]>
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 | */ | |
b95f1b31 | 12 | #include <linux/export.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> | |
44110fe3 | 32 | #include <linux/cpuset.h> |
2f718ffc | 33 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
8a9f3ccd | 34 | #include <linux/memcontrol.h> |
c515e1fd | 35 | #include <linux/cleancache.h> |
0f8053a5 NP |
36 | #include "internal.h" |
37 | ||
fe0bfaaf RJ |
38 | #define CREATE_TRACE_POINTS |
39 | #include <trace/events/filemap.h> | |
40 | ||
1da177e4 | 41 | /* |
1da177e4 LT |
42 | * FIXME: remove all knowledge of the buffer layer from the core VM |
43 | */ | |
148f948b | 44 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 45 | |
1da177e4 LT |
46 | #include <asm/mman.h> |
47 | ||
48 | /* | |
49 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
50 | * though. | |
51 | * | |
52 | * Shared mappings now work. 15.8.1995 Bruno. | |
53 | * | |
54 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
55 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
56 | * | |
57 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
58 | */ | |
59 | ||
60 | /* | |
61 | * Lock ordering: | |
62 | * | |
3d48ae45 | 63 | * ->i_mmap_mutex (truncate_pagecache) |
1da177e4 | 64 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 HD |
65 | * ->swap_lock (exclusive_swap_page, others) |
66 | * ->mapping->tree_lock | |
1da177e4 | 67 | * |
1b1dcc1b | 68 | * ->i_mutex |
3d48ae45 | 69 | * ->i_mmap_mutex (truncate->unmap_mapping_range) |
1da177e4 LT |
70 | * |
71 | * ->mmap_sem | |
3d48ae45 | 72 | * ->i_mmap_mutex |
b8072f09 | 73 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
74 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
75 | * | |
76 | * ->mmap_sem | |
77 | * ->lock_page (access_process_vm) | |
78 | * | |
82591e6e NP |
79 | * ->i_mutex (generic_file_buffered_write) |
80 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) | |
1da177e4 | 81 | * |
f758eeab | 82 | * bdi->wb.list_lock |
a66979ab | 83 | * sb_lock (fs/fs-writeback.c) |
1da177e4 LT |
84 | * ->mapping->tree_lock (__sync_single_inode) |
85 | * | |
3d48ae45 | 86 | * ->i_mmap_mutex |
1da177e4 LT |
87 | * ->anon_vma.lock (vma_adjust) |
88 | * | |
89 | * ->anon_vma.lock | |
b8072f09 | 90 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 91 | * |
b8072f09 | 92 | * ->page_table_lock or pte_lock |
5d337b91 | 93 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
94 | * ->private_lock (try_to_unmap_one) |
95 | * ->tree_lock (try_to_unmap_one) | |
96 | * ->zone.lru_lock (follow_page->mark_page_accessed) | |
053837fc | 97 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
1da177e4 LT |
98 | * ->private_lock (page_remove_rmap->set_page_dirty) |
99 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
f758eeab | 100 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 101 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
f758eeab | 102 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 103 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
104 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
105 | * | |
9a3c531d AK |
106 | * ->i_mmap_mutex |
107 | * ->tasklist_lock (memory_failure, collect_procs_ao) | |
1da177e4 LT |
108 | */ |
109 | ||
110 | /* | |
e64a782f | 111 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 112 | * sure the page is locked and that nobody else uses it - or that usage |
19fd6231 | 113 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 114 | */ |
e64a782f | 115 | void __delete_from_page_cache(struct page *page) |
1da177e4 LT |
116 | { |
117 | struct address_space *mapping = page->mapping; | |
118 | ||
fe0bfaaf | 119 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
120 | /* |
121 | * if we're uptodate, flush out into the cleancache, otherwise | |
122 | * invalidate any existing cleancache entries. We can't leave | |
123 | * stale data around in the cleancache once our page is gone | |
124 | */ | |
125 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
126 | cleancache_put_page(page); | |
127 | else | |
3167760f | 128 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 129 | |
1da177e4 LT |
130 | radix_tree_delete(&mapping->page_tree, page->index); |
131 | page->mapping = NULL; | |
b85e0eff | 132 | /* Leave page->index set: truncation lookup relies upon it */ |
1da177e4 | 133 | mapping->nrpages--; |
347ce434 | 134 | __dec_zone_page_state(page, NR_FILE_PAGES); |
4b02108a KM |
135 | if (PageSwapBacked(page)) |
136 | __dec_zone_page_state(page, NR_SHMEM); | |
45426812 | 137 | BUG_ON(page_mapped(page)); |
3a692790 LT |
138 | |
139 | /* | |
140 | * Some filesystems seem to re-dirty the page even after | |
141 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
142 | * | |
143 | * Fix it up by doing a final dirty accounting check after | |
144 | * having removed the page entirely. | |
145 | */ | |
146 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { | |
147 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
148 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
149 | } | |
1da177e4 LT |
150 | } |
151 | ||
702cfbf9 MK |
152 | /** |
153 | * delete_from_page_cache - delete page from page cache | |
154 | * @page: the page which the kernel is trying to remove from page cache | |
155 | * | |
156 | * This must be called only on pages that have been verified to be in the page | |
157 | * cache and locked. It will never put the page into the free list, the caller | |
158 | * has a reference on the page. | |
159 | */ | |
160 | void delete_from_page_cache(struct page *page) | |
1da177e4 LT |
161 | { |
162 | struct address_space *mapping = page->mapping; | |
6072d13c | 163 | void (*freepage)(struct page *); |
1da177e4 | 164 | |
cd7619d6 | 165 | BUG_ON(!PageLocked(page)); |
1da177e4 | 166 | |
6072d13c | 167 | freepage = mapping->a_ops->freepage; |
19fd6231 | 168 | spin_lock_irq(&mapping->tree_lock); |
e64a782f | 169 | __delete_from_page_cache(page); |
19fd6231 | 170 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 171 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
172 | |
173 | if (freepage) | |
174 | freepage(page); | |
97cecb5a MK |
175 | page_cache_release(page); |
176 | } | |
177 | EXPORT_SYMBOL(delete_from_page_cache); | |
178 | ||
7eaceacc | 179 | static int sleep_on_page(void *word) |
1da177e4 | 180 | { |
1da177e4 LT |
181 | io_schedule(); |
182 | return 0; | |
183 | } | |
184 | ||
7eaceacc | 185 | static int sleep_on_page_killable(void *word) |
2687a356 | 186 | { |
7eaceacc | 187 | sleep_on_page(word); |
2687a356 MW |
188 | return fatal_signal_pending(current) ? -EINTR : 0; |
189 | } | |
190 | ||
1da177e4 | 191 | /** |
485bb99b | 192 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
193 | * @mapping: address space structure to write |
194 | * @start: offset in bytes where the range starts | |
469eb4d0 | 195 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 196 | * @sync_mode: enable synchronous operation |
1da177e4 | 197 | * |
485bb99b RD |
198 | * Start writeback against all of a mapping's dirty pages that lie |
199 | * within the byte offsets <start, end> inclusive. | |
200 | * | |
1da177e4 | 201 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 202 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
203 | * these two operations is that if a dirty page/buffer is encountered, it must |
204 | * be waited upon, and not just skipped over. | |
205 | */ | |
ebcf28e1 AM |
206 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
207 | loff_t end, int sync_mode) | |
1da177e4 LT |
208 | { |
209 | int ret; | |
210 | struct writeback_control wbc = { | |
211 | .sync_mode = sync_mode, | |
05fe478d | 212 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
213 | .range_start = start, |
214 | .range_end = end, | |
1da177e4 LT |
215 | }; |
216 | ||
217 | if (!mapping_cap_writeback_dirty(mapping)) | |
218 | return 0; | |
219 | ||
220 | ret = do_writepages(mapping, &wbc); | |
221 | return ret; | |
222 | } | |
223 | ||
224 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
225 | int sync_mode) | |
226 | { | |
111ebb6e | 227 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
228 | } |
229 | ||
230 | int filemap_fdatawrite(struct address_space *mapping) | |
231 | { | |
232 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
233 | } | |
234 | EXPORT_SYMBOL(filemap_fdatawrite); | |
235 | ||
f4c0a0fd | 236 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 237 | loff_t end) |
1da177e4 LT |
238 | { |
239 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
240 | } | |
f4c0a0fd | 241 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 242 | |
485bb99b RD |
243 | /** |
244 | * filemap_flush - mostly a non-blocking flush | |
245 | * @mapping: target address_space | |
246 | * | |
1da177e4 LT |
247 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
248 | * purposes - I/O may not be started against all dirty pages. | |
249 | */ | |
250 | int filemap_flush(struct address_space *mapping) | |
251 | { | |
252 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
253 | } | |
254 | EXPORT_SYMBOL(filemap_flush); | |
255 | ||
485bb99b | 256 | /** |
94004ed7 CH |
257 | * filemap_fdatawait_range - wait for writeback to complete |
258 | * @mapping: address space structure to wait for | |
259 | * @start_byte: offset in bytes where the range starts | |
260 | * @end_byte: offset in bytes where the range ends (inclusive) | |
485bb99b | 261 | * |
94004ed7 CH |
262 | * Walk the list of under-writeback pages of the given address space |
263 | * in the given range and wait for all of them. | |
1da177e4 | 264 | */ |
94004ed7 CH |
265 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
266 | loff_t end_byte) | |
1da177e4 | 267 | { |
94004ed7 CH |
268 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
269 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
270 | struct pagevec pvec; |
271 | int nr_pages; | |
272 | int ret = 0; | |
1da177e4 | 273 | |
94004ed7 | 274 | if (end_byte < start_byte) |
1da177e4 LT |
275 | return 0; |
276 | ||
277 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
278 | while ((index <= end) && |
279 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
280 | PAGECACHE_TAG_WRITEBACK, | |
281 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
282 | unsigned i; | |
283 | ||
284 | for (i = 0; i < nr_pages; i++) { | |
285 | struct page *page = pvec.pages[i]; | |
286 | ||
287 | /* until radix tree lookup accepts end_index */ | |
288 | if (page->index > end) | |
289 | continue; | |
290 | ||
291 | wait_on_page_writeback(page); | |
212260aa | 292 | if (TestClearPageError(page)) |
1da177e4 LT |
293 | ret = -EIO; |
294 | } | |
295 | pagevec_release(&pvec); | |
296 | cond_resched(); | |
297 | } | |
298 | ||
299 | /* Check for outstanding write errors */ | |
300 | if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
301 | ret = -ENOSPC; | |
302 | if (test_and_clear_bit(AS_EIO, &mapping->flags)) | |
303 | ret = -EIO; | |
304 | ||
305 | return ret; | |
306 | } | |
d3bccb6f JK |
307 | EXPORT_SYMBOL(filemap_fdatawait_range); |
308 | ||
1da177e4 | 309 | /** |
485bb99b | 310 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 311 | * @mapping: address space structure to wait for |
485bb99b RD |
312 | * |
313 | * Walk the list of under-writeback pages of the given address space | |
314 | * and wait for all of them. | |
1da177e4 LT |
315 | */ |
316 | int filemap_fdatawait(struct address_space *mapping) | |
317 | { | |
318 | loff_t i_size = i_size_read(mapping->host); | |
319 | ||
320 | if (i_size == 0) | |
321 | return 0; | |
322 | ||
94004ed7 | 323 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
324 | } |
325 | EXPORT_SYMBOL(filemap_fdatawait); | |
326 | ||
327 | int filemap_write_and_wait(struct address_space *mapping) | |
328 | { | |
28fd1298 | 329 | int err = 0; |
1da177e4 LT |
330 | |
331 | if (mapping->nrpages) { | |
28fd1298 OH |
332 | err = filemap_fdatawrite(mapping); |
333 | /* | |
334 | * Even if the above returned error, the pages may be | |
335 | * written partially (e.g. -ENOSPC), so we wait for it. | |
336 | * But the -EIO is special case, it may indicate the worst | |
337 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
338 | */ | |
339 | if (err != -EIO) { | |
340 | int err2 = filemap_fdatawait(mapping); | |
341 | if (!err) | |
342 | err = err2; | |
343 | } | |
1da177e4 | 344 | } |
28fd1298 | 345 | return err; |
1da177e4 | 346 | } |
28fd1298 | 347 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 348 | |
485bb99b RD |
349 | /** |
350 | * filemap_write_and_wait_range - write out & wait on a file range | |
351 | * @mapping: the address_space for the pages | |
352 | * @lstart: offset in bytes where the range starts | |
353 | * @lend: offset in bytes where the range ends (inclusive) | |
354 | * | |
469eb4d0 AM |
355 | * Write out and wait upon file offsets lstart->lend, inclusive. |
356 | * | |
357 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
358 | * that this function can be used to write to the very end-of-file (end = -1). | |
359 | */ | |
1da177e4 LT |
360 | int filemap_write_and_wait_range(struct address_space *mapping, |
361 | loff_t lstart, loff_t lend) | |
362 | { | |
28fd1298 | 363 | int err = 0; |
1da177e4 LT |
364 | |
365 | if (mapping->nrpages) { | |
28fd1298 OH |
366 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
367 | WB_SYNC_ALL); | |
368 | /* See comment of filemap_write_and_wait() */ | |
369 | if (err != -EIO) { | |
94004ed7 CH |
370 | int err2 = filemap_fdatawait_range(mapping, |
371 | lstart, lend); | |
28fd1298 OH |
372 | if (!err) |
373 | err = err2; | |
374 | } | |
1da177e4 | 375 | } |
28fd1298 | 376 | return err; |
1da177e4 | 377 | } |
f6995585 | 378 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 379 | |
ef6a3c63 MS |
380 | /** |
381 | * replace_page_cache_page - replace a pagecache page with a new one | |
382 | * @old: page to be replaced | |
383 | * @new: page to replace with | |
384 | * @gfp_mask: allocation mode | |
385 | * | |
386 | * This function replaces a page in the pagecache with a new one. On | |
387 | * success it acquires the pagecache reference for the new page and | |
388 | * drops it for the old page. Both the old and new pages must be | |
389 | * locked. This function does not add the new page to the LRU, the | |
390 | * caller must do that. | |
391 | * | |
392 | * The remove + add is atomic. The only way this function can fail is | |
393 | * memory allocation failure. | |
394 | */ | |
395 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
396 | { | |
397 | int error; | |
ef6a3c63 MS |
398 | |
399 | VM_BUG_ON(!PageLocked(old)); | |
400 | VM_BUG_ON(!PageLocked(new)); | |
401 | VM_BUG_ON(new->mapping); | |
402 | ||
ef6a3c63 MS |
403 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
404 | if (!error) { | |
405 | struct address_space *mapping = old->mapping; | |
406 | void (*freepage)(struct page *); | |
407 | ||
408 | pgoff_t offset = old->index; | |
409 | freepage = mapping->a_ops->freepage; | |
410 | ||
411 | page_cache_get(new); | |
412 | new->mapping = mapping; | |
413 | new->index = offset; | |
414 | ||
415 | spin_lock_irq(&mapping->tree_lock); | |
e64a782f | 416 | __delete_from_page_cache(old); |
ef6a3c63 MS |
417 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
418 | BUG_ON(error); | |
419 | mapping->nrpages++; | |
420 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
421 | if (PageSwapBacked(new)) | |
422 | __inc_zone_page_state(new, NR_SHMEM); | |
423 | spin_unlock_irq(&mapping->tree_lock); | |
ab936cbc KH |
424 | /* mem_cgroup codes must not be called under tree_lock */ |
425 | mem_cgroup_replace_page_cache(old, new); | |
ef6a3c63 MS |
426 | radix_tree_preload_end(); |
427 | if (freepage) | |
428 | freepage(old); | |
429 | page_cache_release(old); | |
ef6a3c63 MS |
430 | } |
431 | ||
432 | return error; | |
433 | } | |
434 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
435 | ||
485bb99b | 436 | /** |
e286781d | 437 | * add_to_page_cache_locked - add a locked page to the pagecache |
485bb99b RD |
438 | * @page: page to add |
439 | * @mapping: the page's address_space | |
440 | * @offset: page index | |
441 | * @gfp_mask: page allocation mode | |
442 | * | |
e286781d | 443 | * This function is used to add a page to the pagecache. It must be locked. |
1da177e4 LT |
444 | * This function does not add the page to the LRU. The caller must do that. |
445 | */ | |
e286781d | 446 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, |
6daa0e28 | 447 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 448 | { |
e286781d NP |
449 | int error; |
450 | ||
451 | VM_BUG_ON(!PageLocked(page)); | |
31475dd6 | 452 | VM_BUG_ON(PageSwapBacked(page)); |
e286781d NP |
453 | |
454 | error = mem_cgroup_cache_charge(page, current->mm, | |
2c26fdd7 | 455 | gfp_mask & GFP_RECLAIM_MASK); |
35c754d7 BS |
456 | if (error) |
457 | goto out; | |
1da177e4 | 458 | |
35c754d7 | 459 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
1da177e4 | 460 | if (error == 0) { |
e286781d NP |
461 | page_cache_get(page); |
462 | page->mapping = mapping; | |
463 | page->index = offset; | |
464 | ||
19fd6231 | 465 | spin_lock_irq(&mapping->tree_lock); |
1da177e4 | 466 | error = radix_tree_insert(&mapping->page_tree, offset, page); |
e286781d | 467 | if (likely(!error)) { |
1da177e4 | 468 | mapping->nrpages++; |
347ce434 | 469 | __inc_zone_page_state(page, NR_FILE_PAGES); |
e767e056 | 470 | spin_unlock_irq(&mapping->tree_lock); |
fe0bfaaf | 471 | trace_mm_filemap_add_to_page_cache(page); |
e286781d NP |
472 | } else { |
473 | page->mapping = NULL; | |
b85e0eff | 474 | /* Leave page->index set: truncation relies upon it */ |
e767e056 | 475 | spin_unlock_irq(&mapping->tree_lock); |
69029cd5 | 476 | mem_cgroup_uncharge_cache_page(page); |
e286781d NP |
477 | page_cache_release(page); |
478 | } | |
1da177e4 | 479 | radix_tree_preload_end(); |
35c754d7 | 480 | } else |
69029cd5 | 481 | mem_cgroup_uncharge_cache_page(page); |
8a9f3ccd | 482 | out: |
1da177e4 LT |
483 | return error; |
484 | } | |
e286781d | 485 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
486 | |
487 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 488 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 489 | { |
4f98a2fe RR |
490 | int ret; |
491 | ||
4f98a2fe | 492 | ret = add_to_page_cache(page, mapping, offset, gfp_mask); |
31475dd6 HD |
493 | if (ret == 0) |
494 | lru_cache_add_file(page); | |
1da177e4 LT |
495 | return ret; |
496 | } | |
18bc0bbd | 497 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 498 | |
44110fe3 | 499 | #ifdef CONFIG_NUMA |
2ae88149 | 500 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 501 | { |
c0ff7453 MX |
502 | int n; |
503 | struct page *page; | |
504 | ||
44110fe3 | 505 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
506 | unsigned int cpuset_mems_cookie; |
507 | do { | |
508 | cpuset_mems_cookie = get_mems_allowed(); | |
509 | n = cpuset_mem_spread_node(); | |
510 | page = alloc_pages_exact_node(n, gfp, 0); | |
511 | } while (!put_mems_allowed(cpuset_mems_cookie) && !page); | |
512 | ||
c0ff7453 | 513 | return page; |
44110fe3 | 514 | } |
2ae88149 | 515 | return alloc_pages(gfp, 0); |
44110fe3 | 516 | } |
2ae88149 | 517 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
518 | #endif |
519 | ||
1da177e4 LT |
520 | /* |
521 | * In order to wait for pages to become available there must be | |
522 | * waitqueues associated with pages. By using a hash table of | |
523 | * waitqueues where the bucket discipline is to maintain all | |
524 | * waiters on the same queue and wake all when any of the pages | |
525 | * become available, and for the woken contexts to check to be | |
526 | * sure the appropriate page became available, this saves space | |
527 | * at a cost of "thundering herd" phenomena during rare hash | |
528 | * collisions. | |
529 | */ | |
530 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
531 | { | |
532 | const struct zone *zone = page_zone(page); | |
533 | ||
534 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
535 | } | |
536 | ||
537 | static inline void wake_up_page(struct page *page, int bit) | |
538 | { | |
539 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); | |
540 | } | |
541 | ||
920c7a5d | 542 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
543 | { |
544 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
545 | ||
546 | if (test_bit(bit_nr, &page->flags)) | |
7eaceacc | 547 | __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
548 | TASK_UNINTERRUPTIBLE); |
549 | } | |
550 | EXPORT_SYMBOL(wait_on_page_bit); | |
551 | ||
f62e00cc KM |
552 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
553 | { | |
554 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
555 | ||
556 | if (!test_bit(bit_nr, &page->flags)) | |
557 | return 0; | |
558 | ||
559 | return __wait_on_bit(page_waitqueue(page), &wait, | |
560 | sleep_on_page_killable, TASK_KILLABLE); | |
561 | } | |
562 | ||
385e1ca5 DH |
563 | /** |
564 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
565 | * @page: Page defining the wait queue of interest |
566 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
567 | * |
568 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
569 | */ | |
570 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
571 | { | |
572 | wait_queue_head_t *q = page_waitqueue(page); | |
573 | unsigned long flags; | |
574 | ||
575 | spin_lock_irqsave(&q->lock, flags); | |
576 | __add_wait_queue(q, waiter); | |
577 | spin_unlock_irqrestore(&q->lock, flags); | |
578 | } | |
579 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
580 | ||
1da177e4 | 581 | /** |
485bb99b | 582 | * unlock_page - unlock a locked page |
1da177e4 LT |
583 | * @page: the page |
584 | * | |
585 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
586 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
587 | * mechananism between PageLocked pages and PageWriteback pages is shared. | |
588 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. | |
589 | * | |
8413ac9d NP |
590 | * The mb is necessary to enforce ordering between the clear_bit and the read |
591 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 592 | */ |
920c7a5d | 593 | void unlock_page(struct page *page) |
1da177e4 | 594 | { |
8413ac9d NP |
595 | VM_BUG_ON(!PageLocked(page)); |
596 | clear_bit_unlock(PG_locked, &page->flags); | |
597 | smp_mb__after_clear_bit(); | |
1da177e4 LT |
598 | wake_up_page(page, PG_locked); |
599 | } | |
600 | EXPORT_SYMBOL(unlock_page); | |
601 | ||
485bb99b RD |
602 | /** |
603 | * end_page_writeback - end writeback against a page | |
604 | * @page: the page | |
1da177e4 LT |
605 | */ |
606 | void end_page_writeback(struct page *page) | |
607 | { | |
ac6aadb2 MS |
608 | if (TestClearPageReclaim(page)) |
609 | rotate_reclaimable_page(page); | |
610 | ||
611 | if (!test_clear_page_writeback(page)) | |
612 | BUG(); | |
613 | ||
1da177e4 LT |
614 | smp_mb__after_clear_bit(); |
615 | wake_up_page(page, PG_writeback); | |
616 | } | |
617 | EXPORT_SYMBOL(end_page_writeback); | |
618 | ||
485bb99b RD |
619 | /** |
620 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
621 | * @page: the page to lock | |
1da177e4 | 622 | */ |
920c7a5d | 623 | void __lock_page(struct page *page) |
1da177e4 LT |
624 | { |
625 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
626 | ||
7eaceacc | 627 | __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
628 | TASK_UNINTERRUPTIBLE); |
629 | } | |
630 | EXPORT_SYMBOL(__lock_page); | |
631 | ||
b5606c2d | 632 | int __lock_page_killable(struct page *page) |
2687a356 MW |
633 | { |
634 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
635 | ||
636 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
7eaceacc | 637 | sleep_on_page_killable, TASK_KILLABLE); |
2687a356 | 638 | } |
18bc0bbd | 639 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 640 | |
d065bd81 ML |
641 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
642 | unsigned int flags) | |
643 | { | |
37b23e05 KM |
644 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
645 | /* | |
646 | * CAUTION! In this case, mmap_sem is not released | |
647 | * even though return 0. | |
648 | */ | |
649 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
650 | return 0; | |
651 | ||
652 | up_read(&mm->mmap_sem); | |
653 | if (flags & FAULT_FLAG_KILLABLE) | |
654 | wait_on_page_locked_killable(page); | |
655 | else | |
318b275f | 656 | wait_on_page_locked(page); |
d065bd81 | 657 | return 0; |
37b23e05 KM |
658 | } else { |
659 | if (flags & FAULT_FLAG_KILLABLE) { | |
660 | int ret; | |
661 | ||
662 | ret = __lock_page_killable(page); | |
663 | if (ret) { | |
664 | up_read(&mm->mmap_sem); | |
665 | return 0; | |
666 | } | |
667 | } else | |
668 | __lock_page(page); | |
669 | return 1; | |
d065bd81 ML |
670 | } |
671 | } | |
672 | ||
485bb99b RD |
673 | /** |
674 | * find_get_page - find and get a page reference | |
675 | * @mapping: the address_space to search | |
676 | * @offset: the page index | |
677 | * | |
da6052f7 NP |
678 | * Is there a pagecache struct page at the given (mapping, offset) tuple? |
679 | * If yes, increment its refcount and return it; if no, return NULL. | |
1da177e4 | 680 | */ |
a60637c8 | 681 | struct page *find_get_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 682 | { |
a60637c8 | 683 | void **pagep; |
1da177e4 LT |
684 | struct page *page; |
685 | ||
a60637c8 NP |
686 | rcu_read_lock(); |
687 | repeat: | |
688 | page = NULL; | |
689 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
690 | if (pagep) { | |
691 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
692 | if (unlikely(!page)) |
693 | goto out; | |
a2c16d6c | 694 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
695 | if (radix_tree_deref_retry(page)) |
696 | goto repeat; | |
697 | /* | |
698 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
699 | * here as an exceptional entry: so return it without | |
700 | * attempting to raise page count. | |
701 | */ | |
702 | goto out; | |
a2c16d6c | 703 | } |
a60637c8 NP |
704 | if (!page_cache_get_speculative(page)) |
705 | goto repeat; | |
706 | ||
707 | /* | |
708 | * Has the page moved? | |
709 | * This is part of the lockless pagecache protocol. See | |
710 | * include/linux/pagemap.h for details. | |
711 | */ | |
712 | if (unlikely(page != *pagep)) { | |
713 | page_cache_release(page); | |
714 | goto repeat; | |
715 | } | |
716 | } | |
27d20fdd | 717 | out: |
a60637c8 NP |
718 | rcu_read_unlock(); |
719 | ||
1da177e4 LT |
720 | return page; |
721 | } | |
1da177e4 LT |
722 | EXPORT_SYMBOL(find_get_page); |
723 | ||
1da177e4 LT |
724 | /** |
725 | * find_lock_page - locate, pin and lock a pagecache page | |
67be2dd1 MW |
726 | * @mapping: the address_space to search |
727 | * @offset: the page index | |
1da177e4 LT |
728 | * |
729 | * Locates the desired pagecache page, locks it, increments its reference | |
730 | * count and returns its address. | |
731 | * | |
732 | * Returns zero if the page was not present. find_lock_page() may sleep. | |
733 | */ | |
a60637c8 | 734 | struct page *find_lock_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 LT |
735 | { |
736 | struct page *page; | |
737 | ||
1da177e4 | 738 | repeat: |
a60637c8 | 739 | page = find_get_page(mapping, offset); |
a2c16d6c | 740 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
741 | lock_page(page); |
742 | /* Has the page been truncated? */ | |
743 | if (unlikely(page->mapping != mapping)) { | |
744 | unlock_page(page); | |
745 | page_cache_release(page); | |
746 | goto repeat; | |
1da177e4 | 747 | } |
a60637c8 | 748 | VM_BUG_ON(page->index != offset); |
1da177e4 | 749 | } |
1da177e4 LT |
750 | return page; |
751 | } | |
1da177e4 LT |
752 | EXPORT_SYMBOL(find_lock_page); |
753 | ||
754 | /** | |
755 | * find_or_create_page - locate or add a pagecache page | |
67be2dd1 MW |
756 | * @mapping: the page's address_space |
757 | * @index: the page's index into the mapping | |
758 | * @gfp_mask: page allocation mode | |
1da177e4 LT |
759 | * |
760 | * Locates a page in the pagecache. If the page is not present, a new page | |
761 | * is allocated using @gfp_mask and is added to the pagecache and to the VM's | |
762 | * LRU list. The returned page is locked and has its reference count | |
763 | * incremented. | |
764 | * | |
765 | * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic | |
766 | * allocation! | |
767 | * | |
768 | * find_or_create_page() returns the desired page's address, or zero on | |
769 | * memory exhaustion. | |
770 | */ | |
771 | struct page *find_or_create_page(struct address_space *mapping, | |
57f6b96c | 772 | pgoff_t index, gfp_t gfp_mask) |
1da177e4 | 773 | { |
eb2be189 | 774 | struct page *page; |
1da177e4 LT |
775 | int err; |
776 | repeat: | |
777 | page = find_lock_page(mapping, index); | |
778 | if (!page) { | |
eb2be189 NP |
779 | page = __page_cache_alloc(gfp_mask); |
780 | if (!page) | |
781 | return NULL; | |
67d58ac4 NP |
782 | /* |
783 | * We want a regular kernel memory (not highmem or DMA etc) | |
784 | * allocation for the radix tree nodes, but we need to honour | |
785 | * the context-specific requirements the caller has asked for. | |
786 | * GFP_RECLAIM_MASK collects those requirements. | |
787 | */ | |
788 | err = add_to_page_cache_lru(page, mapping, index, | |
789 | (gfp_mask & GFP_RECLAIM_MASK)); | |
eb2be189 NP |
790 | if (unlikely(err)) { |
791 | page_cache_release(page); | |
792 | page = NULL; | |
793 | if (err == -EEXIST) | |
794 | goto repeat; | |
1da177e4 | 795 | } |
1da177e4 | 796 | } |
1da177e4 LT |
797 | return page; |
798 | } | |
1da177e4 LT |
799 | EXPORT_SYMBOL(find_or_create_page); |
800 | ||
801 | /** | |
802 | * find_get_pages - gang pagecache lookup | |
803 | * @mapping: The address_space to search | |
804 | * @start: The starting page index | |
805 | * @nr_pages: The maximum number of pages | |
806 | * @pages: Where the resulting pages are placed | |
807 | * | |
808 | * find_get_pages() will search for and return a group of up to | |
809 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
810 | * find_get_pages() takes a reference against the returned pages. | |
811 | * | |
812 | * The search returns a group of mapping-contiguous pages with ascending | |
813 | * indexes. There may be holes in the indices due to not-present pages. | |
814 | * | |
815 | * find_get_pages() returns the number of pages which were found. | |
816 | */ | |
817 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
818 | unsigned int nr_pages, struct page **pages) | |
819 | { | |
0fc9d104 KK |
820 | struct radix_tree_iter iter; |
821 | void **slot; | |
822 | unsigned ret = 0; | |
823 | ||
824 | if (unlikely(!nr_pages)) | |
825 | return 0; | |
a60637c8 NP |
826 | |
827 | rcu_read_lock(); | |
828 | restart: | |
0fc9d104 | 829 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
a60637c8 NP |
830 | struct page *page; |
831 | repeat: | |
0fc9d104 | 832 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
833 | if (unlikely(!page)) |
834 | continue; | |
9d8aa4ea | 835 | |
a2c16d6c | 836 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
837 | if (radix_tree_deref_retry(page)) { |
838 | /* | |
839 | * Transient condition which can only trigger | |
840 | * when entry at index 0 moves out of or back | |
841 | * to root: none yet gotten, safe to restart. | |
842 | */ | |
0fc9d104 | 843 | WARN_ON(iter.index); |
8079b1c8 HD |
844 | goto restart; |
845 | } | |
a2c16d6c | 846 | /* |
8079b1c8 HD |
847 | * Otherwise, shmem/tmpfs must be storing a swap entry |
848 | * here as an exceptional entry: so skip over it - | |
849 | * we only reach this from invalidate_mapping_pages(). | |
a2c16d6c | 850 | */ |
8079b1c8 | 851 | continue; |
27d20fdd | 852 | } |
a60637c8 NP |
853 | |
854 | if (!page_cache_get_speculative(page)) | |
855 | goto repeat; | |
856 | ||
857 | /* Has the page moved? */ | |
0fc9d104 | 858 | if (unlikely(page != *slot)) { |
a60637c8 NP |
859 | page_cache_release(page); |
860 | goto repeat; | |
861 | } | |
1da177e4 | 862 | |
a60637c8 | 863 | pages[ret] = page; |
0fc9d104 KK |
864 | if (++ret == nr_pages) |
865 | break; | |
a60637c8 | 866 | } |
5b280c0c | 867 | |
a60637c8 | 868 | rcu_read_unlock(); |
1da177e4 LT |
869 | return ret; |
870 | } | |
871 | ||
ebf43500 JA |
872 | /** |
873 | * find_get_pages_contig - gang contiguous pagecache lookup | |
874 | * @mapping: The address_space to search | |
875 | * @index: The starting page index | |
876 | * @nr_pages: The maximum number of pages | |
877 | * @pages: Where the resulting pages are placed | |
878 | * | |
879 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
880 | * that the returned number of pages are guaranteed to be contiguous. | |
881 | * | |
882 | * find_get_pages_contig() returns the number of pages which were found. | |
883 | */ | |
884 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
885 | unsigned int nr_pages, struct page **pages) | |
886 | { | |
0fc9d104 KK |
887 | struct radix_tree_iter iter; |
888 | void **slot; | |
889 | unsigned int ret = 0; | |
890 | ||
891 | if (unlikely(!nr_pages)) | |
892 | return 0; | |
a60637c8 NP |
893 | |
894 | rcu_read_lock(); | |
895 | restart: | |
0fc9d104 | 896 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
a60637c8 NP |
897 | struct page *page; |
898 | repeat: | |
0fc9d104 KK |
899 | page = radix_tree_deref_slot(slot); |
900 | /* The hole, there no reason to continue */ | |
a60637c8 | 901 | if (unlikely(!page)) |
0fc9d104 | 902 | break; |
9d8aa4ea | 903 | |
a2c16d6c | 904 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
905 | if (radix_tree_deref_retry(page)) { |
906 | /* | |
907 | * Transient condition which can only trigger | |
908 | * when entry at index 0 moves out of or back | |
909 | * to root: none yet gotten, safe to restart. | |
910 | */ | |
911 | goto restart; | |
912 | } | |
a2c16d6c | 913 | /* |
8079b1c8 HD |
914 | * Otherwise, shmem/tmpfs must be storing a swap entry |
915 | * here as an exceptional entry: so stop looking for | |
916 | * contiguous pages. | |
a2c16d6c | 917 | */ |
8079b1c8 | 918 | break; |
a2c16d6c | 919 | } |
ebf43500 | 920 | |
a60637c8 NP |
921 | if (!page_cache_get_speculative(page)) |
922 | goto repeat; | |
923 | ||
924 | /* Has the page moved? */ | |
0fc9d104 | 925 | if (unlikely(page != *slot)) { |
a60637c8 NP |
926 | page_cache_release(page); |
927 | goto repeat; | |
928 | } | |
929 | ||
9cbb4cb2 NP |
930 | /* |
931 | * must check mapping and index after taking the ref. | |
932 | * otherwise we can get both false positives and false | |
933 | * negatives, which is just confusing to the caller. | |
934 | */ | |
0fc9d104 | 935 | if (page->mapping == NULL || page->index != iter.index) { |
9cbb4cb2 NP |
936 | page_cache_release(page); |
937 | break; | |
938 | } | |
939 | ||
a60637c8 | 940 | pages[ret] = page; |
0fc9d104 KK |
941 | if (++ret == nr_pages) |
942 | break; | |
ebf43500 | 943 | } |
a60637c8 NP |
944 | rcu_read_unlock(); |
945 | return ret; | |
ebf43500 | 946 | } |
ef71c15c | 947 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 948 | |
485bb99b RD |
949 | /** |
950 | * find_get_pages_tag - find and return pages that match @tag | |
951 | * @mapping: the address_space to search | |
952 | * @index: the starting page index | |
953 | * @tag: the tag index | |
954 | * @nr_pages: the maximum number of pages | |
955 | * @pages: where the resulting pages are placed | |
956 | * | |
1da177e4 | 957 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 958 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
959 | */ |
960 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
961 | int tag, unsigned int nr_pages, struct page **pages) | |
962 | { | |
0fc9d104 KK |
963 | struct radix_tree_iter iter; |
964 | void **slot; | |
965 | unsigned ret = 0; | |
966 | ||
967 | if (unlikely(!nr_pages)) | |
968 | return 0; | |
a60637c8 NP |
969 | |
970 | rcu_read_lock(); | |
971 | restart: | |
0fc9d104 KK |
972 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
973 | &iter, *index, tag) { | |
a60637c8 NP |
974 | struct page *page; |
975 | repeat: | |
0fc9d104 | 976 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
977 | if (unlikely(!page)) |
978 | continue; | |
9d8aa4ea | 979 | |
a2c16d6c | 980 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
981 | if (radix_tree_deref_retry(page)) { |
982 | /* | |
983 | * Transient condition which can only trigger | |
984 | * when entry at index 0 moves out of or back | |
985 | * to root: none yet gotten, safe to restart. | |
986 | */ | |
987 | goto restart; | |
988 | } | |
a2c16d6c | 989 | /* |
8079b1c8 HD |
990 | * This function is never used on a shmem/tmpfs |
991 | * mapping, so a swap entry won't be found here. | |
a2c16d6c | 992 | */ |
8079b1c8 | 993 | BUG(); |
a2c16d6c | 994 | } |
a60637c8 NP |
995 | |
996 | if (!page_cache_get_speculative(page)) | |
997 | goto repeat; | |
998 | ||
999 | /* Has the page moved? */ | |
0fc9d104 | 1000 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1001 | page_cache_release(page); |
1002 | goto repeat; | |
1003 | } | |
1004 | ||
1005 | pages[ret] = page; | |
0fc9d104 KK |
1006 | if (++ret == nr_pages) |
1007 | break; | |
a60637c8 | 1008 | } |
5b280c0c | 1009 | |
a60637c8 | 1010 | rcu_read_unlock(); |
1da177e4 | 1011 | |
1da177e4 LT |
1012 | if (ret) |
1013 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1014 | |
1da177e4 LT |
1015 | return ret; |
1016 | } | |
ef71c15c | 1017 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1018 | |
485bb99b RD |
1019 | /** |
1020 | * grab_cache_page_nowait - returns locked page at given index in given cache | |
1021 | * @mapping: target address_space | |
1022 | * @index: the page index | |
1023 | * | |
72fd4a35 | 1024 | * Same as grab_cache_page(), but do not wait if the page is unavailable. |
1da177e4 LT |
1025 | * This is intended for speculative data generators, where the data can |
1026 | * be regenerated if the page couldn't be grabbed. This routine should | |
1027 | * be safe to call while holding the lock for another page. | |
1028 | * | |
1029 | * Clear __GFP_FS when allocating the page to avoid recursion into the fs | |
1030 | * and deadlock against the caller's locked page. | |
1031 | */ | |
1032 | struct page * | |
57f6b96c | 1033 | grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) |
1da177e4 LT |
1034 | { |
1035 | struct page *page = find_get_page(mapping, index); | |
1da177e4 LT |
1036 | |
1037 | if (page) { | |
529ae9aa | 1038 | if (trylock_page(page)) |
1da177e4 LT |
1039 | return page; |
1040 | page_cache_release(page); | |
1041 | return NULL; | |
1042 | } | |
2ae88149 | 1043 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); |
67d58ac4 | 1044 | if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) { |
1da177e4 LT |
1045 | page_cache_release(page); |
1046 | page = NULL; | |
1047 | } | |
1048 | return page; | |
1049 | } | |
1da177e4 LT |
1050 | EXPORT_SYMBOL(grab_cache_page_nowait); |
1051 | ||
76d42bd9 WF |
1052 | /* |
1053 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1054 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1055 | * | |
1056 | * ---R__________________________________________B__________ | |
1057 | * ^ reading here ^ bad block(assume 4k) | |
1058 | * | |
1059 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1060 | * => failing the whole request => read(R) => read(R+1) => | |
1061 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1062 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1063 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1064 | * | |
1065 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1066 | */ | |
1067 | static void shrink_readahead_size_eio(struct file *filp, | |
1068 | struct file_ra_state *ra) | |
1069 | { | |
76d42bd9 | 1070 | ra->ra_pages /= 4; |
76d42bd9 WF |
1071 | } |
1072 | ||
485bb99b | 1073 | /** |
36e78914 | 1074 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1075 | * @filp: the file to read |
1076 | * @ppos: current file position | |
1077 | * @desc: read_descriptor | |
1078 | * @actor: read method | |
1079 | * | |
1da177e4 | 1080 | * This is a generic file read routine, and uses the |
485bb99b | 1081 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1082 | * |
1083 | * This is really ugly. But the goto's actually try to clarify some | |
1084 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1085 | */ |
36e78914 CH |
1086 | static void do_generic_file_read(struct file *filp, loff_t *ppos, |
1087 | read_descriptor_t *desc, read_actor_t actor) | |
1da177e4 | 1088 | { |
36e78914 | 1089 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1090 | struct inode *inode = mapping->host; |
36e78914 | 1091 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1092 | pgoff_t index; |
1093 | pgoff_t last_index; | |
1094 | pgoff_t prev_index; | |
1095 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1096 | unsigned int prev_offset; |
1da177e4 | 1097 | int error; |
1da177e4 | 1098 | |
1da177e4 | 1099 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1100 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1101 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
1da177e4 LT |
1102 | last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1103 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1104 | ||
1da177e4 LT |
1105 | for (;;) { |
1106 | struct page *page; | |
57f6b96c | 1107 | pgoff_t end_index; |
a32ea1e1 | 1108 | loff_t isize; |
1da177e4 LT |
1109 | unsigned long nr, ret; |
1110 | ||
1da177e4 | 1111 | cond_resched(); |
1da177e4 LT |
1112 | find_page: |
1113 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1114 | if (!page) { |
cf914a7d | 1115 | page_cache_sync_readahead(mapping, |
7ff81078 | 1116 | ra, filp, |
3ea89ee8 FW |
1117 | index, last_index - index); |
1118 | page = find_get_page(mapping, index); | |
1119 | if (unlikely(page == NULL)) | |
1120 | goto no_cached_page; | |
1121 | } | |
1122 | if (PageReadahead(page)) { | |
cf914a7d | 1123 | page_cache_async_readahead(mapping, |
7ff81078 | 1124 | ra, filp, page, |
3ea89ee8 | 1125 | index, last_index - index); |
1da177e4 | 1126 | } |
8ab22b9a HH |
1127 | if (!PageUptodate(page)) { |
1128 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1129 | !mapping->a_ops->is_partially_uptodate) | |
1130 | goto page_not_up_to_date; | |
529ae9aa | 1131 | if (!trylock_page(page)) |
8ab22b9a | 1132 | goto page_not_up_to_date; |
8d056cb9 DH |
1133 | /* Did it get truncated before we got the lock? */ |
1134 | if (!page->mapping) | |
1135 | goto page_not_up_to_date_locked; | |
8ab22b9a HH |
1136 | if (!mapping->a_ops->is_partially_uptodate(page, |
1137 | desc, offset)) | |
1138 | goto page_not_up_to_date_locked; | |
1139 | unlock_page(page); | |
1140 | } | |
1da177e4 | 1141 | page_ok: |
a32ea1e1 N |
1142 | /* |
1143 | * i_size must be checked after we know the page is Uptodate. | |
1144 | * | |
1145 | * Checking i_size after the check allows us to calculate | |
1146 | * the correct value for "nr", which means the zero-filled | |
1147 | * part of the page is not copied back to userspace (unless | |
1148 | * another truncate extends the file - this is desired though). | |
1149 | */ | |
1150 | ||
1151 | isize = i_size_read(inode); | |
1152 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1153 | if (unlikely(!isize || index > end_index)) { | |
1154 | page_cache_release(page); | |
1155 | goto out; | |
1156 | } | |
1157 | ||
1158 | /* nr is the maximum number of bytes to copy from this page */ | |
1159 | nr = PAGE_CACHE_SIZE; | |
1160 | if (index == end_index) { | |
1161 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1162 | if (nr <= offset) { | |
1163 | page_cache_release(page); | |
1164 | goto out; | |
1165 | } | |
1166 | } | |
1167 | nr = nr - offset; | |
1da177e4 LT |
1168 | |
1169 | /* If users can be writing to this page using arbitrary | |
1170 | * virtual addresses, take care about potential aliasing | |
1171 | * before reading the page on the kernel side. | |
1172 | */ | |
1173 | if (mapping_writably_mapped(mapping)) | |
1174 | flush_dcache_page(page); | |
1175 | ||
1176 | /* | |
ec0f1637 JK |
1177 | * When a sequential read accesses a page several times, |
1178 | * only mark it as accessed the first time. | |
1da177e4 | 1179 | */ |
ec0f1637 | 1180 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1181 | mark_page_accessed(page); |
1182 | prev_index = index; | |
1183 | ||
1184 | /* | |
1185 | * Ok, we have the page, and it's up-to-date, so | |
1186 | * now we can copy it to user space... | |
1187 | * | |
1188 | * The actor routine returns how many bytes were actually used.. | |
1189 | * NOTE! This may not be the same as how much of a user buffer | |
1190 | * we filled up (we may be padding etc), so we can only update | |
1191 | * "pos" here (the actor routine has to update the user buffer | |
1192 | * pointers and the remaining count). | |
1193 | */ | |
1194 | ret = actor(desc, page, offset, nr); | |
1195 | offset += ret; | |
1196 | index += offset >> PAGE_CACHE_SHIFT; | |
1197 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1198 | prev_offset = offset; |
1da177e4 LT |
1199 | |
1200 | page_cache_release(page); | |
1201 | if (ret == nr && desc->count) | |
1202 | continue; | |
1203 | goto out; | |
1204 | ||
1205 | page_not_up_to_date: | |
1206 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1207 | error = lock_page_killable(page); |
1208 | if (unlikely(error)) | |
1209 | goto readpage_error; | |
1da177e4 | 1210 | |
8ab22b9a | 1211 | page_not_up_to_date_locked: |
da6052f7 | 1212 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1213 | if (!page->mapping) { |
1214 | unlock_page(page); | |
1215 | page_cache_release(page); | |
1216 | continue; | |
1217 | } | |
1218 | ||
1219 | /* Did somebody else fill it already? */ | |
1220 | if (PageUptodate(page)) { | |
1221 | unlock_page(page); | |
1222 | goto page_ok; | |
1223 | } | |
1224 | ||
1225 | readpage: | |
91803b49 JM |
1226 | /* |
1227 | * A previous I/O error may have been due to temporary | |
1228 | * failures, eg. multipath errors. | |
1229 | * PG_error will be set again if readpage fails. | |
1230 | */ | |
1231 | ClearPageError(page); | |
1da177e4 LT |
1232 | /* Start the actual read. The read will unlock the page. */ |
1233 | error = mapping->a_ops->readpage(filp, page); | |
1234 | ||
994fc28c ZB |
1235 | if (unlikely(error)) { |
1236 | if (error == AOP_TRUNCATED_PAGE) { | |
1237 | page_cache_release(page); | |
1238 | goto find_page; | |
1239 | } | |
1da177e4 | 1240 | goto readpage_error; |
994fc28c | 1241 | } |
1da177e4 LT |
1242 | |
1243 | if (!PageUptodate(page)) { | |
85462323 ON |
1244 | error = lock_page_killable(page); |
1245 | if (unlikely(error)) | |
1246 | goto readpage_error; | |
1da177e4 LT |
1247 | if (!PageUptodate(page)) { |
1248 | if (page->mapping == NULL) { | |
1249 | /* | |
2ecdc82e | 1250 | * invalidate_mapping_pages got it |
1da177e4 LT |
1251 | */ |
1252 | unlock_page(page); | |
1253 | page_cache_release(page); | |
1254 | goto find_page; | |
1255 | } | |
1256 | unlock_page(page); | |
7ff81078 | 1257 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1258 | error = -EIO; |
1259 | goto readpage_error; | |
1da177e4 LT |
1260 | } |
1261 | unlock_page(page); | |
1262 | } | |
1263 | ||
1da177e4 LT |
1264 | goto page_ok; |
1265 | ||
1266 | readpage_error: | |
1267 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1268 | desc->error = error; | |
1269 | page_cache_release(page); | |
1270 | goto out; | |
1271 | ||
1272 | no_cached_page: | |
1273 | /* | |
1274 | * Ok, it wasn't cached, so we need to create a new | |
1275 | * page.. | |
1276 | */ | |
eb2be189 NP |
1277 | page = page_cache_alloc_cold(mapping); |
1278 | if (!page) { | |
1279 | desc->error = -ENOMEM; | |
1280 | goto out; | |
1da177e4 | 1281 | } |
eb2be189 | 1282 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1283 | index, GFP_KERNEL); |
1284 | if (error) { | |
eb2be189 | 1285 | page_cache_release(page); |
1da177e4 LT |
1286 | if (error == -EEXIST) |
1287 | goto find_page; | |
1288 | desc->error = error; | |
1289 | goto out; | |
1290 | } | |
1da177e4 LT |
1291 | goto readpage; |
1292 | } | |
1293 | ||
1294 | out: | |
7ff81078 FW |
1295 | ra->prev_pos = prev_index; |
1296 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1297 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1298 | |
f4e6b498 | 1299 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1300 | file_accessed(filp); |
1da177e4 | 1301 | } |
1da177e4 LT |
1302 | |
1303 | int file_read_actor(read_descriptor_t *desc, struct page *page, | |
1304 | unsigned long offset, unsigned long size) | |
1305 | { | |
1306 | char *kaddr; | |
1307 | unsigned long left, count = desc->count; | |
1308 | ||
1309 | if (size > count) | |
1310 | size = count; | |
1311 | ||
1312 | /* | |
1313 | * Faults on the destination of a read are common, so do it before | |
1314 | * taking the kmap. | |
1315 | */ | |
1316 | if (!fault_in_pages_writeable(desc->arg.buf, size)) { | |
9b04c5fe | 1317 | kaddr = kmap_atomic(page); |
1da177e4 LT |
1318 | left = __copy_to_user_inatomic(desc->arg.buf, |
1319 | kaddr + offset, size); | |
9b04c5fe | 1320 | kunmap_atomic(kaddr); |
1da177e4 LT |
1321 | if (left == 0) |
1322 | goto success; | |
1323 | } | |
1324 | ||
1325 | /* Do it the slow way */ | |
1326 | kaddr = kmap(page); | |
1327 | left = __copy_to_user(desc->arg.buf, kaddr + offset, size); | |
1328 | kunmap(page); | |
1329 | ||
1330 | if (left) { | |
1331 | size -= left; | |
1332 | desc->error = -EFAULT; | |
1333 | } | |
1334 | success: | |
1335 | desc->count = count - size; | |
1336 | desc->written += size; | |
1337 | desc->arg.buf += size; | |
1338 | return size; | |
1339 | } | |
1340 | ||
0ceb3314 DM |
1341 | /* |
1342 | * Performs necessary checks before doing a write | |
1343 | * @iov: io vector request | |
1344 | * @nr_segs: number of segments in the iovec | |
1345 | * @count: number of bytes to write | |
1346 | * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE | |
1347 | * | |
1348 | * Adjust number of segments and amount of bytes to write (nr_segs should be | |
1349 | * properly initialized first). Returns appropriate error code that caller | |
1350 | * should return or zero in case that write should be allowed. | |
1351 | */ | |
1352 | int generic_segment_checks(const struct iovec *iov, | |
1353 | unsigned long *nr_segs, size_t *count, int access_flags) | |
1354 | { | |
1355 | unsigned long seg; | |
1356 | size_t cnt = 0; | |
1357 | for (seg = 0; seg < *nr_segs; seg++) { | |
1358 | const struct iovec *iv = &iov[seg]; | |
1359 | ||
1360 | /* | |
1361 | * If any segment has a negative length, or the cumulative | |
1362 | * length ever wraps negative then return -EINVAL. | |
1363 | */ | |
1364 | cnt += iv->iov_len; | |
1365 | if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) | |
1366 | return -EINVAL; | |
1367 | if (access_ok(access_flags, iv->iov_base, iv->iov_len)) | |
1368 | continue; | |
1369 | if (seg == 0) | |
1370 | return -EFAULT; | |
1371 | *nr_segs = seg; | |
1372 | cnt -= iv->iov_len; /* This segment is no good */ | |
1373 | break; | |
1374 | } | |
1375 | *count = cnt; | |
1376 | return 0; | |
1377 | } | |
1378 | EXPORT_SYMBOL(generic_segment_checks); | |
1379 | ||
485bb99b | 1380 | /** |
b2abacf3 | 1381 | * generic_file_aio_read - generic filesystem read routine |
485bb99b RD |
1382 | * @iocb: kernel I/O control block |
1383 | * @iov: io vector request | |
1384 | * @nr_segs: number of segments in the iovec | |
b2abacf3 | 1385 | * @pos: current file position |
485bb99b | 1386 | * |
1da177e4 LT |
1387 | * This is the "read()" routine for all filesystems |
1388 | * that can use the page cache directly. | |
1389 | */ | |
1390 | ssize_t | |
543ade1f BP |
1391 | generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
1392 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
1393 | { |
1394 | struct file *filp = iocb->ki_filp; | |
1395 | ssize_t retval; | |
66f998f6 | 1396 | unsigned long seg = 0; |
1da177e4 | 1397 | size_t count; |
543ade1f | 1398 | loff_t *ppos = &iocb->ki_pos; |
1da177e4 LT |
1399 | |
1400 | count = 0; | |
0ceb3314 DM |
1401 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); |
1402 | if (retval) | |
1403 | return retval; | |
1da177e4 LT |
1404 | |
1405 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
1406 | if (filp->f_flags & O_DIRECT) { | |
543ade1f | 1407 | loff_t size; |
1da177e4 LT |
1408 | struct address_space *mapping; |
1409 | struct inode *inode; | |
1410 | ||
1411 | mapping = filp->f_mapping; | |
1412 | inode = mapping->host; | |
1da177e4 LT |
1413 | if (!count) |
1414 | goto out; /* skip atime */ | |
1415 | size = i_size_read(inode); | |
1416 | if (pos < size) { | |
48b47c56 NP |
1417 | retval = filemap_write_and_wait_range(mapping, pos, |
1418 | pos + iov_length(iov, nr_segs) - 1); | |
a969e903 CH |
1419 | if (!retval) { |
1420 | retval = mapping->a_ops->direct_IO(READ, iocb, | |
1421 | iov, pos, nr_segs); | |
1422 | } | |
66f998f6 | 1423 | if (retval > 0) { |
1da177e4 | 1424 | *ppos = pos + retval; |
66f998f6 JB |
1425 | count -= retval; |
1426 | } | |
1427 | ||
1428 | /* | |
1429 | * Btrfs can have a short DIO read if we encounter | |
1430 | * compressed extents, so if there was an error, or if | |
1431 | * we've already read everything we wanted to, or if | |
1432 | * there was a short read because we hit EOF, go ahead | |
1433 | * and return. Otherwise fallthrough to buffered io for | |
1434 | * the rest of the read. | |
1435 | */ | |
1436 | if (retval < 0 || !count || *ppos >= size) { | |
11fa977e HD |
1437 | file_accessed(filp); |
1438 | goto out; | |
1439 | } | |
0e0bcae3 | 1440 | } |
1da177e4 LT |
1441 | } |
1442 | ||
66f998f6 | 1443 | count = retval; |
11fa977e HD |
1444 | for (seg = 0; seg < nr_segs; seg++) { |
1445 | read_descriptor_t desc; | |
66f998f6 JB |
1446 | loff_t offset = 0; |
1447 | ||
1448 | /* | |
1449 | * If we did a short DIO read we need to skip the section of the | |
1450 | * iov that we've already read data into. | |
1451 | */ | |
1452 | if (count) { | |
1453 | if (count > iov[seg].iov_len) { | |
1454 | count -= iov[seg].iov_len; | |
1455 | continue; | |
1456 | } | |
1457 | offset = count; | |
1458 | count = 0; | |
1459 | } | |
1da177e4 | 1460 | |
11fa977e | 1461 | desc.written = 0; |
66f998f6 JB |
1462 | desc.arg.buf = iov[seg].iov_base + offset; |
1463 | desc.count = iov[seg].iov_len - offset; | |
11fa977e HD |
1464 | if (desc.count == 0) |
1465 | continue; | |
1466 | desc.error = 0; | |
1467 | do_generic_file_read(filp, ppos, &desc, file_read_actor); | |
1468 | retval += desc.written; | |
1469 | if (desc.error) { | |
1470 | retval = retval ?: desc.error; | |
1471 | break; | |
1da177e4 | 1472 | } |
11fa977e HD |
1473 | if (desc.count > 0) |
1474 | break; | |
1da177e4 LT |
1475 | } |
1476 | out: | |
1477 | return retval; | |
1478 | } | |
1da177e4 LT |
1479 | EXPORT_SYMBOL(generic_file_aio_read); |
1480 | ||
1da177e4 | 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; | |
275b12bf WF |
1531 | if (!ra->ra_pages) |
1532 | return; | |
ef00e08e | 1533 | |
2cbea1d3 | 1534 | if (VM_SequentialReadHint(vma)) { |
7ffc59b4 WF |
1535 | page_cache_sync_readahead(mapping, ra, file, offset, |
1536 | ra->ra_pages); | |
ef00e08e LT |
1537 | return; |
1538 | } | |
1539 | ||
207d04ba AK |
1540 | /* Avoid banging the cache line if not needed */ |
1541 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1542 | ra->mmap_miss++; |
1543 | ||
1544 | /* | |
1545 | * Do we miss much more than hit in this file? If so, | |
1546 | * stop bothering with read-ahead. It will only hurt. | |
1547 | */ | |
1548 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1549 | return; | |
1550 | ||
d30a1100 WF |
1551 | /* |
1552 | * mmap read-around | |
1553 | */ | |
ef00e08e | 1554 | ra_pages = max_sane_readahead(ra->ra_pages); |
275b12bf WF |
1555 | ra->start = max_t(long, 0, offset - ra_pages / 2); |
1556 | ra->size = ra_pages; | |
2cbea1d3 | 1557 | ra->async_size = ra_pages / 4; |
275b12bf | 1558 | ra_submit(ra, mapping, file); |
ef00e08e LT |
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 | 1614 | page = find_get_page(mapping, offset); |
45cac65b | 1615 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
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); |
45cac65b | 1621 | } else if (!page) { |
ef00e08e LT |
1622 | /* No page in the page cache at all */ |
1623 | do_sync_mmap_readahead(vma, ra, file, offset); | |
1624 | count_vm_event(PGMAJFAULT); | |
456f998e | 1625 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
1626 | ret = VM_FAULT_MAJOR; |
1627 | retry_find: | |
b522c94d | 1628 | page = find_get_page(mapping, offset); |
1da177e4 LT |
1629 | if (!page) |
1630 | goto no_cached_page; | |
1631 | } | |
1632 | ||
d88c0922 ML |
1633 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
1634 | page_cache_release(page); | |
d065bd81 | 1635 | return ret | VM_FAULT_RETRY; |
d88c0922 | 1636 | } |
b522c94d ML |
1637 | |
1638 | /* Did it get truncated? */ | |
1639 | if (unlikely(page->mapping != mapping)) { | |
1640 | unlock_page(page); | |
1641 | put_page(page); | |
1642 | goto retry_find; | |
1643 | } | |
1644 | VM_BUG_ON(page->index != offset); | |
1645 | ||
1da177e4 | 1646 | /* |
d00806b1 NP |
1647 | * We have a locked page in the page cache, now we need to check |
1648 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1649 | */ |
d00806b1 | 1650 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1651 | goto page_not_uptodate; |
1652 | ||
ef00e08e LT |
1653 | /* |
1654 | * Found the page and have a reference on it. | |
1655 | * We must recheck i_size under page lock. | |
1656 | */ | |
d00806b1 | 1657 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1658 | if (unlikely(offset >= size)) { |
d00806b1 | 1659 | unlock_page(page); |
745ad48e | 1660 | page_cache_release(page); |
5307cc1a | 1661 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1662 | } |
1663 | ||
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 | ||
4fcf1c62 JK |
1716 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
1717 | { | |
1718 | struct page *page = vmf->page; | |
496ad9aa | 1719 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
1720 | int ret = VM_FAULT_LOCKED; |
1721 | ||
14da9200 | 1722 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
1723 | file_update_time(vma->vm_file); |
1724 | lock_page(page); | |
1725 | if (page->mapping != inode->i_mapping) { | |
1726 | unlock_page(page); | |
1727 | ret = VM_FAULT_NOPAGE; | |
1728 | goto out; | |
1729 | } | |
14da9200 JK |
1730 | /* |
1731 | * We mark the page dirty already here so that when freeze is in | |
1732 | * progress, we are guaranteed that writeback during freezing will | |
1733 | * see the dirty page and writeprotect it again. | |
1734 | */ | |
1735 | set_page_dirty(page); | |
1d1d1a76 | 1736 | wait_for_stable_page(page); |
4fcf1c62 | 1737 | out: |
14da9200 | 1738 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
1739 | return ret; |
1740 | } | |
1741 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
1742 | ||
f0f37e2f | 1743 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 1744 | .fault = filemap_fault, |
4fcf1c62 | 1745 | .page_mkwrite = filemap_page_mkwrite, |
0b173bc4 | 1746 | .remap_pages = generic_file_remap_pages, |
1da177e4 LT |
1747 | }; |
1748 | ||
1749 | /* This is used for a general mmap of a disk file */ | |
1750 | ||
1751 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1752 | { | |
1753 | struct address_space *mapping = file->f_mapping; | |
1754 | ||
1755 | if (!mapping->a_ops->readpage) | |
1756 | return -ENOEXEC; | |
1757 | file_accessed(file); | |
1758 | vma->vm_ops = &generic_file_vm_ops; | |
1759 | return 0; | |
1760 | } | |
1da177e4 LT |
1761 | |
1762 | /* | |
1763 | * This is for filesystems which do not implement ->writepage. | |
1764 | */ | |
1765 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
1766 | { | |
1767 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
1768 | return -EINVAL; | |
1769 | return generic_file_mmap(file, vma); | |
1770 | } | |
1771 | #else | |
1772 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1773 | { | |
1774 | return -ENOSYS; | |
1775 | } | |
1776 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
1777 | { | |
1778 | return -ENOSYS; | |
1779 | } | |
1780 | #endif /* CONFIG_MMU */ | |
1781 | ||
1782 | EXPORT_SYMBOL(generic_file_mmap); | |
1783 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
1784 | ||
6fe6900e | 1785 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 1786 | pgoff_t index, |
5e5358e7 | 1787 | int (*filler)(void *, struct page *), |
0531b2aa LT |
1788 | void *data, |
1789 | gfp_t gfp) | |
1da177e4 | 1790 | { |
eb2be189 | 1791 | struct page *page; |
1da177e4 LT |
1792 | int err; |
1793 | repeat: | |
1794 | page = find_get_page(mapping, index); | |
1795 | if (!page) { | |
0531b2aa | 1796 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
1797 | if (!page) |
1798 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 1799 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 NP |
1800 | if (unlikely(err)) { |
1801 | page_cache_release(page); | |
1802 | if (err == -EEXIST) | |
1803 | goto repeat; | |
1da177e4 | 1804 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
1805 | return ERR_PTR(err); |
1806 | } | |
1da177e4 LT |
1807 | err = filler(data, page); |
1808 | if (err < 0) { | |
1809 | page_cache_release(page); | |
1810 | page = ERR_PTR(err); | |
1811 | } | |
1812 | } | |
1da177e4 LT |
1813 | return page; |
1814 | } | |
1815 | ||
0531b2aa | 1816 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 1817 | pgoff_t index, |
5e5358e7 | 1818 | int (*filler)(void *, struct page *), |
0531b2aa LT |
1819 | void *data, |
1820 | gfp_t gfp) | |
1821 | ||
1da177e4 LT |
1822 | { |
1823 | struct page *page; | |
1824 | int err; | |
1825 | ||
1826 | retry: | |
0531b2aa | 1827 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 1828 | if (IS_ERR(page)) |
c855ff37 | 1829 | return page; |
1da177e4 LT |
1830 | if (PageUptodate(page)) |
1831 | goto out; | |
1832 | ||
1833 | lock_page(page); | |
1834 | if (!page->mapping) { | |
1835 | unlock_page(page); | |
1836 | page_cache_release(page); | |
1837 | goto retry; | |
1838 | } | |
1839 | if (PageUptodate(page)) { | |
1840 | unlock_page(page); | |
1841 | goto out; | |
1842 | } | |
1843 | err = filler(data, page); | |
1844 | if (err < 0) { | |
1845 | page_cache_release(page); | |
c855ff37 | 1846 | return ERR_PTR(err); |
1da177e4 | 1847 | } |
c855ff37 | 1848 | out: |
6fe6900e NP |
1849 | mark_page_accessed(page); |
1850 | return page; | |
1851 | } | |
0531b2aa LT |
1852 | |
1853 | /** | |
1854 | * read_cache_page_async - read into page cache, fill it if needed | |
1855 | * @mapping: the page's address_space | |
1856 | * @index: the page index | |
1857 | * @filler: function to perform the read | |
5e5358e7 | 1858 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa LT |
1859 | * |
1860 | * Same as read_cache_page, but don't wait for page to become unlocked | |
1861 | * after submitting it to the filler. | |
1862 | * | |
1863 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1864 | * not set, try to fill the page but don't wait for it to become unlocked. | |
1865 | * | |
1866 | * If the page does not get brought uptodate, return -EIO. | |
1867 | */ | |
1868 | struct page *read_cache_page_async(struct address_space *mapping, | |
1869 | pgoff_t index, | |
5e5358e7 | 1870 | int (*filler)(void *, struct page *), |
0531b2aa LT |
1871 | void *data) |
1872 | { | |
1873 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
1874 | } | |
6fe6900e NP |
1875 | EXPORT_SYMBOL(read_cache_page_async); |
1876 | ||
0531b2aa LT |
1877 | static struct page *wait_on_page_read(struct page *page) |
1878 | { | |
1879 | if (!IS_ERR(page)) { | |
1880 | wait_on_page_locked(page); | |
1881 | if (!PageUptodate(page)) { | |
1882 | page_cache_release(page); | |
1883 | page = ERR_PTR(-EIO); | |
1884 | } | |
1885 | } | |
1886 | return page; | |
1887 | } | |
1888 | ||
1889 | /** | |
1890 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
1891 | * @mapping: the page's address_space | |
1892 | * @index: the page index | |
1893 | * @gfp: the page allocator flags to use if allocating | |
1894 | * | |
1895 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 1896 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
1897 | * |
1898 | * If the page does not get brought uptodate, return -EIO. | |
1899 | */ | |
1900 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
1901 | pgoff_t index, | |
1902 | gfp_t gfp) | |
1903 | { | |
1904 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
1905 | ||
1906 | return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp)); | |
1907 | } | |
1908 | EXPORT_SYMBOL(read_cache_page_gfp); | |
1909 | ||
6fe6900e NP |
1910 | /** |
1911 | * read_cache_page - read into page cache, fill it if needed | |
1912 | * @mapping: the page's address_space | |
1913 | * @index: the page index | |
1914 | * @filler: function to perform the read | |
5e5358e7 | 1915 | * @data: first arg to filler(data, page) function, often left as NULL |
6fe6900e NP |
1916 | * |
1917 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1918 | * not set, try to fill the page then wait for it to become unlocked. | |
1919 | * | |
1920 | * If the page does not get brought uptodate, return -EIO. | |
1921 | */ | |
1922 | struct page *read_cache_page(struct address_space *mapping, | |
57f6b96c | 1923 | pgoff_t index, |
5e5358e7 | 1924 | int (*filler)(void *, struct page *), |
6fe6900e NP |
1925 | void *data) |
1926 | { | |
0531b2aa | 1927 | return wait_on_page_read(read_cache_page_async(mapping, index, filler, data)); |
1da177e4 | 1928 | } |
1da177e4 LT |
1929 | EXPORT_SYMBOL(read_cache_page); |
1930 | ||
2f718ffc | 1931 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
1da177e4 LT |
1932 | const struct iovec *iov, size_t base, size_t bytes) |
1933 | { | |
f1800536 | 1934 | size_t copied = 0, left = 0; |
1da177e4 LT |
1935 | |
1936 | while (bytes) { | |
1937 | char __user *buf = iov->iov_base + base; | |
1938 | int copy = min(bytes, iov->iov_len - base); | |
1939 | ||
1940 | base = 0; | |
f1800536 | 1941 | left = __copy_from_user_inatomic(vaddr, buf, copy); |
1da177e4 LT |
1942 | copied += copy; |
1943 | bytes -= copy; | |
1944 | vaddr += copy; | |
1945 | iov++; | |
1946 | ||
01408c49 | 1947 | if (unlikely(left)) |
1da177e4 | 1948 | break; |
1da177e4 LT |
1949 | } |
1950 | return copied - left; | |
1951 | } | |
1952 | ||
2f718ffc NP |
1953 | /* |
1954 | * Copy as much as we can into the page and return the number of bytes which | |
af901ca1 | 1955 | * were successfully copied. If a fault is encountered then return the number of |
2f718ffc NP |
1956 | * bytes which were copied. |
1957 | */ | |
1958 | size_t iov_iter_copy_from_user_atomic(struct page *page, | |
1959 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
1960 | { | |
1961 | char *kaddr; | |
1962 | size_t copied; | |
1963 | ||
1964 | BUG_ON(!in_atomic()); | |
9b04c5fe | 1965 | kaddr = kmap_atomic(page); |
2f718ffc NP |
1966 | if (likely(i->nr_segs == 1)) { |
1967 | int left; | |
1968 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 1969 | left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); |
2f718ffc NP |
1970 | copied = bytes - left; |
1971 | } else { | |
1972 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
1973 | i->iov, i->iov_offset, bytes); | |
1974 | } | |
9b04c5fe | 1975 | kunmap_atomic(kaddr); |
2f718ffc NP |
1976 | |
1977 | return copied; | |
1978 | } | |
89e10787 | 1979 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); |
2f718ffc NP |
1980 | |
1981 | /* | |
1982 | * This has the same sideeffects and return value as | |
1983 | * iov_iter_copy_from_user_atomic(). | |
1984 | * The difference is that it attempts to resolve faults. | |
1985 | * Page must not be locked. | |
1986 | */ | |
1987 | size_t iov_iter_copy_from_user(struct page *page, | |
1988 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
1989 | { | |
1990 | char *kaddr; | |
1991 | size_t copied; | |
1992 | ||
1993 | kaddr = kmap(page); | |
1994 | if (likely(i->nr_segs == 1)) { | |
1995 | int left; | |
1996 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 1997 | left = __copy_from_user(kaddr + offset, buf, bytes); |
2f718ffc NP |
1998 | copied = bytes - left; |
1999 | } else { | |
2000 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2001 | i->iov, i->iov_offset, bytes); | |
2002 | } | |
2003 | kunmap(page); | |
2004 | return copied; | |
2005 | } | |
89e10787 | 2006 | EXPORT_SYMBOL(iov_iter_copy_from_user); |
2f718ffc | 2007 | |
f7009264 | 2008 | void iov_iter_advance(struct iov_iter *i, size_t bytes) |
2f718ffc | 2009 | { |
f7009264 NP |
2010 | BUG_ON(i->count < bytes); |
2011 | ||
2f718ffc NP |
2012 | if (likely(i->nr_segs == 1)) { |
2013 | i->iov_offset += bytes; | |
f7009264 | 2014 | i->count -= bytes; |
2f718ffc NP |
2015 | } else { |
2016 | const struct iovec *iov = i->iov; | |
2017 | size_t base = i->iov_offset; | |
39be79c1 | 2018 | unsigned long nr_segs = i->nr_segs; |
2f718ffc | 2019 | |
124d3b70 NP |
2020 | /* |
2021 | * The !iov->iov_len check ensures we skip over unlikely | |
f7009264 | 2022 | * zero-length segments (without overruning the iovec). |
124d3b70 | 2023 | */ |
94ad374a | 2024 | while (bytes || unlikely(i->count && !iov->iov_len)) { |
f7009264 | 2025 | int copy; |
2f718ffc | 2026 | |
f7009264 NP |
2027 | copy = min(bytes, iov->iov_len - base); |
2028 | BUG_ON(!i->count || i->count < copy); | |
2029 | i->count -= copy; | |
2f718ffc NP |
2030 | bytes -= copy; |
2031 | base += copy; | |
2032 | if (iov->iov_len == base) { | |
2033 | iov++; | |
39be79c1 | 2034 | nr_segs--; |
2f718ffc NP |
2035 | base = 0; |
2036 | } | |
2037 | } | |
2038 | i->iov = iov; | |
2039 | i->iov_offset = base; | |
39be79c1 | 2040 | i->nr_segs = nr_segs; |
2f718ffc NP |
2041 | } |
2042 | } | |
89e10787 | 2043 | EXPORT_SYMBOL(iov_iter_advance); |
2f718ffc | 2044 | |
afddba49 NP |
2045 | /* |
2046 | * Fault in the first iovec of the given iov_iter, to a maximum length | |
2047 | * of bytes. Returns 0 on success, or non-zero if the memory could not be | |
2048 | * accessed (ie. because it is an invalid address). | |
2049 | * | |
2050 | * writev-intensive code may want this to prefault several iovecs -- that | |
2051 | * would be possible (callers must not rely on the fact that _only_ the | |
2052 | * first iovec will be faulted with the current implementation). | |
2053 | */ | |
2054 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) | |
2f718ffc | 2055 | { |
2f718ffc | 2056 | char __user *buf = i->iov->iov_base + i->iov_offset; |
afddba49 NP |
2057 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); |
2058 | return fault_in_pages_readable(buf, bytes); | |
2f718ffc | 2059 | } |
89e10787 | 2060 | EXPORT_SYMBOL(iov_iter_fault_in_readable); |
2f718ffc NP |
2061 | |
2062 | /* | |
2063 | * Return the count of just the current iov_iter segment. | |
2064 | */ | |
d28574e0 | 2065 | size_t iov_iter_single_seg_count(const struct iov_iter *i) |
2f718ffc NP |
2066 | { |
2067 | const struct iovec *iov = i->iov; | |
2068 | if (i->nr_segs == 1) | |
2069 | return i->count; | |
2070 | else | |
2071 | return min(i->count, iov->iov_len - i->iov_offset); | |
2072 | } | |
89e10787 | 2073 | EXPORT_SYMBOL(iov_iter_single_seg_count); |
2f718ffc | 2074 | |
1da177e4 LT |
2075 | /* |
2076 | * Performs necessary checks before doing a write | |
2077 | * | |
485bb99b | 2078 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2079 | * Returns appropriate error code that caller should return or |
2080 | * zero in case that write should be allowed. | |
2081 | */ | |
2082 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) | |
2083 | { | |
2084 | struct inode *inode = file->f_mapping->host; | |
59e99e5b | 2085 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
1da177e4 LT |
2086 | |
2087 | if (unlikely(*pos < 0)) | |
2088 | return -EINVAL; | |
2089 | ||
1da177e4 LT |
2090 | if (!isblk) { |
2091 | /* FIXME: this is for backwards compatibility with 2.4 */ | |
2092 | if (file->f_flags & O_APPEND) | |
2093 | *pos = i_size_read(inode); | |
2094 | ||
2095 | if (limit != RLIM_INFINITY) { | |
2096 | if (*pos >= limit) { | |
2097 | send_sig(SIGXFSZ, current, 0); | |
2098 | return -EFBIG; | |
2099 | } | |
2100 | if (*count > limit - (typeof(limit))*pos) { | |
2101 | *count = limit - (typeof(limit))*pos; | |
2102 | } | |
2103 | } | |
2104 | } | |
2105 | ||
2106 | /* | |
2107 | * LFS rule | |
2108 | */ | |
2109 | if (unlikely(*pos + *count > MAX_NON_LFS && | |
2110 | !(file->f_flags & O_LARGEFILE))) { | |
2111 | if (*pos >= MAX_NON_LFS) { | |
1da177e4 LT |
2112 | return -EFBIG; |
2113 | } | |
2114 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { | |
2115 | *count = MAX_NON_LFS - (unsigned long)*pos; | |
2116 | } | |
2117 | } | |
2118 | ||
2119 | /* | |
2120 | * Are we about to exceed the fs block limit ? | |
2121 | * | |
2122 | * If we have written data it becomes a short write. If we have | |
2123 | * exceeded without writing data we send a signal and return EFBIG. | |
2124 | * Linus frestrict idea will clean these up nicely.. | |
2125 | */ | |
2126 | if (likely(!isblk)) { | |
2127 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { | |
2128 | if (*count || *pos > inode->i_sb->s_maxbytes) { | |
1da177e4 LT |
2129 | return -EFBIG; |
2130 | } | |
2131 | /* zero-length writes at ->s_maxbytes are OK */ | |
2132 | } | |
2133 | ||
2134 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) | |
2135 | *count = inode->i_sb->s_maxbytes - *pos; | |
2136 | } else { | |
9361401e | 2137 | #ifdef CONFIG_BLOCK |
1da177e4 LT |
2138 | loff_t isize; |
2139 | if (bdev_read_only(I_BDEV(inode))) | |
2140 | return -EPERM; | |
2141 | isize = i_size_read(inode); | |
2142 | if (*pos >= isize) { | |
2143 | if (*count || *pos > isize) | |
2144 | return -ENOSPC; | |
2145 | } | |
2146 | ||
2147 | if (*pos + *count > isize) | |
2148 | *count = isize - *pos; | |
9361401e DH |
2149 | #else |
2150 | return -EPERM; | |
2151 | #endif | |
1da177e4 LT |
2152 | } |
2153 | return 0; | |
2154 | } | |
2155 | EXPORT_SYMBOL(generic_write_checks); | |
2156 | ||
afddba49 NP |
2157 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2158 | loff_t pos, unsigned len, unsigned flags, | |
2159 | struct page **pagep, void **fsdata) | |
2160 | { | |
2161 | const struct address_space_operations *aops = mapping->a_ops; | |
2162 | ||
4e02ed4b | 2163 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2164 | pagep, fsdata); |
afddba49 NP |
2165 | } |
2166 | EXPORT_SYMBOL(pagecache_write_begin); | |
2167 | ||
2168 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2169 | loff_t pos, unsigned len, unsigned copied, | |
2170 | struct page *page, void *fsdata) | |
2171 | { | |
2172 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2173 | |
4e02ed4b NP |
2174 | mark_page_accessed(page); |
2175 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); | |
afddba49 NP |
2176 | } |
2177 | EXPORT_SYMBOL(pagecache_write_end); | |
2178 | ||
1da177e4 LT |
2179 | ssize_t |
2180 | generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, | |
2181 | unsigned long *nr_segs, loff_t pos, loff_t *ppos, | |
2182 | size_t count, size_t ocount) | |
2183 | { | |
2184 | struct file *file = iocb->ki_filp; | |
2185 | struct address_space *mapping = file->f_mapping; | |
2186 | struct inode *inode = mapping->host; | |
2187 | ssize_t written; | |
a969e903 CH |
2188 | size_t write_len; |
2189 | pgoff_t end; | |
1da177e4 LT |
2190 | |
2191 | if (count != ocount) | |
2192 | *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); | |
2193 | ||
a969e903 CH |
2194 | write_len = iov_length(iov, *nr_segs); |
2195 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; | |
a969e903 | 2196 | |
48b47c56 | 2197 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2198 | if (written) |
2199 | goto out; | |
2200 | ||
2201 | /* | |
2202 | * After a write we want buffered reads to be sure to go to disk to get | |
2203 | * the new data. We invalidate clean cached page from the region we're | |
2204 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2205 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2206 | */ |
2207 | if (mapping->nrpages) { | |
2208 | written = invalidate_inode_pages2_range(mapping, | |
2209 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2210 | /* |
2211 | * If a page can not be invalidated, return 0 to fall back | |
2212 | * to buffered write. | |
2213 | */ | |
2214 | if (written) { | |
2215 | if (written == -EBUSY) | |
2216 | return 0; | |
a969e903 | 2217 | goto out; |
6ccfa806 | 2218 | } |
a969e903 CH |
2219 | } |
2220 | ||
2221 | written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs); | |
2222 | ||
2223 | /* | |
2224 | * Finally, try again to invalidate clean pages which might have been | |
2225 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2226 | * if the source of the write was an mmap'ed region of the file | |
2227 | * we're writing. Either one is a pretty crazy thing to do, | |
2228 | * so we don't support it 100%. If this invalidation | |
2229 | * fails, tough, the write still worked... | |
2230 | */ | |
2231 | if (mapping->nrpages) { | |
2232 | invalidate_inode_pages2_range(mapping, | |
2233 | pos >> PAGE_CACHE_SHIFT, end); | |
2234 | } | |
2235 | ||
1da177e4 | 2236 | if (written > 0) { |
0116651c NK |
2237 | pos += written; |
2238 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { | |
2239 | i_size_write(inode, pos); | |
1da177e4 LT |
2240 | mark_inode_dirty(inode); |
2241 | } | |
0116651c | 2242 | *ppos = pos; |
1da177e4 | 2243 | } |
a969e903 | 2244 | out: |
1da177e4 LT |
2245 | return written; |
2246 | } | |
2247 | EXPORT_SYMBOL(generic_file_direct_write); | |
2248 | ||
eb2be189 NP |
2249 | /* |
2250 | * Find or create a page at the given pagecache position. Return the locked | |
2251 | * page. This function is specifically for buffered writes. | |
2252 | */ | |
54566b2c NP |
2253 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2254 | pgoff_t index, unsigned flags) | |
eb2be189 NP |
2255 | { |
2256 | int status; | |
0faa70cb | 2257 | gfp_t gfp_mask; |
eb2be189 | 2258 | struct page *page; |
54566b2c | 2259 | gfp_t gfp_notmask = 0; |
0faa70cb | 2260 | |
1010bb1b FW |
2261 | gfp_mask = mapping_gfp_mask(mapping); |
2262 | if (mapping_cap_account_dirty(mapping)) | |
2263 | gfp_mask |= __GFP_WRITE; | |
54566b2c NP |
2264 | if (flags & AOP_FLAG_NOFS) |
2265 | gfp_notmask = __GFP_FS; | |
eb2be189 NP |
2266 | repeat: |
2267 | page = find_lock_page(mapping, index); | |
c585a267 | 2268 | if (page) |
3d08bcc8 | 2269 | goto found; |
eb2be189 | 2270 | |
0faa70cb | 2271 | page = __page_cache_alloc(gfp_mask & ~gfp_notmask); |
eb2be189 NP |
2272 | if (!page) |
2273 | return NULL; | |
54566b2c NP |
2274 | status = add_to_page_cache_lru(page, mapping, index, |
2275 | GFP_KERNEL & ~gfp_notmask); | |
eb2be189 NP |
2276 | if (unlikely(status)) { |
2277 | page_cache_release(page); | |
2278 | if (status == -EEXIST) | |
2279 | goto repeat; | |
2280 | return NULL; | |
2281 | } | |
3d08bcc8 | 2282 | found: |
1d1d1a76 | 2283 | wait_for_stable_page(page); |
eb2be189 NP |
2284 | return page; |
2285 | } | |
54566b2c | 2286 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2287 | |
afddba49 NP |
2288 | static ssize_t generic_perform_write(struct file *file, |
2289 | struct iov_iter *i, loff_t pos) | |
2290 | { | |
2291 | struct address_space *mapping = file->f_mapping; | |
2292 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2293 | long status = 0; | |
2294 | ssize_t written = 0; | |
674b892e NP |
2295 | unsigned int flags = 0; |
2296 | ||
2297 | /* | |
2298 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2299 | */ | |
2300 | if (segment_eq(get_fs(), KERNEL_DS)) | |
2301 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
afddba49 NP |
2302 | |
2303 | do { | |
2304 | struct page *page; | |
afddba49 NP |
2305 | unsigned long offset; /* Offset into pagecache page */ |
2306 | unsigned long bytes; /* Bytes to write to page */ | |
2307 | size_t copied; /* Bytes copied from user */ | |
2308 | void *fsdata; | |
2309 | ||
2310 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2311 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2312 | iov_iter_count(i)); | |
2313 | ||
2314 | again: | |
afddba49 NP |
2315 | /* |
2316 | * Bring in the user page that we will copy from _first_. | |
2317 | * Otherwise there's a nasty deadlock on copying from the | |
2318 | * same page as we're writing to, without it being marked | |
2319 | * up-to-date. | |
2320 | * | |
2321 | * Not only is this an optimisation, but it is also required | |
2322 | * to check that the address is actually valid, when atomic | |
2323 | * usercopies are used, below. | |
2324 | */ | |
2325 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2326 | status = -EFAULT; | |
2327 | break; | |
2328 | } | |
2329 | ||
674b892e | 2330 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 NP |
2331 | &page, &fsdata); |
2332 | if (unlikely(status)) | |
2333 | break; | |
2334 | ||
931e80e4 | 2335 | if (mapping_writably_mapped(mapping)) |
2336 | flush_dcache_page(page); | |
2337 | ||
afddba49 NP |
2338 | pagefault_disable(); |
2339 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); | |
2340 | pagefault_enable(); | |
2341 | flush_dcache_page(page); | |
2342 | ||
c8236db9 | 2343 | mark_page_accessed(page); |
afddba49 NP |
2344 | status = a_ops->write_end(file, mapping, pos, bytes, copied, |
2345 | page, fsdata); | |
2346 | if (unlikely(status < 0)) | |
2347 | break; | |
2348 | copied = status; | |
2349 | ||
2350 | cond_resched(); | |
2351 | ||
124d3b70 | 2352 | iov_iter_advance(i, copied); |
afddba49 NP |
2353 | if (unlikely(copied == 0)) { |
2354 | /* | |
2355 | * If we were unable to copy any data at all, we must | |
2356 | * fall back to a single segment length write. | |
2357 | * | |
2358 | * If we didn't fallback here, we could livelock | |
2359 | * because not all segments in the iov can be copied at | |
2360 | * once without a pagefault. | |
2361 | */ | |
2362 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2363 | iov_iter_single_seg_count(i)); | |
2364 | goto again; | |
2365 | } | |
afddba49 NP |
2366 | pos += copied; |
2367 | written += copied; | |
2368 | ||
2369 | balance_dirty_pages_ratelimited(mapping); | |
a50527b1 JK |
2370 | if (fatal_signal_pending(current)) { |
2371 | status = -EINTR; | |
2372 | break; | |
2373 | } | |
afddba49 NP |
2374 | } while (iov_iter_count(i)); |
2375 | ||
2376 | return written ? written : status; | |
2377 | } | |
2378 | ||
2379 | ssize_t | |
2380 | generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, | |
2381 | unsigned long nr_segs, loff_t pos, loff_t *ppos, | |
2382 | size_t count, ssize_t written) | |
2383 | { | |
2384 | struct file *file = iocb->ki_filp; | |
afddba49 NP |
2385 | ssize_t status; |
2386 | struct iov_iter i; | |
2387 | ||
2388 | iov_iter_init(&i, iov, nr_segs, count, written); | |
4e02ed4b | 2389 | status = generic_perform_write(file, &i, pos); |
1da177e4 | 2390 | |
1da177e4 | 2391 | if (likely(status >= 0)) { |
afddba49 NP |
2392 | written += status; |
2393 | *ppos = pos + status; | |
1da177e4 LT |
2394 | } |
2395 | ||
1da177e4 LT |
2396 | return written ? written : status; |
2397 | } | |
2398 | EXPORT_SYMBOL(generic_file_buffered_write); | |
2399 | ||
e4dd9de3 JK |
2400 | /** |
2401 | * __generic_file_aio_write - write data to a file | |
2402 | * @iocb: IO state structure (file, offset, etc.) | |
2403 | * @iov: vector with data to write | |
2404 | * @nr_segs: number of segments in the vector | |
2405 | * @ppos: position where to write | |
2406 | * | |
2407 | * This function does all the work needed for actually writing data to a | |
2408 | * file. It does all basic checks, removes SUID from the file, updates | |
2409 | * modification times and calls proper subroutines depending on whether we | |
2410 | * do direct IO or a standard buffered write. | |
2411 | * | |
2412 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2413 | * object which does not need locking at all. | |
2414 | * | |
2415 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2416 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2417 | * avoid syncing under i_mutex. | |
2418 | */ | |
2419 | ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
2420 | unsigned long nr_segs, loff_t *ppos) | |
1da177e4 LT |
2421 | { |
2422 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2423 | struct address_space * mapping = file->f_mapping; |
1da177e4 LT |
2424 | size_t ocount; /* original count */ |
2425 | size_t count; /* after file limit checks */ | |
2426 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2427 | loff_t pos; |
2428 | ssize_t written; | |
2429 | ssize_t err; | |
2430 | ||
2431 | ocount = 0; | |
0ceb3314 DM |
2432 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
2433 | if (err) | |
2434 | return err; | |
1da177e4 LT |
2435 | |
2436 | count = ocount; | |
2437 | pos = *ppos; | |
2438 | ||
1da177e4 LT |
2439 | /* We can write back this queue in page reclaim */ |
2440 | current->backing_dev_info = mapping->backing_dev_info; | |
2441 | written = 0; | |
2442 | ||
2443 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
2444 | if (err) | |
2445 | goto out; | |
2446 | ||
2447 | if (count == 0) | |
2448 | goto out; | |
2449 | ||
2f1936b8 | 2450 | err = file_remove_suid(file); |
1da177e4 LT |
2451 | if (err) |
2452 | goto out; | |
2453 | ||
c3b2da31 JB |
2454 | err = file_update_time(file); |
2455 | if (err) | |
2456 | goto out; | |
1da177e4 LT |
2457 | |
2458 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
2459 | if (unlikely(file->f_flags & O_DIRECT)) { | |
fb5527e6 JM |
2460 | loff_t endbyte; |
2461 | ssize_t written_buffered; | |
2462 | ||
2463 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, | |
2464 | ppos, count, ocount); | |
1da177e4 LT |
2465 | if (written < 0 || written == count) |
2466 | goto out; | |
2467 | /* | |
2468 | * direct-io write to a hole: fall through to buffered I/O | |
2469 | * for completing the rest of the request. | |
2470 | */ | |
2471 | pos += written; | |
2472 | count -= written; | |
fb5527e6 JM |
2473 | written_buffered = generic_file_buffered_write(iocb, iov, |
2474 | nr_segs, pos, ppos, count, | |
2475 | written); | |
2476 | /* | |
2477 | * If generic_file_buffered_write() retuned a synchronous error | |
2478 | * then we want to return the number of bytes which were | |
2479 | * direct-written, or the error code if that was zero. Note | |
2480 | * that this differs from normal direct-io semantics, which | |
2481 | * will return -EFOO even if some bytes were written. | |
2482 | */ | |
2483 | if (written_buffered < 0) { | |
2484 | err = written_buffered; | |
2485 | goto out; | |
2486 | } | |
1da177e4 | 2487 | |
fb5527e6 JM |
2488 | /* |
2489 | * We need to ensure that the page cache pages are written to | |
2490 | * disk and invalidated to preserve the expected O_DIRECT | |
2491 | * semantics. | |
2492 | */ | |
2493 | endbyte = pos + written_buffered - written - 1; | |
c05c4edd | 2494 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); |
fb5527e6 JM |
2495 | if (err == 0) { |
2496 | written = written_buffered; | |
2497 | invalidate_mapping_pages(mapping, | |
2498 | pos >> PAGE_CACHE_SHIFT, | |
2499 | endbyte >> PAGE_CACHE_SHIFT); | |
2500 | } else { | |
2501 | /* | |
2502 | * We don't know how much we wrote, so just return | |
2503 | * the number of bytes which were direct-written | |
2504 | */ | |
2505 | } | |
2506 | } else { | |
2507 | written = generic_file_buffered_write(iocb, iov, nr_segs, | |
2508 | pos, ppos, count, written); | |
2509 | } | |
1da177e4 LT |
2510 | out: |
2511 | current->backing_dev_info = NULL; | |
2512 | return written ? written : err; | |
2513 | } | |
e4dd9de3 JK |
2514 | EXPORT_SYMBOL(__generic_file_aio_write); |
2515 | ||
e4dd9de3 JK |
2516 | /** |
2517 | * generic_file_aio_write - write data to a file | |
2518 | * @iocb: IO state structure | |
2519 | * @iov: vector with data to write | |
2520 | * @nr_segs: number of segments in the vector | |
2521 | * @pos: position in file where to write | |
2522 | * | |
2523 | * This is a wrapper around __generic_file_aio_write() to be used by most | |
2524 | * filesystems. It takes care of syncing the file in case of O_SYNC file | |
2525 | * and acquires i_mutex as needed. | |
2526 | */ | |
027445c3 BP |
2527 | ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
2528 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
2529 | { |
2530 | struct file *file = iocb->ki_filp; | |
148f948b | 2531 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2532 | ssize_t ret; |
1da177e4 LT |
2533 | |
2534 | BUG_ON(iocb->ki_pos != pos); | |
2535 | ||
14da9200 | 2536 | sb_start_write(inode->i_sb); |
1b1dcc1b | 2537 | mutex_lock(&inode->i_mutex); |
e4dd9de3 | 2538 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); |
1b1dcc1b | 2539 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 2540 | |
148f948b | 2541 | if (ret > 0 || ret == -EIOCBQUEUED) { |
1da177e4 LT |
2542 | ssize_t err; |
2543 | ||
148f948b | 2544 | err = generic_write_sync(file, pos, ret); |
c7b50db2 | 2545 | if (err < 0 && ret > 0) |
1da177e4 LT |
2546 | ret = err; |
2547 | } | |
14da9200 | 2548 | sb_end_write(inode->i_sb); |
1da177e4 LT |
2549 | return ret; |
2550 | } | |
2551 | EXPORT_SYMBOL(generic_file_aio_write); | |
2552 | ||
cf9a2ae8 DH |
2553 | /** |
2554 | * try_to_release_page() - release old fs-specific metadata on a page | |
2555 | * | |
2556 | * @page: the page which the kernel is trying to free | |
2557 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2558 | * | |
2559 | * The address_space is to try to release any data against the page | |
2560 | * (presumably at page->private). If the release was successful, return `1'. | |
2561 | * Otherwise return zero. | |
2562 | * | |
266cf658 DH |
2563 | * This may also be called if PG_fscache is set on a page, indicating that the |
2564 | * page is known to the local caching routines. | |
2565 | * | |
cf9a2ae8 | 2566 | * The @gfp_mask argument specifies whether I/O may be performed to release |
3f31fddf | 2567 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
cf9a2ae8 | 2568 | * |
cf9a2ae8 DH |
2569 | */ |
2570 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2571 | { | |
2572 | struct address_space * const mapping = page->mapping; | |
2573 | ||
2574 | BUG_ON(!PageLocked(page)); | |
2575 | if (PageWriteback(page)) | |
2576 | return 0; | |
2577 | ||
2578 | if (mapping && mapping->a_ops->releasepage) | |
2579 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2580 | return try_to_free_buffers(page); | |
2581 | } | |
2582 | ||
2583 | EXPORT_SYMBOL(try_to_release_page); |