]>
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 | * | |
a66979ab DC |
83 | * inode_wb_list_lock |
84 | * sb_lock (fs/fs-writeback.c) | |
1da177e4 LT |
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) | |
a66979ab | 101 | * inode_wb_list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 102 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
a66979ab | 103 | * inode_wb_list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 104 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
105 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
106 | * | |
6a46079c AK |
107 | * (code doesn't rely on that order, so you could switch it around) |
108 | * ->tasklist_lock (memory_failure, collect_procs_ao) | |
109 | * ->i_mmap_lock | |
1da177e4 LT |
110 | */ |
111 | ||
112 | /* | |
e64a782f | 113 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 114 | * sure the page is locked and that nobody else uses it - or that usage |
19fd6231 | 115 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 116 | */ |
e64a782f | 117 | void __delete_from_page_cache(struct page *page) |
1da177e4 LT |
118 | { |
119 | struct address_space *mapping = page->mapping; | |
120 | ||
121 | radix_tree_delete(&mapping->page_tree, page->index); | |
122 | page->mapping = NULL; | |
123 | mapping->nrpages--; | |
347ce434 | 124 | __dec_zone_page_state(page, NR_FILE_PAGES); |
4b02108a KM |
125 | if (PageSwapBacked(page)) |
126 | __dec_zone_page_state(page, NR_SHMEM); | |
45426812 | 127 | BUG_ON(page_mapped(page)); |
3a692790 LT |
128 | |
129 | /* | |
130 | * Some filesystems seem to re-dirty the page even after | |
131 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
132 | * | |
133 | * Fix it up by doing a final dirty accounting check after | |
134 | * having removed the page entirely. | |
135 | */ | |
136 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { | |
137 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
138 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
139 | } | |
1da177e4 LT |
140 | } |
141 | ||
702cfbf9 MK |
142 | /** |
143 | * delete_from_page_cache - delete page from page cache | |
144 | * @page: the page which the kernel is trying to remove from page cache | |
145 | * | |
146 | * This must be called only on pages that have been verified to be in the page | |
147 | * cache and locked. It will never put the page into the free list, the caller | |
148 | * has a reference on the page. | |
149 | */ | |
150 | void delete_from_page_cache(struct page *page) | |
1da177e4 LT |
151 | { |
152 | struct address_space *mapping = page->mapping; | |
6072d13c | 153 | void (*freepage)(struct page *); |
1da177e4 | 154 | |
cd7619d6 | 155 | BUG_ON(!PageLocked(page)); |
1da177e4 | 156 | |
6072d13c | 157 | freepage = mapping->a_ops->freepage; |
19fd6231 | 158 | spin_lock_irq(&mapping->tree_lock); |
e64a782f | 159 | __delete_from_page_cache(page); |
19fd6231 | 160 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 161 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
162 | |
163 | if (freepage) | |
164 | freepage(page); | |
97cecb5a MK |
165 | page_cache_release(page); |
166 | } | |
167 | EXPORT_SYMBOL(delete_from_page_cache); | |
168 | ||
7eaceacc | 169 | static int sleep_on_page(void *word) |
1da177e4 | 170 | { |
1da177e4 LT |
171 | io_schedule(); |
172 | return 0; | |
173 | } | |
174 | ||
7eaceacc | 175 | static int sleep_on_page_killable(void *word) |
2687a356 | 176 | { |
7eaceacc | 177 | sleep_on_page(word); |
2687a356 MW |
178 | return fatal_signal_pending(current) ? -EINTR : 0; |
179 | } | |
180 | ||
1da177e4 | 181 | /** |
485bb99b | 182 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
183 | * @mapping: address space structure to write |
184 | * @start: offset in bytes where the range starts | |
469eb4d0 | 185 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 186 | * @sync_mode: enable synchronous operation |
1da177e4 | 187 | * |
485bb99b RD |
188 | * Start writeback against all of a mapping's dirty pages that lie |
189 | * within the byte offsets <start, end> inclusive. | |
190 | * | |
1da177e4 | 191 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 192 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
193 | * these two operations is that if a dirty page/buffer is encountered, it must |
194 | * be waited upon, and not just skipped over. | |
195 | */ | |
ebcf28e1 AM |
196 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
197 | loff_t end, int sync_mode) | |
1da177e4 LT |
198 | { |
199 | int ret; | |
200 | struct writeback_control wbc = { | |
201 | .sync_mode = sync_mode, | |
05fe478d | 202 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
203 | .range_start = start, |
204 | .range_end = end, | |
1da177e4 LT |
205 | }; |
206 | ||
207 | if (!mapping_cap_writeback_dirty(mapping)) | |
208 | return 0; | |
209 | ||
210 | ret = do_writepages(mapping, &wbc); | |
211 | return ret; | |
212 | } | |
213 | ||
214 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
215 | int sync_mode) | |
216 | { | |
111ebb6e | 217 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
218 | } |
219 | ||
220 | int filemap_fdatawrite(struct address_space *mapping) | |
221 | { | |
222 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
223 | } | |
224 | EXPORT_SYMBOL(filemap_fdatawrite); | |
225 | ||
f4c0a0fd | 226 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 227 | loff_t end) |
1da177e4 LT |
228 | { |
229 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
230 | } | |
f4c0a0fd | 231 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 232 | |
485bb99b RD |
233 | /** |
234 | * filemap_flush - mostly a non-blocking flush | |
235 | * @mapping: target address_space | |
236 | * | |
1da177e4 LT |
237 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
238 | * purposes - I/O may not be started against all dirty pages. | |
239 | */ | |
240 | int filemap_flush(struct address_space *mapping) | |
241 | { | |
242 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
243 | } | |
244 | EXPORT_SYMBOL(filemap_flush); | |
245 | ||
485bb99b | 246 | /** |
94004ed7 CH |
247 | * filemap_fdatawait_range - wait for writeback to complete |
248 | * @mapping: address space structure to wait for | |
249 | * @start_byte: offset in bytes where the range starts | |
250 | * @end_byte: offset in bytes where the range ends (inclusive) | |
485bb99b | 251 | * |
94004ed7 CH |
252 | * Walk the list of under-writeback pages of the given address space |
253 | * in the given range and wait for all of them. | |
1da177e4 | 254 | */ |
94004ed7 CH |
255 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
256 | loff_t end_byte) | |
1da177e4 | 257 | { |
94004ed7 CH |
258 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
259 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
260 | struct pagevec pvec; |
261 | int nr_pages; | |
262 | int ret = 0; | |
1da177e4 | 263 | |
94004ed7 | 264 | if (end_byte < start_byte) |
1da177e4 LT |
265 | return 0; |
266 | ||
267 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
268 | while ((index <= end) && |
269 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
270 | PAGECACHE_TAG_WRITEBACK, | |
271 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
272 | unsigned i; | |
273 | ||
274 | for (i = 0; i < nr_pages; i++) { | |
275 | struct page *page = pvec.pages[i]; | |
276 | ||
277 | /* until radix tree lookup accepts end_index */ | |
278 | if (page->index > end) | |
279 | continue; | |
280 | ||
281 | wait_on_page_writeback(page); | |
212260aa | 282 | if (TestClearPageError(page)) |
1da177e4 LT |
283 | ret = -EIO; |
284 | } | |
285 | pagevec_release(&pvec); | |
286 | cond_resched(); | |
287 | } | |
288 | ||
289 | /* Check for outstanding write errors */ | |
290 | if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
291 | ret = -ENOSPC; | |
292 | if (test_and_clear_bit(AS_EIO, &mapping->flags)) | |
293 | ret = -EIO; | |
294 | ||
295 | return ret; | |
296 | } | |
d3bccb6f JK |
297 | EXPORT_SYMBOL(filemap_fdatawait_range); |
298 | ||
1da177e4 | 299 | /** |
485bb99b | 300 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 301 | * @mapping: address space structure to wait for |
485bb99b RD |
302 | * |
303 | * Walk the list of under-writeback pages of the given address space | |
304 | * and wait for all of them. | |
1da177e4 LT |
305 | */ |
306 | int filemap_fdatawait(struct address_space *mapping) | |
307 | { | |
308 | loff_t i_size = i_size_read(mapping->host); | |
309 | ||
310 | if (i_size == 0) | |
311 | return 0; | |
312 | ||
94004ed7 | 313 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
314 | } |
315 | EXPORT_SYMBOL(filemap_fdatawait); | |
316 | ||
317 | int filemap_write_and_wait(struct address_space *mapping) | |
318 | { | |
28fd1298 | 319 | int err = 0; |
1da177e4 LT |
320 | |
321 | if (mapping->nrpages) { | |
28fd1298 OH |
322 | err = filemap_fdatawrite(mapping); |
323 | /* | |
324 | * Even if the above returned error, the pages may be | |
325 | * written partially (e.g. -ENOSPC), so we wait for it. | |
326 | * But the -EIO is special case, it may indicate the worst | |
327 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
328 | */ | |
329 | if (err != -EIO) { | |
330 | int err2 = filemap_fdatawait(mapping); | |
331 | if (!err) | |
332 | err = err2; | |
333 | } | |
1da177e4 | 334 | } |
28fd1298 | 335 | return err; |
1da177e4 | 336 | } |
28fd1298 | 337 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 338 | |
485bb99b RD |
339 | /** |
340 | * filemap_write_and_wait_range - write out & wait on a file range | |
341 | * @mapping: the address_space for the pages | |
342 | * @lstart: offset in bytes where the range starts | |
343 | * @lend: offset in bytes where the range ends (inclusive) | |
344 | * | |
469eb4d0 AM |
345 | * Write out and wait upon file offsets lstart->lend, inclusive. |
346 | * | |
347 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
348 | * that this function can be used to write to the very end-of-file (end = -1). | |
349 | */ | |
1da177e4 LT |
350 | int filemap_write_and_wait_range(struct address_space *mapping, |
351 | loff_t lstart, loff_t lend) | |
352 | { | |
28fd1298 | 353 | int err = 0; |
1da177e4 LT |
354 | |
355 | if (mapping->nrpages) { | |
28fd1298 OH |
356 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
357 | WB_SYNC_ALL); | |
358 | /* See comment of filemap_write_and_wait() */ | |
359 | if (err != -EIO) { | |
94004ed7 CH |
360 | int err2 = filemap_fdatawait_range(mapping, |
361 | lstart, lend); | |
28fd1298 OH |
362 | if (!err) |
363 | err = err2; | |
364 | } | |
1da177e4 | 365 | } |
28fd1298 | 366 | return err; |
1da177e4 | 367 | } |
f6995585 | 368 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 369 | |
ef6a3c63 MS |
370 | /** |
371 | * replace_page_cache_page - replace a pagecache page with a new one | |
372 | * @old: page to be replaced | |
373 | * @new: page to replace with | |
374 | * @gfp_mask: allocation mode | |
375 | * | |
376 | * This function replaces a page in the pagecache with a new one. On | |
377 | * success it acquires the pagecache reference for the new page and | |
378 | * drops it for the old page. Both the old and new pages must be | |
379 | * locked. This function does not add the new page to the LRU, the | |
380 | * caller must do that. | |
381 | * | |
382 | * The remove + add is atomic. The only way this function can fail is | |
383 | * memory allocation failure. | |
384 | */ | |
385 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
386 | { | |
387 | int error; | |
388 | struct mem_cgroup *memcg = NULL; | |
389 | ||
390 | VM_BUG_ON(!PageLocked(old)); | |
391 | VM_BUG_ON(!PageLocked(new)); | |
392 | VM_BUG_ON(new->mapping); | |
393 | ||
394 | /* | |
395 | * This is not page migration, but prepare_migration and | |
396 | * end_migration does enough work for charge replacement. | |
397 | * | |
398 | * In the longer term we probably want a specialized function | |
399 | * for moving the charge from old to new in a more efficient | |
400 | * manner. | |
401 | */ | |
402 | error = mem_cgroup_prepare_migration(old, new, &memcg, gfp_mask); | |
403 | if (error) | |
404 | return error; | |
405 | ||
406 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); | |
407 | if (!error) { | |
408 | struct address_space *mapping = old->mapping; | |
409 | void (*freepage)(struct page *); | |
410 | ||
411 | pgoff_t offset = old->index; | |
412 | freepage = mapping->a_ops->freepage; | |
413 | ||
414 | page_cache_get(new); | |
415 | new->mapping = mapping; | |
416 | new->index = offset; | |
417 | ||
418 | spin_lock_irq(&mapping->tree_lock); | |
e64a782f | 419 | __delete_from_page_cache(old); |
ef6a3c63 MS |
420 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
421 | BUG_ON(error); | |
422 | mapping->nrpages++; | |
423 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
424 | if (PageSwapBacked(new)) | |
425 | __inc_zone_page_state(new, NR_SHMEM); | |
426 | spin_unlock_irq(&mapping->tree_lock); | |
427 | radix_tree_preload_end(); | |
428 | if (freepage) | |
429 | freepage(old); | |
430 | page_cache_release(old); | |
431 | mem_cgroup_end_migration(memcg, old, new, true); | |
432 | } else { | |
433 | mem_cgroup_end_migration(memcg, old, new, false); | |
434 | } | |
435 | ||
436 | return error; | |
437 | } | |
438 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
439 | ||
485bb99b | 440 | /** |
e286781d | 441 | * add_to_page_cache_locked - add a locked page to the pagecache |
485bb99b RD |
442 | * @page: page to add |
443 | * @mapping: the page's address_space | |
444 | * @offset: page index | |
445 | * @gfp_mask: page allocation mode | |
446 | * | |
e286781d | 447 | * This function is used to add a page to the pagecache. It must be locked. |
1da177e4 LT |
448 | * This function does not add the page to the LRU. The caller must do that. |
449 | */ | |
e286781d | 450 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, |
6daa0e28 | 451 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 452 | { |
e286781d NP |
453 | int error; |
454 | ||
455 | VM_BUG_ON(!PageLocked(page)); | |
456 | ||
457 | error = mem_cgroup_cache_charge(page, current->mm, | |
2c26fdd7 | 458 | gfp_mask & GFP_RECLAIM_MASK); |
35c754d7 BS |
459 | if (error) |
460 | goto out; | |
1da177e4 | 461 | |
35c754d7 | 462 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
1da177e4 | 463 | if (error == 0) { |
e286781d NP |
464 | page_cache_get(page); |
465 | page->mapping = mapping; | |
466 | page->index = offset; | |
467 | ||
19fd6231 | 468 | spin_lock_irq(&mapping->tree_lock); |
1da177e4 | 469 | error = radix_tree_insert(&mapping->page_tree, offset, page); |
e286781d | 470 | if (likely(!error)) { |
1da177e4 | 471 | mapping->nrpages++; |
347ce434 | 472 | __inc_zone_page_state(page, NR_FILE_PAGES); |
4b02108a KM |
473 | if (PageSwapBacked(page)) |
474 | __inc_zone_page_state(page, NR_SHMEM); | |
e767e056 | 475 | spin_unlock_irq(&mapping->tree_lock); |
e286781d NP |
476 | } else { |
477 | page->mapping = NULL; | |
e767e056 | 478 | spin_unlock_irq(&mapping->tree_lock); |
69029cd5 | 479 | mem_cgroup_uncharge_cache_page(page); |
e286781d NP |
480 | page_cache_release(page); |
481 | } | |
1da177e4 | 482 | radix_tree_preload_end(); |
35c754d7 | 483 | } else |
69029cd5 | 484 | mem_cgroup_uncharge_cache_page(page); |
8a9f3ccd | 485 | out: |
1da177e4 LT |
486 | return error; |
487 | } | |
e286781d | 488 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
489 | |
490 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 491 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 492 | { |
4f98a2fe RR |
493 | int ret; |
494 | ||
495 | /* | |
496 | * Splice_read and readahead add shmem/tmpfs pages into the page cache | |
497 | * before shmem_readpage has a chance to mark them as SwapBacked: they | |
e9d6c157 | 498 | * need to go on the anon lru below, and mem_cgroup_cache_charge |
4f98a2fe RR |
499 | * (called in add_to_page_cache) needs to know where they're going too. |
500 | */ | |
501 | if (mapping_cap_swap_backed(mapping)) | |
502 | SetPageSwapBacked(page); | |
503 | ||
504 | ret = add_to_page_cache(page, mapping, offset, gfp_mask); | |
505 | if (ret == 0) { | |
506 | if (page_is_file_cache(page)) | |
507 | lru_cache_add_file(page); | |
508 | else | |
e9d6c157 | 509 | lru_cache_add_anon(page); |
4f98a2fe | 510 | } |
1da177e4 LT |
511 | return ret; |
512 | } | |
18bc0bbd | 513 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 514 | |
44110fe3 | 515 | #ifdef CONFIG_NUMA |
2ae88149 | 516 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 517 | { |
c0ff7453 MX |
518 | int n; |
519 | struct page *page; | |
520 | ||
44110fe3 | 521 | if (cpuset_do_page_mem_spread()) { |
c0ff7453 MX |
522 | get_mems_allowed(); |
523 | n = cpuset_mem_spread_node(); | |
524 | page = alloc_pages_exact_node(n, gfp, 0); | |
525 | put_mems_allowed(); | |
526 | return page; | |
44110fe3 | 527 | } |
2ae88149 | 528 | return alloc_pages(gfp, 0); |
44110fe3 | 529 | } |
2ae88149 | 530 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
531 | #endif |
532 | ||
1da177e4 LT |
533 | /* |
534 | * In order to wait for pages to become available there must be | |
535 | * waitqueues associated with pages. By using a hash table of | |
536 | * waitqueues where the bucket discipline is to maintain all | |
537 | * waiters on the same queue and wake all when any of the pages | |
538 | * become available, and for the woken contexts to check to be | |
539 | * sure the appropriate page became available, this saves space | |
540 | * at a cost of "thundering herd" phenomena during rare hash | |
541 | * collisions. | |
542 | */ | |
543 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
544 | { | |
545 | const struct zone *zone = page_zone(page); | |
546 | ||
547 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
548 | } | |
549 | ||
550 | static inline void wake_up_page(struct page *page, int bit) | |
551 | { | |
552 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); | |
553 | } | |
554 | ||
920c7a5d | 555 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
556 | { |
557 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
558 | ||
559 | if (test_bit(bit_nr, &page->flags)) | |
7eaceacc | 560 | __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
561 | TASK_UNINTERRUPTIBLE); |
562 | } | |
563 | EXPORT_SYMBOL(wait_on_page_bit); | |
564 | ||
385e1ca5 DH |
565 | /** |
566 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
567 | * @page: Page defining the wait queue of interest |
568 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
569 | * |
570 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
571 | */ | |
572 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
573 | { | |
574 | wait_queue_head_t *q = page_waitqueue(page); | |
575 | unsigned long flags; | |
576 | ||
577 | spin_lock_irqsave(&q->lock, flags); | |
578 | __add_wait_queue(q, waiter); | |
579 | spin_unlock_irqrestore(&q->lock, flags); | |
580 | } | |
581 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
582 | ||
1da177e4 | 583 | /** |
485bb99b | 584 | * unlock_page - unlock a locked page |
1da177e4 LT |
585 | * @page: the page |
586 | * | |
587 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
588 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
589 | * mechananism between PageLocked pages and PageWriteback pages is shared. | |
590 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. | |
591 | * | |
8413ac9d NP |
592 | * The mb is necessary to enforce ordering between the clear_bit and the read |
593 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 594 | */ |
920c7a5d | 595 | void unlock_page(struct page *page) |
1da177e4 | 596 | { |
8413ac9d NP |
597 | VM_BUG_ON(!PageLocked(page)); |
598 | clear_bit_unlock(PG_locked, &page->flags); | |
599 | smp_mb__after_clear_bit(); | |
1da177e4 LT |
600 | wake_up_page(page, PG_locked); |
601 | } | |
602 | EXPORT_SYMBOL(unlock_page); | |
603 | ||
485bb99b RD |
604 | /** |
605 | * end_page_writeback - end writeback against a page | |
606 | * @page: the page | |
1da177e4 LT |
607 | */ |
608 | void end_page_writeback(struct page *page) | |
609 | { | |
ac6aadb2 MS |
610 | if (TestClearPageReclaim(page)) |
611 | rotate_reclaimable_page(page); | |
612 | ||
613 | if (!test_clear_page_writeback(page)) | |
614 | BUG(); | |
615 | ||
1da177e4 LT |
616 | smp_mb__after_clear_bit(); |
617 | wake_up_page(page, PG_writeback); | |
618 | } | |
619 | EXPORT_SYMBOL(end_page_writeback); | |
620 | ||
485bb99b RD |
621 | /** |
622 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
623 | * @page: the page to lock | |
1da177e4 | 624 | */ |
920c7a5d | 625 | void __lock_page(struct page *page) |
1da177e4 LT |
626 | { |
627 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
628 | ||
7eaceacc | 629 | __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
630 | TASK_UNINTERRUPTIBLE); |
631 | } | |
632 | EXPORT_SYMBOL(__lock_page); | |
633 | ||
b5606c2d | 634 | int __lock_page_killable(struct page *page) |
2687a356 MW |
635 | { |
636 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
637 | ||
638 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
7eaceacc | 639 | sleep_on_page_killable, TASK_KILLABLE); |
2687a356 | 640 | } |
18bc0bbd | 641 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 642 | |
d065bd81 ML |
643 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
644 | unsigned int flags) | |
645 | { | |
646 | if (!(flags & FAULT_FLAG_ALLOW_RETRY)) { | |
647 | __lock_page(page); | |
648 | return 1; | |
649 | } else { | |
318b275f GN |
650 | if (!(flags & FAULT_FLAG_RETRY_NOWAIT)) { |
651 | up_read(&mm->mmap_sem); | |
652 | wait_on_page_locked(page); | |
653 | } | |
d065bd81 ML |
654 | return 0; |
655 | } | |
656 | } | |
657 | ||
485bb99b RD |
658 | /** |
659 | * find_get_page - find and get a page reference | |
660 | * @mapping: the address_space to search | |
661 | * @offset: the page index | |
662 | * | |
da6052f7 NP |
663 | * Is there a pagecache struct page at the given (mapping, offset) tuple? |
664 | * If yes, increment its refcount and return it; if no, return NULL. | |
1da177e4 | 665 | */ |
a60637c8 | 666 | struct page *find_get_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 667 | { |
a60637c8 | 668 | void **pagep; |
1da177e4 LT |
669 | struct page *page; |
670 | ||
a60637c8 NP |
671 | rcu_read_lock(); |
672 | repeat: | |
673 | page = NULL; | |
674 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
675 | if (pagep) { | |
676 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
677 | if (unlikely(!page)) |
678 | goto out; | |
679 | if (radix_tree_deref_retry(page)) | |
a60637c8 NP |
680 | goto repeat; |
681 | ||
682 | if (!page_cache_get_speculative(page)) | |
683 | goto repeat; | |
684 | ||
685 | /* | |
686 | * Has the page moved? | |
687 | * This is part of the lockless pagecache protocol. See | |
688 | * include/linux/pagemap.h for details. | |
689 | */ | |
690 | if (unlikely(page != *pagep)) { | |
691 | page_cache_release(page); | |
692 | goto repeat; | |
693 | } | |
694 | } | |
27d20fdd | 695 | out: |
a60637c8 NP |
696 | rcu_read_unlock(); |
697 | ||
1da177e4 LT |
698 | return page; |
699 | } | |
1da177e4 LT |
700 | EXPORT_SYMBOL(find_get_page); |
701 | ||
1da177e4 LT |
702 | /** |
703 | * find_lock_page - locate, pin and lock a pagecache page | |
67be2dd1 MW |
704 | * @mapping: the address_space to search |
705 | * @offset: the page index | |
1da177e4 LT |
706 | * |
707 | * Locates the desired pagecache page, locks it, increments its reference | |
708 | * count and returns its address. | |
709 | * | |
710 | * Returns zero if the page was not present. find_lock_page() may sleep. | |
711 | */ | |
a60637c8 | 712 | struct page *find_lock_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 LT |
713 | { |
714 | struct page *page; | |
715 | ||
1da177e4 | 716 | repeat: |
a60637c8 | 717 | page = find_get_page(mapping, offset); |
1da177e4 | 718 | if (page) { |
a60637c8 NP |
719 | lock_page(page); |
720 | /* Has the page been truncated? */ | |
721 | if (unlikely(page->mapping != mapping)) { | |
722 | unlock_page(page); | |
723 | page_cache_release(page); | |
724 | goto repeat; | |
1da177e4 | 725 | } |
a60637c8 | 726 | VM_BUG_ON(page->index != offset); |
1da177e4 | 727 | } |
1da177e4 LT |
728 | return page; |
729 | } | |
1da177e4 LT |
730 | EXPORT_SYMBOL(find_lock_page); |
731 | ||
732 | /** | |
733 | * find_or_create_page - locate or add a pagecache page | |
67be2dd1 MW |
734 | * @mapping: the page's address_space |
735 | * @index: the page's index into the mapping | |
736 | * @gfp_mask: page allocation mode | |
1da177e4 LT |
737 | * |
738 | * Locates a page in the pagecache. If the page is not present, a new page | |
739 | * is allocated using @gfp_mask and is added to the pagecache and to the VM's | |
740 | * LRU list. The returned page is locked and has its reference count | |
741 | * incremented. | |
742 | * | |
743 | * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic | |
744 | * allocation! | |
745 | * | |
746 | * find_or_create_page() returns the desired page's address, or zero on | |
747 | * memory exhaustion. | |
748 | */ | |
749 | struct page *find_or_create_page(struct address_space *mapping, | |
57f6b96c | 750 | pgoff_t index, gfp_t gfp_mask) |
1da177e4 | 751 | { |
eb2be189 | 752 | struct page *page; |
1da177e4 LT |
753 | int err; |
754 | repeat: | |
755 | page = find_lock_page(mapping, index); | |
756 | if (!page) { | |
eb2be189 NP |
757 | page = __page_cache_alloc(gfp_mask); |
758 | if (!page) | |
759 | return NULL; | |
67d58ac4 NP |
760 | /* |
761 | * We want a regular kernel memory (not highmem or DMA etc) | |
762 | * allocation for the radix tree nodes, but we need to honour | |
763 | * the context-specific requirements the caller has asked for. | |
764 | * GFP_RECLAIM_MASK collects those requirements. | |
765 | */ | |
766 | err = add_to_page_cache_lru(page, mapping, index, | |
767 | (gfp_mask & GFP_RECLAIM_MASK)); | |
eb2be189 NP |
768 | if (unlikely(err)) { |
769 | page_cache_release(page); | |
770 | page = NULL; | |
771 | if (err == -EEXIST) | |
772 | goto repeat; | |
1da177e4 | 773 | } |
1da177e4 | 774 | } |
1da177e4 LT |
775 | return page; |
776 | } | |
1da177e4 LT |
777 | EXPORT_SYMBOL(find_or_create_page); |
778 | ||
779 | /** | |
780 | * find_get_pages - gang pagecache lookup | |
781 | * @mapping: The address_space to search | |
782 | * @start: The starting page index | |
783 | * @nr_pages: The maximum number of pages | |
784 | * @pages: Where the resulting pages are placed | |
785 | * | |
786 | * find_get_pages() will search for and return a group of up to | |
787 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
788 | * find_get_pages() takes a reference against the returned pages. | |
789 | * | |
790 | * The search returns a group of mapping-contiguous pages with ascending | |
791 | * indexes. There may be holes in the indices due to not-present pages. | |
792 | * | |
793 | * find_get_pages() returns the number of pages which were found. | |
794 | */ | |
795 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
796 | unsigned int nr_pages, struct page **pages) | |
797 | { | |
798 | unsigned int i; | |
799 | unsigned int ret; | |
a60637c8 NP |
800 | unsigned int nr_found; |
801 | ||
802 | rcu_read_lock(); | |
803 | restart: | |
804 | nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, | |
805 | (void ***)pages, start, nr_pages); | |
806 | ret = 0; | |
807 | for (i = 0; i < nr_found; i++) { | |
808 | struct page *page; | |
809 | repeat: | |
810 | page = radix_tree_deref_slot((void **)pages[i]); | |
811 | if (unlikely(!page)) | |
812 | continue; | |
9d8aa4ea HD |
813 | |
814 | /* | |
815 | * This can only trigger when the entry at index 0 moves out | |
816 | * of or back to the root: none yet gotten, safe to restart. | |
817 | */ | |
27d20fdd | 818 | if (radix_tree_deref_retry(page)) { |
9d8aa4ea | 819 | WARN_ON(start | i); |
a60637c8 | 820 | goto restart; |
27d20fdd | 821 | } |
a60637c8 NP |
822 | |
823 | if (!page_cache_get_speculative(page)) | |
824 | goto repeat; | |
825 | ||
826 | /* Has the page moved? */ | |
827 | if (unlikely(page != *((void **)pages[i]))) { | |
828 | page_cache_release(page); | |
829 | goto repeat; | |
830 | } | |
1da177e4 | 831 | |
a60637c8 NP |
832 | pages[ret] = page; |
833 | ret++; | |
834 | } | |
5b280c0c HD |
835 | |
836 | /* | |
837 | * If all entries were removed before we could secure them, | |
838 | * try again, because callers stop trying once 0 is returned. | |
839 | */ | |
840 | if (unlikely(!ret && nr_found)) | |
841 | goto restart; | |
a60637c8 | 842 | rcu_read_unlock(); |
1da177e4 LT |
843 | return ret; |
844 | } | |
845 | ||
ebf43500 JA |
846 | /** |
847 | * find_get_pages_contig - gang contiguous pagecache lookup | |
848 | * @mapping: The address_space to search | |
849 | * @index: The starting page index | |
850 | * @nr_pages: The maximum number of pages | |
851 | * @pages: Where the resulting pages are placed | |
852 | * | |
853 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
854 | * that the returned number of pages are guaranteed to be contiguous. | |
855 | * | |
856 | * find_get_pages_contig() returns the number of pages which were found. | |
857 | */ | |
858 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
859 | unsigned int nr_pages, struct page **pages) | |
860 | { | |
861 | unsigned int i; | |
862 | unsigned int ret; | |
a60637c8 NP |
863 | unsigned int nr_found; |
864 | ||
865 | rcu_read_lock(); | |
866 | restart: | |
867 | nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, | |
868 | (void ***)pages, index, nr_pages); | |
869 | ret = 0; | |
870 | for (i = 0; i < nr_found; i++) { | |
871 | struct page *page; | |
872 | repeat: | |
873 | page = radix_tree_deref_slot((void **)pages[i]); | |
874 | if (unlikely(!page)) | |
875 | continue; | |
9d8aa4ea HD |
876 | |
877 | /* | |
878 | * This can only trigger when the entry at index 0 moves out | |
879 | * of or back to the root: none yet gotten, safe to restart. | |
880 | */ | |
27d20fdd | 881 | if (radix_tree_deref_retry(page)) |
a60637c8 | 882 | goto restart; |
ebf43500 | 883 | |
a60637c8 NP |
884 | if (!page_cache_get_speculative(page)) |
885 | goto repeat; | |
886 | ||
887 | /* Has the page moved? */ | |
888 | if (unlikely(page != *((void **)pages[i]))) { | |
889 | page_cache_release(page); | |
890 | goto repeat; | |
891 | } | |
892 | ||
9cbb4cb2 NP |
893 | /* |
894 | * must check mapping and index after taking the ref. | |
895 | * otherwise we can get both false positives and false | |
896 | * negatives, which is just confusing to the caller. | |
897 | */ | |
898 | if (page->mapping == NULL || page->index != index) { | |
899 | page_cache_release(page); | |
900 | break; | |
901 | } | |
902 | ||
a60637c8 NP |
903 | pages[ret] = page; |
904 | ret++; | |
ebf43500 JA |
905 | index++; |
906 | } | |
a60637c8 NP |
907 | rcu_read_unlock(); |
908 | return ret; | |
ebf43500 | 909 | } |
ef71c15c | 910 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 911 | |
485bb99b RD |
912 | /** |
913 | * find_get_pages_tag - find and return pages that match @tag | |
914 | * @mapping: the address_space to search | |
915 | * @index: the starting page index | |
916 | * @tag: the tag index | |
917 | * @nr_pages: the maximum number of pages | |
918 | * @pages: where the resulting pages are placed | |
919 | * | |
1da177e4 | 920 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 921 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
922 | */ |
923 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
924 | int tag, unsigned int nr_pages, struct page **pages) | |
925 | { | |
926 | unsigned int i; | |
927 | unsigned int ret; | |
a60637c8 NP |
928 | unsigned int nr_found; |
929 | ||
930 | rcu_read_lock(); | |
931 | restart: | |
932 | nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree, | |
933 | (void ***)pages, *index, nr_pages, tag); | |
934 | ret = 0; | |
935 | for (i = 0; i < nr_found; i++) { | |
936 | struct page *page; | |
937 | repeat: | |
938 | page = radix_tree_deref_slot((void **)pages[i]); | |
939 | if (unlikely(!page)) | |
940 | continue; | |
9d8aa4ea HD |
941 | |
942 | /* | |
943 | * This can only trigger when the entry at index 0 moves out | |
944 | * of or back to the root: none yet gotten, safe to restart. | |
945 | */ | |
27d20fdd | 946 | if (radix_tree_deref_retry(page)) |
a60637c8 NP |
947 | goto restart; |
948 | ||
949 | if (!page_cache_get_speculative(page)) | |
950 | goto repeat; | |
951 | ||
952 | /* Has the page moved? */ | |
953 | if (unlikely(page != *((void **)pages[i]))) { | |
954 | page_cache_release(page); | |
955 | goto repeat; | |
956 | } | |
957 | ||
958 | pages[ret] = page; | |
959 | ret++; | |
960 | } | |
5b280c0c HD |
961 | |
962 | /* | |
963 | * If all entries were removed before we could secure them, | |
964 | * try again, because callers stop trying once 0 is returned. | |
965 | */ | |
966 | if (unlikely(!ret && nr_found)) | |
967 | goto restart; | |
a60637c8 | 968 | rcu_read_unlock(); |
1da177e4 | 969 | |
1da177e4 LT |
970 | if (ret) |
971 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 972 | |
1da177e4 LT |
973 | return ret; |
974 | } | |
ef71c15c | 975 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 976 | |
485bb99b RD |
977 | /** |
978 | * grab_cache_page_nowait - returns locked page at given index in given cache | |
979 | * @mapping: target address_space | |
980 | * @index: the page index | |
981 | * | |
72fd4a35 | 982 | * Same as grab_cache_page(), but do not wait if the page is unavailable. |
1da177e4 LT |
983 | * This is intended for speculative data generators, where the data can |
984 | * be regenerated if the page couldn't be grabbed. This routine should | |
985 | * be safe to call while holding the lock for another page. | |
986 | * | |
987 | * Clear __GFP_FS when allocating the page to avoid recursion into the fs | |
988 | * and deadlock against the caller's locked page. | |
989 | */ | |
990 | struct page * | |
57f6b96c | 991 | grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) |
1da177e4 LT |
992 | { |
993 | struct page *page = find_get_page(mapping, index); | |
1da177e4 LT |
994 | |
995 | if (page) { | |
529ae9aa | 996 | if (trylock_page(page)) |
1da177e4 LT |
997 | return page; |
998 | page_cache_release(page); | |
999 | return NULL; | |
1000 | } | |
2ae88149 | 1001 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); |
67d58ac4 | 1002 | if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) { |
1da177e4 LT |
1003 | page_cache_release(page); |
1004 | page = NULL; | |
1005 | } | |
1006 | return page; | |
1007 | } | |
1da177e4 LT |
1008 | EXPORT_SYMBOL(grab_cache_page_nowait); |
1009 | ||
76d42bd9 WF |
1010 | /* |
1011 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1012 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1013 | * | |
1014 | * ---R__________________________________________B__________ | |
1015 | * ^ reading here ^ bad block(assume 4k) | |
1016 | * | |
1017 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1018 | * => failing the whole request => read(R) => read(R+1) => | |
1019 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1020 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1021 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1022 | * | |
1023 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1024 | */ | |
1025 | static void shrink_readahead_size_eio(struct file *filp, | |
1026 | struct file_ra_state *ra) | |
1027 | { | |
76d42bd9 | 1028 | ra->ra_pages /= 4; |
76d42bd9 WF |
1029 | } |
1030 | ||
485bb99b | 1031 | /** |
36e78914 | 1032 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1033 | * @filp: the file to read |
1034 | * @ppos: current file position | |
1035 | * @desc: read_descriptor | |
1036 | * @actor: read method | |
1037 | * | |
1da177e4 | 1038 | * This is a generic file read routine, and uses the |
485bb99b | 1039 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1040 | * |
1041 | * This is really ugly. But the goto's actually try to clarify some | |
1042 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1043 | */ |
36e78914 CH |
1044 | static void do_generic_file_read(struct file *filp, loff_t *ppos, |
1045 | read_descriptor_t *desc, read_actor_t actor) | |
1da177e4 | 1046 | { |
36e78914 | 1047 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1048 | struct inode *inode = mapping->host; |
36e78914 | 1049 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1050 | pgoff_t index; |
1051 | pgoff_t last_index; | |
1052 | pgoff_t prev_index; | |
1053 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1054 | unsigned int prev_offset; |
1da177e4 | 1055 | int error; |
1da177e4 | 1056 | |
1da177e4 | 1057 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1058 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1059 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
1da177e4 LT |
1060 | last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1061 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1062 | ||
1da177e4 LT |
1063 | for (;;) { |
1064 | struct page *page; | |
57f6b96c | 1065 | pgoff_t end_index; |
a32ea1e1 | 1066 | loff_t isize; |
1da177e4 LT |
1067 | unsigned long nr, ret; |
1068 | ||
1da177e4 | 1069 | cond_resched(); |
1da177e4 LT |
1070 | find_page: |
1071 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1072 | if (!page) { |
cf914a7d | 1073 | page_cache_sync_readahead(mapping, |
7ff81078 | 1074 | ra, filp, |
3ea89ee8 FW |
1075 | index, last_index - index); |
1076 | page = find_get_page(mapping, index); | |
1077 | if (unlikely(page == NULL)) | |
1078 | goto no_cached_page; | |
1079 | } | |
1080 | if (PageReadahead(page)) { | |
cf914a7d | 1081 | page_cache_async_readahead(mapping, |
7ff81078 | 1082 | ra, filp, page, |
3ea89ee8 | 1083 | index, last_index - index); |
1da177e4 | 1084 | } |
8ab22b9a HH |
1085 | if (!PageUptodate(page)) { |
1086 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1087 | !mapping->a_ops->is_partially_uptodate) | |
1088 | goto page_not_up_to_date; | |
529ae9aa | 1089 | if (!trylock_page(page)) |
8ab22b9a | 1090 | goto page_not_up_to_date; |
8d056cb9 DH |
1091 | /* Did it get truncated before we got the lock? */ |
1092 | if (!page->mapping) | |
1093 | goto page_not_up_to_date_locked; | |
8ab22b9a HH |
1094 | if (!mapping->a_ops->is_partially_uptodate(page, |
1095 | desc, offset)) | |
1096 | goto page_not_up_to_date_locked; | |
1097 | unlock_page(page); | |
1098 | } | |
1da177e4 | 1099 | page_ok: |
a32ea1e1 N |
1100 | /* |
1101 | * i_size must be checked after we know the page is Uptodate. | |
1102 | * | |
1103 | * Checking i_size after the check allows us to calculate | |
1104 | * the correct value for "nr", which means the zero-filled | |
1105 | * part of the page is not copied back to userspace (unless | |
1106 | * another truncate extends the file - this is desired though). | |
1107 | */ | |
1108 | ||
1109 | isize = i_size_read(inode); | |
1110 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1111 | if (unlikely(!isize || index > end_index)) { | |
1112 | page_cache_release(page); | |
1113 | goto out; | |
1114 | } | |
1115 | ||
1116 | /* nr is the maximum number of bytes to copy from this page */ | |
1117 | nr = PAGE_CACHE_SIZE; | |
1118 | if (index == end_index) { | |
1119 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1120 | if (nr <= offset) { | |
1121 | page_cache_release(page); | |
1122 | goto out; | |
1123 | } | |
1124 | } | |
1125 | nr = nr - offset; | |
1da177e4 LT |
1126 | |
1127 | /* If users can be writing to this page using arbitrary | |
1128 | * virtual addresses, take care about potential aliasing | |
1129 | * before reading the page on the kernel side. | |
1130 | */ | |
1131 | if (mapping_writably_mapped(mapping)) | |
1132 | flush_dcache_page(page); | |
1133 | ||
1134 | /* | |
ec0f1637 JK |
1135 | * When a sequential read accesses a page several times, |
1136 | * only mark it as accessed the first time. | |
1da177e4 | 1137 | */ |
ec0f1637 | 1138 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1139 | mark_page_accessed(page); |
1140 | prev_index = index; | |
1141 | ||
1142 | /* | |
1143 | * Ok, we have the page, and it's up-to-date, so | |
1144 | * now we can copy it to user space... | |
1145 | * | |
1146 | * The actor routine returns how many bytes were actually used.. | |
1147 | * NOTE! This may not be the same as how much of a user buffer | |
1148 | * we filled up (we may be padding etc), so we can only update | |
1149 | * "pos" here (the actor routine has to update the user buffer | |
1150 | * pointers and the remaining count). | |
1151 | */ | |
1152 | ret = actor(desc, page, offset, nr); | |
1153 | offset += ret; | |
1154 | index += offset >> PAGE_CACHE_SHIFT; | |
1155 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1156 | prev_offset = offset; |
1da177e4 LT |
1157 | |
1158 | page_cache_release(page); | |
1159 | if (ret == nr && desc->count) | |
1160 | continue; | |
1161 | goto out; | |
1162 | ||
1163 | page_not_up_to_date: | |
1164 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1165 | error = lock_page_killable(page); |
1166 | if (unlikely(error)) | |
1167 | goto readpage_error; | |
1da177e4 | 1168 | |
8ab22b9a | 1169 | page_not_up_to_date_locked: |
da6052f7 | 1170 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1171 | if (!page->mapping) { |
1172 | unlock_page(page); | |
1173 | page_cache_release(page); | |
1174 | continue; | |
1175 | } | |
1176 | ||
1177 | /* Did somebody else fill it already? */ | |
1178 | if (PageUptodate(page)) { | |
1179 | unlock_page(page); | |
1180 | goto page_ok; | |
1181 | } | |
1182 | ||
1183 | readpage: | |
91803b49 JM |
1184 | /* |
1185 | * A previous I/O error may have been due to temporary | |
1186 | * failures, eg. multipath errors. | |
1187 | * PG_error will be set again if readpage fails. | |
1188 | */ | |
1189 | ClearPageError(page); | |
1da177e4 LT |
1190 | /* Start the actual read. The read will unlock the page. */ |
1191 | error = mapping->a_ops->readpage(filp, page); | |
1192 | ||
994fc28c ZB |
1193 | if (unlikely(error)) { |
1194 | if (error == AOP_TRUNCATED_PAGE) { | |
1195 | page_cache_release(page); | |
1196 | goto find_page; | |
1197 | } | |
1da177e4 | 1198 | goto readpage_error; |
994fc28c | 1199 | } |
1da177e4 LT |
1200 | |
1201 | if (!PageUptodate(page)) { | |
85462323 ON |
1202 | error = lock_page_killable(page); |
1203 | if (unlikely(error)) | |
1204 | goto readpage_error; | |
1da177e4 LT |
1205 | if (!PageUptodate(page)) { |
1206 | if (page->mapping == NULL) { | |
1207 | /* | |
2ecdc82e | 1208 | * invalidate_mapping_pages got it |
1da177e4 LT |
1209 | */ |
1210 | unlock_page(page); | |
1211 | page_cache_release(page); | |
1212 | goto find_page; | |
1213 | } | |
1214 | unlock_page(page); | |
7ff81078 | 1215 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1216 | error = -EIO; |
1217 | goto readpage_error; | |
1da177e4 LT |
1218 | } |
1219 | unlock_page(page); | |
1220 | } | |
1221 | ||
1da177e4 LT |
1222 | goto page_ok; |
1223 | ||
1224 | readpage_error: | |
1225 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1226 | desc->error = error; | |
1227 | page_cache_release(page); | |
1228 | goto out; | |
1229 | ||
1230 | no_cached_page: | |
1231 | /* | |
1232 | * Ok, it wasn't cached, so we need to create a new | |
1233 | * page.. | |
1234 | */ | |
eb2be189 NP |
1235 | page = page_cache_alloc_cold(mapping); |
1236 | if (!page) { | |
1237 | desc->error = -ENOMEM; | |
1238 | goto out; | |
1da177e4 | 1239 | } |
eb2be189 | 1240 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1241 | index, GFP_KERNEL); |
1242 | if (error) { | |
eb2be189 | 1243 | page_cache_release(page); |
1da177e4 LT |
1244 | if (error == -EEXIST) |
1245 | goto find_page; | |
1246 | desc->error = error; | |
1247 | goto out; | |
1248 | } | |
1da177e4 LT |
1249 | goto readpage; |
1250 | } | |
1251 | ||
1252 | out: | |
7ff81078 FW |
1253 | ra->prev_pos = prev_index; |
1254 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1255 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1256 | |
f4e6b498 | 1257 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1258 | file_accessed(filp); |
1da177e4 | 1259 | } |
1da177e4 LT |
1260 | |
1261 | int file_read_actor(read_descriptor_t *desc, struct page *page, | |
1262 | unsigned long offset, unsigned long size) | |
1263 | { | |
1264 | char *kaddr; | |
1265 | unsigned long left, count = desc->count; | |
1266 | ||
1267 | if (size > count) | |
1268 | size = count; | |
1269 | ||
1270 | /* | |
1271 | * Faults on the destination of a read are common, so do it before | |
1272 | * taking the kmap. | |
1273 | */ | |
1274 | if (!fault_in_pages_writeable(desc->arg.buf, size)) { | |
1275 | kaddr = kmap_atomic(page, KM_USER0); | |
1276 | left = __copy_to_user_inatomic(desc->arg.buf, | |
1277 | kaddr + offset, size); | |
1278 | kunmap_atomic(kaddr, KM_USER0); | |
1279 | if (left == 0) | |
1280 | goto success; | |
1281 | } | |
1282 | ||
1283 | /* Do it the slow way */ | |
1284 | kaddr = kmap(page); | |
1285 | left = __copy_to_user(desc->arg.buf, kaddr + offset, size); | |
1286 | kunmap(page); | |
1287 | ||
1288 | if (left) { | |
1289 | size -= left; | |
1290 | desc->error = -EFAULT; | |
1291 | } | |
1292 | success: | |
1293 | desc->count = count - size; | |
1294 | desc->written += size; | |
1295 | desc->arg.buf += size; | |
1296 | return size; | |
1297 | } | |
1298 | ||
0ceb3314 DM |
1299 | /* |
1300 | * Performs necessary checks before doing a write | |
1301 | * @iov: io vector request | |
1302 | * @nr_segs: number of segments in the iovec | |
1303 | * @count: number of bytes to write | |
1304 | * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE | |
1305 | * | |
1306 | * Adjust number of segments and amount of bytes to write (nr_segs should be | |
1307 | * properly initialized first). Returns appropriate error code that caller | |
1308 | * should return or zero in case that write should be allowed. | |
1309 | */ | |
1310 | int generic_segment_checks(const struct iovec *iov, | |
1311 | unsigned long *nr_segs, size_t *count, int access_flags) | |
1312 | { | |
1313 | unsigned long seg; | |
1314 | size_t cnt = 0; | |
1315 | for (seg = 0; seg < *nr_segs; seg++) { | |
1316 | const struct iovec *iv = &iov[seg]; | |
1317 | ||
1318 | /* | |
1319 | * If any segment has a negative length, or the cumulative | |
1320 | * length ever wraps negative then return -EINVAL. | |
1321 | */ | |
1322 | cnt += iv->iov_len; | |
1323 | if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) | |
1324 | return -EINVAL; | |
1325 | if (access_ok(access_flags, iv->iov_base, iv->iov_len)) | |
1326 | continue; | |
1327 | if (seg == 0) | |
1328 | return -EFAULT; | |
1329 | *nr_segs = seg; | |
1330 | cnt -= iv->iov_len; /* This segment is no good */ | |
1331 | break; | |
1332 | } | |
1333 | *count = cnt; | |
1334 | return 0; | |
1335 | } | |
1336 | EXPORT_SYMBOL(generic_segment_checks); | |
1337 | ||
485bb99b | 1338 | /** |
b2abacf3 | 1339 | * generic_file_aio_read - generic filesystem read routine |
485bb99b RD |
1340 | * @iocb: kernel I/O control block |
1341 | * @iov: io vector request | |
1342 | * @nr_segs: number of segments in the iovec | |
b2abacf3 | 1343 | * @pos: current file position |
485bb99b | 1344 | * |
1da177e4 LT |
1345 | * This is the "read()" routine for all filesystems |
1346 | * that can use the page cache directly. | |
1347 | */ | |
1348 | ssize_t | |
543ade1f BP |
1349 | generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
1350 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
1351 | { |
1352 | struct file *filp = iocb->ki_filp; | |
1353 | ssize_t retval; | |
66f998f6 | 1354 | unsigned long seg = 0; |
1da177e4 | 1355 | size_t count; |
543ade1f | 1356 | loff_t *ppos = &iocb->ki_pos; |
55602dd6 | 1357 | struct blk_plug plug; |
1da177e4 LT |
1358 | |
1359 | count = 0; | |
0ceb3314 DM |
1360 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); |
1361 | if (retval) | |
1362 | return retval; | |
1da177e4 | 1363 | |
55602dd6 JA |
1364 | blk_start_plug(&plug); |
1365 | ||
1da177e4 LT |
1366 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ |
1367 | if (filp->f_flags & O_DIRECT) { | |
543ade1f | 1368 | loff_t size; |
1da177e4 LT |
1369 | struct address_space *mapping; |
1370 | struct inode *inode; | |
1371 | ||
1372 | mapping = filp->f_mapping; | |
1373 | inode = mapping->host; | |
1da177e4 LT |
1374 | if (!count) |
1375 | goto out; /* skip atime */ | |
1376 | size = i_size_read(inode); | |
1377 | if (pos < size) { | |
48b47c56 NP |
1378 | retval = filemap_write_and_wait_range(mapping, pos, |
1379 | pos + iov_length(iov, nr_segs) - 1); | |
a969e903 CH |
1380 | if (!retval) { |
1381 | retval = mapping->a_ops->direct_IO(READ, iocb, | |
1382 | iov, pos, nr_segs); | |
1383 | } | |
66f998f6 | 1384 | if (retval > 0) { |
1da177e4 | 1385 | *ppos = pos + retval; |
66f998f6 JB |
1386 | count -= retval; |
1387 | } | |
1388 | ||
1389 | /* | |
1390 | * Btrfs can have a short DIO read if we encounter | |
1391 | * compressed extents, so if there was an error, or if | |
1392 | * we've already read everything we wanted to, or if | |
1393 | * there was a short read because we hit EOF, go ahead | |
1394 | * and return. Otherwise fallthrough to buffered io for | |
1395 | * the rest of the read. | |
1396 | */ | |
1397 | if (retval < 0 || !count || *ppos >= size) { | |
11fa977e HD |
1398 | file_accessed(filp); |
1399 | goto out; | |
1400 | } | |
0e0bcae3 | 1401 | } |
1da177e4 LT |
1402 | } |
1403 | ||
66f998f6 | 1404 | count = retval; |
11fa977e HD |
1405 | for (seg = 0; seg < nr_segs; seg++) { |
1406 | read_descriptor_t desc; | |
66f998f6 JB |
1407 | loff_t offset = 0; |
1408 | ||
1409 | /* | |
1410 | * If we did a short DIO read we need to skip the section of the | |
1411 | * iov that we've already read data into. | |
1412 | */ | |
1413 | if (count) { | |
1414 | if (count > iov[seg].iov_len) { | |
1415 | count -= iov[seg].iov_len; | |
1416 | continue; | |
1417 | } | |
1418 | offset = count; | |
1419 | count = 0; | |
1420 | } | |
1da177e4 | 1421 | |
11fa977e | 1422 | desc.written = 0; |
66f998f6 JB |
1423 | desc.arg.buf = iov[seg].iov_base + offset; |
1424 | desc.count = iov[seg].iov_len - offset; | |
11fa977e HD |
1425 | if (desc.count == 0) |
1426 | continue; | |
1427 | desc.error = 0; | |
1428 | do_generic_file_read(filp, ppos, &desc, file_read_actor); | |
1429 | retval += desc.written; | |
1430 | if (desc.error) { | |
1431 | retval = retval ?: desc.error; | |
1432 | break; | |
1da177e4 | 1433 | } |
11fa977e HD |
1434 | if (desc.count > 0) |
1435 | break; | |
1da177e4 LT |
1436 | } |
1437 | out: | |
55602dd6 | 1438 | blk_finish_plug(&plug); |
1da177e4 LT |
1439 | return retval; |
1440 | } | |
1da177e4 LT |
1441 | EXPORT_SYMBOL(generic_file_aio_read); |
1442 | ||
1da177e4 LT |
1443 | static ssize_t |
1444 | do_readahead(struct address_space *mapping, struct file *filp, | |
57f6b96c | 1445 | pgoff_t index, unsigned long nr) |
1da177e4 LT |
1446 | { |
1447 | if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage) | |
1448 | return -EINVAL; | |
1449 | ||
f7e839dd | 1450 | force_page_cache_readahead(mapping, filp, index, nr); |
1da177e4 LT |
1451 | return 0; |
1452 | } | |
1453 | ||
6673e0c3 | 1454 | SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count) |
1da177e4 LT |
1455 | { |
1456 | ssize_t ret; | |
1457 | struct file *file; | |
1458 | ||
1459 | ret = -EBADF; | |
1460 | file = fget(fd); | |
1461 | if (file) { | |
1462 | if (file->f_mode & FMODE_READ) { | |
1463 | struct address_space *mapping = file->f_mapping; | |
57f6b96c FW |
1464 | pgoff_t start = offset >> PAGE_CACHE_SHIFT; |
1465 | pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
1466 | unsigned long len = end - start + 1; |
1467 | ret = do_readahead(mapping, file, start, len); | |
1468 | } | |
1469 | fput(file); | |
1470 | } | |
1471 | return ret; | |
1472 | } | |
6673e0c3 HC |
1473 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
1474 | asmlinkage long SyS_readahead(long fd, loff_t offset, long count) | |
1475 | { | |
1476 | return SYSC_readahead((int) fd, offset, (size_t) count); | |
1477 | } | |
1478 | SYSCALL_ALIAS(sys_readahead, SyS_readahead); | |
1479 | #endif | |
1da177e4 LT |
1480 | |
1481 | #ifdef CONFIG_MMU | |
485bb99b RD |
1482 | /** |
1483 | * page_cache_read - adds requested page to the page cache if not already there | |
1484 | * @file: file to read | |
1485 | * @offset: page index | |
1486 | * | |
1da177e4 LT |
1487 | * This adds the requested page to the page cache if it isn't already there, |
1488 | * and schedules an I/O to read in its contents from disk. | |
1489 | */ | |
920c7a5d | 1490 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1491 | { |
1492 | struct address_space *mapping = file->f_mapping; | |
1493 | struct page *page; | |
994fc28c | 1494 | int ret; |
1da177e4 | 1495 | |
994fc28c ZB |
1496 | do { |
1497 | page = page_cache_alloc_cold(mapping); | |
1498 | if (!page) | |
1499 | return -ENOMEM; | |
1500 | ||
1501 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); | |
1502 | if (ret == 0) | |
1503 | ret = mapping->a_ops->readpage(file, page); | |
1504 | else if (ret == -EEXIST) | |
1505 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1506 | |
1da177e4 | 1507 | page_cache_release(page); |
1da177e4 | 1508 | |
994fc28c ZB |
1509 | } while (ret == AOP_TRUNCATED_PAGE); |
1510 | ||
1511 | return ret; | |
1da177e4 LT |
1512 | } |
1513 | ||
1514 | #define MMAP_LOTSAMISS (100) | |
1515 | ||
ef00e08e LT |
1516 | /* |
1517 | * Synchronous readahead happens when we don't even find | |
1518 | * a page in the page cache at all. | |
1519 | */ | |
1520 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1521 | struct file_ra_state *ra, | |
1522 | struct file *file, | |
1523 | pgoff_t offset) | |
1524 | { | |
1525 | unsigned long ra_pages; | |
1526 | struct address_space *mapping = file->f_mapping; | |
1527 | ||
1528 | /* If we don't want any read-ahead, don't bother */ | |
1529 | if (VM_RandomReadHint(vma)) | |
1530 | return; | |
1531 | ||
70ac23cf WF |
1532 | if (VM_SequentialReadHint(vma) || |
1533 | offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) { | |
7ffc59b4 WF |
1534 | page_cache_sync_readahead(mapping, ra, file, offset, |
1535 | ra->ra_pages); | |
ef00e08e LT |
1536 | return; |
1537 | } | |
1538 | ||
1539 | if (ra->mmap_miss < INT_MAX) | |
1540 | ra->mmap_miss++; | |
1541 | ||
1542 | /* | |
1543 | * Do we miss much more than hit in this file? If so, | |
1544 | * stop bothering with read-ahead. It will only hurt. | |
1545 | */ | |
1546 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1547 | return; | |
1548 | ||
d30a1100 WF |
1549 | /* |
1550 | * mmap read-around | |
1551 | */ | |
ef00e08e LT |
1552 | ra_pages = max_sane_readahead(ra->ra_pages); |
1553 | if (ra_pages) { | |
d30a1100 WF |
1554 | ra->start = max_t(long, 0, offset - ra_pages/2); |
1555 | ra->size = ra_pages; | |
1556 | ra->async_size = 0; | |
1557 | ra_submit(ra, mapping, file); | |
ef00e08e LT |
1558 | } |
1559 | } | |
1560 | ||
1561 | /* | |
1562 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1563 | * so we want to possibly extend the readahead further.. | |
1564 | */ | |
1565 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1566 | struct file_ra_state *ra, | |
1567 | struct file *file, | |
1568 | struct page *page, | |
1569 | pgoff_t offset) | |
1570 | { | |
1571 | struct address_space *mapping = file->f_mapping; | |
1572 | ||
1573 | /* If we don't want any read-ahead, don't bother */ | |
1574 | if (VM_RandomReadHint(vma)) | |
1575 | return; | |
1576 | if (ra->mmap_miss > 0) | |
1577 | ra->mmap_miss--; | |
1578 | if (PageReadahead(page)) | |
2fad6f5d WF |
1579 | page_cache_async_readahead(mapping, ra, file, |
1580 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1581 | } |
1582 | ||
485bb99b | 1583 | /** |
54cb8821 | 1584 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1585 | * @vma: vma in which the fault was taken |
1586 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1587 | * |
54cb8821 | 1588 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1589 | * mapped memory region to read in file data during a page fault. |
1590 | * | |
1591 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1592 | * it in the page cache, and handles the special cases reasonably without | |
1593 | * having a lot of duplicated code. | |
1594 | */ | |
d0217ac0 | 1595 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1596 | { |
1597 | int error; | |
54cb8821 | 1598 | struct file *file = vma->vm_file; |
1da177e4 LT |
1599 | struct address_space *mapping = file->f_mapping; |
1600 | struct file_ra_state *ra = &file->f_ra; | |
1601 | struct inode *inode = mapping->host; | |
ef00e08e | 1602 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 1603 | struct page *page; |
2004dc8e | 1604 | pgoff_t size; |
83c54070 | 1605 | int ret = 0; |
1da177e4 | 1606 | |
1da177e4 | 1607 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1608 | if (offset >= size) |
5307cc1a | 1609 | return VM_FAULT_SIGBUS; |
1da177e4 | 1610 | |
1da177e4 LT |
1611 | /* |
1612 | * Do we have something in the page cache already? | |
1613 | */ | |
ef00e08e LT |
1614 | page = find_get_page(mapping, offset); |
1615 | if (likely(page)) { | |
1da177e4 | 1616 | /* |
ef00e08e LT |
1617 | * We found the page, so try async readahead before |
1618 | * waiting for the lock. | |
1da177e4 | 1619 | */ |
ef00e08e | 1620 | do_async_mmap_readahead(vma, ra, file, page, offset); |
ef00e08e LT |
1621 | } else { |
1622 | /* No page in the page cache at all */ | |
1623 | do_sync_mmap_readahead(vma, ra, file, offset); | |
1624 | count_vm_event(PGMAJFAULT); | |
1625 | ret = VM_FAULT_MAJOR; | |
1626 | retry_find: | |
b522c94d | 1627 | page = find_get_page(mapping, offset); |
1da177e4 LT |
1628 | if (!page) |
1629 | goto no_cached_page; | |
1630 | } | |
1631 | ||
d88c0922 ML |
1632 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
1633 | page_cache_release(page); | |
d065bd81 | 1634 | return ret | VM_FAULT_RETRY; |
d88c0922 | 1635 | } |
b522c94d ML |
1636 | |
1637 | /* Did it get truncated? */ | |
1638 | if (unlikely(page->mapping != mapping)) { | |
1639 | unlock_page(page); | |
1640 | put_page(page); | |
1641 | goto retry_find; | |
1642 | } | |
1643 | VM_BUG_ON(page->index != offset); | |
1644 | ||
1da177e4 | 1645 | /* |
d00806b1 NP |
1646 | * We have a locked page in the page cache, now we need to check |
1647 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1648 | */ |
d00806b1 | 1649 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1650 | goto page_not_uptodate; |
1651 | ||
ef00e08e LT |
1652 | /* |
1653 | * Found the page and have a reference on it. | |
1654 | * We must recheck i_size under page lock. | |
1655 | */ | |
d00806b1 | 1656 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1657 | if (unlikely(offset >= size)) { |
d00806b1 | 1658 | unlock_page(page); |
745ad48e | 1659 | page_cache_release(page); |
5307cc1a | 1660 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1661 | } |
1662 | ||
ef00e08e | 1663 | ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT; |
d0217ac0 | 1664 | vmf->page = page; |
83c54070 | 1665 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 1666 | |
1da177e4 LT |
1667 | no_cached_page: |
1668 | /* | |
1669 | * We're only likely to ever get here if MADV_RANDOM is in | |
1670 | * effect. | |
1671 | */ | |
ef00e08e | 1672 | error = page_cache_read(file, offset); |
1da177e4 LT |
1673 | |
1674 | /* | |
1675 | * The page we want has now been added to the page cache. | |
1676 | * In the unlikely event that someone removed it in the | |
1677 | * meantime, we'll just come back here and read it again. | |
1678 | */ | |
1679 | if (error >= 0) | |
1680 | goto retry_find; | |
1681 | ||
1682 | /* | |
1683 | * An error return from page_cache_read can result if the | |
1684 | * system is low on memory, or a problem occurs while trying | |
1685 | * to schedule I/O. | |
1686 | */ | |
1687 | if (error == -ENOMEM) | |
d0217ac0 NP |
1688 | return VM_FAULT_OOM; |
1689 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1690 | |
1691 | page_not_uptodate: | |
1da177e4 LT |
1692 | /* |
1693 | * Umm, take care of errors if the page isn't up-to-date. | |
1694 | * Try to re-read it _once_. We do this synchronously, | |
1695 | * because there really aren't any performance issues here | |
1696 | * and we need to check for errors. | |
1697 | */ | |
1da177e4 | 1698 | ClearPageError(page); |
994fc28c | 1699 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
1700 | if (!error) { |
1701 | wait_on_page_locked(page); | |
1702 | if (!PageUptodate(page)) | |
1703 | error = -EIO; | |
1704 | } | |
d00806b1 NP |
1705 | page_cache_release(page); |
1706 | ||
1707 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 1708 | goto retry_find; |
1da177e4 | 1709 | |
d00806b1 | 1710 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 1711 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 1712 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
1713 | } |
1714 | EXPORT_SYMBOL(filemap_fault); | |
1715 | ||
f0f37e2f | 1716 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 1717 | .fault = filemap_fault, |
1da177e4 LT |
1718 | }; |
1719 | ||
1720 | /* This is used for a general mmap of a disk file */ | |
1721 | ||
1722 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1723 | { | |
1724 | struct address_space *mapping = file->f_mapping; | |
1725 | ||
1726 | if (!mapping->a_ops->readpage) | |
1727 | return -ENOEXEC; | |
1728 | file_accessed(file); | |
1729 | vma->vm_ops = &generic_file_vm_ops; | |
d0217ac0 | 1730 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
1731 | return 0; |
1732 | } | |
1da177e4 LT |
1733 | |
1734 | /* | |
1735 | * This is for filesystems which do not implement ->writepage. | |
1736 | */ | |
1737 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
1738 | { | |
1739 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
1740 | return -EINVAL; | |
1741 | return generic_file_mmap(file, vma); | |
1742 | } | |
1743 | #else | |
1744 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1745 | { | |
1746 | return -ENOSYS; | |
1747 | } | |
1748 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
1749 | { | |
1750 | return -ENOSYS; | |
1751 | } | |
1752 | #endif /* CONFIG_MMU */ | |
1753 | ||
1754 | EXPORT_SYMBOL(generic_file_mmap); | |
1755 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
1756 | ||
6fe6900e | 1757 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 1758 | pgoff_t index, |
1da177e4 | 1759 | int (*filler)(void *,struct page*), |
0531b2aa LT |
1760 | void *data, |
1761 | gfp_t gfp) | |
1da177e4 | 1762 | { |
eb2be189 | 1763 | struct page *page; |
1da177e4 LT |
1764 | int err; |
1765 | repeat: | |
1766 | page = find_get_page(mapping, index); | |
1767 | if (!page) { | |
0531b2aa | 1768 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
1769 | if (!page) |
1770 | return ERR_PTR(-ENOMEM); | |
1771 | err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL); | |
1772 | if (unlikely(err)) { | |
1773 | page_cache_release(page); | |
1774 | if (err == -EEXIST) | |
1775 | goto repeat; | |
1da177e4 | 1776 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
1777 | return ERR_PTR(err); |
1778 | } | |
1da177e4 LT |
1779 | err = filler(data, page); |
1780 | if (err < 0) { | |
1781 | page_cache_release(page); | |
1782 | page = ERR_PTR(err); | |
1783 | } | |
1784 | } | |
1da177e4 LT |
1785 | return page; |
1786 | } | |
1787 | ||
0531b2aa | 1788 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 1789 | pgoff_t index, |
1da177e4 | 1790 | int (*filler)(void *,struct page*), |
0531b2aa LT |
1791 | void *data, |
1792 | gfp_t gfp) | |
1793 | ||
1da177e4 LT |
1794 | { |
1795 | struct page *page; | |
1796 | int err; | |
1797 | ||
1798 | retry: | |
0531b2aa | 1799 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 1800 | if (IS_ERR(page)) |
c855ff37 | 1801 | return page; |
1da177e4 LT |
1802 | if (PageUptodate(page)) |
1803 | goto out; | |
1804 | ||
1805 | lock_page(page); | |
1806 | if (!page->mapping) { | |
1807 | unlock_page(page); | |
1808 | page_cache_release(page); | |
1809 | goto retry; | |
1810 | } | |
1811 | if (PageUptodate(page)) { | |
1812 | unlock_page(page); | |
1813 | goto out; | |
1814 | } | |
1815 | err = filler(data, page); | |
1816 | if (err < 0) { | |
1817 | page_cache_release(page); | |
c855ff37 | 1818 | return ERR_PTR(err); |
1da177e4 | 1819 | } |
c855ff37 | 1820 | out: |
6fe6900e NP |
1821 | mark_page_accessed(page); |
1822 | return page; | |
1823 | } | |
0531b2aa LT |
1824 | |
1825 | /** | |
1826 | * read_cache_page_async - read into page cache, fill it if needed | |
1827 | * @mapping: the page's address_space | |
1828 | * @index: the page index | |
1829 | * @filler: function to perform the read | |
1830 | * @data: destination for read data | |
1831 | * | |
1832 | * Same as read_cache_page, but don't wait for page to become unlocked | |
1833 | * after submitting it to the filler. | |
1834 | * | |
1835 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1836 | * not set, try to fill the page but don't wait for it to become unlocked. | |
1837 | * | |
1838 | * If the page does not get brought uptodate, return -EIO. | |
1839 | */ | |
1840 | struct page *read_cache_page_async(struct address_space *mapping, | |
1841 | pgoff_t index, | |
1842 | int (*filler)(void *,struct page*), | |
1843 | void *data) | |
1844 | { | |
1845 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
1846 | } | |
6fe6900e NP |
1847 | EXPORT_SYMBOL(read_cache_page_async); |
1848 | ||
0531b2aa LT |
1849 | static struct page *wait_on_page_read(struct page *page) |
1850 | { | |
1851 | if (!IS_ERR(page)) { | |
1852 | wait_on_page_locked(page); | |
1853 | if (!PageUptodate(page)) { | |
1854 | page_cache_release(page); | |
1855 | page = ERR_PTR(-EIO); | |
1856 | } | |
1857 | } | |
1858 | return page; | |
1859 | } | |
1860 | ||
1861 | /** | |
1862 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
1863 | * @mapping: the page's address_space | |
1864 | * @index: the page index | |
1865 | * @gfp: the page allocator flags to use if allocating | |
1866 | * | |
1867 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
1868 | * any new page allocations done using the specified allocation flags. Note | |
1869 | * that the Radix tree operations will still use GFP_KERNEL, so you can't | |
1870 | * expect to do this atomically or anything like that - but you can pass in | |
1871 | * other page requirements. | |
1872 | * | |
1873 | * If the page does not get brought uptodate, return -EIO. | |
1874 | */ | |
1875 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
1876 | pgoff_t index, | |
1877 | gfp_t gfp) | |
1878 | { | |
1879 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
1880 | ||
1881 | return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp)); | |
1882 | } | |
1883 | EXPORT_SYMBOL(read_cache_page_gfp); | |
1884 | ||
6fe6900e NP |
1885 | /** |
1886 | * read_cache_page - read into page cache, fill it if needed | |
1887 | * @mapping: the page's address_space | |
1888 | * @index: the page index | |
1889 | * @filler: function to perform the read | |
1890 | * @data: destination for read data | |
1891 | * | |
1892 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1893 | * not set, try to fill the page then wait for it to become unlocked. | |
1894 | * | |
1895 | * If the page does not get brought uptodate, return -EIO. | |
1896 | */ | |
1897 | struct page *read_cache_page(struct address_space *mapping, | |
57f6b96c | 1898 | pgoff_t index, |
6fe6900e NP |
1899 | int (*filler)(void *,struct page*), |
1900 | void *data) | |
1901 | { | |
0531b2aa | 1902 | return wait_on_page_read(read_cache_page_async(mapping, index, filler, data)); |
1da177e4 | 1903 | } |
1da177e4 LT |
1904 | EXPORT_SYMBOL(read_cache_page); |
1905 | ||
1da177e4 LT |
1906 | /* |
1907 | * The logic we want is | |
1908 | * | |
1909 | * if suid or (sgid and xgrp) | |
1910 | * remove privs | |
1911 | */ | |
01de85e0 | 1912 | int should_remove_suid(struct dentry *dentry) |
1da177e4 LT |
1913 | { |
1914 | mode_t mode = dentry->d_inode->i_mode; | |
1915 | int kill = 0; | |
1da177e4 LT |
1916 | |
1917 | /* suid always must be killed */ | |
1918 | if (unlikely(mode & S_ISUID)) | |
1919 | kill = ATTR_KILL_SUID; | |
1920 | ||
1921 | /* | |
1922 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1923 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1924 | */ | |
1925 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1926 | kill |= ATTR_KILL_SGID; | |
1927 | ||
7f5ff766 | 1928 | if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) |
01de85e0 | 1929 | return kill; |
1da177e4 | 1930 | |
01de85e0 JA |
1931 | return 0; |
1932 | } | |
d23a147b | 1933 | EXPORT_SYMBOL(should_remove_suid); |
01de85e0 | 1934 | |
7f3d4ee1 | 1935 | static int __remove_suid(struct dentry *dentry, int kill) |
01de85e0 JA |
1936 | { |
1937 | struct iattr newattrs; | |
1938 | ||
1939 | newattrs.ia_valid = ATTR_FORCE | kill; | |
1940 | return notify_change(dentry, &newattrs); | |
1941 | } | |
1942 | ||
2f1936b8 | 1943 | int file_remove_suid(struct file *file) |
01de85e0 | 1944 | { |
2f1936b8 | 1945 | struct dentry *dentry = file->f_path.dentry; |
b5376771 SH |
1946 | int killsuid = should_remove_suid(dentry); |
1947 | int killpriv = security_inode_need_killpriv(dentry); | |
1948 | int error = 0; | |
01de85e0 | 1949 | |
b5376771 SH |
1950 | if (killpriv < 0) |
1951 | return killpriv; | |
1952 | if (killpriv) | |
1953 | error = security_inode_killpriv(dentry); | |
1954 | if (!error && killsuid) | |
1955 | error = __remove_suid(dentry, killsuid); | |
01de85e0 | 1956 | |
b5376771 | 1957 | return error; |
1da177e4 | 1958 | } |
2f1936b8 | 1959 | EXPORT_SYMBOL(file_remove_suid); |
1da177e4 | 1960 | |
2f718ffc | 1961 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
1da177e4 LT |
1962 | const struct iovec *iov, size_t base, size_t bytes) |
1963 | { | |
f1800536 | 1964 | size_t copied = 0, left = 0; |
1da177e4 LT |
1965 | |
1966 | while (bytes) { | |
1967 | char __user *buf = iov->iov_base + base; | |
1968 | int copy = min(bytes, iov->iov_len - base); | |
1969 | ||
1970 | base = 0; | |
f1800536 | 1971 | left = __copy_from_user_inatomic(vaddr, buf, copy); |
1da177e4 LT |
1972 | copied += copy; |
1973 | bytes -= copy; | |
1974 | vaddr += copy; | |
1975 | iov++; | |
1976 | ||
01408c49 | 1977 | if (unlikely(left)) |
1da177e4 | 1978 | break; |
1da177e4 LT |
1979 | } |
1980 | return copied - left; | |
1981 | } | |
1982 | ||
2f718ffc NP |
1983 | /* |
1984 | * Copy as much as we can into the page and return the number of bytes which | |
af901ca1 | 1985 | * were successfully copied. If a fault is encountered then return the number of |
2f718ffc NP |
1986 | * bytes which were copied. |
1987 | */ | |
1988 | size_t iov_iter_copy_from_user_atomic(struct page *page, | |
1989 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
1990 | { | |
1991 | char *kaddr; | |
1992 | size_t copied; | |
1993 | ||
1994 | BUG_ON(!in_atomic()); | |
1995 | kaddr = kmap_atomic(page, KM_USER0); | |
1996 | if (likely(i->nr_segs == 1)) { | |
1997 | int left; | |
1998 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 1999 | left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); |
2f718ffc NP |
2000 | copied = bytes - left; |
2001 | } else { | |
2002 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2003 | i->iov, i->iov_offset, bytes); | |
2004 | } | |
2005 | kunmap_atomic(kaddr, KM_USER0); | |
2006 | ||
2007 | return copied; | |
2008 | } | |
89e10787 | 2009 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); |
2f718ffc NP |
2010 | |
2011 | /* | |
2012 | * This has the same sideeffects and return value as | |
2013 | * iov_iter_copy_from_user_atomic(). | |
2014 | * The difference is that it attempts to resolve faults. | |
2015 | * Page must not be locked. | |
2016 | */ | |
2017 | size_t iov_iter_copy_from_user(struct page *page, | |
2018 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
2019 | { | |
2020 | char *kaddr; | |
2021 | size_t copied; | |
2022 | ||
2023 | kaddr = kmap(page); | |
2024 | if (likely(i->nr_segs == 1)) { | |
2025 | int left; | |
2026 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 2027 | left = __copy_from_user(kaddr + offset, buf, bytes); |
2f718ffc NP |
2028 | copied = bytes - left; |
2029 | } else { | |
2030 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2031 | i->iov, i->iov_offset, bytes); | |
2032 | } | |
2033 | kunmap(page); | |
2034 | return copied; | |
2035 | } | |
89e10787 | 2036 | EXPORT_SYMBOL(iov_iter_copy_from_user); |
2f718ffc | 2037 | |
f7009264 | 2038 | void iov_iter_advance(struct iov_iter *i, size_t bytes) |
2f718ffc | 2039 | { |
f7009264 NP |
2040 | BUG_ON(i->count < bytes); |
2041 | ||
2f718ffc NP |
2042 | if (likely(i->nr_segs == 1)) { |
2043 | i->iov_offset += bytes; | |
f7009264 | 2044 | i->count -= bytes; |
2f718ffc NP |
2045 | } else { |
2046 | const struct iovec *iov = i->iov; | |
2047 | size_t base = i->iov_offset; | |
2048 | ||
124d3b70 NP |
2049 | /* |
2050 | * The !iov->iov_len check ensures we skip over unlikely | |
f7009264 | 2051 | * zero-length segments (without overruning the iovec). |
124d3b70 | 2052 | */ |
94ad374a | 2053 | while (bytes || unlikely(i->count && !iov->iov_len)) { |
f7009264 | 2054 | int copy; |
2f718ffc | 2055 | |
f7009264 NP |
2056 | copy = min(bytes, iov->iov_len - base); |
2057 | BUG_ON(!i->count || i->count < copy); | |
2058 | i->count -= copy; | |
2f718ffc NP |
2059 | bytes -= copy; |
2060 | base += copy; | |
2061 | if (iov->iov_len == base) { | |
2062 | iov++; | |
2063 | base = 0; | |
2064 | } | |
2065 | } | |
2066 | i->iov = iov; | |
2067 | i->iov_offset = base; | |
2068 | } | |
2069 | } | |
89e10787 | 2070 | EXPORT_SYMBOL(iov_iter_advance); |
2f718ffc | 2071 | |
afddba49 NP |
2072 | /* |
2073 | * Fault in the first iovec of the given iov_iter, to a maximum length | |
2074 | * of bytes. Returns 0 on success, or non-zero if the memory could not be | |
2075 | * accessed (ie. because it is an invalid address). | |
2076 | * | |
2077 | * writev-intensive code may want this to prefault several iovecs -- that | |
2078 | * would be possible (callers must not rely on the fact that _only_ the | |
2079 | * first iovec will be faulted with the current implementation). | |
2080 | */ | |
2081 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) | |
2f718ffc | 2082 | { |
2f718ffc | 2083 | char __user *buf = i->iov->iov_base + i->iov_offset; |
afddba49 NP |
2084 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); |
2085 | return fault_in_pages_readable(buf, bytes); | |
2f718ffc | 2086 | } |
89e10787 | 2087 | EXPORT_SYMBOL(iov_iter_fault_in_readable); |
2f718ffc NP |
2088 | |
2089 | /* | |
2090 | * Return the count of just the current iov_iter segment. | |
2091 | */ | |
2092 | size_t iov_iter_single_seg_count(struct iov_iter *i) | |
2093 | { | |
2094 | const struct iovec *iov = i->iov; | |
2095 | if (i->nr_segs == 1) | |
2096 | return i->count; | |
2097 | else | |
2098 | return min(i->count, iov->iov_len - i->iov_offset); | |
2099 | } | |
89e10787 | 2100 | EXPORT_SYMBOL(iov_iter_single_seg_count); |
2f718ffc | 2101 | |
1da177e4 LT |
2102 | /* |
2103 | * Performs necessary checks before doing a write | |
2104 | * | |
485bb99b | 2105 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2106 | * Returns appropriate error code that caller should return or |
2107 | * zero in case that write should be allowed. | |
2108 | */ | |
2109 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) | |
2110 | { | |
2111 | struct inode *inode = file->f_mapping->host; | |
59e99e5b | 2112 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
1da177e4 LT |
2113 | |
2114 | if (unlikely(*pos < 0)) | |
2115 | return -EINVAL; | |
2116 | ||
1da177e4 LT |
2117 | if (!isblk) { |
2118 | /* FIXME: this is for backwards compatibility with 2.4 */ | |
2119 | if (file->f_flags & O_APPEND) | |
2120 | *pos = i_size_read(inode); | |
2121 | ||
2122 | if (limit != RLIM_INFINITY) { | |
2123 | if (*pos >= limit) { | |
2124 | send_sig(SIGXFSZ, current, 0); | |
2125 | return -EFBIG; | |
2126 | } | |
2127 | if (*count > limit - (typeof(limit))*pos) { | |
2128 | *count = limit - (typeof(limit))*pos; | |
2129 | } | |
2130 | } | |
2131 | } | |
2132 | ||
2133 | /* | |
2134 | * LFS rule | |
2135 | */ | |
2136 | if (unlikely(*pos + *count > MAX_NON_LFS && | |
2137 | !(file->f_flags & O_LARGEFILE))) { | |
2138 | if (*pos >= MAX_NON_LFS) { | |
1da177e4 LT |
2139 | return -EFBIG; |
2140 | } | |
2141 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { | |
2142 | *count = MAX_NON_LFS - (unsigned long)*pos; | |
2143 | } | |
2144 | } | |
2145 | ||
2146 | /* | |
2147 | * Are we about to exceed the fs block limit ? | |
2148 | * | |
2149 | * If we have written data it becomes a short write. If we have | |
2150 | * exceeded without writing data we send a signal and return EFBIG. | |
2151 | * Linus frestrict idea will clean these up nicely.. | |
2152 | */ | |
2153 | if (likely(!isblk)) { | |
2154 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { | |
2155 | if (*count || *pos > inode->i_sb->s_maxbytes) { | |
1da177e4 LT |
2156 | return -EFBIG; |
2157 | } | |
2158 | /* zero-length writes at ->s_maxbytes are OK */ | |
2159 | } | |
2160 | ||
2161 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) | |
2162 | *count = inode->i_sb->s_maxbytes - *pos; | |
2163 | } else { | |
9361401e | 2164 | #ifdef CONFIG_BLOCK |
1da177e4 LT |
2165 | loff_t isize; |
2166 | if (bdev_read_only(I_BDEV(inode))) | |
2167 | return -EPERM; | |
2168 | isize = i_size_read(inode); | |
2169 | if (*pos >= isize) { | |
2170 | if (*count || *pos > isize) | |
2171 | return -ENOSPC; | |
2172 | } | |
2173 | ||
2174 | if (*pos + *count > isize) | |
2175 | *count = isize - *pos; | |
9361401e DH |
2176 | #else |
2177 | return -EPERM; | |
2178 | #endif | |
1da177e4 LT |
2179 | } |
2180 | return 0; | |
2181 | } | |
2182 | EXPORT_SYMBOL(generic_write_checks); | |
2183 | ||
afddba49 NP |
2184 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2185 | loff_t pos, unsigned len, unsigned flags, | |
2186 | struct page **pagep, void **fsdata) | |
2187 | { | |
2188 | const struct address_space_operations *aops = mapping->a_ops; | |
2189 | ||
4e02ed4b | 2190 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2191 | pagep, fsdata); |
afddba49 NP |
2192 | } |
2193 | EXPORT_SYMBOL(pagecache_write_begin); | |
2194 | ||
2195 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2196 | loff_t pos, unsigned len, unsigned copied, | |
2197 | struct page *page, void *fsdata) | |
2198 | { | |
2199 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2200 | |
4e02ed4b NP |
2201 | mark_page_accessed(page); |
2202 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); | |
afddba49 NP |
2203 | } |
2204 | EXPORT_SYMBOL(pagecache_write_end); | |
2205 | ||
1da177e4 LT |
2206 | ssize_t |
2207 | generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, | |
2208 | unsigned long *nr_segs, loff_t pos, loff_t *ppos, | |
2209 | size_t count, size_t ocount) | |
2210 | { | |
2211 | struct file *file = iocb->ki_filp; | |
2212 | struct address_space *mapping = file->f_mapping; | |
2213 | struct inode *inode = mapping->host; | |
2214 | ssize_t written; | |
a969e903 CH |
2215 | size_t write_len; |
2216 | pgoff_t end; | |
1da177e4 LT |
2217 | |
2218 | if (count != ocount) | |
2219 | *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); | |
2220 | ||
a969e903 CH |
2221 | write_len = iov_length(iov, *nr_segs); |
2222 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; | |
a969e903 | 2223 | |
48b47c56 | 2224 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2225 | if (written) |
2226 | goto out; | |
2227 | ||
2228 | /* | |
2229 | * After a write we want buffered reads to be sure to go to disk to get | |
2230 | * the new data. We invalidate clean cached page from the region we're | |
2231 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2232 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2233 | */ |
2234 | if (mapping->nrpages) { | |
2235 | written = invalidate_inode_pages2_range(mapping, | |
2236 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2237 | /* |
2238 | * If a page can not be invalidated, return 0 to fall back | |
2239 | * to buffered write. | |
2240 | */ | |
2241 | if (written) { | |
2242 | if (written == -EBUSY) | |
2243 | return 0; | |
a969e903 | 2244 | goto out; |
6ccfa806 | 2245 | } |
a969e903 CH |
2246 | } |
2247 | ||
2248 | written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs); | |
2249 | ||
2250 | /* | |
2251 | * Finally, try again to invalidate clean pages which might have been | |
2252 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2253 | * if the source of the write was an mmap'ed region of the file | |
2254 | * we're writing. Either one is a pretty crazy thing to do, | |
2255 | * so we don't support it 100%. If this invalidation | |
2256 | * fails, tough, the write still worked... | |
2257 | */ | |
2258 | if (mapping->nrpages) { | |
2259 | invalidate_inode_pages2_range(mapping, | |
2260 | pos >> PAGE_CACHE_SHIFT, end); | |
2261 | } | |
2262 | ||
1da177e4 | 2263 | if (written > 0) { |
0116651c NK |
2264 | pos += written; |
2265 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { | |
2266 | i_size_write(inode, pos); | |
1da177e4 LT |
2267 | mark_inode_dirty(inode); |
2268 | } | |
0116651c | 2269 | *ppos = pos; |
1da177e4 | 2270 | } |
a969e903 | 2271 | out: |
1da177e4 LT |
2272 | return written; |
2273 | } | |
2274 | EXPORT_SYMBOL(generic_file_direct_write); | |
2275 | ||
eb2be189 NP |
2276 | /* |
2277 | * Find or create a page at the given pagecache position. Return the locked | |
2278 | * page. This function is specifically for buffered writes. | |
2279 | */ | |
54566b2c NP |
2280 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2281 | pgoff_t index, unsigned flags) | |
eb2be189 NP |
2282 | { |
2283 | int status; | |
2284 | struct page *page; | |
54566b2c NP |
2285 | gfp_t gfp_notmask = 0; |
2286 | if (flags & AOP_FLAG_NOFS) | |
2287 | gfp_notmask = __GFP_FS; | |
eb2be189 NP |
2288 | repeat: |
2289 | page = find_lock_page(mapping, index); | |
c585a267 | 2290 | if (page) |
eb2be189 NP |
2291 | return page; |
2292 | ||
54566b2c | 2293 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask); |
eb2be189 NP |
2294 | if (!page) |
2295 | return NULL; | |
54566b2c NP |
2296 | status = add_to_page_cache_lru(page, mapping, index, |
2297 | GFP_KERNEL & ~gfp_notmask); | |
eb2be189 NP |
2298 | if (unlikely(status)) { |
2299 | page_cache_release(page); | |
2300 | if (status == -EEXIST) | |
2301 | goto repeat; | |
2302 | return NULL; | |
2303 | } | |
2304 | return page; | |
2305 | } | |
54566b2c | 2306 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2307 | |
afddba49 NP |
2308 | static ssize_t generic_perform_write(struct file *file, |
2309 | struct iov_iter *i, loff_t pos) | |
2310 | { | |
2311 | struct address_space *mapping = file->f_mapping; | |
2312 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2313 | long status = 0; | |
2314 | ssize_t written = 0; | |
674b892e NP |
2315 | unsigned int flags = 0; |
2316 | ||
2317 | /* | |
2318 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2319 | */ | |
2320 | if (segment_eq(get_fs(), KERNEL_DS)) | |
2321 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
afddba49 NP |
2322 | |
2323 | do { | |
2324 | struct page *page; | |
afddba49 NP |
2325 | unsigned long offset; /* Offset into pagecache page */ |
2326 | unsigned long bytes; /* Bytes to write to page */ | |
2327 | size_t copied; /* Bytes copied from user */ | |
2328 | void *fsdata; | |
2329 | ||
2330 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2331 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2332 | iov_iter_count(i)); | |
2333 | ||
2334 | again: | |
2335 | ||
2336 | /* | |
2337 | * Bring in the user page that we will copy from _first_. | |
2338 | * Otherwise there's a nasty deadlock on copying from the | |
2339 | * same page as we're writing to, without it being marked | |
2340 | * up-to-date. | |
2341 | * | |
2342 | * Not only is this an optimisation, but it is also required | |
2343 | * to check that the address is actually valid, when atomic | |
2344 | * usercopies are used, below. | |
2345 | */ | |
2346 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2347 | status = -EFAULT; | |
2348 | break; | |
2349 | } | |
2350 | ||
674b892e | 2351 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 NP |
2352 | &page, &fsdata); |
2353 | if (unlikely(status)) | |
2354 | break; | |
2355 | ||
931e80e4 | 2356 | if (mapping_writably_mapped(mapping)) |
2357 | flush_dcache_page(page); | |
2358 | ||
afddba49 NP |
2359 | pagefault_disable(); |
2360 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); | |
2361 | pagefault_enable(); | |
2362 | flush_dcache_page(page); | |
2363 | ||
c8236db9 | 2364 | mark_page_accessed(page); |
afddba49 NP |
2365 | status = a_ops->write_end(file, mapping, pos, bytes, copied, |
2366 | page, fsdata); | |
2367 | if (unlikely(status < 0)) | |
2368 | break; | |
2369 | copied = status; | |
2370 | ||
2371 | cond_resched(); | |
2372 | ||
124d3b70 | 2373 | iov_iter_advance(i, copied); |
afddba49 NP |
2374 | if (unlikely(copied == 0)) { |
2375 | /* | |
2376 | * If we were unable to copy any data at all, we must | |
2377 | * fall back to a single segment length write. | |
2378 | * | |
2379 | * If we didn't fallback here, we could livelock | |
2380 | * because not all segments in the iov can be copied at | |
2381 | * once without a pagefault. | |
2382 | */ | |
2383 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2384 | iov_iter_single_seg_count(i)); | |
2385 | goto again; | |
2386 | } | |
afddba49 NP |
2387 | pos += copied; |
2388 | written += copied; | |
2389 | ||
2390 | balance_dirty_pages_ratelimited(mapping); | |
2391 | ||
2392 | } while (iov_iter_count(i)); | |
2393 | ||
2394 | return written ? written : status; | |
2395 | } | |
2396 | ||
2397 | ssize_t | |
2398 | generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, | |
2399 | unsigned long nr_segs, loff_t pos, loff_t *ppos, | |
2400 | size_t count, ssize_t written) | |
2401 | { | |
2402 | struct file *file = iocb->ki_filp; | |
afddba49 NP |
2403 | ssize_t status; |
2404 | struct iov_iter i; | |
2405 | ||
2406 | iov_iter_init(&i, iov, nr_segs, count, written); | |
4e02ed4b | 2407 | status = generic_perform_write(file, &i, pos); |
1da177e4 | 2408 | |
1da177e4 | 2409 | if (likely(status >= 0)) { |
afddba49 NP |
2410 | written += status; |
2411 | *ppos = pos + status; | |
1da177e4 LT |
2412 | } |
2413 | ||
1da177e4 LT |
2414 | return written ? written : status; |
2415 | } | |
2416 | EXPORT_SYMBOL(generic_file_buffered_write); | |
2417 | ||
e4dd9de3 JK |
2418 | /** |
2419 | * __generic_file_aio_write - write data to a file | |
2420 | * @iocb: IO state structure (file, offset, etc.) | |
2421 | * @iov: vector with data to write | |
2422 | * @nr_segs: number of segments in the vector | |
2423 | * @ppos: position where to write | |
2424 | * | |
2425 | * This function does all the work needed for actually writing data to a | |
2426 | * file. It does all basic checks, removes SUID from the file, updates | |
2427 | * modification times and calls proper subroutines depending on whether we | |
2428 | * do direct IO or a standard buffered write. | |
2429 | * | |
2430 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2431 | * object which does not need locking at all. | |
2432 | * | |
2433 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2434 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2435 | * avoid syncing under i_mutex. | |
2436 | */ | |
2437 | ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
2438 | unsigned long nr_segs, loff_t *ppos) | |
1da177e4 LT |
2439 | { |
2440 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2441 | struct address_space * mapping = file->f_mapping; |
1da177e4 LT |
2442 | size_t ocount; /* original count */ |
2443 | size_t count; /* after file limit checks */ | |
2444 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2445 | loff_t pos; |
2446 | ssize_t written; | |
2447 | ssize_t err; | |
2448 | ||
2449 | ocount = 0; | |
0ceb3314 DM |
2450 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
2451 | if (err) | |
2452 | return err; | |
1da177e4 LT |
2453 | |
2454 | count = ocount; | |
2455 | pos = *ppos; | |
2456 | ||
2457 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
2458 | ||
2459 | /* We can write back this queue in page reclaim */ | |
2460 | current->backing_dev_info = mapping->backing_dev_info; | |
2461 | written = 0; | |
2462 | ||
2463 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
2464 | if (err) | |
2465 | goto out; | |
2466 | ||
2467 | if (count == 0) | |
2468 | goto out; | |
2469 | ||
2f1936b8 | 2470 | err = file_remove_suid(file); |
1da177e4 LT |
2471 | if (err) |
2472 | goto out; | |
2473 | ||
870f4817 | 2474 | file_update_time(file); |
1da177e4 LT |
2475 | |
2476 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
2477 | if (unlikely(file->f_flags & O_DIRECT)) { | |
fb5527e6 JM |
2478 | loff_t endbyte; |
2479 | ssize_t written_buffered; | |
2480 | ||
2481 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, | |
2482 | ppos, count, ocount); | |
1da177e4 LT |
2483 | if (written < 0 || written == count) |
2484 | goto out; | |
2485 | /* | |
2486 | * direct-io write to a hole: fall through to buffered I/O | |
2487 | * for completing the rest of the request. | |
2488 | */ | |
2489 | pos += written; | |
2490 | count -= written; | |
fb5527e6 JM |
2491 | written_buffered = generic_file_buffered_write(iocb, iov, |
2492 | nr_segs, pos, ppos, count, | |
2493 | written); | |
2494 | /* | |
2495 | * If generic_file_buffered_write() retuned a synchronous error | |
2496 | * then we want to return the number of bytes which were | |
2497 | * direct-written, or the error code if that was zero. Note | |
2498 | * that this differs from normal direct-io semantics, which | |
2499 | * will return -EFOO even if some bytes were written. | |
2500 | */ | |
2501 | if (written_buffered < 0) { | |
2502 | err = written_buffered; | |
2503 | goto out; | |
2504 | } | |
1da177e4 | 2505 | |
fb5527e6 JM |
2506 | /* |
2507 | * We need to ensure that the page cache pages are written to | |
2508 | * disk and invalidated to preserve the expected O_DIRECT | |
2509 | * semantics. | |
2510 | */ | |
2511 | endbyte = pos + written_buffered - written - 1; | |
c05c4edd | 2512 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); |
fb5527e6 JM |
2513 | if (err == 0) { |
2514 | written = written_buffered; | |
2515 | invalidate_mapping_pages(mapping, | |
2516 | pos >> PAGE_CACHE_SHIFT, | |
2517 | endbyte >> PAGE_CACHE_SHIFT); | |
2518 | } else { | |
2519 | /* | |
2520 | * We don't know how much we wrote, so just return | |
2521 | * the number of bytes which were direct-written | |
2522 | */ | |
2523 | } | |
2524 | } else { | |
2525 | written = generic_file_buffered_write(iocb, iov, nr_segs, | |
2526 | pos, ppos, count, written); | |
2527 | } | |
1da177e4 LT |
2528 | out: |
2529 | current->backing_dev_info = NULL; | |
2530 | return written ? written : err; | |
2531 | } | |
e4dd9de3 JK |
2532 | EXPORT_SYMBOL(__generic_file_aio_write); |
2533 | ||
e4dd9de3 JK |
2534 | /** |
2535 | * generic_file_aio_write - write data to a file | |
2536 | * @iocb: IO state structure | |
2537 | * @iov: vector with data to write | |
2538 | * @nr_segs: number of segments in the vector | |
2539 | * @pos: position in file where to write | |
2540 | * | |
2541 | * This is a wrapper around __generic_file_aio_write() to be used by most | |
2542 | * filesystems. It takes care of syncing the file in case of O_SYNC file | |
2543 | * and acquires i_mutex as needed. | |
2544 | */ | |
027445c3 BP |
2545 | ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
2546 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
2547 | { |
2548 | struct file *file = iocb->ki_filp; | |
148f948b | 2549 | struct inode *inode = file->f_mapping->host; |
55602dd6 | 2550 | struct blk_plug plug; |
1da177e4 | 2551 | ssize_t ret; |
1da177e4 LT |
2552 | |
2553 | BUG_ON(iocb->ki_pos != pos); | |
2554 | ||
1b1dcc1b | 2555 | mutex_lock(&inode->i_mutex); |
55602dd6 | 2556 | blk_start_plug(&plug); |
e4dd9de3 | 2557 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); |
1b1dcc1b | 2558 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 2559 | |
148f948b | 2560 | if (ret > 0 || ret == -EIOCBQUEUED) { |
1da177e4 LT |
2561 | ssize_t err; |
2562 | ||
148f948b | 2563 | err = generic_write_sync(file, pos, ret); |
c7b50db2 | 2564 | if (err < 0 && ret > 0) |
1da177e4 LT |
2565 | ret = err; |
2566 | } | |
55602dd6 | 2567 | blk_finish_plug(&plug); |
1da177e4 LT |
2568 | return ret; |
2569 | } | |
2570 | EXPORT_SYMBOL(generic_file_aio_write); | |
2571 | ||
cf9a2ae8 DH |
2572 | /** |
2573 | * try_to_release_page() - release old fs-specific metadata on a page | |
2574 | * | |
2575 | * @page: the page which the kernel is trying to free | |
2576 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2577 | * | |
2578 | * The address_space is to try to release any data against the page | |
2579 | * (presumably at page->private). If the release was successful, return `1'. | |
2580 | * Otherwise return zero. | |
2581 | * | |
266cf658 DH |
2582 | * This may also be called if PG_fscache is set on a page, indicating that the |
2583 | * page is known to the local caching routines. | |
2584 | * | |
cf9a2ae8 | 2585 | * The @gfp_mask argument specifies whether I/O may be performed to release |
3f31fddf | 2586 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
cf9a2ae8 | 2587 | * |
cf9a2ae8 DH |
2588 | */ |
2589 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2590 | { | |
2591 | struct address_space * const mapping = page->mapping; | |
2592 | ||
2593 | BUG_ON(!PageLocked(page)); | |
2594 | if (PageWriteback(page)) | |
2595 | return 0; | |
2596 | ||
2597 | if (mapping && mapping->a_ops->releasepage) | |
2598 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2599 | return try_to_free_buffers(page); | |
2600 | } | |
2601 | ||
2602 | EXPORT_SYMBOL(try_to_release_page); |