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> |
c59ede7b | 16 | #include <linux/capability.h> |
1da177e4 | 17 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 18 | #include <linux/gfp.h> |
1da177e4 LT |
19 | #include <linux/mm.h> |
20 | #include <linux/swap.h> | |
21 | #include <linux/mman.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/file.h> | |
24 | #include <linux/uio.h> | |
25 | #include <linux/hash.h> | |
26 | #include <linux/writeback.h> | |
53253383 | 27 | #include <linux/backing-dev.h> |
1da177e4 LT |
28 | #include <linux/pagevec.h> |
29 | #include <linux/blkdev.h> | |
30 | #include <linux/security.h> | |
44110fe3 | 31 | #include <linux/cpuset.h> |
2f718ffc | 32 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
00501b53 | 33 | #include <linux/hugetlb.h> |
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 | ||
1cac41cb MB |
39 | #ifdef CONFIG_SDP |
40 | #include <sdp/cache_cleanup.h> | |
41 | #endif | |
42 | ||
fe0bfaaf RJ |
43 | #define CREATE_TRACE_POINTS |
44 | #include <trace/events/filemap.h> | |
45 | ||
1da177e4 | 46 | /* |
1da177e4 LT |
47 | * FIXME: remove all knowledge of the buffer layer from the core VM |
48 | */ | |
148f948b | 49 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 50 | |
1da177e4 LT |
51 | #include <asm/mman.h> |
52 | ||
53 | /* | |
54 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
55 | * though. | |
56 | * | |
57 | * Shared mappings now work. 15.8.1995 Bruno. | |
58 | * | |
59 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
60 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
61 | * | |
62 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
63 | */ | |
64 | ||
65 | /* | |
66 | * Lock ordering: | |
67 | * | |
c8c06efa | 68 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 69 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 HD |
70 | * ->swap_lock (exclusive_swap_page, others) |
71 | * ->mapping->tree_lock | |
1da177e4 | 72 | * |
1b1dcc1b | 73 | * ->i_mutex |
c8c06efa | 74 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
1da177e4 LT |
75 | * |
76 | * ->mmap_sem | |
c8c06efa | 77 | * ->i_mmap_rwsem |
b8072f09 | 78 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
79 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
80 | * | |
81 | * ->mmap_sem | |
82 | * ->lock_page (access_process_vm) | |
83 | * | |
ccad2365 | 84 | * ->i_mutex (generic_perform_write) |
82591e6e | 85 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
1da177e4 | 86 | * |
f758eeab | 87 | * bdi->wb.list_lock |
a66979ab | 88 | * sb_lock (fs/fs-writeback.c) |
1da177e4 LT |
89 | * ->mapping->tree_lock (__sync_single_inode) |
90 | * | |
c8c06efa | 91 | * ->i_mmap_rwsem |
1da177e4 LT |
92 | * ->anon_vma.lock (vma_adjust) |
93 | * | |
94 | * ->anon_vma.lock | |
b8072f09 | 95 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 96 | * |
b8072f09 | 97 | * ->page_table_lock or pte_lock |
5d337b91 | 98 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
99 | * ->private_lock (try_to_unmap_one) |
100 | * ->tree_lock (try_to_unmap_one) | |
101 | * ->zone.lru_lock (follow_page->mark_page_accessed) | |
053837fc | 102 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
1da177e4 LT |
103 | * ->private_lock (page_remove_rmap->set_page_dirty) |
104 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
f758eeab | 105 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 106 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
c4843a75 | 107 | * ->memcg->move_lock (page_remove_rmap->mem_cgroup_begin_page_stat) |
f758eeab | 108 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 109 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
110 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
111 | * | |
c8c06efa | 112 | * ->i_mmap_rwsem |
9a3c531d | 113 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
114 | */ |
115 | ||
f84311d7 JW |
116 | static int page_cache_tree_insert(struct address_space *mapping, |
117 | struct page *page, void **shadowp) | |
118 | { | |
119 | struct radix_tree_node *node; | |
120 | void **slot; | |
121 | int error; | |
122 | ||
123 | error = __radix_tree_create(&mapping->page_tree, page->index, | |
124 | &node, &slot); | |
125 | if (error) | |
126 | return error; | |
127 | if (*slot) { | |
128 | void *p; | |
129 | ||
130 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
131 | if (!radix_tree_exceptional_entry(p)) | |
132 | return -EEXIST; | |
133 | if (shadowp) | |
134 | *shadowp = p; | |
135 | mapping->nrshadows--; | |
136 | if (node) | |
137 | workingset_node_shadows_dec(node); | |
138 | } | |
139 | radix_tree_replace_slot(slot, page); | |
140 | mapping->nrpages++; | |
141 | if (node) { | |
142 | workingset_node_pages_inc(node); | |
143 | /* | |
144 | * Don't track node that contains actual pages. | |
145 | * | |
146 | * Avoid acquiring the list_lru lock if already | |
147 | * untracked. The list_empty() test is safe as | |
148 | * node->private_list is protected by | |
149 | * mapping->tree_lock. | |
150 | */ | |
151 | if (!list_empty(&node->private_list)) | |
152 | list_lru_del(&workingset_shadow_nodes, | |
153 | &node->private_list); | |
154 | } | |
155 | return 0; | |
156 | } | |
157 | ||
91b0abe3 JW |
158 | static void page_cache_tree_delete(struct address_space *mapping, |
159 | struct page *page, void *shadow) | |
160 | { | |
449dd698 JW |
161 | struct radix_tree_node *node; |
162 | unsigned long index; | |
163 | unsigned int offset; | |
164 | unsigned int tag; | |
165 | void **slot; | |
91b0abe3 | 166 | |
449dd698 JW |
167 | VM_BUG_ON(!PageLocked(page)); |
168 | ||
169 | __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot); | |
170 | ||
b52b7b5a JW |
171 | if (!node) { |
172 | /* | |
173 | * We need a node to properly account shadow | |
174 | * entries. Don't plant any without. XXX | |
175 | */ | |
176 | shadow = NULL; | |
177 | } | |
178 | ||
449dd698 | 179 | if (shadow) { |
91b0abe3 JW |
180 | mapping->nrshadows++; |
181 | /* | |
182 | * Make sure the nrshadows update is committed before | |
183 | * the nrpages update so that final truncate racing | |
184 | * with reclaim does not see both counters 0 at the | |
185 | * same time and miss a shadow entry. | |
186 | */ | |
187 | smp_wmb(); | |
449dd698 | 188 | } |
91b0abe3 | 189 | mapping->nrpages--; |
449dd698 JW |
190 | |
191 | if (!node) { | |
192 | /* Clear direct pointer tags in root node */ | |
193 | mapping->page_tree.gfp_mask &= __GFP_BITS_MASK; | |
194 | radix_tree_replace_slot(slot, shadow); | |
195 | return; | |
196 | } | |
197 | ||
198 | /* Clear tree tags for the removed page */ | |
199 | index = page->index; | |
200 | offset = index & RADIX_TREE_MAP_MASK; | |
201 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { | |
202 | if (test_bit(offset, node->tags[tag])) | |
203 | radix_tree_tag_clear(&mapping->page_tree, index, tag); | |
204 | } | |
205 | ||
206 | /* Delete page, swap shadow entry */ | |
207 | radix_tree_replace_slot(slot, shadow); | |
208 | workingset_node_pages_dec(node); | |
209 | if (shadow) | |
210 | workingset_node_shadows_inc(node); | |
211 | else | |
212 | if (__radix_tree_delete_node(&mapping->page_tree, node)) | |
213 | return; | |
214 | ||
215 | /* | |
216 | * Track node that only contains shadow entries. | |
217 | * | |
218 | * Avoid acquiring the list_lru lock if already tracked. The | |
219 | * list_empty() test is safe as node->private_list is | |
220 | * protected by mapping->tree_lock. | |
221 | */ | |
222 | if (!workingset_node_pages(node) && | |
223 | list_empty(&node->private_list)) { | |
224 | node->private_data = mapping; | |
225 | list_lru_add(&workingset_shadow_nodes, &node->private_list); | |
226 | } | |
91b0abe3 JW |
227 | } |
228 | ||
1da177e4 | 229 | /* |
e64a782f | 230 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 231 | * sure the page is locked and that nobody else uses it - or that usage |
c4843a75 GT |
232 | * is safe. The caller must hold the mapping's tree_lock and |
233 | * mem_cgroup_begin_page_stat(). | |
1da177e4 | 234 | */ |
c4843a75 GT |
235 | void __delete_from_page_cache(struct page *page, void *shadow, |
236 | struct mem_cgroup *memcg) | |
1da177e4 LT |
237 | { |
238 | struct address_space *mapping = page->mapping; | |
239 | ||
1cac41cb MB |
240 | #ifdef CONFIG_SDP |
241 | if(mapping_sensitive(mapping)) | |
242 | sdp_page_cleanup(page); | |
243 | #endif | |
244 | ||
fe0bfaaf | 245 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
246 | /* |
247 | * if we're uptodate, flush out into the cleancache, otherwise | |
248 | * invalidate any existing cleancache entries. We can't leave | |
249 | * stale data around in the cleancache once our page is gone | |
250 | */ | |
251 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
252 | cleancache_put_page(page); | |
253 | else | |
3167760f | 254 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 255 | |
91b0abe3 JW |
256 | page_cache_tree_delete(mapping, page, shadow); |
257 | ||
1da177e4 | 258 | page->mapping = NULL; |
b85e0eff | 259 | /* Leave page->index set: truncation lookup relies upon it */ |
91b0abe3 | 260 | |
4165b9b4 MH |
261 | /* hugetlb pages do not participate in page cache accounting. */ |
262 | if (!PageHuge(page)) | |
263 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
4b02108a KM |
264 | if (PageSwapBacked(page)) |
265 | __dec_zone_page_state(page, NR_SHMEM); | |
45426812 | 266 | BUG_ON(page_mapped(page)); |
3a692790 LT |
267 | |
268 | /* | |
b9ea2515 KK |
269 | * At this point page must be either written or cleaned by truncate. |
270 | * Dirty page here signals a bug and loss of unwritten data. | |
3a692790 | 271 | * |
b9ea2515 KK |
272 | * This fixes dirty accounting after removing the page entirely but |
273 | * leaves PageDirty set: it has no effect for truncated page and | |
274 | * anyway will be cleared before returning page into buddy allocator. | |
3a692790 | 275 | */ |
b9ea2515 | 276 | if (WARN_ON_ONCE(PageDirty(page))) |
682aa8e1 TH |
277 | account_page_cleaned(page, mapping, memcg, |
278 | inode_to_wb(mapping->host)); | |
1da177e4 LT |
279 | } |
280 | ||
702cfbf9 MK |
281 | /** |
282 | * delete_from_page_cache - delete page from page cache | |
283 | * @page: the page which the kernel is trying to remove from page cache | |
284 | * | |
285 | * This must be called only on pages that have been verified to be in the page | |
286 | * cache and locked. It will never put the page into the free list, the caller | |
287 | * has a reference on the page. | |
288 | */ | |
289 | void delete_from_page_cache(struct page *page) | |
1da177e4 LT |
290 | { |
291 | struct address_space *mapping = page->mapping; | |
c4843a75 GT |
292 | struct mem_cgroup *memcg; |
293 | unsigned long flags; | |
294 | ||
6072d13c | 295 | void (*freepage)(struct page *); |
1da177e4 | 296 | |
cd7619d6 | 297 | BUG_ON(!PageLocked(page)); |
1da177e4 | 298 | |
6072d13c | 299 | freepage = mapping->a_ops->freepage; |
c4843a75 GT |
300 | |
301 | memcg = mem_cgroup_begin_page_stat(page); | |
302 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
303 | __delete_from_page_cache(page, NULL, memcg); | |
304 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
305 | mem_cgroup_end_page_stat(memcg); | |
6072d13c LT |
306 | |
307 | if (freepage) | |
308 | freepage(page); | |
97cecb5a MK |
309 | page_cache_release(page); |
310 | } | |
311 | EXPORT_SYMBOL(delete_from_page_cache); | |
312 | ||
865ffef3 DM |
313 | static int filemap_check_errors(struct address_space *mapping) |
314 | { | |
315 | int ret = 0; | |
316 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
317 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
318 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 319 | ret = -ENOSPC; |
7fcbbaf1 JA |
320 | if (test_bit(AS_EIO, &mapping->flags) && |
321 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
322 | ret = -EIO; |
323 | return ret; | |
324 | } | |
325 | ||
1da177e4 | 326 | /** |
485bb99b | 327 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
328 | * @mapping: address space structure to write |
329 | * @start: offset in bytes where the range starts | |
469eb4d0 | 330 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 331 | * @sync_mode: enable synchronous operation |
1da177e4 | 332 | * |
485bb99b RD |
333 | * Start writeback against all of a mapping's dirty pages that lie |
334 | * within the byte offsets <start, end> inclusive. | |
335 | * | |
1da177e4 | 336 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 337 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
338 | * these two operations is that if a dirty page/buffer is encountered, it must |
339 | * be waited upon, and not just skipped over. | |
340 | */ | |
ebcf28e1 AM |
341 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
342 | loff_t end, int sync_mode) | |
1da177e4 LT |
343 | { |
344 | int ret; | |
345 | struct writeback_control wbc = { | |
346 | .sync_mode = sync_mode, | |
05fe478d | 347 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
348 | .range_start = start, |
349 | .range_end = end, | |
1da177e4 LT |
350 | }; |
351 | ||
352 | if (!mapping_cap_writeback_dirty(mapping)) | |
353 | return 0; | |
354 | ||
b16b1deb | 355 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 356 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 357 | wbc_detach_inode(&wbc); |
1da177e4 LT |
358 | return ret; |
359 | } | |
360 | ||
361 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
362 | int sync_mode) | |
363 | { | |
111ebb6e | 364 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
365 | } |
366 | ||
367 | int filemap_fdatawrite(struct address_space *mapping) | |
368 | { | |
369 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
370 | } | |
371 | EXPORT_SYMBOL(filemap_fdatawrite); | |
372 | ||
f4c0a0fd | 373 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 374 | loff_t end) |
1da177e4 LT |
375 | { |
376 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
377 | } | |
f4c0a0fd | 378 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 379 | |
485bb99b RD |
380 | /** |
381 | * filemap_flush - mostly a non-blocking flush | |
382 | * @mapping: target address_space | |
383 | * | |
1da177e4 LT |
384 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
385 | * purposes - I/O may not be started against all dirty pages. | |
386 | */ | |
387 | int filemap_flush(struct address_space *mapping) | |
388 | { | |
389 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
390 | } | |
391 | EXPORT_SYMBOL(filemap_flush); | |
392 | ||
aa750fd7 JN |
393 | static int __filemap_fdatawait_range(struct address_space *mapping, |
394 | loff_t start_byte, loff_t end_byte) | |
1da177e4 | 395 | { |
94004ed7 CH |
396 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
397 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
398 | struct pagevec pvec; |
399 | int nr_pages; | |
aa750fd7 | 400 | int ret = 0; |
1da177e4 | 401 | |
94004ed7 | 402 | if (end_byte < start_byte) |
865ffef3 | 403 | goto out; |
1da177e4 LT |
404 | |
405 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
406 | while ((index <= end) && |
407 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
408 | PAGECACHE_TAG_WRITEBACK, | |
409 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
410 | unsigned i; | |
411 | ||
412 | for (i = 0; i < nr_pages; i++) { | |
413 | struct page *page = pvec.pages[i]; | |
414 | ||
415 | /* until radix tree lookup accepts end_index */ | |
416 | if (page->index > end) | |
417 | continue; | |
418 | ||
419 | wait_on_page_writeback(page); | |
212260aa | 420 | if (TestClearPageError(page)) |
1da177e4 LT |
421 | ret = -EIO; |
422 | } | |
423 | pagevec_release(&pvec); | |
424 | cond_resched(); | |
425 | } | |
865ffef3 | 426 | out: |
aa750fd7 JN |
427 | return ret; |
428 | } | |
429 | ||
430 | /** | |
431 | * filemap_fdatawait_range - wait for writeback to complete | |
432 | * @mapping: address space structure to wait for | |
433 | * @start_byte: offset in bytes where the range starts | |
434 | * @end_byte: offset in bytes where the range ends (inclusive) | |
435 | * | |
436 | * Walk the list of under-writeback pages of the given address space | |
437 | * in the given range and wait for all of them. Check error status of | |
438 | * the address space and return it. | |
439 | * | |
440 | * Since the error status of the address space is cleared by this function, | |
441 | * callers are responsible for checking the return value and handling and/or | |
442 | * reporting the error. | |
443 | */ | |
444 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
445 | loff_t end_byte) | |
446 | { | |
447 | int ret, ret2; | |
448 | ||
449 | ret = __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
865ffef3 DM |
450 | ret2 = filemap_check_errors(mapping); |
451 | if (!ret) | |
452 | ret = ret2; | |
1da177e4 LT |
453 | |
454 | return ret; | |
455 | } | |
d3bccb6f JK |
456 | EXPORT_SYMBOL(filemap_fdatawait_range); |
457 | ||
aa750fd7 JN |
458 | /** |
459 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
460 | * @mapping: address space structure to wait for | |
461 | * | |
462 | * Walk the list of under-writeback pages of the given address space | |
463 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
464 | * does not clear error status of the address space. | |
465 | * | |
466 | * Use this function if callers don't handle errors themselves. Expected | |
467 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
468 | * fsfreeze(8) | |
469 | */ | |
470 | void filemap_fdatawait_keep_errors(struct address_space *mapping) | |
471 | { | |
472 | loff_t i_size = i_size_read(mapping->host); | |
473 | ||
474 | if (i_size == 0) | |
475 | return; | |
476 | ||
477 | __filemap_fdatawait_range(mapping, 0, i_size - 1); | |
478 | } | |
479 | ||
1da177e4 | 480 | /** |
485bb99b | 481 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 482 | * @mapping: address space structure to wait for |
485bb99b RD |
483 | * |
484 | * Walk the list of under-writeback pages of the given address space | |
aa750fd7 JN |
485 | * and wait for all of them. Check error status of the address space |
486 | * and return it. | |
487 | * | |
488 | * Since the error status of the address space is cleared by this function, | |
489 | * callers are responsible for checking the return value and handling and/or | |
490 | * reporting the error. | |
1da177e4 LT |
491 | */ |
492 | int filemap_fdatawait(struct address_space *mapping) | |
493 | { | |
494 | loff_t i_size = i_size_read(mapping->host); | |
495 | ||
496 | if (i_size == 0) | |
497 | return 0; | |
498 | ||
94004ed7 | 499 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
500 | } |
501 | EXPORT_SYMBOL(filemap_fdatawait); | |
502 | ||
503 | int filemap_write_and_wait(struct address_space *mapping) | |
504 | { | |
28fd1298 | 505 | int err = 0; |
1da177e4 LT |
506 | |
507 | if (mapping->nrpages) { | |
28fd1298 OH |
508 | err = filemap_fdatawrite(mapping); |
509 | /* | |
510 | * Even if the above returned error, the pages may be | |
511 | * written partially (e.g. -ENOSPC), so we wait for it. | |
512 | * But the -EIO is special case, it may indicate the worst | |
513 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
514 | */ | |
515 | if (err != -EIO) { | |
516 | int err2 = filemap_fdatawait(mapping); | |
517 | if (!err) | |
518 | err = err2; | |
519 | } | |
865ffef3 DM |
520 | } else { |
521 | err = filemap_check_errors(mapping); | |
1da177e4 | 522 | } |
28fd1298 | 523 | return err; |
1da177e4 | 524 | } |
28fd1298 | 525 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 526 | |
485bb99b RD |
527 | /** |
528 | * filemap_write_and_wait_range - write out & wait on a file range | |
529 | * @mapping: the address_space for the pages | |
530 | * @lstart: offset in bytes where the range starts | |
531 | * @lend: offset in bytes where the range ends (inclusive) | |
532 | * | |
469eb4d0 AM |
533 | * Write out and wait upon file offsets lstart->lend, inclusive. |
534 | * | |
535 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
536 | * that this function can be used to write to the very end-of-file (end = -1). | |
537 | */ | |
1da177e4 LT |
538 | int filemap_write_and_wait_range(struct address_space *mapping, |
539 | loff_t lstart, loff_t lend) | |
540 | { | |
28fd1298 | 541 | int err = 0; |
1da177e4 LT |
542 | |
543 | if (mapping->nrpages) { | |
28fd1298 OH |
544 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
545 | WB_SYNC_ALL); | |
546 | /* See comment of filemap_write_and_wait() */ | |
547 | if (err != -EIO) { | |
94004ed7 CH |
548 | int err2 = filemap_fdatawait_range(mapping, |
549 | lstart, lend); | |
28fd1298 OH |
550 | if (!err) |
551 | err = err2; | |
552 | } | |
865ffef3 DM |
553 | } else { |
554 | err = filemap_check_errors(mapping); | |
1da177e4 | 555 | } |
28fd1298 | 556 | return err; |
1da177e4 | 557 | } |
f6995585 | 558 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 559 | |
ef6a3c63 MS |
560 | /** |
561 | * replace_page_cache_page - replace a pagecache page with a new one | |
562 | * @old: page to be replaced | |
563 | * @new: page to replace with | |
564 | * @gfp_mask: allocation mode | |
565 | * | |
566 | * This function replaces a page in the pagecache with a new one. On | |
567 | * success it acquires the pagecache reference for the new page and | |
568 | * drops it for the old page. Both the old and new pages must be | |
569 | * locked. This function does not add the new page to the LRU, the | |
570 | * caller must do that. | |
571 | * | |
572 | * The remove + add is atomic. The only way this function can fail is | |
573 | * memory allocation failure. | |
574 | */ | |
575 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
576 | { | |
577 | int error; | |
ef6a3c63 | 578 | |
309381fe SL |
579 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
580 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
581 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 582 | |
ef6a3c63 MS |
583 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
584 | if (!error) { | |
585 | struct address_space *mapping = old->mapping; | |
586 | void (*freepage)(struct page *); | |
c4843a75 GT |
587 | struct mem_cgroup *memcg; |
588 | unsigned long flags; | |
ef6a3c63 MS |
589 | |
590 | pgoff_t offset = old->index; | |
591 | freepage = mapping->a_ops->freepage; | |
592 | ||
593 | page_cache_get(new); | |
594 | new->mapping = mapping; | |
595 | new->index = offset; | |
596 | ||
c4843a75 GT |
597 | memcg = mem_cgroup_begin_page_stat(old); |
598 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
599 | __delete_from_page_cache(old, NULL, memcg); | |
f84311d7 | 600 | error = page_cache_tree_insert(mapping, new, NULL); |
ef6a3c63 | 601 | BUG_ON(error); |
4165b9b4 MH |
602 | |
603 | /* | |
604 | * hugetlb pages do not participate in page cache accounting. | |
605 | */ | |
606 | if (!PageHuge(new)) | |
607 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
ef6a3c63 MS |
608 | if (PageSwapBacked(new)) |
609 | __inc_zone_page_state(new, NR_SHMEM); | |
c4843a75 GT |
610 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
611 | mem_cgroup_end_page_stat(memcg); | |
45637bab | 612 | mem_cgroup_replace_page(old, new); |
ef6a3c63 MS |
613 | radix_tree_preload_end(); |
614 | if (freepage) | |
615 | freepage(old); | |
616 | page_cache_release(old); | |
ef6a3c63 MS |
617 | } |
618 | ||
619 | return error; | |
620 | } | |
621 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
622 | ||
a528910e JW |
623 | static int __add_to_page_cache_locked(struct page *page, |
624 | struct address_space *mapping, | |
625 | pgoff_t offset, gfp_t gfp_mask, | |
626 | void **shadowp) | |
1da177e4 | 627 | { |
00501b53 JW |
628 | int huge = PageHuge(page); |
629 | struct mem_cgroup *memcg; | |
e286781d NP |
630 | int error; |
631 | ||
309381fe SL |
632 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
633 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d | 634 | |
00501b53 JW |
635 | if (!huge) { |
636 | error = mem_cgroup_try_charge(page, current->mm, | |
637 | gfp_mask, &memcg); | |
638 | if (error) | |
639 | return error; | |
640 | } | |
1da177e4 | 641 | |
5e4c0d97 | 642 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 643 | if (error) { |
00501b53 JW |
644 | if (!huge) |
645 | mem_cgroup_cancel_charge(page, memcg); | |
66a0c8ee KS |
646 | return error; |
647 | } | |
648 | ||
649 | page_cache_get(page); | |
650 | page->mapping = mapping; | |
651 | page->index = offset; | |
652 | ||
653 | spin_lock_irq(&mapping->tree_lock); | |
a528910e | 654 | error = page_cache_tree_insert(mapping, page, shadowp); |
66a0c8ee KS |
655 | radix_tree_preload_end(); |
656 | if (unlikely(error)) | |
657 | goto err_insert; | |
4165b9b4 MH |
658 | |
659 | /* hugetlb pages do not participate in page cache accounting. */ | |
660 | if (!huge) | |
661 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
66a0c8ee | 662 | spin_unlock_irq(&mapping->tree_lock); |
00501b53 JW |
663 | if (!huge) |
664 | mem_cgroup_commit_charge(page, memcg, false); | |
66a0c8ee KS |
665 | trace_mm_filemap_add_to_page_cache(page); |
666 | return 0; | |
667 | err_insert: | |
668 | page->mapping = NULL; | |
669 | /* Leave page->index set: truncation relies upon it */ | |
670 | spin_unlock_irq(&mapping->tree_lock); | |
00501b53 JW |
671 | if (!huge) |
672 | mem_cgroup_cancel_charge(page, memcg); | |
66a0c8ee | 673 | page_cache_release(page); |
1da177e4 LT |
674 | return error; |
675 | } | |
a528910e JW |
676 | |
677 | /** | |
678 | * add_to_page_cache_locked - add a locked page to the pagecache | |
679 | * @page: page to add | |
680 | * @mapping: the page's address_space | |
681 | * @offset: page index | |
682 | * @gfp_mask: page allocation mode | |
683 | * | |
684 | * This function is used to add a page to the pagecache. It must be locked. | |
685 | * This function does not add the page to the LRU. The caller must do that. | |
686 | */ | |
687 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
688 | pgoff_t offset, gfp_t gfp_mask) | |
689 | { | |
690 | return __add_to_page_cache_locked(page, mapping, offset, | |
691 | gfp_mask, NULL); | |
692 | } | |
e286781d | 693 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
694 | |
695 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 696 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 697 | { |
a528910e | 698 | void *shadow = NULL; |
4f98a2fe RR |
699 | int ret; |
700 | ||
a528910e JW |
701 | __set_page_locked(page); |
702 | ret = __add_to_page_cache_locked(page, mapping, offset, | |
703 | gfp_mask, &shadow); | |
704 | if (unlikely(ret)) | |
705 | __clear_page_locked(page); | |
706 | else { | |
707 | /* | |
708 | * The page might have been evicted from cache only | |
709 | * recently, in which case it should be activated like | |
710 | * any other repeatedly accessed page. | |
711 | */ | |
712 | if (shadow && workingset_refault(shadow)) { | |
713 | SetPageActive(page); | |
714 | workingset_activation(page); | |
715 | } else | |
716 | ClearPageActive(page); | |
717 | lru_cache_add(page); | |
718 | } | |
1da177e4 LT |
719 | return ret; |
720 | } | |
18bc0bbd | 721 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 722 | |
44110fe3 | 723 | #ifdef CONFIG_NUMA |
2ae88149 | 724 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 725 | { |
c0ff7453 MX |
726 | int n; |
727 | struct page *page; | |
728 | ||
44110fe3 | 729 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
730 | unsigned int cpuset_mems_cookie; |
731 | do { | |
d26914d1 | 732 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 733 | n = cpuset_mem_spread_node(); |
96db800f | 734 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 735 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 736 | |
c0ff7453 | 737 | return page; |
44110fe3 | 738 | } |
2ae88149 | 739 | return alloc_pages(gfp, 0); |
44110fe3 | 740 | } |
2ae88149 | 741 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
742 | #endif |
743 | ||
1da177e4 LT |
744 | /* |
745 | * In order to wait for pages to become available there must be | |
746 | * waitqueues associated with pages. By using a hash table of | |
747 | * waitqueues where the bucket discipline is to maintain all | |
748 | * waiters on the same queue and wake all when any of the pages | |
749 | * become available, and for the woken contexts to check to be | |
750 | * sure the appropriate page became available, this saves space | |
751 | * at a cost of "thundering herd" phenomena during rare hash | |
752 | * collisions. | |
753 | */ | |
a4796e37 | 754 | wait_queue_head_t *page_waitqueue(struct page *page) |
1da177e4 LT |
755 | { |
756 | const struct zone *zone = page_zone(page); | |
757 | ||
758 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
759 | } | |
a4796e37 | 760 | EXPORT_SYMBOL(page_waitqueue); |
1da177e4 | 761 | |
920c7a5d | 762 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
763 | { |
764 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
765 | ||
766 | if (test_bit(bit_nr, &page->flags)) | |
74316201 | 767 | __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
768 | TASK_UNINTERRUPTIBLE); |
769 | } | |
770 | EXPORT_SYMBOL(wait_on_page_bit); | |
771 | ||
f62e00cc KM |
772 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
773 | { | |
774 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
775 | ||
776 | if (!test_bit(bit_nr, &page->flags)) | |
777 | return 0; | |
778 | ||
779 | return __wait_on_bit(page_waitqueue(page), &wait, | |
74316201 | 780 | bit_wait_io, TASK_KILLABLE); |
f62e00cc KM |
781 | } |
782 | ||
cbbce822 N |
783 | int wait_on_page_bit_killable_timeout(struct page *page, |
784 | int bit_nr, unsigned long timeout) | |
785 | { | |
786 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
787 | ||
788 | wait.key.timeout = jiffies + timeout; | |
789 | if (!test_bit(bit_nr, &page->flags)) | |
790 | return 0; | |
791 | return __wait_on_bit(page_waitqueue(page), &wait, | |
792 | bit_wait_io_timeout, TASK_KILLABLE); | |
793 | } | |
794 | EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout); | |
795 | ||
385e1ca5 DH |
796 | /** |
797 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
798 | * @page: Page defining the wait queue of interest |
799 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
800 | * |
801 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
802 | */ | |
803 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
804 | { | |
805 | wait_queue_head_t *q = page_waitqueue(page); | |
806 | unsigned long flags; | |
807 | ||
808 | spin_lock_irqsave(&q->lock, flags); | |
809 | __add_wait_queue(q, waiter); | |
810 | spin_unlock_irqrestore(&q->lock, flags); | |
811 | } | |
812 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
813 | ||
1da177e4 | 814 | /** |
485bb99b | 815 | * unlock_page - unlock a locked page |
1da177e4 LT |
816 | * @page: the page |
817 | * | |
818 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
819 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 820 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
821 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
822 | * | |
8413ac9d NP |
823 | * The mb is necessary to enforce ordering between the clear_bit and the read |
824 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 825 | */ |
920c7a5d | 826 | void unlock_page(struct page *page) |
1da177e4 | 827 | { |
309381fe | 828 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
8413ac9d | 829 | clear_bit_unlock(PG_locked, &page->flags); |
4e857c58 | 830 | smp_mb__after_atomic(); |
1da177e4 LT |
831 | wake_up_page(page, PG_locked); |
832 | } | |
833 | EXPORT_SYMBOL(unlock_page); | |
834 | ||
485bb99b RD |
835 | /** |
836 | * end_page_writeback - end writeback against a page | |
837 | * @page: the page | |
1da177e4 LT |
838 | */ |
839 | void end_page_writeback(struct page *page) | |
840 | { | |
888cf2db MG |
841 | /* |
842 | * TestClearPageReclaim could be used here but it is an atomic | |
843 | * operation and overkill in this particular case. Failing to | |
844 | * shuffle a page marked for immediate reclaim is too mild to | |
845 | * justify taking an atomic operation penalty at the end of | |
846 | * ever page writeback. | |
847 | */ | |
848 | if (PageReclaim(page)) { | |
849 | ClearPageReclaim(page); | |
ac6aadb2 | 850 | rotate_reclaimable_page(page); |
888cf2db | 851 | } |
ac6aadb2 MS |
852 | |
853 | if (!test_clear_page_writeback(page)) | |
854 | BUG(); | |
855 | ||
4e857c58 | 856 | smp_mb__after_atomic(); |
1da177e4 LT |
857 | wake_up_page(page, PG_writeback); |
858 | } | |
859 | EXPORT_SYMBOL(end_page_writeback); | |
860 | ||
57d99845 MW |
861 | /* |
862 | * After completing I/O on a page, call this routine to update the page | |
863 | * flags appropriately | |
864 | */ | |
865 | void page_endio(struct page *page, int rw, int err) | |
866 | { | |
867 | if (rw == READ) { | |
868 | if (!err) { | |
869 | SetPageUptodate(page); | |
870 | } else { | |
871 | ClearPageUptodate(page); | |
872 | SetPageError(page); | |
873 | } | |
874 | unlock_page(page); | |
875 | } else { /* rw == WRITE */ | |
876 | if (err) { | |
c5c893e7 MK |
877 | struct address_space *mapping; |
878 | ||
57d99845 | 879 | SetPageError(page); |
c5c893e7 MK |
880 | mapping = page_mapping(page); |
881 | if (mapping) | |
882 | mapping_set_error(mapping, err); | |
57d99845 MW |
883 | } |
884 | end_page_writeback(page); | |
885 | } | |
886 | } | |
887 | EXPORT_SYMBOL_GPL(page_endio); | |
888 | ||
485bb99b RD |
889 | /** |
890 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
891 | * @page: the page to lock | |
1da177e4 | 892 | */ |
920c7a5d | 893 | void __lock_page(struct page *page) |
1da177e4 LT |
894 | { |
895 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
896 | ||
74316201 | 897 | __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
898 | TASK_UNINTERRUPTIBLE); |
899 | } | |
900 | EXPORT_SYMBOL(__lock_page); | |
901 | ||
b5606c2d | 902 | int __lock_page_killable(struct page *page) |
2687a356 MW |
903 | { |
904 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
905 | ||
906 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
74316201 | 907 | bit_wait_io, TASK_KILLABLE); |
2687a356 | 908 | } |
18bc0bbd | 909 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 910 | |
9a95f3cf PC |
911 | /* |
912 | * Return values: | |
913 | * 1 - page is locked; mmap_sem is still held. | |
914 | * 0 - page is not locked. | |
915 | * mmap_sem has been released (up_read()), unless flags had both | |
916 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
917 | * which case mmap_sem is still held. | |
918 | * | |
919 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
920 | * with the page locked and the mmap_sem unperturbed. | |
921 | */ | |
d065bd81 ML |
922 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
923 | unsigned int flags) | |
924 | { | |
37b23e05 KM |
925 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
926 | /* | |
927 | * CAUTION! In this case, mmap_sem is not released | |
928 | * even though return 0. | |
929 | */ | |
930 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
931 | return 0; | |
932 | ||
933 | up_read(&mm->mmap_sem); | |
934 | if (flags & FAULT_FLAG_KILLABLE) | |
935 | wait_on_page_locked_killable(page); | |
936 | else | |
318b275f | 937 | wait_on_page_locked(page); |
d065bd81 | 938 | return 0; |
37b23e05 KM |
939 | } else { |
940 | if (flags & FAULT_FLAG_KILLABLE) { | |
941 | int ret; | |
942 | ||
943 | ret = __lock_page_killable(page); | |
944 | if (ret) { | |
945 | up_read(&mm->mmap_sem); | |
946 | return 0; | |
947 | } | |
948 | } else | |
949 | __lock_page(page); | |
950 | return 1; | |
d065bd81 ML |
951 | } |
952 | } | |
953 | ||
e7b563bb JW |
954 | /** |
955 | * page_cache_next_hole - find the next hole (not-present entry) | |
956 | * @mapping: mapping | |
957 | * @index: index | |
958 | * @max_scan: maximum range to search | |
959 | * | |
960 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
961 | * lowest indexed hole. | |
962 | * | |
963 | * Returns: the index of the hole if found, otherwise returns an index | |
964 | * outside of the set specified (in which case 'return - index >= | |
965 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
966 | * be returned. | |
967 | * | |
968 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
969 | * like radix_tree_gang_lookup, this will not atomically search a | |
970 | * snapshot of the tree at a single point in time. For example, if a | |
971 | * hole is created at index 5, then subsequently a hole is created at | |
972 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
973 | * if called under rcu_read_lock. | |
974 | */ | |
975 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
976 | pgoff_t index, unsigned long max_scan) | |
977 | { | |
978 | unsigned long i; | |
979 | ||
980 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
981 | struct page *page; |
982 | ||
983 | page = radix_tree_lookup(&mapping->page_tree, index); | |
984 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
985 | break; |
986 | index++; | |
987 | if (index == 0) | |
988 | break; | |
989 | } | |
990 | ||
991 | return index; | |
992 | } | |
993 | EXPORT_SYMBOL(page_cache_next_hole); | |
994 | ||
995 | /** | |
996 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
997 | * @mapping: mapping | |
998 | * @index: index | |
999 | * @max_scan: maximum range to search | |
1000 | * | |
1001 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
1002 | * the first hole. | |
1003 | * | |
1004 | * Returns: the index of the hole if found, otherwise returns an index | |
1005 | * outside of the set specified (in which case 'index - return >= | |
1006 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
1007 | * will be returned. | |
1008 | * | |
1009 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
1010 | * like radix_tree_gang_lookup, this will not atomically search a | |
1011 | * snapshot of the tree at a single point in time. For example, if a | |
1012 | * hole is created at index 10, then subsequently a hole is created at | |
1013 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
1014 | * called under rcu_read_lock. | |
1015 | */ | |
1016 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
1017 | pgoff_t index, unsigned long max_scan) | |
1018 | { | |
1019 | unsigned long i; | |
1020 | ||
1021 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
1022 | struct page *page; |
1023 | ||
1024 | page = radix_tree_lookup(&mapping->page_tree, index); | |
1025 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
1026 | break; |
1027 | index--; | |
1028 | if (index == ULONG_MAX) | |
1029 | break; | |
1030 | } | |
1031 | ||
1032 | return index; | |
1033 | } | |
1034 | EXPORT_SYMBOL(page_cache_prev_hole); | |
1035 | ||
485bb99b | 1036 | /** |
0cd6144a | 1037 | * find_get_entry - find and get a page cache entry |
485bb99b | 1038 | * @mapping: the address_space to search |
0cd6144a JW |
1039 | * @offset: the page cache index |
1040 | * | |
1041 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1042 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1043 | * |
139b6a6f JW |
1044 | * If the slot holds a shadow entry of a previously evicted page, or a |
1045 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1046 | * |
1047 | * Otherwise, %NULL is returned. | |
1da177e4 | 1048 | */ |
0cd6144a | 1049 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1050 | { |
a60637c8 | 1051 | void **pagep; |
1da177e4 LT |
1052 | struct page *page; |
1053 | ||
a60637c8 NP |
1054 | rcu_read_lock(); |
1055 | repeat: | |
1056 | page = NULL; | |
1057 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
1058 | if (pagep) { | |
1059 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
1060 | if (unlikely(!page)) |
1061 | goto out; | |
a2c16d6c | 1062 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1063 | if (radix_tree_deref_retry(page)) |
1064 | goto repeat; | |
1065 | /* | |
139b6a6f JW |
1066 | * A shadow entry of a recently evicted page, |
1067 | * or a swap entry from shmem/tmpfs. Return | |
1068 | * it without attempting to raise page count. | |
8079b1c8 HD |
1069 | */ |
1070 | goto out; | |
a2c16d6c | 1071 | } |
a60637c8 NP |
1072 | if (!page_cache_get_speculative(page)) |
1073 | goto repeat; | |
1074 | ||
1075 | /* | |
1076 | * Has the page moved? | |
1077 | * This is part of the lockless pagecache protocol. See | |
1078 | * include/linux/pagemap.h for details. | |
1079 | */ | |
1080 | if (unlikely(page != *pagep)) { | |
1081 | page_cache_release(page); | |
1082 | goto repeat; | |
1083 | } | |
1084 | } | |
27d20fdd | 1085 | out: |
a60637c8 NP |
1086 | rcu_read_unlock(); |
1087 | ||
1da177e4 LT |
1088 | return page; |
1089 | } | |
0cd6144a | 1090 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1091 | |
0cd6144a JW |
1092 | /** |
1093 | * find_lock_entry - locate, pin and lock a page cache entry | |
1094 | * @mapping: the address_space to search | |
1095 | * @offset: the page cache index | |
1096 | * | |
1097 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1098 | * page cache page, it is returned locked and with an increased | |
1099 | * refcount. | |
1100 | * | |
139b6a6f JW |
1101 | * If the slot holds a shadow entry of a previously evicted page, or a |
1102 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1103 | * |
1104 | * Otherwise, %NULL is returned. | |
1105 | * | |
1106 | * find_lock_entry() may sleep. | |
1107 | */ | |
1108 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1109 | { |
1110 | struct page *page; | |
1111 | ||
1da177e4 | 1112 | repeat: |
0cd6144a | 1113 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1114 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1115 | lock_page(page); |
1116 | /* Has the page been truncated? */ | |
1117 | if (unlikely(page->mapping != mapping)) { | |
1118 | unlock_page(page); | |
1119 | page_cache_release(page); | |
1120 | goto repeat; | |
1da177e4 | 1121 | } |
309381fe | 1122 | VM_BUG_ON_PAGE(page->index != offset, page); |
1da177e4 | 1123 | } |
1da177e4 LT |
1124 | return page; |
1125 | } | |
0cd6144a JW |
1126 | EXPORT_SYMBOL(find_lock_entry); |
1127 | ||
1128 | /** | |
2457aec6 | 1129 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1130 | * @mapping: the address_space to search |
1131 | * @offset: the page index | |
2457aec6 | 1132 | * @fgp_flags: PCG flags |
45f87de5 | 1133 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
0cd6144a | 1134 | * |
2457aec6 | 1135 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1136 | * |
75325189 | 1137 | * PCG flags modify how the page is returned. |
0cd6144a | 1138 | * |
2457aec6 MG |
1139 | * FGP_ACCESSED: the page will be marked accessed |
1140 | * FGP_LOCK: Page is return locked | |
1141 | * FGP_CREAT: If page is not present then a new page is allocated using | |
45f87de5 MH |
1142 | * @gfp_mask and added to the page cache and the VM's LRU |
1143 | * list. The page is returned locked and with an increased | |
1144 | * refcount. Otherwise, %NULL is returned. | |
1da177e4 | 1145 | * |
2457aec6 MG |
1146 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1147 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1148 | * |
2457aec6 | 1149 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1150 | */ |
2457aec6 | 1151 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
45f87de5 | 1152 | int fgp_flags, gfp_t gfp_mask) |
1da177e4 | 1153 | { |
eb2be189 | 1154 | struct page *page; |
2457aec6 | 1155 | |
1da177e4 | 1156 | repeat: |
2457aec6 MG |
1157 | page = find_get_entry(mapping, offset); |
1158 | if (radix_tree_exceptional_entry(page)) | |
1159 | page = NULL; | |
1160 | if (!page) | |
1161 | goto no_page; | |
1162 | ||
1163 | if (fgp_flags & FGP_LOCK) { | |
1164 | if (fgp_flags & FGP_NOWAIT) { | |
1165 | if (!trylock_page(page)) { | |
1166 | page_cache_release(page); | |
1167 | return NULL; | |
1168 | } | |
1169 | } else { | |
1170 | lock_page(page); | |
1171 | } | |
1172 | ||
1173 | /* Has the page been truncated? */ | |
1174 | if (unlikely(page->mapping != mapping)) { | |
1175 | unlock_page(page); | |
1176 | page_cache_release(page); | |
1177 | goto repeat; | |
1178 | } | |
1179 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1180 | } | |
1181 | ||
1182 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1183 | mark_page_accessed(page); | |
1184 | ||
1185 | no_page: | |
1186 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1187 | int err; | |
1188 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1189 | gfp_mask |= __GFP_WRITE; |
1190 | if (fgp_flags & FGP_NOFS) | |
1191 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1192 | |
45f87de5 | 1193 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1194 | if (!page) |
1195 | return NULL; | |
2457aec6 MG |
1196 | |
1197 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1198 | fgp_flags |= FGP_LOCK; | |
1199 | ||
eb39d618 | 1200 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1201 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1202 | __SetPageReferenced(page); |
2457aec6 | 1203 | |
45f87de5 MH |
1204 | err = add_to_page_cache_lru(page, mapping, offset, |
1205 | gfp_mask & GFP_RECLAIM_MASK); | |
eb2be189 NP |
1206 | if (unlikely(err)) { |
1207 | page_cache_release(page); | |
1208 | page = NULL; | |
1209 | if (err == -EEXIST) | |
1210 | goto repeat; | |
1da177e4 | 1211 | } |
1da177e4 | 1212 | } |
2457aec6 | 1213 | |
1da177e4 LT |
1214 | return page; |
1215 | } | |
2457aec6 | 1216 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1217 | |
0cd6144a JW |
1218 | /** |
1219 | * find_get_entries - gang pagecache lookup | |
1220 | * @mapping: The address_space to search | |
1221 | * @start: The starting page cache index | |
1222 | * @nr_entries: The maximum number of entries | |
1223 | * @entries: Where the resulting entries are placed | |
1224 | * @indices: The cache indices corresponding to the entries in @entries | |
1225 | * | |
1226 | * find_get_entries() will search for and return a group of up to | |
1227 | * @nr_entries entries in the mapping. The entries are placed at | |
1228 | * @entries. find_get_entries() takes a reference against any actual | |
1229 | * pages it returns. | |
1230 | * | |
1231 | * The search returns a group of mapping-contiguous page cache entries | |
1232 | * with ascending indexes. There may be holes in the indices due to | |
1233 | * not-present pages. | |
1234 | * | |
139b6a6f JW |
1235 | * Any shadow entries of evicted pages, or swap entries from |
1236 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1237 | * |
1238 | * find_get_entries() returns the number of pages and shadow entries | |
1239 | * which were found. | |
1240 | */ | |
1241 | unsigned find_get_entries(struct address_space *mapping, | |
1242 | pgoff_t start, unsigned int nr_entries, | |
1243 | struct page **entries, pgoff_t *indices) | |
1244 | { | |
1245 | void **slot; | |
1246 | unsigned int ret = 0; | |
1247 | struct radix_tree_iter iter; | |
1248 | ||
1249 | if (!nr_entries) | |
1250 | return 0; | |
1251 | ||
1252 | rcu_read_lock(); | |
1253 | restart: | |
1254 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { | |
1255 | struct page *page; | |
1256 | repeat: | |
1257 | page = radix_tree_deref_slot(slot); | |
1258 | if (unlikely(!page)) | |
1259 | continue; | |
1260 | if (radix_tree_exception(page)) { | |
1261 | if (radix_tree_deref_retry(page)) | |
1262 | goto restart; | |
1263 | /* | |
139b6a6f JW |
1264 | * A shadow entry of a recently evicted page, |
1265 | * or a swap entry from shmem/tmpfs. Return | |
1266 | * it without attempting to raise page count. | |
0cd6144a JW |
1267 | */ |
1268 | goto export; | |
1269 | } | |
1270 | if (!page_cache_get_speculative(page)) | |
1271 | goto repeat; | |
1272 | ||
1273 | /* Has the page moved? */ | |
1274 | if (unlikely(page != *slot)) { | |
1275 | page_cache_release(page); | |
1276 | goto repeat; | |
1277 | } | |
1278 | export: | |
1279 | indices[ret] = iter.index; | |
1280 | entries[ret] = page; | |
1281 | if (++ret == nr_entries) | |
1282 | break; | |
1283 | } | |
1284 | rcu_read_unlock(); | |
1285 | return ret; | |
1286 | } | |
1287 | ||
1da177e4 LT |
1288 | /** |
1289 | * find_get_pages - gang pagecache lookup | |
1290 | * @mapping: The address_space to search | |
1291 | * @start: The starting page index | |
1292 | * @nr_pages: The maximum number of pages | |
1293 | * @pages: Where the resulting pages are placed | |
1294 | * | |
1295 | * find_get_pages() will search for and return a group of up to | |
1296 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
1297 | * find_get_pages() takes a reference against the returned pages. | |
1298 | * | |
1299 | * The search returns a group of mapping-contiguous pages with ascending | |
1300 | * indexes. There may be holes in the indices due to not-present pages. | |
1301 | * | |
1302 | * find_get_pages() returns the number of pages which were found. | |
1303 | */ | |
1304 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
1305 | unsigned int nr_pages, struct page **pages) | |
1306 | { | |
0fc9d104 KK |
1307 | struct radix_tree_iter iter; |
1308 | void **slot; | |
1309 | unsigned ret = 0; | |
1310 | ||
1311 | if (unlikely(!nr_pages)) | |
1312 | return 0; | |
a60637c8 NP |
1313 | |
1314 | rcu_read_lock(); | |
1315 | restart: | |
0fc9d104 | 1316 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
a60637c8 NP |
1317 | struct page *page; |
1318 | repeat: | |
0fc9d104 | 1319 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1320 | if (unlikely(!page)) |
1321 | continue; | |
9d8aa4ea | 1322 | |
a2c16d6c | 1323 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1324 | if (radix_tree_deref_retry(page)) { |
1325 | /* | |
1326 | * Transient condition which can only trigger | |
1327 | * when entry at index 0 moves out of or back | |
1328 | * to root: none yet gotten, safe to restart. | |
1329 | */ | |
0fc9d104 | 1330 | WARN_ON(iter.index); |
8079b1c8 HD |
1331 | goto restart; |
1332 | } | |
a2c16d6c | 1333 | /* |
139b6a6f JW |
1334 | * A shadow entry of a recently evicted page, |
1335 | * or a swap entry from shmem/tmpfs. Skip | |
1336 | * over it. | |
a2c16d6c | 1337 | */ |
8079b1c8 | 1338 | continue; |
27d20fdd | 1339 | } |
a60637c8 NP |
1340 | |
1341 | if (!page_cache_get_speculative(page)) | |
1342 | goto repeat; | |
1343 | ||
1344 | /* Has the page moved? */ | |
0fc9d104 | 1345 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1346 | page_cache_release(page); |
1347 | goto repeat; | |
1348 | } | |
1da177e4 | 1349 | |
a60637c8 | 1350 | pages[ret] = page; |
0fc9d104 KK |
1351 | if (++ret == nr_pages) |
1352 | break; | |
a60637c8 | 1353 | } |
5b280c0c | 1354 | |
a60637c8 | 1355 | rcu_read_unlock(); |
1da177e4 LT |
1356 | return ret; |
1357 | } | |
1358 | ||
ebf43500 JA |
1359 | /** |
1360 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1361 | * @mapping: The address_space to search | |
1362 | * @index: The starting page index | |
1363 | * @nr_pages: The maximum number of pages | |
1364 | * @pages: Where the resulting pages are placed | |
1365 | * | |
1366 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1367 | * that the returned number of pages are guaranteed to be contiguous. | |
1368 | * | |
1369 | * find_get_pages_contig() returns the number of pages which were found. | |
1370 | */ | |
1371 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1372 | unsigned int nr_pages, struct page **pages) | |
1373 | { | |
0fc9d104 KK |
1374 | struct radix_tree_iter iter; |
1375 | void **slot; | |
1376 | unsigned int ret = 0; | |
1377 | ||
1378 | if (unlikely(!nr_pages)) | |
1379 | return 0; | |
a60637c8 NP |
1380 | |
1381 | rcu_read_lock(); | |
1382 | restart: | |
0fc9d104 | 1383 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
a60637c8 NP |
1384 | struct page *page; |
1385 | repeat: | |
0fc9d104 KK |
1386 | page = radix_tree_deref_slot(slot); |
1387 | /* The hole, there no reason to continue */ | |
a60637c8 | 1388 | if (unlikely(!page)) |
0fc9d104 | 1389 | break; |
9d8aa4ea | 1390 | |
a2c16d6c | 1391 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1392 | if (radix_tree_deref_retry(page)) { |
1393 | /* | |
1394 | * Transient condition which can only trigger | |
1395 | * when entry at index 0 moves out of or back | |
1396 | * to root: none yet gotten, safe to restart. | |
1397 | */ | |
1398 | goto restart; | |
1399 | } | |
a2c16d6c | 1400 | /* |
139b6a6f JW |
1401 | * A shadow entry of a recently evicted page, |
1402 | * or a swap entry from shmem/tmpfs. Stop | |
1403 | * looking for contiguous pages. | |
a2c16d6c | 1404 | */ |
8079b1c8 | 1405 | break; |
a2c16d6c | 1406 | } |
ebf43500 | 1407 | |
a60637c8 NP |
1408 | if (!page_cache_get_speculative(page)) |
1409 | goto repeat; | |
1410 | ||
1411 | /* Has the page moved? */ | |
0fc9d104 | 1412 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1413 | page_cache_release(page); |
1414 | goto repeat; | |
1415 | } | |
1416 | ||
9cbb4cb2 NP |
1417 | /* |
1418 | * must check mapping and index after taking the ref. | |
1419 | * otherwise we can get both false positives and false | |
1420 | * negatives, which is just confusing to the caller. | |
1421 | */ | |
0fc9d104 | 1422 | if (page->mapping == NULL || page->index != iter.index) { |
9cbb4cb2 NP |
1423 | page_cache_release(page); |
1424 | break; | |
1425 | } | |
1426 | ||
a60637c8 | 1427 | pages[ret] = page; |
0fc9d104 KK |
1428 | if (++ret == nr_pages) |
1429 | break; | |
ebf43500 | 1430 | } |
a60637c8 NP |
1431 | rcu_read_unlock(); |
1432 | return ret; | |
ebf43500 | 1433 | } |
ef71c15c | 1434 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1435 | |
485bb99b RD |
1436 | /** |
1437 | * find_get_pages_tag - find and return pages that match @tag | |
1438 | * @mapping: the address_space to search | |
1439 | * @index: the starting page index | |
1440 | * @tag: the tag index | |
1441 | * @nr_pages: the maximum number of pages | |
1442 | * @pages: where the resulting pages are placed | |
1443 | * | |
1da177e4 | 1444 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1445 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
1446 | */ |
1447 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
1448 | int tag, unsigned int nr_pages, struct page **pages) | |
1449 | { | |
0fc9d104 KK |
1450 | struct radix_tree_iter iter; |
1451 | void **slot; | |
1452 | unsigned ret = 0; | |
1453 | ||
1454 | if (unlikely(!nr_pages)) | |
1455 | return 0; | |
a60637c8 NP |
1456 | |
1457 | rcu_read_lock(); | |
1458 | restart: | |
0fc9d104 KK |
1459 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1460 | &iter, *index, tag) { | |
a60637c8 NP |
1461 | struct page *page; |
1462 | repeat: | |
0fc9d104 | 1463 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1464 | if (unlikely(!page)) |
1465 | continue; | |
9d8aa4ea | 1466 | |
a2c16d6c | 1467 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1468 | if (radix_tree_deref_retry(page)) { |
1469 | /* | |
1470 | * Transient condition which can only trigger | |
1471 | * when entry at index 0 moves out of or back | |
1472 | * to root: none yet gotten, safe to restart. | |
1473 | */ | |
1474 | goto restart; | |
1475 | } | |
a2c16d6c | 1476 | /* |
139b6a6f JW |
1477 | * A shadow entry of a recently evicted page. |
1478 | * | |
1479 | * Those entries should never be tagged, but | |
1480 | * this tree walk is lockless and the tags are | |
1481 | * looked up in bulk, one radix tree node at a | |
1482 | * time, so there is a sizable window for page | |
1483 | * reclaim to evict a page we saw tagged. | |
1484 | * | |
1485 | * Skip over it. | |
a2c16d6c | 1486 | */ |
139b6a6f | 1487 | continue; |
a2c16d6c | 1488 | } |
a60637c8 NP |
1489 | |
1490 | if (!page_cache_get_speculative(page)) | |
1491 | goto repeat; | |
1492 | ||
1493 | /* Has the page moved? */ | |
0fc9d104 | 1494 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1495 | page_cache_release(page); |
1496 | goto repeat; | |
1497 | } | |
1498 | ||
1499 | pages[ret] = page; | |
0fc9d104 KK |
1500 | if (++ret == nr_pages) |
1501 | break; | |
a60637c8 | 1502 | } |
5b280c0c | 1503 | |
a60637c8 | 1504 | rcu_read_unlock(); |
1da177e4 | 1505 | |
1da177e4 LT |
1506 | if (ret) |
1507 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1508 | |
1da177e4 LT |
1509 | return ret; |
1510 | } | |
ef71c15c | 1511 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1512 | |
76d42bd9 WF |
1513 | /* |
1514 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1515 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1516 | * | |
1517 | * ---R__________________________________________B__________ | |
1518 | * ^ reading here ^ bad block(assume 4k) | |
1519 | * | |
1520 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1521 | * => failing the whole request => read(R) => read(R+1) => | |
1522 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1523 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1524 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1525 | * | |
1526 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1527 | */ | |
1528 | static void shrink_readahead_size_eio(struct file *filp, | |
1529 | struct file_ra_state *ra) | |
1530 | { | |
76d42bd9 | 1531 | ra->ra_pages /= 4; |
76d42bd9 WF |
1532 | } |
1533 | ||
485bb99b | 1534 | /** |
36e78914 | 1535 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1536 | * @filp: the file to read |
1537 | * @ppos: current file position | |
6e58e79d AV |
1538 | * @iter: data destination |
1539 | * @written: already copied | |
485bb99b | 1540 | * |
1da177e4 | 1541 | * This is a generic file read routine, and uses the |
485bb99b | 1542 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1543 | * |
1544 | * This is really ugly. But the goto's actually try to clarify some | |
1545 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1546 | */ |
6e58e79d AV |
1547 | static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos, |
1548 | struct iov_iter *iter, ssize_t written) | |
1da177e4 | 1549 | { |
36e78914 | 1550 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1551 | struct inode *inode = mapping->host; |
36e78914 | 1552 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1553 | pgoff_t index; |
1554 | pgoff_t last_index; | |
1555 | pgoff_t prev_index; | |
1556 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1557 | unsigned int prev_offset; |
6e58e79d | 1558 | int error = 0; |
1da177e4 | 1559 | |
1da177e4 | 1560 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1561 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1562 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
6e58e79d | 1563 | last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1da177e4 LT |
1564 | offset = *ppos & ~PAGE_CACHE_MASK; |
1565 | ||
1da177e4 LT |
1566 | for (;;) { |
1567 | struct page *page; | |
57f6b96c | 1568 | pgoff_t end_index; |
a32ea1e1 | 1569 | loff_t isize; |
1da177e4 LT |
1570 | unsigned long nr, ret; |
1571 | ||
1da177e4 | 1572 | cond_resched(); |
1da177e4 | 1573 | find_page: |
4025ab36 MH |
1574 | if (fatal_signal_pending(current)) { |
1575 | error = -EINTR; | |
1576 | goto out; | |
1577 | } | |
1578 | ||
1da177e4 | 1579 | page = find_get_page(mapping, index); |
3ea89ee8 | 1580 | if (!page) { |
cf914a7d | 1581 | page_cache_sync_readahead(mapping, |
7ff81078 | 1582 | ra, filp, |
3ea89ee8 FW |
1583 | index, last_index - index); |
1584 | page = find_get_page(mapping, index); | |
1585 | if (unlikely(page == NULL)) | |
1586 | goto no_cached_page; | |
1587 | } | |
1588 | if (PageReadahead(page)) { | |
cf914a7d | 1589 | page_cache_async_readahead(mapping, |
7ff81078 | 1590 | ra, filp, page, |
3ea89ee8 | 1591 | index, last_index - index); |
1da177e4 | 1592 | } |
8ab22b9a HH |
1593 | if (!PageUptodate(page)) { |
1594 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1595 | !mapping->a_ops->is_partially_uptodate) | |
1596 | goto page_not_up_to_date; | |
529ae9aa | 1597 | if (!trylock_page(page)) |
8ab22b9a | 1598 | goto page_not_up_to_date; |
8d056cb9 DH |
1599 | /* Did it get truncated before we got the lock? */ |
1600 | if (!page->mapping) | |
1601 | goto page_not_up_to_date_locked; | |
8ab22b9a | 1602 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 1603 | offset, iter->count)) |
8ab22b9a HH |
1604 | goto page_not_up_to_date_locked; |
1605 | unlock_page(page); | |
1606 | } | |
1da177e4 | 1607 | page_ok: |
a32ea1e1 N |
1608 | /* |
1609 | * i_size must be checked after we know the page is Uptodate. | |
1610 | * | |
1611 | * Checking i_size after the check allows us to calculate | |
1612 | * the correct value for "nr", which means the zero-filled | |
1613 | * part of the page is not copied back to userspace (unless | |
1614 | * another truncate extends the file - this is desired though). | |
1615 | */ | |
1616 | ||
1617 | isize = i_size_read(inode); | |
1618 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1619 | if (unlikely(!isize || index > end_index)) { | |
1620 | page_cache_release(page); | |
1621 | goto out; | |
1622 | } | |
1623 | ||
1624 | /* nr is the maximum number of bytes to copy from this page */ | |
1625 | nr = PAGE_CACHE_SIZE; | |
1626 | if (index == end_index) { | |
1627 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1628 | if (nr <= offset) { | |
1629 | page_cache_release(page); | |
1630 | goto out; | |
1631 | } | |
1632 | } | |
1633 | nr = nr - offset; | |
1da177e4 LT |
1634 | |
1635 | /* If users can be writing to this page using arbitrary | |
1636 | * virtual addresses, take care about potential aliasing | |
1637 | * before reading the page on the kernel side. | |
1638 | */ | |
1639 | if (mapping_writably_mapped(mapping)) | |
1640 | flush_dcache_page(page); | |
1641 | ||
1642 | /* | |
ec0f1637 JK |
1643 | * When a sequential read accesses a page several times, |
1644 | * only mark it as accessed the first time. | |
1da177e4 | 1645 | */ |
ec0f1637 | 1646 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1647 | mark_page_accessed(page); |
1648 | prev_index = index; | |
1649 | ||
1650 | /* | |
1651 | * Ok, we have the page, and it's up-to-date, so | |
1652 | * now we can copy it to user space... | |
1da177e4 | 1653 | */ |
6e58e79d AV |
1654 | |
1655 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 LT |
1656 | offset += ret; |
1657 | index += offset >> PAGE_CACHE_SHIFT; | |
1658 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1659 | prev_offset = offset; |
1da177e4 LT |
1660 | |
1661 | page_cache_release(page); | |
6e58e79d AV |
1662 | written += ret; |
1663 | if (!iov_iter_count(iter)) | |
1664 | goto out; | |
1665 | if (ret < nr) { | |
1666 | error = -EFAULT; | |
1667 | goto out; | |
1668 | } | |
1669 | continue; | |
1da177e4 LT |
1670 | |
1671 | page_not_up_to_date: | |
1672 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1673 | error = lock_page_killable(page); |
1674 | if (unlikely(error)) | |
1675 | goto readpage_error; | |
1da177e4 | 1676 | |
8ab22b9a | 1677 | page_not_up_to_date_locked: |
da6052f7 | 1678 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1679 | if (!page->mapping) { |
1680 | unlock_page(page); | |
1681 | page_cache_release(page); | |
1682 | continue; | |
1683 | } | |
1684 | ||
1685 | /* Did somebody else fill it already? */ | |
1686 | if (PageUptodate(page)) { | |
1687 | unlock_page(page); | |
1688 | goto page_ok; | |
1689 | } | |
1690 | ||
1691 | readpage: | |
91803b49 JM |
1692 | /* |
1693 | * A previous I/O error may have been due to temporary | |
1694 | * failures, eg. multipath errors. | |
1695 | * PG_error will be set again if readpage fails. | |
1696 | */ | |
1697 | ClearPageError(page); | |
1da177e4 LT |
1698 | /* Start the actual read. The read will unlock the page. */ |
1699 | error = mapping->a_ops->readpage(filp, page); | |
1700 | ||
994fc28c ZB |
1701 | if (unlikely(error)) { |
1702 | if (error == AOP_TRUNCATED_PAGE) { | |
1703 | page_cache_release(page); | |
6e58e79d | 1704 | error = 0; |
994fc28c ZB |
1705 | goto find_page; |
1706 | } | |
1da177e4 | 1707 | goto readpage_error; |
994fc28c | 1708 | } |
1da177e4 LT |
1709 | |
1710 | if (!PageUptodate(page)) { | |
85462323 ON |
1711 | error = lock_page_killable(page); |
1712 | if (unlikely(error)) | |
1713 | goto readpage_error; | |
1da177e4 LT |
1714 | if (!PageUptodate(page)) { |
1715 | if (page->mapping == NULL) { | |
1716 | /* | |
2ecdc82e | 1717 | * invalidate_mapping_pages got it |
1da177e4 LT |
1718 | */ |
1719 | unlock_page(page); | |
1720 | page_cache_release(page); | |
1721 | goto find_page; | |
1722 | } | |
1723 | unlock_page(page); | |
7ff81078 | 1724 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1725 | error = -EIO; |
1726 | goto readpage_error; | |
1da177e4 LT |
1727 | } |
1728 | unlock_page(page); | |
1729 | } | |
1730 | ||
1da177e4 LT |
1731 | goto page_ok; |
1732 | ||
1733 | readpage_error: | |
1734 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1da177e4 LT |
1735 | page_cache_release(page); |
1736 | goto out; | |
1737 | ||
1738 | no_cached_page: | |
1739 | /* | |
1740 | * Ok, it wasn't cached, so we need to create a new | |
1741 | * page.. | |
1742 | */ | |
eb2be189 NP |
1743 | page = page_cache_alloc_cold(mapping); |
1744 | if (!page) { | |
6e58e79d | 1745 | error = -ENOMEM; |
eb2be189 | 1746 | goto out; |
1da177e4 | 1747 | } |
6afdb859 | 1748 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 1749 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 1750 | if (error) { |
eb2be189 | 1751 | page_cache_release(page); |
6e58e79d AV |
1752 | if (error == -EEXIST) { |
1753 | error = 0; | |
1da177e4 | 1754 | goto find_page; |
6e58e79d | 1755 | } |
1da177e4 LT |
1756 | goto out; |
1757 | } | |
1da177e4 LT |
1758 | goto readpage; |
1759 | } | |
1760 | ||
1761 | out: | |
7ff81078 FW |
1762 | ra->prev_pos = prev_index; |
1763 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1764 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1765 | |
f4e6b498 | 1766 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1767 | file_accessed(filp); |
6e58e79d | 1768 | return written ? written : error; |
1da177e4 LT |
1769 | } |
1770 | ||
485bb99b | 1771 | /** |
6abd2322 | 1772 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 1773 | * @iocb: kernel I/O control block |
6abd2322 | 1774 | * @iter: destination for the data read |
485bb99b | 1775 | * |
6abd2322 | 1776 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
1777 | * that can use the page cache directly. |
1778 | */ | |
1779 | ssize_t | |
ed978a81 | 1780 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 1781 | { |
ed978a81 | 1782 | struct file *file = iocb->ki_filp; |
cb66a7a1 | 1783 | ssize_t retval = 0; |
543ade1f | 1784 | loff_t *ppos = &iocb->ki_pos; |
ed978a81 | 1785 | loff_t pos = *ppos; |
1da177e4 | 1786 | |
2ba48ce5 | 1787 | if (iocb->ki_flags & IOCB_DIRECT) { |
ed978a81 AV |
1788 | struct address_space *mapping = file->f_mapping; |
1789 | struct inode *inode = mapping->host; | |
1790 | size_t count = iov_iter_count(iter); | |
543ade1f | 1791 | loff_t size; |
1da177e4 | 1792 | |
1da177e4 LT |
1793 | if (!count) |
1794 | goto out; /* skip atime */ | |
1795 | size = i_size_read(inode); | |
9fe55eea | 1796 | retval = filemap_write_and_wait_range(mapping, pos, |
a6cbcd4a | 1797 | pos + count - 1); |
9fe55eea | 1798 | if (!retval) { |
ed978a81 | 1799 | struct iov_iter data = *iter; |
22c6186e | 1800 | retval = mapping->a_ops->direct_IO(iocb, &data, pos); |
9fe55eea | 1801 | } |
d8d3d94b | 1802 | |
9fe55eea SW |
1803 | if (retval > 0) { |
1804 | *ppos = pos + retval; | |
ed978a81 | 1805 | iov_iter_advance(iter, retval); |
9fe55eea | 1806 | } |
66f998f6 | 1807 | |
9fe55eea SW |
1808 | /* |
1809 | * Btrfs can have a short DIO read if we encounter | |
1810 | * compressed extents, so if there was an error, or if | |
1811 | * we've already read everything we wanted to, or if | |
1812 | * there was a short read because we hit EOF, go ahead | |
1813 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
1814 | * the rest of the read. Buffered reads will not work for |
1815 | * DAX files, so don't bother trying. | |
9fe55eea | 1816 | */ |
fbbbad4b MW |
1817 | if (retval < 0 || !iov_iter_count(iter) || *ppos >= size || |
1818 | IS_DAX(inode)) { | |
ed978a81 | 1819 | file_accessed(file); |
9fe55eea | 1820 | goto out; |
0e0bcae3 | 1821 | } |
1da177e4 LT |
1822 | } |
1823 | ||
ed978a81 | 1824 | retval = do_generic_file_read(file, ppos, iter, retval); |
1da177e4 LT |
1825 | out: |
1826 | return retval; | |
1827 | } | |
ed978a81 | 1828 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 1829 | |
1da177e4 | 1830 | #ifdef CONFIG_MMU |
485bb99b RD |
1831 | /** |
1832 | * page_cache_read - adds requested page to the page cache if not already there | |
1833 | * @file: file to read | |
1834 | * @offset: page index | |
1835 | * | |
1da177e4 LT |
1836 | * This adds the requested page to the page cache if it isn't already there, |
1837 | * and schedules an I/O to read in its contents from disk. | |
1838 | */ | |
920c7a5d | 1839 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1840 | { |
1841 | struct address_space *mapping = file->f_mapping; | |
99dadfdd | 1842 | struct page *page; |
994fc28c | 1843 | int ret; |
1da177e4 | 1844 | |
994fc28c ZB |
1845 | do { |
1846 | page = page_cache_alloc_cold(mapping); | |
1847 | if (!page) | |
1848 | return -ENOMEM; | |
1849 | ||
6afdb859 | 1850 | ret = add_to_page_cache_lru(page, mapping, offset, |
c62d2555 | 1851 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
994fc28c ZB |
1852 | if (ret == 0) |
1853 | ret = mapping->a_ops->readpage(file, page); | |
1854 | else if (ret == -EEXIST) | |
1855 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1856 | |
1da177e4 | 1857 | page_cache_release(page); |
1da177e4 | 1858 | |
994fc28c | 1859 | } while (ret == AOP_TRUNCATED_PAGE); |
99dadfdd | 1860 | |
994fc28c | 1861 | return ret; |
1da177e4 LT |
1862 | } |
1863 | ||
1864 | #define MMAP_LOTSAMISS (100) | |
1865 | ||
ef00e08e LT |
1866 | /* |
1867 | * Synchronous readahead happens when we don't even find | |
1868 | * a page in the page cache at all. | |
1869 | */ | |
1870 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1871 | struct file_ra_state *ra, | |
1872 | struct file *file, | |
1873 | pgoff_t offset) | |
1874 | { | |
1cac41cb | 1875 | unsigned long ra_pages; |
ef00e08e LT |
1876 | struct address_space *mapping = file->f_mapping; |
1877 | ||
1878 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1879 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 1880 | return; |
275b12bf WF |
1881 | if (!ra->ra_pages) |
1882 | return; | |
ef00e08e | 1883 | |
64363aad | 1884 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
1885 | page_cache_sync_readahead(mapping, ra, file, offset, |
1886 | ra->ra_pages); | |
ef00e08e LT |
1887 | return; |
1888 | } | |
1889 | ||
207d04ba AK |
1890 | /* Avoid banging the cache line if not needed */ |
1891 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1892 | ra->mmap_miss++; |
1893 | ||
1894 | /* | |
1895 | * Do we miss much more than hit in this file? If so, | |
1896 | * stop bothering with read-ahead. It will only hurt. | |
1897 | */ | |
1898 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1899 | return; | |
1900 | ||
d30a1100 WF |
1901 | /* |
1902 | * mmap read-around | |
1903 | */ | |
1cac41cb MB |
1904 | #if CONFIG_MMAP_READAROUND_LIMIT == 0 |
1905 | ra_pages = ra->ra_pages; | |
1906 | #else | |
1907 | if (ra->ra_pages > CONFIG_MMAP_READAROUND_LIMIT) | |
1908 | ra_pages = CONFIG_MMAP_READAROUND_LIMIT; | |
1909 | else | |
1910 | ra_pages = ra->ra_pages; | |
1911 | #endif | |
1912 | ra->start = max_t(long, 0, offset - ra_pages / 2); | |
1913 | ra->size = ra_pages; | |
1914 | ra->async_size = ra_pages / 4; | |
275b12bf | 1915 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1916 | } |
1917 | ||
1918 | /* | |
1919 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1920 | * so we want to possibly extend the readahead further.. | |
1921 | */ | |
1922 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1923 | struct file_ra_state *ra, | |
1924 | struct file *file, | |
1925 | struct page *page, | |
1926 | pgoff_t offset) | |
1927 | { | |
1928 | struct address_space *mapping = file->f_mapping; | |
1929 | ||
1930 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1931 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
1932 | return; |
1933 | if (ra->mmap_miss > 0) | |
1934 | ra->mmap_miss--; | |
1935 | if (PageReadahead(page)) | |
2fad6f5d WF |
1936 | page_cache_async_readahead(mapping, ra, file, |
1937 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1938 | } |
1939 | ||
485bb99b | 1940 | /** |
54cb8821 | 1941 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1942 | * @vma: vma in which the fault was taken |
1943 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1944 | * |
54cb8821 | 1945 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1946 | * mapped memory region to read in file data during a page fault. |
1947 | * | |
1948 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1949 | * it in the page cache, and handles the special cases reasonably without | |
1950 | * having a lot of duplicated code. | |
9a95f3cf PC |
1951 | * |
1952 | * vma->vm_mm->mmap_sem must be held on entry. | |
1953 | * | |
1954 | * If our return value has VM_FAULT_RETRY set, it's because | |
1955 | * lock_page_or_retry() returned 0. | |
1956 | * The mmap_sem has usually been released in this case. | |
1957 | * See __lock_page_or_retry() for the exception. | |
1958 | * | |
1959 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
1960 | * has not been released. | |
1961 | * | |
1962 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
1da177e4 | 1963 | */ |
d0217ac0 | 1964 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1965 | { |
1966 | int error; | |
54cb8821 | 1967 | struct file *file = vma->vm_file; |
1da177e4 LT |
1968 | struct address_space *mapping = file->f_mapping; |
1969 | struct file_ra_state *ra = &file->f_ra; | |
1970 | struct inode *inode = mapping->host; | |
ef00e08e | 1971 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 1972 | struct page *page; |
99e3e53f | 1973 | loff_t size; |
83c54070 | 1974 | int ret = 0; |
1da177e4 | 1975 | |
99e3e53f KS |
1976 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
1977 | if (offset >= size >> PAGE_CACHE_SHIFT) | |
5307cc1a | 1978 | return VM_FAULT_SIGBUS; |
1da177e4 | 1979 | |
1da177e4 | 1980 | /* |
49426420 | 1981 | * Do we have something in the page cache already? |
1da177e4 | 1982 | */ |
ef00e08e | 1983 | page = find_get_page(mapping, offset); |
45cac65b | 1984 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 1985 | /* |
ef00e08e LT |
1986 | * We found the page, so try async readahead before |
1987 | * waiting for the lock. | |
1da177e4 | 1988 | */ |
ef00e08e | 1989 | do_async_mmap_readahead(vma, ra, file, page, offset); |
45cac65b | 1990 | } else if (!page) { |
ef00e08e LT |
1991 | /* No page in the page cache at all */ |
1992 | do_sync_mmap_readahead(vma, ra, file, offset); | |
1993 | count_vm_event(PGMAJFAULT); | |
456f998e | 1994 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
1995 | ret = VM_FAULT_MAJOR; |
1996 | retry_find: | |
b522c94d | 1997 | page = find_get_page(mapping, offset); |
1da177e4 LT |
1998 | if (!page) |
1999 | goto no_cached_page; | |
2000 | } | |
2001 | ||
d88c0922 ML |
2002 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
2003 | page_cache_release(page); | |
d065bd81 | 2004 | return ret | VM_FAULT_RETRY; |
d88c0922 | 2005 | } |
b522c94d ML |
2006 | |
2007 | /* Did it get truncated? */ | |
2008 | if (unlikely(page->mapping != mapping)) { | |
2009 | unlock_page(page); | |
2010 | put_page(page); | |
2011 | goto retry_find; | |
2012 | } | |
309381fe | 2013 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 2014 | |
1da177e4 | 2015 | /* |
d00806b1 NP |
2016 | * We have a locked page in the page cache, now we need to check |
2017 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2018 | */ |
d00806b1 | 2019 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2020 | goto page_not_uptodate; |
2021 | ||
ef00e08e LT |
2022 | /* |
2023 | * Found the page and have a reference on it. | |
2024 | * We must recheck i_size under page lock. | |
2025 | */ | |
99e3e53f KS |
2026 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
2027 | if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) { | |
d00806b1 | 2028 | unlock_page(page); |
745ad48e | 2029 | page_cache_release(page); |
5307cc1a | 2030 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2031 | } |
2032 | ||
d0217ac0 | 2033 | vmf->page = page; |
83c54070 | 2034 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2035 | |
1da177e4 LT |
2036 | no_cached_page: |
2037 | /* | |
2038 | * We're only likely to ever get here if MADV_RANDOM is in | |
2039 | * effect. | |
2040 | */ | |
ef00e08e | 2041 | error = page_cache_read(file, offset); |
1da177e4 LT |
2042 | |
2043 | /* | |
2044 | * The page we want has now been added to the page cache. | |
2045 | * In the unlikely event that someone removed it in the | |
2046 | * meantime, we'll just come back here and read it again. | |
2047 | */ | |
2048 | if (error >= 0) | |
2049 | goto retry_find; | |
2050 | ||
2051 | /* | |
2052 | * An error return from page_cache_read can result if the | |
2053 | * system is low on memory, or a problem occurs while trying | |
2054 | * to schedule I/O. | |
2055 | */ | |
2056 | if (error == -ENOMEM) | |
d0217ac0 NP |
2057 | return VM_FAULT_OOM; |
2058 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
2059 | |
2060 | page_not_uptodate: | |
1da177e4 LT |
2061 | /* |
2062 | * Umm, take care of errors if the page isn't up-to-date. | |
2063 | * Try to re-read it _once_. We do this synchronously, | |
2064 | * because there really aren't any performance issues here | |
2065 | * and we need to check for errors. | |
2066 | */ | |
1da177e4 | 2067 | ClearPageError(page); |
994fc28c | 2068 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2069 | if (!error) { |
2070 | wait_on_page_locked(page); | |
2071 | if (!PageUptodate(page)) | |
2072 | error = -EIO; | |
2073 | } | |
d00806b1 NP |
2074 | page_cache_release(page); |
2075 | ||
2076 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2077 | goto retry_find; |
1da177e4 | 2078 | |
d00806b1 | 2079 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 2080 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 2081 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
2082 | } |
2083 | EXPORT_SYMBOL(filemap_fault); | |
2084 | ||
f1820361 KS |
2085 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
2086 | { | |
2087 | struct radix_tree_iter iter; | |
2088 | void **slot; | |
2089 | struct file *file = vma->vm_file; | |
2090 | struct address_space *mapping = file->f_mapping; | |
2091 | loff_t size; | |
2092 | struct page *page; | |
2093 | unsigned long address = (unsigned long) vmf->virtual_address; | |
2094 | unsigned long addr; | |
2095 | pte_t *pte; | |
2096 | ||
2097 | rcu_read_lock(); | |
2098 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) { | |
2099 | if (iter.index > vmf->max_pgoff) | |
2100 | break; | |
2101 | repeat: | |
2102 | page = radix_tree_deref_slot(slot); | |
2103 | if (unlikely(!page)) | |
2104 | goto next; | |
2105 | if (radix_tree_exception(page)) { | |
2106 | if (radix_tree_deref_retry(page)) | |
2107 | break; | |
2108 | else | |
2109 | goto next; | |
2110 | } | |
2111 | ||
2112 | if (!page_cache_get_speculative(page)) | |
2113 | goto repeat; | |
2114 | ||
2115 | /* Has the page moved? */ | |
2116 | if (unlikely(page != *slot)) { | |
2117 | page_cache_release(page); | |
2118 | goto repeat; | |
2119 | } | |
2120 | ||
2121 | if (!PageUptodate(page) || | |
2122 | PageReadahead(page) || | |
2123 | PageHWPoison(page)) | |
2124 | goto skip; | |
2125 | if (!trylock_page(page)) | |
2126 | goto skip; | |
2127 | ||
2128 | if (page->mapping != mapping || !PageUptodate(page)) | |
2129 | goto unlock; | |
2130 | ||
99e3e53f KS |
2131 | size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE); |
2132 | if (page->index >= size >> PAGE_CACHE_SHIFT) | |
f1820361 KS |
2133 | goto unlock; |
2134 | ||
2135 | pte = vmf->pte + page->index - vmf->pgoff; | |
2136 | if (!pte_none(*pte)) | |
2137 | goto unlock; | |
2138 | ||
2139 | if (file->f_ra.mmap_miss > 0) | |
2140 | file->f_ra.mmap_miss--; | |
2141 | addr = address + (page->index - vmf->pgoff) * PAGE_SIZE; | |
2142 | do_set_pte(vma, addr, page, pte, false, false); | |
2143 | unlock_page(page); | |
2144 | goto next; | |
2145 | unlock: | |
2146 | unlock_page(page); | |
2147 | skip: | |
2148 | page_cache_release(page); | |
2149 | next: | |
2150 | if (iter.index == vmf->max_pgoff) | |
2151 | break; | |
2152 | } | |
2153 | rcu_read_unlock(); | |
2154 | } | |
2155 | EXPORT_SYMBOL(filemap_map_pages); | |
2156 | ||
4fcf1c62 JK |
2157 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2158 | { | |
2159 | struct page *page = vmf->page; | |
496ad9aa | 2160 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
2161 | int ret = VM_FAULT_LOCKED; |
2162 | ||
14da9200 | 2163 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
2164 | file_update_time(vma->vm_file); |
2165 | lock_page(page); | |
2166 | if (page->mapping != inode->i_mapping) { | |
2167 | unlock_page(page); | |
2168 | ret = VM_FAULT_NOPAGE; | |
2169 | goto out; | |
2170 | } | |
14da9200 JK |
2171 | /* |
2172 | * We mark the page dirty already here so that when freeze is in | |
2173 | * progress, we are guaranteed that writeback during freezing will | |
2174 | * see the dirty page and writeprotect it again. | |
2175 | */ | |
2176 | set_page_dirty(page); | |
1d1d1a76 | 2177 | wait_for_stable_page(page); |
4fcf1c62 | 2178 | out: |
14da9200 | 2179 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2180 | return ret; |
2181 | } | |
2182 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2183 | ||
f0f37e2f | 2184 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2185 | .fault = filemap_fault, |
f1820361 | 2186 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2187 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2188 | }; |
2189 | ||
2190 | /* This is used for a general mmap of a disk file */ | |
2191 | ||
2192 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2193 | { | |
2194 | struct address_space *mapping = file->f_mapping; | |
2195 | ||
2196 | if (!mapping->a_ops->readpage) | |
2197 | return -ENOEXEC; | |
2198 | file_accessed(file); | |
2199 | vma->vm_ops = &generic_file_vm_ops; | |
2200 | return 0; | |
2201 | } | |
1da177e4 LT |
2202 | |
2203 | /* | |
2204 | * This is for filesystems which do not implement ->writepage. | |
2205 | */ | |
2206 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2207 | { | |
2208 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2209 | return -EINVAL; | |
2210 | return generic_file_mmap(file, vma); | |
2211 | } | |
2212 | #else | |
2213 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2214 | { | |
2215 | return -ENOSYS; | |
2216 | } | |
2217 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2218 | { | |
2219 | return -ENOSYS; | |
2220 | } | |
2221 | #endif /* CONFIG_MMU */ | |
2222 | ||
2223 | EXPORT_SYMBOL(generic_file_mmap); | |
2224 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2225 | ||
67f9fd91 SL |
2226 | static struct page *wait_on_page_read(struct page *page) |
2227 | { | |
2228 | if (!IS_ERR(page)) { | |
2229 | wait_on_page_locked(page); | |
2230 | if (!PageUptodate(page)) { | |
2231 | page_cache_release(page); | |
2232 | page = ERR_PTR(-EIO); | |
2233 | } | |
2234 | } | |
2235 | return page; | |
2236 | } | |
2237 | ||
6fe6900e | 2238 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 2239 | pgoff_t index, |
5e5358e7 | 2240 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2241 | void *data, |
2242 | gfp_t gfp) | |
1da177e4 | 2243 | { |
eb2be189 | 2244 | struct page *page; |
1da177e4 LT |
2245 | int err; |
2246 | repeat: | |
2247 | page = find_get_page(mapping, index); | |
2248 | if (!page) { | |
0531b2aa | 2249 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
2250 | if (!page) |
2251 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2252 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 NP |
2253 | if (unlikely(err)) { |
2254 | page_cache_release(page); | |
2255 | if (err == -EEXIST) | |
2256 | goto repeat; | |
1da177e4 | 2257 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2258 | return ERR_PTR(err); |
2259 | } | |
1da177e4 LT |
2260 | err = filler(data, page); |
2261 | if (err < 0) { | |
2262 | page_cache_release(page); | |
2263 | page = ERR_PTR(err); | |
67f9fd91 SL |
2264 | } else { |
2265 | page = wait_on_page_read(page); | |
1da177e4 LT |
2266 | } |
2267 | } | |
1da177e4 LT |
2268 | return page; |
2269 | } | |
2270 | ||
0531b2aa | 2271 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2272 | pgoff_t index, |
5e5358e7 | 2273 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2274 | void *data, |
2275 | gfp_t gfp) | |
2276 | ||
1da177e4 LT |
2277 | { |
2278 | struct page *page; | |
2279 | int err; | |
2280 | ||
2281 | retry: | |
0531b2aa | 2282 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 2283 | if (IS_ERR(page)) |
c855ff37 | 2284 | return page; |
1da177e4 LT |
2285 | if (PageUptodate(page)) |
2286 | goto out; | |
2287 | ||
2288 | lock_page(page); | |
2289 | if (!page->mapping) { | |
2290 | unlock_page(page); | |
2291 | page_cache_release(page); | |
2292 | goto retry; | |
2293 | } | |
2294 | if (PageUptodate(page)) { | |
2295 | unlock_page(page); | |
2296 | goto out; | |
2297 | } | |
2298 | err = filler(data, page); | |
2299 | if (err < 0) { | |
2300 | page_cache_release(page); | |
c855ff37 | 2301 | return ERR_PTR(err); |
67f9fd91 SL |
2302 | } else { |
2303 | page = wait_on_page_read(page); | |
2304 | if (IS_ERR(page)) | |
2305 | return page; | |
1da177e4 | 2306 | } |
c855ff37 | 2307 | out: |
6fe6900e NP |
2308 | mark_page_accessed(page); |
2309 | return page; | |
2310 | } | |
0531b2aa LT |
2311 | |
2312 | /** | |
67f9fd91 | 2313 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2314 | * @mapping: the page's address_space |
2315 | * @index: the page index | |
2316 | * @filler: function to perform the read | |
5e5358e7 | 2317 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2318 | * |
0531b2aa | 2319 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2320 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2321 | * |
2322 | * If the page does not get brought uptodate, return -EIO. | |
2323 | */ | |
67f9fd91 | 2324 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2325 | pgoff_t index, |
5e5358e7 | 2326 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2327 | void *data) |
2328 | { | |
2329 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2330 | } | |
67f9fd91 | 2331 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2332 | |
2333 | /** | |
2334 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2335 | * @mapping: the page's address_space | |
2336 | * @index: the page index | |
2337 | * @gfp: the page allocator flags to use if allocating | |
2338 | * | |
2339 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2340 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2341 | * |
2342 | * If the page does not get brought uptodate, return -EIO. | |
2343 | */ | |
2344 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2345 | pgoff_t index, | |
2346 | gfp_t gfp) | |
2347 | { | |
2348 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2349 | ||
67f9fd91 | 2350 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2351 | } |
2352 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2353 | ||
1da177e4 LT |
2354 | /* |
2355 | * Performs necessary checks before doing a write | |
2356 | * | |
485bb99b | 2357 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2358 | * Returns appropriate error code that caller should return or |
2359 | * zero in case that write should be allowed. | |
2360 | */ | |
3309dd04 | 2361 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2362 | { |
3309dd04 | 2363 | struct file *file = iocb->ki_filp; |
1da177e4 | 2364 | struct inode *inode = file->f_mapping->host; |
59e99e5b | 2365 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
3309dd04 | 2366 | loff_t pos; |
1da177e4 | 2367 | |
3309dd04 AV |
2368 | if (!iov_iter_count(from)) |
2369 | return 0; | |
1da177e4 | 2370 | |
0fa6b005 | 2371 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2372 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2373 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2374 | |
3309dd04 | 2375 | pos = iocb->ki_pos; |
1da177e4 | 2376 | |
0fa6b005 | 2377 | if (limit != RLIM_INFINITY) { |
3309dd04 | 2378 | if (iocb->ki_pos >= limit) { |
0fa6b005 AV |
2379 | send_sig(SIGXFSZ, current, 0); |
2380 | return -EFBIG; | |
1da177e4 | 2381 | } |
3309dd04 | 2382 | iov_iter_truncate(from, limit - (unsigned long)pos); |
1da177e4 LT |
2383 | } |
2384 | ||
2385 | /* | |
2386 | * LFS rule | |
2387 | */ | |
3309dd04 | 2388 | if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS && |
1da177e4 | 2389 | !(file->f_flags & O_LARGEFILE))) { |
3309dd04 | 2390 | if (pos >= MAX_NON_LFS) |
1da177e4 | 2391 | return -EFBIG; |
3309dd04 | 2392 | iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos); |
1da177e4 LT |
2393 | } |
2394 | ||
2395 | /* | |
2396 | * Are we about to exceed the fs block limit ? | |
2397 | * | |
2398 | * If we have written data it becomes a short write. If we have | |
2399 | * exceeded without writing data we send a signal and return EFBIG. | |
2400 | * Linus frestrict idea will clean these up nicely.. | |
2401 | */ | |
3309dd04 AV |
2402 | if (unlikely(pos >= inode->i_sb->s_maxbytes)) |
2403 | return -EFBIG; | |
1da177e4 | 2404 | |
3309dd04 AV |
2405 | iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos); |
2406 | return iov_iter_count(from); | |
1da177e4 LT |
2407 | } |
2408 | EXPORT_SYMBOL(generic_write_checks); | |
2409 | ||
afddba49 NP |
2410 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2411 | loff_t pos, unsigned len, unsigned flags, | |
2412 | struct page **pagep, void **fsdata) | |
2413 | { | |
2414 | const struct address_space_operations *aops = mapping->a_ops; | |
2415 | ||
4e02ed4b | 2416 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2417 | pagep, fsdata); |
afddba49 NP |
2418 | } |
2419 | EXPORT_SYMBOL(pagecache_write_begin); | |
2420 | ||
2421 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2422 | loff_t pos, unsigned len, unsigned copied, | |
2423 | struct page *page, void *fsdata) | |
2424 | { | |
2425 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2426 | |
4e02ed4b | 2427 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2428 | } |
2429 | EXPORT_SYMBOL(pagecache_write_end); | |
2430 | ||
1da177e4 | 2431 | ssize_t |
0c949334 | 2432 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos) |
1da177e4 LT |
2433 | { |
2434 | struct file *file = iocb->ki_filp; | |
2435 | struct address_space *mapping = file->f_mapping; | |
2436 | struct inode *inode = mapping->host; | |
2437 | ssize_t written; | |
a969e903 CH |
2438 | size_t write_len; |
2439 | pgoff_t end; | |
26978b8b | 2440 | struct iov_iter data; |
1da177e4 | 2441 | |
0c949334 | 2442 | write_len = iov_iter_count(from); |
a969e903 | 2443 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; |
a969e903 | 2444 | |
48b47c56 | 2445 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2446 | if (written) |
2447 | goto out; | |
2448 | ||
2449 | /* | |
2450 | * After a write we want buffered reads to be sure to go to disk to get | |
2451 | * the new data. We invalidate clean cached page from the region we're | |
2452 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2453 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2454 | */ |
2455 | if (mapping->nrpages) { | |
2456 | written = invalidate_inode_pages2_range(mapping, | |
2457 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2458 | /* |
2459 | * If a page can not be invalidated, return 0 to fall back | |
2460 | * to buffered write. | |
2461 | */ | |
2462 | if (written) { | |
2463 | if (written == -EBUSY) | |
2464 | return 0; | |
a969e903 | 2465 | goto out; |
6ccfa806 | 2466 | } |
a969e903 CH |
2467 | } |
2468 | ||
26978b8b | 2469 | data = *from; |
22c6186e | 2470 | written = mapping->a_ops->direct_IO(iocb, &data, pos); |
a969e903 CH |
2471 | |
2472 | /* | |
2473 | * Finally, try again to invalidate clean pages which might have been | |
2474 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2475 | * if the source of the write was an mmap'ed region of the file | |
2476 | * we're writing. Either one is a pretty crazy thing to do, | |
2477 | * so we don't support it 100%. If this invalidation | |
2478 | * fails, tough, the write still worked... | |
2479 | */ | |
2480 | if (mapping->nrpages) { | |
2481 | invalidate_inode_pages2_range(mapping, | |
2482 | pos >> PAGE_CACHE_SHIFT, end); | |
2483 | } | |
2484 | ||
1da177e4 | 2485 | if (written > 0) { |
0116651c | 2486 | pos += written; |
f8579f86 | 2487 | iov_iter_advance(from, written); |
0116651c NK |
2488 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
2489 | i_size_write(inode, pos); | |
1da177e4 LT |
2490 | mark_inode_dirty(inode); |
2491 | } | |
5cb6c6c7 | 2492 | iocb->ki_pos = pos; |
1da177e4 | 2493 | } |
a969e903 | 2494 | out: |
1da177e4 LT |
2495 | return written; |
2496 | } | |
2497 | EXPORT_SYMBOL(generic_file_direct_write); | |
2498 | ||
eb2be189 NP |
2499 | /* |
2500 | * Find or create a page at the given pagecache position. Return the locked | |
2501 | * page. This function is specifically for buffered writes. | |
2502 | */ | |
54566b2c NP |
2503 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2504 | pgoff_t index, unsigned flags) | |
eb2be189 | 2505 | { |
eb2be189 | 2506 | struct page *page; |
2457aec6 | 2507 | int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT; |
0faa70cb | 2508 | |
54566b2c | 2509 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
2510 | fgp_flags |= FGP_NOFS; |
2511 | ||
2512 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 2513 | mapping_gfp_mask(mapping)); |
c585a267 | 2514 | if (page) |
2457aec6 | 2515 | wait_for_stable_page(page); |
eb2be189 | 2516 | |
eb2be189 NP |
2517 | return page; |
2518 | } | |
54566b2c | 2519 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2520 | |
3b93f911 | 2521 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
2522 | struct iov_iter *i, loff_t pos) |
2523 | { | |
2524 | struct address_space *mapping = file->f_mapping; | |
2525 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2526 | long status = 0; | |
2527 | ssize_t written = 0; | |
674b892e NP |
2528 | unsigned int flags = 0; |
2529 | ||
2530 | /* | |
2531 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2532 | */ | |
777eda2c | 2533 | if (!iter_is_iovec(i)) |
674b892e | 2534 | flags |= AOP_FLAG_UNINTERRUPTIBLE; |
afddba49 NP |
2535 | |
2536 | do { | |
2537 | struct page *page; | |
afddba49 NP |
2538 | unsigned long offset; /* Offset into pagecache page */ |
2539 | unsigned long bytes; /* Bytes to write to page */ | |
2540 | size_t copied; /* Bytes copied from user */ | |
2541 | void *fsdata; | |
2542 | ||
2543 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2544 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2545 | iov_iter_count(i)); | |
2546 | ||
2547 | again: | |
00a3d660 LT |
2548 | /* |
2549 | * Bring in the user page that we will copy from _first_. | |
2550 | * Otherwise there's a nasty deadlock on copying from the | |
2551 | * same page as we're writing to, without it being marked | |
2552 | * up-to-date. | |
2553 | * | |
2554 | * Not only is this an optimisation, but it is also required | |
2555 | * to check that the address is actually valid, when atomic | |
2556 | * usercopies are used, below. | |
2557 | */ | |
2558 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2559 | status = -EFAULT; | |
2560 | break; | |
2561 | } | |
2562 | ||
296291cd JK |
2563 | if (fatal_signal_pending(current)) { |
2564 | status = -EINTR; | |
2565 | break; | |
2566 | } | |
2567 | ||
674b892e | 2568 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 2569 | &page, &fsdata); |
2457aec6 | 2570 | if (unlikely(status < 0)) |
afddba49 NP |
2571 | break; |
2572 | ||
931e80e4 | 2573 | if (mapping_writably_mapped(mapping)) |
2574 | flush_dcache_page(page); | |
00a3d660 | 2575 | |
afddba49 | 2576 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
2577 | flush_dcache_page(page); |
2578 | ||
2579 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
2580 | page, fsdata); | |
2581 | if (unlikely(status < 0)) | |
2582 | break; | |
2583 | copied = status; | |
2584 | ||
2585 | cond_resched(); | |
2586 | ||
124d3b70 | 2587 | iov_iter_advance(i, copied); |
afddba49 NP |
2588 | if (unlikely(copied == 0)) { |
2589 | /* | |
2590 | * If we were unable to copy any data at all, we must | |
2591 | * fall back to a single segment length write. | |
2592 | * | |
2593 | * If we didn't fallback here, we could livelock | |
2594 | * because not all segments in the iov can be copied at | |
2595 | * once without a pagefault. | |
2596 | */ | |
2597 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2598 | iov_iter_single_seg_count(i)); | |
2599 | goto again; | |
2600 | } | |
afddba49 NP |
2601 | pos += copied; |
2602 | written += copied; | |
2603 | ||
2604 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
2605 | } while (iov_iter_count(i)); |
2606 | ||
2607 | return written ? written : status; | |
2608 | } | |
3b93f911 | 2609 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 2610 | |
e4dd9de3 | 2611 | /** |
8174202b | 2612 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 2613 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 2614 | * @from: iov_iter with data to write |
e4dd9de3 JK |
2615 | * |
2616 | * This function does all the work needed for actually writing data to a | |
2617 | * file. It does all basic checks, removes SUID from the file, updates | |
2618 | * modification times and calls proper subroutines depending on whether we | |
2619 | * do direct IO or a standard buffered write. | |
2620 | * | |
2621 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2622 | * object which does not need locking at all. | |
2623 | * | |
2624 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2625 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2626 | * avoid syncing under i_mutex. | |
2627 | */ | |
8174202b | 2628 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2629 | { |
2630 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2631 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 2632 | struct inode *inode = mapping->host; |
3b93f911 | 2633 | ssize_t written = 0; |
1da177e4 | 2634 | ssize_t err; |
3b93f911 | 2635 | ssize_t status; |
1da177e4 | 2636 | |
1da177e4 | 2637 | /* We can write back this queue in page reclaim */ |
de1414a6 | 2638 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 2639 | err = file_remove_privs(file); |
1da177e4 LT |
2640 | if (err) |
2641 | goto out; | |
2642 | ||
c3b2da31 JB |
2643 | err = file_update_time(file); |
2644 | if (err) | |
2645 | goto out; | |
1da177e4 | 2646 | |
2ba48ce5 | 2647 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 2648 | loff_t pos, endbyte; |
fb5527e6 | 2649 | |
0b8def9d | 2650 | written = generic_file_direct_write(iocb, from, iocb->ki_pos); |
1da177e4 | 2651 | /* |
fbbbad4b MW |
2652 | * If the write stopped short of completing, fall back to |
2653 | * buffered writes. Some filesystems do this for writes to | |
2654 | * holes, for example. For DAX files, a buffered write will | |
2655 | * not succeed (even if it did, DAX does not handle dirty | |
2656 | * page-cache pages correctly). | |
1da177e4 | 2657 | */ |
0b8def9d | 2658 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
2659 | goto out; |
2660 | ||
0b8def9d | 2661 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 2662 | /* |
3b93f911 | 2663 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
2664 | * then we want to return the number of bytes which were |
2665 | * direct-written, or the error code if that was zero. Note | |
2666 | * that this differs from normal direct-io semantics, which | |
2667 | * will return -EFOO even if some bytes were written. | |
2668 | */ | |
60bb4529 | 2669 | if (unlikely(status < 0)) { |
3b93f911 | 2670 | err = status; |
fb5527e6 JM |
2671 | goto out; |
2672 | } | |
fb5527e6 JM |
2673 | /* |
2674 | * We need to ensure that the page cache pages are written to | |
2675 | * disk and invalidated to preserve the expected O_DIRECT | |
2676 | * semantics. | |
2677 | */ | |
3b93f911 | 2678 | endbyte = pos + status - 1; |
0b8def9d | 2679 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 2680 | if (err == 0) { |
0b8def9d | 2681 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 2682 | written += status; |
fb5527e6 JM |
2683 | invalidate_mapping_pages(mapping, |
2684 | pos >> PAGE_CACHE_SHIFT, | |
2685 | endbyte >> PAGE_CACHE_SHIFT); | |
2686 | } else { | |
2687 | /* | |
2688 | * We don't know how much we wrote, so just return | |
2689 | * the number of bytes which were direct-written | |
2690 | */ | |
2691 | } | |
2692 | } else { | |
0b8def9d AV |
2693 | written = generic_perform_write(file, from, iocb->ki_pos); |
2694 | if (likely(written > 0)) | |
2695 | iocb->ki_pos += written; | |
fb5527e6 | 2696 | } |
1da177e4 LT |
2697 | out: |
2698 | current->backing_dev_info = NULL; | |
2699 | return written ? written : err; | |
2700 | } | |
8174202b | 2701 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 2702 | |
e4dd9de3 | 2703 | /** |
8174202b | 2704 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 2705 | * @iocb: IO state structure |
8174202b | 2706 | * @from: iov_iter with data to write |
e4dd9de3 | 2707 | * |
8174202b | 2708 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
2709 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
2710 | * and acquires i_mutex as needed. | |
2711 | */ | |
8174202b | 2712 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2713 | { |
2714 | struct file *file = iocb->ki_filp; | |
148f948b | 2715 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2716 | ssize_t ret; |
1da177e4 | 2717 | |
1b1dcc1b | 2718 | mutex_lock(&inode->i_mutex); |
3309dd04 AV |
2719 | ret = generic_write_checks(iocb, from); |
2720 | if (ret > 0) | |
5f380c7f | 2721 | ret = __generic_file_write_iter(iocb, from); |
1b1dcc1b | 2722 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 2723 | |
02afc27f | 2724 | if (ret > 0) { |
1da177e4 LT |
2725 | ssize_t err; |
2726 | ||
d311d79d AV |
2727 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
2728 | if (err < 0) | |
1da177e4 LT |
2729 | ret = err; |
2730 | } | |
2731 | return ret; | |
2732 | } | |
8174202b | 2733 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 2734 | |
cf9a2ae8 DH |
2735 | /** |
2736 | * try_to_release_page() - release old fs-specific metadata on a page | |
2737 | * | |
2738 | * @page: the page which the kernel is trying to free | |
2739 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2740 | * | |
2741 | * The address_space is to try to release any data against the page | |
2742 | * (presumably at page->private). If the release was successful, return `1'. | |
2743 | * Otherwise return zero. | |
2744 | * | |
266cf658 DH |
2745 | * This may also be called if PG_fscache is set on a page, indicating that the |
2746 | * page is known to the local caching routines. | |
2747 | * | |
cf9a2ae8 | 2748 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 2749 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 2750 | * |
cf9a2ae8 DH |
2751 | */ |
2752 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2753 | { | |
2754 | struct address_space * const mapping = page->mapping; | |
2755 | ||
2756 | BUG_ON(!PageLocked(page)); | |
2757 | if (PageWriteback(page)) | |
2758 | return 0; | |
2759 | ||
2760 | if (mapping && mapping->a_ops->releasepage) | |
2761 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2762 | return try_to_free_buffers(page); | |
2763 | } | |
2764 | ||
2765 | EXPORT_SYMBOL(try_to_release_page); |