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 | 12 | #include <linux/module.h> |
1da177e4 LT |
13 | #include <linux/compiler.h> |
14 | #include <linux/fs.h> | |
c22ce143 | 15 | #include <linux/uaccess.h> |
1da177e4 | 16 | #include <linux/aio.h> |
c59ede7b | 17 | #include <linux/capability.h> |
1da177e4 | 18 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 19 | #include <linux/gfp.h> |
1da177e4 LT |
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> |
4f98a2fe | 36 | #include <linux/mm_inline.h> /* for page_is_file_cache() */ |
0f8053a5 NP |
37 | #include "internal.h" |
38 | ||
1da177e4 | 39 | /* |
1da177e4 LT |
40 | * FIXME: remove all knowledge of the buffer layer from the core VM |
41 | */ | |
148f948b | 42 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 43 | |
1da177e4 LT |
44 | #include <asm/mman.h> |
45 | ||
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 | * | |
25d9e2d1 | 61 | * ->i_mmap_lock (truncate_pagecache) |
1da177e4 | 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 | * | |
6a46079c AK |
105 | * (code doesn't rely on that order, so you could switch it around) |
106 | * ->tasklist_lock (memory_failure, collect_procs_ao) | |
107 | * ->i_mmap_lock | |
1da177e4 LT |
108 | */ |
109 | ||
110 | /* | |
111 | * Remove a page from the page cache and free it. Caller has to make | |
112 | * sure the page is locked and that nobody else uses it - or that usage | |
19fd6231 | 113 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 LT |
114 | */ |
115 | void __remove_from_page_cache(struct page *page) | |
116 | { | |
117 | struct address_space *mapping = page->mapping; | |
118 | ||
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); |
4b02108a KM |
123 | if (PageSwapBacked(page)) |
124 | __dec_zone_page_state(page, NR_SHMEM); | |
45426812 | 125 | BUG_ON(page_mapped(page)); |
3a692790 LT |
126 | |
127 | /* | |
128 | * Some filesystems seem to re-dirty the page even after | |
129 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
130 | * | |
131 | * Fix it up by doing a final dirty accounting check after | |
132 | * having removed the page entirely. | |
133 | */ | |
134 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { | |
135 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
136 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
137 | } | |
1da177e4 LT |
138 | } |
139 | ||
140 | void remove_from_page_cache(struct page *page) | |
141 | { | |
142 | struct address_space *mapping = page->mapping; | |
6072d13c | 143 | void (*freepage)(struct page *); |
1da177e4 | 144 | |
cd7619d6 | 145 | BUG_ON(!PageLocked(page)); |
1da177e4 | 146 | |
6072d13c | 147 | freepage = mapping->a_ops->freepage; |
19fd6231 | 148 | spin_lock_irq(&mapping->tree_lock); |
1da177e4 | 149 | __remove_from_page_cache(page); |
19fd6231 | 150 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 151 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
152 | |
153 | if (freepage) | |
154 | freepage(page); | |
1da177e4 | 155 | } |
a52116ab | 156 | EXPORT_SYMBOL(remove_from_page_cache); |
1da177e4 LT |
157 | |
158 | static int sync_page(void *word) | |
159 | { | |
160 | struct address_space *mapping; | |
161 | struct page *page; | |
162 | ||
07808b74 | 163 | page = container_of((unsigned long *)word, struct page, flags); |
1da177e4 LT |
164 | |
165 | /* | |
dd1d5afc WLII |
166 | * page_mapping() is being called without PG_locked held. |
167 | * Some knowledge of the state and use of the page is used to | |
168 | * reduce the requirements down to a memory barrier. | |
169 | * The danger here is of a stale page_mapping() return value | |
170 | * indicating a struct address_space different from the one it's | |
171 | * associated with when it is associated with one. | |
172 | * After smp_mb(), it's either the correct page_mapping() for | |
173 | * the page, or an old page_mapping() and the page's own | |
174 | * page_mapping() has gone NULL. | |
175 | * The ->sync_page() address_space operation must tolerate | |
176 | * page_mapping() going NULL. By an amazing coincidence, | |
177 | * this comes about because none of the users of the page | |
178 | * in the ->sync_page() methods make essential use of the | |
179 | * page_mapping(), merely passing the page down to the backing | |
180 | * device's unplug functions when it's non-NULL, which in turn | |
4c21e2f2 | 181 | * ignore it for all cases but swap, where only page_private(page) is |
dd1d5afc WLII |
182 | * of interest. When page_mapping() does go NULL, the entire |
183 | * call stack gracefully ignores the page and returns. | |
184 | * -- wli | |
1da177e4 LT |
185 | */ |
186 | smp_mb(); | |
187 | mapping = page_mapping(page); | |
188 | if (mapping && mapping->a_ops && mapping->a_ops->sync_page) | |
189 | mapping->a_ops->sync_page(page); | |
190 | io_schedule(); | |
191 | return 0; | |
192 | } | |
193 | ||
2687a356 MW |
194 | static int sync_page_killable(void *word) |
195 | { | |
196 | sync_page(word); | |
197 | return fatal_signal_pending(current) ? -EINTR : 0; | |
198 | } | |
199 | ||
1da177e4 | 200 | /** |
485bb99b | 201 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
202 | * @mapping: address space structure to write |
203 | * @start: offset in bytes where the range starts | |
469eb4d0 | 204 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 205 | * @sync_mode: enable synchronous operation |
1da177e4 | 206 | * |
485bb99b RD |
207 | * Start writeback against all of a mapping's dirty pages that lie |
208 | * within the byte offsets <start, end> inclusive. | |
209 | * | |
1da177e4 | 210 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 211 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
212 | * these two operations is that if a dirty page/buffer is encountered, it must |
213 | * be waited upon, and not just skipped over. | |
214 | */ | |
ebcf28e1 AM |
215 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
216 | loff_t end, int sync_mode) | |
1da177e4 LT |
217 | { |
218 | int ret; | |
219 | struct writeback_control wbc = { | |
220 | .sync_mode = sync_mode, | |
05fe478d | 221 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
222 | .range_start = start, |
223 | .range_end = end, | |
1da177e4 LT |
224 | }; |
225 | ||
226 | if (!mapping_cap_writeback_dirty(mapping)) | |
227 | return 0; | |
228 | ||
229 | ret = do_writepages(mapping, &wbc); | |
230 | return ret; | |
231 | } | |
232 | ||
233 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
234 | int sync_mode) | |
235 | { | |
111ebb6e | 236 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
237 | } |
238 | ||
239 | int filemap_fdatawrite(struct address_space *mapping) | |
240 | { | |
241 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
242 | } | |
243 | EXPORT_SYMBOL(filemap_fdatawrite); | |
244 | ||
f4c0a0fd | 245 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 246 | loff_t end) |
1da177e4 LT |
247 | { |
248 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
249 | } | |
f4c0a0fd | 250 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 251 | |
485bb99b RD |
252 | /** |
253 | * filemap_flush - mostly a non-blocking flush | |
254 | * @mapping: target address_space | |
255 | * | |
1da177e4 LT |
256 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
257 | * purposes - I/O may not be started against all dirty pages. | |
258 | */ | |
259 | int filemap_flush(struct address_space *mapping) | |
260 | { | |
261 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
262 | } | |
263 | EXPORT_SYMBOL(filemap_flush); | |
264 | ||
485bb99b | 265 | /** |
94004ed7 CH |
266 | * filemap_fdatawait_range - wait for writeback to complete |
267 | * @mapping: address space structure to wait for | |
268 | * @start_byte: offset in bytes where the range starts | |
269 | * @end_byte: offset in bytes where the range ends (inclusive) | |
485bb99b | 270 | * |
94004ed7 CH |
271 | * Walk the list of under-writeback pages of the given address space |
272 | * in the given range and wait for all of them. | |
1da177e4 | 273 | */ |
94004ed7 CH |
274 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
275 | loff_t end_byte) | |
1da177e4 | 276 | { |
94004ed7 CH |
277 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
278 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
279 | struct pagevec pvec; |
280 | int nr_pages; | |
281 | int ret = 0; | |
1da177e4 | 282 | |
94004ed7 | 283 | if (end_byte < start_byte) |
1da177e4 LT |
284 | return 0; |
285 | ||
286 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
287 | while ((index <= end) && |
288 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
289 | PAGECACHE_TAG_WRITEBACK, | |
290 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
291 | unsigned i; | |
292 | ||
293 | for (i = 0; i < nr_pages; i++) { | |
294 | struct page *page = pvec.pages[i]; | |
295 | ||
296 | /* until radix tree lookup accepts end_index */ | |
297 | if (page->index > end) | |
298 | continue; | |
299 | ||
300 | wait_on_page_writeback(page); | |
212260aa | 301 | if (TestClearPageError(page)) |
1da177e4 LT |
302 | ret = -EIO; |
303 | } | |
304 | pagevec_release(&pvec); | |
305 | cond_resched(); | |
306 | } | |
307 | ||
308 | /* Check for outstanding write errors */ | |
309 | if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
310 | ret = -ENOSPC; | |
311 | if (test_and_clear_bit(AS_EIO, &mapping->flags)) | |
312 | ret = -EIO; | |
313 | ||
314 | return ret; | |
315 | } | |
d3bccb6f JK |
316 | EXPORT_SYMBOL(filemap_fdatawait_range); |
317 | ||
1da177e4 | 318 | /** |
485bb99b | 319 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 320 | * @mapping: address space structure to wait for |
485bb99b RD |
321 | * |
322 | * Walk the list of under-writeback pages of the given address space | |
323 | * and wait for all of them. | |
1da177e4 LT |
324 | */ |
325 | int filemap_fdatawait(struct address_space *mapping) | |
326 | { | |
327 | loff_t i_size = i_size_read(mapping->host); | |
328 | ||
329 | if (i_size == 0) | |
330 | return 0; | |
331 | ||
94004ed7 | 332 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
333 | } |
334 | EXPORT_SYMBOL(filemap_fdatawait); | |
335 | ||
336 | int filemap_write_and_wait(struct address_space *mapping) | |
337 | { | |
28fd1298 | 338 | int err = 0; |
1da177e4 LT |
339 | |
340 | if (mapping->nrpages) { | |
28fd1298 OH |
341 | err = filemap_fdatawrite(mapping); |
342 | /* | |
343 | * Even if the above returned error, the pages may be | |
344 | * written partially (e.g. -ENOSPC), so we wait for it. | |
345 | * But the -EIO is special case, it may indicate the worst | |
346 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
347 | */ | |
348 | if (err != -EIO) { | |
349 | int err2 = filemap_fdatawait(mapping); | |
350 | if (!err) | |
351 | err = err2; | |
352 | } | |
1da177e4 | 353 | } |
28fd1298 | 354 | return err; |
1da177e4 | 355 | } |
28fd1298 | 356 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 357 | |
485bb99b RD |
358 | /** |
359 | * filemap_write_and_wait_range - write out & wait on a file range | |
360 | * @mapping: the address_space for the pages | |
361 | * @lstart: offset in bytes where the range starts | |
362 | * @lend: offset in bytes where the range ends (inclusive) | |
363 | * | |
469eb4d0 AM |
364 | * Write out and wait upon file offsets lstart->lend, inclusive. |
365 | * | |
366 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
367 | * that this function can be used to write to the very end-of-file (end = -1). | |
368 | */ | |
1da177e4 LT |
369 | int filemap_write_and_wait_range(struct address_space *mapping, |
370 | loff_t lstart, loff_t lend) | |
371 | { | |
28fd1298 | 372 | int err = 0; |
1da177e4 LT |
373 | |
374 | if (mapping->nrpages) { | |
28fd1298 OH |
375 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
376 | WB_SYNC_ALL); | |
377 | /* See comment of filemap_write_and_wait() */ | |
378 | if (err != -EIO) { | |
94004ed7 CH |
379 | int err2 = filemap_fdatawait_range(mapping, |
380 | lstart, lend); | |
28fd1298 OH |
381 | if (!err) |
382 | err = err2; | |
383 | } | |
1da177e4 | 384 | } |
28fd1298 | 385 | return err; |
1da177e4 | 386 | } |
f6995585 | 387 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 388 | |
ef6a3c63 MS |
389 | /** |
390 | * replace_page_cache_page - replace a pagecache page with a new one | |
391 | * @old: page to be replaced | |
392 | * @new: page to replace with | |
393 | * @gfp_mask: allocation mode | |
394 | * | |
395 | * This function replaces a page in the pagecache with a new one. On | |
396 | * success it acquires the pagecache reference for the new page and | |
397 | * drops it for the old page. Both the old and new pages must be | |
398 | * locked. This function does not add the new page to the LRU, the | |
399 | * caller must do that. | |
400 | * | |
401 | * The remove + add is atomic. The only way this function can fail is | |
402 | * memory allocation failure. | |
403 | */ | |
404 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
405 | { | |
406 | int error; | |
407 | struct mem_cgroup *memcg = NULL; | |
408 | ||
409 | VM_BUG_ON(!PageLocked(old)); | |
410 | VM_BUG_ON(!PageLocked(new)); | |
411 | VM_BUG_ON(new->mapping); | |
412 | ||
413 | /* | |
414 | * This is not page migration, but prepare_migration and | |
415 | * end_migration does enough work for charge replacement. | |
416 | * | |
417 | * In the longer term we probably want a specialized function | |
418 | * for moving the charge from old to new in a more efficient | |
419 | * manner. | |
420 | */ | |
421 | error = mem_cgroup_prepare_migration(old, new, &memcg, gfp_mask); | |
422 | if (error) | |
423 | return error; | |
424 | ||
425 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); | |
426 | if (!error) { | |
427 | struct address_space *mapping = old->mapping; | |
428 | void (*freepage)(struct page *); | |
429 | ||
430 | pgoff_t offset = old->index; | |
431 | freepage = mapping->a_ops->freepage; | |
432 | ||
433 | page_cache_get(new); | |
434 | new->mapping = mapping; | |
435 | new->index = offset; | |
436 | ||
437 | spin_lock_irq(&mapping->tree_lock); | |
438 | __remove_from_page_cache(old); | |
439 | error = radix_tree_insert(&mapping->page_tree, offset, new); | |
440 | BUG_ON(error); | |
441 | mapping->nrpages++; | |
442 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
443 | if (PageSwapBacked(new)) | |
444 | __inc_zone_page_state(new, NR_SHMEM); | |
445 | spin_unlock_irq(&mapping->tree_lock); | |
446 | radix_tree_preload_end(); | |
447 | if (freepage) | |
448 | freepage(old); | |
449 | page_cache_release(old); | |
450 | mem_cgroup_end_migration(memcg, old, new, true); | |
451 | } else { | |
452 | mem_cgroup_end_migration(memcg, old, new, false); | |
453 | } | |
454 | ||
455 | return error; | |
456 | } | |
457 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
458 | ||
485bb99b | 459 | /** |
e286781d | 460 | * add_to_page_cache_locked - add a locked page to the pagecache |
485bb99b RD |
461 | * @page: page to add |
462 | * @mapping: the page's address_space | |
463 | * @offset: page index | |
464 | * @gfp_mask: page allocation mode | |
465 | * | |
e286781d | 466 | * This function is used to add a page to the pagecache. It must be locked. |
1da177e4 LT |
467 | * This function does not add the page to the LRU. The caller must do that. |
468 | */ | |
e286781d | 469 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, |
6daa0e28 | 470 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 471 | { |
e286781d NP |
472 | int error; |
473 | ||
474 | VM_BUG_ON(!PageLocked(page)); | |
475 | ||
476 | error = mem_cgroup_cache_charge(page, current->mm, | |
2c26fdd7 | 477 | gfp_mask & GFP_RECLAIM_MASK); |
35c754d7 BS |
478 | if (error) |
479 | goto out; | |
1da177e4 | 480 | |
35c754d7 | 481 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
1da177e4 | 482 | if (error == 0) { |
e286781d NP |
483 | page_cache_get(page); |
484 | page->mapping = mapping; | |
485 | page->index = offset; | |
486 | ||
19fd6231 | 487 | spin_lock_irq(&mapping->tree_lock); |
1da177e4 | 488 | error = radix_tree_insert(&mapping->page_tree, offset, page); |
e286781d | 489 | if (likely(!error)) { |
1da177e4 | 490 | mapping->nrpages++; |
347ce434 | 491 | __inc_zone_page_state(page, NR_FILE_PAGES); |
4b02108a KM |
492 | if (PageSwapBacked(page)) |
493 | __inc_zone_page_state(page, NR_SHMEM); | |
e767e056 | 494 | spin_unlock_irq(&mapping->tree_lock); |
e286781d NP |
495 | } else { |
496 | page->mapping = NULL; | |
e767e056 | 497 | spin_unlock_irq(&mapping->tree_lock); |
69029cd5 | 498 | mem_cgroup_uncharge_cache_page(page); |
e286781d NP |
499 | page_cache_release(page); |
500 | } | |
1da177e4 | 501 | radix_tree_preload_end(); |
35c754d7 | 502 | } else |
69029cd5 | 503 | mem_cgroup_uncharge_cache_page(page); |
8a9f3ccd | 504 | out: |
1da177e4 LT |
505 | return error; |
506 | } | |
e286781d | 507 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
508 | |
509 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 510 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 511 | { |
4f98a2fe RR |
512 | int ret; |
513 | ||
514 | /* | |
515 | * Splice_read and readahead add shmem/tmpfs pages into the page cache | |
516 | * before shmem_readpage has a chance to mark them as SwapBacked: they | |
e9d6c157 | 517 | * need to go on the anon lru below, and mem_cgroup_cache_charge |
4f98a2fe RR |
518 | * (called in add_to_page_cache) needs to know where they're going too. |
519 | */ | |
520 | if (mapping_cap_swap_backed(mapping)) | |
521 | SetPageSwapBacked(page); | |
522 | ||
523 | ret = add_to_page_cache(page, mapping, offset, gfp_mask); | |
524 | if (ret == 0) { | |
525 | if (page_is_file_cache(page)) | |
526 | lru_cache_add_file(page); | |
527 | else | |
e9d6c157 | 528 | lru_cache_add_anon(page); |
4f98a2fe | 529 | } |
1da177e4 LT |
530 | return ret; |
531 | } | |
18bc0bbd | 532 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 533 | |
44110fe3 | 534 | #ifdef CONFIG_NUMA |
2ae88149 | 535 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 536 | { |
c0ff7453 MX |
537 | int n; |
538 | struct page *page; | |
539 | ||
44110fe3 | 540 | if (cpuset_do_page_mem_spread()) { |
c0ff7453 MX |
541 | get_mems_allowed(); |
542 | n = cpuset_mem_spread_node(); | |
543 | page = alloc_pages_exact_node(n, gfp, 0); | |
544 | put_mems_allowed(); | |
545 | return page; | |
44110fe3 | 546 | } |
2ae88149 | 547 | return alloc_pages(gfp, 0); |
44110fe3 | 548 | } |
2ae88149 | 549 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
550 | #endif |
551 | ||
db37648c NP |
552 | static int __sleep_on_page_lock(void *word) |
553 | { | |
554 | io_schedule(); | |
555 | return 0; | |
556 | } | |
557 | ||
1da177e4 LT |
558 | /* |
559 | * In order to wait for pages to become available there must be | |
560 | * waitqueues associated with pages. By using a hash table of | |
561 | * waitqueues where the bucket discipline is to maintain all | |
562 | * waiters on the same queue and wake all when any of the pages | |
563 | * become available, and for the woken contexts to check to be | |
564 | * sure the appropriate page became available, this saves space | |
565 | * at a cost of "thundering herd" phenomena during rare hash | |
566 | * collisions. | |
567 | */ | |
568 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
569 | { | |
570 | const struct zone *zone = page_zone(page); | |
571 | ||
572 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
573 | } | |
574 | ||
575 | static inline void wake_up_page(struct page *page, int bit) | |
576 | { | |
577 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); | |
578 | } | |
579 | ||
920c7a5d | 580 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
581 | { |
582 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
583 | ||
584 | if (test_bit(bit_nr, &page->flags)) | |
585 | __wait_on_bit(page_waitqueue(page), &wait, sync_page, | |
586 | TASK_UNINTERRUPTIBLE); | |
587 | } | |
588 | EXPORT_SYMBOL(wait_on_page_bit); | |
589 | ||
385e1ca5 DH |
590 | /** |
591 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
592 | * @page: Page defining the wait queue of interest |
593 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
594 | * |
595 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
596 | */ | |
597 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
598 | { | |
599 | wait_queue_head_t *q = page_waitqueue(page); | |
600 | unsigned long flags; | |
601 | ||
602 | spin_lock_irqsave(&q->lock, flags); | |
603 | __add_wait_queue(q, waiter); | |
604 | spin_unlock_irqrestore(&q->lock, flags); | |
605 | } | |
606 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
607 | ||
1da177e4 | 608 | /** |
485bb99b | 609 | * unlock_page - unlock a locked page |
1da177e4 LT |
610 | * @page: the page |
611 | * | |
612 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
613 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
614 | * mechananism between PageLocked pages and PageWriteback pages is shared. | |
615 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. | |
616 | * | |
8413ac9d NP |
617 | * The mb is necessary to enforce ordering between the clear_bit and the read |
618 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 619 | */ |
920c7a5d | 620 | void unlock_page(struct page *page) |
1da177e4 | 621 | { |
8413ac9d NP |
622 | VM_BUG_ON(!PageLocked(page)); |
623 | clear_bit_unlock(PG_locked, &page->flags); | |
624 | smp_mb__after_clear_bit(); | |
1da177e4 LT |
625 | wake_up_page(page, PG_locked); |
626 | } | |
627 | EXPORT_SYMBOL(unlock_page); | |
628 | ||
485bb99b RD |
629 | /** |
630 | * end_page_writeback - end writeback against a page | |
631 | * @page: the page | |
1da177e4 LT |
632 | */ |
633 | void end_page_writeback(struct page *page) | |
634 | { | |
ac6aadb2 MS |
635 | if (TestClearPageReclaim(page)) |
636 | rotate_reclaimable_page(page); | |
637 | ||
638 | if (!test_clear_page_writeback(page)) | |
639 | BUG(); | |
640 | ||
1da177e4 LT |
641 | smp_mb__after_clear_bit(); |
642 | wake_up_page(page, PG_writeback); | |
643 | } | |
644 | EXPORT_SYMBOL(end_page_writeback); | |
645 | ||
485bb99b RD |
646 | /** |
647 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
648 | * @page: the page to lock | |
1da177e4 | 649 | * |
485bb99b | 650 | * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some |
1da177e4 LT |
651 | * random driver's requestfn sets TASK_RUNNING, we could busywait. However |
652 | * chances are that on the second loop, the block layer's plug list is empty, | |
653 | * so sync_page() will then return in state TASK_UNINTERRUPTIBLE. | |
654 | */ | |
920c7a5d | 655 | void __lock_page(struct page *page) |
1da177e4 LT |
656 | { |
657 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
658 | ||
659 | __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page, | |
660 | TASK_UNINTERRUPTIBLE); | |
661 | } | |
662 | EXPORT_SYMBOL(__lock_page); | |
663 | ||
b5606c2d | 664 | int __lock_page_killable(struct page *page) |
2687a356 MW |
665 | { |
666 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
667 | ||
668 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
669 | sync_page_killable, TASK_KILLABLE); | |
670 | } | |
18bc0bbd | 671 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 672 | |
7682486b RD |
673 | /** |
674 | * __lock_page_nosync - get a lock on the page, without calling sync_page() | |
675 | * @page: the page to lock | |
676 | * | |
db37648c NP |
677 | * Variant of lock_page that does not require the caller to hold a reference |
678 | * on the page's mapping. | |
679 | */ | |
920c7a5d | 680 | void __lock_page_nosync(struct page *page) |
db37648c NP |
681 | { |
682 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
683 | __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock, | |
684 | TASK_UNINTERRUPTIBLE); | |
685 | } | |
686 | ||
d065bd81 ML |
687 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
688 | unsigned int flags) | |
689 | { | |
690 | if (!(flags & FAULT_FLAG_ALLOW_RETRY)) { | |
691 | __lock_page(page); | |
692 | return 1; | |
693 | } else { | |
318b275f GN |
694 | if (!(flags & FAULT_FLAG_RETRY_NOWAIT)) { |
695 | up_read(&mm->mmap_sem); | |
696 | wait_on_page_locked(page); | |
697 | } | |
d065bd81 ML |
698 | return 0; |
699 | } | |
700 | } | |
701 | ||
485bb99b RD |
702 | /** |
703 | * find_get_page - find and get a page reference | |
704 | * @mapping: the address_space to search | |
705 | * @offset: the page index | |
706 | * | |
da6052f7 NP |
707 | * Is there a pagecache struct page at the given (mapping, offset) tuple? |
708 | * If yes, increment its refcount and return it; if no, return NULL. | |
1da177e4 | 709 | */ |
a60637c8 | 710 | struct page *find_get_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 711 | { |
a60637c8 | 712 | void **pagep; |
1da177e4 LT |
713 | struct page *page; |
714 | ||
a60637c8 NP |
715 | rcu_read_lock(); |
716 | repeat: | |
717 | page = NULL; | |
718 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
719 | if (pagep) { | |
720 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
721 | if (unlikely(!page)) |
722 | goto out; | |
723 | if (radix_tree_deref_retry(page)) | |
a60637c8 NP |
724 | goto repeat; |
725 | ||
726 | if (!page_cache_get_speculative(page)) | |
727 | goto repeat; | |
728 | ||
729 | /* | |
730 | * Has the page moved? | |
731 | * This is part of the lockless pagecache protocol. See | |
732 | * include/linux/pagemap.h for details. | |
733 | */ | |
734 | if (unlikely(page != *pagep)) { | |
735 | page_cache_release(page); | |
736 | goto repeat; | |
737 | } | |
738 | } | |
27d20fdd | 739 | out: |
a60637c8 NP |
740 | rcu_read_unlock(); |
741 | ||
1da177e4 LT |
742 | return page; |
743 | } | |
1da177e4 LT |
744 | EXPORT_SYMBOL(find_get_page); |
745 | ||
1da177e4 LT |
746 | /** |
747 | * find_lock_page - locate, pin and lock a pagecache page | |
67be2dd1 MW |
748 | * @mapping: the address_space to search |
749 | * @offset: the page index | |
1da177e4 LT |
750 | * |
751 | * Locates the desired pagecache page, locks it, increments its reference | |
752 | * count and returns its address. | |
753 | * | |
754 | * Returns zero if the page was not present. find_lock_page() may sleep. | |
755 | */ | |
a60637c8 | 756 | struct page *find_lock_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 LT |
757 | { |
758 | struct page *page; | |
759 | ||
1da177e4 | 760 | repeat: |
a60637c8 | 761 | page = find_get_page(mapping, offset); |
1da177e4 | 762 | if (page) { |
a60637c8 NP |
763 | lock_page(page); |
764 | /* Has the page been truncated? */ | |
765 | if (unlikely(page->mapping != mapping)) { | |
766 | unlock_page(page); | |
767 | page_cache_release(page); | |
768 | goto repeat; | |
1da177e4 | 769 | } |
a60637c8 | 770 | VM_BUG_ON(page->index != offset); |
1da177e4 | 771 | } |
1da177e4 LT |
772 | return page; |
773 | } | |
1da177e4 LT |
774 | EXPORT_SYMBOL(find_lock_page); |
775 | ||
776 | /** | |
777 | * find_or_create_page - locate or add a pagecache page | |
67be2dd1 MW |
778 | * @mapping: the page's address_space |
779 | * @index: the page's index into the mapping | |
780 | * @gfp_mask: page allocation mode | |
1da177e4 LT |
781 | * |
782 | * Locates a page in the pagecache. If the page is not present, a new page | |
783 | * is allocated using @gfp_mask and is added to the pagecache and to the VM's | |
784 | * LRU list. The returned page is locked and has its reference count | |
785 | * incremented. | |
786 | * | |
787 | * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic | |
788 | * allocation! | |
789 | * | |
790 | * find_or_create_page() returns the desired page's address, or zero on | |
791 | * memory exhaustion. | |
792 | */ | |
793 | struct page *find_or_create_page(struct address_space *mapping, | |
57f6b96c | 794 | pgoff_t index, gfp_t gfp_mask) |
1da177e4 | 795 | { |
eb2be189 | 796 | struct page *page; |
1da177e4 LT |
797 | int err; |
798 | repeat: | |
799 | page = find_lock_page(mapping, index); | |
800 | if (!page) { | |
eb2be189 NP |
801 | page = __page_cache_alloc(gfp_mask); |
802 | if (!page) | |
803 | return NULL; | |
67d58ac4 NP |
804 | /* |
805 | * We want a regular kernel memory (not highmem or DMA etc) | |
806 | * allocation for the radix tree nodes, but we need to honour | |
807 | * the context-specific requirements the caller has asked for. | |
808 | * GFP_RECLAIM_MASK collects those requirements. | |
809 | */ | |
810 | err = add_to_page_cache_lru(page, mapping, index, | |
811 | (gfp_mask & GFP_RECLAIM_MASK)); | |
eb2be189 NP |
812 | if (unlikely(err)) { |
813 | page_cache_release(page); | |
814 | page = NULL; | |
815 | if (err == -EEXIST) | |
816 | goto repeat; | |
1da177e4 | 817 | } |
1da177e4 | 818 | } |
1da177e4 LT |
819 | return page; |
820 | } | |
1da177e4 LT |
821 | EXPORT_SYMBOL(find_or_create_page); |
822 | ||
823 | /** | |
824 | * find_get_pages - gang pagecache lookup | |
825 | * @mapping: The address_space to search | |
826 | * @start: The starting page index | |
827 | * @nr_pages: The maximum number of pages | |
828 | * @pages: Where the resulting pages are placed | |
829 | * | |
830 | * find_get_pages() will search for and return a group of up to | |
831 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
832 | * find_get_pages() takes a reference against the returned pages. | |
833 | * | |
834 | * The search returns a group of mapping-contiguous pages with ascending | |
835 | * indexes. There may be holes in the indices due to not-present pages. | |
836 | * | |
837 | * find_get_pages() returns the number of pages which were found. | |
838 | */ | |
839 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
840 | unsigned int nr_pages, struct page **pages) | |
841 | { | |
842 | unsigned int i; | |
843 | unsigned int ret; | |
a60637c8 NP |
844 | unsigned int nr_found; |
845 | ||
846 | rcu_read_lock(); | |
847 | restart: | |
848 | nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, | |
849 | (void ***)pages, start, nr_pages); | |
850 | ret = 0; | |
851 | for (i = 0; i < nr_found; i++) { | |
852 | struct page *page; | |
853 | repeat: | |
854 | page = radix_tree_deref_slot((void **)pages[i]); | |
855 | if (unlikely(!page)) | |
856 | continue; | |
27d20fdd NP |
857 | if (radix_tree_deref_retry(page)) { |
858 | if (ret) | |
859 | start = pages[ret-1]->index; | |
a60637c8 | 860 | goto restart; |
27d20fdd | 861 | } |
a60637c8 NP |
862 | |
863 | if (!page_cache_get_speculative(page)) | |
864 | goto repeat; | |
865 | ||
866 | /* Has the page moved? */ | |
867 | if (unlikely(page != *((void **)pages[i]))) { | |
868 | page_cache_release(page); | |
869 | goto repeat; | |
870 | } | |
1da177e4 | 871 | |
a60637c8 NP |
872 | pages[ret] = page; |
873 | ret++; | |
874 | } | |
875 | rcu_read_unlock(); | |
1da177e4 LT |
876 | return ret; |
877 | } | |
878 | ||
ebf43500 JA |
879 | /** |
880 | * find_get_pages_contig - gang contiguous pagecache lookup | |
881 | * @mapping: The address_space to search | |
882 | * @index: The starting page index | |
883 | * @nr_pages: The maximum number of pages | |
884 | * @pages: Where the resulting pages are placed | |
885 | * | |
886 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
887 | * that the returned number of pages are guaranteed to be contiguous. | |
888 | * | |
889 | * find_get_pages_contig() returns the number of pages which were found. | |
890 | */ | |
891 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
892 | unsigned int nr_pages, struct page **pages) | |
893 | { | |
894 | unsigned int i; | |
895 | unsigned int ret; | |
a60637c8 NP |
896 | unsigned int nr_found; |
897 | ||
898 | rcu_read_lock(); | |
899 | restart: | |
900 | nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, | |
901 | (void ***)pages, index, nr_pages); | |
902 | ret = 0; | |
903 | for (i = 0; i < nr_found; i++) { | |
904 | struct page *page; | |
905 | repeat: | |
906 | page = radix_tree_deref_slot((void **)pages[i]); | |
907 | if (unlikely(!page)) | |
908 | continue; | |
27d20fdd | 909 | if (radix_tree_deref_retry(page)) |
a60637c8 | 910 | goto restart; |
ebf43500 | 911 | |
a60637c8 NP |
912 | if (!page_cache_get_speculative(page)) |
913 | goto repeat; | |
914 | ||
915 | /* Has the page moved? */ | |
916 | if (unlikely(page != *((void **)pages[i]))) { | |
917 | page_cache_release(page); | |
918 | goto repeat; | |
919 | } | |
920 | ||
9cbb4cb2 NP |
921 | /* |
922 | * must check mapping and index after taking the ref. | |
923 | * otherwise we can get both false positives and false | |
924 | * negatives, which is just confusing to the caller. | |
925 | */ | |
926 | if (page->mapping == NULL || page->index != index) { | |
927 | page_cache_release(page); | |
928 | break; | |
929 | } | |
930 | ||
a60637c8 NP |
931 | pages[ret] = page; |
932 | ret++; | |
ebf43500 JA |
933 | index++; |
934 | } | |
a60637c8 NP |
935 | rcu_read_unlock(); |
936 | return ret; | |
ebf43500 | 937 | } |
ef71c15c | 938 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 939 | |
485bb99b RD |
940 | /** |
941 | * find_get_pages_tag - find and return pages that match @tag | |
942 | * @mapping: the address_space to search | |
943 | * @index: the starting page index | |
944 | * @tag: the tag index | |
945 | * @nr_pages: the maximum number of pages | |
946 | * @pages: where the resulting pages are placed | |
947 | * | |
1da177e4 | 948 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 949 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
950 | */ |
951 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
952 | int tag, unsigned int nr_pages, struct page **pages) | |
953 | { | |
954 | unsigned int i; | |
955 | unsigned int ret; | |
a60637c8 NP |
956 | unsigned int nr_found; |
957 | ||
958 | rcu_read_lock(); | |
959 | restart: | |
960 | nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree, | |
961 | (void ***)pages, *index, nr_pages, tag); | |
962 | ret = 0; | |
963 | for (i = 0; i < nr_found; i++) { | |
964 | struct page *page; | |
965 | repeat: | |
966 | page = radix_tree_deref_slot((void **)pages[i]); | |
967 | if (unlikely(!page)) | |
968 | continue; | |
27d20fdd | 969 | if (radix_tree_deref_retry(page)) |
a60637c8 NP |
970 | goto restart; |
971 | ||
972 | if (!page_cache_get_speculative(page)) | |
973 | goto repeat; | |
974 | ||
975 | /* Has the page moved? */ | |
976 | if (unlikely(page != *((void **)pages[i]))) { | |
977 | page_cache_release(page); | |
978 | goto repeat; | |
979 | } | |
980 | ||
981 | pages[ret] = page; | |
982 | ret++; | |
983 | } | |
984 | rcu_read_unlock(); | |
1da177e4 | 985 | |
1da177e4 LT |
986 | if (ret) |
987 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 988 | |
1da177e4 LT |
989 | return ret; |
990 | } | |
ef71c15c | 991 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 992 | |
485bb99b RD |
993 | /** |
994 | * grab_cache_page_nowait - returns locked page at given index in given cache | |
995 | * @mapping: target address_space | |
996 | * @index: the page index | |
997 | * | |
72fd4a35 | 998 | * Same as grab_cache_page(), but do not wait if the page is unavailable. |
1da177e4 LT |
999 | * This is intended for speculative data generators, where the data can |
1000 | * be regenerated if the page couldn't be grabbed. This routine should | |
1001 | * be safe to call while holding the lock for another page. | |
1002 | * | |
1003 | * Clear __GFP_FS when allocating the page to avoid recursion into the fs | |
1004 | * and deadlock against the caller's locked page. | |
1005 | */ | |
1006 | struct page * | |
57f6b96c | 1007 | grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) |
1da177e4 LT |
1008 | { |
1009 | struct page *page = find_get_page(mapping, index); | |
1da177e4 LT |
1010 | |
1011 | if (page) { | |
529ae9aa | 1012 | if (trylock_page(page)) |
1da177e4 LT |
1013 | return page; |
1014 | page_cache_release(page); | |
1015 | return NULL; | |
1016 | } | |
2ae88149 | 1017 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); |
67d58ac4 | 1018 | if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) { |
1da177e4 LT |
1019 | page_cache_release(page); |
1020 | page = NULL; | |
1021 | } | |
1022 | return page; | |
1023 | } | |
1da177e4 LT |
1024 | EXPORT_SYMBOL(grab_cache_page_nowait); |
1025 | ||
76d42bd9 WF |
1026 | /* |
1027 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1028 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1029 | * | |
1030 | * ---R__________________________________________B__________ | |
1031 | * ^ reading here ^ bad block(assume 4k) | |
1032 | * | |
1033 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1034 | * => failing the whole request => read(R) => read(R+1) => | |
1035 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1036 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1037 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1038 | * | |
1039 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1040 | */ | |
1041 | static void shrink_readahead_size_eio(struct file *filp, | |
1042 | struct file_ra_state *ra) | |
1043 | { | |
76d42bd9 | 1044 | ra->ra_pages /= 4; |
76d42bd9 WF |
1045 | } |
1046 | ||
485bb99b | 1047 | /** |
36e78914 | 1048 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1049 | * @filp: the file to read |
1050 | * @ppos: current file position | |
1051 | * @desc: read_descriptor | |
1052 | * @actor: read method | |
1053 | * | |
1da177e4 | 1054 | * This is a generic file read routine, and uses the |
485bb99b | 1055 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1056 | * |
1057 | * This is really ugly. But the goto's actually try to clarify some | |
1058 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1059 | */ |
36e78914 CH |
1060 | static void do_generic_file_read(struct file *filp, loff_t *ppos, |
1061 | read_descriptor_t *desc, read_actor_t actor) | |
1da177e4 | 1062 | { |
36e78914 | 1063 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1064 | struct inode *inode = mapping->host; |
36e78914 | 1065 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1066 | pgoff_t index; |
1067 | pgoff_t last_index; | |
1068 | pgoff_t prev_index; | |
1069 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1070 | unsigned int prev_offset; |
1da177e4 | 1071 | int error; |
1da177e4 | 1072 | |
1da177e4 | 1073 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1074 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1075 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
1da177e4 LT |
1076 | last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1077 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1078 | ||
1da177e4 LT |
1079 | for (;;) { |
1080 | struct page *page; | |
57f6b96c | 1081 | pgoff_t end_index; |
a32ea1e1 | 1082 | loff_t isize; |
1da177e4 LT |
1083 | unsigned long nr, ret; |
1084 | ||
1da177e4 | 1085 | cond_resched(); |
1da177e4 LT |
1086 | find_page: |
1087 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1088 | if (!page) { |
cf914a7d | 1089 | page_cache_sync_readahead(mapping, |
7ff81078 | 1090 | ra, filp, |
3ea89ee8 FW |
1091 | index, last_index - index); |
1092 | page = find_get_page(mapping, index); | |
1093 | if (unlikely(page == NULL)) | |
1094 | goto no_cached_page; | |
1095 | } | |
1096 | if (PageReadahead(page)) { | |
cf914a7d | 1097 | page_cache_async_readahead(mapping, |
7ff81078 | 1098 | ra, filp, page, |
3ea89ee8 | 1099 | index, last_index - index); |
1da177e4 | 1100 | } |
8ab22b9a HH |
1101 | if (!PageUptodate(page)) { |
1102 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1103 | !mapping->a_ops->is_partially_uptodate) | |
1104 | goto page_not_up_to_date; | |
529ae9aa | 1105 | if (!trylock_page(page)) |
8ab22b9a | 1106 | goto page_not_up_to_date; |
8d056cb9 DH |
1107 | /* Did it get truncated before we got the lock? */ |
1108 | if (!page->mapping) | |
1109 | goto page_not_up_to_date_locked; | |
8ab22b9a HH |
1110 | if (!mapping->a_ops->is_partially_uptodate(page, |
1111 | desc, offset)) | |
1112 | goto page_not_up_to_date_locked; | |
1113 | unlock_page(page); | |
1114 | } | |
1da177e4 | 1115 | page_ok: |
a32ea1e1 N |
1116 | /* |
1117 | * i_size must be checked after we know the page is Uptodate. | |
1118 | * | |
1119 | * Checking i_size after the check allows us to calculate | |
1120 | * the correct value for "nr", which means the zero-filled | |
1121 | * part of the page is not copied back to userspace (unless | |
1122 | * another truncate extends the file - this is desired though). | |
1123 | */ | |
1124 | ||
1125 | isize = i_size_read(inode); | |
1126 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1127 | if (unlikely(!isize || index > end_index)) { | |
1128 | page_cache_release(page); | |
1129 | goto out; | |
1130 | } | |
1131 | ||
1132 | /* nr is the maximum number of bytes to copy from this page */ | |
1133 | nr = PAGE_CACHE_SIZE; | |
1134 | if (index == end_index) { | |
1135 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1136 | if (nr <= offset) { | |
1137 | page_cache_release(page); | |
1138 | goto out; | |
1139 | } | |
1140 | } | |
1141 | nr = nr - offset; | |
1da177e4 LT |
1142 | |
1143 | /* If users can be writing to this page using arbitrary | |
1144 | * virtual addresses, take care about potential aliasing | |
1145 | * before reading the page on the kernel side. | |
1146 | */ | |
1147 | if (mapping_writably_mapped(mapping)) | |
1148 | flush_dcache_page(page); | |
1149 | ||
1150 | /* | |
ec0f1637 JK |
1151 | * When a sequential read accesses a page several times, |
1152 | * only mark it as accessed the first time. | |
1da177e4 | 1153 | */ |
ec0f1637 | 1154 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1155 | mark_page_accessed(page); |
1156 | prev_index = index; | |
1157 | ||
1158 | /* | |
1159 | * Ok, we have the page, and it's up-to-date, so | |
1160 | * now we can copy it to user space... | |
1161 | * | |
1162 | * The actor routine returns how many bytes were actually used.. | |
1163 | * NOTE! This may not be the same as how much of a user buffer | |
1164 | * we filled up (we may be padding etc), so we can only update | |
1165 | * "pos" here (the actor routine has to update the user buffer | |
1166 | * pointers and the remaining count). | |
1167 | */ | |
1168 | ret = actor(desc, page, offset, nr); | |
1169 | offset += ret; | |
1170 | index += offset >> PAGE_CACHE_SHIFT; | |
1171 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1172 | prev_offset = offset; |
1da177e4 LT |
1173 | |
1174 | page_cache_release(page); | |
1175 | if (ret == nr && desc->count) | |
1176 | continue; | |
1177 | goto out; | |
1178 | ||
1179 | page_not_up_to_date: | |
1180 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1181 | error = lock_page_killable(page); |
1182 | if (unlikely(error)) | |
1183 | goto readpage_error; | |
1da177e4 | 1184 | |
8ab22b9a | 1185 | page_not_up_to_date_locked: |
da6052f7 | 1186 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1187 | if (!page->mapping) { |
1188 | unlock_page(page); | |
1189 | page_cache_release(page); | |
1190 | continue; | |
1191 | } | |
1192 | ||
1193 | /* Did somebody else fill it already? */ | |
1194 | if (PageUptodate(page)) { | |
1195 | unlock_page(page); | |
1196 | goto page_ok; | |
1197 | } | |
1198 | ||
1199 | readpage: | |
91803b49 JM |
1200 | /* |
1201 | * A previous I/O error may have been due to temporary | |
1202 | * failures, eg. multipath errors. | |
1203 | * PG_error will be set again if readpage fails. | |
1204 | */ | |
1205 | ClearPageError(page); | |
1da177e4 LT |
1206 | /* Start the actual read. The read will unlock the page. */ |
1207 | error = mapping->a_ops->readpage(filp, page); | |
1208 | ||
994fc28c ZB |
1209 | if (unlikely(error)) { |
1210 | if (error == AOP_TRUNCATED_PAGE) { | |
1211 | page_cache_release(page); | |
1212 | goto find_page; | |
1213 | } | |
1da177e4 | 1214 | goto readpage_error; |
994fc28c | 1215 | } |
1da177e4 LT |
1216 | |
1217 | if (!PageUptodate(page)) { | |
85462323 ON |
1218 | error = lock_page_killable(page); |
1219 | if (unlikely(error)) | |
1220 | goto readpage_error; | |
1da177e4 LT |
1221 | if (!PageUptodate(page)) { |
1222 | if (page->mapping == NULL) { | |
1223 | /* | |
2ecdc82e | 1224 | * invalidate_mapping_pages got it |
1da177e4 LT |
1225 | */ |
1226 | unlock_page(page); | |
1227 | page_cache_release(page); | |
1228 | goto find_page; | |
1229 | } | |
1230 | unlock_page(page); | |
7ff81078 | 1231 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1232 | error = -EIO; |
1233 | goto readpage_error; | |
1da177e4 LT |
1234 | } |
1235 | unlock_page(page); | |
1236 | } | |
1237 | ||
1da177e4 LT |
1238 | goto page_ok; |
1239 | ||
1240 | readpage_error: | |
1241 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1242 | desc->error = error; | |
1243 | page_cache_release(page); | |
1244 | goto out; | |
1245 | ||
1246 | no_cached_page: | |
1247 | /* | |
1248 | * Ok, it wasn't cached, so we need to create a new | |
1249 | * page.. | |
1250 | */ | |
eb2be189 NP |
1251 | page = page_cache_alloc_cold(mapping); |
1252 | if (!page) { | |
1253 | desc->error = -ENOMEM; | |
1254 | goto out; | |
1da177e4 | 1255 | } |
eb2be189 | 1256 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1257 | index, GFP_KERNEL); |
1258 | if (error) { | |
eb2be189 | 1259 | page_cache_release(page); |
1da177e4 LT |
1260 | if (error == -EEXIST) |
1261 | goto find_page; | |
1262 | desc->error = error; | |
1263 | goto out; | |
1264 | } | |
1da177e4 LT |
1265 | goto readpage; |
1266 | } | |
1267 | ||
1268 | out: | |
7ff81078 FW |
1269 | ra->prev_pos = prev_index; |
1270 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1271 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1272 | |
f4e6b498 | 1273 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1274 | file_accessed(filp); |
1da177e4 | 1275 | } |
1da177e4 LT |
1276 | |
1277 | int file_read_actor(read_descriptor_t *desc, struct page *page, | |
1278 | unsigned long offset, unsigned long size) | |
1279 | { | |
1280 | char *kaddr; | |
1281 | unsigned long left, count = desc->count; | |
1282 | ||
1283 | if (size > count) | |
1284 | size = count; | |
1285 | ||
1286 | /* | |
1287 | * Faults on the destination of a read are common, so do it before | |
1288 | * taking the kmap. | |
1289 | */ | |
1290 | if (!fault_in_pages_writeable(desc->arg.buf, size)) { | |
1291 | kaddr = kmap_atomic(page, KM_USER0); | |
1292 | left = __copy_to_user_inatomic(desc->arg.buf, | |
1293 | kaddr + offset, size); | |
1294 | kunmap_atomic(kaddr, KM_USER0); | |
1295 | if (left == 0) | |
1296 | goto success; | |
1297 | } | |
1298 | ||
1299 | /* Do it the slow way */ | |
1300 | kaddr = kmap(page); | |
1301 | left = __copy_to_user(desc->arg.buf, kaddr + offset, size); | |
1302 | kunmap(page); | |
1303 | ||
1304 | if (left) { | |
1305 | size -= left; | |
1306 | desc->error = -EFAULT; | |
1307 | } | |
1308 | success: | |
1309 | desc->count = count - size; | |
1310 | desc->written += size; | |
1311 | desc->arg.buf += size; | |
1312 | return size; | |
1313 | } | |
1314 | ||
0ceb3314 DM |
1315 | /* |
1316 | * Performs necessary checks before doing a write | |
1317 | * @iov: io vector request | |
1318 | * @nr_segs: number of segments in the iovec | |
1319 | * @count: number of bytes to write | |
1320 | * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE | |
1321 | * | |
1322 | * Adjust number of segments and amount of bytes to write (nr_segs should be | |
1323 | * properly initialized first). Returns appropriate error code that caller | |
1324 | * should return or zero in case that write should be allowed. | |
1325 | */ | |
1326 | int generic_segment_checks(const struct iovec *iov, | |
1327 | unsigned long *nr_segs, size_t *count, int access_flags) | |
1328 | { | |
1329 | unsigned long seg; | |
1330 | size_t cnt = 0; | |
1331 | for (seg = 0; seg < *nr_segs; seg++) { | |
1332 | const struct iovec *iv = &iov[seg]; | |
1333 | ||
1334 | /* | |
1335 | * If any segment has a negative length, or the cumulative | |
1336 | * length ever wraps negative then return -EINVAL. | |
1337 | */ | |
1338 | cnt += iv->iov_len; | |
1339 | if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) | |
1340 | return -EINVAL; | |
1341 | if (access_ok(access_flags, iv->iov_base, iv->iov_len)) | |
1342 | continue; | |
1343 | if (seg == 0) | |
1344 | return -EFAULT; | |
1345 | *nr_segs = seg; | |
1346 | cnt -= iv->iov_len; /* This segment is no good */ | |
1347 | break; | |
1348 | } | |
1349 | *count = cnt; | |
1350 | return 0; | |
1351 | } | |
1352 | EXPORT_SYMBOL(generic_segment_checks); | |
1353 | ||
485bb99b | 1354 | /** |
b2abacf3 | 1355 | * generic_file_aio_read - generic filesystem read routine |
485bb99b RD |
1356 | * @iocb: kernel I/O control block |
1357 | * @iov: io vector request | |
1358 | * @nr_segs: number of segments in the iovec | |
b2abacf3 | 1359 | * @pos: current file position |
485bb99b | 1360 | * |
1da177e4 LT |
1361 | * This is the "read()" routine for all filesystems |
1362 | * that can use the page cache directly. | |
1363 | */ | |
1364 | ssize_t | |
543ade1f BP |
1365 | generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
1366 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
1367 | { |
1368 | struct file *filp = iocb->ki_filp; | |
1369 | ssize_t retval; | |
66f998f6 | 1370 | unsigned long seg = 0; |
1da177e4 | 1371 | size_t count; |
543ade1f | 1372 | loff_t *ppos = &iocb->ki_pos; |
1da177e4 LT |
1373 | |
1374 | count = 0; | |
0ceb3314 DM |
1375 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); |
1376 | if (retval) | |
1377 | return retval; | |
1da177e4 LT |
1378 | |
1379 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
1380 | if (filp->f_flags & O_DIRECT) { | |
543ade1f | 1381 | loff_t size; |
1da177e4 LT |
1382 | struct address_space *mapping; |
1383 | struct inode *inode; | |
1384 | ||
1385 | mapping = filp->f_mapping; | |
1386 | inode = mapping->host; | |
1da177e4 LT |
1387 | if (!count) |
1388 | goto out; /* skip atime */ | |
1389 | size = i_size_read(inode); | |
1390 | if (pos < size) { | |
48b47c56 NP |
1391 | retval = filemap_write_and_wait_range(mapping, pos, |
1392 | pos + iov_length(iov, nr_segs) - 1); | |
a969e903 CH |
1393 | if (!retval) { |
1394 | retval = mapping->a_ops->direct_IO(READ, iocb, | |
1395 | iov, pos, nr_segs); | |
1396 | } | |
66f998f6 | 1397 | if (retval > 0) { |
1da177e4 | 1398 | *ppos = pos + retval; |
66f998f6 JB |
1399 | count -= retval; |
1400 | } | |
1401 | ||
1402 | /* | |
1403 | * Btrfs can have a short DIO read if we encounter | |
1404 | * compressed extents, so if there was an error, or if | |
1405 | * we've already read everything we wanted to, or if | |
1406 | * there was a short read because we hit EOF, go ahead | |
1407 | * and return. Otherwise fallthrough to buffered io for | |
1408 | * the rest of the read. | |
1409 | */ | |
1410 | if (retval < 0 || !count || *ppos >= size) { | |
11fa977e HD |
1411 | file_accessed(filp); |
1412 | goto out; | |
1413 | } | |
0e0bcae3 | 1414 | } |
1da177e4 LT |
1415 | } |
1416 | ||
66f998f6 | 1417 | count = retval; |
11fa977e HD |
1418 | for (seg = 0; seg < nr_segs; seg++) { |
1419 | read_descriptor_t desc; | |
66f998f6 JB |
1420 | loff_t offset = 0; |
1421 | ||
1422 | /* | |
1423 | * If we did a short DIO read we need to skip the section of the | |
1424 | * iov that we've already read data into. | |
1425 | */ | |
1426 | if (count) { | |
1427 | if (count > iov[seg].iov_len) { | |
1428 | count -= iov[seg].iov_len; | |
1429 | continue; | |
1430 | } | |
1431 | offset = count; | |
1432 | count = 0; | |
1433 | } | |
1da177e4 | 1434 | |
11fa977e | 1435 | desc.written = 0; |
66f998f6 JB |
1436 | desc.arg.buf = iov[seg].iov_base + offset; |
1437 | desc.count = iov[seg].iov_len - offset; | |
11fa977e HD |
1438 | if (desc.count == 0) |
1439 | continue; | |
1440 | desc.error = 0; | |
1441 | do_generic_file_read(filp, ppos, &desc, file_read_actor); | |
1442 | retval += desc.written; | |
1443 | if (desc.error) { | |
1444 | retval = retval ?: desc.error; | |
1445 | break; | |
1da177e4 | 1446 | } |
11fa977e HD |
1447 | if (desc.count > 0) |
1448 | break; | |
1da177e4 LT |
1449 | } |
1450 | out: | |
1451 | return retval; | |
1452 | } | |
1da177e4 LT |
1453 | EXPORT_SYMBOL(generic_file_aio_read); |
1454 | ||
1da177e4 LT |
1455 | static ssize_t |
1456 | do_readahead(struct address_space *mapping, struct file *filp, | |
57f6b96c | 1457 | pgoff_t index, unsigned long nr) |
1da177e4 LT |
1458 | { |
1459 | if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage) | |
1460 | return -EINVAL; | |
1461 | ||
f7e839dd | 1462 | force_page_cache_readahead(mapping, filp, index, nr); |
1da177e4 LT |
1463 | return 0; |
1464 | } | |
1465 | ||
6673e0c3 | 1466 | SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count) |
1da177e4 LT |
1467 | { |
1468 | ssize_t ret; | |
1469 | struct file *file; | |
1470 | ||
1471 | ret = -EBADF; | |
1472 | file = fget(fd); | |
1473 | if (file) { | |
1474 | if (file->f_mode & FMODE_READ) { | |
1475 | struct address_space *mapping = file->f_mapping; | |
57f6b96c FW |
1476 | pgoff_t start = offset >> PAGE_CACHE_SHIFT; |
1477 | pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
1478 | unsigned long len = end - start + 1; |
1479 | ret = do_readahead(mapping, file, start, len); | |
1480 | } | |
1481 | fput(file); | |
1482 | } | |
1483 | return ret; | |
1484 | } | |
6673e0c3 HC |
1485 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
1486 | asmlinkage long SyS_readahead(long fd, loff_t offset, long count) | |
1487 | { | |
1488 | return SYSC_readahead((int) fd, offset, (size_t) count); | |
1489 | } | |
1490 | SYSCALL_ALIAS(sys_readahead, SyS_readahead); | |
1491 | #endif | |
1da177e4 LT |
1492 | |
1493 | #ifdef CONFIG_MMU | |
485bb99b RD |
1494 | /** |
1495 | * page_cache_read - adds requested page to the page cache if not already there | |
1496 | * @file: file to read | |
1497 | * @offset: page index | |
1498 | * | |
1da177e4 LT |
1499 | * This adds the requested page to the page cache if it isn't already there, |
1500 | * and schedules an I/O to read in its contents from disk. | |
1501 | */ | |
920c7a5d | 1502 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1503 | { |
1504 | struct address_space *mapping = file->f_mapping; | |
1505 | struct page *page; | |
994fc28c | 1506 | int ret; |
1da177e4 | 1507 | |
994fc28c ZB |
1508 | do { |
1509 | page = page_cache_alloc_cold(mapping); | |
1510 | if (!page) | |
1511 | return -ENOMEM; | |
1512 | ||
1513 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); | |
1514 | if (ret == 0) | |
1515 | ret = mapping->a_ops->readpage(file, page); | |
1516 | else if (ret == -EEXIST) | |
1517 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1518 | |
1da177e4 | 1519 | page_cache_release(page); |
1da177e4 | 1520 | |
994fc28c ZB |
1521 | } while (ret == AOP_TRUNCATED_PAGE); |
1522 | ||
1523 | return ret; | |
1da177e4 LT |
1524 | } |
1525 | ||
1526 | #define MMAP_LOTSAMISS (100) | |
1527 | ||
ef00e08e LT |
1528 | /* |
1529 | * Synchronous readahead happens when we don't even find | |
1530 | * a page in the page cache at all. | |
1531 | */ | |
1532 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1533 | struct file_ra_state *ra, | |
1534 | struct file *file, | |
1535 | pgoff_t offset) | |
1536 | { | |
1537 | unsigned long ra_pages; | |
1538 | struct address_space *mapping = file->f_mapping; | |
1539 | ||
1540 | /* If we don't want any read-ahead, don't bother */ | |
1541 | if (VM_RandomReadHint(vma)) | |
1542 | return; | |
1543 | ||
70ac23cf WF |
1544 | if (VM_SequentialReadHint(vma) || |
1545 | offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) { | |
7ffc59b4 WF |
1546 | page_cache_sync_readahead(mapping, ra, file, offset, |
1547 | ra->ra_pages); | |
ef00e08e LT |
1548 | return; |
1549 | } | |
1550 | ||
1551 | if (ra->mmap_miss < INT_MAX) | |
1552 | ra->mmap_miss++; | |
1553 | ||
1554 | /* | |
1555 | * Do we miss much more than hit in this file? If so, | |
1556 | * stop bothering with read-ahead. It will only hurt. | |
1557 | */ | |
1558 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1559 | return; | |
1560 | ||
d30a1100 WF |
1561 | /* |
1562 | * mmap read-around | |
1563 | */ | |
ef00e08e LT |
1564 | ra_pages = max_sane_readahead(ra->ra_pages); |
1565 | if (ra_pages) { | |
d30a1100 WF |
1566 | ra->start = max_t(long, 0, offset - ra_pages/2); |
1567 | ra->size = ra_pages; | |
1568 | ra->async_size = 0; | |
1569 | ra_submit(ra, mapping, file); | |
ef00e08e LT |
1570 | } |
1571 | } | |
1572 | ||
1573 | /* | |
1574 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1575 | * so we want to possibly extend the readahead further.. | |
1576 | */ | |
1577 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1578 | struct file_ra_state *ra, | |
1579 | struct file *file, | |
1580 | struct page *page, | |
1581 | pgoff_t offset) | |
1582 | { | |
1583 | struct address_space *mapping = file->f_mapping; | |
1584 | ||
1585 | /* If we don't want any read-ahead, don't bother */ | |
1586 | if (VM_RandomReadHint(vma)) | |
1587 | return; | |
1588 | if (ra->mmap_miss > 0) | |
1589 | ra->mmap_miss--; | |
1590 | if (PageReadahead(page)) | |
2fad6f5d WF |
1591 | page_cache_async_readahead(mapping, ra, file, |
1592 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1593 | } |
1594 | ||
485bb99b | 1595 | /** |
54cb8821 | 1596 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1597 | * @vma: vma in which the fault was taken |
1598 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1599 | * |
54cb8821 | 1600 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1601 | * mapped memory region to read in file data during a page fault. |
1602 | * | |
1603 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1604 | * it in the page cache, and handles the special cases reasonably without | |
1605 | * having a lot of duplicated code. | |
1606 | */ | |
d0217ac0 | 1607 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1608 | { |
1609 | int error; | |
54cb8821 | 1610 | struct file *file = vma->vm_file; |
1da177e4 LT |
1611 | struct address_space *mapping = file->f_mapping; |
1612 | struct file_ra_state *ra = &file->f_ra; | |
1613 | struct inode *inode = mapping->host; | |
ef00e08e | 1614 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 1615 | struct page *page; |
2004dc8e | 1616 | pgoff_t size; |
83c54070 | 1617 | int ret = 0; |
1da177e4 | 1618 | |
1da177e4 | 1619 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1620 | if (offset >= size) |
5307cc1a | 1621 | return VM_FAULT_SIGBUS; |
1da177e4 | 1622 | |
1da177e4 LT |
1623 | /* |
1624 | * Do we have something in the page cache already? | |
1625 | */ | |
ef00e08e LT |
1626 | page = find_get_page(mapping, offset); |
1627 | if (likely(page)) { | |
1da177e4 | 1628 | /* |
ef00e08e LT |
1629 | * We found the page, so try async readahead before |
1630 | * waiting for the lock. | |
1da177e4 | 1631 | */ |
ef00e08e | 1632 | do_async_mmap_readahead(vma, ra, file, page, offset); |
ef00e08e LT |
1633 | } else { |
1634 | /* No page in the page cache at all */ | |
1635 | do_sync_mmap_readahead(vma, ra, file, offset); | |
1636 | count_vm_event(PGMAJFAULT); | |
1637 | ret = VM_FAULT_MAJOR; | |
1638 | retry_find: | |
b522c94d | 1639 | page = find_get_page(mapping, offset); |
1da177e4 LT |
1640 | if (!page) |
1641 | goto no_cached_page; | |
1642 | } | |
1643 | ||
d88c0922 ML |
1644 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
1645 | page_cache_release(page); | |
d065bd81 | 1646 | return ret | VM_FAULT_RETRY; |
d88c0922 | 1647 | } |
b522c94d ML |
1648 | |
1649 | /* Did it get truncated? */ | |
1650 | if (unlikely(page->mapping != mapping)) { | |
1651 | unlock_page(page); | |
1652 | put_page(page); | |
1653 | goto retry_find; | |
1654 | } | |
1655 | VM_BUG_ON(page->index != offset); | |
1656 | ||
1da177e4 | 1657 | /* |
d00806b1 NP |
1658 | * We have a locked page in the page cache, now we need to check |
1659 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1660 | */ |
d00806b1 | 1661 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1662 | goto page_not_uptodate; |
1663 | ||
ef00e08e LT |
1664 | /* |
1665 | * Found the page and have a reference on it. | |
1666 | * We must recheck i_size under page lock. | |
1667 | */ | |
d00806b1 | 1668 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1669 | if (unlikely(offset >= size)) { |
d00806b1 | 1670 | unlock_page(page); |
745ad48e | 1671 | page_cache_release(page); |
5307cc1a | 1672 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1673 | } |
1674 | ||
ef00e08e | 1675 | ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT; |
d0217ac0 | 1676 | vmf->page = page; |
83c54070 | 1677 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 1678 | |
1da177e4 LT |
1679 | no_cached_page: |
1680 | /* | |
1681 | * We're only likely to ever get here if MADV_RANDOM is in | |
1682 | * effect. | |
1683 | */ | |
ef00e08e | 1684 | error = page_cache_read(file, offset); |
1da177e4 LT |
1685 | |
1686 | /* | |
1687 | * The page we want has now been added to the page cache. | |
1688 | * In the unlikely event that someone removed it in the | |
1689 | * meantime, we'll just come back here and read it again. | |
1690 | */ | |
1691 | if (error >= 0) | |
1692 | goto retry_find; | |
1693 | ||
1694 | /* | |
1695 | * An error return from page_cache_read can result if the | |
1696 | * system is low on memory, or a problem occurs while trying | |
1697 | * to schedule I/O. | |
1698 | */ | |
1699 | if (error == -ENOMEM) | |
d0217ac0 NP |
1700 | return VM_FAULT_OOM; |
1701 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1702 | |
1703 | page_not_uptodate: | |
1da177e4 LT |
1704 | /* |
1705 | * Umm, take care of errors if the page isn't up-to-date. | |
1706 | * Try to re-read it _once_. We do this synchronously, | |
1707 | * because there really aren't any performance issues here | |
1708 | * and we need to check for errors. | |
1709 | */ | |
1da177e4 | 1710 | ClearPageError(page); |
994fc28c | 1711 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
1712 | if (!error) { |
1713 | wait_on_page_locked(page); | |
1714 | if (!PageUptodate(page)) | |
1715 | error = -EIO; | |
1716 | } | |
d00806b1 NP |
1717 | page_cache_release(page); |
1718 | ||
1719 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 1720 | goto retry_find; |
1da177e4 | 1721 | |
d00806b1 | 1722 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 1723 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 1724 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
1725 | } |
1726 | EXPORT_SYMBOL(filemap_fault); | |
1727 | ||
f0f37e2f | 1728 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 1729 | .fault = filemap_fault, |
1da177e4 LT |
1730 | }; |
1731 | ||
1732 | /* This is used for a general mmap of a disk file */ | |
1733 | ||
1734 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1735 | { | |
1736 | struct address_space *mapping = file->f_mapping; | |
1737 | ||
1738 | if (!mapping->a_ops->readpage) | |
1739 | return -ENOEXEC; | |
1740 | file_accessed(file); | |
1741 | vma->vm_ops = &generic_file_vm_ops; | |
d0217ac0 | 1742 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
1743 | return 0; |
1744 | } | |
1da177e4 LT |
1745 | |
1746 | /* | |
1747 | * This is for filesystems which do not implement ->writepage. | |
1748 | */ | |
1749 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
1750 | { | |
1751 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
1752 | return -EINVAL; | |
1753 | return generic_file_mmap(file, vma); | |
1754 | } | |
1755 | #else | |
1756 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1757 | { | |
1758 | return -ENOSYS; | |
1759 | } | |
1760 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
1761 | { | |
1762 | return -ENOSYS; | |
1763 | } | |
1764 | #endif /* CONFIG_MMU */ | |
1765 | ||
1766 | EXPORT_SYMBOL(generic_file_mmap); | |
1767 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
1768 | ||
6fe6900e | 1769 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 1770 | pgoff_t index, |
1da177e4 | 1771 | int (*filler)(void *,struct page*), |
0531b2aa LT |
1772 | void *data, |
1773 | gfp_t gfp) | |
1da177e4 | 1774 | { |
eb2be189 | 1775 | struct page *page; |
1da177e4 LT |
1776 | int err; |
1777 | repeat: | |
1778 | page = find_get_page(mapping, index); | |
1779 | if (!page) { | |
0531b2aa | 1780 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
1781 | if (!page) |
1782 | return ERR_PTR(-ENOMEM); | |
1783 | err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL); | |
1784 | if (unlikely(err)) { | |
1785 | page_cache_release(page); | |
1786 | if (err == -EEXIST) | |
1787 | goto repeat; | |
1da177e4 | 1788 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
1789 | return ERR_PTR(err); |
1790 | } | |
1da177e4 LT |
1791 | err = filler(data, page); |
1792 | if (err < 0) { | |
1793 | page_cache_release(page); | |
1794 | page = ERR_PTR(err); | |
1795 | } | |
1796 | } | |
1da177e4 LT |
1797 | return page; |
1798 | } | |
1799 | ||
0531b2aa | 1800 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 1801 | pgoff_t index, |
1da177e4 | 1802 | int (*filler)(void *,struct page*), |
0531b2aa LT |
1803 | void *data, |
1804 | gfp_t gfp) | |
1805 | ||
1da177e4 LT |
1806 | { |
1807 | struct page *page; | |
1808 | int err; | |
1809 | ||
1810 | retry: | |
0531b2aa | 1811 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 1812 | if (IS_ERR(page)) |
c855ff37 | 1813 | return page; |
1da177e4 LT |
1814 | if (PageUptodate(page)) |
1815 | goto out; | |
1816 | ||
1817 | lock_page(page); | |
1818 | if (!page->mapping) { | |
1819 | unlock_page(page); | |
1820 | page_cache_release(page); | |
1821 | goto retry; | |
1822 | } | |
1823 | if (PageUptodate(page)) { | |
1824 | unlock_page(page); | |
1825 | goto out; | |
1826 | } | |
1827 | err = filler(data, page); | |
1828 | if (err < 0) { | |
1829 | page_cache_release(page); | |
c855ff37 | 1830 | return ERR_PTR(err); |
1da177e4 | 1831 | } |
c855ff37 | 1832 | out: |
6fe6900e NP |
1833 | mark_page_accessed(page); |
1834 | return page; | |
1835 | } | |
0531b2aa LT |
1836 | |
1837 | /** | |
1838 | * read_cache_page_async - read into page cache, fill it if needed | |
1839 | * @mapping: the page's address_space | |
1840 | * @index: the page index | |
1841 | * @filler: function to perform the read | |
1842 | * @data: destination for read data | |
1843 | * | |
1844 | * Same as read_cache_page, but don't wait for page to become unlocked | |
1845 | * after submitting it to the filler. | |
1846 | * | |
1847 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1848 | * not set, try to fill the page but don't wait for it to become unlocked. | |
1849 | * | |
1850 | * If the page does not get brought uptodate, return -EIO. | |
1851 | */ | |
1852 | struct page *read_cache_page_async(struct address_space *mapping, | |
1853 | pgoff_t index, | |
1854 | int (*filler)(void *,struct page*), | |
1855 | void *data) | |
1856 | { | |
1857 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
1858 | } | |
6fe6900e NP |
1859 | EXPORT_SYMBOL(read_cache_page_async); |
1860 | ||
0531b2aa LT |
1861 | static struct page *wait_on_page_read(struct page *page) |
1862 | { | |
1863 | if (!IS_ERR(page)) { | |
1864 | wait_on_page_locked(page); | |
1865 | if (!PageUptodate(page)) { | |
1866 | page_cache_release(page); | |
1867 | page = ERR_PTR(-EIO); | |
1868 | } | |
1869 | } | |
1870 | return page; | |
1871 | } | |
1872 | ||
1873 | /** | |
1874 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
1875 | * @mapping: the page's address_space | |
1876 | * @index: the page index | |
1877 | * @gfp: the page allocator flags to use if allocating | |
1878 | * | |
1879 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
1880 | * any new page allocations done using the specified allocation flags. Note | |
1881 | * that the Radix tree operations will still use GFP_KERNEL, so you can't | |
1882 | * expect to do this atomically or anything like that - but you can pass in | |
1883 | * other page requirements. | |
1884 | * | |
1885 | * If the page does not get brought uptodate, return -EIO. | |
1886 | */ | |
1887 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
1888 | pgoff_t index, | |
1889 | gfp_t gfp) | |
1890 | { | |
1891 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
1892 | ||
1893 | return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp)); | |
1894 | } | |
1895 | EXPORT_SYMBOL(read_cache_page_gfp); | |
1896 | ||
6fe6900e NP |
1897 | /** |
1898 | * read_cache_page - read into page cache, fill it if needed | |
1899 | * @mapping: the page's address_space | |
1900 | * @index: the page index | |
1901 | * @filler: function to perform the read | |
1902 | * @data: destination for read data | |
1903 | * | |
1904 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1905 | * not set, try to fill the page then wait for it to become unlocked. | |
1906 | * | |
1907 | * If the page does not get brought uptodate, return -EIO. | |
1908 | */ | |
1909 | struct page *read_cache_page(struct address_space *mapping, | |
57f6b96c | 1910 | pgoff_t index, |
6fe6900e NP |
1911 | int (*filler)(void *,struct page*), |
1912 | void *data) | |
1913 | { | |
0531b2aa | 1914 | return wait_on_page_read(read_cache_page_async(mapping, index, filler, data)); |
1da177e4 | 1915 | } |
1da177e4 LT |
1916 | EXPORT_SYMBOL(read_cache_page); |
1917 | ||
1da177e4 LT |
1918 | /* |
1919 | * The logic we want is | |
1920 | * | |
1921 | * if suid or (sgid and xgrp) | |
1922 | * remove privs | |
1923 | */ | |
01de85e0 | 1924 | int should_remove_suid(struct dentry *dentry) |
1da177e4 LT |
1925 | { |
1926 | mode_t mode = dentry->d_inode->i_mode; | |
1927 | int kill = 0; | |
1da177e4 LT |
1928 | |
1929 | /* suid always must be killed */ | |
1930 | if (unlikely(mode & S_ISUID)) | |
1931 | kill = ATTR_KILL_SUID; | |
1932 | ||
1933 | /* | |
1934 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1935 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1936 | */ | |
1937 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1938 | kill |= ATTR_KILL_SGID; | |
1939 | ||
7f5ff766 | 1940 | if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) |
01de85e0 | 1941 | return kill; |
1da177e4 | 1942 | |
01de85e0 JA |
1943 | return 0; |
1944 | } | |
d23a147b | 1945 | EXPORT_SYMBOL(should_remove_suid); |
01de85e0 | 1946 | |
7f3d4ee1 | 1947 | static int __remove_suid(struct dentry *dentry, int kill) |
01de85e0 JA |
1948 | { |
1949 | struct iattr newattrs; | |
1950 | ||
1951 | newattrs.ia_valid = ATTR_FORCE | kill; | |
1952 | return notify_change(dentry, &newattrs); | |
1953 | } | |
1954 | ||
2f1936b8 | 1955 | int file_remove_suid(struct file *file) |
01de85e0 | 1956 | { |
2f1936b8 | 1957 | struct dentry *dentry = file->f_path.dentry; |
b5376771 SH |
1958 | int killsuid = should_remove_suid(dentry); |
1959 | int killpriv = security_inode_need_killpriv(dentry); | |
1960 | int error = 0; | |
01de85e0 | 1961 | |
b5376771 SH |
1962 | if (killpriv < 0) |
1963 | return killpriv; | |
1964 | if (killpriv) | |
1965 | error = security_inode_killpriv(dentry); | |
1966 | if (!error && killsuid) | |
1967 | error = __remove_suid(dentry, killsuid); | |
01de85e0 | 1968 | |
b5376771 | 1969 | return error; |
1da177e4 | 1970 | } |
2f1936b8 | 1971 | EXPORT_SYMBOL(file_remove_suid); |
1da177e4 | 1972 | |
2f718ffc | 1973 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
1da177e4 LT |
1974 | const struct iovec *iov, size_t base, size_t bytes) |
1975 | { | |
f1800536 | 1976 | size_t copied = 0, left = 0; |
1da177e4 LT |
1977 | |
1978 | while (bytes) { | |
1979 | char __user *buf = iov->iov_base + base; | |
1980 | int copy = min(bytes, iov->iov_len - base); | |
1981 | ||
1982 | base = 0; | |
f1800536 | 1983 | left = __copy_from_user_inatomic(vaddr, buf, copy); |
1da177e4 LT |
1984 | copied += copy; |
1985 | bytes -= copy; | |
1986 | vaddr += copy; | |
1987 | iov++; | |
1988 | ||
01408c49 | 1989 | if (unlikely(left)) |
1da177e4 | 1990 | break; |
1da177e4 LT |
1991 | } |
1992 | return copied - left; | |
1993 | } | |
1994 | ||
2f718ffc NP |
1995 | /* |
1996 | * Copy as much as we can into the page and return the number of bytes which | |
af901ca1 | 1997 | * were successfully copied. If a fault is encountered then return the number of |
2f718ffc NP |
1998 | * bytes which were copied. |
1999 | */ | |
2000 | size_t iov_iter_copy_from_user_atomic(struct page *page, | |
2001 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
2002 | { | |
2003 | char *kaddr; | |
2004 | size_t copied; | |
2005 | ||
2006 | BUG_ON(!in_atomic()); | |
2007 | kaddr = kmap_atomic(page, KM_USER0); | |
2008 | if (likely(i->nr_segs == 1)) { | |
2009 | int left; | |
2010 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 2011 | left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); |
2f718ffc NP |
2012 | copied = bytes - left; |
2013 | } else { | |
2014 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2015 | i->iov, i->iov_offset, bytes); | |
2016 | } | |
2017 | kunmap_atomic(kaddr, KM_USER0); | |
2018 | ||
2019 | return copied; | |
2020 | } | |
89e10787 | 2021 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); |
2f718ffc NP |
2022 | |
2023 | /* | |
2024 | * This has the same sideeffects and return value as | |
2025 | * iov_iter_copy_from_user_atomic(). | |
2026 | * The difference is that it attempts to resolve faults. | |
2027 | * Page must not be locked. | |
2028 | */ | |
2029 | size_t iov_iter_copy_from_user(struct page *page, | |
2030 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
2031 | { | |
2032 | char *kaddr; | |
2033 | size_t copied; | |
2034 | ||
2035 | kaddr = kmap(page); | |
2036 | if (likely(i->nr_segs == 1)) { | |
2037 | int left; | |
2038 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 2039 | left = __copy_from_user(kaddr + offset, buf, bytes); |
2f718ffc NP |
2040 | copied = bytes - left; |
2041 | } else { | |
2042 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2043 | i->iov, i->iov_offset, bytes); | |
2044 | } | |
2045 | kunmap(page); | |
2046 | return copied; | |
2047 | } | |
89e10787 | 2048 | EXPORT_SYMBOL(iov_iter_copy_from_user); |
2f718ffc | 2049 | |
f7009264 | 2050 | void iov_iter_advance(struct iov_iter *i, size_t bytes) |
2f718ffc | 2051 | { |
f7009264 NP |
2052 | BUG_ON(i->count < bytes); |
2053 | ||
2f718ffc NP |
2054 | if (likely(i->nr_segs == 1)) { |
2055 | i->iov_offset += bytes; | |
f7009264 | 2056 | i->count -= bytes; |
2f718ffc NP |
2057 | } else { |
2058 | const struct iovec *iov = i->iov; | |
2059 | size_t base = i->iov_offset; | |
2060 | ||
124d3b70 NP |
2061 | /* |
2062 | * The !iov->iov_len check ensures we skip over unlikely | |
f7009264 | 2063 | * zero-length segments (without overruning the iovec). |
124d3b70 | 2064 | */ |
94ad374a | 2065 | while (bytes || unlikely(i->count && !iov->iov_len)) { |
f7009264 | 2066 | int copy; |
2f718ffc | 2067 | |
f7009264 NP |
2068 | copy = min(bytes, iov->iov_len - base); |
2069 | BUG_ON(!i->count || i->count < copy); | |
2070 | i->count -= copy; | |
2f718ffc NP |
2071 | bytes -= copy; |
2072 | base += copy; | |
2073 | if (iov->iov_len == base) { | |
2074 | iov++; | |
2075 | base = 0; | |
2076 | } | |
2077 | } | |
2078 | i->iov = iov; | |
2079 | i->iov_offset = base; | |
2080 | } | |
2081 | } | |
89e10787 | 2082 | EXPORT_SYMBOL(iov_iter_advance); |
2f718ffc | 2083 | |
afddba49 NP |
2084 | /* |
2085 | * Fault in the first iovec of the given iov_iter, to a maximum length | |
2086 | * of bytes. Returns 0 on success, or non-zero if the memory could not be | |
2087 | * accessed (ie. because it is an invalid address). | |
2088 | * | |
2089 | * writev-intensive code may want this to prefault several iovecs -- that | |
2090 | * would be possible (callers must not rely on the fact that _only_ the | |
2091 | * first iovec will be faulted with the current implementation). | |
2092 | */ | |
2093 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) | |
2f718ffc | 2094 | { |
2f718ffc | 2095 | char __user *buf = i->iov->iov_base + i->iov_offset; |
afddba49 NP |
2096 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); |
2097 | return fault_in_pages_readable(buf, bytes); | |
2f718ffc | 2098 | } |
89e10787 | 2099 | EXPORT_SYMBOL(iov_iter_fault_in_readable); |
2f718ffc NP |
2100 | |
2101 | /* | |
2102 | * Return the count of just the current iov_iter segment. | |
2103 | */ | |
2104 | size_t iov_iter_single_seg_count(struct iov_iter *i) | |
2105 | { | |
2106 | const struct iovec *iov = i->iov; | |
2107 | if (i->nr_segs == 1) | |
2108 | return i->count; | |
2109 | else | |
2110 | return min(i->count, iov->iov_len - i->iov_offset); | |
2111 | } | |
89e10787 | 2112 | EXPORT_SYMBOL(iov_iter_single_seg_count); |
2f718ffc | 2113 | |
1da177e4 LT |
2114 | /* |
2115 | * Performs necessary checks before doing a write | |
2116 | * | |
485bb99b | 2117 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2118 | * Returns appropriate error code that caller should return or |
2119 | * zero in case that write should be allowed. | |
2120 | */ | |
2121 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) | |
2122 | { | |
2123 | struct inode *inode = file->f_mapping->host; | |
59e99e5b | 2124 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
1da177e4 LT |
2125 | |
2126 | if (unlikely(*pos < 0)) | |
2127 | return -EINVAL; | |
2128 | ||
1da177e4 LT |
2129 | if (!isblk) { |
2130 | /* FIXME: this is for backwards compatibility with 2.4 */ | |
2131 | if (file->f_flags & O_APPEND) | |
2132 | *pos = i_size_read(inode); | |
2133 | ||
2134 | if (limit != RLIM_INFINITY) { | |
2135 | if (*pos >= limit) { | |
2136 | send_sig(SIGXFSZ, current, 0); | |
2137 | return -EFBIG; | |
2138 | } | |
2139 | if (*count > limit - (typeof(limit))*pos) { | |
2140 | *count = limit - (typeof(limit))*pos; | |
2141 | } | |
2142 | } | |
2143 | } | |
2144 | ||
2145 | /* | |
2146 | * LFS rule | |
2147 | */ | |
2148 | if (unlikely(*pos + *count > MAX_NON_LFS && | |
2149 | !(file->f_flags & O_LARGEFILE))) { | |
2150 | if (*pos >= MAX_NON_LFS) { | |
1da177e4 LT |
2151 | return -EFBIG; |
2152 | } | |
2153 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { | |
2154 | *count = MAX_NON_LFS - (unsigned long)*pos; | |
2155 | } | |
2156 | } | |
2157 | ||
2158 | /* | |
2159 | * Are we about to exceed the fs block limit ? | |
2160 | * | |
2161 | * If we have written data it becomes a short write. If we have | |
2162 | * exceeded without writing data we send a signal and return EFBIG. | |
2163 | * Linus frestrict idea will clean these up nicely.. | |
2164 | */ | |
2165 | if (likely(!isblk)) { | |
2166 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { | |
2167 | if (*count || *pos > inode->i_sb->s_maxbytes) { | |
1da177e4 LT |
2168 | return -EFBIG; |
2169 | } | |
2170 | /* zero-length writes at ->s_maxbytes are OK */ | |
2171 | } | |
2172 | ||
2173 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) | |
2174 | *count = inode->i_sb->s_maxbytes - *pos; | |
2175 | } else { | |
9361401e | 2176 | #ifdef CONFIG_BLOCK |
1da177e4 LT |
2177 | loff_t isize; |
2178 | if (bdev_read_only(I_BDEV(inode))) | |
2179 | return -EPERM; | |
2180 | isize = i_size_read(inode); | |
2181 | if (*pos >= isize) { | |
2182 | if (*count || *pos > isize) | |
2183 | return -ENOSPC; | |
2184 | } | |
2185 | ||
2186 | if (*pos + *count > isize) | |
2187 | *count = isize - *pos; | |
9361401e DH |
2188 | #else |
2189 | return -EPERM; | |
2190 | #endif | |
1da177e4 LT |
2191 | } |
2192 | return 0; | |
2193 | } | |
2194 | EXPORT_SYMBOL(generic_write_checks); | |
2195 | ||
afddba49 NP |
2196 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2197 | loff_t pos, unsigned len, unsigned flags, | |
2198 | struct page **pagep, void **fsdata) | |
2199 | { | |
2200 | const struct address_space_operations *aops = mapping->a_ops; | |
2201 | ||
4e02ed4b | 2202 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2203 | pagep, fsdata); |
afddba49 NP |
2204 | } |
2205 | EXPORT_SYMBOL(pagecache_write_begin); | |
2206 | ||
2207 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2208 | loff_t pos, unsigned len, unsigned copied, | |
2209 | struct page *page, void *fsdata) | |
2210 | { | |
2211 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2212 | |
4e02ed4b NP |
2213 | mark_page_accessed(page); |
2214 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); | |
afddba49 NP |
2215 | } |
2216 | EXPORT_SYMBOL(pagecache_write_end); | |
2217 | ||
1da177e4 LT |
2218 | ssize_t |
2219 | generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, | |
2220 | unsigned long *nr_segs, loff_t pos, loff_t *ppos, | |
2221 | size_t count, size_t ocount) | |
2222 | { | |
2223 | struct file *file = iocb->ki_filp; | |
2224 | struct address_space *mapping = file->f_mapping; | |
2225 | struct inode *inode = mapping->host; | |
2226 | ssize_t written; | |
a969e903 CH |
2227 | size_t write_len; |
2228 | pgoff_t end; | |
1da177e4 LT |
2229 | |
2230 | if (count != ocount) | |
2231 | *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); | |
2232 | ||
a969e903 CH |
2233 | write_len = iov_length(iov, *nr_segs); |
2234 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; | |
a969e903 | 2235 | |
48b47c56 | 2236 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2237 | if (written) |
2238 | goto out; | |
2239 | ||
2240 | /* | |
2241 | * After a write we want buffered reads to be sure to go to disk to get | |
2242 | * the new data. We invalidate clean cached page from the region we're | |
2243 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2244 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2245 | */ |
2246 | if (mapping->nrpages) { | |
2247 | written = invalidate_inode_pages2_range(mapping, | |
2248 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2249 | /* |
2250 | * If a page can not be invalidated, return 0 to fall back | |
2251 | * to buffered write. | |
2252 | */ | |
2253 | if (written) { | |
2254 | if (written == -EBUSY) | |
2255 | return 0; | |
a969e903 | 2256 | goto out; |
6ccfa806 | 2257 | } |
a969e903 CH |
2258 | } |
2259 | ||
2260 | written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs); | |
2261 | ||
2262 | /* | |
2263 | * Finally, try again to invalidate clean pages which might have been | |
2264 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2265 | * if the source of the write was an mmap'ed region of the file | |
2266 | * we're writing. Either one is a pretty crazy thing to do, | |
2267 | * so we don't support it 100%. If this invalidation | |
2268 | * fails, tough, the write still worked... | |
2269 | */ | |
2270 | if (mapping->nrpages) { | |
2271 | invalidate_inode_pages2_range(mapping, | |
2272 | pos >> PAGE_CACHE_SHIFT, end); | |
2273 | } | |
2274 | ||
1da177e4 | 2275 | if (written > 0) { |
0116651c NK |
2276 | pos += written; |
2277 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { | |
2278 | i_size_write(inode, pos); | |
1da177e4 LT |
2279 | mark_inode_dirty(inode); |
2280 | } | |
0116651c | 2281 | *ppos = pos; |
1da177e4 | 2282 | } |
a969e903 | 2283 | out: |
1da177e4 LT |
2284 | return written; |
2285 | } | |
2286 | EXPORT_SYMBOL(generic_file_direct_write); | |
2287 | ||
eb2be189 NP |
2288 | /* |
2289 | * Find or create a page at the given pagecache position. Return the locked | |
2290 | * page. This function is specifically for buffered writes. | |
2291 | */ | |
54566b2c NP |
2292 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2293 | pgoff_t index, unsigned flags) | |
eb2be189 NP |
2294 | { |
2295 | int status; | |
2296 | struct page *page; | |
54566b2c NP |
2297 | gfp_t gfp_notmask = 0; |
2298 | if (flags & AOP_FLAG_NOFS) | |
2299 | gfp_notmask = __GFP_FS; | |
eb2be189 NP |
2300 | repeat: |
2301 | page = find_lock_page(mapping, index); | |
c585a267 | 2302 | if (page) |
eb2be189 NP |
2303 | return page; |
2304 | ||
54566b2c | 2305 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask); |
eb2be189 NP |
2306 | if (!page) |
2307 | return NULL; | |
54566b2c NP |
2308 | status = add_to_page_cache_lru(page, mapping, index, |
2309 | GFP_KERNEL & ~gfp_notmask); | |
eb2be189 NP |
2310 | if (unlikely(status)) { |
2311 | page_cache_release(page); | |
2312 | if (status == -EEXIST) | |
2313 | goto repeat; | |
2314 | return NULL; | |
2315 | } | |
2316 | return page; | |
2317 | } | |
54566b2c | 2318 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2319 | |
afddba49 NP |
2320 | static ssize_t generic_perform_write(struct file *file, |
2321 | struct iov_iter *i, loff_t pos) | |
2322 | { | |
2323 | struct address_space *mapping = file->f_mapping; | |
2324 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2325 | long status = 0; | |
2326 | ssize_t written = 0; | |
674b892e NP |
2327 | unsigned int flags = 0; |
2328 | ||
2329 | /* | |
2330 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2331 | */ | |
2332 | if (segment_eq(get_fs(), KERNEL_DS)) | |
2333 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
afddba49 NP |
2334 | |
2335 | do { | |
2336 | struct page *page; | |
afddba49 NP |
2337 | unsigned long offset; /* Offset into pagecache page */ |
2338 | unsigned long bytes; /* Bytes to write to page */ | |
2339 | size_t copied; /* Bytes copied from user */ | |
2340 | void *fsdata; | |
2341 | ||
2342 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2343 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2344 | iov_iter_count(i)); | |
2345 | ||
2346 | again: | |
2347 | ||
2348 | /* | |
2349 | * Bring in the user page that we will copy from _first_. | |
2350 | * Otherwise there's a nasty deadlock on copying from the | |
2351 | * same page as we're writing to, without it being marked | |
2352 | * up-to-date. | |
2353 | * | |
2354 | * Not only is this an optimisation, but it is also required | |
2355 | * to check that the address is actually valid, when atomic | |
2356 | * usercopies are used, below. | |
2357 | */ | |
2358 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2359 | status = -EFAULT; | |
2360 | break; | |
2361 | } | |
2362 | ||
674b892e | 2363 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 NP |
2364 | &page, &fsdata); |
2365 | if (unlikely(status)) | |
2366 | break; | |
2367 | ||
931e80e4 | 2368 | if (mapping_writably_mapped(mapping)) |
2369 | flush_dcache_page(page); | |
2370 | ||
afddba49 NP |
2371 | pagefault_disable(); |
2372 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); | |
2373 | pagefault_enable(); | |
2374 | flush_dcache_page(page); | |
2375 | ||
c8236db9 | 2376 | mark_page_accessed(page); |
afddba49 NP |
2377 | status = a_ops->write_end(file, mapping, pos, bytes, copied, |
2378 | page, fsdata); | |
2379 | if (unlikely(status < 0)) | |
2380 | break; | |
2381 | copied = status; | |
2382 | ||
2383 | cond_resched(); | |
2384 | ||
124d3b70 | 2385 | iov_iter_advance(i, copied); |
afddba49 NP |
2386 | if (unlikely(copied == 0)) { |
2387 | /* | |
2388 | * If we were unable to copy any data at all, we must | |
2389 | * fall back to a single segment length write. | |
2390 | * | |
2391 | * If we didn't fallback here, we could livelock | |
2392 | * because not all segments in the iov can be copied at | |
2393 | * once without a pagefault. | |
2394 | */ | |
2395 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2396 | iov_iter_single_seg_count(i)); | |
2397 | goto again; | |
2398 | } | |
afddba49 NP |
2399 | pos += copied; |
2400 | written += copied; | |
2401 | ||
2402 | balance_dirty_pages_ratelimited(mapping); | |
2403 | ||
2404 | } while (iov_iter_count(i)); | |
2405 | ||
2406 | return written ? written : status; | |
2407 | } | |
2408 | ||
2409 | ssize_t | |
2410 | generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, | |
2411 | unsigned long nr_segs, loff_t pos, loff_t *ppos, | |
2412 | size_t count, ssize_t written) | |
2413 | { | |
2414 | struct file *file = iocb->ki_filp; | |
afddba49 NP |
2415 | ssize_t status; |
2416 | struct iov_iter i; | |
2417 | ||
2418 | iov_iter_init(&i, iov, nr_segs, count, written); | |
4e02ed4b | 2419 | status = generic_perform_write(file, &i, pos); |
1da177e4 | 2420 | |
1da177e4 | 2421 | if (likely(status >= 0)) { |
afddba49 NP |
2422 | written += status; |
2423 | *ppos = pos + status; | |
1da177e4 LT |
2424 | } |
2425 | ||
1da177e4 LT |
2426 | return written ? written : status; |
2427 | } | |
2428 | EXPORT_SYMBOL(generic_file_buffered_write); | |
2429 | ||
e4dd9de3 JK |
2430 | /** |
2431 | * __generic_file_aio_write - write data to a file | |
2432 | * @iocb: IO state structure (file, offset, etc.) | |
2433 | * @iov: vector with data to write | |
2434 | * @nr_segs: number of segments in the vector | |
2435 | * @ppos: position where to write | |
2436 | * | |
2437 | * This function does all the work needed for actually writing data to a | |
2438 | * file. It does all basic checks, removes SUID from the file, updates | |
2439 | * modification times and calls proper subroutines depending on whether we | |
2440 | * do direct IO or a standard buffered write. | |
2441 | * | |
2442 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2443 | * object which does not need locking at all. | |
2444 | * | |
2445 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2446 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2447 | * avoid syncing under i_mutex. | |
2448 | */ | |
2449 | ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
2450 | unsigned long nr_segs, loff_t *ppos) | |
1da177e4 LT |
2451 | { |
2452 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2453 | struct address_space * mapping = file->f_mapping; |
1da177e4 LT |
2454 | size_t ocount; /* original count */ |
2455 | size_t count; /* after file limit checks */ | |
2456 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2457 | loff_t pos; |
2458 | ssize_t written; | |
2459 | ssize_t err; | |
2460 | ||
2461 | ocount = 0; | |
0ceb3314 DM |
2462 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
2463 | if (err) | |
2464 | return err; | |
1da177e4 LT |
2465 | |
2466 | count = ocount; | |
2467 | pos = *ppos; | |
2468 | ||
2469 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
2470 | ||
2471 | /* We can write back this queue in page reclaim */ | |
2472 | current->backing_dev_info = mapping->backing_dev_info; | |
2473 | written = 0; | |
2474 | ||
2475 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
2476 | if (err) | |
2477 | goto out; | |
2478 | ||
2479 | if (count == 0) | |
2480 | goto out; | |
2481 | ||
2f1936b8 | 2482 | err = file_remove_suid(file); |
1da177e4 LT |
2483 | if (err) |
2484 | goto out; | |
2485 | ||
870f4817 | 2486 | file_update_time(file); |
1da177e4 LT |
2487 | |
2488 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
2489 | if (unlikely(file->f_flags & O_DIRECT)) { | |
fb5527e6 JM |
2490 | loff_t endbyte; |
2491 | ssize_t written_buffered; | |
2492 | ||
2493 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, | |
2494 | ppos, count, ocount); | |
1da177e4 LT |
2495 | if (written < 0 || written == count) |
2496 | goto out; | |
2497 | /* | |
2498 | * direct-io write to a hole: fall through to buffered I/O | |
2499 | * for completing the rest of the request. | |
2500 | */ | |
2501 | pos += written; | |
2502 | count -= written; | |
fb5527e6 JM |
2503 | written_buffered = generic_file_buffered_write(iocb, iov, |
2504 | nr_segs, pos, ppos, count, | |
2505 | written); | |
2506 | /* | |
2507 | * If generic_file_buffered_write() retuned a synchronous error | |
2508 | * then we want to return the number of bytes which were | |
2509 | * direct-written, or the error code if that was zero. Note | |
2510 | * that this differs from normal direct-io semantics, which | |
2511 | * will return -EFOO even if some bytes were written. | |
2512 | */ | |
2513 | if (written_buffered < 0) { | |
2514 | err = written_buffered; | |
2515 | goto out; | |
2516 | } | |
1da177e4 | 2517 | |
fb5527e6 JM |
2518 | /* |
2519 | * We need to ensure that the page cache pages are written to | |
2520 | * disk and invalidated to preserve the expected O_DIRECT | |
2521 | * semantics. | |
2522 | */ | |
2523 | endbyte = pos + written_buffered - written - 1; | |
c05c4edd | 2524 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); |
fb5527e6 JM |
2525 | if (err == 0) { |
2526 | written = written_buffered; | |
2527 | invalidate_mapping_pages(mapping, | |
2528 | pos >> PAGE_CACHE_SHIFT, | |
2529 | endbyte >> PAGE_CACHE_SHIFT); | |
2530 | } else { | |
2531 | /* | |
2532 | * We don't know how much we wrote, so just return | |
2533 | * the number of bytes which were direct-written | |
2534 | */ | |
2535 | } | |
2536 | } else { | |
2537 | written = generic_file_buffered_write(iocb, iov, nr_segs, | |
2538 | pos, ppos, count, written); | |
2539 | } | |
1da177e4 LT |
2540 | out: |
2541 | current->backing_dev_info = NULL; | |
2542 | return written ? written : err; | |
2543 | } | |
e4dd9de3 JK |
2544 | EXPORT_SYMBOL(__generic_file_aio_write); |
2545 | ||
e4dd9de3 JK |
2546 | /** |
2547 | * generic_file_aio_write - write data to a file | |
2548 | * @iocb: IO state structure | |
2549 | * @iov: vector with data to write | |
2550 | * @nr_segs: number of segments in the vector | |
2551 | * @pos: position in file where to write | |
2552 | * | |
2553 | * This is a wrapper around __generic_file_aio_write() to be used by most | |
2554 | * filesystems. It takes care of syncing the file in case of O_SYNC file | |
2555 | * and acquires i_mutex as needed. | |
2556 | */ | |
027445c3 BP |
2557 | ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
2558 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
2559 | { |
2560 | struct file *file = iocb->ki_filp; | |
148f948b | 2561 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2562 | ssize_t ret; |
1da177e4 LT |
2563 | |
2564 | BUG_ON(iocb->ki_pos != pos); | |
2565 | ||
1b1dcc1b | 2566 | mutex_lock(&inode->i_mutex); |
e4dd9de3 | 2567 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); |
1b1dcc1b | 2568 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 2569 | |
148f948b | 2570 | if (ret > 0 || ret == -EIOCBQUEUED) { |
1da177e4 LT |
2571 | ssize_t err; |
2572 | ||
148f948b | 2573 | err = generic_write_sync(file, pos, ret); |
c7b50db2 | 2574 | if (err < 0 && ret > 0) |
1da177e4 LT |
2575 | ret = err; |
2576 | } | |
2577 | return ret; | |
2578 | } | |
2579 | EXPORT_SYMBOL(generic_file_aio_write); | |
2580 | ||
cf9a2ae8 DH |
2581 | /** |
2582 | * try_to_release_page() - release old fs-specific metadata on a page | |
2583 | * | |
2584 | * @page: the page which the kernel is trying to free | |
2585 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2586 | * | |
2587 | * The address_space is to try to release any data against the page | |
2588 | * (presumably at page->private). If the release was successful, return `1'. | |
2589 | * Otherwise return zero. | |
2590 | * | |
266cf658 DH |
2591 | * This may also be called if PG_fscache is set on a page, indicating that the |
2592 | * page is known to the local caching routines. | |
2593 | * | |
cf9a2ae8 | 2594 | * The @gfp_mask argument specifies whether I/O may be performed to release |
3f31fddf | 2595 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
cf9a2ae8 | 2596 | * |
cf9a2ae8 DH |
2597 | */ |
2598 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2599 | { | |
2600 | struct address_space * const mapping = page->mapping; | |
2601 | ||
2602 | BUG_ON(!PageLocked(page)); | |
2603 | if (PageWriteback(page)) | |
2604 | return 0; | |
2605 | ||
2606 | if (mapping && mapping->a_ops->releasepage) | |
2607 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2608 | return try_to_free_buffers(page); | |
2609 | } | |
2610 | ||
2611 | EXPORT_SYMBOL(try_to_release_page); |