_from = round_up(from, bsize); | |
612 | if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1))) | |
613 | return; | |
614 | ||
615 | index = from >> PAGE_CACHE_SHIFT; | |
616 | page = find_lock_page(inode->i_mapping, index); | |
617 | /* Page not cached? Nothing to do */ | |
618 | if (!page) | |
619 | return; | |
620 | /* | |
621 | * See clear_page_dirty_for_io() for details why set_page_dirty() | |
622 | * is needed. | |
623 | */ | |
624 | if (page_mkclean(page)) | |
625 | set_page_dirty(page); | |
626 | unlock_page(page); | |
627 | page_cache_release(page); | |
628 | } | |
629 | EXPORT_SYMBOL(pagecache_isize_extended); | |
630 | ||
623e3db9 HD |
631 | /** |
632 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched | |
633 | * @inode: inode | |
634 | * @lstart: offset of beginning of hole | |
635 | * @lend: offset of last byte of hole | |
636 | * | |
637 | * This function should typically be called before the filesystem | |
638 | * releases resources associated with the freed range (eg. deallocates | |
639 | * blocks). This way, pagecache will always stay logically coherent | |
640 | * with on-disk format, and the filesystem would not have to deal with | |
641 | * situations such as writepage being called for a page that has already | |
642 | * had its underlying blocks deallocated. | |
643 | */ | |
644 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) | |
645 | { | |
646 | struct address_space *mapping = inode->i_mapping; | |
647 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); | |
648 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; | |
649 | /* | |
650 | * This rounding is currently just for example: unmap_mapping_range | |
651 | * expands its hole outwards, whereas we want it to contract the hole | |
652 | * inwards. However, existing callers of truncate_pagecache_range are | |
653 | * doing their own page rounding first; and truncate_inode_pages_range | |
654 | * currently BUGs if lend is not pagealigned-1 (it handles partial | |
655 | * page at start of hole, but not partial page at end of hole). Note | |
656 | * unmap_mapping_range allows holelen 0 for all, and we allow lend -1. | |
657 | */ | |
658 | ||
659 | /* | |
660 | * Unlike in truncate_pagecache, unmap_mapping_range is called only | |
661 | * once (before truncating pagecache), and without "even_cows" flag: | |
662 | * hole-punching should not remove private COWed pages from the hole. | |
663 | */ | |
664 | if ((u64)unmap_end > (u64)unmap_start) | |
665 | unmap_mapping_range(mapping, unmap_start, | |
666 | 1 + unmap_end - unmap_start, 0); | |
667 | truncate_inode_pages_range(mapping, lstart, lend); | |
668 | } | |
669 | EXPORT_SYMBOL(truncate_pagecache_range); |