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> |
f1820361 | 36 | #include <linux/rmap.h> |
0f8053a5 NP |
37 | #include "internal.h" |
38 | ||
fe0bfaaf RJ |
39 | #define CREATE_TRACE_POINTS |
40 | #include <trace/events/filemap.h> | |
41 | ||
1da177e4 | 42 | /* |
1da177e4 LT |
43 | * FIXME: remove all knowledge of the buffer layer from the core VM |
44 | */ | |
148f948b | 45 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 46 | |
1da177e4 LT |
47 | #include <asm/mman.h> |
48 | ||
49 | /* | |
50 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
51 | * though. | |
52 | * | |
53 | * Shared mappings now work. 15.8.1995 Bruno. | |
54 | * | |
55 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
56 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
57 | * | |
58 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
59 | */ | |
60 | ||
61 | /* | |
62 | * Lock ordering: | |
63 | * | |
3d48ae45 | 64 | * ->i_mmap_mutex (truncate_pagecache) |
1da177e4 | 65 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 HD |
66 | * ->swap_lock (exclusive_swap_page, others) |
67 | * ->mapping->tree_lock | |
1da177e4 | 68 | * |
1b1dcc1b | 69 | * ->i_mutex |
3d48ae45 | 70 | * ->i_mmap_mutex (truncate->unmap_mapping_range) |
1da177e4 LT |
71 | * |
72 | * ->mmap_sem | |
3d48ae45 | 73 | * ->i_mmap_mutex |
b8072f09 | 74 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
75 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
76 | * | |
77 | * ->mmap_sem | |
78 | * ->lock_page (access_process_vm) | |
79 | * | |
ccad2365 | 80 | * ->i_mutex (generic_perform_write) |
82591e6e | 81 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
1da177e4 | 82 | * |
f758eeab | 83 | * bdi->wb.list_lock |
a66979ab | 84 | * sb_lock (fs/fs-writeback.c) |
1da177e4 LT |
85 | * ->mapping->tree_lock (__sync_single_inode) |
86 | * | |
3d48ae45 | 87 | * ->i_mmap_mutex |
1da177e4 LT |
88 | * ->anon_vma.lock (vma_adjust) |
89 | * | |
90 | * ->anon_vma.lock | |
b8072f09 | 91 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 92 | * |
b8072f09 | 93 | * ->page_table_lock or pte_lock |
5d337b91 | 94 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
95 | * ->private_lock (try_to_unmap_one) |
96 | * ->tree_lock (try_to_unmap_one) | |
97 | * ->zone.lru_lock (follow_page->mark_page_accessed) | |
053837fc | 98 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
1da177e4 LT |
99 | * ->private_lock (page_remove_rmap->set_page_dirty) |
100 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
f758eeab | 101 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 102 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
f758eeab | 103 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 104 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
105 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
106 | * | |
9a3c531d AK |
107 | * ->i_mmap_mutex |
108 | * ->tasklist_lock (memory_failure, collect_procs_ao) | |
1da177e4 LT |
109 | */ |
110 | ||
91b0abe3 JW |
111 | static void page_cache_tree_delete(struct address_space *mapping, |
112 | struct page *page, void *shadow) | |
113 | { | |
449dd698 JW |
114 | struct radix_tree_node *node; |
115 | unsigned long index; | |
116 | unsigned int offset; | |
117 | unsigned int tag; | |
118 | void **slot; | |
91b0abe3 | 119 | |
449dd698 JW |
120 | VM_BUG_ON(!PageLocked(page)); |
121 | ||
122 | __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot); | |
123 | ||
124 | if (shadow) { | |
91b0abe3 JW |
125 | mapping->nrshadows++; |
126 | /* | |
127 | * Make sure the nrshadows update is committed before | |
128 | * the nrpages update so that final truncate racing | |
129 | * with reclaim does not see both counters 0 at the | |
130 | * same time and miss a shadow entry. | |
131 | */ | |
132 | smp_wmb(); | |
449dd698 | 133 | } |
91b0abe3 | 134 | mapping->nrpages--; |
449dd698 JW |
135 | |
136 | if (!node) { | |
137 | /* Clear direct pointer tags in root node */ | |
138 | mapping->page_tree.gfp_mask &= __GFP_BITS_MASK; | |
139 | radix_tree_replace_slot(slot, shadow); | |
140 | return; | |
141 | } | |
142 | ||
143 | /* Clear tree tags for the removed page */ | |
144 | index = page->index; | |
145 | offset = index & RADIX_TREE_MAP_MASK; | |
146 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { | |
147 | if (test_bit(offset, node->tags[tag])) | |
148 | radix_tree_tag_clear(&mapping->page_tree, index, tag); | |
149 | } | |
150 | ||
151 | /* Delete page, swap shadow entry */ | |
152 | radix_tree_replace_slot(slot, shadow); | |
153 | workingset_node_pages_dec(node); | |
154 | if (shadow) | |
155 | workingset_node_shadows_inc(node); | |
156 | else | |
157 | if (__radix_tree_delete_node(&mapping->page_tree, node)) | |
158 | return; | |
159 | ||
160 | /* | |
161 | * Track node that only contains shadow entries. | |
162 | * | |
163 | * Avoid acquiring the list_lru lock if already tracked. The | |
164 | * list_empty() test is safe as node->private_list is | |
165 | * protected by mapping->tree_lock. | |
166 | */ | |
167 | if (!workingset_node_pages(node) && | |
168 | list_empty(&node->private_list)) { | |
169 | node->private_data = mapping; | |
170 | list_lru_add(&workingset_shadow_nodes, &node->private_list); | |
171 | } | |
91b0abe3 JW |
172 | } |
173 | ||
1da177e4 | 174 | /* |
e64a782f | 175 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 176 | * sure the page is locked and that nobody else uses it - or that usage |
19fd6231 | 177 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 178 | */ |
91b0abe3 | 179 | void __delete_from_page_cache(struct page *page, void *shadow) |
1da177e4 LT |
180 | { |
181 | struct address_space *mapping = page->mapping; | |
182 | ||
fe0bfaaf | 183 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
184 | /* |
185 | * if we're uptodate, flush out into the cleancache, otherwise | |
186 | * invalidate any existing cleancache entries. We can't leave | |
187 | * stale data around in the cleancache once our page is gone | |
188 | */ | |
189 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
190 | cleancache_put_page(page); | |
191 | else | |
3167760f | 192 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 193 | |
91b0abe3 JW |
194 | page_cache_tree_delete(mapping, page, shadow); |
195 | ||
1da177e4 | 196 | page->mapping = NULL; |
b85e0eff | 197 | /* Leave page->index set: truncation lookup relies upon it */ |
91b0abe3 | 198 | |
347ce434 | 199 | __dec_zone_page_state(page, NR_FILE_PAGES); |
4b02108a KM |
200 | if (PageSwapBacked(page)) |
201 | __dec_zone_page_state(page, NR_SHMEM); | |
45426812 | 202 | BUG_ON(page_mapped(page)); |
3a692790 LT |
203 | |
204 | /* | |
205 | * Some filesystems seem to re-dirty the page even after | |
206 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
207 | * | |
208 | * Fix it up by doing a final dirty accounting check after | |
209 | * having removed the page entirely. | |
210 | */ | |
211 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { | |
212 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
213 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
214 | } | |
1da177e4 LT |
215 | } |
216 | ||
702cfbf9 MK |
217 | /** |
218 | * delete_from_page_cache - delete page from page cache | |
219 | * @page: the page which the kernel is trying to remove from page cache | |
220 | * | |
221 | * This must be called only on pages that have been verified to be in the page | |
222 | * cache and locked. It will never put the page into the free list, the caller | |
223 | * has a reference on the page. | |
224 | */ | |
225 | void delete_from_page_cache(struct page *page) | |
1da177e4 LT |
226 | { |
227 | struct address_space *mapping = page->mapping; | |
6072d13c | 228 | void (*freepage)(struct page *); |
1da177e4 | 229 | |
cd7619d6 | 230 | BUG_ON(!PageLocked(page)); |
1da177e4 | 231 | |
6072d13c | 232 | freepage = mapping->a_ops->freepage; |
19fd6231 | 233 | spin_lock_irq(&mapping->tree_lock); |
91b0abe3 | 234 | __delete_from_page_cache(page, NULL); |
19fd6231 | 235 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 236 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
237 | |
238 | if (freepage) | |
239 | freepage(page); | |
97cecb5a MK |
240 | page_cache_release(page); |
241 | } | |
242 | EXPORT_SYMBOL(delete_from_page_cache); | |
243 | ||
7eaceacc | 244 | static int sleep_on_page(void *word) |
1da177e4 | 245 | { |
1da177e4 LT |
246 | io_schedule(); |
247 | return 0; | |
248 | } | |
249 | ||
7eaceacc | 250 | static int sleep_on_page_killable(void *word) |
2687a356 | 251 | { |
7eaceacc | 252 | sleep_on_page(word); |
2687a356 MW |
253 | return fatal_signal_pending(current) ? -EINTR : 0; |
254 | } | |
255 | ||
865ffef3 DM |
256 | static int filemap_check_errors(struct address_space *mapping) |
257 | { | |
258 | int ret = 0; | |
259 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
260 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
261 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 262 | ret = -ENOSPC; |
7fcbbaf1 JA |
263 | if (test_bit(AS_EIO, &mapping->flags) && |
264 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
265 | ret = -EIO; |
266 | return ret; | |
267 | } | |
268 | ||
1da177e4 | 269 | /** |
485bb99b | 270 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
271 | * @mapping: address space structure to write |
272 | * @start: offset in bytes where the range starts | |
469eb4d0 | 273 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 274 | * @sync_mode: enable synchronous operation |
1da177e4 | 275 | * |
485bb99b RD |
276 | * Start writeback against all of a mapping's dirty pages that lie |
277 | * within the byte offsets <start, end> inclusive. | |
278 | * | |
1da177e4 | 279 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 280 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
281 | * these two operations is that if a dirty page/buffer is encountered, it must |
282 | * be waited upon, and not just skipped over. | |
283 | */ | |
ebcf28e1 AM |
284 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
285 | loff_t end, int sync_mode) | |
1da177e4 LT |
286 | { |
287 | int ret; | |
288 | struct writeback_control wbc = { | |
289 | .sync_mode = sync_mode, | |
05fe478d | 290 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
291 | .range_start = start, |
292 | .range_end = end, | |
1da177e4 LT |
293 | }; |
294 | ||
295 | if (!mapping_cap_writeback_dirty(mapping)) | |
296 | return 0; | |
297 | ||
298 | ret = do_writepages(mapping, &wbc); | |
299 | return ret; | |
300 | } | |
301 | ||
302 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
303 | int sync_mode) | |
304 | { | |
111ebb6e | 305 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
306 | } |
307 | ||
308 | int filemap_fdatawrite(struct address_space *mapping) | |
309 | { | |
310 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
311 | } | |
312 | EXPORT_SYMBOL(filemap_fdatawrite); | |
313 | ||
f4c0a0fd | 314 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 315 | loff_t end) |
1da177e4 LT |
316 | { |
317 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
318 | } | |
f4c0a0fd | 319 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 320 | |
485bb99b RD |
321 | /** |
322 | * filemap_flush - mostly a non-blocking flush | |
323 | * @mapping: target address_space | |
324 | * | |
1da177e4 LT |
325 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
326 | * purposes - I/O may not be started against all dirty pages. | |
327 | */ | |
328 | int filemap_flush(struct address_space *mapping) | |
329 | { | |
330 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
331 | } | |
332 | EXPORT_SYMBOL(filemap_flush); | |
333 | ||
485bb99b | 334 | /** |
94004ed7 CH |
335 | * filemap_fdatawait_range - wait for writeback to complete |
336 | * @mapping: address space structure to wait for | |
337 | * @start_byte: offset in bytes where the range starts | |
338 | * @end_byte: offset in bytes where the range ends (inclusive) | |
485bb99b | 339 | * |
94004ed7 CH |
340 | * Walk the list of under-writeback pages of the given address space |
341 | * in the given range and wait for all of them. | |
1da177e4 | 342 | */ |
94004ed7 CH |
343 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
344 | loff_t end_byte) | |
1da177e4 | 345 | { |
94004ed7 CH |
346 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
347 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
348 | struct pagevec pvec; |
349 | int nr_pages; | |
865ffef3 | 350 | int ret2, ret = 0; |
1da177e4 | 351 | |
94004ed7 | 352 | if (end_byte < start_byte) |
865ffef3 | 353 | goto out; |
1da177e4 LT |
354 | |
355 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
356 | while ((index <= end) && |
357 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
358 | PAGECACHE_TAG_WRITEBACK, | |
359 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
360 | unsigned i; | |
361 | ||
362 | for (i = 0; i < nr_pages; i++) { | |
363 | struct page *page = pvec.pages[i]; | |
364 | ||
365 | /* until radix tree lookup accepts end_index */ | |
366 | if (page->index > end) | |
367 | continue; | |
368 | ||
369 | wait_on_page_writeback(page); | |
212260aa | 370 | if (TestClearPageError(page)) |
1da177e4 LT |
371 | ret = -EIO; |
372 | } | |
373 | pagevec_release(&pvec); | |
374 | cond_resched(); | |
375 | } | |
865ffef3 DM |
376 | out: |
377 | ret2 = filemap_check_errors(mapping); | |
378 | if (!ret) | |
379 | ret = ret2; | |
1da177e4 LT |
380 | |
381 | return ret; | |
382 | } | |
d3bccb6f JK |
383 | EXPORT_SYMBOL(filemap_fdatawait_range); |
384 | ||
1da177e4 | 385 | /** |
485bb99b | 386 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 387 | * @mapping: address space structure to wait for |
485bb99b RD |
388 | * |
389 | * Walk the list of under-writeback pages of the given address space | |
390 | * and wait for all of them. | |
1da177e4 LT |
391 | */ |
392 | int filemap_fdatawait(struct address_space *mapping) | |
393 | { | |
394 | loff_t i_size = i_size_read(mapping->host); | |
395 | ||
396 | if (i_size == 0) | |
397 | return 0; | |
398 | ||
94004ed7 | 399 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
400 | } |
401 | EXPORT_SYMBOL(filemap_fdatawait); | |
402 | ||
403 | int filemap_write_and_wait(struct address_space *mapping) | |
404 | { | |
28fd1298 | 405 | int err = 0; |
1da177e4 LT |
406 | |
407 | if (mapping->nrpages) { | |
28fd1298 OH |
408 | err = filemap_fdatawrite(mapping); |
409 | /* | |
410 | * Even if the above returned error, the pages may be | |
411 | * written partially (e.g. -ENOSPC), so we wait for it. | |
412 | * But the -EIO is special case, it may indicate the worst | |
413 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
414 | */ | |
415 | if (err != -EIO) { | |
416 | int err2 = filemap_fdatawait(mapping); | |
417 | if (!err) | |
418 | err = err2; | |
419 | } | |
865ffef3 DM |
420 | } else { |
421 | err = filemap_check_errors(mapping); | |
1da177e4 | 422 | } |
28fd1298 | 423 | return err; |
1da177e4 | 424 | } |
28fd1298 | 425 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 426 | |
485bb99b RD |
427 | /** |
428 | * filemap_write_and_wait_range - write out & wait on a file range | |
429 | * @mapping: the address_space for the pages | |
430 | * @lstart: offset in bytes where the range starts | |
431 | * @lend: offset in bytes where the range ends (inclusive) | |
432 | * | |
469eb4d0 AM |
433 | * Write out and wait upon file offsets lstart->lend, inclusive. |
434 | * | |
435 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
436 | * that this function can be used to write to the very end-of-file (end = -1). | |
437 | */ | |
1da177e4 LT |
438 | int filemap_write_and_wait_range(struct address_space *mapping, |
439 | loff_t lstart, loff_t lend) | |
440 | { | |
28fd1298 | 441 | int err = 0; |
1da177e4 LT |
442 | |
443 | if (mapping->nrpages) { | |
28fd1298 OH |
444 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
445 | WB_SYNC_ALL); | |
446 | /* See comment of filemap_write_and_wait() */ | |
447 | if (err != -EIO) { | |
94004ed7 CH |
448 | int err2 = filemap_fdatawait_range(mapping, |
449 | lstart, lend); | |
28fd1298 OH |
450 | if (!err) |
451 | err = err2; | |
452 | } | |
865ffef3 DM |
453 | } else { |
454 | err = filemap_check_errors(mapping); | |
1da177e4 | 455 | } |
28fd1298 | 456 | return err; |
1da177e4 | 457 | } |
f6995585 | 458 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 459 | |
ef6a3c63 MS |
460 | /** |
461 | * replace_page_cache_page - replace a pagecache page with a new one | |
462 | * @old: page to be replaced | |
463 | * @new: page to replace with | |
464 | * @gfp_mask: allocation mode | |
465 | * | |
466 | * This function replaces a page in the pagecache with a new one. On | |
467 | * success it acquires the pagecache reference for the new page and | |
468 | * drops it for the old page. Both the old and new pages must be | |
469 | * locked. This function does not add the new page to the LRU, the | |
470 | * caller must do that. | |
471 | * | |
472 | * The remove + add is atomic. The only way this function can fail is | |
473 | * memory allocation failure. | |
474 | */ | |
475 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
476 | { | |
477 | int error; | |
ef6a3c63 | 478 | |
309381fe SL |
479 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
480 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
481 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 482 | |
ef6a3c63 MS |
483 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
484 | if (!error) { | |
485 | struct address_space *mapping = old->mapping; | |
486 | void (*freepage)(struct page *); | |
487 | ||
488 | pgoff_t offset = old->index; | |
489 | freepage = mapping->a_ops->freepage; | |
490 | ||
491 | page_cache_get(new); | |
492 | new->mapping = mapping; | |
493 | new->index = offset; | |
494 | ||
495 | spin_lock_irq(&mapping->tree_lock); | |
91b0abe3 | 496 | __delete_from_page_cache(old, NULL); |
ef6a3c63 MS |
497 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
498 | BUG_ON(error); | |
499 | mapping->nrpages++; | |
500 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
501 | if (PageSwapBacked(new)) | |
502 | __inc_zone_page_state(new, NR_SHMEM); | |
503 | spin_unlock_irq(&mapping->tree_lock); | |
ab936cbc KH |
504 | /* mem_cgroup codes must not be called under tree_lock */ |
505 | mem_cgroup_replace_page_cache(old, new); | |
ef6a3c63 MS |
506 | radix_tree_preload_end(); |
507 | if (freepage) | |
508 | freepage(old); | |
509 | page_cache_release(old); | |
ef6a3c63 MS |
510 | } |
511 | ||
512 | return error; | |
513 | } | |
514 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
515 | ||
0cd6144a | 516 | static int page_cache_tree_insert(struct address_space *mapping, |
a528910e | 517 | struct page *page, void **shadowp) |
0cd6144a | 518 | { |
449dd698 | 519 | struct radix_tree_node *node; |
0cd6144a JW |
520 | void **slot; |
521 | int error; | |
522 | ||
449dd698 JW |
523 | error = __radix_tree_create(&mapping->page_tree, page->index, |
524 | &node, &slot); | |
525 | if (error) | |
526 | return error; | |
527 | if (*slot) { | |
0cd6144a JW |
528 | void *p; |
529 | ||
530 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
531 | if (!radix_tree_exceptional_entry(p)) | |
532 | return -EEXIST; | |
a528910e JW |
533 | if (shadowp) |
534 | *shadowp = p; | |
449dd698 JW |
535 | mapping->nrshadows--; |
536 | if (node) | |
537 | workingset_node_shadows_dec(node); | |
0cd6144a | 538 | } |
449dd698 JW |
539 | radix_tree_replace_slot(slot, page); |
540 | mapping->nrpages++; | |
541 | if (node) { | |
542 | workingset_node_pages_inc(node); | |
543 | /* | |
544 | * Don't track node that contains actual pages. | |
545 | * | |
546 | * Avoid acquiring the list_lru lock if already | |
547 | * untracked. The list_empty() test is safe as | |
548 | * node->private_list is protected by | |
549 | * mapping->tree_lock. | |
550 | */ | |
551 | if (!list_empty(&node->private_list)) | |
552 | list_lru_del(&workingset_shadow_nodes, | |
553 | &node->private_list); | |
554 | } | |
555 | return 0; | |
0cd6144a JW |
556 | } |
557 | ||
a528910e JW |
558 | static int __add_to_page_cache_locked(struct page *page, |
559 | struct address_space *mapping, | |
560 | pgoff_t offset, gfp_t gfp_mask, | |
561 | void **shadowp) | |
1da177e4 | 562 | { |
e286781d NP |
563 | int error; |
564 | ||
309381fe SL |
565 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
566 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d | 567 | |
d715ae08 | 568 | error = mem_cgroup_charge_file(page, current->mm, |
2c26fdd7 | 569 | gfp_mask & GFP_RECLAIM_MASK); |
35c754d7 | 570 | if (error) |
66a0c8ee | 571 | return error; |
1da177e4 | 572 | |
5e4c0d97 | 573 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 574 | if (error) { |
69029cd5 | 575 | mem_cgroup_uncharge_cache_page(page); |
66a0c8ee KS |
576 | return error; |
577 | } | |
578 | ||
579 | page_cache_get(page); | |
580 | page->mapping = mapping; | |
581 | page->index = offset; | |
582 | ||
583 | spin_lock_irq(&mapping->tree_lock); | |
a528910e | 584 | error = page_cache_tree_insert(mapping, page, shadowp); |
66a0c8ee KS |
585 | radix_tree_preload_end(); |
586 | if (unlikely(error)) | |
587 | goto err_insert; | |
66a0c8ee KS |
588 | __inc_zone_page_state(page, NR_FILE_PAGES); |
589 | spin_unlock_irq(&mapping->tree_lock); | |
590 | trace_mm_filemap_add_to_page_cache(page); | |
591 | return 0; | |
592 | err_insert: | |
593 | page->mapping = NULL; | |
594 | /* Leave page->index set: truncation relies upon it */ | |
595 | spin_unlock_irq(&mapping->tree_lock); | |
596 | mem_cgroup_uncharge_cache_page(page); | |
597 | page_cache_release(page); | |
1da177e4 LT |
598 | return error; |
599 | } | |
a528910e JW |
600 | |
601 | /** | |
602 | * add_to_page_cache_locked - add a locked page to the pagecache | |
603 | * @page: page to add | |
604 | * @mapping: the page's address_space | |
605 | * @offset: page index | |
606 | * @gfp_mask: page allocation mode | |
607 | * | |
608 | * This function is used to add a page to the pagecache. It must be locked. | |
609 | * This function does not add the page to the LRU. The caller must do that. | |
610 | */ | |
611 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
612 | pgoff_t offset, gfp_t gfp_mask) | |
613 | { | |
614 | return __add_to_page_cache_locked(page, mapping, offset, | |
615 | gfp_mask, NULL); | |
616 | } | |
e286781d | 617 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
618 | |
619 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 620 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 621 | { |
a528910e | 622 | void *shadow = NULL; |
4f98a2fe RR |
623 | int ret; |
624 | ||
a528910e JW |
625 | __set_page_locked(page); |
626 | ret = __add_to_page_cache_locked(page, mapping, offset, | |
627 | gfp_mask, &shadow); | |
628 | if (unlikely(ret)) | |
629 | __clear_page_locked(page); | |
630 | else { | |
631 | /* | |
632 | * The page might have been evicted from cache only | |
633 | * recently, in which case it should be activated like | |
634 | * any other repeatedly accessed page. | |
635 | */ | |
636 | if (shadow && workingset_refault(shadow)) { | |
637 | SetPageActive(page); | |
638 | workingset_activation(page); | |
639 | } else | |
640 | ClearPageActive(page); | |
641 | lru_cache_add(page); | |
642 | } | |
1da177e4 LT |
643 | return ret; |
644 | } | |
18bc0bbd | 645 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 646 | |
44110fe3 | 647 | #ifdef CONFIG_NUMA |
2ae88149 | 648 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 649 | { |
c0ff7453 MX |
650 | int n; |
651 | struct page *page; | |
652 | ||
44110fe3 | 653 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
654 | unsigned int cpuset_mems_cookie; |
655 | do { | |
d26914d1 | 656 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 MG |
657 | n = cpuset_mem_spread_node(); |
658 | page = alloc_pages_exact_node(n, gfp, 0); | |
d26914d1 | 659 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 660 | |
c0ff7453 | 661 | return page; |
44110fe3 | 662 | } |
2ae88149 | 663 | return alloc_pages(gfp, 0); |
44110fe3 | 664 | } |
2ae88149 | 665 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
666 | #endif |
667 | ||
1da177e4 LT |
668 | /* |
669 | * In order to wait for pages to become available there must be | |
670 | * waitqueues associated with pages. By using a hash table of | |
671 | * waitqueues where the bucket discipline is to maintain all | |
672 | * waiters on the same queue and wake all when any of the pages | |
673 | * become available, and for the woken contexts to check to be | |
674 | * sure the appropriate page became available, this saves space | |
675 | * at a cost of "thundering herd" phenomena during rare hash | |
676 | * collisions. | |
677 | */ | |
678 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
679 | { | |
680 | const struct zone *zone = page_zone(page); | |
681 | ||
682 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
683 | } | |
684 | ||
685 | static inline void wake_up_page(struct page *page, int bit) | |
686 | { | |
687 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); | |
688 | } | |
689 | ||
920c7a5d | 690 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
691 | { |
692 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
693 | ||
694 | if (test_bit(bit_nr, &page->flags)) | |
7eaceacc | 695 | __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
696 | TASK_UNINTERRUPTIBLE); |
697 | } | |
698 | EXPORT_SYMBOL(wait_on_page_bit); | |
699 | ||
f62e00cc KM |
700 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
701 | { | |
702 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
703 | ||
704 | if (!test_bit(bit_nr, &page->flags)) | |
705 | return 0; | |
706 | ||
707 | return __wait_on_bit(page_waitqueue(page), &wait, | |
708 | sleep_on_page_killable, TASK_KILLABLE); | |
709 | } | |
710 | ||
385e1ca5 DH |
711 | /** |
712 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
713 | * @page: Page defining the wait queue of interest |
714 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
715 | * |
716 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
717 | */ | |
718 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
719 | { | |
720 | wait_queue_head_t *q = page_waitqueue(page); | |
721 | unsigned long flags; | |
722 | ||
723 | spin_lock_irqsave(&q->lock, flags); | |
724 | __add_wait_queue(q, waiter); | |
725 | spin_unlock_irqrestore(&q->lock, flags); | |
726 | } | |
727 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
728 | ||
1da177e4 | 729 | /** |
485bb99b | 730 | * unlock_page - unlock a locked page |
1da177e4 LT |
731 | * @page: the page |
732 | * | |
733 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
734 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
735 | * mechananism between PageLocked pages and PageWriteback pages is shared. | |
736 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. | |
737 | * | |
8413ac9d NP |
738 | * The mb is necessary to enforce ordering between the clear_bit and the read |
739 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 740 | */ |
920c7a5d | 741 | void unlock_page(struct page *page) |
1da177e4 | 742 | { |
309381fe | 743 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
8413ac9d | 744 | clear_bit_unlock(PG_locked, &page->flags); |
4e857c58 | 745 | smp_mb__after_atomic(); |
1da177e4 LT |
746 | wake_up_page(page, PG_locked); |
747 | } | |
748 | EXPORT_SYMBOL(unlock_page); | |
749 | ||
485bb99b RD |
750 | /** |
751 | * end_page_writeback - end writeback against a page | |
752 | * @page: the page | |
1da177e4 LT |
753 | */ |
754 | void end_page_writeback(struct page *page) | |
755 | { | |
888cf2db MG |
756 | /* |
757 | * TestClearPageReclaim could be used here but it is an atomic | |
758 | * operation and overkill in this particular case. Failing to | |
759 | * shuffle a page marked for immediate reclaim is too mild to | |
760 | * justify taking an atomic operation penalty at the end of | |
761 | * ever page writeback. | |
762 | */ | |
763 | if (PageReclaim(page)) { | |
764 | ClearPageReclaim(page); | |
ac6aadb2 | 765 | rotate_reclaimable_page(page); |
888cf2db | 766 | } |
ac6aadb2 MS |
767 | |
768 | if (!test_clear_page_writeback(page)) | |
769 | BUG(); | |
770 | ||
4e857c58 | 771 | smp_mb__after_atomic(); |
1da177e4 LT |
772 | wake_up_page(page, PG_writeback); |
773 | } | |
774 | EXPORT_SYMBOL(end_page_writeback); | |
775 | ||
57d99845 MW |
776 | /* |
777 | * After completing I/O on a page, call this routine to update the page | |
778 | * flags appropriately | |
779 | */ | |
780 | void page_endio(struct page *page, int rw, int err) | |
781 | { | |
782 | if (rw == READ) { | |
783 | if (!err) { | |
784 | SetPageUptodate(page); | |
785 | } else { | |
786 | ClearPageUptodate(page); | |
787 | SetPageError(page); | |
788 | } | |
789 | unlock_page(page); | |
790 | } else { /* rw == WRITE */ | |
791 | if (err) { | |
792 | SetPageError(page); | |
793 | if (page->mapping) | |
794 | mapping_set_error(page->mapping, err); | |
795 | } | |
796 | end_page_writeback(page); | |
797 | } | |
798 | } | |
799 | EXPORT_SYMBOL_GPL(page_endio); | |
800 | ||
485bb99b RD |
801 | /** |
802 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
803 | * @page: the page to lock | |
1da177e4 | 804 | */ |
920c7a5d | 805 | void __lock_page(struct page *page) |
1da177e4 LT |
806 | { |
807 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
808 | ||
7eaceacc | 809 | __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
810 | TASK_UNINTERRUPTIBLE); |
811 | } | |
812 | EXPORT_SYMBOL(__lock_page); | |
813 | ||
b5606c2d | 814 | int __lock_page_killable(struct page *page) |
2687a356 MW |
815 | { |
816 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
817 | ||
818 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
7eaceacc | 819 | sleep_on_page_killable, TASK_KILLABLE); |
2687a356 | 820 | } |
18bc0bbd | 821 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 822 | |
d065bd81 ML |
823 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
824 | unsigned int flags) | |
825 | { | |
37b23e05 KM |
826 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
827 | /* | |
828 | * CAUTION! In this case, mmap_sem is not released | |
829 | * even though return 0. | |
830 | */ | |
831 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
832 | return 0; | |
833 | ||
834 | up_read(&mm->mmap_sem); | |
835 | if (flags & FAULT_FLAG_KILLABLE) | |
836 | wait_on_page_locked_killable(page); | |
837 | else | |
318b275f | 838 | wait_on_page_locked(page); |
d065bd81 | 839 | return 0; |
37b23e05 KM |
840 | } else { |
841 | if (flags & FAULT_FLAG_KILLABLE) { | |
842 | int ret; | |
843 | ||
844 | ret = __lock_page_killable(page); | |
845 | if (ret) { | |
846 | up_read(&mm->mmap_sem); | |
847 | return 0; | |
848 | } | |
849 | } else | |
850 | __lock_page(page); | |
851 | return 1; | |
d065bd81 ML |
852 | } |
853 | } | |
854 | ||
e7b563bb JW |
855 | /** |
856 | * page_cache_next_hole - find the next hole (not-present entry) | |
857 | * @mapping: mapping | |
858 | * @index: index | |
859 | * @max_scan: maximum range to search | |
860 | * | |
861 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
862 | * lowest indexed hole. | |
863 | * | |
864 | * Returns: the index of the hole if found, otherwise returns an index | |
865 | * outside of the set specified (in which case 'return - index >= | |
866 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
867 | * be returned. | |
868 | * | |
869 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
870 | * like radix_tree_gang_lookup, this will not atomically search a | |
871 | * snapshot of the tree at a single point in time. For example, if a | |
872 | * hole is created at index 5, then subsequently a hole is created at | |
873 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
874 | * if called under rcu_read_lock. | |
875 | */ | |
876 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
877 | pgoff_t index, unsigned long max_scan) | |
878 | { | |
879 | unsigned long i; | |
880 | ||
881 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
882 | struct page *page; |
883 | ||
884 | page = radix_tree_lookup(&mapping->page_tree, index); | |
885 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
886 | break; |
887 | index++; | |
888 | if (index == 0) | |
889 | break; | |
890 | } | |
891 | ||
892 | return index; | |
893 | } | |
894 | EXPORT_SYMBOL(page_cache_next_hole); | |
895 | ||
896 | /** | |
897 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
898 | * @mapping: mapping | |
899 | * @index: index | |
900 | * @max_scan: maximum range to search | |
901 | * | |
902 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
903 | * the first hole. | |
904 | * | |
905 | * Returns: the index of the hole if found, otherwise returns an index | |
906 | * outside of the set specified (in which case 'index - return >= | |
907 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
908 | * will be returned. | |
909 | * | |
910 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
911 | * like radix_tree_gang_lookup, this will not atomically search a | |
912 | * snapshot of the tree at a single point in time. For example, if a | |
913 | * hole is created at index 10, then subsequently a hole is created at | |
914 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
915 | * called under rcu_read_lock. | |
916 | */ | |
917 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
918 | pgoff_t index, unsigned long max_scan) | |
919 | { | |
920 | unsigned long i; | |
921 | ||
922 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
923 | struct page *page; |
924 | ||
925 | page = radix_tree_lookup(&mapping->page_tree, index); | |
926 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
927 | break; |
928 | index--; | |
929 | if (index == ULONG_MAX) | |
930 | break; | |
931 | } | |
932 | ||
933 | return index; | |
934 | } | |
935 | EXPORT_SYMBOL(page_cache_prev_hole); | |
936 | ||
485bb99b | 937 | /** |
0cd6144a | 938 | * find_get_entry - find and get a page cache entry |
485bb99b | 939 | * @mapping: the address_space to search |
0cd6144a JW |
940 | * @offset: the page cache index |
941 | * | |
942 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
943 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 944 | * |
139b6a6f JW |
945 | * If the slot holds a shadow entry of a previously evicted page, or a |
946 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
947 | * |
948 | * Otherwise, %NULL is returned. | |
1da177e4 | 949 | */ |
0cd6144a | 950 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 951 | { |
a60637c8 | 952 | void **pagep; |
1da177e4 LT |
953 | struct page *page; |
954 | ||
a60637c8 NP |
955 | rcu_read_lock(); |
956 | repeat: | |
957 | page = NULL; | |
958 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
959 | if (pagep) { | |
960 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
961 | if (unlikely(!page)) |
962 | goto out; | |
a2c16d6c | 963 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
964 | if (radix_tree_deref_retry(page)) |
965 | goto repeat; | |
966 | /* | |
139b6a6f JW |
967 | * A shadow entry of a recently evicted page, |
968 | * or a swap entry from shmem/tmpfs. Return | |
969 | * it without attempting to raise page count. | |
8079b1c8 HD |
970 | */ |
971 | goto out; | |
a2c16d6c | 972 | } |
a60637c8 NP |
973 | if (!page_cache_get_speculative(page)) |
974 | goto repeat; | |
975 | ||
976 | /* | |
977 | * Has the page moved? | |
978 | * This is part of the lockless pagecache protocol. See | |
979 | * include/linux/pagemap.h for details. | |
980 | */ | |
981 | if (unlikely(page != *pagep)) { | |
982 | page_cache_release(page); | |
983 | goto repeat; | |
984 | } | |
985 | } | |
27d20fdd | 986 | out: |
a60637c8 NP |
987 | rcu_read_unlock(); |
988 | ||
1da177e4 LT |
989 | return page; |
990 | } | |
0cd6144a | 991 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 992 | |
0cd6144a JW |
993 | /** |
994 | * find_lock_entry - locate, pin and lock a page cache entry | |
995 | * @mapping: the address_space to search | |
996 | * @offset: the page cache index | |
997 | * | |
998 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
999 | * page cache page, it is returned locked and with an increased | |
1000 | * refcount. | |
1001 | * | |
139b6a6f JW |
1002 | * If the slot holds a shadow entry of a previously evicted page, or a |
1003 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1004 | * |
1005 | * Otherwise, %NULL is returned. | |
1006 | * | |
1007 | * find_lock_entry() may sleep. | |
1008 | */ | |
1009 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1010 | { |
1011 | struct page *page; | |
1012 | ||
1da177e4 | 1013 | repeat: |
0cd6144a | 1014 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1015 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1016 | lock_page(page); |
1017 | /* Has the page been truncated? */ | |
1018 | if (unlikely(page->mapping != mapping)) { | |
1019 | unlock_page(page); | |
1020 | page_cache_release(page); | |
1021 | goto repeat; | |
1da177e4 | 1022 | } |
309381fe | 1023 | VM_BUG_ON_PAGE(page->index != offset, page); |
1da177e4 | 1024 | } |
1da177e4 LT |
1025 | return page; |
1026 | } | |
0cd6144a JW |
1027 | EXPORT_SYMBOL(find_lock_entry); |
1028 | ||
1029 | /** | |
2457aec6 | 1030 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1031 | * @mapping: the address_space to search |
1032 | * @offset: the page index | |
2457aec6 MG |
1033 | * @fgp_flags: PCG flags |
1034 | * @gfp_mask: gfp mask to use if a page is to be allocated | |
0cd6144a | 1035 | * |
2457aec6 | 1036 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1037 | * |
2457aec6 | 1038 | * PCG flags modify how the page is returned |
0cd6144a | 1039 | * |
2457aec6 MG |
1040 | * FGP_ACCESSED: the page will be marked accessed |
1041 | * FGP_LOCK: Page is return locked | |
1042 | * FGP_CREAT: If page is not present then a new page is allocated using | |
1043 | * @gfp_mask and added to the page cache and the VM's LRU | |
1044 | * list. The page is returned locked and with an increased | |
1045 | * refcount. Otherwise, %NULL is returned. | |
1da177e4 | 1046 | * |
2457aec6 MG |
1047 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1048 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1049 | * |
2457aec6 | 1050 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1051 | */ |
2457aec6 MG |
1052 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
1053 | int fgp_flags, gfp_t cache_gfp_mask, gfp_t radix_gfp_mask) | |
1da177e4 | 1054 | { |
eb2be189 | 1055 | struct page *page; |
2457aec6 | 1056 | |
1da177e4 | 1057 | repeat: |
2457aec6 MG |
1058 | page = find_get_entry(mapping, offset); |
1059 | if (radix_tree_exceptional_entry(page)) | |
1060 | page = NULL; | |
1061 | if (!page) | |
1062 | goto no_page; | |
1063 | ||
1064 | if (fgp_flags & FGP_LOCK) { | |
1065 | if (fgp_flags & FGP_NOWAIT) { | |
1066 | if (!trylock_page(page)) { | |
1067 | page_cache_release(page); | |
1068 | return NULL; | |
1069 | } | |
1070 | } else { | |
1071 | lock_page(page); | |
1072 | } | |
1073 | ||
1074 | /* Has the page been truncated? */ | |
1075 | if (unlikely(page->mapping != mapping)) { | |
1076 | unlock_page(page); | |
1077 | page_cache_release(page); | |
1078 | goto repeat; | |
1079 | } | |
1080 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1081 | } | |
1082 | ||
1083 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1084 | mark_page_accessed(page); | |
1085 | ||
1086 | no_page: | |
1087 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1088 | int err; | |
1089 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
1090 | cache_gfp_mask |= __GFP_WRITE; | |
1091 | if (fgp_flags & FGP_NOFS) { | |
1092 | cache_gfp_mask &= ~__GFP_FS; | |
1093 | radix_gfp_mask &= ~__GFP_FS; | |
1094 | } | |
1095 | ||
1096 | page = __page_cache_alloc(cache_gfp_mask); | |
eb2be189 NP |
1097 | if (!page) |
1098 | return NULL; | |
2457aec6 MG |
1099 | |
1100 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1101 | fgp_flags |= FGP_LOCK; | |
1102 | ||
1103 | /* Init accessed so avoit atomic mark_page_accessed later */ | |
1104 | if (fgp_flags & FGP_ACCESSED) | |
1105 | init_page_accessed(page); | |
1106 | ||
1107 | err = add_to_page_cache_lru(page, mapping, offset, radix_gfp_mask); | |
eb2be189 NP |
1108 | if (unlikely(err)) { |
1109 | page_cache_release(page); | |
1110 | page = NULL; | |
1111 | if (err == -EEXIST) | |
1112 | goto repeat; | |
1da177e4 | 1113 | } |
1da177e4 | 1114 | } |
2457aec6 | 1115 | |
1da177e4 LT |
1116 | return page; |
1117 | } | |
2457aec6 | 1118 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1119 | |
0cd6144a JW |
1120 | /** |
1121 | * find_get_entries - gang pagecache lookup | |
1122 | * @mapping: The address_space to search | |
1123 | * @start: The starting page cache index | |
1124 | * @nr_entries: The maximum number of entries | |
1125 | * @entries: Where the resulting entries are placed | |
1126 | * @indices: The cache indices corresponding to the entries in @entries | |
1127 | * | |
1128 | * find_get_entries() will search for and return a group of up to | |
1129 | * @nr_entries entries in the mapping. The entries are placed at | |
1130 | * @entries. find_get_entries() takes a reference against any actual | |
1131 | * pages it returns. | |
1132 | * | |
1133 | * The search returns a group of mapping-contiguous page cache entries | |
1134 | * with ascending indexes. There may be holes in the indices due to | |
1135 | * not-present pages. | |
1136 | * | |
139b6a6f JW |
1137 | * Any shadow entries of evicted pages, or swap entries from |
1138 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1139 | * |
1140 | * find_get_entries() returns the number of pages and shadow entries | |
1141 | * which were found. | |
1142 | */ | |
1143 | unsigned find_get_entries(struct address_space *mapping, | |
1144 | pgoff_t start, unsigned int nr_entries, | |
1145 | struct page **entries, pgoff_t *indices) | |
1146 | { | |
1147 | void **slot; | |
1148 | unsigned int ret = 0; | |
1149 | struct radix_tree_iter iter; | |
1150 | ||
1151 | if (!nr_entries) | |
1152 | return 0; | |
1153 | ||
1154 | rcu_read_lock(); | |
1155 | restart: | |
1156 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { | |
1157 | struct page *page; | |
1158 | repeat: | |
1159 | page = radix_tree_deref_slot(slot); | |
1160 | if (unlikely(!page)) | |
1161 | continue; | |
1162 | if (radix_tree_exception(page)) { | |
1163 | if (radix_tree_deref_retry(page)) | |
1164 | goto restart; | |
1165 | /* | |
139b6a6f JW |
1166 | * A shadow entry of a recently evicted page, |
1167 | * or a swap entry from shmem/tmpfs. Return | |
1168 | * it without attempting to raise page count. | |
0cd6144a JW |
1169 | */ |
1170 | goto export; | |
1171 | } | |
1172 | if (!page_cache_get_speculative(page)) | |
1173 | goto repeat; | |
1174 | ||
1175 | /* Has the page moved? */ | |
1176 | if (unlikely(page != *slot)) { | |
1177 | page_cache_release(page); | |
1178 | goto repeat; | |
1179 | } | |
1180 | export: | |
1181 | indices[ret] = iter.index; | |
1182 | entries[ret] = page; | |
1183 | if (++ret == nr_entries) | |
1184 | break; | |
1185 | } | |
1186 | rcu_read_unlock(); | |
1187 | return ret; | |
1188 | } | |
1189 | ||
1da177e4 LT |
1190 | /** |
1191 | * find_get_pages - gang pagecache lookup | |
1192 | * @mapping: The address_space to search | |
1193 | * @start: The starting page index | |
1194 | * @nr_pages: The maximum number of pages | |
1195 | * @pages: Where the resulting pages are placed | |
1196 | * | |
1197 | * find_get_pages() will search for and return a group of up to | |
1198 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
1199 | * find_get_pages() takes a reference against the returned pages. | |
1200 | * | |
1201 | * The search returns a group of mapping-contiguous pages with ascending | |
1202 | * indexes. There may be holes in the indices due to not-present pages. | |
1203 | * | |
1204 | * find_get_pages() returns the number of pages which were found. | |
1205 | */ | |
1206 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
1207 | unsigned int nr_pages, struct page **pages) | |
1208 | { | |
0fc9d104 KK |
1209 | struct radix_tree_iter iter; |
1210 | void **slot; | |
1211 | unsigned ret = 0; | |
1212 | ||
1213 | if (unlikely(!nr_pages)) | |
1214 | return 0; | |
a60637c8 NP |
1215 | |
1216 | rcu_read_lock(); | |
1217 | restart: | |
0fc9d104 | 1218 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
a60637c8 NP |
1219 | struct page *page; |
1220 | repeat: | |
0fc9d104 | 1221 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1222 | if (unlikely(!page)) |
1223 | continue; | |
9d8aa4ea | 1224 | |
a2c16d6c | 1225 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1226 | if (radix_tree_deref_retry(page)) { |
1227 | /* | |
1228 | * Transient condition which can only trigger | |
1229 | * when entry at index 0 moves out of or back | |
1230 | * to root: none yet gotten, safe to restart. | |
1231 | */ | |
0fc9d104 | 1232 | WARN_ON(iter.index); |
8079b1c8 HD |
1233 | goto restart; |
1234 | } | |
a2c16d6c | 1235 | /* |
139b6a6f JW |
1236 | * A shadow entry of a recently evicted page, |
1237 | * or a swap entry from shmem/tmpfs. Skip | |
1238 | * over it. | |
a2c16d6c | 1239 | */ |
8079b1c8 | 1240 | continue; |
27d20fdd | 1241 | } |
a60637c8 NP |
1242 | |
1243 | if (!page_cache_get_speculative(page)) | |
1244 | goto repeat; | |
1245 | ||
1246 | /* Has the page moved? */ | |
0fc9d104 | 1247 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1248 | page_cache_release(page); |
1249 | goto repeat; | |
1250 | } | |
1da177e4 | 1251 | |
a60637c8 | 1252 | pages[ret] = page; |
0fc9d104 KK |
1253 | if (++ret == nr_pages) |
1254 | break; | |
a60637c8 | 1255 | } |
5b280c0c | 1256 | |
a60637c8 | 1257 | rcu_read_unlock(); |
1da177e4 LT |
1258 | return ret; |
1259 | } | |
1260 | ||
ebf43500 JA |
1261 | /** |
1262 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1263 | * @mapping: The address_space to search | |
1264 | * @index: The starting page index | |
1265 | * @nr_pages: The maximum number of pages | |
1266 | * @pages: Where the resulting pages are placed | |
1267 | * | |
1268 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1269 | * that the returned number of pages are guaranteed to be contiguous. | |
1270 | * | |
1271 | * find_get_pages_contig() returns the number of pages which were found. | |
1272 | */ | |
1273 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1274 | unsigned int nr_pages, struct page **pages) | |
1275 | { | |
0fc9d104 KK |
1276 | struct radix_tree_iter iter; |
1277 | void **slot; | |
1278 | unsigned int ret = 0; | |
1279 | ||
1280 | if (unlikely(!nr_pages)) | |
1281 | return 0; | |
a60637c8 NP |
1282 | |
1283 | rcu_read_lock(); | |
1284 | restart: | |
0fc9d104 | 1285 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
a60637c8 NP |
1286 | struct page *page; |
1287 | repeat: | |
0fc9d104 KK |
1288 | page = radix_tree_deref_slot(slot); |
1289 | /* The hole, there no reason to continue */ | |
a60637c8 | 1290 | if (unlikely(!page)) |
0fc9d104 | 1291 | break; |
9d8aa4ea | 1292 | |
a2c16d6c | 1293 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1294 | if (radix_tree_deref_retry(page)) { |
1295 | /* | |
1296 | * Transient condition which can only trigger | |
1297 | * when entry at index 0 moves out of or back | |
1298 | * to root: none yet gotten, safe to restart. | |
1299 | */ | |
1300 | goto restart; | |
1301 | } | |
a2c16d6c | 1302 | /* |
139b6a6f JW |
1303 | * A shadow entry of a recently evicted page, |
1304 | * or a swap entry from shmem/tmpfs. Stop | |
1305 | * looking for contiguous pages. | |
a2c16d6c | 1306 | */ |
8079b1c8 | 1307 | break; |
a2c16d6c | 1308 | } |
ebf43500 | 1309 | |
a60637c8 NP |
1310 | if (!page_cache_get_speculative(page)) |
1311 | goto repeat; | |
1312 | ||
1313 | /* Has the page moved? */ | |
0fc9d104 | 1314 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1315 | page_cache_release(page); |
1316 | goto repeat; | |
1317 | } | |
1318 | ||
9cbb4cb2 NP |
1319 | /* |
1320 | * must check mapping and index after taking the ref. | |
1321 | * otherwise we can get both false positives and false | |
1322 | * negatives, which is just confusing to the caller. | |
1323 | */ | |
0fc9d104 | 1324 | if (page->mapping == NULL || page->index != iter.index) { |
9cbb4cb2 NP |
1325 | page_cache_release(page); |
1326 | break; | |
1327 | } | |
1328 | ||
a60637c8 | 1329 | pages[ret] = page; |
0fc9d104 KK |
1330 | if (++ret == nr_pages) |
1331 | break; | |
ebf43500 | 1332 | } |
a60637c8 NP |
1333 | rcu_read_unlock(); |
1334 | return ret; | |
ebf43500 | 1335 | } |
ef71c15c | 1336 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1337 | |
485bb99b RD |
1338 | /** |
1339 | * find_get_pages_tag - find and return pages that match @tag | |
1340 | * @mapping: the address_space to search | |
1341 | * @index: the starting page index | |
1342 | * @tag: the tag index | |
1343 | * @nr_pages: the maximum number of pages | |
1344 | * @pages: where the resulting pages are placed | |
1345 | * | |
1da177e4 | 1346 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1347 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
1348 | */ |
1349 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
1350 | int tag, unsigned int nr_pages, struct page **pages) | |
1351 | { | |
0fc9d104 KK |
1352 | struct radix_tree_iter iter; |
1353 | void **slot; | |
1354 | unsigned ret = 0; | |
1355 | ||
1356 | if (unlikely(!nr_pages)) | |
1357 | return 0; | |
a60637c8 NP |
1358 | |
1359 | rcu_read_lock(); | |
1360 | restart: | |
0fc9d104 KK |
1361 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1362 | &iter, *index, tag) { | |
a60637c8 NP |
1363 | struct page *page; |
1364 | repeat: | |
0fc9d104 | 1365 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1366 | if (unlikely(!page)) |
1367 | continue; | |
9d8aa4ea | 1368 | |
a2c16d6c | 1369 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1370 | if (radix_tree_deref_retry(page)) { |
1371 | /* | |
1372 | * Transient condition which can only trigger | |
1373 | * when entry at index 0 moves out of or back | |
1374 | * to root: none yet gotten, safe to restart. | |
1375 | */ | |
1376 | goto restart; | |
1377 | } | |
a2c16d6c | 1378 | /* |
139b6a6f JW |
1379 | * A shadow entry of a recently evicted page. |
1380 | * | |
1381 | * Those entries should never be tagged, but | |
1382 | * this tree walk is lockless and the tags are | |
1383 | * looked up in bulk, one radix tree node at a | |
1384 | * time, so there is a sizable window for page | |
1385 | * reclaim to evict a page we saw tagged. | |
1386 | * | |
1387 | * Skip over it. | |
a2c16d6c | 1388 | */ |
139b6a6f | 1389 | continue; |
a2c16d6c | 1390 | } |
a60637c8 NP |
1391 | |
1392 | if (!page_cache_get_speculative(page)) | |
1393 | goto repeat; | |
1394 | ||
1395 | /* Has the page moved? */ | |
0fc9d104 | 1396 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1397 | page_cache_release(page); |
1398 | goto repeat; | |
1399 | } | |
1400 | ||
1401 | pages[ret] = page; | |
0fc9d104 KK |
1402 | if (++ret == nr_pages) |
1403 | break; | |
a60637c8 | 1404 | } |
5b280c0c | 1405 | |
a60637c8 | 1406 | rcu_read_unlock(); |
1da177e4 | 1407 | |
1da177e4 LT |
1408 | if (ret) |
1409 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1410 | |
1da177e4 LT |
1411 | return ret; |
1412 | } | |
ef71c15c | 1413 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1414 | |
76d42bd9 WF |
1415 | /* |
1416 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1417 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1418 | * | |
1419 | * ---R__________________________________________B__________ | |
1420 | * ^ reading here ^ bad block(assume 4k) | |
1421 | * | |
1422 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1423 | * => failing the whole request => read(R) => read(R+1) => | |
1424 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1425 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1426 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1427 | * | |
1428 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1429 | */ | |
1430 | static void shrink_readahead_size_eio(struct file *filp, | |
1431 | struct file_ra_state *ra) | |
1432 | { | |
76d42bd9 | 1433 | ra->ra_pages /= 4; |
76d42bd9 WF |
1434 | } |
1435 | ||
485bb99b | 1436 | /** |
36e78914 | 1437 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1438 | * @filp: the file to read |
1439 | * @ppos: current file position | |
6e58e79d AV |
1440 | * @iter: data destination |
1441 | * @written: already copied | |
485bb99b | 1442 | * |
1da177e4 | 1443 | * This is a generic file read routine, and uses the |
485bb99b | 1444 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1445 | * |
1446 | * This is really ugly. But the goto's actually try to clarify some | |
1447 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1448 | */ |
6e58e79d AV |
1449 | static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos, |
1450 | struct iov_iter *iter, ssize_t written) | |
1da177e4 | 1451 | { |
36e78914 | 1452 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1453 | struct inode *inode = mapping->host; |
36e78914 | 1454 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1455 | pgoff_t index; |
1456 | pgoff_t last_index; | |
1457 | pgoff_t prev_index; | |
1458 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1459 | unsigned int prev_offset; |
6e58e79d | 1460 | int error = 0; |
1da177e4 | 1461 | |
1da177e4 | 1462 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1463 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1464 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
6e58e79d | 1465 | last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1da177e4 LT |
1466 | offset = *ppos & ~PAGE_CACHE_MASK; |
1467 | ||
1da177e4 LT |
1468 | for (;;) { |
1469 | struct page *page; | |
57f6b96c | 1470 | pgoff_t end_index; |
a32ea1e1 | 1471 | loff_t isize; |
1da177e4 LT |
1472 | unsigned long nr, ret; |
1473 | ||
1da177e4 | 1474 | cond_resched(); |
1da177e4 LT |
1475 | find_page: |
1476 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1477 | if (!page) { |
cf914a7d | 1478 | page_cache_sync_readahead(mapping, |
7ff81078 | 1479 | ra, filp, |
3ea89ee8 FW |
1480 | index, last_index - index); |
1481 | page = find_get_page(mapping, index); | |
1482 | if (unlikely(page == NULL)) | |
1483 | goto no_cached_page; | |
1484 | } | |
1485 | if (PageReadahead(page)) { | |
cf914a7d | 1486 | page_cache_async_readahead(mapping, |
7ff81078 | 1487 | ra, filp, page, |
3ea89ee8 | 1488 | index, last_index - index); |
1da177e4 | 1489 | } |
8ab22b9a HH |
1490 | if (!PageUptodate(page)) { |
1491 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1492 | !mapping->a_ops->is_partially_uptodate) | |
1493 | goto page_not_up_to_date; | |
529ae9aa | 1494 | if (!trylock_page(page)) |
8ab22b9a | 1495 | goto page_not_up_to_date; |
8d056cb9 DH |
1496 | /* Did it get truncated before we got the lock? */ |
1497 | if (!page->mapping) | |
1498 | goto page_not_up_to_date_locked; | |
8ab22b9a | 1499 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 1500 | offset, iter->count)) |
8ab22b9a HH |
1501 | goto page_not_up_to_date_locked; |
1502 | unlock_page(page); | |
1503 | } | |
1da177e4 | 1504 | page_ok: |
a32ea1e1 N |
1505 | /* |
1506 | * i_size must be checked after we know the page is Uptodate. | |
1507 | * | |
1508 | * Checking i_size after the check allows us to calculate | |
1509 | * the correct value for "nr", which means the zero-filled | |
1510 | * part of the page is not copied back to userspace (unless | |
1511 | * another truncate extends the file - this is desired though). | |
1512 | */ | |
1513 | ||
1514 | isize = i_size_read(inode); | |
1515 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1516 | if (unlikely(!isize || index > end_index)) { | |
1517 | page_cache_release(page); | |
1518 | goto out; | |
1519 | } | |
1520 | ||
1521 | /* nr is the maximum number of bytes to copy from this page */ | |
1522 | nr = PAGE_CACHE_SIZE; | |
1523 | if (index == end_index) { | |
1524 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1525 | if (nr <= offset) { | |
1526 | page_cache_release(page); | |
1527 | goto out; | |
1528 | } | |
1529 | } | |
1530 | nr = nr - offset; | |
1da177e4 LT |
1531 | |
1532 | /* If users can be writing to this page using arbitrary | |
1533 | * virtual addresses, take care about potential aliasing | |
1534 | * before reading the page on the kernel side. | |
1535 | */ | |
1536 | if (mapping_writably_mapped(mapping)) | |
1537 | flush_dcache_page(page); | |
1538 | ||
1539 | /* | |
ec0f1637 JK |
1540 | * When a sequential read accesses a page several times, |
1541 | * only mark it as accessed the first time. | |
1da177e4 | 1542 | */ |
ec0f1637 | 1543 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1544 | mark_page_accessed(page); |
1545 | prev_index = index; | |
1546 | ||
1547 | /* | |
1548 | * Ok, we have the page, and it's up-to-date, so | |
1549 | * now we can copy it to user space... | |
1da177e4 | 1550 | */ |
6e58e79d AV |
1551 | |
1552 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 LT |
1553 | offset += ret; |
1554 | index += offset >> PAGE_CACHE_SHIFT; | |
1555 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1556 | prev_offset = offset; |
1da177e4 LT |
1557 | |
1558 | page_cache_release(page); | |
6e58e79d AV |
1559 | written += ret; |
1560 | if (!iov_iter_count(iter)) | |
1561 | goto out; | |
1562 | if (ret < nr) { | |
1563 | error = -EFAULT; | |
1564 | goto out; | |
1565 | } | |
1566 | continue; | |
1da177e4 LT |
1567 | |
1568 | page_not_up_to_date: | |
1569 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1570 | error = lock_page_killable(page); |
1571 | if (unlikely(error)) | |
1572 | goto readpage_error; | |
1da177e4 | 1573 | |
8ab22b9a | 1574 | page_not_up_to_date_locked: |
da6052f7 | 1575 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1576 | if (!page->mapping) { |
1577 | unlock_page(page); | |
1578 | page_cache_release(page); | |
1579 | continue; | |
1580 | } | |
1581 | ||
1582 | /* Did somebody else fill it already? */ | |
1583 | if (PageUptodate(page)) { | |
1584 | unlock_page(page); | |
1585 | goto page_ok; | |
1586 | } | |
1587 | ||
1588 | readpage: | |
91803b49 JM |
1589 | /* |
1590 | * A previous I/O error may have been due to temporary | |
1591 | * failures, eg. multipath errors. | |
1592 | * PG_error will be set again if readpage fails. | |
1593 | */ | |
1594 | ClearPageError(page); | |
1da177e4 LT |
1595 | /* Start the actual read. The read will unlock the page. */ |
1596 | error = mapping->a_ops->readpage(filp, page); | |
1597 | ||
994fc28c ZB |
1598 | if (unlikely(error)) { |
1599 | if (error == AOP_TRUNCATED_PAGE) { | |
1600 | page_cache_release(page); | |
6e58e79d | 1601 | error = 0; |
994fc28c ZB |
1602 | goto find_page; |
1603 | } | |
1da177e4 | 1604 | goto readpage_error; |
994fc28c | 1605 | } |
1da177e4 LT |
1606 | |
1607 | if (!PageUptodate(page)) { | |
85462323 ON |
1608 | error = lock_page_killable(page); |
1609 | if (unlikely(error)) | |
1610 | goto readpage_error; | |
1da177e4 LT |
1611 | if (!PageUptodate(page)) { |
1612 | if (page->mapping == NULL) { | |
1613 | /* | |
2ecdc82e | 1614 | * invalidate_mapping_pages got it |
1da177e4 LT |
1615 | */ |
1616 | unlock_page(page); | |
1617 | page_cache_release(page); | |
1618 | goto find_page; | |
1619 | } | |
1620 | unlock_page(page); | |
7ff81078 | 1621 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1622 | error = -EIO; |
1623 | goto readpage_error; | |
1da177e4 LT |
1624 | } |
1625 | unlock_page(page); | |
1626 | } | |
1627 | ||
1da177e4 LT |
1628 | goto page_ok; |
1629 | ||
1630 | readpage_error: | |
1631 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1da177e4 LT |
1632 | page_cache_release(page); |
1633 | goto out; | |
1634 | ||
1635 | no_cached_page: | |
1636 | /* | |
1637 | * Ok, it wasn't cached, so we need to create a new | |
1638 | * page.. | |
1639 | */ | |
eb2be189 NP |
1640 | page = page_cache_alloc_cold(mapping); |
1641 | if (!page) { | |
6e58e79d | 1642 | error = -ENOMEM; |
eb2be189 | 1643 | goto out; |
1da177e4 | 1644 | } |
eb2be189 | 1645 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1646 | index, GFP_KERNEL); |
1647 | if (error) { | |
eb2be189 | 1648 | page_cache_release(page); |
6e58e79d AV |
1649 | if (error == -EEXIST) { |
1650 | error = 0; | |
1da177e4 | 1651 | goto find_page; |
6e58e79d | 1652 | } |
1da177e4 LT |
1653 | goto out; |
1654 | } | |
1da177e4 LT |
1655 | goto readpage; |
1656 | } | |
1657 | ||
1658 | out: | |
7ff81078 FW |
1659 | ra->prev_pos = prev_index; |
1660 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1661 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1662 | |
f4e6b498 | 1663 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1664 | file_accessed(filp); |
6e58e79d | 1665 | return written ? written : error; |
1da177e4 LT |
1666 | } |
1667 | ||
485bb99b | 1668 | /** |
6abd2322 | 1669 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 1670 | * @iocb: kernel I/O control block |
6abd2322 | 1671 | * @iter: destination for the data read |
485bb99b | 1672 | * |
6abd2322 | 1673 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
1674 | * that can use the page cache directly. |
1675 | */ | |
1676 | ssize_t | |
ed978a81 | 1677 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 1678 | { |
ed978a81 | 1679 | struct file *file = iocb->ki_filp; |
cb66a7a1 | 1680 | ssize_t retval = 0; |
543ade1f | 1681 | loff_t *ppos = &iocb->ki_pos; |
ed978a81 | 1682 | loff_t pos = *ppos; |
1da177e4 LT |
1683 | |
1684 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
ed978a81 AV |
1685 | if (file->f_flags & O_DIRECT) { |
1686 | struct address_space *mapping = file->f_mapping; | |
1687 | struct inode *inode = mapping->host; | |
1688 | size_t count = iov_iter_count(iter); | |
543ade1f | 1689 | loff_t size; |
1da177e4 | 1690 | |
1da177e4 LT |
1691 | if (!count) |
1692 | goto out; /* skip atime */ | |
1693 | size = i_size_read(inode); | |
9fe55eea | 1694 | retval = filemap_write_and_wait_range(mapping, pos, |
a6cbcd4a | 1695 | pos + count - 1); |
9fe55eea | 1696 | if (!retval) { |
ed978a81 | 1697 | struct iov_iter data = *iter; |
26978b8b | 1698 | retval = mapping->a_ops->direct_IO(READ, iocb, &data, pos); |
9fe55eea | 1699 | } |
d8d3d94b | 1700 | |
9fe55eea SW |
1701 | if (retval > 0) { |
1702 | *ppos = pos + retval; | |
ed978a81 | 1703 | iov_iter_advance(iter, retval); |
9fe55eea | 1704 | } |
66f998f6 | 1705 | |
9fe55eea SW |
1706 | /* |
1707 | * Btrfs can have a short DIO read if we encounter | |
1708 | * compressed extents, so if there was an error, or if | |
1709 | * we've already read everything we wanted to, or if | |
1710 | * there was a short read because we hit EOF, go ahead | |
1711 | * and return. Otherwise fallthrough to buffered io for | |
1712 | * the rest of the read. | |
1713 | */ | |
ed978a81 AV |
1714 | if (retval < 0 || !iov_iter_count(iter) || *ppos >= size) { |
1715 | file_accessed(file); | |
9fe55eea | 1716 | goto out; |
0e0bcae3 | 1717 | } |
1da177e4 LT |
1718 | } |
1719 | ||
ed978a81 | 1720 | retval = do_generic_file_read(file, ppos, iter, retval); |
1da177e4 LT |
1721 | out: |
1722 | return retval; | |
1723 | } | |
ed978a81 | 1724 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 1725 | |
1da177e4 | 1726 | #ifdef CONFIG_MMU |
485bb99b RD |
1727 | /** |
1728 | * page_cache_read - adds requested page to the page cache if not already there | |
1729 | * @file: file to read | |
1730 | * @offset: page index | |
1731 | * | |
1da177e4 LT |
1732 | * This adds the requested page to the page cache if it isn't already there, |
1733 | * and schedules an I/O to read in its contents from disk. | |
1734 | */ | |
920c7a5d | 1735 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1736 | { |
1737 | struct address_space *mapping = file->f_mapping; | |
1738 | struct page *page; | |
994fc28c | 1739 | int ret; |
1da177e4 | 1740 | |
994fc28c ZB |
1741 | do { |
1742 | page = page_cache_alloc_cold(mapping); | |
1743 | if (!page) | |
1744 | return -ENOMEM; | |
1745 | ||
1746 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); | |
1747 | if (ret == 0) | |
1748 | ret = mapping->a_ops->readpage(file, page); | |
1749 | else if (ret == -EEXIST) | |
1750 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1751 | |
1da177e4 | 1752 | page_cache_release(page); |
1da177e4 | 1753 | |
994fc28c ZB |
1754 | } while (ret == AOP_TRUNCATED_PAGE); |
1755 | ||
1756 | return ret; | |
1da177e4 LT |
1757 | } |
1758 | ||
1759 | #define MMAP_LOTSAMISS (100) | |
1760 | ||
ef00e08e LT |
1761 | /* |
1762 | * Synchronous readahead happens when we don't even find | |
1763 | * a page in the page cache at all. | |
1764 | */ | |
1765 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1766 | struct file_ra_state *ra, | |
1767 | struct file *file, | |
1768 | pgoff_t offset) | |
1769 | { | |
1770 | unsigned long ra_pages; | |
1771 | struct address_space *mapping = file->f_mapping; | |
1772 | ||
1773 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1774 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 1775 | return; |
275b12bf WF |
1776 | if (!ra->ra_pages) |
1777 | return; | |
ef00e08e | 1778 | |
64363aad | 1779 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
1780 | page_cache_sync_readahead(mapping, ra, file, offset, |
1781 | ra->ra_pages); | |
ef00e08e LT |
1782 | return; |
1783 | } | |
1784 | ||
207d04ba AK |
1785 | /* Avoid banging the cache line if not needed */ |
1786 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1787 | ra->mmap_miss++; |
1788 | ||
1789 | /* | |
1790 | * Do we miss much more than hit in this file? If so, | |
1791 | * stop bothering with read-ahead. It will only hurt. | |
1792 | */ | |
1793 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1794 | return; | |
1795 | ||
d30a1100 WF |
1796 | /* |
1797 | * mmap read-around | |
1798 | */ | |
ef00e08e | 1799 | ra_pages = max_sane_readahead(ra->ra_pages); |
275b12bf WF |
1800 | ra->start = max_t(long, 0, offset - ra_pages / 2); |
1801 | ra->size = ra_pages; | |
2cbea1d3 | 1802 | ra->async_size = ra_pages / 4; |
275b12bf | 1803 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1804 | } |
1805 | ||
1806 | /* | |
1807 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1808 | * so we want to possibly extend the readahead further.. | |
1809 | */ | |
1810 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1811 | struct file_ra_state *ra, | |
1812 | struct file *file, | |
1813 | struct page *page, | |
1814 | pgoff_t offset) | |
1815 | { | |
1816 | struct address_space *mapping = file->f_mapping; | |
1817 | ||
1818 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1819 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
1820 | return; |
1821 | if (ra->mmap_miss > 0) | |
1822 | ra->mmap_miss--; | |
1823 | if (PageReadahead(page)) | |
2fad6f5d WF |
1824 | page_cache_async_readahead(mapping, ra, file, |
1825 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1826 | } |
1827 | ||
485bb99b | 1828 | /** |
54cb8821 | 1829 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1830 | * @vma: vma in which the fault was taken |
1831 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1832 | * |
54cb8821 | 1833 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1834 | * mapped memory region to read in file data during a page fault. |
1835 | * | |
1836 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1837 | * it in the page cache, and handles the special cases reasonably without | |
1838 | * having a lot of duplicated code. | |
1839 | */ | |
d0217ac0 | 1840 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1841 | { |
1842 | int error; | |
54cb8821 | 1843 | struct file *file = vma->vm_file; |
1da177e4 LT |
1844 | struct address_space *mapping = file->f_mapping; |
1845 | struct file_ra_state *ra = &file->f_ra; | |
1846 | struct inode *inode = mapping->host; | |
ef00e08e | 1847 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 1848 | struct page *page; |
99e3e53f | 1849 | loff_t size; |
83c54070 | 1850 | int ret = 0; |
1da177e4 | 1851 | |
99e3e53f KS |
1852 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
1853 | if (offset >= size >> PAGE_CACHE_SHIFT) | |
5307cc1a | 1854 | return VM_FAULT_SIGBUS; |
1da177e4 | 1855 | |
1da177e4 | 1856 | /* |
49426420 | 1857 | * Do we have something in the page cache already? |
1da177e4 | 1858 | */ |
ef00e08e | 1859 | page = find_get_page(mapping, offset); |
45cac65b | 1860 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 1861 | /* |
ef00e08e LT |
1862 | * We found the page, so try async readahead before |
1863 | * waiting for the lock. | |
1da177e4 | 1864 | */ |
ef00e08e | 1865 | do_async_mmap_readahead(vma, ra, file, page, offset); |
45cac65b | 1866 | } else if (!page) { |
ef00e08e LT |
1867 | /* No page in the page cache at all */ |
1868 | do_sync_mmap_readahead(vma, ra, file, offset); | |
1869 | count_vm_event(PGMAJFAULT); | |
456f998e | 1870 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
1871 | ret = VM_FAULT_MAJOR; |
1872 | retry_find: | |
b522c94d | 1873 | page = find_get_page(mapping, offset); |
1da177e4 LT |
1874 | if (!page) |
1875 | goto no_cached_page; | |
1876 | } | |
1877 | ||
d88c0922 ML |
1878 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
1879 | page_cache_release(page); | |
d065bd81 | 1880 | return ret | VM_FAULT_RETRY; |
d88c0922 | 1881 | } |
b522c94d ML |
1882 | |
1883 | /* Did it get truncated? */ | |
1884 | if (unlikely(page->mapping != mapping)) { | |
1885 | unlock_page(page); | |
1886 | put_page(page); | |
1887 | goto retry_find; | |
1888 | } | |
309381fe | 1889 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 1890 | |
1da177e4 | 1891 | /* |
d00806b1 NP |
1892 | * We have a locked page in the page cache, now we need to check |
1893 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1894 | */ |
d00806b1 | 1895 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1896 | goto page_not_uptodate; |
1897 | ||
ef00e08e LT |
1898 | /* |
1899 | * Found the page and have a reference on it. | |
1900 | * We must recheck i_size under page lock. | |
1901 | */ | |
99e3e53f KS |
1902 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
1903 | if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) { | |
d00806b1 | 1904 | unlock_page(page); |
745ad48e | 1905 | page_cache_release(page); |
5307cc1a | 1906 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1907 | } |
1908 | ||
d0217ac0 | 1909 | vmf->page = page; |
83c54070 | 1910 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 1911 | |
1da177e4 LT |
1912 | no_cached_page: |
1913 | /* | |
1914 | * We're only likely to ever get here if MADV_RANDOM is in | |
1915 | * effect. | |
1916 | */ | |
ef00e08e | 1917 | error = page_cache_read(file, offset); |
1da177e4 LT |
1918 | |
1919 | /* | |
1920 | * The page we want has now been added to the page cache. | |
1921 | * In the unlikely event that someone removed it in the | |
1922 | * meantime, we'll just come back here and read it again. | |
1923 | */ | |
1924 | if (error >= 0) | |
1925 | goto retry_find; | |
1926 | ||
1927 | /* | |
1928 | * An error return from page_cache_read can result if the | |
1929 | * system is low on memory, or a problem occurs while trying | |
1930 | * to schedule I/O. | |
1931 | */ | |
1932 | if (error == -ENOMEM) | |
d0217ac0 NP |
1933 | return VM_FAULT_OOM; |
1934 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1935 | |
1936 | page_not_uptodate: | |
1da177e4 LT |
1937 | /* |
1938 | * Umm, take care of errors if the page isn't up-to-date. | |
1939 | * Try to re-read it _once_. We do this synchronously, | |
1940 | * because there really aren't any performance issues here | |
1941 | * and we need to check for errors. | |
1942 | */ | |
1da177e4 | 1943 | ClearPageError(page); |
994fc28c | 1944 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
1945 | if (!error) { |
1946 | wait_on_page_locked(page); | |
1947 | if (!PageUptodate(page)) | |
1948 | error = -EIO; | |
1949 | } | |
d00806b1 NP |
1950 | page_cache_release(page); |
1951 | ||
1952 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 1953 | goto retry_find; |
1da177e4 | 1954 | |
d00806b1 | 1955 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 1956 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 1957 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
1958 | } |
1959 | EXPORT_SYMBOL(filemap_fault); | |
1960 | ||
f1820361 KS |
1961 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
1962 | { | |
1963 | struct radix_tree_iter iter; | |
1964 | void **slot; | |
1965 | struct file *file = vma->vm_file; | |
1966 | struct address_space *mapping = file->f_mapping; | |
1967 | loff_t size; | |
1968 | struct page *page; | |
1969 | unsigned long address = (unsigned long) vmf->virtual_address; | |
1970 | unsigned long addr; | |
1971 | pte_t *pte; | |
1972 | ||
1973 | rcu_read_lock(); | |
1974 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) { | |
1975 | if (iter.index > vmf->max_pgoff) | |
1976 | break; | |
1977 | repeat: | |
1978 | page = radix_tree_deref_slot(slot); | |
1979 | if (unlikely(!page)) | |
1980 | goto next; | |
1981 | if (radix_tree_exception(page)) { | |
1982 | if (radix_tree_deref_retry(page)) | |
1983 | break; | |
1984 | else | |
1985 | goto next; | |
1986 | } | |
1987 | ||
1988 | if (!page_cache_get_speculative(page)) | |
1989 | goto repeat; | |
1990 | ||
1991 | /* Has the page moved? */ | |
1992 | if (unlikely(page != *slot)) { | |
1993 | page_cache_release(page); | |
1994 | goto repeat; | |
1995 | } | |
1996 | ||
1997 | if (!PageUptodate(page) || | |
1998 | PageReadahead(page) || | |
1999 | PageHWPoison(page)) | |
2000 | goto skip; | |
2001 | if (!trylock_page(page)) | |
2002 | goto skip; | |
2003 | ||
2004 | if (page->mapping != mapping || !PageUptodate(page)) | |
2005 | goto unlock; | |
2006 | ||
99e3e53f KS |
2007 | size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE); |
2008 | if (page->index >= size >> PAGE_CACHE_SHIFT) | |
f1820361 KS |
2009 | goto unlock; |
2010 | ||
2011 | pte = vmf->pte + page->index - vmf->pgoff; | |
2012 | if (!pte_none(*pte)) | |
2013 | goto unlock; | |
2014 | ||
2015 | if (file->f_ra.mmap_miss > 0) | |
2016 | file->f_ra.mmap_miss--; | |
2017 | addr = address + (page->index - vmf->pgoff) * PAGE_SIZE; | |
2018 | do_set_pte(vma, addr, page, pte, false, false); | |
2019 | unlock_page(page); | |
2020 | goto next; | |
2021 | unlock: | |
2022 | unlock_page(page); | |
2023 | skip: | |
2024 | page_cache_release(page); | |
2025 | next: | |
2026 | if (iter.index == vmf->max_pgoff) | |
2027 | break; | |
2028 | } | |
2029 | rcu_read_unlock(); | |
2030 | } | |
2031 | EXPORT_SYMBOL(filemap_map_pages); | |
2032 | ||
4fcf1c62 JK |
2033 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2034 | { | |
2035 | struct page *page = vmf->page; | |
496ad9aa | 2036 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
2037 | int ret = VM_FAULT_LOCKED; |
2038 | ||
14da9200 | 2039 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
2040 | file_update_time(vma->vm_file); |
2041 | lock_page(page); | |
2042 | if (page->mapping != inode->i_mapping) { | |
2043 | unlock_page(page); | |
2044 | ret = VM_FAULT_NOPAGE; | |
2045 | goto out; | |
2046 | } | |
14da9200 JK |
2047 | /* |
2048 | * We mark the page dirty already here so that when freeze is in | |
2049 | * progress, we are guaranteed that writeback during freezing will | |
2050 | * see the dirty page and writeprotect it again. | |
2051 | */ | |
2052 | set_page_dirty(page); | |
1d1d1a76 | 2053 | wait_for_stable_page(page); |
4fcf1c62 | 2054 | out: |
14da9200 | 2055 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2056 | return ret; |
2057 | } | |
2058 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2059 | ||
f0f37e2f | 2060 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2061 | .fault = filemap_fault, |
f1820361 | 2062 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2063 | .page_mkwrite = filemap_page_mkwrite, |
0b173bc4 | 2064 | .remap_pages = generic_file_remap_pages, |
1da177e4 LT |
2065 | }; |
2066 | ||
2067 | /* This is used for a general mmap of a disk file */ | |
2068 | ||
2069 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2070 | { | |
2071 | struct address_space *mapping = file->f_mapping; | |
2072 | ||
2073 | if (!mapping->a_ops->readpage) | |
2074 | return -ENOEXEC; | |
2075 | file_accessed(file); | |
2076 | vma->vm_ops = &generic_file_vm_ops; | |
2077 | return 0; | |
2078 | } | |
1da177e4 LT |
2079 | |
2080 | /* | |
2081 | * This is for filesystems which do not implement ->writepage. | |
2082 | */ | |
2083 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2084 | { | |
2085 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2086 | return -EINVAL; | |
2087 | return generic_file_mmap(file, vma); | |
2088 | } | |
2089 | #else | |
2090 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2091 | { | |
2092 | return -ENOSYS; | |
2093 | } | |
2094 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2095 | { | |
2096 | return -ENOSYS; | |
2097 | } | |
2098 | #endif /* CONFIG_MMU */ | |
2099 | ||
2100 | EXPORT_SYMBOL(generic_file_mmap); | |
2101 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2102 | ||
67f9fd91 SL |
2103 | static struct page *wait_on_page_read(struct page *page) |
2104 | { | |
2105 | if (!IS_ERR(page)) { | |
2106 | wait_on_page_locked(page); | |
2107 | if (!PageUptodate(page)) { | |
2108 | page_cache_release(page); | |
2109 | page = ERR_PTR(-EIO); | |
2110 | } | |
2111 | } | |
2112 | return page; | |
2113 | } | |
2114 | ||
6fe6900e | 2115 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 2116 | pgoff_t index, |
5e5358e7 | 2117 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2118 | void *data, |
2119 | gfp_t gfp) | |
1da177e4 | 2120 | { |
eb2be189 | 2121 | struct page *page; |
1da177e4 LT |
2122 | int err; |
2123 | repeat: | |
2124 | page = find_get_page(mapping, index); | |
2125 | if (!page) { | |
0531b2aa | 2126 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
2127 | if (!page) |
2128 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2129 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 NP |
2130 | if (unlikely(err)) { |
2131 | page_cache_release(page); | |
2132 | if (err == -EEXIST) | |
2133 | goto repeat; | |
1da177e4 | 2134 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2135 | return ERR_PTR(err); |
2136 | } | |
1da177e4 LT |
2137 | err = filler(data, page); |
2138 | if (err < 0) { | |
2139 | page_cache_release(page); | |
2140 | page = ERR_PTR(err); | |
67f9fd91 SL |
2141 | } else { |
2142 | page = wait_on_page_read(page); | |
1da177e4 LT |
2143 | } |
2144 | } | |
1da177e4 LT |
2145 | return page; |
2146 | } | |
2147 | ||
0531b2aa | 2148 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2149 | pgoff_t index, |
5e5358e7 | 2150 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2151 | void *data, |
2152 | gfp_t gfp) | |
2153 | ||
1da177e4 LT |
2154 | { |
2155 | struct page *page; | |
2156 | int err; | |
2157 | ||
2158 | retry: | |
0531b2aa | 2159 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 2160 | if (IS_ERR(page)) |
c855ff37 | 2161 | return page; |
1da177e4 LT |
2162 | if (PageUptodate(page)) |
2163 | goto out; | |
2164 | ||
2165 | lock_page(page); | |
2166 | if (!page->mapping) { | |
2167 | unlock_page(page); | |
2168 | page_cache_release(page); | |
2169 | goto retry; | |
2170 | } | |
2171 | if (PageUptodate(page)) { | |
2172 | unlock_page(page); | |
2173 | goto out; | |
2174 | } | |
2175 | err = filler(data, page); | |
2176 | if (err < 0) { | |
2177 | page_cache_release(page); | |
c855ff37 | 2178 | return ERR_PTR(err); |
67f9fd91 SL |
2179 | } else { |
2180 | page = wait_on_page_read(page); | |
2181 | if (IS_ERR(page)) | |
2182 | return page; | |
1da177e4 | 2183 | } |
c855ff37 | 2184 | out: |
6fe6900e NP |
2185 | mark_page_accessed(page); |
2186 | return page; | |
2187 | } | |
0531b2aa LT |
2188 | |
2189 | /** | |
67f9fd91 | 2190 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2191 | * @mapping: the page's address_space |
2192 | * @index: the page index | |
2193 | * @filler: function to perform the read | |
5e5358e7 | 2194 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2195 | * |
0531b2aa | 2196 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2197 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2198 | * |
2199 | * If the page does not get brought uptodate, return -EIO. | |
2200 | */ | |
67f9fd91 | 2201 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2202 | pgoff_t index, |
5e5358e7 | 2203 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2204 | void *data) |
2205 | { | |
2206 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2207 | } | |
67f9fd91 | 2208 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2209 | |
2210 | /** | |
2211 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2212 | * @mapping: the page's address_space | |
2213 | * @index: the page index | |
2214 | * @gfp: the page allocator flags to use if allocating | |
2215 | * | |
2216 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2217 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2218 | * |
2219 | * If the page does not get brought uptodate, return -EIO. | |
2220 | */ | |
2221 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2222 | pgoff_t index, | |
2223 | gfp_t gfp) | |
2224 | { | |
2225 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2226 | ||
67f9fd91 | 2227 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2228 | } |
2229 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2230 | ||
1da177e4 LT |
2231 | /* |
2232 | * Performs necessary checks before doing a write | |
2233 | * | |
485bb99b | 2234 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2235 | * Returns appropriate error code that caller should return or |
2236 | * zero in case that write should be allowed. | |
2237 | */ | |
2238 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) | |
2239 | { | |
2240 | struct inode *inode = file->f_mapping->host; | |
59e99e5b | 2241 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
1da177e4 LT |
2242 | |
2243 | if (unlikely(*pos < 0)) | |
2244 | return -EINVAL; | |
2245 | ||
1da177e4 LT |
2246 | if (!isblk) { |
2247 | /* FIXME: this is for backwards compatibility with 2.4 */ | |
2248 | if (file->f_flags & O_APPEND) | |
2249 | *pos = i_size_read(inode); | |
2250 | ||
2251 | if (limit != RLIM_INFINITY) { | |
2252 | if (*pos >= limit) { | |
2253 | send_sig(SIGXFSZ, current, 0); | |
2254 | return -EFBIG; | |
2255 | } | |
2256 | if (*count > limit - (typeof(limit))*pos) { | |
2257 | *count = limit - (typeof(limit))*pos; | |
2258 | } | |
2259 | } | |
2260 | } | |
2261 | ||
2262 | /* | |
2263 | * LFS rule | |
2264 | */ | |
2265 | if (unlikely(*pos + *count > MAX_NON_LFS && | |
2266 | !(file->f_flags & O_LARGEFILE))) { | |
2267 | if (*pos >= MAX_NON_LFS) { | |
1da177e4 LT |
2268 | return -EFBIG; |
2269 | } | |
2270 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { | |
2271 | *count = MAX_NON_LFS - (unsigned long)*pos; | |
2272 | } | |
2273 | } | |
2274 | ||
2275 | /* | |
2276 | * Are we about to exceed the fs block limit ? | |
2277 | * | |
2278 | * If we have written data it becomes a short write. If we have | |
2279 | * exceeded without writing data we send a signal and return EFBIG. | |
2280 | * Linus frestrict idea will clean these up nicely.. | |
2281 | */ | |
2282 | if (likely(!isblk)) { | |
2283 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { | |
2284 | if (*count || *pos > inode->i_sb->s_maxbytes) { | |
1da177e4 LT |
2285 | return -EFBIG; |
2286 | } | |
2287 | /* zero-length writes at ->s_maxbytes are OK */ | |
2288 | } | |
2289 | ||
2290 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) | |
2291 | *count = inode->i_sb->s_maxbytes - *pos; | |
2292 | } else { | |
9361401e | 2293 | #ifdef CONFIG_BLOCK |
1da177e4 LT |
2294 | loff_t isize; |
2295 | if (bdev_read_only(I_BDEV(inode))) | |
2296 | return -EPERM; | |
2297 | isize = i_size_read(inode); | |
2298 | if (*pos >= isize) { | |
2299 | if (*count || *pos > isize) | |
2300 | return -ENOSPC; | |
2301 | } | |
2302 | ||
2303 | if (*pos + *count > isize) | |
2304 | *count = isize - *pos; | |
9361401e DH |
2305 | #else |
2306 | return -EPERM; | |
2307 | #endif | |
1da177e4 LT |
2308 | } |
2309 | return 0; | |
2310 | } | |
2311 | EXPORT_SYMBOL(generic_write_checks); | |
2312 | ||
afddba49 NP |
2313 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2314 | loff_t pos, unsigned len, unsigned flags, | |
2315 | struct page **pagep, void **fsdata) | |
2316 | { | |
2317 | const struct address_space_operations *aops = mapping->a_ops; | |
2318 | ||
4e02ed4b | 2319 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2320 | pagep, fsdata); |
afddba49 NP |
2321 | } |
2322 | EXPORT_SYMBOL(pagecache_write_begin); | |
2323 | ||
2324 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2325 | loff_t pos, unsigned len, unsigned copied, | |
2326 | struct page *page, void *fsdata) | |
2327 | { | |
2328 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2329 | |
4e02ed4b | 2330 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2331 | } |
2332 | EXPORT_SYMBOL(pagecache_write_end); | |
2333 | ||
1da177e4 | 2334 | ssize_t |
0c949334 | 2335 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos) |
1da177e4 LT |
2336 | { |
2337 | struct file *file = iocb->ki_filp; | |
2338 | struct address_space *mapping = file->f_mapping; | |
2339 | struct inode *inode = mapping->host; | |
2340 | ssize_t written; | |
a969e903 CH |
2341 | size_t write_len; |
2342 | pgoff_t end; | |
26978b8b | 2343 | struct iov_iter data; |
1da177e4 | 2344 | |
0c949334 | 2345 | write_len = iov_iter_count(from); |
a969e903 | 2346 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; |
a969e903 | 2347 | |
48b47c56 | 2348 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2349 | if (written) |
2350 | goto out; | |
2351 | ||
2352 | /* | |
2353 | * After a write we want buffered reads to be sure to go to disk to get | |
2354 | * the new data. We invalidate clean cached page from the region we're | |
2355 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2356 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2357 | */ |
2358 | if (mapping->nrpages) { | |
2359 | written = invalidate_inode_pages2_range(mapping, | |
2360 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2361 | /* |
2362 | * If a page can not be invalidated, return 0 to fall back | |
2363 | * to buffered write. | |
2364 | */ | |
2365 | if (written) { | |
2366 | if (written == -EBUSY) | |
2367 | return 0; | |
a969e903 | 2368 | goto out; |
6ccfa806 | 2369 | } |
a969e903 CH |
2370 | } |
2371 | ||
26978b8b AV |
2372 | data = *from; |
2373 | written = mapping->a_ops->direct_IO(WRITE, iocb, &data, pos); | |
a969e903 CH |
2374 | |
2375 | /* | |
2376 | * Finally, try again to invalidate clean pages which might have been | |
2377 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2378 | * if the source of the write was an mmap'ed region of the file | |
2379 | * we're writing. Either one is a pretty crazy thing to do, | |
2380 | * so we don't support it 100%. If this invalidation | |
2381 | * fails, tough, the write still worked... | |
2382 | */ | |
2383 | if (mapping->nrpages) { | |
2384 | invalidate_inode_pages2_range(mapping, | |
2385 | pos >> PAGE_CACHE_SHIFT, end); | |
2386 | } | |
2387 | ||
1da177e4 | 2388 | if (written > 0) { |
0116651c | 2389 | pos += written; |
f8579f86 | 2390 | iov_iter_advance(from, written); |
0116651c NK |
2391 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
2392 | i_size_write(inode, pos); | |
1da177e4 LT |
2393 | mark_inode_dirty(inode); |
2394 | } | |
5cb6c6c7 | 2395 | iocb->ki_pos = pos; |
1da177e4 | 2396 | } |
a969e903 | 2397 | out: |
1da177e4 LT |
2398 | return written; |
2399 | } | |
2400 | EXPORT_SYMBOL(generic_file_direct_write); | |
2401 | ||
eb2be189 NP |
2402 | /* |
2403 | * Find or create a page at the given pagecache position. Return the locked | |
2404 | * page. This function is specifically for buffered writes. | |
2405 | */ | |
54566b2c NP |
2406 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2407 | pgoff_t index, unsigned flags) | |
eb2be189 | 2408 | { |
eb2be189 | 2409 | struct page *page; |
2457aec6 | 2410 | int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT; |
0faa70cb | 2411 | |
54566b2c | 2412 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
2413 | fgp_flags |= FGP_NOFS; |
2414 | ||
2415 | page = pagecache_get_page(mapping, index, fgp_flags, | |
2416 | mapping_gfp_mask(mapping), | |
2417 | GFP_KERNEL); | |
c585a267 | 2418 | if (page) |
2457aec6 | 2419 | wait_for_stable_page(page); |
eb2be189 | 2420 | |
eb2be189 NP |
2421 | return page; |
2422 | } | |
54566b2c | 2423 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2424 | |
3b93f911 | 2425 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
2426 | struct iov_iter *i, loff_t pos) |
2427 | { | |
2428 | struct address_space *mapping = file->f_mapping; | |
2429 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2430 | long status = 0; | |
2431 | ssize_t written = 0; | |
674b892e NP |
2432 | unsigned int flags = 0; |
2433 | ||
2434 | /* | |
2435 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2436 | */ | |
2437 | if (segment_eq(get_fs(), KERNEL_DS)) | |
2438 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
afddba49 NP |
2439 | |
2440 | do { | |
2441 | struct page *page; | |
afddba49 NP |
2442 | unsigned long offset; /* Offset into pagecache page */ |
2443 | unsigned long bytes; /* Bytes to write to page */ | |
2444 | size_t copied; /* Bytes copied from user */ | |
2445 | void *fsdata; | |
2446 | ||
2447 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2448 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2449 | iov_iter_count(i)); | |
2450 | ||
2451 | again: | |
afddba49 NP |
2452 | /* |
2453 | * Bring in the user page that we will copy from _first_. | |
2454 | * Otherwise there's a nasty deadlock on copying from the | |
2455 | * same page as we're writing to, without it being marked | |
2456 | * up-to-date. | |
2457 | * | |
2458 | * Not only is this an optimisation, but it is also required | |
2459 | * to check that the address is actually valid, when atomic | |
2460 | * usercopies are used, below. | |
2461 | */ | |
2462 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2463 | status = -EFAULT; | |
2464 | break; | |
2465 | } | |
2466 | ||
674b892e | 2467 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 2468 | &page, &fsdata); |
2457aec6 | 2469 | if (unlikely(status < 0)) |
afddba49 NP |
2470 | break; |
2471 | ||
931e80e4 | 2472 | if (mapping_writably_mapped(mapping)) |
2473 | flush_dcache_page(page); | |
2474 | ||
afddba49 | 2475 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
2476 | flush_dcache_page(page); |
2477 | ||
2478 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
2479 | page, fsdata); | |
2480 | if (unlikely(status < 0)) | |
2481 | break; | |
2482 | copied = status; | |
2483 | ||
2484 | cond_resched(); | |
2485 | ||
124d3b70 | 2486 | iov_iter_advance(i, copied); |
afddba49 NP |
2487 | if (unlikely(copied == 0)) { |
2488 | /* | |
2489 | * If we were unable to copy any data at all, we must | |
2490 | * fall back to a single segment length write. | |
2491 | * | |
2492 | * If we didn't fallback here, we could livelock | |
2493 | * because not all segments in the iov can be copied at | |
2494 | * once without a pagefault. | |
2495 | */ | |
2496 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2497 | iov_iter_single_seg_count(i)); | |
2498 | goto again; | |
2499 | } | |
afddba49 NP |
2500 | pos += copied; |
2501 | written += copied; | |
2502 | ||
2503 | balance_dirty_pages_ratelimited(mapping); | |
a50527b1 JK |
2504 | if (fatal_signal_pending(current)) { |
2505 | status = -EINTR; | |
2506 | break; | |
2507 | } | |
afddba49 NP |
2508 | } while (iov_iter_count(i)); |
2509 | ||
2510 | return written ? written : status; | |
2511 | } | |
3b93f911 | 2512 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 2513 | |
e4dd9de3 | 2514 | /** |
8174202b | 2515 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 2516 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 2517 | * @from: iov_iter with data to write |
e4dd9de3 JK |
2518 | * |
2519 | * This function does all the work needed for actually writing data to a | |
2520 | * file. It does all basic checks, removes SUID from the file, updates | |
2521 | * modification times and calls proper subroutines depending on whether we | |
2522 | * do direct IO or a standard buffered write. | |
2523 | * | |
2524 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2525 | * object which does not need locking at all. | |
2526 | * | |
2527 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2528 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2529 | * avoid syncing under i_mutex. | |
2530 | */ | |
8174202b | 2531 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2532 | { |
2533 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2534 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 2535 | struct inode *inode = mapping->host; |
41fc56d5 | 2536 | loff_t pos = iocb->ki_pos; |
3b93f911 | 2537 | ssize_t written = 0; |
1da177e4 | 2538 | ssize_t err; |
3b93f911 | 2539 | ssize_t status; |
8174202b | 2540 | size_t count = iov_iter_count(from); |
1da177e4 | 2541 | |
1da177e4 LT |
2542 | /* We can write back this queue in page reclaim */ |
2543 | current->backing_dev_info = mapping->backing_dev_info; | |
1da177e4 LT |
2544 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); |
2545 | if (err) | |
2546 | goto out; | |
2547 | ||
2548 | if (count == 0) | |
2549 | goto out; | |
2550 | ||
8174202b | 2551 | iov_iter_truncate(from, count); |
0c949334 | 2552 | |
2f1936b8 | 2553 | err = file_remove_suid(file); |
1da177e4 LT |
2554 | if (err) |
2555 | goto out; | |
2556 | ||
c3b2da31 JB |
2557 | err = file_update_time(file); |
2558 | if (err) | |
2559 | goto out; | |
1da177e4 LT |
2560 | |
2561 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
2562 | if (unlikely(file->f_flags & O_DIRECT)) { | |
fb5527e6 | 2563 | loff_t endbyte; |
fb5527e6 | 2564 | |
8174202b | 2565 | written = generic_file_direct_write(iocb, from, pos); |
1da177e4 LT |
2566 | if (written < 0 || written == count) |
2567 | goto out; | |
3b93f911 | 2568 | |
1da177e4 LT |
2569 | /* |
2570 | * direct-io write to a hole: fall through to buffered I/O | |
2571 | * for completing the rest of the request. | |
2572 | */ | |
2573 | pos += written; | |
2574 | count -= written; | |
3b93f911 | 2575 | |
8174202b | 2576 | status = generic_perform_write(file, from, pos); |
fb5527e6 | 2577 | /* |
3b93f911 | 2578 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
2579 | * then we want to return the number of bytes which were |
2580 | * direct-written, or the error code if that was zero. Note | |
2581 | * that this differs from normal direct-io semantics, which | |
2582 | * will return -EFOO even if some bytes were written. | |
2583 | */ | |
3b93f911 AV |
2584 | if (unlikely(status < 0) && !written) { |
2585 | err = status; | |
fb5527e6 JM |
2586 | goto out; |
2587 | } | |
3b93f911 | 2588 | iocb->ki_pos = pos + status; |
fb5527e6 JM |
2589 | /* |
2590 | * We need to ensure that the page cache pages are written to | |
2591 | * disk and invalidated to preserve the expected O_DIRECT | |
2592 | * semantics. | |
2593 | */ | |
3b93f911 | 2594 | endbyte = pos + status - 1; |
c05c4edd | 2595 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); |
fb5527e6 | 2596 | if (err == 0) { |
3b93f911 | 2597 | written += status; |
fb5527e6 JM |
2598 | invalidate_mapping_pages(mapping, |
2599 | pos >> PAGE_CACHE_SHIFT, | |
2600 | endbyte >> PAGE_CACHE_SHIFT); | |
2601 | } else { | |
2602 | /* | |
2603 | * We don't know how much we wrote, so just return | |
2604 | * the number of bytes which were direct-written | |
2605 | */ | |
2606 | } | |
2607 | } else { | |
8174202b | 2608 | written = generic_perform_write(file, from, pos); |
3b93f911 AV |
2609 | if (likely(written >= 0)) |
2610 | iocb->ki_pos = pos + written; | |
fb5527e6 | 2611 | } |
1da177e4 LT |
2612 | out: |
2613 | current->backing_dev_info = NULL; | |
2614 | return written ? written : err; | |
2615 | } | |
8174202b | 2616 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 2617 | |
e4dd9de3 | 2618 | /** |
8174202b | 2619 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 2620 | * @iocb: IO state structure |
8174202b | 2621 | * @from: iov_iter with data to write |
e4dd9de3 | 2622 | * |
8174202b | 2623 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
2624 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
2625 | * and acquires i_mutex as needed. | |
2626 | */ | |
8174202b | 2627 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2628 | { |
2629 | struct file *file = iocb->ki_filp; | |
148f948b | 2630 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2631 | ssize_t ret; |
1da177e4 | 2632 | |
1b1dcc1b | 2633 | mutex_lock(&inode->i_mutex); |
8174202b | 2634 | ret = __generic_file_write_iter(iocb, from); |
1b1dcc1b | 2635 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 2636 | |
02afc27f | 2637 | if (ret > 0) { |
1da177e4 LT |
2638 | ssize_t err; |
2639 | ||
d311d79d AV |
2640 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
2641 | if (err < 0) | |
1da177e4 LT |
2642 | ret = err; |
2643 | } | |
2644 | return ret; | |
2645 | } | |
8174202b | 2646 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 2647 | |
cf9a2ae8 DH |
2648 | /** |
2649 | * try_to_release_page() - release old fs-specific metadata on a page | |
2650 | * | |
2651 | * @page: the page which the kernel is trying to free | |
2652 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2653 | * | |
2654 | * The address_space is to try to release any data against the page | |
2655 | * (presumably at page->private). If the release was successful, return `1'. | |
2656 | * Otherwise return zero. | |
2657 | * | |
266cf658 DH |
2658 | * This may also be called if PG_fscache is set on a page, indicating that the |
2659 | * page is known to the local caching routines. | |
2660 | * | |
cf9a2ae8 | 2661 | * The @gfp_mask argument specifies whether I/O may be performed to release |
3f31fddf | 2662 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
cf9a2ae8 | 2663 | * |
cf9a2ae8 DH |
2664 | */ |
2665 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2666 | { | |
2667 | struct address_space * const mapping = page->mapping; | |
2668 | ||
2669 | BUG_ON(!PageLocked(page)); | |
2670 | if (PageWriteback(page)) | |
2671 | return 0; | |
2672 | ||
2673 | if (mapping && mapping->a_ops->releasepage) | |
2674 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2675 | return try_to_free_buffers(page); | |
2676 | } | |
2677 | ||
2678 | EXPORT_SYMBOL(try_to_release_page); |