]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/filemap.c | |
3 | * | |
4 | * Copyright (C) 1994-1999 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * This file handles the generic file mmap semantics used by | |
9 | * most "normal" filesystems (but you don't /have/ to use this: | |
10 | * the NFS filesystem used to do this differently, for example) | |
11 | */ | |
1da177e4 LT |
12 | #include <linux/module.h> |
13 | #include <linux/slab.h> | |
14 | #include <linux/compiler.h> | |
15 | #include <linux/fs.h> | |
c22ce143 | 16 | #include <linux/uaccess.h> |
1da177e4 | 17 | #include <linux/aio.h> |
c59ede7b | 18 | #include <linux/capability.h> |
1da177e4 LT |
19 | #include <linux/kernel_stat.h> |
20 | #include <linux/mm.h> | |
21 | #include <linux/swap.h> | |
22 | #include <linux/mman.h> | |
23 | #include <linux/pagemap.h> | |
24 | #include <linux/file.h> | |
25 | #include <linux/uio.h> | |
26 | #include <linux/hash.h> | |
27 | #include <linux/writeback.h> | |
53253383 | 28 | #include <linux/backing-dev.h> |
1da177e4 LT |
29 | #include <linux/pagevec.h> |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/security.h> | |
32 | #include <linux/syscalls.h> | |
44110fe3 | 33 | #include <linux/cpuset.h> |
2f718ffc | 34 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
8a9f3ccd | 35 | #include <linux/memcontrol.h> |
0f8053a5 NP |
36 | #include "internal.h" |
37 | ||
1da177e4 | 38 | /* |
1da177e4 LT |
39 | * FIXME: remove all knowledge of the buffer layer from the core VM |
40 | */ | |
41 | #include <linux/buffer_head.h> /* for generic_osync_inode */ | |
42 | ||
1da177e4 LT |
43 | #include <asm/mman.h> |
44 | ||
5ce7852c | 45 | |
1da177e4 LT |
46 | /* |
47 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
48 | * though. | |
49 | * | |
50 | * Shared mappings now work. 15.8.1995 Bruno. | |
51 | * | |
52 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
53 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
54 | * | |
55 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
56 | */ | |
57 | ||
58 | /* | |
59 | * Lock ordering: | |
60 | * | |
61 | * ->i_mmap_lock (vmtruncate) | |
62 | * ->private_lock (__free_pte->__set_page_dirty_buffers) | |
5d337b91 HD |
63 | * ->swap_lock (exclusive_swap_page, others) |
64 | * ->mapping->tree_lock | |
1da177e4 | 65 | * |
1b1dcc1b | 66 | * ->i_mutex |
1da177e4 LT |
67 | * ->i_mmap_lock (truncate->unmap_mapping_range) |
68 | * | |
69 | * ->mmap_sem | |
70 | * ->i_mmap_lock | |
b8072f09 | 71 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
72 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
73 | * | |
74 | * ->mmap_sem | |
75 | * ->lock_page (access_process_vm) | |
76 | * | |
82591e6e NP |
77 | * ->i_mutex (generic_file_buffered_write) |
78 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) | |
1da177e4 | 79 | * |
1b1dcc1b | 80 | * ->i_mutex |
1da177e4 LT |
81 | * ->i_alloc_sem (various) |
82 | * | |
83 | * ->inode_lock | |
84 | * ->sb_lock (fs/fs-writeback.c) | |
85 | * ->mapping->tree_lock (__sync_single_inode) | |
86 | * | |
87 | * ->i_mmap_lock | |
88 | * ->anon_vma.lock (vma_adjust) | |
89 | * | |
90 | * ->anon_vma.lock | |
b8072f09 | 91 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 92 | * |
b8072f09 | 93 | * ->page_table_lock or pte_lock |
5d337b91 | 94 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
95 | * ->private_lock (try_to_unmap_one) |
96 | * ->tree_lock (try_to_unmap_one) | |
97 | * ->zone.lru_lock (follow_page->mark_page_accessed) | |
053837fc | 98 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
1da177e4 LT |
99 | * ->private_lock (page_remove_rmap->set_page_dirty) |
100 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
101 | * ->inode_lock (page_remove_rmap->set_page_dirty) | |
102 | * ->inode_lock (zap_pte_range->set_page_dirty) | |
103 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) | |
104 | * | |
105 | * ->task->proc_lock | |
106 | * ->dcache_lock (proc_pid_lookup) | |
107 | */ | |
108 | ||
109 | /* | |
110 | * Remove a page from the page cache and free it. Caller has to make | |
111 | * sure the page is locked and that nobody else uses it - or that usage | |
19fd6231 | 112 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 LT |
113 | */ |
114 | void __remove_from_page_cache(struct page *page) | |
115 | { | |
116 | struct address_space *mapping = page->mapping; | |
117 | ||
69029cd5 | 118 | mem_cgroup_uncharge_cache_page(page); |
1da177e4 LT |
119 | radix_tree_delete(&mapping->page_tree, page->index); |
120 | page->mapping = NULL; | |
121 | mapping->nrpages--; | |
347ce434 | 122 | __dec_zone_page_state(page, NR_FILE_PAGES); |
45426812 | 123 | BUG_ON(page_mapped(page)); |
3a692790 LT |
124 | |
125 | /* | |
126 | * Some filesystems seem to re-dirty the page even after | |
127 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
128 | * | |
129 | * Fix it up by doing a final dirty accounting check after | |
130 | * having removed the page entirely. | |
131 | */ | |
132 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { | |
133 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
134 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
135 | } | |
1da177e4 LT |
136 | } |
137 | ||
138 | void remove_from_page_cache(struct page *page) | |
139 | { | |
140 | struct address_space *mapping = page->mapping; | |
141 | ||
cd7619d6 | 142 | BUG_ON(!PageLocked(page)); |
1da177e4 | 143 | |
19fd6231 | 144 | spin_lock_irq(&mapping->tree_lock); |
1da177e4 | 145 | __remove_from_page_cache(page); |
19fd6231 | 146 | spin_unlock_irq(&mapping->tree_lock); |
1da177e4 LT |
147 | } |
148 | ||
149 | static int sync_page(void *word) | |
150 | { | |
151 | struct address_space *mapping; | |
152 | struct page *page; | |
153 | ||
07808b74 | 154 | page = container_of((unsigned long *)word, struct page, flags); |
1da177e4 LT |
155 | |
156 | /* | |
dd1d5afc NYC |
157 | * page_mapping() is being called without PG_locked held. |
158 | * Some knowledge of the state and use of the page is used to | |
159 | * reduce the requirements down to a memory barrier. | |
160 | * The danger here is of a stale page_mapping() return value | |
161 | * indicating a struct address_space different from the one it's | |
162 | * associated with when it is associated with one. | |
163 | * After smp_mb(), it's either the correct page_mapping() for | |
164 | * the page, or an old page_mapping() and the page's own | |
165 | * page_mapping() has gone NULL. | |
166 | * The ->sync_page() address_space operation must tolerate | |
167 | * page_mapping() going NULL. By an amazing coincidence, | |
168 | * this comes about because none of the users of the page | |
169 | * in the ->sync_page() methods make essential use of the | |
170 | * page_mapping(), merely passing the page down to the backing | |
171 | * device's unplug functions when it's non-NULL, which in turn | |
4c21e2f2 | 172 | * ignore it for all cases but swap, where only page_private(page) is |
dd1d5afc NYC |
173 | * of interest. When page_mapping() does go NULL, the entire |
174 | * call stack gracefully ignores the page and returns. | |
175 | * -- wli | |
1da177e4 LT |
176 | */ |
177 | smp_mb(); | |
178 | mapping = page_mapping(page); | |
179 | if (mapping && mapping->a_ops && mapping->a_ops->sync_page) | |
180 | mapping->a_ops->sync_page(page); | |
181 | io_schedule(); | |
182 | return 0; | |
183 | } | |
184 | ||
2687a356 MW |
185 | static int sync_page_killable(void *word) |
186 | { | |
187 | sync_page(word); | |
188 | return fatal_signal_pending(current) ? -EINTR : 0; | |
189 | } | |
190 | ||
1da177e4 | 191 | /** |
485bb99b | 192 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
193 | * @mapping: address space structure to write |
194 | * @start: offset in bytes where the range starts | |
469eb4d0 | 195 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 196 | * @sync_mode: enable synchronous operation |
1da177e4 | 197 | * |
485bb99b RD |
198 | * Start writeback against all of a mapping's dirty pages that lie |
199 | * within the byte offsets <start, end> inclusive. | |
200 | * | |
1da177e4 | 201 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 202 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
203 | * these two operations is that if a dirty page/buffer is encountered, it must |
204 | * be waited upon, and not just skipped over. | |
205 | */ | |
ebcf28e1 AM |
206 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
207 | loff_t end, int sync_mode) | |
1da177e4 LT |
208 | { |
209 | int ret; | |
210 | struct writeback_control wbc = { | |
211 | .sync_mode = sync_mode, | |
212 | .nr_to_write = mapping->nrpages * 2, | |
111ebb6e OH |
213 | .range_start = start, |
214 | .range_end = end, | |
1da177e4 LT |
215 | }; |
216 | ||
217 | if (!mapping_cap_writeback_dirty(mapping)) | |
218 | return 0; | |
219 | ||
220 | ret = do_writepages(mapping, &wbc); | |
221 | return ret; | |
222 | } | |
223 | ||
224 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
225 | int sync_mode) | |
226 | { | |
111ebb6e | 227 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
228 | } |
229 | ||
230 | int filemap_fdatawrite(struct address_space *mapping) | |
231 | { | |
232 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
233 | } | |
234 | EXPORT_SYMBOL(filemap_fdatawrite); | |
235 | ||
f4c0a0fd | 236 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 237 | loff_t end) |
1da177e4 LT |
238 | { |
239 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
240 | } | |
f4c0a0fd | 241 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 242 | |
485bb99b RD |
243 | /** |
244 | * filemap_flush - mostly a non-blocking flush | |
245 | * @mapping: target address_space | |
246 | * | |
1da177e4 LT |
247 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
248 | * purposes - I/O may not be started against all dirty pages. | |
249 | */ | |
250 | int filemap_flush(struct address_space *mapping) | |
251 | { | |
252 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
253 | } | |
254 | EXPORT_SYMBOL(filemap_flush); | |
255 | ||
485bb99b RD |
256 | /** |
257 | * wait_on_page_writeback_range - wait for writeback to complete | |
258 | * @mapping: target address_space | |
259 | * @start: beginning page index | |
260 | * @end: ending page index | |
261 | * | |
1da177e4 LT |
262 | * Wait for writeback to complete against pages indexed by start->end |
263 | * inclusive | |
264 | */ | |
ebcf28e1 | 265 | int wait_on_page_writeback_range(struct address_space *mapping, |
1da177e4 LT |
266 | pgoff_t start, pgoff_t end) |
267 | { | |
268 | struct pagevec pvec; | |
269 | int nr_pages; | |
270 | int ret = 0; | |
271 | pgoff_t index; | |
272 | ||
273 | if (end < start) | |
274 | return 0; | |
275 | ||
276 | pagevec_init(&pvec, 0); | |
277 | index = start; | |
278 | while ((index <= end) && | |
279 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
280 | PAGECACHE_TAG_WRITEBACK, | |
281 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
282 | unsigned i; | |
283 | ||
284 | for (i = 0; i < nr_pages; i++) { | |
285 | struct page *page = pvec.pages[i]; | |
286 | ||
287 | /* until radix tree lookup accepts end_index */ | |
288 | if (page->index > end) | |
289 | continue; | |
290 | ||
291 | wait_on_page_writeback(page); | |
292 | if (PageError(page)) | |
293 | ret = -EIO; | |
294 | } | |
295 | pagevec_release(&pvec); | |
296 | cond_resched(); | |
297 | } | |
298 | ||
299 | /* Check for outstanding write errors */ | |
300 | if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
301 | ret = -ENOSPC; | |
302 | if (test_and_clear_bit(AS_EIO, &mapping->flags)) | |
303 | ret = -EIO; | |
304 | ||
305 | return ret; | |
306 | } | |
307 | ||
485bb99b RD |
308 | /** |
309 | * sync_page_range - write and wait on all pages in the passed range | |
310 | * @inode: target inode | |
311 | * @mapping: target address_space | |
312 | * @pos: beginning offset in pages to write | |
313 | * @count: number of bytes to write | |
314 | * | |
1da177e4 LT |
315 | * Write and wait upon all the pages in the passed range. This is a "data |
316 | * integrity" operation. It waits upon in-flight writeout before starting and | |
317 | * waiting upon new writeout. If there was an IO error, return it. | |
318 | * | |
1b1dcc1b | 319 | * We need to re-take i_mutex during the generic_osync_inode list walk because |
1da177e4 LT |
320 | * it is otherwise livelockable. |
321 | */ | |
322 | int sync_page_range(struct inode *inode, struct address_space *mapping, | |
268fc16e | 323 | loff_t pos, loff_t count) |
1da177e4 LT |
324 | { |
325 | pgoff_t start = pos >> PAGE_CACHE_SHIFT; | |
326 | pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT; | |
327 | int ret; | |
328 | ||
329 | if (!mapping_cap_writeback_dirty(mapping) || !count) | |
330 | return 0; | |
331 | ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1); | |
332 | if (ret == 0) { | |
1b1dcc1b | 333 | mutex_lock(&inode->i_mutex); |
1da177e4 | 334 | ret = generic_osync_inode(inode, mapping, OSYNC_METADATA); |
1b1dcc1b | 335 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
336 | } |
337 | if (ret == 0) | |
338 | ret = wait_on_page_writeback_range(mapping, start, end); | |
339 | return ret; | |
340 | } | |
341 | EXPORT_SYMBOL(sync_page_range); | |
342 | ||
485bb99b | 343 | /** |
7682486b | 344 | * sync_page_range_nolock - write & wait on all pages in the passed range without locking |
485bb99b RD |
345 | * @inode: target inode |
346 | * @mapping: target address_space | |
347 | * @pos: beginning offset in pages to write | |
348 | * @count: number of bytes to write | |
349 | * | |
72fd4a35 | 350 | * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea |
1da177e4 LT |
351 | * as it forces O_SYNC writers to different parts of the same file |
352 | * to be serialised right until io completion. | |
353 | */ | |
268fc16e OH |
354 | int sync_page_range_nolock(struct inode *inode, struct address_space *mapping, |
355 | loff_t pos, loff_t count) | |
1da177e4 LT |
356 | { |
357 | pgoff_t start = pos >> PAGE_CACHE_SHIFT; | |
358 | pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT; | |
359 | int ret; | |
360 | ||
361 | if (!mapping_cap_writeback_dirty(mapping) || !count) | |
362 | return 0; | |
363 | ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1); | |
364 | if (ret == 0) | |
365 | ret = generic_osync_inode(inode, mapping, OSYNC_METADATA); | |
366 | if (ret == 0) | |
367 | ret = wait_on_page_writeback_range(mapping, start, end); | |
368 | return ret; | |
369 | } | |
268fc16e | 370 | EXPORT_SYMBOL(sync_page_range_nolock); |
1da177e4 LT |
371 | |
372 | /** | |
485bb99b | 373 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 374 | * @mapping: address space structure to wait for |
485bb99b RD |
375 | * |
376 | * Walk the list of under-writeback pages of the given address space | |
377 | * and wait for all of them. | |
1da177e4 LT |
378 | */ |
379 | int filemap_fdatawait(struct address_space *mapping) | |
380 | { | |
381 | loff_t i_size = i_size_read(mapping->host); | |
382 | ||
383 | if (i_size == 0) | |
384 | return 0; | |
385 | ||
386 | return wait_on_page_writeback_range(mapping, 0, | |
387 | (i_size - 1) >> PAGE_CACHE_SHIFT); | |
388 | } | |
389 | EXPORT_SYMBOL(filemap_fdatawait); | |
390 | ||
391 | int filemap_write_and_wait(struct address_space *mapping) | |
392 | { | |
28fd1298 | 393 | int err = 0; |
1da177e4 LT |
394 | |
395 | if (mapping->nrpages) { | |
28fd1298 OH |
396 | err = filemap_fdatawrite(mapping); |
397 | /* | |
398 | * Even if the above returned error, the pages may be | |
399 | * written partially (e.g. -ENOSPC), so we wait for it. | |
400 | * But the -EIO is special case, it may indicate the worst | |
401 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
402 | */ | |
403 | if (err != -EIO) { | |
404 | int err2 = filemap_fdatawait(mapping); | |
405 | if (!err) | |
406 | err = err2; | |
407 | } | |
1da177e4 | 408 | } |
28fd1298 | 409 | return err; |
1da177e4 | 410 | } |
28fd1298 | 411 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 412 | |
485bb99b RD |
413 | /** |
414 | * filemap_write_and_wait_range - write out & wait on a file range | |
415 | * @mapping: the address_space for the pages | |
416 | * @lstart: offset in bytes where the range starts | |
417 | * @lend: offset in bytes where the range ends (inclusive) | |
418 | * | |
469eb4d0 AM |
419 | * Write out and wait upon file offsets lstart->lend, inclusive. |
420 | * | |
421 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
422 | * that this function can be used to write to the very end-of-file (end = -1). | |
423 | */ | |
1da177e4 LT |
424 | int filemap_write_and_wait_range(struct address_space *mapping, |
425 | loff_t lstart, loff_t lend) | |
426 | { | |
28fd1298 | 427 | int err = 0; |
1da177e4 LT |
428 | |
429 | if (mapping->nrpages) { | |
28fd1298 OH |
430 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
431 | WB_SYNC_ALL); | |
432 | /* See comment of filemap_write_and_wait() */ | |
433 | if (err != -EIO) { | |
434 | int err2 = wait_on_page_writeback_range(mapping, | |
435 | lstart >> PAGE_CACHE_SHIFT, | |
436 | lend >> PAGE_CACHE_SHIFT); | |
437 | if (!err) | |
438 | err = err2; | |
439 | } | |
1da177e4 | 440 | } |
28fd1298 | 441 | return err; |
1da177e4 LT |
442 | } |
443 | ||
485bb99b | 444 | /** |
e286781d | 445 | * add_to_page_cache_locked - add a locked page to the pagecache |
485bb99b RD |
446 | * @page: page to add |
447 | * @mapping: the page's address_space | |
448 | * @offset: page index | |
449 | * @gfp_mask: page allocation mode | |
450 | * | |
e286781d | 451 | * This function is used to add a page to the pagecache. It must be locked. |
1da177e4 LT |
452 | * This function does not add the page to the LRU. The caller must do that. |
453 | */ | |
e286781d | 454 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, |
6daa0e28 | 455 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 456 | { |
e286781d NP |
457 | int error; |
458 | ||
459 | VM_BUG_ON(!PageLocked(page)); | |
460 | ||
461 | error = mem_cgroup_cache_charge(page, current->mm, | |
4c6bc8dd | 462 | gfp_mask & ~__GFP_HIGHMEM); |
35c754d7 BS |
463 | if (error) |
464 | goto out; | |
1da177e4 | 465 | |
35c754d7 | 466 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
1da177e4 | 467 | if (error == 0) { |
e286781d NP |
468 | page_cache_get(page); |
469 | page->mapping = mapping; | |
470 | page->index = offset; | |
471 | ||
19fd6231 | 472 | spin_lock_irq(&mapping->tree_lock); |
1da177e4 | 473 | error = radix_tree_insert(&mapping->page_tree, offset, page); |
e286781d | 474 | if (likely(!error)) { |
1da177e4 | 475 | mapping->nrpages++; |
347ce434 | 476 | __inc_zone_page_state(page, NR_FILE_PAGES); |
e286781d NP |
477 | } else { |
478 | page->mapping = NULL; | |
69029cd5 | 479 | mem_cgroup_uncharge_cache_page(page); |
e286781d NP |
480 | page_cache_release(page); |
481 | } | |
8a9f3ccd | 482 | |
19fd6231 | 483 | spin_unlock_irq(&mapping->tree_lock); |
1da177e4 | 484 | radix_tree_preload_end(); |
35c754d7 | 485 | } else |
69029cd5 | 486 | mem_cgroup_uncharge_cache_page(page); |
8a9f3ccd | 487 | out: |
1da177e4 LT |
488 | return error; |
489 | } | |
e286781d | 490 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
491 | |
492 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 493 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 LT |
494 | { |
495 | int ret = add_to_page_cache(page, mapping, offset, gfp_mask); | |
496 | if (ret == 0) | |
497 | lru_cache_add(page); | |
498 | return ret; | |
499 | } | |
500 | ||
44110fe3 | 501 | #ifdef CONFIG_NUMA |
2ae88149 | 502 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 PJ |
503 | { |
504 | if (cpuset_do_page_mem_spread()) { | |
505 | int n = cpuset_mem_spread_node(); | |
2ae88149 | 506 | return alloc_pages_node(n, gfp, 0); |
44110fe3 | 507 | } |
2ae88149 | 508 | return alloc_pages(gfp, 0); |
44110fe3 | 509 | } |
2ae88149 | 510 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
511 | #endif |
512 | ||
db37648c NP |
513 | static int __sleep_on_page_lock(void *word) |
514 | { | |
515 | io_schedule(); | |
516 | return 0; | |
517 | } | |
518 | ||
1da177e4 LT |
519 | /* |
520 | * In order to wait for pages to become available there must be | |
521 | * waitqueues associated with pages. By using a hash table of | |
522 | * waitqueues where the bucket discipline is to maintain all | |
523 | * waiters on the same queue and wake all when any of the pages | |
524 | * become available, and for the woken contexts to check to be | |
525 | * sure the appropriate page became available, this saves space | |
526 | * at a cost of "thundering herd" phenomena during rare hash | |
527 | * collisions. | |
528 | */ | |
529 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
530 | { | |
531 | const struct zone *zone = page_zone(page); | |
532 | ||
533 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
534 | } | |
535 | ||
536 | static inline void wake_up_page(struct page *page, int bit) | |
537 | { | |
538 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); | |
539 | } | |
540 | ||
920c7a5d | 541 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
542 | { |
543 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
544 | ||
545 | if (test_bit(bit_nr, &page->flags)) | |
546 | __wait_on_bit(page_waitqueue(page), &wait, sync_page, | |
547 | TASK_UNINTERRUPTIBLE); | |
548 | } | |
549 | EXPORT_SYMBOL(wait_on_page_bit); | |
550 | ||
551 | /** | |
485bb99b | 552 | * unlock_page - unlock a locked page |
1da177e4 LT |
553 | * @page: the page |
554 | * | |
555 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
556 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
557 | * mechananism between PageLocked pages and PageWriteback pages is shared. | |
558 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. | |
559 | * | |
560 | * The first mb is necessary to safely close the critical section opened by the | |
529ae9aa NP |
561 | * test_and_set_bit() to lock the page; the second mb is necessary to enforce |
562 | * ordering between the clear_bit and the read of the waitqueue (to avoid SMP | |
563 | * races with a parallel wait_on_page_locked()). | |
1da177e4 | 564 | */ |
920c7a5d | 565 | void unlock_page(struct page *page) |
1da177e4 LT |
566 | { |
567 | smp_mb__before_clear_bit(); | |
529ae9aa | 568 | if (!test_and_clear_bit(PG_locked, &page->flags)) |
1da177e4 LT |
569 | BUG(); |
570 | smp_mb__after_clear_bit(); | |
571 | wake_up_page(page, PG_locked); | |
572 | } | |
573 | EXPORT_SYMBOL(unlock_page); | |
574 | ||
485bb99b RD |
575 | /** |
576 | * end_page_writeback - end writeback against a page | |
577 | * @page: the page | |
1da177e4 LT |
578 | */ |
579 | void end_page_writeback(struct page *page) | |
580 | { | |
ac6aadb2 MS |
581 | if (TestClearPageReclaim(page)) |
582 | rotate_reclaimable_page(page); | |
583 | ||
584 | if (!test_clear_page_writeback(page)) | |
585 | BUG(); | |
586 | ||
1da177e4 LT |
587 | smp_mb__after_clear_bit(); |
588 | wake_up_page(page, PG_writeback); | |
589 | } | |
590 | EXPORT_SYMBOL(end_page_writeback); | |
591 | ||
485bb99b RD |
592 | /** |
593 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
594 | * @page: the page to lock | |
1da177e4 | 595 | * |
485bb99b | 596 | * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some |
1da177e4 LT |
597 | * random driver's requestfn sets TASK_RUNNING, we could busywait. However |
598 | * chances are that on the second loop, the block layer's plug list is empty, | |
599 | * so sync_page() will then return in state TASK_UNINTERRUPTIBLE. | |
600 | */ | |
920c7a5d | 601 | void __lock_page(struct page *page) |
1da177e4 LT |
602 | { |
603 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
604 | ||
605 | __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page, | |
606 | TASK_UNINTERRUPTIBLE); | |
607 | } | |
608 | EXPORT_SYMBOL(__lock_page); | |
609 | ||
b5606c2d | 610 | int __lock_page_killable(struct page *page) |
2687a356 MW |
611 | { |
612 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
613 | ||
614 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
615 | sync_page_killable, TASK_KILLABLE); | |
616 | } | |
617 | ||
7682486b RD |
618 | /** |
619 | * __lock_page_nosync - get a lock on the page, without calling sync_page() | |
620 | * @page: the page to lock | |
621 | * | |
db37648c NP |
622 | * Variant of lock_page that does not require the caller to hold a reference |
623 | * on the page's mapping. | |
624 | */ | |
920c7a5d | 625 | void __lock_page_nosync(struct page *page) |
db37648c NP |
626 | { |
627 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
628 | __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock, | |
629 | TASK_UNINTERRUPTIBLE); | |
630 | } | |
631 | ||
485bb99b RD |
632 | /** |
633 | * find_get_page - find and get a page reference | |
634 | * @mapping: the address_space to search | |
635 | * @offset: the page index | |
636 | * | |
da6052f7 NP |
637 | * Is there a pagecache struct page at the given (mapping, offset) tuple? |
638 | * If yes, increment its refcount and return it; if no, return NULL. | |
1da177e4 | 639 | */ |
a60637c8 | 640 | struct page *find_get_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 641 | { |
a60637c8 | 642 | void **pagep; |
1da177e4 LT |
643 | struct page *page; |
644 | ||
a60637c8 NP |
645 | rcu_read_lock(); |
646 | repeat: | |
647 | page = NULL; | |
648 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
649 | if (pagep) { | |
650 | page = radix_tree_deref_slot(pagep); | |
651 | if (unlikely(!page || page == RADIX_TREE_RETRY)) | |
652 | goto repeat; | |
653 | ||
654 | if (!page_cache_get_speculative(page)) | |
655 | goto repeat; | |
656 | ||
657 | /* | |
658 | * Has the page moved? | |
659 | * This is part of the lockless pagecache protocol. See | |
660 | * include/linux/pagemap.h for details. | |
661 | */ | |
662 | if (unlikely(page != *pagep)) { | |
663 | page_cache_release(page); | |
664 | goto repeat; | |
665 | } | |
666 | } | |
667 | rcu_read_unlock(); | |
668 | ||
1da177e4 LT |
669 | return page; |
670 | } | |
1da177e4 LT |
671 | EXPORT_SYMBOL(find_get_page); |
672 | ||
1da177e4 LT |
673 | /** |
674 | * find_lock_page - locate, pin and lock a pagecache page | |
67be2dd1 MW |
675 | * @mapping: the address_space to search |
676 | * @offset: the page index | |
1da177e4 LT |
677 | * |
678 | * Locates the desired pagecache page, locks it, increments its reference | |
679 | * count and returns its address. | |
680 | * | |
681 | * Returns zero if the page was not present. find_lock_page() may sleep. | |
682 | */ | |
a60637c8 | 683 | struct page *find_lock_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 LT |
684 | { |
685 | struct page *page; | |
686 | ||
1da177e4 | 687 | repeat: |
a60637c8 | 688 | page = find_get_page(mapping, offset); |
1da177e4 | 689 | if (page) { |
a60637c8 NP |
690 | lock_page(page); |
691 | /* Has the page been truncated? */ | |
692 | if (unlikely(page->mapping != mapping)) { | |
693 | unlock_page(page); | |
694 | page_cache_release(page); | |
695 | goto repeat; | |
1da177e4 | 696 | } |
a60637c8 | 697 | VM_BUG_ON(page->index != offset); |
1da177e4 | 698 | } |
1da177e4 LT |
699 | return page; |
700 | } | |
1da177e4 LT |
701 | EXPORT_SYMBOL(find_lock_page); |
702 | ||
703 | /** | |
704 | * find_or_create_page - locate or add a pagecache page | |
67be2dd1 MW |
705 | * @mapping: the page's address_space |
706 | * @index: the page's index into the mapping | |
707 | * @gfp_mask: page allocation mode | |
1da177e4 LT |
708 | * |
709 | * Locates a page in the pagecache. If the page is not present, a new page | |
710 | * is allocated using @gfp_mask and is added to the pagecache and to the VM's | |
711 | * LRU list. The returned page is locked and has its reference count | |
712 | * incremented. | |
713 | * | |
714 | * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic | |
715 | * allocation! | |
716 | * | |
717 | * find_or_create_page() returns the desired page's address, or zero on | |
718 | * memory exhaustion. | |
719 | */ | |
720 | struct page *find_or_create_page(struct address_space *mapping, | |
57f6b96c | 721 | pgoff_t index, gfp_t gfp_mask) |
1da177e4 | 722 | { |
eb2be189 | 723 | struct page *page; |
1da177e4 LT |
724 | int err; |
725 | repeat: | |
726 | page = find_lock_page(mapping, index); | |
727 | if (!page) { | |
eb2be189 NP |
728 | page = __page_cache_alloc(gfp_mask); |
729 | if (!page) | |
730 | return NULL; | |
731 | err = add_to_page_cache_lru(page, mapping, index, gfp_mask); | |
732 | if (unlikely(err)) { | |
733 | page_cache_release(page); | |
734 | page = NULL; | |
735 | if (err == -EEXIST) | |
736 | goto repeat; | |
1da177e4 | 737 | } |
1da177e4 | 738 | } |
1da177e4 LT |
739 | return page; |
740 | } | |
1da177e4 LT |
741 | EXPORT_SYMBOL(find_or_create_page); |
742 | ||
743 | /** | |
744 | * find_get_pages - gang pagecache lookup | |
745 | * @mapping: The address_space to search | |
746 | * @start: The starting page index | |
747 | * @nr_pages: The maximum number of pages | |
748 | * @pages: Where the resulting pages are placed | |
749 | * | |
750 | * find_get_pages() will search for and return a group of up to | |
751 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
752 | * find_get_pages() takes a reference against the returned pages. | |
753 | * | |
754 | * The search returns a group of mapping-contiguous pages with ascending | |
755 | * indexes. There may be holes in the indices due to not-present pages. | |
756 | * | |
757 | * find_get_pages() returns the number of pages which were found. | |
758 | */ | |
759 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
760 | unsigned int nr_pages, struct page **pages) | |
761 | { | |
762 | unsigned int i; | |
763 | unsigned int ret; | |
a60637c8 NP |
764 | unsigned int nr_found; |
765 | ||
766 | rcu_read_lock(); | |
767 | restart: | |
768 | nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, | |
769 | (void ***)pages, start, nr_pages); | |
770 | ret = 0; | |
771 | for (i = 0; i < nr_found; i++) { | |
772 | struct page *page; | |
773 | repeat: | |
774 | page = radix_tree_deref_slot((void **)pages[i]); | |
775 | if (unlikely(!page)) | |
776 | continue; | |
777 | /* | |
778 | * this can only trigger if nr_found == 1, making livelock | |
779 | * a non issue. | |
780 | */ | |
781 | if (unlikely(page == RADIX_TREE_RETRY)) | |
782 | goto restart; | |
783 | ||
784 | if (!page_cache_get_speculative(page)) | |
785 | goto repeat; | |
786 | ||
787 | /* Has the page moved? */ | |
788 | if (unlikely(page != *((void **)pages[i]))) { | |
789 | page_cache_release(page); | |
790 | goto repeat; | |
791 | } | |
1da177e4 | 792 | |
a60637c8 NP |
793 | pages[ret] = page; |
794 | ret++; | |
795 | } | |
796 | rcu_read_unlock(); | |
1da177e4 LT |
797 | return ret; |
798 | } | |
799 | ||
ebf43500 JA |
800 | /** |
801 | * find_get_pages_contig - gang contiguous pagecache lookup | |
802 | * @mapping: The address_space to search | |
803 | * @index: The starting page index | |
804 | * @nr_pages: The maximum number of pages | |
805 | * @pages: Where the resulting pages are placed | |
806 | * | |
807 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
808 | * that the returned number of pages are guaranteed to be contiguous. | |
809 | * | |
810 | * find_get_pages_contig() returns the number of pages which were found. | |
811 | */ | |
812 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
813 | unsigned int nr_pages, struct page **pages) | |
814 | { | |
815 | unsigned int i; | |
816 | unsigned int ret; | |
a60637c8 NP |
817 | unsigned int nr_found; |
818 | ||
819 | rcu_read_lock(); | |
820 | restart: | |
821 | nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, | |
822 | (void ***)pages, index, nr_pages); | |
823 | ret = 0; | |
824 | for (i = 0; i < nr_found; i++) { | |
825 | struct page *page; | |
826 | repeat: | |
827 | page = radix_tree_deref_slot((void **)pages[i]); | |
828 | if (unlikely(!page)) | |
829 | continue; | |
830 | /* | |
831 | * this can only trigger if nr_found == 1, making livelock | |
832 | * a non issue. | |
833 | */ | |
834 | if (unlikely(page == RADIX_TREE_RETRY)) | |
835 | goto restart; | |
ebf43500 | 836 | |
a60637c8 | 837 | if (page->mapping == NULL || page->index != index) |
ebf43500 JA |
838 | break; |
839 | ||
a60637c8 NP |
840 | if (!page_cache_get_speculative(page)) |
841 | goto repeat; | |
842 | ||
843 | /* Has the page moved? */ | |
844 | if (unlikely(page != *((void **)pages[i]))) { | |
845 | page_cache_release(page); | |
846 | goto repeat; | |
847 | } | |
848 | ||
849 | pages[ret] = page; | |
850 | ret++; | |
ebf43500 JA |
851 | index++; |
852 | } | |
a60637c8 NP |
853 | rcu_read_unlock(); |
854 | return ret; | |
ebf43500 | 855 | } |
ef71c15c | 856 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 857 | |
485bb99b RD |
858 | /** |
859 | * find_get_pages_tag - find and return pages that match @tag | |
860 | * @mapping: the address_space to search | |
861 | * @index: the starting page index | |
862 | * @tag: the tag index | |
863 | * @nr_pages: the maximum number of pages | |
864 | * @pages: where the resulting pages are placed | |
865 | * | |
1da177e4 | 866 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 867 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
868 | */ |
869 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
870 | int tag, unsigned int nr_pages, struct page **pages) | |
871 | { | |
872 | unsigned int i; | |
873 | unsigned int ret; | |
a60637c8 NP |
874 | unsigned int nr_found; |
875 | ||
876 | rcu_read_lock(); | |
877 | restart: | |
878 | nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree, | |
879 | (void ***)pages, *index, nr_pages, tag); | |
880 | ret = 0; | |
881 | for (i = 0; i < nr_found; i++) { | |
882 | struct page *page; | |
883 | repeat: | |
884 | page = radix_tree_deref_slot((void **)pages[i]); | |
885 | if (unlikely(!page)) | |
886 | continue; | |
887 | /* | |
888 | * this can only trigger if nr_found == 1, making livelock | |
889 | * a non issue. | |
890 | */ | |
891 | if (unlikely(page == RADIX_TREE_RETRY)) | |
892 | goto restart; | |
893 | ||
894 | if (!page_cache_get_speculative(page)) | |
895 | goto repeat; | |
896 | ||
897 | /* Has the page moved? */ | |
898 | if (unlikely(page != *((void **)pages[i]))) { | |
899 | page_cache_release(page); | |
900 | goto repeat; | |
901 | } | |
902 | ||
903 | pages[ret] = page; | |
904 | ret++; | |
905 | } | |
906 | rcu_read_unlock(); | |
1da177e4 | 907 | |
1da177e4 LT |
908 | if (ret) |
909 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 910 | |
1da177e4 LT |
911 | return ret; |
912 | } | |
ef71c15c | 913 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 914 | |
485bb99b RD |
915 | /** |
916 | * grab_cache_page_nowait - returns locked page at given index in given cache | |
917 | * @mapping: target address_space | |
918 | * @index: the page index | |
919 | * | |
72fd4a35 | 920 | * Same as grab_cache_page(), but do not wait if the page is unavailable. |
1da177e4 LT |
921 | * This is intended for speculative data generators, where the data can |
922 | * be regenerated if the page couldn't be grabbed. This routine should | |
923 | * be safe to call while holding the lock for another page. | |
924 | * | |
925 | * Clear __GFP_FS when allocating the page to avoid recursion into the fs | |
926 | * and deadlock against the caller's locked page. | |
927 | */ | |
928 | struct page * | |
57f6b96c | 929 | grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) |
1da177e4 LT |
930 | { |
931 | struct page *page = find_get_page(mapping, index); | |
1da177e4 LT |
932 | |
933 | if (page) { | |
529ae9aa | 934 | if (trylock_page(page)) |
1da177e4 LT |
935 | return page; |
936 | page_cache_release(page); | |
937 | return NULL; | |
938 | } | |
2ae88149 NP |
939 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); |
940 | if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) { | |
1da177e4 LT |
941 | page_cache_release(page); |
942 | page = NULL; | |
943 | } | |
944 | return page; | |
945 | } | |
1da177e4 LT |
946 | EXPORT_SYMBOL(grab_cache_page_nowait); |
947 | ||
76d42bd9 WF |
948 | /* |
949 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
950 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
951 | * | |
952 | * ---R__________________________________________B__________ | |
953 | * ^ reading here ^ bad block(assume 4k) | |
954 | * | |
955 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
956 | * => failing the whole request => read(R) => read(R+1) => | |
957 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
958 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
959 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
960 | * | |
961 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
962 | */ | |
963 | static void shrink_readahead_size_eio(struct file *filp, | |
964 | struct file_ra_state *ra) | |
965 | { | |
966 | if (!ra->ra_pages) | |
967 | return; | |
968 | ||
969 | ra->ra_pages /= 4; | |
76d42bd9 WF |
970 | } |
971 | ||
485bb99b | 972 | /** |
36e78914 | 973 | * do_generic_file_read - generic file read routine |
485bb99b RD |
974 | * @filp: the file to read |
975 | * @ppos: current file position | |
976 | * @desc: read_descriptor | |
977 | * @actor: read method | |
978 | * | |
1da177e4 | 979 | * This is a generic file read routine, and uses the |
485bb99b | 980 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
981 | * |
982 | * This is really ugly. But the goto's actually try to clarify some | |
983 | * of the logic when it comes to error handling etc. | |
1da177e4 | 984 | */ |
36e78914 CH |
985 | static void do_generic_file_read(struct file *filp, loff_t *ppos, |
986 | read_descriptor_t *desc, read_actor_t actor) | |
1da177e4 | 987 | { |
36e78914 | 988 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 989 | struct inode *inode = mapping->host; |
36e78914 | 990 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
991 | pgoff_t index; |
992 | pgoff_t last_index; | |
993 | pgoff_t prev_index; | |
994 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 995 | unsigned int prev_offset; |
1da177e4 | 996 | int error; |
1da177e4 | 997 | |
1da177e4 | 998 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
999 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1000 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
1da177e4 LT |
1001 | last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1002 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1003 | ||
1da177e4 LT |
1004 | for (;;) { |
1005 | struct page *page; | |
57f6b96c | 1006 | pgoff_t end_index; |
a32ea1e1 | 1007 | loff_t isize; |
1da177e4 LT |
1008 | unsigned long nr, ret; |
1009 | ||
1da177e4 | 1010 | cond_resched(); |
1da177e4 LT |
1011 | find_page: |
1012 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1013 | if (!page) { |
cf914a7d | 1014 | page_cache_sync_readahead(mapping, |
7ff81078 | 1015 | ra, filp, |
3ea89ee8 FW |
1016 | index, last_index - index); |
1017 | page = find_get_page(mapping, index); | |
1018 | if (unlikely(page == NULL)) | |
1019 | goto no_cached_page; | |
1020 | } | |
1021 | if (PageReadahead(page)) { | |
cf914a7d | 1022 | page_cache_async_readahead(mapping, |
7ff81078 | 1023 | ra, filp, page, |
3ea89ee8 | 1024 | index, last_index - index); |
1da177e4 | 1025 | } |
8ab22b9a HH |
1026 | if (!PageUptodate(page)) { |
1027 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1028 | !mapping->a_ops->is_partially_uptodate) | |
1029 | goto page_not_up_to_date; | |
529ae9aa | 1030 | if (!trylock_page(page)) |
8ab22b9a HH |
1031 | goto page_not_up_to_date; |
1032 | if (!mapping->a_ops->is_partially_uptodate(page, | |
1033 | desc, offset)) | |
1034 | goto page_not_up_to_date_locked; | |
1035 | unlock_page(page); | |
1036 | } | |
1da177e4 | 1037 | page_ok: |
a32ea1e1 N |
1038 | /* |
1039 | * i_size must be checked after we know the page is Uptodate. | |
1040 | * | |
1041 | * Checking i_size after the check allows us to calculate | |
1042 | * the correct value for "nr", which means the zero-filled | |
1043 | * part of the page is not copied back to userspace (unless | |
1044 | * another truncate extends the file - this is desired though). | |
1045 | */ | |
1046 | ||
1047 | isize = i_size_read(inode); | |
1048 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1049 | if (unlikely(!isize || index > end_index)) { | |
1050 | page_cache_release(page); | |
1051 | goto out; | |
1052 | } | |
1053 | ||
1054 | /* nr is the maximum number of bytes to copy from this page */ | |
1055 | nr = PAGE_CACHE_SIZE; | |
1056 | if (index == end_index) { | |
1057 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1058 | if (nr <= offset) { | |
1059 | page_cache_release(page); | |
1060 | goto out; | |
1061 | } | |
1062 | } | |
1063 | nr = nr - offset; | |
1da177e4 LT |
1064 | |
1065 | /* If users can be writing to this page using arbitrary | |
1066 | * virtual addresses, take care about potential aliasing | |
1067 | * before reading the page on the kernel side. | |
1068 | */ | |
1069 | if (mapping_writably_mapped(mapping)) | |
1070 | flush_dcache_page(page); | |
1071 | ||
1072 | /* | |
ec0f1637 JK |
1073 | * When a sequential read accesses a page several times, |
1074 | * only mark it as accessed the first time. | |
1da177e4 | 1075 | */ |
ec0f1637 | 1076 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1077 | mark_page_accessed(page); |
1078 | prev_index = index; | |
1079 | ||
1080 | /* | |
1081 | * Ok, we have the page, and it's up-to-date, so | |
1082 | * now we can copy it to user space... | |
1083 | * | |
1084 | * The actor routine returns how many bytes were actually used.. | |
1085 | * NOTE! This may not be the same as how much of a user buffer | |
1086 | * we filled up (we may be padding etc), so we can only update | |
1087 | * "pos" here (the actor routine has to update the user buffer | |
1088 | * pointers and the remaining count). | |
1089 | */ | |
1090 | ret = actor(desc, page, offset, nr); | |
1091 | offset += ret; | |
1092 | index += offset >> PAGE_CACHE_SHIFT; | |
1093 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1094 | prev_offset = offset; |
1da177e4 LT |
1095 | |
1096 | page_cache_release(page); | |
1097 | if (ret == nr && desc->count) | |
1098 | continue; | |
1099 | goto out; | |
1100 | ||
1101 | page_not_up_to_date: | |
1102 | /* Get exclusive access to the page ... */ | |
0b94e97a MW |
1103 | if (lock_page_killable(page)) |
1104 | goto readpage_eio; | |
1da177e4 | 1105 | |
8ab22b9a | 1106 | page_not_up_to_date_locked: |
da6052f7 | 1107 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1108 | if (!page->mapping) { |
1109 | unlock_page(page); | |
1110 | page_cache_release(page); | |
1111 | continue; | |
1112 | } | |
1113 | ||
1114 | /* Did somebody else fill it already? */ | |
1115 | if (PageUptodate(page)) { | |
1116 | unlock_page(page); | |
1117 | goto page_ok; | |
1118 | } | |
1119 | ||
1120 | readpage: | |
1121 | /* Start the actual read. The read will unlock the page. */ | |
1122 | error = mapping->a_ops->readpage(filp, page); | |
1123 | ||
994fc28c ZB |
1124 | if (unlikely(error)) { |
1125 | if (error == AOP_TRUNCATED_PAGE) { | |
1126 | page_cache_release(page); | |
1127 | goto find_page; | |
1128 | } | |
1da177e4 | 1129 | goto readpage_error; |
994fc28c | 1130 | } |
1da177e4 LT |
1131 | |
1132 | if (!PageUptodate(page)) { | |
0b94e97a MW |
1133 | if (lock_page_killable(page)) |
1134 | goto readpage_eio; | |
1da177e4 LT |
1135 | if (!PageUptodate(page)) { |
1136 | if (page->mapping == NULL) { | |
1137 | /* | |
1138 | * invalidate_inode_pages got it | |
1139 | */ | |
1140 | unlock_page(page); | |
1141 | page_cache_release(page); | |
1142 | goto find_page; | |
1143 | } | |
1144 | unlock_page(page); | |
7ff81078 | 1145 | shrink_readahead_size_eio(filp, ra); |
0b94e97a | 1146 | goto readpage_eio; |
1da177e4 LT |
1147 | } |
1148 | unlock_page(page); | |
1149 | } | |
1150 | ||
1da177e4 LT |
1151 | goto page_ok; |
1152 | ||
0b94e97a MW |
1153 | readpage_eio: |
1154 | error = -EIO; | |
1da177e4 LT |
1155 | readpage_error: |
1156 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1157 | desc->error = error; | |
1158 | page_cache_release(page); | |
1159 | goto out; | |
1160 | ||
1161 | no_cached_page: | |
1162 | /* | |
1163 | * Ok, it wasn't cached, so we need to create a new | |
1164 | * page.. | |
1165 | */ | |
eb2be189 NP |
1166 | page = page_cache_alloc_cold(mapping); |
1167 | if (!page) { | |
1168 | desc->error = -ENOMEM; | |
1169 | goto out; | |
1da177e4 | 1170 | } |
eb2be189 | 1171 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1172 | index, GFP_KERNEL); |
1173 | if (error) { | |
eb2be189 | 1174 | page_cache_release(page); |
1da177e4 LT |
1175 | if (error == -EEXIST) |
1176 | goto find_page; | |
1177 | desc->error = error; | |
1178 | goto out; | |
1179 | } | |
1da177e4 LT |
1180 | goto readpage; |
1181 | } | |
1182 | ||
1183 | out: | |
7ff81078 FW |
1184 | ra->prev_pos = prev_index; |
1185 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1186 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1187 | |
f4e6b498 | 1188 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
1da177e4 LT |
1189 | if (filp) |
1190 | file_accessed(filp); | |
1191 | } | |
1da177e4 LT |
1192 | |
1193 | int file_read_actor(read_descriptor_t *desc, struct page *page, | |
1194 | unsigned long offset, unsigned long size) | |
1195 | { | |
1196 | char *kaddr; | |
1197 | unsigned long left, count = desc->count; | |
1198 | ||
1199 | if (size > count) | |
1200 | size = count; | |
1201 | ||
1202 | /* | |
1203 | * Faults on the destination of a read are common, so do it before | |
1204 | * taking the kmap. | |
1205 | */ | |
1206 | if (!fault_in_pages_writeable(desc->arg.buf, size)) { | |
1207 | kaddr = kmap_atomic(page, KM_USER0); | |
1208 | left = __copy_to_user_inatomic(desc->arg.buf, | |
1209 | kaddr + offset, size); | |
1210 | kunmap_atomic(kaddr, KM_USER0); | |
1211 | if (left == 0) | |
1212 | goto success; | |
1213 | } | |
1214 | ||
1215 | /* Do it the slow way */ | |
1216 | kaddr = kmap(page); | |
1217 | left = __copy_to_user(desc->arg.buf, kaddr + offset, size); | |
1218 | kunmap(page); | |
1219 | ||
1220 | if (left) { | |
1221 | size -= left; | |
1222 | desc->error = -EFAULT; | |
1223 | } | |
1224 | success: | |
1225 | desc->count = count - size; | |
1226 | desc->written += size; | |
1227 | desc->arg.buf += size; | |
1228 | return size; | |
1229 | } | |
1230 | ||
0ceb3314 DM |
1231 | /* |
1232 | * Performs necessary checks before doing a write | |
1233 | * @iov: io vector request | |
1234 | * @nr_segs: number of segments in the iovec | |
1235 | * @count: number of bytes to write | |
1236 | * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE | |
1237 | * | |
1238 | * Adjust number of segments and amount of bytes to write (nr_segs should be | |
1239 | * properly initialized first). Returns appropriate error code that caller | |
1240 | * should return or zero in case that write should be allowed. | |
1241 | */ | |
1242 | int generic_segment_checks(const struct iovec *iov, | |
1243 | unsigned long *nr_segs, size_t *count, int access_flags) | |
1244 | { | |
1245 | unsigned long seg; | |
1246 | size_t cnt = 0; | |
1247 | for (seg = 0; seg < *nr_segs; seg++) { | |
1248 | const struct iovec *iv = &iov[seg]; | |
1249 | ||
1250 | /* | |
1251 | * If any segment has a negative length, or the cumulative | |
1252 | * length ever wraps negative then return -EINVAL. | |
1253 | */ | |
1254 | cnt += iv->iov_len; | |
1255 | if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) | |
1256 | return -EINVAL; | |
1257 | if (access_ok(access_flags, iv->iov_base, iv->iov_len)) | |
1258 | continue; | |
1259 | if (seg == 0) | |
1260 | return -EFAULT; | |
1261 | *nr_segs = seg; | |
1262 | cnt -= iv->iov_len; /* This segment is no good */ | |
1263 | break; | |
1264 | } | |
1265 | *count = cnt; | |
1266 | return 0; | |
1267 | } | |
1268 | EXPORT_SYMBOL(generic_segment_checks); | |
1269 | ||
485bb99b | 1270 | /** |
b2abacf3 | 1271 | * generic_file_aio_read - generic filesystem read routine |
485bb99b RD |
1272 | * @iocb: kernel I/O control block |
1273 | * @iov: io vector request | |
1274 | * @nr_segs: number of segments in the iovec | |
b2abacf3 | 1275 | * @pos: current file position |
485bb99b | 1276 | * |
1da177e4 LT |
1277 | * This is the "read()" routine for all filesystems |
1278 | * that can use the page cache directly. | |
1279 | */ | |
1280 | ssize_t | |
543ade1f BP |
1281 | generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
1282 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
1283 | { |
1284 | struct file *filp = iocb->ki_filp; | |
1285 | ssize_t retval; | |
1286 | unsigned long seg; | |
1287 | size_t count; | |
543ade1f | 1288 | loff_t *ppos = &iocb->ki_pos; |
1da177e4 LT |
1289 | |
1290 | count = 0; | |
0ceb3314 DM |
1291 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); |
1292 | if (retval) | |
1293 | return retval; | |
1da177e4 LT |
1294 | |
1295 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
1296 | if (filp->f_flags & O_DIRECT) { | |
543ade1f | 1297 | loff_t size; |
1da177e4 LT |
1298 | struct address_space *mapping; |
1299 | struct inode *inode; | |
1300 | ||
1301 | mapping = filp->f_mapping; | |
1302 | inode = mapping->host; | |
1da177e4 LT |
1303 | if (!count) |
1304 | goto out; /* skip atime */ | |
1305 | size = i_size_read(inode); | |
1306 | if (pos < size) { | |
a969e903 CH |
1307 | retval = filemap_write_and_wait(mapping); |
1308 | if (!retval) { | |
1309 | retval = mapping->a_ops->direct_IO(READ, iocb, | |
1310 | iov, pos, nr_segs); | |
1311 | } | |
1da177e4 LT |
1312 | if (retval > 0) |
1313 | *ppos = pos + retval; | |
11fa977e HD |
1314 | if (retval) { |
1315 | file_accessed(filp); | |
1316 | goto out; | |
1317 | } | |
0e0bcae3 | 1318 | } |
1da177e4 LT |
1319 | } |
1320 | ||
11fa977e HD |
1321 | for (seg = 0; seg < nr_segs; seg++) { |
1322 | read_descriptor_t desc; | |
1da177e4 | 1323 | |
11fa977e HD |
1324 | desc.written = 0; |
1325 | desc.arg.buf = iov[seg].iov_base; | |
1326 | desc.count = iov[seg].iov_len; | |
1327 | if (desc.count == 0) | |
1328 | continue; | |
1329 | desc.error = 0; | |
1330 | do_generic_file_read(filp, ppos, &desc, file_read_actor); | |
1331 | retval += desc.written; | |
1332 | if (desc.error) { | |
1333 | retval = retval ?: desc.error; | |
1334 | break; | |
1da177e4 | 1335 | } |
11fa977e HD |
1336 | if (desc.count > 0) |
1337 | break; | |
1da177e4 LT |
1338 | } |
1339 | out: | |
1340 | return retval; | |
1341 | } | |
1da177e4 LT |
1342 | EXPORT_SYMBOL(generic_file_aio_read); |
1343 | ||
1da177e4 LT |
1344 | static ssize_t |
1345 | do_readahead(struct address_space *mapping, struct file *filp, | |
57f6b96c | 1346 | pgoff_t index, unsigned long nr) |
1da177e4 LT |
1347 | { |
1348 | if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage) | |
1349 | return -EINVAL; | |
1350 | ||
1351 | force_page_cache_readahead(mapping, filp, index, | |
1352 | max_sane_readahead(nr)); | |
1353 | return 0; | |
1354 | } | |
1355 | ||
1356 | asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count) | |
1357 | { | |
1358 | ssize_t ret; | |
1359 | struct file *file; | |
1360 | ||
1361 | ret = -EBADF; | |
1362 | file = fget(fd); | |
1363 | if (file) { | |
1364 | if (file->f_mode & FMODE_READ) { | |
1365 | struct address_space *mapping = file->f_mapping; | |
57f6b96c FW |
1366 | pgoff_t start = offset >> PAGE_CACHE_SHIFT; |
1367 | pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
1368 | unsigned long len = end - start + 1; |
1369 | ret = do_readahead(mapping, file, start, len); | |
1370 | } | |
1371 | fput(file); | |
1372 | } | |
1373 | return ret; | |
1374 | } | |
1375 | ||
1376 | #ifdef CONFIG_MMU | |
485bb99b RD |
1377 | /** |
1378 | * page_cache_read - adds requested page to the page cache if not already there | |
1379 | * @file: file to read | |
1380 | * @offset: page index | |
1381 | * | |
1da177e4 LT |
1382 | * This adds the requested page to the page cache if it isn't already there, |
1383 | * and schedules an I/O to read in its contents from disk. | |
1384 | */ | |
920c7a5d | 1385 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1386 | { |
1387 | struct address_space *mapping = file->f_mapping; | |
1388 | struct page *page; | |
994fc28c | 1389 | int ret; |
1da177e4 | 1390 | |
994fc28c ZB |
1391 | do { |
1392 | page = page_cache_alloc_cold(mapping); | |
1393 | if (!page) | |
1394 | return -ENOMEM; | |
1395 | ||
1396 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); | |
1397 | if (ret == 0) | |
1398 | ret = mapping->a_ops->readpage(file, page); | |
1399 | else if (ret == -EEXIST) | |
1400 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1401 | |
1da177e4 | 1402 | page_cache_release(page); |
1da177e4 | 1403 | |
994fc28c ZB |
1404 | } while (ret == AOP_TRUNCATED_PAGE); |
1405 | ||
1406 | return ret; | |
1da177e4 LT |
1407 | } |
1408 | ||
1409 | #define MMAP_LOTSAMISS (100) | |
1410 | ||
485bb99b | 1411 | /** |
54cb8821 | 1412 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1413 | * @vma: vma in which the fault was taken |
1414 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1415 | * |
54cb8821 | 1416 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1417 | * mapped memory region to read in file data during a page fault. |
1418 | * | |
1419 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1420 | * it in the page cache, and handles the special cases reasonably without | |
1421 | * having a lot of duplicated code. | |
1422 | */ | |
d0217ac0 | 1423 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1424 | { |
1425 | int error; | |
54cb8821 | 1426 | struct file *file = vma->vm_file; |
1da177e4 LT |
1427 | struct address_space *mapping = file->f_mapping; |
1428 | struct file_ra_state *ra = &file->f_ra; | |
1429 | struct inode *inode = mapping->host; | |
1430 | struct page *page; | |
2004dc8e | 1431 | pgoff_t size; |
54cb8821 | 1432 | int did_readaround = 0; |
83c54070 | 1433 | int ret = 0; |
1da177e4 | 1434 | |
1da177e4 | 1435 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
d0217ac0 | 1436 | if (vmf->pgoff >= size) |
5307cc1a | 1437 | return VM_FAULT_SIGBUS; |
1da177e4 LT |
1438 | |
1439 | /* If we don't want any read-ahead, don't bother */ | |
54cb8821 | 1440 | if (VM_RandomReadHint(vma)) |
1da177e4 LT |
1441 | goto no_cached_page; |
1442 | ||
1da177e4 LT |
1443 | /* |
1444 | * Do we have something in the page cache already? | |
1445 | */ | |
1446 | retry_find: | |
d0217ac0 | 1447 | page = find_lock_page(mapping, vmf->pgoff); |
3ea89ee8 FW |
1448 | /* |
1449 | * For sequential accesses, we use the generic readahead logic. | |
1450 | */ | |
1451 | if (VM_SequentialReadHint(vma)) { | |
1452 | if (!page) { | |
cf914a7d | 1453 | page_cache_sync_readahead(mapping, ra, file, |
3ea89ee8 FW |
1454 | vmf->pgoff, 1); |
1455 | page = find_lock_page(mapping, vmf->pgoff); | |
1456 | if (!page) | |
1457 | goto no_cached_page; | |
1458 | } | |
1459 | if (PageReadahead(page)) { | |
cf914a7d | 1460 | page_cache_async_readahead(mapping, ra, file, page, |
3ea89ee8 FW |
1461 | vmf->pgoff, 1); |
1462 | } | |
1463 | } | |
1464 | ||
1da177e4 LT |
1465 | if (!page) { |
1466 | unsigned long ra_pages; | |
1467 | ||
1da177e4 LT |
1468 | ra->mmap_miss++; |
1469 | ||
1470 | /* | |
1471 | * Do we miss much more than hit in this file? If so, | |
1472 | * stop bothering with read-ahead. It will only hurt. | |
1473 | */ | |
0bb7ba6b | 1474 | if (ra->mmap_miss > MMAP_LOTSAMISS) |
1da177e4 LT |
1475 | goto no_cached_page; |
1476 | ||
1477 | /* | |
1478 | * To keep the pgmajfault counter straight, we need to | |
1479 | * check did_readaround, as this is an inner loop. | |
1480 | */ | |
1481 | if (!did_readaround) { | |
d0217ac0 | 1482 | ret = VM_FAULT_MAJOR; |
f8891e5e | 1483 | count_vm_event(PGMAJFAULT); |
1da177e4 LT |
1484 | } |
1485 | did_readaround = 1; | |
1486 | ra_pages = max_sane_readahead(file->f_ra.ra_pages); | |
1487 | if (ra_pages) { | |
1488 | pgoff_t start = 0; | |
1489 | ||
d0217ac0 NP |
1490 | if (vmf->pgoff > ra_pages / 2) |
1491 | start = vmf->pgoff - ra_pages / 2; | |
1da177e4 LT |
1492 | do_page_cache_readahead(mapping, file, start, ra_pages); |
1493 | } | |
d0217ac0 | 1494 | page = find_lock_page(mapping, vmf->pgoff); |
1da177e4 LT |
1495 | if (!page) |
1496 | goto no_cached_page; | |
1497 | } | |
1498 | ||
1499 | if (!did_readaround) | |
0bb7ba6b | 1500 | ra->mmap_miss--; |
1da177e4 LT |
1501 | |
1502 | /* | |
d00806b1 NP |
1503 | * We have a locked page in the page cache, now we need to check |
1504 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1505 | */ |
d00806b1 | 1506 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1507 | goto page_not_uptodate; |
1508 | ||
d00806b1 NP |
1509 | /* Must recheck i_size under page lock */ |
1510 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
d0217ac0 | 1511 | if (unlikely(vmf->pgoff >= size)) { |
d00806b1 | 1512 | unlock_page(page); |
745ad48e | 1513 | page_cache_release(page); |
5307cc1a | 1514 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1515 | } |
1516 | ||
1da177e4 LT |
1517 | /* |
1518 | * Found the page and have a reference on it. | |
1519 | */ | |
1520 | mark_page_accessed(page); | |
f4e6b498 | 1521 | ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT; |
d0217ac0 | 1522 | vmf->page = page; |
83c54070 | 1523 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 1524 | |
1da177e4 LT |
1525 | no_cached_page: |
1526 | /* | |
1527 | * We're only likely to ever get here if MADV_RANDOM is in | |
1528 | * effect. | |
1529 | */ | |
d0217ac0 | 1530 | error = page_cache_read(file, vmf->pgoff); |
1da177e4 LT |
1531 | |
1532 | /* | |
1533 | * The page we want has now been added to the page cache. | |
1534 | * In the unlikely event that someone removed it in the | |
1535 | * meantime, we'll just come back here and read it again. | |
1536 | */ | |
1537 | if (error >= 0) | |
1538 | goto retry_find; | |
1539 | ||
1540 | /* | |
1541 | * An error return from page_cache_read can result if the | |
1542 | * system is low on memory, or a problem occurs while trying | |
1543 | * to schedule I/O. | |
1544 | */ | |
1545 | if (error == -ENOMEM) | |
d0217ac0 NP |
1546 | return VM_FAULT_OOM; |
1547 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1548 | |
1549 | page_not_uptodate: | |
d00806b1 | 1550 | /* IO error path */ |
1da177e4 | 1551 | if (!did_readaround) { |
d0217ac0 | 1552 | ret = VM_FAULT_MAJOR; |
f8891e5e | 1553 | count_vm_event(PGMAJFAULT); |
1da177e4 | 1554 | } |
1da177e4 LT |
1555 | |
1556 | /* | |
1557 | * Umm, take care of errors if the page isn't up-to-date. | |
1558 | * Try to re-read it _once_. We do this synchronously, | |
1559 | * because there really aren't any performance issues here | |
1560 | * and we need to check for errors. | |
1561 | */ | |
1da177e4 | 1562 | ClearPageError(page); |
994fc28c | 1563 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
1564 | if (!error) { |
1565 | wait_on_page_locked(page); | |
1566 | if (!PageUptodate(page)) | |
1567 | error = -EIO; | |
1568 | } | |
d00806b1 NP |
1569 | page_cache_release(page); |
1570 | ||
1571 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 1572 | goto retry_find; |
1da177e4 | 1573 | |
d00806b1 | 1574 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 1575 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 1576 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
1577 | } |
1578 | EXPORT_SYMBOL(filemap_fault); | |
1579 | ||
1da177e4 | 1580 | struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 1581 | .fault = filemap_fault, |
1da177e4 LT |
1582 | }; |
1583 | ||
1584 | /* This is used for a general mmap of a disk file */ | |
1585 | ||
1586 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1587 | { | |
1588 | struct address_space *mapping = file->f_mapping; | |
1589 | ||
1590 | if (!mapping->a_ops->readpage) | |
1591 | return -ENOEXEC; | |
1592 | file_accessed(file); | |
1593 | vma->vm_ops = &generic_file_vm_ops; | |
d0217ac0 | 1594 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
1595 | return 0; |
1596 | } | |
1da177e4 LT |
1597 | |
1598 | /* | |
1599 | * This is for filesystems which do not implement ->writepage. | |
1600 | */ | |
1601 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
1602 | { | |
1603 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
1604 | return -EINVAL; | |
1605 | return generic_file_mmap(file, vma); | |
1606 | } | |
1607 | #else | |
1608 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1609 | { | |
1610 | return -ENOSYS; | |
1611 | } | |
1612 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
1613 | { | |
1614 | return -ENOSYS; | |
1615 | } | |
1616 | #endif /* CONFIG_MMU */ | |
1617 | ||
1618 | EXPORT_SYMBOL(generic_file_mmap); | |
1619 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
1620 | ||
6fe6900e | 1621 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 1622 | pgoff_t index, |
1da177e4 LT |
1623 | int (*filler)(void *,struct page*), |
1624 | void *data) | |
1625 | { | |
eb2be189 | 1626 | struct page *page; |
1da177e4 LT |
1627 | int err; |
1628 | repeat: | |
1629 | page = find_get_page(mapping, index); | |
1630 | if (!page) { | |
eb2be189 NP |
1631 | page = page_cache_alloc_cold(mapping); |
1632 | if (!page) | |
1633 | return ERR_PTR(-ENOMEM); | |
1634 | err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL); | |
1635 | if (unlikely(err)) { | |
1636 | page_cache_release(page); | |
1637 | if (err == -EEXIST) | |
1638 | goto repeat; | |
1da177e4 | 1639 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
1640 | return ERR_PTR(err); |
1641 | } | |
1da177e4 LT |
1642 | err = filler(data, page); |
1643 | if (err < 0) { | |
1644 | page_cache_release(page); | |
1645 | page = ERR_PTR(err); | |
1646 | } | |
1647 | } | |
1da177e4 LT |
1648 | return page; |
1649 | } | |
1650 | ||
7682486b RD |
1651 | /** |
1652 | * read_cache_page_async - read into page cache, fill it if needed | |
1653 | * @mapping: the page's address_space | |
1654 | * @index: the page index | |
1655 | * @filler: function to perform the read | |
1656 | * @data: destination for read data | |
1657 | * | |
6fe6900e NP |
1658 | * Same as read_cache_page, but don't wait for page to become unlocked |
1659 | * after submitting it to the filler. | |
7682486b RD |
1660 | * |
1661 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1662 | * not set, try to fill the page but don't wait for it to become unlocked. | |
1663 | * | |
1664 | * If the page does not get brought uptodate, return -EIO. | |
1da177e4 | 1665 | */ |
6fe6900e | 1666 | struct page *read_cache_page_async(struct address_space *mapping, |
57f6b96c | 1667 | pgoff_t index, |
1da177e4 LT |
1668 | int (*filler)(void *,struct page*), |
1669 | void *data) | |
1670 | { | |
1671 | struct page *page; | |
1672 | int err; | |
1673 | ||
1674 | retry: | |
1675 | page = __read_cache_page(mapping, index, filler, data); | |
1676 | if (IS_ERR(page)) | |
c855ff37 | 1677 | return page; |
1da177e4 LT |
1678 | if (PageUptodate(page)) |
1679 | goto out; | |
1680 | ||
1681 | lock_page(page); | |
1682 | if (!page->mapping) { | |
1683 | unlock_page(page); | |
1684 | page_cache_release(page); | |
1685 | goto retry; | |
1686 | } | |
1687 | if (PageUptodate(page)) { | |
1688 | unlock_page(page); | |
1689 | goto out; | |
1690 | } | |
1691 | err = filler(data, page); | |
1692 | if (err < 0) { | |
1693 | page_cache_release(page); | |
c855ff37 | 1694 | return ERR_PTR(err); |
1da177e4 | 1695 | } |
c855ff37 | 1696 | out: |
6fe6900e NP |
1697 | mark_page_accessed(page); |
1698 | return page; | |
1699 | } | |
1700 | EXPORT_SYMBOL(read_cache_page_async); | |
1701 | ||
1702 | /** | |
1703 | * read_cache_page - read into page cache, fill it if needed | |
1704 | * @mapping: the page's address_space | |
1705 | * @index: the page index | |
1706 | * @filler: function to perform the read | |
1707 | * @data: destination for read data | |
1708 | * | |
1709 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1710 | * not set, try to fill the page then wait for it to become unlocked. | |
1711 | * | |
1712 | * If the page does not get brought uptodate, return -EIO. | |
1713 | */ | |
1714 | struct page *read_cache_page(struct address_space *mapping, | |
57f6b96c | 1715 | pgoff_t index, |
6fe6900e NP |
1716 | int (*filler)(void *,struct page*), |
1717 | void *data) | |
1718 | { | |
1719 | struct page *page; | |
1720 | ||
1721 | page = read_cache_page_async(mapping, index, filler, data); | |
1722 | if (IS_ERR(page)) | |
1723 | goto out; | |
1724 | wait_on_page_locked(page); | |
1725 | if (!PageUptodate(page)) { | |
1726 | page_cache_release(page); | |
1727 | page = ERR_PTR(-EIO); | |
1728 | } | |
1da177e4 LT |
1729 | out: |
1730 | return page; | |
1731 | } | |
1da177e4 LT |
1732 | EXPORT_SYMBOL(read_cache_page); |
1733 | ||
1da177e4 LT |
1734 | /* |
1735 | * The logic we want is | |
1736 | * | |
1737 | * if suid or (sgid and xgrp) | |
1738 | * remove privs | |
1739 | */ | |
01de85e0 | 1740 | int should_remove_suid(struct dentry *dentry) |
1da177e4 LT |
1741 | { |
1742 | mode_t mode = dentry->d_inode->i_mode; | |
1743 | int kill = 0; | |
1da177e4 LT |
1744 | |
1745 | /* suid always must be killed */ | |
1746 | if (unlikely(mode & S_ISUID)) | |
1747 | kill = ATTR_KILL_SUID; | |
1748 | ||
1749 | /* | |
1750 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1751 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1752 | */ | |
1753 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1754 | kill |= ATTR_KILL_SGID; | |
1755 | ||
01de85e0 JA |
1756 | if (unlikely(kill && !capable(CAP_FSETID))) |
1757 | return kill; | |
1da177e4 | 1758 | |
01de85e0 JA |
1759 | return 0; |
1760 | } | |
d23a147b | 1761 | EXPORT_SYMBOL(should_remove_suid); |
01de85e0 | 1762 | |
7f3d4ee1 | 1763 | static int __remove_suid(struct dentry *dentry, int kill) |
01de85e0 JA |
1764 | { |
1765 | struct iattr newattrs; | |
1766 | ||
1767 | newattrs.ia_valid = ATTR_FORCE | kill; | |
1768 | return notify_change(dentry, &newattrs); | |
1769 | } | |
1770 | ||
2f1936b8 | 1771 | int file_remove_suid(struct file *file) |
01de85e0 | 1772 | { |
2f1936b8 | 1773 | struct dentry *dentry = file->f_path.dentry; |
b5376771 SH |
1774 | int killsuid = should_remove_suid(dentry); |
1775 | int killpriv = security_inode_need_killpriv(dentry); | |
1776 | int error = 0; | |
01de85e0 | 1777 | |
b5376771 SH |
1778 | if (killpriv < 0) |
1779 | return killpriv; | |
1780 | if (killpriv) | |
1781 | error = security_inode_killpriv(dentry); | |
1782 | if (!error && killsuid) | |
1783 | error = __remove_suid(dentry, killsuid); | |
01de85e0 | 1784 | |
b5376771 | 1785 | return error; |
1da177e4 | 1786 | } |
2f1936b8 | 1787 | EXPORT_SYMBOL(file_remove_suid); |
1da177e4 | 1788 | |
2f718ffc | 1789 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
1da177e4 LT |
1790 | const struct iovec *iov, size_t base, size_t bytes) |
1791 | { | |
1792 | size_t copied = 0, left = 0; | |
1793 | ||
1794 | while (bytes) { | |
1795 | char __user *buf = iov->iov_base + base; | |
1796 | int copy = min(bytes, iov->iov_len - base); | |
1797 | ||
1798 | base = 0; | |
c22ce143 | 1799 | left = __copy_from_user_inatomic_nocache(vaddr, buf, copy); |
1da177e4 LT |
1800 | copied += copy; |
1801 | bytes -= copy; | |
1802 | vaddr += copy; | |
1803 | iov++; | |
1804 | ||
01408c49 | 1805 | if (unlikely(left)) |
1da177e4 | 1806 | break; |
1da177e4 LT |
1807 | } |
1808 | return copied - left; | |
1809 | } | |
1810 | ||
2f718ffc NP |
1811 | /* |
1812 | * Copy as much as we can into the page and return the number of bytes which | |
1813 | * were sucessfully copied. If a fault is encountered then return the number of | |
1814 | * bytes which were copied. | |
1815 | */ | |
1816 | size_t iov_iter_copy_from_user_atomic(struct page *page, | |
1817 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
1818 | { | |
1819 | char *kaddr; | |
1820 | size_t copied; | |
1821 | ||
1822 | BUG_ON(!in_atomic()); | |
1823 | kaddr = kmap_atomic(page, KM_USER0); | |
1824 | if (likely(i->nr_segs == 1)) { | |
1825 | int left; | |
1826 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
1827 | left = __copy_from_user_inatomic_nocache(kaddr + offset, | |
1828 | buf, bytes); | |
1829 | copied = bytes - left; | |
1830 | } else { | |
1831 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
1832 | i->iov, i->iov_offset, bytes); | |
1833 | } | |
1834 | kunmap_atomic(kaddr, KM_USER0); | |
1835 | ||
1836 | return copied; | |
1837 | } | |
89e10787 | 1838 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); |
2f718ffc NP |
1839 | |
1840 | /* | |
1841 | * This has the same sideeffects and return value as | |
1842 | * iov_iter_copy_from_user_atomic(). | |
1843 | * The difference is that it attempts to resolve faults. | |
1844 | * Page must not be locked. | |
1845 | */ | |
1846 | size_t iov_iter_copy_from_user(struct page *page, | |
1847 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
1848 | { | |
1849 | char *kaddr; | |
1850 | size_t copied; | |
1851 | ||
1852 | kaddr = kmap(page); | |
1853 | if (likely(i->nr_segs == 1)) { | |
1854 | int left; | |
1855 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
1856 | left = __copy_from_user_nocache(kaddr + offset, buf, bytes); | |
1857 | copied = bytes - left; | |
1858 | } else { | |
1859 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
1860 | i->iov, i->iov_offset, bytes); | |
1861 | } | |
1862 | kunmap(page); | |
1863 | return copied; | |
1864 | } | |
89e10787 | 1865 | EXPORT_SYMBOL(iov_iter_copy_from_user); |
2f718ffc | 1866 | |
f7009264 | 1867 | void iov_iter_advance(struct iov_iter *i, size_t bytes) |
2f718ffc | 1868 | { |
f7009264 NP |
1869 | BUG_ON(i->count < bytes); |
1870 | ||
2f718ffc NP |
1871 | if (likely(i->nr_segs == 1)) { |
1872 | i->iov_offset += bytes; | |
f7009264 | 1873 | i->count -= bytes; |
2f718ffc NP |
1874 | } else { |
1875 | const struct iovec *iov = i->iov; | |
1876 | size_t base = i->iov_offset; | |
1877 | ||
124d3b70 NP |
1878 | /* |
1879 | * The !iov->iov_len check ensures we skip over unlikely | |
f7009264 | 1880 | * zero-length segments (without overruning the iovec). |
124d3b70 | 1881 | */ |
94ad374a | 1882 | while (bytes || unlikely(i->count && !iov->iov_len)) { |
f7009264 | 1883 | int copy; |
2f718ffc | 1884 | |
f7009264 NP |
1885 | copy = min(bytes, iov->iov_len - base); |
1886 | BUG_ON(!i->count || i->count < copy); | |
1887 | i->count -= copy; | |
2f718ffc NP |
1888 | bytes -= copy; |
1889 | base += copy; | |
1890 | if (iov->iov_len == base) { | |
1891 | iov++; | |
1892 | base = 0; | |
1893 | } | |
1894 | } | |
1895 | i->iov = iov; | |
1896 | i->iov_offset = base; | |
1897 | } | |
1898 | } | |
89e10787 | 1899 | EXPORT_SYMBOL(iov_iter_advance); |
2f718ffc | 1900 | |
afddba49 NP |
1901 | /* |
1902 | * Fault in the first iovec of the given iov_iter, to a maximum length | |
1903 | * of bytes. Returns 0 on success, or non-zero if the memory could not be | |
1904 | * accessed (ie. because it is an invalid address). | |
1905 | * | |
1906 | * writev-intensive code may want this to prefault several iovecs -- that | |
1907 | * would be possible (callers must not rely on the fact that _only_ the | |
1908 | * first iovec will be faulted with the current implementation). | |
1909 | */ | |
1910 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) | |
2f718ffc | 1911 | { |
2f718ffc | 1912 | char __user *buf = i->iov->iov_base + i->iov_offset; |
afddba49 NP |
1913 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); |
1914 | return fault_in_pages_readable(buf, bytes); | |
2f718ffc | 1915 | } |
89e10787 | 1916 | EXPORT_SYMBOL(iov_iter_fault_in_readable); |
2f718ffc NP |
1917 | |
1918 | /* | |
1919 | * Return the count of just the current iov_iter segment. | |
1920 | */ | |
1921 | size_t iov_iter_single_seg_count(struct iov_iter *i) | |
1922 | { | |
1923 | const struct iovec *iov = i->iov; | |
1924 | if (i->nr_segs == 1) | |
1925 | return i->count; | |
1926 | else | |
1927 | return min(i->count, iov->iov_len - i->iov_offset); | |
1928 | } | |
89e10787 | 1929 | EXPORT_SYMBOL(iov_iter_single_seg_count); |
2f718ffc | 1930 | |
1da177e4 LT |
1931 | /* |
1932 | * Performs necessary checks before doing a write | |
1933 | * | |
485bb99b | 1934 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
1935 | * Returns appropriate error code that caller should return or |
1936 | * zero in case that write should be allowed. | |
1937 | */ | |
1938 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) | |
1939 | { | |
1940 | struct inode *inode = file->f_mapping->host; | |
1941 | unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
1942 | ||
1943 | if (unlikely(*pos < 0)) | |
1944 | return -EINVAL; | |
1945 | ||
1da177e4 LT |
1946 | if (!isblk) { |
1947 | /* FIXME: this is for backwards compatibility with 2.4 */ | |
1948 | if (file->f_flags & O_APPEND) | |
1949 | *pos = i_size_read(inode); | |
1950 | ||
1951 | if (limit != RLIM_INFINITY) { | |
1952 | if (*pos >= limit) { | |
1953 | send_sig(SIGXFSZ, current, 0); | |
1954 | return -EFBIG; | |
1955 | } | |
1956 | if (*count > limit - (typeof(limit))*pos) { | |
1957 | *count = limit - (typeof(limit))*pos; | |
1958 | } | |
1959 | } | |
1960 | } | |
1961 | ||
1962 | /* | |
1963 | * LFS rule | |
1964 | */ | |
1965 | if (unlikely(*pos + *count > MAX_NON_LFS && | |
1966 | !(file->f_flags & O_LARGEFILE))) { | |
1967 | if (*pos >= MAX_NON_LFS) { | |
1da177e4 LT |
1968 | return -EFBIG; |
1969 | } | |
1970 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { | |
1971 | *count = MAX_NON_LFS - (unsigned long)*pos; | |
1972 | } | |
1973 | } | |
1974 | ||
1975 | /* | |
1976 | * Are we about to exceed the fs block limit ? | |
1977 | * | |
1978 | * If we have written data it becomes a short write. If we have | |
1979 | * exceeded without writing data we send a signal and return EFBIG. | |
1980 | * Linus frestrict idea will clean these up nicely.. | |
1981 | */ | |
1982 | if (likely(!isblk)) { | |
1983 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { | |
1984 | if (*count || *pos > inode->i_sb->s_maxbytes) { | |
1da177e4 LT |
1985 | return -EFBIG; |
1986 | } | |
1987 | /* zero-length writes at ->s_maxbytes are OK */ | |
1988 | } | |
1989 | ||
1990 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) | |
1991 | *count = inode->i_sb->s_maxbytes - *pos; | |
1992 | } else { | |
9361401e | 1993 | #ifdef CONFIG_BLOCK |
1da177e4 LT |
1994 | loff_t isize; |
1995 | if (bdev_read_only(I_BDEV(inode))) | |
1996 | return -EPERM; | |
1997 | isize = i_size_read(inode); | |
1998 | if (*pos >= isize) { | |
1999 | if (*count || *pos > isize) | |
2000 | return -ENOSPC; | |
2001 | } | |
2002 | ||
2003 | if (*pos + *count > isize) | |
2004 | *count = isize - *pos; | |
9361401e DH |
2005 | #else |
2006 | return -EPERM; | |
2007 | #endif | |
1da177e4 LT |
2008 | } |
2009 | return 0; | |
2010 | } | |
2011 | EXPORT_SYMBOL(generic_write_checks); | |
2012 | ||
afddba49 NP |
2013 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2014 | loff_t pos, unsigned len, unsigned flags, | |
2015 | struct page **pagep, void **fsdata) | |
2016 | { | |
2017 | const struct address_space_operations *aops = mapping->a_ops; | |
2018 | ||
2019 | if (aops->write_begin) { | |
2020 | return aops->write_begin(file, mapping, pos, len, flags, | |
2021 | pagep, fsdata); | |
2022 | } else { | |
2023 | int ret; | |
2024 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
2025 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); | |
2026 | struct inode *inode = mapping->host; | |
2027 | struct page *page; | |
2028 | again: | |
2029 | page = __grab_cache_page(mapping, index); | |
2030 | *pagep = page; | |
2031 | if (!page) | |
2032 | return -ENOMEM; | |
2033 | ||
2034 | if (flags & AOP_FLAG_UNINTERRUPTIBLE && !PageUptodate(page)) { | |
2035 | /* | |
2036 | * There is no way to resolve a short write situation | |
2037 | * for a !Uptodate page (except by double copying in | |
2038 | * the caller done by generic_perform_write_2copy). | |
2039 | * | |
2040 | * Instead, we have to bring it uptodate here. | |
2041 | */ | |
2042 | ret = aops->readpage(file, page); | |
2043 | page_cache_release(page); | |
2044 | if (ret) { | |
2045 | if (ret == AOP_TRUNCATED_PAGE) | |
2046 | goto again; | |
2047 | return ret; | |
2048 | } | |
2049 | goto again; | |
2050 | } | |
2051 | ||
2052 | ret = aops->prepare_write(file, page, offset, offset+len); | |
2053 | if (ret) { | |
55144768 | 2054 | unlock_page(page); |
afddba49 NP |
2055 | page_cache_release(page); |
2056 | if (pos + len > inode->i_size) | |
2057 | vmtruncate(inode, inode->i_size); | |
afddba49 NP |
2058 | } |
2059 | return ret; | |
2060 | } | |
2061 | } | |
2062 | EXPORT_SYMBOL(pagecache_write_begin); | |
2063 | ||
2064 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2065 | loff_t pos, unsigned len, unsigned copied, | |
2066 | struct page *page, void *fsdata) | |
2067 | { | |
2068 | const struct address_space_operations *aops = mapping->a_ops; | |
2069 | int ret; | |
2070 | ||
2071 | if (aops->write_end) { | |
2072 | mark_page_accessed(page); | |
2073 | ret = aops->write_end(file, mapping, pos, len, copied, | |
2074 | page, fsdata); | |
2075 | } else { | |
2076 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); | |
2077 | struct inode *inode = mapping->host; | |
2078 | ||
2079 | flush_dcache_page(page); | |
2080 | ret = aops->commit_write(file, page, offset, offset+len); | |
2081 | unlock_page(page); | |
2082 | mark_page_accessed(page); | |
2083 | page_cache_release(page); | |
afddba49 NP |
2084 | |
2085 | if (ret < 0) { | |
2086 | if (pos + len > inode->i_size) | |
2087 | vmtruncate(inode, inode->i_size); | |
2088 | } else if (ret > 0) | |
2089 | ret = min_t(size_t, copied, ret); | |
2090 | else | |
2091 | ret = copied; | |
2092 | } | |
2093 | ||
2094 | return ret; | |
2095 | } | |
2096 | EXPORT_SYMBOL(pagecache_write_end); | |
2097 | ||
1da177e4 LT |
2098 | ssize_t |
2099 | generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, | |
2100 | unsigned long *nr_segs, loff_t pos, loff_t *ppos, | |
2101 | size_t count, size_t ocount) | |
2102 | { | |
2103 | struct file *file = iocb->ki_filp; | |
2104 | struct address_space *mapping = file->f_mapping; | |
2105 | struct inode *inode = mapping->host; | |
2106 | ssize_t written; | |
a969e903 CH |
2107 | size_t write_len; |
2108 | pgoff_t end; | |
1da177e4 LT |
2109 | |
2110 | if (count != ocount) | |
2111 | *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); | |
2112 | ||
a969e903 CH |
2113 | /* |
2114 | * Unmap all mmappings of the file up-front. | |
2115 | * | |
2116 | * This will cause any pte dirty bits to be propagated into the | |
2117 | * pageframes for the subsequent filemap_write_and_wait(). | |
2118 | */ | |
2119 | write_len = iov_length(iov, *nr_segs); | |
2120 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; | |
2121 | if (mapping_mapped(mapping)) | |
2122 | unmap_mapping_range(mapping, pos, write_len, 0); | |
2123 | ||
2124 | written = filemap_write_and_wait(mapping); | |
2125 | if (written) | |
2126 | goto out; | |
2127 | ||
2128 | /* | |
2129 | * After a write we want buffered reads to be sure to go to disk to get | |
2130 | * the new data. We invalidate clean cached page from the region we're | |
2131 | * about to write. We do this *before* the write so that we can return | |
2132 | * -EIO without clobbering -EIOCBQUEUED from ->direct_IO(). | |
2133 | */ | |
2134 | if (mapping->nrpages) { | |
2135 | written = invalidate_inode_pages2_range(mapping, | |
2136 | pos >> PAGE_CACHE_SHIFT, end); | |
2137 | if (written) | |
2138 | goto out; | |
2139 | } | |
2140 | ||
2141 | written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs); | |
2142 | ||
2143 | /* | |
2144 | * Finally, try again to invalidate clean pages which might have been | |
2145 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2146 | * if the source of the write was an mmap'ed region of the file | |
2147 | * we're writing. Either one is a pretty crazy thing to do, | |
2148 | * so we don't support it 100%. If this invalidation | |
2149 | * fails, tough, the write still worked... | |
2150 | */ | |
2151 | if (mapping->nrpages) { | |
2152 | invalidate_inode_pages2_range(mapping, | |
2153 | pos >> PAGE_CACHE_SHIFT, end); | |
2154 | } | |
2155 | ||
1da177e4 LT |
2156 | if (written > 0) { |
2157 | loff_t end = pos + written; | |
2158 | if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { | |
2159 | i_size_write(inode, end); | |
2160 | mark_inode_dirty(inode); | |
2161 | } | |
2162 | *ppos = end; | |
2163 | } | |
2164 | ||
2165 | /* | |
2166 | * Sync the fs metadata but not the minor inode changes and | |
2167 | * of course not the data as we did direct DMA for the IO. | |
1b1dcc1b | 2168 | * i_mutex is held, which protects generic_osync_inode() from |
8459d86a | 2169 | * livelocking. AIO O_DIRECT ops attempt to sync metadata here. |
1da177e4 | 2170 | */ |
a969e903 | 2171 | out: |
8459d86a ZB |
2172 | if ((written >= 0 || written == -EIOCBQUEUED) && |
2173 | ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
1e8a81c5 HH |
2174 | int err = generic_osync_inode(inode, mapping, OSYNC_METADATA); |
2175 | if (err < 0) | |
2176 | written = err; | |
2177 | } | |
1da177e4 LT |
2178 | return written; |
2179 | } | |
2180 | EXPORT_SYMBOL(generic_file_direct_write); | |
2181 | ||
eb2be189 NP |
2182 | /* |
2183 | * Find or create a page at the given pagecache position. Return the locked | |
2184 | * page. This function is specifically for buffered writes. | |
2185 | */ | |
afddba49 | 2186 | struct page *__grab_cache_page(struct address_space *mapping, pgoff_t index) |
eb2be189 NP |
2187 | { |
2188 | int status; | |
2189 | struct page *page; | |
2190 | repeat: | |
2191 | page = find_lock_page(mapping, index); | |
2192 | if (likely(page)) | |
2193 | return page; | |
2194 | ||
2195 | page = page_cache_alloc(mapping); | |
2196 | if (!page) | |
2197 | return NULL; | |
2198 | status = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL); | |
2199 | if (unlikely(status)) { | |
2200 | page_cache_release(page); | |
2201 | if (status == -EEXIST) | |
2202 | goto repeat; | |
2203 | return NULL; | |
2204 | } | |
2205 | return page; | |
2206 | } | |
afddba49 | 2207 | EXPORT_SYMBOL(__grab_cache_page); |
eb2be189 | 2208 | |
afddba49 NP |
2209 | static ssize_t generic_perform_write_2copy(struct file *file, |
2210 | struct iov_iter *i, loff_t pos) | |
1da177e4 | 2211 | { |
ae37461c | 2212 | struct address_space *mapping = file->f_mapping; |
f5e54d6e | 2213 | const struct address_space_operations *a_ops = mapping->a_ops; |
afddba49 NP |
2214 | struct inode *inode = mapping->host; |
2215 | long status = 0; | |
2216 | ssize_t written = 0; | |
1da177e4 LT |
2217 | |
2218 | do { | |
08291429 | 2219 | struct page *src_page; |
eb2be189 | 2220 | struct page *page; |
ae37461c AM |
2221 | pgoff_t index; /* Pagecache index for current page */ |
2222 | unsigned long offset; /* Offset into pagecache page */ | |
08291429 | 2223 | unsigned long bytes; /* Bytes to write to page */ |
ae37461c | 2224 | size_t copied; /* Bytes copied from user */ |
1da177e4 | 2225 | |
ae37461c | 2226 | offset = (pos & (PAGE_CACHE_SIZE - 1)); |
1da177e4 | 2227 | index = pos >> PAGE_CACHE_SHIFT; |
2f718ffc | 2228 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
afddba49 | 2229 | iov_iter_count(i)); |
41cb8ac0 | 2230 | |
08291429 NP |
2231 | /* |
2232 | * a non-NULL src_page indicates that we're doing the | |
2233 | * copy via get_user_pages and kmap. | |
2234 | */ | |
2235 | src_page = NULL; | |
2236 | ||
41cb8ac0 NP |
2237 | /* |
2238 | * Bring in the user page that we will copy from _first_. | |
2239 | * Otherwise there's a nasty deadlock on copying from the | |
2240 | * same page as we're writing to, without it being marked | |
2241 | * up-to-date. | |
08291429 NP |
2242 | * |
2243 | * Not only is this an optimisation, but it is also required | |
2244 | * to check that the address is actually valid, when atomic | |
2245 | * usercopies are used, below. | |
41cb8ac0 | 2246 | */ |
afddba49 | 2247 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { |
08291429 NP |
2248 | status = -EFAULT; |
2249 | break; | |
2250 | } | |
eb2be189 NP |
2251 | |
2252 | page = __grab_cache_page(mapping, index); | |
1da177e4 LT |
2253 | if (!page) { |
2254 | status = -ENOMEM; | |
2255 | break; | |
2256 | } | |
2257 | ||
08291429 NP |
2258 | /* |
2259 | * non-uptodate pages cannot cope with short copies, and we | |
2260 | * cannot take a pagefault with the destination page locked. | |
2261 | * So pin the source page to copy it. | |
2262 | */ | |
674b892e | 2263 | if (!PageUptodate(page) && !segment_eq(get_fs(), KERNEL_DS)) { |
08291429 NP |
2264 | unlock_page(page); |
2265 | ||
2266 | src_page = alloc_page(GFP_KERNEL); | |
2267 | if (!src_page) { | |
2268 | page_cache_release(page); | |
2269 | status = -ENOMEM; | |
2270 | break; | |
2271 | } | |
2272 | ||
2273 | /* | |
2274 | * Cannot get_user_pages with a page locked for the | |
2275 | * same reason as we can't take a page fault with a | |
2276 | * page locked (as explained below). | |
2277 | */ | |
afddba49 | 2278 | copied = iov_iter_copy_from_user(src_page, i, |
2f718ffc | 2279 | offset, bytes); |
08291429 NP |
2280 | if (unlikely(copied == 0)) { |
2281 | status = -EFAULT; | |
2282 | page_cache_release(page); | |
2283 | page_cache_release(src_page); | |
2284 | break; | |
2285 | } | |
2286 | bytes = copied; | |
2287 | ||
2288 | lock_page(page); | |
2289 | /* | |
2290 | * Can't handle the page going uptodate here, because | |
2291 | * that means we would use non-atomic usercopies, which | |
2292 | * zero out the tail of the page, which can cause | |
2293 | * zeroes to become transiently visible. We could just | |
2294 | * use a non-zeroing copy, but the APIs aren't too | |
2295 | * consistent. | |
2296 | */ | |
2297 | if (unlikely(!page->mapping || PageUptodate(page))) { | |
2298 | unlock_page(page); | |
2299 | page_cache_release(page); | |
2300 | page_cache_release(src_page); | |
2301 | continue; | |
2302 | } | |
08291429 NP |
2303 | } |
2304 | ||
1da177e4 | 2305 | status = a_ops->prepare_write(file, page, offset, offset+bytes); |
64649a58 NP |
2306 | if (unlikely(status)) |
2307 | goto fs_write_aop_error; | |
994fc28c | 2308 | |
08291429 NP |
2309 | if (!src_page) { |
2310 | /* | |
2311 | * Must not enter the pagefault handler here, because | |
2312 | * we hold the page lock, so we might recursively | |
2313 | * deadlock on the same lock, or get an ABBA deadlock | |
2314 | * against a different lock, or against the mmap_sem | |
2315 | * (which nests outside the page lock). So increment | |
2316 | * preempt count, and use _atomic usercopies. | |
2317 | * | |
2318 | * The page is uptodate so we are OK to encounter a | |
2319 | * short copy: if unmodified parts of the page are | |
2320 | * marked dirty and written out to disk, it doesn't | |
2321 | * really matter. | |
2322 | */ | |
2323 | pagefault_disable(); | |
afddba49 | 2324 | copied = iov_iter_copy_from_user_atomic(page, i, |
2f718ffc | 2325 | offset, bytes); |
08291429 NP |
2326 | pagefault_enable(); |
2327 | } else { | |
2328 | void *src, *dst; | |
2329 | src = kmap_atomic(src_page, KM_USER0); | |
2330 | dst = kmap_atomic(page, KM_USER1); | |
2331 | memcpy(dst + offset, src + offset, bytes); | |
2332 | kunmap_atomic(dst, KM_USER1); | |
2333 | kunmap_atomic(src, KM_USER0); | |
2334 | copied = bytes; | |
2335 | } | |
1da177e4 | 2336 | flush_dcache_page(page); |
4a9e5ef1 | 2337 | |
1da177e4 | 2338 | status = a_ops->commit_write(file, page, offset, offset+bytes); |
55144768 | 2339 | if (unlikely(status < 0)) |
64649a58 | 2340 | goto fs_write_aop_error; |
64649a58 | 2341 | if (unlikely(status > 0)) /* filesystem did partial write */ |
08291429 NP |
2342 | copied = min_t(size_t, copied, status); |
2343 | ||
2344 | unlock_page(page); | |
2345 | mark_page_accessed(page); | |
2346 | page_cache_release(page); | |
2347 | if (src_page) | |
2348 | page_cache_release(src_page); | |
64649a58 | 2349 | |
afddba49 | 2350 | iov_iter_advance(i, copied); |
4a9e5ef1 | 2351 | pos += copied; |
afddba49 | 2352 | written += copied; |
4a9e5ef1 | 2353 | |
1da177e4 LT |
2354 | balance_dirty_pages_ratelimited(mapping); |
2355 | cond_resched(); | |
64649a58 NP |
2356 | continue; |
2357 | ||
2358 | fs_write_aop_error: | |
55144768 | 2359 | unlock_page(page); |
64649a58 | 2360 | page_cache_release(page); |
08291429 NP |
2361 | if (src_page) |
2362 | page_cache_release(src_page); | |
64649a58 NP |
2363 | |
2364 | /* | |
2365 | * prepare_write() may have instantiated a few blocks | |
2366 | * outside i_size. Trim these off again. Don't need | |
2367 | * i_size_read because we hold i_mutex. | |
2368 | */ | |
2369 | if (pos + bytes > inode->i_size) | |
2370 | vmtruncate(inode, inode->i_size); | |
55144768 | 2371 | break; |
afddba49 NP |
2372 | } while (iov_iter_count(i)); |
2373 | ||
2374 | return written ? written : status; | |
2375 | } | |
2376 | ||
2377 | static ssize_t generic_perform_write(struct file *file, | |
2378 | struct iov_iter *i, loff_t pos) | |
2379 | { | |
2380 | struct address_space *mapping = file->f_mapping; | |
2381 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2382 | long status = 0; | |
2383 | ssize_t written = 0; | |
674b892e NP |
2384 | unsigned int flags = 0; |
2385 | ||
2386 | /* | |
2387 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2388 | */ | |
2389 | if (segment_eq(get_fs(), KERNEL_DS)) | |
2390 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
afddba49 NP |
2391 | |
2392 | do { | |
2393 | struct page *page; | |
2394 | pgoff_t index; /* Pagecache index for current page */ | |
2395 | unsigned long offset; /* Offset into pagecache page */ | |
2396 | unsigned long bytes; /* Bytes to write to page */ | |
2397 | size_t copied; /* Bytes copied from user */ | |
2398 | void *fsdata; | |
2399 | ||
2400 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
2401 | index = pos >> PAGE_CACHE_SHIFT; | |
2402 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2403 | iov_iter_count(i)); | |
2404 | ||
2405 | again: | |
2406 | ||
2407 | /* | |
2408 | * Bring in the user page that we will copy from _first_. | |
2409 | * Otherwise there's a nasty deadlock on copying from the | |
2410 | * same page as we're writing to, without it being marked | |
2411 | * up-to-date. | |
2412 | * | |
2413 | * Not only is this an optimisation, but it is also required | |
2414 | * to check that the address is actually valid, when atomic | |
2415 | * usercopies are used, below. | |
2416 | */ | |
2417 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2418 | status = -EFAULT; | |
2419 | break; | |
2420 | } | |
2421 | ||
674b892e | 2422 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 NP |
2423 | &page, &fsdata); |
2424 | if (unlikely(status)) | |
2425 | break; | |
2426 | ||
2427 | pagefault_disable(); | |
2428 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); | |
2429 | pagefault_enable(); | |
2430 | flush_dcache_page(page); | |
2431 | ||
2432 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
2433 | page, fsdata); | |
2434 | if (unlikely(status < 0)) | |
2435 | break; | |
2436 | copied = status; | |
2437 | ||
2438 | cond_resched(); | |
2439 | ||
124d3b70 | 2440 | iov_iter_advance(i, copied); |
afddba49 NP |
2441 | if (unlikely(copied == 0)) { |
2442 | /* | |
2443 | * If we were unable to copy any data at all, we must | |
2444 | * fall back to a single segment length write. | |
2445 | * | |
2446 | * If we didn't fallback here, we could livelock | |
2447 | * because not all segments in the iov can be copied at | |
2448 | * once without a pagefault. | |
2449 | */ | |
2450 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2451 | iov_iter_single_seg_count(i)); | |
2452 | goto again; | |
2453 | } | |
afddba49 NP |
2454 | pos += copied; |
2455 | written += copied; | |
2456 | ||
2457 | balance_dirty_pages_ratelimited(mapping); | |
2458 | ||
2459 | } while (iov_iter_count(i)); | |
2460 | ||
2461 | return written ? written : status; | |
2462 | } | |
2463 | ||
2464 | ssize_t | |
2465 | generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, | |
2466 | unsigned long nr_segs, loff_t pos, loff_t *ppos, | |
2467 | size_t count, ssize_t written) | |
2468 | { | |
2469 | struct file *file = iocb->ki_filp; | |
2470 | struct address_space *mapping = file->f_mapping; | |
2471 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2472 | struct inode *inode = mapping->host; | |
2473 | ssize_t status; | |
2474 | struct iov_iter i; | |
2475 | ||
2476 | iov_iter_init(&i, iov, nr_segs, count, written); | |
2477 | if (a_ops->write_begin) | |
2478 | status = generic_perform_write(file, &i, pos); | |
2479 | else | |
2480 | status = generic_perform_write_2copy(file, &i, pos); | |
1da177e4 | 2481 | |
1da177e4 | 2482 | if (likely(status >= 0)) { |
afddba49 NP |
2483 | written += status; |
2484 | *ppos = pos + status; | |
2485 | ||
2486 | /* | |
2487 | * For now, when the user asks for O_SYNC, we'll actually give | |
2488 | * O_DSYNC | |
2489 | */ | |
1da177e4 LT |
2490 | if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) { |
2491 | if (!a_ops->writepage || !is_sync_kiocb(iocb)) | |
2492 | status = generic_osync_inode(inode, mapping, | |
2493 | OSYNC_METADATA|OSYNC_DATA); | |
2494 | } | |
2495 | } | |
2496 | ||
2497 | /* | |
2498 | * If we get here for O_DIRECT writes then we must have fallen through | |
2499 | * to buffered writes (block instantiation inside i_size). So we sync | |
2500 | * the file data here, to try to honour O_DIRECT expectations. | |
2501 | */ | |
2502 | if (unlikely(file->f_flags & O_DIRECT) && written) | |
2503 | status = filemap_write_and_wait(mapping); | |
2504 | ||
1da177e4 LT |
2505 | return written ? written : status; |
2506 | } | |
2507 | EXPORT_SYMBOL(generic_file_buffered_write); | |
2508 | ||
5ce7852c | 2509 | static ssize_t |
1da177e4 LT |
2510 | __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov, |
2511 | unsigned long nr_segs, loff_t *ppos) | |
2512 | { | |
2513 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2514 | struct address_space * mapping = file->f_mapping; |
1da177e4 LT |
2515 | size_t ocount; /* original count */ |
2516 | size_t count; /* after file limit checks */ | |
2517 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2518 | loff_t pos; |
2519 | ssize_t written; | |
2520 | ssize_t err; | |
2521 | ||
2522 | ocount = 0; | |
0ceb3314 DM |
2523 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
2524 | if (err) | |
2525 | return err; | |
1da177e4 LT |
2526 | |
2527 | count = ocount; | |
2528 | pos = *ppos; | |
2529 | ||
2530 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
2531 | ||
2532 | /* We can write back this queue in page reclaim */ | |
2533 | current->backing_dev_info = mapping->backing_dev_info; | |
2534 | written = 0; | |
2535 | ||
2536 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
2537 | if (err) | |
2538 | goto out; | |
2539 | ||
2540 | if (count == 0) | |
2541 | goto out; | |
2542 | ||
2f1936b8 | 2543 | err = file_remove_suid(file); |
1da177e4 LT |
2544 | if (err) |
2545 | goto out; | |
2546 | ||
870f4817 | 2547 | file_update_time(file); |
1da177e4 LT |
2548 | |
2549 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
2550 | if (unlikely(file->f_flags & O_DIRECT)) { | |
fb5527e6 JM |
2551 | loff_t endbyte; |
2552 | ssize_t written_buffered; | |
2553 | ||
2554 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, | |
2555 | ppos, count, ocount); | |
1da177e4 LT |
2556 | if (written < 0 || written == count) |
2557 | goto out; | |
2558 | /* | |
2559 | * direct-io write to a hole: fall through to buffered I/O | |
2560 | * for completing the rest of the request. | |
2561 | */ | |
2562 | pos += written; | |
2563 | count -= written; | |
fb5527e6 JM |
2564 | written_buffered = generic_file_buffered_write(iocb, iov, |
2565 | nr_segs, pos, ppos, count, | |
2566 | written); | |
2567 | /* | |
2568 | * If generic_file_buffered_write() retuned a synchronous error | |
2569 | * then we want to return the number of bytes which were | |
2570 | * direct-written, or the error code if that was zero. Note | |
2571 | * that this differs from normal direct-io semantics, which | |
2572 | * will return -EFOO even if some bytes were written. | |
2573 | */ | |
2574 | if (written_buffered < 0) { | |
2575 | err = written_buffered; | |
2576 | goto out; | |
2577 | } | |
1da177e4 | 2578 | |
fb5527e6 JM |
2579 | /* |
2580 | * We need to ensure that the page cache pages are written to | |
2581 | * disk and invalidated to preserve the expected O_DIRECT | |
2582 | * semantics. | |
2583 | */ | |
2584 | endbyte = pos + written_buffered - written - 1; | |
ef51c976 MF |
2585 | err = do_sync_mapping_range(file->f_mapping, pos, endbyte, |
2586 | SYNC_FILE_RANGE_WAIT_BEFORE| | |
2587 | SYNC_FILE_RANGE_WRITE| | |
2588 | SYNC_FILE_RANGE_WAIT_AFTER); | |
fb5527e6 JM |
2589 | if (err == 0) { |
2590 | written = written_buffered; | |
2591 | invalidate_mapping_pages(mapping, | |
2592 | pos >> PAGE_CACHE_SHIFT, | |
2593 | endbyte >> PAGE_CACHE_SHIFT); | |
2594 | } else { | |
2595 | /* | |
2596 | * We don't know how much we wrote, so just return | |
2597 | * the number of bytes which were direct-written | |
2598 | */ | |
2599 | } | |
2600 | } else { | |
2601 | written = generic_file_buffered_write(iocb, iov, nr_segs, | |
2602 | pos, ppos, count, written); | |
2603 | } | |
1da177e4 LT |
2604 | out: |
2605 | current->backing_dev_info = NULL; | |
2606 | return written ? written : err; | |
2607 | } | |
1da177e4 | 2608 | |
027445c3 BP |
2609 | ssize_t generic_file_aio_write_nolock(struct kiocb *iocb, |
2610 | const struct iovec *iov, unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
2611 | { |
2612 | struct file *file = iocb->ki_filp; | |
2613 | struct address_space *mapping = file->f_mapping; | |
2614 | struct inode *inode = mapping->host; | |
2615 | ssize_t ret; | |
1da177e4 | 2616 | |
027445c3 BP |
2617 | BUG_ON(iocb->ki_pos != pos); |
2618 | ||
2619 | ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, | |
2620 | &iocb->ki_pos); | |
1da177e4 LT |
2621 | |
2622 | if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
027445c3 | 2623 | ssize_t err; |
1da177e4 LT |
2624 | |
2625 | err = sync_page_range_nolock(inode, mapping, pos, ret); | |
2626 | if (err < 0) | |
2627 | ret = err; | |
2628 | } | |
2629 | return ret; | |
2630 | } | |
027445c3 | 2631 | EXPORT_SYMBOL(generic_file_aio_write_nolock); |
1da177e4 | 2632 | |
027445c3 BP |
2633 | ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
2634 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
2635 | { |
2636 | struct file *file = iocb->ki_filp; | |
2637 | struct address_space *mapping = file->f_mapping; | |
2638 | struct inode *inode = mapping->host; | |
2639 | ssize_t ret; | |
1da177e4 LT |
2640 | |
2641 | BUG_ON(iocb->ki_pos != pos); | |
2642 | ||
1b1dcc1b | 2643 | mutex_lock(&inode->i_mutex); |
027445c3 BP |
2644 | ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, |
2645 | &iocb->ki_pos); | |
1b1dcc1b | 2646 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
2647 | |
2648 | if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
2649 | ssize_t err; | |
2650 | ||
2651 | err = sync_page_range(inode, mapping, pos, ret); | |
2652 | if (err < 0) | |
2653 | ret = err; | |
2654 | } | |
2655 | return ret; | |
2656 | } | |
2657 | EXPORT_SYMBOL(generic_file_aio_write); | |
2658 | ||
cf9a2ae8 DH |
2659 | /** |
2660 | * try_to_release_page() - release old fs-specific metadata on a page | |
2661 | * | |
2662 | * @page: the page which the kernel is trying to free | |
2663 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2664 | * | |
2665 | * The address_space is to try to release any data against the page | |
2666 | * (presumably at page->private). If the release was successful, return `1'. | |
2667 | * Otherwise return zero. | |
2668 | * | |
2669 | * The @gfp_mask argument specifies whether I/O may be performed to release | |
3f31fddf | 2670 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
cf9a2ae8 | 2671 | * |
cf9a2ae8 DH |
2672 | */ |
2673 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2674 | { | |
2675 | struct address_space * const mapping = page->mapping; | |
2676 | ||
2677 | BUG_ON(!PageLocked(page)); | |
2678 | if (PageWriteback(page)) | |
2679 | return 0; | |
2680 | ||
2681 | if (mapping && mapping->a_ops->releasepage) | |
2682 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2683 | return try_to_free_buffers(page); | |
2684 | } | |
2685 | ||
2686 | EXPORT_SYMBOL(try_to_release_page); |