]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/filemap.c | |
3 | * | |
4 | * Copyright (C) 1994-1999 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * This file handles the generic file mmap semantics used by | |
9 | * most "normal" filesystems (but you don't /have/ to use this: | |
10 | * the NFS filesystem used to do this differently, for example) | |
11 | */ | |
b95f1b31 | 12 | #include <linux/export.h> |
1da177e4 LT |
13 | #include <linux/compiler.h> |
14 | #include <linux/fs.h> | |
c22ce143 | 15 | #include <linux/uaccess.h> |
1da177e4 | 16 | #include <linux/aio.h> |
c59ede7b | 17 | #include <linux/capability.h> |
1da177e4 | 18 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 19 | #include <linux/gfp.h> |
1da177e4 LT |
20 | #include <linux/mm.h> |
21 | #include <linux/swap.h> | |
22 | #include <linux/mman.h> | |
23 | #include <linux/pagemap.h> | |
24 | #include <linux/file.h> | |
25 | #include <linux/uio.h> | |
26 | #include <linux/hash.h> | |
27 | #include <linux/writeback.h> | |
53253383 | 28 | #include <linux/backing-dev.h> |
1da177e4 LT |
29 | #include <linux/pagevec.h> |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/security.h> | |
44110fe3 | 32 | #include <linux/cpuset.h> |
2f718ffc | 33 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
00501b53 | 34 | #include <linux/hugetlb.h> |
8a9f3ccd | 35 | #include <linux/memcontrol.h> |
c515e1fd | 36 | #include <linux/cleancache.h> |
f1820361 | 37 | #include <linux/rmap.h> |
0f8053a5 NP |
38 | #include "internal.h" |
39 | ||
fe0bfaaf RJ |
40 | #define CREATE_TRACE_POINTS |
41 | #include <trace/events/filemap.h> | |
42 | ||
1da177e4 | 43 | /* |
1da177e4 LT |
44 | * FIXME: remove all knowledge of the buffer layer from the core VM |
45 | */ | |
148f948b | 46 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 47 | |
1da177e4 LT |
48 | #include <asm/mman.h> |
49 | ||
50 | /* | |
51 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
52 | * though. | |
53 | * | |
54 | * Shared mappings now work. 15.8.1995 Bruno. | |
55 | * | |
56 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
57 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
58 | * | |
59 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
60 | */ | |
61 | ||
62 | /* | |
63 | * Lock ordering: | |
64 | * | |
3d48ae45 | 65 | * ->i_mmap_mutex (truncate_pagecache) |
1da177e4 | 66 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 HD |
67 | * ->swap_lock (exclusive_swap_page, others) |
68 | * ->mapping->tree_lock | |
1da177e4 | 69 | * |
1b1dcc1b | 70 | * ->i_mutex |
3d48ae45 | 71 | * ->i_mmap_mutex (truncate->unmap_mapping_range) |
1da177e4 LT |
72 | * |
73 | * ->mmap_sem | |
3d48ae45 | 74 | * ->i_mmap_mutex |
b8072f09 | 75 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
76 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
77 | * | |
78 | * ->mmap_sem | |
79 | * ->lock_page (access_process_vm) | |
80 | * | |
ccad2365 | 81 | * ->i_mutex (generic_perform_write) |
82591e6e | 82 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
1da177e4 | 83 | * |
f758eeab | 84 | * bdi->wb.list_lock |
a66979ab | 85 | * sb_lock (fs/fs-writeback.c) |
1da177e4 LT |
86 | * ->mapping->tree_lock (__sync_single_inode) |
87 | * | |
3d48ae45 | 88 | * ->i_mmap_mutex |
1da177e4 LT |
89 | * ->anon_vma.lock (vma_adjust) |
90 | * | |
91 | * ->anon_vma.lock | |
b8072f09 | 92 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 93 | * |
b8072f09 | 94 | * ->page_table_lock or pte_lock |
5d337b91 | 95 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
96 | * ->private_lock (try_to_unmap_one) |
97 | * ->tree_lock (try_to_unmap_one) | |
98 | * ->zone.lru_lock (follow_page->mark_page_accessed) | |
053837fc | 99 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
1da177e4 LT |
100 | * ->private_lock (page_remove_rmap->set_page_dirty) |
101 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
f758eeab | 102 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 103 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
f758eeab | 104 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 105 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
106 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
107 | * | |
9a3c531d AK |
108 | * ->i_mmap_mutex |
109 | * ->tasklist_lock (memory_failure, collect_procs_ao) | |
1da177e4 LT |
110 | */ |
111 | ||
91b0abe3 JW |
112 | static void page_cache_tree_delete(struct address_space *mapping, |
113 | struct page *page, void *shadow) | |
114 | { | |
449dd698 JW |
115 | struct radix_tree_node *node; |
116 | unsigned long index; | |
117 | unsigned int offset; | |
118 | unsigned int tag; | |
119 | void **slot; | |
91b0abe3 | 120 | |
449dd698 JW |
121 | VM_BUG_ON(!PageLocked(page)); |
122 | ||
123 | __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot); | |
124 | ||
125 | if (shadow) { | |
91b0abe3 JW |
126 | mapping->nrshadows++; |
127 | /* | |
128 | * Make sure the nrshadows update is committed before | |
129 | * the nrpages update so that final truncate racing | |
130 | * with reclaim does not see both counters 0 at the | |
131 | * same time and miss a shadow entry. | |
132 | */ | |
133 | smp_wmb(); | |
449dd698 | 134 | } |
91b0abe3 | 135 | mapping->nrpages--; |
449dd698 JW |
136 | |
137 | if (!node) { | |
138 | /* Clear direct pointer tags in root node */ | |
139 | mapping->page_tree.gfp_mask &= __GFP_BITS_MASK; | |
140 | radix_tree_replace_slot(slot, shadow); | |
141 | return; | |
142 | } | |
143 | ||
144 | /* Clear tree tags for the removed page */ | |
145 | index = page->index; | |
146 | offset = index & RADIX_TREE_MAP_MASK; | |
147 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { | |
148 | if (test_bit(offset, node->tags[tag])) | |
149 | radix_tree_tag_clear(&mapping->page_tree, index, tag); | |
150 | } | |
151 | ||
152 | /* Delete page, swap shadow entry */ | |
153 | radix_tree_replace_slot(slot, shadow); | |
154 | workingset_node_pages_dec(node); | |
155 | if (shadow) | |
156 | workingset_node_shadows_inc(node); | |
157 | else | |
158 | if (__radix_tree_delete_node(&mapping->page_tree, node)) | |
159 | return; | |
160 | ||
161 | /* | |
162 | * Track node that only contains shadow entries. | |
163 | * | |
164 | * Avoid acquiring the list_lru lock if already tracked. The | |
165 | * list_empty() test is safe as node->private_list is | |
166 | * protected by mapping->tree_lock. | |
167 | */ | |
168 | if (!workingset_node_pages(node) && | |
169 | list_empty(&node->private_list)) { | |
170 | node->private_data = mapping; | |
171 | list_lru_add(&workingset_shadow_nodes, &node->private_list); | |
172 | } | |
91b0abe3 JW |
173 | } |
174 | ||
1da177e4 | 175 | /* |
e64a782f | 176 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 177 | * sure the page is locked and that nobody else uses it - or that usage |
19fd6231 | 178 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 179 | */ |
91b0abe3 | 180 | void __delete_from_page_cache(struct page *page, void *shadow) |
1da177e4 LT |
181 | { |
182 | struct address_space *mapping = page->mapping; | |
183 | ||
fe0bfaaf | 184 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
185 | /* |
186 | * if we're uptodate, flush out into the cleancache, otherwise | |
187 | * invalidate any existing cleancache entries. We can't leave | |
188 | * stale data around in the cleancache once our page is gone | |
189 | */ | |
190 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
191 | cleancache_put_page(page); | |
192 | else | |
3167760f | 193 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 194 | |
91b0abe3 JW |
195 | page_cache_tree_delete(mapping, page, shadow); |
196 | ||
1da177e4 | 197 | page->mapping = NULL; |
b85e0eff | 198 | /* Leave page->index set: truncation lookup relies upon it */ |
91b0abe3 | 199 | |
347ce434 | 200 | __dec_zone_page_state(page, NR_FILE_PAGES); |
4b02108a KM |
201 | if (PageSwapBacked(page)) |
202 | __dec_zone_page_state(page, NR_SHMEM); | |
45426812 | 203 | BUG_ON(page_mapped(page)); |
3a692790 LT |
204 | |
205 | /* | |
206 | * Some filesystems seem to re-dirty the page even after | |
207 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
208 | * | |
209 | * Fix it up by doing a final dirty accounting check after | |
210 | * having removed the page entirely. | |
211 | */ | |
212 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { | |
213 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
214 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
215 | } | |
1da177e4 LT |
216 | } |
217 | ||
702cfbf9 MK |
218 | /** |
219 | * delete_from_page_cache - delete page from page cache | |
220 | * @page: the page which the kernel is trying to remove from page cache | |
221 | * | |
222 | * This must be called only on pages that have been verified to be in the page | |
223 | * cache and locked. It will never put the page into the free list, the caller | |
224 | * has a reference on the page. | |
225 | */ | |
226 | void delete_from_page_cache(struct page *page) | |
1da177e4 LT |
227 | { |
228 | struct address_space *mapping = page->mapping; | |
6072d13c | 229 | void (*freepage)(struct page *); |
1da177e4 | 230 | |
cd7619d6 | 231 | BUG_ON(!PageLocked(page)); |
1da177e4 | 232 | |
6072d13c | 233 | freepage = mapping->a_ops->freepage; |
19fd6231 | 234 | spin_lock_irq(&mapping->tree_lock); |
91b0abe3 | 235 | __delete_from_page_cache(page, NULL); |
19fd6231 | 236 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 237 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
238 | |
239 | if (freepage) | |
240 | freepage(page); | |
97cecb5a MK |
241 | page_cache_release(page); |
242 | } | |
243 | EXPORT_SYMBOL(delete_from_page_cache); | |
244 | ||
865ffef3 DM |
245 | static int filemap_check_errors(struct address_space *mapping) |
246 | { | |
247 | int ret = 0; | |
248 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
249 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
250 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 251 | ret = -ENOSPC; |
7fcbbaf1 JA |
252 | if (test_bit(AS_EIO, &mapping->flags) && |
253 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
254 | ret = -EIO; |
255 | return ret; | |
256 | } | |
257 | ||
1da177e4 | 258 | /** |
485bb99b | 259 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
260 | * @mapping: address space structure to write |
261 | * @start: offset in bytes where the range starts | |
469eb4d0 | 262 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 263 | * @sync_mode: enable synchronous operation |
1da177e4 | 264 | * |
485bb99b RD |
265 | * Start writeback against all of a mapping's dirty pages that lie |
266 | * within the byte offsets <start, end> inclusive. | |
267 | * | |
1da177e4 | 268 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 269 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
270 | * these two operations is that if a dirty page/buffer is encountered, it must |
271 | * be waited upon, and not just skipped over. | |
272 | */ | |
ebcf28e1 AM |
273 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
274 | loff_t end, int sync_mode) | |
1da177e4 LT |
275 | { |
276 | int ret; | |
277 | struct writeback_control wbc = { | |
278 | .sync_mode = sync_mode, | |
05fe478d | 279 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
280 | .range_start = start, |
281 | .range_end = end, | |
1da177e4 LT |
282 | }; |
283 | ||
284 | if (!mapping_cap_writeback_dirty(mapping)) | |
285 | return 0; | |
286 | ||
287 | ret = do_writepages(mapping, &wbc); | |
288 | return ret; | |
289 | } | |
290 | ||
291 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
292 | int sync_mode) | |
293 | { | |
111ebb6e | 294 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
295 | } |
296 | ||
297 | int filemap_fdatawrite(struct address_space *mapping) | |
298 | { | |
299 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
300 | } | |
301 | EXPORT_SYMBOL(filemap_fdatawrite); | |
302 | ||
f4c0a0fd | 303 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 304 | loff_t end) |
1da177e4 LT |
305 | { |
306 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
307 | } | |
f4c0a0fd | 308 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 309 | |
485bb99b RD |
310 | /** |
311 | * filemap_flush - mostly a non-blocking flush | |
312 | * @mapping: target address_space | |
313 | * | |
1da177e4 LT |
314 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
315 | * purposes - I/O may not be started against all dirty pages. | |
316 | */ | |
317 | int filemap_flush(struct address_space *mapping) | |
318 | { | |
319 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
320 | } | |
321 | EXPORT_SYMBOL(filemap_flush); | |
322 | ||
485bb99b | 323 | /** |
94004ed7 CH |
324 | * filemap_fdatawait_range - wait for writeback to complete |
325 | * @mapping: address space structure to wait for | |
326 | * @start_byte: offset in bytes where the range starts | |
327 | * @end_byte: offset in bytes where the range ends (inclusive) | |
485bb99b | 328 | * |
94004ed7 CH |
329 | * Walk the list of under-writeback pages of the given address space |
330 | * in the given range and wait for all of them. | |
1da177e4 | 331 | */ |
94004ed7 CH |
332 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
333 | loff_t end_byte) | |
1da177e4 | 334 | { |
94004ed7 CH |
335 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
336 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
337 | struct pagevec pvec; |
338 | int nr_pages; | |
865ffef3 | 339 | int ret2, ret = 0; |
1da177e4 | 340 | |
94004ed7 | 341 | if (end_byte < start_byte) |
865ffef3 | 342 | goto out; |
1da177e4 LT |
343 | |
344 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
345 | while ((index <= end) && |
346 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
347 | PAGECACHE_TAG_WRITEBACK, | |
348 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
349 | unsigned i; | |
350 | ||
351 | for (i = 0; i < nr_pages; i++) { | |
352 | struct page *page = pvec.pages[i]; | |
353 | ||
354 | /* until radix tree lookup accepts end_index */ | |
355 | if (page->index > end) | |
356 | continue; | |
357 | ||
358 | wait_on_page_writeback(page); | |
212260aa | 359 | if (TestClearPageError(page)) |
1da177e4 LT |
360 | ret = -EIO; |
361 | } | |
362 | pagevec_release(&pvec); | |
363 | cond_resched(); | |
364 | } | |
865ffef3 DM |
365 | out: |
366 | ret2 = filemap_check_errors(mapping); | |
367 | if (!ret) | |
368 | ret = ret2; | |
1da177e4 LT |
369 | |
370 | return ret; | |
371 | } | |
d3bccb6f JK |
372 | EXPORT_SYMBOL(filemap_fdatawait_range); |
373 | ||
1da177e4 | 374 | /** |
485bb99b | 375 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 376 | * @mapping: address space structure to wait for |
485bb99b RD |
377 | * |
378 | * Walk the list of under-writeback pages of the given address space | |
379 | * and wait for all of them. | |
1da177e4 LT |
380 | */ |
381 | int filemap_fdatawait(struct address_space *mapping) | |
382 | { | |
383 | loff_t i_size = i_size_read(mapping->host); | |
384 | ||
385 | if (i_size == 0) | |
386 | return 0; | |
387 | ||
94004ed7 | 388 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
389 | } |
390 | EXPORT_SYMBOL(filemap_fdatawait); | |
391 | ||
392 | int filemap_write_and_wait(struct address_space *mapping) | |
393 | { | |
28fd1298 | 394 | int err = 0; |
1da177e4 LT |
395 | |
396 | if (mapping->nrpages) { | |
28fd1298 OH |
397 | err = filemap_fdatawrite(mapping); |
398 | /* | |
399 | * Even if the above returned error, the pages may be | |
400 | * written partially (e.g. -ENOSPC), so we wait for it. | |
401 | * But the -EIO is special case, it may indicate the worst | |
402 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
403 | */ | |
404 | if (err != -EIO) { | |
405 | int err2 = filemap_fdatawait(mapping); | |
406 | if (!err) | |
407 | err = err2; | |
408 | } | |
865ffef3 DM |
409 | } else { |
410 | err = filemap_check_errors(mapping); | |
1da177e4 | 411 | } |
28fd1298 | 412 | return err; |
1da177e4 | 413 | } |
28fd1298 | 414 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 415 | |
485bb99b RD |
416 | /** |
417 | * filemap_write_and_wait_range - write out & wait on a file range | |
418 | * @mapping: the address_space for the pages | |
419 | * @lstart: offset in bytes where the range starts | |
420 | * @lend: offset in bytes where the range ends (inclusive) | |
421 | * | |
469eb4d0 AM |
422 | * Write out and wait upon file offsets lstart->lend, inclusive. |
423 | * | |
424 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
425 | * that this function can be used to write to the very end-of-file (end = -1). | |
426 | */ | |
1da177e4 LT |
427 | int filemap_write_and_wait_range(struct address_space *mapping, |
428 | loff_t lstart, loff_t lend) | |
429 | { | |
28fd1298 | 430 | int err = 0; |
1da177e4 LT |
431 | |
432 | if (mapping->nrpages) { | |
28fd1298 OH |
433 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
434 | WB_SYNC_ALL); | |
435 | /* See comment of filemap_write_and_wait() */ | |
436 | if (err != -EIO) { | |
94004ed7 CH |
437 | int err2 = filemap_fdatawait_range(mapping, |
438 | lstart, lend); | |
28fd1298 OH |
439 | if (!err) |
440 | err = err2; | |
441 | } | |
865ffef3 DM |
442 | } else { |
443 | err = filemap_check_errors(mapping); | |
1da177e4 | 444 | } |
28fd1298 | 445 | return err; |
1da177e4 | 446 | } |
f6995585 | 447 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 448 | |
ef6a3c63 MS |
449 | /** |
450 | * replace_page_cache_page - replace a pagecache page with a new one | |
451 | * @old: page to be replaced | |
452 | * @new: page to replace with | |
453 | * @gfp_mask: allocation mode | |
454 | * | |
455 | * This function replaces a page in the pagecache with a new one. On | |
456 | * success it acquires the pagecache reference for the new page and | |
457 | * drops it for the old page. Both the old and new pages must be | |
458 | * locked. This function does not add the new page to the LRU, the | |
459 | * caller must do that. | |
460 | * | |
461 | * The remove + add is atomic. The only way this function can fail is | |
462 | * memory allocation failure. | |
463 | */ | |
464 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
465 | { | |
466 | int error; | |
ef6a3c63 | 467 | |
309381fe SL |
468 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
469 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
470 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 471 | |
ef6a3c63 MS |
472 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
473 | if (!error) { | |
474 | struct address_space *mapping = old->mapping; | |
475 | void (*freepage)(struct page *); | |
476 | ||
477 | pgoff_t offset = old->index; | |
478 | freepage = mapping->a_ops->freepage; | |
479 | ||
480 | page_cache_get(new); | |
481 | new->mapping = mapping; | |
482 | new->index = offset; | |
483 | ||
484 | spin_lock_irq(&mapping->tree_lock); | |
91b0abe3 | 485 | __delete_from_page_cache(old, NULL); |
ef6a3c63 MS |
486 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
487 | BUG_ON(error); | |
488 | mapping->nrpages++; | |
489 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
490 | if (PageSwapBacked(new)) | |
491 | __inc_zone_page_state(new, NR_SHMEM); | |
492 | spin_unlock_irq(&mapping->tree_lock); | |
ab936cbc KH |
493 | /* mem_cgroup codes must not be called under tree_lock */ |
494 | mem_cgroup_replace_page_cache(old, new); | |
ef6a3c63 MS |
495 | radix_tree_preload_end(); |
496 | if (freepage) | |
497 | freepage(old); | |
498 | page_cache_release(old); | |
ef6a3c63 MS |
499 | } |
500 | ||
501 | return error; | |
502 | } | |
503 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
504 | ||
0cd6144a | 505 | static int page_cache_tree_insert(struct address_space *mapping, |
a528910e | 506 | struct page *page, void **shadowp) |
0cd6144a | 507 | { |
449dd698 | 508 | struct radix_tree_node *node; |
0cd6144a JW |
509 | void **slot; |
510 | int error; | |
511 | ||
449dd698 JW |
512 | error = __radix_tree_create(&mapping->page_tree, page->index, |
513 | &node, &slot); | |
514 | if (error) | |
515 | return error; | |
516 | if (*slot) { | |
0cd6144a JW |
517 | void *p; |
518 | ||
519 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
520 | if (!radix_tree_exceptional_entry(p)) | |
521 | return -EEXIST; | |
a528910e JW |
522 | if (shadowp) |
523 | *shadowp = p; | |
449dd698 JW |
524 | mapping->nrshadows--; |
525 | if (node) | |
526 | workingset_node_shadows_dec(node); | |
0cd6144a | 527 | } |
449dd698 JW |
528 | radix_tree_replace_slot(slot, page); |
529 | mapping->nrpages++; | |
530 | if (node) { | |
531 | workingset_node_pages_inc(node); | |
532 | /* | |
533 | * Don't track node that contains actual pages. | |
534 | * | |
535 | * Avoid acquiring the list_lru lock if already | |
536 | * untracked. The list_empty() test is safe as | |
537 | * node->private_list is protected by | |
538 | * mapping->tree_lock. | |
539 | */ | |
540 | if (!list_empty(&node->private_list)) | |
541 | list_lru_del(&workingset_shadow_nodes, | |
542 | &node->private_list); | |
543 | } | |
544 | return 0; | |
0cd6144a JW |
545 | } |
546 | ||
a528910e JW |
547 | static int __add_to_page_cache_locked(struct page *page, |
548 | struct address_space *mapping, | |
549 | pgoff_t offset, gfp_t gfp_mask, | |
550 | void **shadowp) | |
1da177e4 | 551 | { |
00501b53 JW |
552 | int huge = PageHuge(page); |
553 | struct mem_cgroup *memcg; | |
e286781d NP |
554 | int error; |
555 | ||
309381fe SL |
556 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
557 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d | 558 | |
00501b53 JW |
559 | if (!huge) { |
560 | error = mem_cgroup_try_charge(page, current->mm, | |
561 | gfp_mask, &memcg); | |
562 | if (error) | |
563 | return error; | |
564 | } | |
1da177e4 | 565 | |
5e4c0d97 | 566 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 567 | if (error) { |
00501b53 JW |
568 | if (!huge) |
569 | mem_cgroup_cancel_charge(page, memcg); | |
66a0c8ee KS |
570 | return error; |
571 | } | |
572 | ||
573 | page_cache_get(page); | |
574 | page->mapping = mapping; | |
575 | page->index = offset; | |
576 | ||
577 | spin_lock_irq(&mapping->tree_lock); | |
a528910e | 578 | error = page_cache_tree_insert(mapping, page, shadowp); |
66a0c8ee KS |
579 | radix_tree_preload_end(); |
580 | if (unlikely(error)) | |
581 | goto err_insert; | |
66a0c8ee KS |
582 | __inc_zone_page_state(page, NR_FILE_PAGES); |
583 | spin_unlock_irq(&mapping->tree_lock); | |
00501b53 JW |
584 | if (!huge) |
585 | mem_cgroup_commit_charge(page, memcg, false); | |
66a0c8ee KS |
586 | trace_mm_filemap_add_to_page_cache(page); |
587 | return 0; | |
588 | err_insert: | |
589 | page->mapping = NULL; | |
590 | /* Leave page->index set: truncation relies upon it */ | |
591 | spin_unlock_irq(&mapping->tree_lock); | |
00501b53 JW |
592 | if (!huge) |
593 | mem_cgroup_cancel_charge(page, memcg); | |
66a0c8ee | 594 | page_cache_release(page); |
1da177e4 LT |
595 | return error; |
596 | } | |
a528910e JW |
597 | |
598 | /** | |
599 | * add_to_page_cache_locked - add a locked page to the pagecache | |
600 | * @page: page to add | |
601 | * @mapping: the page's address_space | |
602 | * @offset: page index | |
603 | * @gfp_mask: page allocation mode | |
604 | * | |
605 | * This function is used to add a page to the pagecache. It must be locked. | |
606 | * This function does not add the page to the LRU. The caller must do that. | |
607 | */ | |
608 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
609 | pgoff_t offset, gfp_t gfp_mask) | |
610 | { | |
611 | return __add_to_page_cache_locked(page, mapping, offset, | |
612 | gfp_mask, NULL); | |
613 | } | |
e286781d | 614 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
615 | |
616 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 617 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 618 | { |
a528910e | 619 | void *shadow = NULL; |
4f98a2fe RR |
620 | int ret; |
621 | ||
a528910e JW |
622 | __set_page_locked(page); |
623 | ret = __add_to_page_cache_locked(page, mapping, offset, | |
624 | gfp_mask, &shadow); | |
625 | if (unlikely(ret)) | |
626 | __clear_page_locked(page); | |
627 | else { | |
628 | /* | |
629 | * The page might have been evicted from cache only | |
630 | * recently, in which case it should be activated like | |
631 | * any other repeatedly accessed page. | |
632 | */ | |
633 | if (shadow && workingset_refault(shadow)) { | |
634 | SetPageActive(page); | |
635 | workingset_activation(page); | |
636 | } else | |
637 | ClearPageActive(page); | |
638 | lru_cache_add(page); | |
639 | } | |
1da177e4 LT |
640 | return ret; |
641 | } | |
18bc0bbd | 642 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 643 | |
44110fe3 | 644 | #ifdef CONFIG_NUMA |
2ae88149 | 645 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 646 | { |
c0ff7453 MX |
647 | int n; |
648 | struct page *page; | |
649 | ||
44110fe3 | 650 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
651 | unsigned int cpuset_mems_cookie; |
652 | do { | |
d26914d1 | 653 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 MG |
654 | n = cpuset_mem_spread_node(); |
655 | page = alloc_pages_exact_node(n, gfp, 0); | |
d26914d1 | 656 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 657 | |
c0ff7453 | 658 | return page; |
44110fe3 | 659 | } |
2ae88149 | 660 | return alloc_pages(gfp, 0); |
44110fe3 | 661 | } |
2ae88149 | 662 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
663 | #endif |
664 | ||
1da177e4 LT |
665 | /* |
666 | * In order to wait for pages to become available there must be | |
667 | * waitqueues associated with pages. By using a hash table of | |
668 | * waitqueues where the bucket discipline is to maintain all | |
669 | * waiters on the same queue and wake all when any of the pages | |
670 | * become available, and for the woken contexts to check to be | |
671 | * sure the appropriate page became available, this saves space | |
672 | * at a cost of "thundering herd" phenomena during rare hash | |
673 | * collisions. | |
674 | */ | |
675 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
676 | { | |
677 | const struct zone *zone = page_zone(page); | |
678 | ||
679 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
680 | } | |
681 | ||
682 | static inline void wake_up_page(struct page *page, int bit) | |
683 | { | |
684 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); | |
685 | } | |
686 | ||
920c7a5d | 687 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
688 | { |
689 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
690 | ||
691 | if (test_bit(bit_nr, &page->flags)) | |
74316201 | 692 | __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
693 | TASK_UNINTERRUPTIBLE); |
694 | } | |
695 | EXPORT_SYMBOL(wait_on_page_bit); | |
696 | ||
f62e00cc KM |
697 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
698 | { | |
699 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
700 | ||
701 | if (!test_bit(bit_nr, &page->flags)) | |
702 | return 0; | |
703 | ||
704 | return __wait_on_bit(page_waitqueue(page), &wait, | |
74316201 | 705 | bit_wait_io, TASK_KILLABLE); |
f62e00cc KM |
706 | } |
707 | ||
385e1ca5 DH |
708 | /** |
709 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
710 | * @page: Page defining the wait queue of interest |
711 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
712 | * |
713 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
714 | */ | |
715 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
716 | { | |
717 | wait_queue_head_t *q = page_waitqueue(page); | |
718 | unsigned long flags; | |
719 | ||
720 | spin_lock_irqsave(&q->lock, flags); | |
721 | __add_wait_queue(q, waiter); | |
722 | spin_unlock_irqrestore(&q->lock, flags); | |
723 | } | |
724 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
725 | ||
1da177e4 | 726 | /** |
485bb99b | 727 | * unlock_page - unlock a locked page |
1da177e4 LT |
728 | * @page: the page |
729 | * | |
730 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
731 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
732 | * mechananism between PageLocked pages and PageWriteback pages is shared. | |
733 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. | |
734 | * | |
8413ac9d NP |
735 | * The mb is necessary to enforce ordering between the clear_bit and the read |
736 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 737 | */ |
920c7a5d | 738 | void unlock_page(struct page *page) |
1da177e4 | 739 | { |
309381fe | 740 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
8413ac9d | 741 | clear_bit_unlock(PG_locked, &page->flags); |
4e857c58 | 742 | smp_mb__after_atomic(); |
1da177e4 LT |
743 | wake_up_page(page, PG_locked); |
744 | } | |
745 | EXPORT_SYMBOL(unlock_page); | |
746 | ||
485bb99b RD |
747 | /** |
748 | * end_page_writeback - end writeback against a page | |
749 | * @page: the page | |
1da177e4 LT |
750 | */ |
751 | void end_page_writeback(struct page *page) | |
752 | { | |
888cf2db MG |
753 | /* |
754 | * TestClearPageReclaim could be used here but it is an atomic | |
755 | * operation and overkill in this particular case. Failing to | |
756 | * shuffle a page marked for immediate reclaim is too mild to | |
757 | * justify taking an atomic operation penalty at the end of | |
758 | * ever page writeback. | |
759 | */ | |
760 | if (PageReclaim(page)) { | |
761 | ClearPageReclaim(page); | |
ac6aadb2 | 762 | rotate_reclaimable_page(page); |
888cf2db | 763 | } |
ac6aadb2 MS |
764 | |
765 | if (!test_clear_page_writeback(page)) | |
766 | BUG(); | |
767 | ||
4e857c58 | 768 | smp_mb__after_atomic(); |
1da177e4 LT |
769 | wake_up_page(page, PG_writeback); |
770 | } | |
771 | EXPORT_SYMBOL(end_page_writeback); | |
772 | ||
57d99845 MW |
773 | /* |
774 | * After completing I/O on a page, call this routine to update the page | |
775 | * flags appropriately | |
776 | */ | |
777 | void page_endio(struct page *page, int rw, int err) | |
778 | { | |
779 | if (rw == READ) { | |
780 | if (!err) { | |
781 | SetPageUptodate(page); | |
782 | } else { | |
783 | ClearPageUptodate(page); | |
784 | SetPageError(page); | |
785 | } | |
786 | unlock_page(page); | |
787 | } else { /* rw == WRITE */ | |
788 | if (err) { | |
789 | SetPageError(page); | |
790 | if (page->mapping) | |
791 | mapping_set_error(page->mapping, err); | |
792 | } | |
793 | end_page_writeback(page); | |
794 | } | |
795 | } | |
796 | EXPORT_SYMBOL_GPL(page_endio); | |
797 | ||
485bb99b RD |
798 | /** |
799 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
800 | * @page: the page to lock | |
1da177e4 | 801 | */ |
920c7a5d | 802 | void __lock_page(struct page *page) |
1da177e4 LT |
803 | { |
804 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
805 | ||
74316201 | 806 | __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
807 | TASK_UNINTERRUPTIBLE); |
808 | } | |
809 | EXPORT_SYMBOL(__lock_page); | |
810 | ||
b5606c2d | 811 | int __lock_page_killable(struct page *page) |
2687a356 MW |
812 | { |
813 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
814 | ||
815 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
74316201 | 816 | bit_wait_io, TASK_KILLABLE); |
2687a356 | 817 | } |
18bc0bbd | 818 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 819 | |
9a95f3cf PC |
820 | /* |
821 | * Return values: | |
822 | * 1 - page is locked; mmap_sem is still held. | |
823 | * 0 - page is not locked. | |
824 | * mmap_sem has been released (up_read()), unless flags had both | |
825 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
826 | * which case mmap_sem is still held. | |
827 | * | |
828 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
829 | * with the page locked and the mmap_sem unperturbed. | |
830 | */ | |
d065bd81 ML |
831 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
832 | unsigned int flags) | |
833 | { | |
37b23e05 KM |
834 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
835 | /* | |
836 | * CAUTION! In this case, mmap_sem is not released | |
837 | * even though return 0. | |
838 | */ | |
839 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
840 | return 0; | |
841 | ||
842 | up_read(&mm->mmap_sem); | |
843 | if (flags & FAULT_FLAG_KILLABLE) | |
844 | wait_on_page_locked_killable(page); | |
845 | else | |
318b275f | 846 | wait_on_page_locked(page); |
d065bd81 | 847 | return 0; |
37b23e05 KM |
848 | } else { |
849 | if (flags & FAULT_FLAG_KILLABLE) { | |
850 | int ret; | |
851 | ||
852 | ret = __lock_page_killable(page); | |
853 | if (ret) { | |
854 | up_read(&mm->mmap_sem); | |
855 | return 0; | |
856 | } | |
857 | } else | |
858 | __lock_page(page); | |
859 | return 1; | |
d065bd81 ML |
860 | } |
861 | } | |
862 | ||
e7b563bb JW |
863 | /** |
864 | * page_cache_next_hole - find the next hole (not-present entry) | |
865 | * @mapping: mapping | |
866 | * @index: index | |
867 | * @max_scan: maximum range to search | |
868 | * | |
869 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
870 | * lowest indexed hole. | |
871 | * | |
872 | * Returns: the index of the hole if found, otherwise returns an index | |
873 | * outside of the set specified (in which case 'return - index >= | |
874 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
875 | * be returned. | |
876 | * | |
877 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
878 | * like radix_tree_gang_lookup, this will not atomically search a | |
879 | * snapshot of the tree at a single point in time. For example, if a | |
880 | * hole is created at index 5, then subsequently a hole is created at | |
881 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
882 | * if called under rcu_read_lock. | |
883 | */ | |
884 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
885 | pgoff_t index, unsigned long max_scan) | |
886 | { | |
887 | unsigned long i; | |
888 | ||
889 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
890 | struct page *page; |
891 | ||
892 | page = radix_tree_lookup(&mapping->page_tree, index); | |
893 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
894 | break; |
895 | index++; | |
896 | if (index == 0) | |
897 | break; | |
898 | } | |
899 | ||
900 | return index; | |
901 | } | |
902 | EXPORT_SYMBOL(page_cache_next_hole); | |
903 | ||
904 | /** | |
905 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
906 | * @mapping: mapping | |
907 | * @index: index | |
908 | * @max_scan: maximum range to search | |
909 | * | |
910 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
911 | * the first hole. | |
912 | * | |
913 | * Returns: the index of the hole if found, otherwise returns an index | |
914 | * outside of the set specified (in which case 'index - return >= | |
915 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
916 | * will be returned. | |
917 | * | |
918 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
919 | * like radix_tree_gang_lookup, this will not atomically search a | |
920 | * snapshot of the tree at a single point in time. For example, if a | |
921 | * hole is created at index 10, then subsequently a hole is created at | |
922 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
923 | * called under rcu_read_lock. | |
924 | */ | |
925 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
926 | pgoff_t index, unsigned long max_scan) | |
927 | { | |
928 | unsigned long i; | |
929 | ||
930 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
931 | struct page *page; |
932 | ||
933 | page = radix_tree_lookup(&mapping->page_tree, index); | |
934 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
935 | break; |
936 | index--; | |
937 | if (index == ULONG_MAX) | |
938 | break; | |
939 | } | |
940 | ||
941 | return index; | |
942 | } | |
943 | EXPORT_SYMBOL(page_cache_prev_hole); | |
944 | ||
485bb99b | 945 | /** |
0cd6144a | 946 | * find_get_entry - find and get a page cache entry |
485bb99b | 947 | * @mapping: the address_space to search |
0cd6144a JW |
948 | * @offset: the page cache index |
949 | * | |
950 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
951 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 952 | * |
139b6a6f JW |
953 | * If the slot holds a shadow entry of a previously evicted page, or a |
954 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
955 | * |
956 | * Otherwise, %NULL is returned. | |
1da177e4 | 957 | */ |
0cd6144a | 958 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 959 | { |
a60637c8 | 960 | void **pagep; |
1da177e4 LT |
961 | struct page *page; |
962 | ||
a60637c8 NP |
963 | rcu_read_lock(); |
964 | repeat: | |
965 | page = NULL; | |
966 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
967 | if (pagep) { | |
968 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
969 | if (unlikely(!page)) |
970 | goto out; | |
a2c16d6c | 971 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
972 | if (radix_tree_deref_retry(page)) |
973 | goto repeat; | |
974 | /* | |
139b6a6f JW |
975 | * A shadow entry of a recently evicted page, |
976 | * or a swap entry from shmem/tmpfs. Return | |
977 | * it without attempting to raise page count. | |
8079b1c8 HD |
978 | */ |
979 | goto out; | |
a2c16d6c | 980 | } |
a60637c8 NP |
981 | if (!page_cache_get_speculative(page)) |
982 | goto repeat; | |
983 | ||
984 | /* | |
985 | * Has the page moved? | |
986 | * This is part of the lockless pagecache protocol. See | |
987 | * include/linux/pagemap.h for details. | |
988 | */ | |
989 | if (unlikely(page != *pagep)) { | |
990 | page_cache_release(page); | |
991 | goto repeat; | |
992 | } | |
993 | } | |
27d20fdd | 994 | out: |
a60637c8 NP |
995 | rcu_read_unlock(); |
996 | ||
1da177e4 LT |
997 | return page; |
998 | } | |
0cd6144a | 999 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1000 | |
0cd6144a JW |
1001 | /** |
1002 | * find_lock_entry - locate, pin and lock a page cache entry | |
1003 | * @mapping: the address_space to search | |
1004 | * @offset: the page cache index | |
1005 | * | |
1006 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1007 | * page cache page, it is returned locked and with an increased | |
1008 | * refcount. | |
1009 | * | |
139b6a6f JW |
1010 | * If the slot holds a shadow entry of a previously evicted page, or a |
1011 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1012 | * |
1013 | * Otherwise, %NULL is returned. | |
1014 | * | |
1015 | * find_lock_entry() may sleep. | |
1016 | */ | |
1017 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1018 | { |
1019 | struct page *page; | |
1020 | ||
1da177e4 | 1021 | repeat: |
0cd6144a | 1022 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1023 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1024 | lock_page(page); |
1025 | /* Has the page been truncated? */ | |
1026 | if (unlikely(page->mapping != mapping)) { | |
1027 | unlock_page(page); | |
1028 | page_cache_release(page); | |
1029 | goto repeat; | |
1da177e4 | 1030 | } |
309381fe | 1031 | VM_BUG_ON_PAGE(page->index != offset, page); |
1da177e4 | 1032 | } |
1da177e4 LT |
1033 | return page; |
1034 | } | |
0cd6144a JW |
1035 | EXPORT_SYMBOL(find_lock_entry); |
1036 | ||
1037 | /** | |
2457aec6 | 1038 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1039 | * @mapping: the address_space to search |
1040 | * @offset: the page index | |
2457aec6 | 1041 | * @fgp_flags: PCG flags |
75325189 RD |
1042 | * @cache_gfp_mask: gfp mask to use for the page cache data page allocation |
1043 | * @radix_gfp_mask: gfp mask to use for radix tree node allocation | |
0cd6144a | 1044 | * |
2457aec6 | 1045 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1046 | * |
75325189 | 1047 | * PCG flags modify how the page is returned. |
0cd6144a | 1048 | * |
2457aec6 MG |
1049 | * FGP_ACCESSED: the page will be marked accessed |
1050 | * FGP_LOCK: Page is return locked | |
1051 | * FGP_CREAT: If page is not present then a new page is allocated using | |
75325189 RD |
1052 | * @cache_gfp_mask and added to the page cache and the VM's LRU |
1053 | * list. If radix tree nodes are allocated during page cache | |
1054 | * insertion then @radix_gfp_mask is used. The page is returned | |
1055 | * locked and with an increased refcount. Otherwise, %NULL is | |
1056 | * returned. | |
1da177e4 | 1057 | * |
2457aec6 MG |
1058 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1059 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1060 | * |
2457aec6 | 1061 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1062 | */ |
2457aec6 MG |
1063 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
1064 | int fgp_flags, gfp_t cache_gfp_mask, gfp_t radix_gfp_mask) | |
1da177e4 | 1065 | { |
eb2be189 | 1066 | struct page *page; |
2457aec6 | 1067 | |
1da177e4 | 1068 | repeat: |
2457aec6 MG |
1069 | page = find_get_entry(mapping, offset); |
1070 | if (radix_tree_exceptional_entry(page)) | |
1071 | page = NULL; | |
1072 | if (!page) | |
1073 | goto no_page; | |
1074 | ||
1075 | if (fgp_flags & FGP_LOCK) { | |
1076 | if (fgp_flags & FGP_NOWAIT) { | |
1077 | if (!trylock_page(page)) { | |
1078 | page_cache_release(page); | |
1079 | return NULL; | |
1080 | } | |
1081 | } else { | |
1082 | lock_page(page); | |
1083 | } | |
1084 | ||
1085 | /* Has the page been truncated? */ | |
1086 | if (unlikely(page->mapping != mapping)) { | |
1087 | unlock_page(page); | |
1088 | page_cache_release(page); | |
1089 | goto repeat; | |
1090 | } | |
1091 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1092 | } | |
1093 | ||
1094 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1095 | mark_page_accessed(page); | |
1096 | ||
1097 | no_page: | |
1098 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1099 | int err; | |
1100 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
1101 | cache_gfp_mask |= __GFP_WRITE; | |
1102 | if (fgp_flags & FGP_NOFS) { | |
1103 | cache_gfp_mask &= ~__GFP_FS; | |
1104 | radix_gfp_mask &= ~__GFP_FS; | |
1105 | } | |
1106 | ||
1107 | page = __page_cache_alloc(cache_gfp_mask); | |
eb2be189 NP |
1108 | if (!page) |
1109 | return NULL; | |
2457aec6 MG |
1110 | |
1111 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1112 | fgp_flags |= FGP_LOCK; | |
1113 | ||
eb39d618 | 1114 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1115 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1116 | __SetPageReferenced(page); |
2457aec6 MG |
1117 | |
1118 | err = add_to_page_cache_lru(page, mapping, offset, radix_gfp_mask); | |
eb2be189 NP |
1119 | if (unlikely(err)) { |
1120 | page_cache_release(page); | |
1121 | page = NULL; | |
1122 | if (err == -EEXIST) | |
1123 | goto repeat; | |
1da177e4 | 1124 | } |
1da177e4 | 1125 | } |
2457aec6 | 1126 | |
1da177e4 LT |
1127 | return page; |
1128 | } | |
2457aec6 | 1129 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1130 | |
0cd6144a JW |
1131 | /** |
1132 | * find_get_entries - gang pagecache lookup | |
1133 | * @mapping: The address_space to search | |
1134 | * @start: The starting page cache index | |
1135 | * @nr_entries: The maximum number of entries | |
1136 | * @entries: Where the resulting entries are placed | |
1137 | * @indices: The cache indices corresponding to the entries in @entries | |
1138 | * | |
1139 | * find_get_entries() will search for and return a group of up to | |
1140 | * @nr_entries entries in the mapping. The entries are placed at | |
1141 | * @entries. find_get_entries() takes a reference against any actual | |
1142 | * pages it returns. | |
1143 | * | |
1144 | * The search returns a group of mapping-contiguous page cache entries | |
1145 | * with ascending indexes. There may be holes in the indices due to | |
1146 | * not-present pages. | |
1147 | * | |
139b6a6f JW |
1148 | * Any shadow entries of evicted pages, or swap entries from |
1149 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1150 | * |
1151 | * find_get_entries() returns the number of pages and shadow entries | |
1152 | * which were found. | |
1153 | */ | |
1154 | unsigned find_get_entries(struct address_space *mapping, | |
1155 | pgoff_t start, unsigned int nr_entries, | |
1156 | struct page **entries, pgoff_t *indices) | |
1157 | { | |
1158 | void **slot; | |
1159 | unsigned int ret = 0; | |
1160 | struct radix_tree_iter iter; | |
1161 | ||
1162 | if (!nr_entries) | |
1163 | return 0; | |
1164 | ||
1165 | rcu_read_lock(); | |
1166 | restart: | |
1167 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { | |
1168 | struct page *page; | |
1169 | repeat: | |
1170 | page = radix_tree_deref_slot(slot); | |
1171 | if (unlikely(!page)) | |
1172 | continue; | |
1173 | if (radix_tree_exception(page)) { | |
1174 | if (radix_tree_deref_retry(page)) | |
1175 | goto restart; | |
1176 | /* | |
139b6a6f JW |
1177 | * A shadow entry of a recently evicted page, |
1178 | * or a swap entry from shmem/tmpfs. Return | |
1179 | * it without attempting to raise page count. | |
0cd6144a JW |
1180 | */ |
1181 | goto export; | |
1182 | } | |
1183 | if (!page_cache_get_speculative(page)) | |
1184 | goto repeat; | |
1185 | ||
1186 | /* Has the page moved? */ | |
1187 | if (unlikely(page != *slot)) { | |
1188 | page_cache_release(page); | |
1189 | goto repeat; | |
1190 | } | |
1191 | export: | |
1192 | indices[ret] = iter.index; | |
1193 | entries[ret] = page; | |
1194 | if (++ret == nr_entries) | |
1195 | break; | |
1196 | } | |
1197 | rcu_read_unlock(); | |
1198 | return ret; | |
1199 | } | |
1200 | ||
1da177e4 LT |
1201 | /** |
1202 | * find_get_pages - gang pagecache lookup | |
1203 | * @mapping: The address_space to search | |
1204 | * @start: The starting page index | |
1205 | * @nr_pages: The maximum number of pages | |
1206 | * @pages: Where the resulting pages are placed | |
1207 | * | |
1208 | * find_get_pages() will search for and return a group of up to | |
1209 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
1210 | * find_get_pages() takes a reference against the returned pages. | |
1211 | * | |
1212 | * The search returns a group of mapping-contiguous pages with ascending | |
1213 | * indexes. There may be holes in the indices due to not-present pages. | |
1214 | * | |
1215 | * find_get_pages() returns the number of pages which were found. | |
1216 | */ | |
1217 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
1218 | unsigned int nr_pages, struct page **pages) | |
1219 | { | |
0fc9d104 KK |
1220 | struct radix_tree_iter iter; |
1221 | void **slot; | |
1222 | unsigned ret = 0; | |
1223 | ||
1224 | if (unlikely(!nr_pages)) | |
1225 | return 0; | |
a60637c8 NP |
1226 | |
1227 | rcu_read_lock(); | |
1228 | restart: | |
0fc9d104 | 1229 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
a60637c8 NP |
1230 | struct page *page; |
1231 | repeat: | |
0fc9d104 | 1232 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1233 | if (unlikely(!page)) |
1234 | continue; | |
9d8aa4ea | 1235 | |
a2c16d6c | 1236 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1237 | if (radix_tree_deref_retry(page)) { |
1238 | /* | |
1239 | * Transient condition which can only trigger | |
1240 | * when entry at index 0 moves out of or back | |
1241 | * to root: none yet gotten, safe to restart. | |
1242 | */ | |
0fc9d104 | 1243 | WARN_ON(iter.index); |
8079b1c8 HD |
1244 | goto restart; |
1245 | } | |
a2c16d6c | 1246 | /* |
139b6a6f JW |
1247 | * A shadow entry of a recently evicted page, |
1248 | * or a swap entry from shmem/tmpfs. Skip | |
1249 | * over it. | |
a2c16d6c | 1250 | */ |
8079b1c8 | 1251 | continue; |
27d20fdd | 1252 | } |
a60637c8 NP |
1253 | |
1254 | if (!page_cache_get_speculative(page)) | |
1255 | goto repeat; | |
1256 | ||
1257 | /* Has the page moved? */ | |
0fc9d104 | 1258 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1259 | page_cache_release(page); |
1260 | goto repeat; | |
1261 | } | |
1da177e4 | 1262 | |
a60637c8 | 1263 | pages[ret] = page; |
0fc9d104 KK |
1264 | if (++ret == nr_pages) |
1265 | break; | |
a60637c8 | 1266 | } |
5b280c0c | 1267 | |
a60637c8 | 1268 | rcu_read_unlock(); |
1da177e4 LT |
1269 | return ret; |
1270 | } | |
1271 | ||
ebf43500 JA |
1272 | /** |
1273 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1274 | * @mapping: The address_space to search | |
1275 | * @index: The starting page index | |
1276 | * @nr_pages: The maximum number of pages | |
1277 | * @pages: Where the resulting pages are placed | |
1278 | * | |
1279 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1280 | * that the returned number of pages are guaranteed to be contiguous. | |
1281 | * | |
1282 | * find_get_pages_contig() returns the number of pages which were found. | |
1283 | */ | |
1284 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1285 | unsigned int nr_pages, struct page **pages) | |
1286 | { | |
0fc9d104 KK |
1287 | struct radix_tree_iter iter; |
1288 | void **slot; | |
1289 | unsigned int ret = 0; | |
1290 | ||
1291 | if (unlikely(!nr_pages)) | |
1292 | return 0; | |
a60637c8 NP |
1293 | |
1294 | rcu_read_lock(); | |
1295 | restart: | |
0fc9d104 | 1296 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
a60637c8 NP |
1297 | struct page *page; |
1298 | repeat: | |
0fc9d104 KK |
1299 | page = radix_tree_deref_slot(slot); |
1300 | /* The hole, there no reason to continue */ | |
a60637c8 | 1301 | if (unlikely(!page)) |
0fc9d104 | 1302 | break; |
9d8aa4ea | 1303 | |
a2c16d6c | 1304 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1305 | if (radix_tree_deref_retry(page)) { |
1306 | /* | |
1307 | * Transient condition which can only trigger | |
1308 | * when entry at index 0 moves out of or back | |
1309 | * to root: none yet gotten, safe to restart. | |
1310 | */ | |
1311 | goto restart; | |
1312 | } | |
a2c16d6c | 1313 | /* |
139b6a6f JW |
1314 | * A shadow entry of a recently evicted page, |
1315 | * or a swap entry from shmem/tmpfs. Stop | |
1316 | * looking for contiguous pages. | |
a2c16d6c | 1317 | */ |
8079b1c8 | 1318 | break; |
a2c16d6c | 1319 | } |
ebf43500 | 1320 | |
a60637c8 NP |
1321 | if (!page_cache_get_speculative(page)) |
1322 | goto repeat; | |
1323 | ||
1324 | /* Has the page moved? */ | |
0fc9d104 | 1325 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1326 | page_cache_release(page); |
1327 | goto repeat; | |
1328 | } | |
1329 | ||
9cbb4cb2 NP |
1330 | /* |
1331 | * must check mapping and index after taking the ref. | |
1332 | * otherwise we can get both false positives and false | |
1333 | * negatives, which is just confusing to the caller. | |
1334 | */ | |
0fc9d104 | 1335 | if (page->mapping == NULL || page->index != iter.index) { |
9cbb4cb2 NP |
1336 | page_cache_release(page); |
1337 | break; | |
1338 | } | |
1339 | ||
a60637c8 | 1340 | pages[ret] = page; |
0fc9d104 KK |
1341 | if (++ret == nr_pages) |
1342 | break; | |
ebf43500 | 1343 | } |
a60637c8 NP |
1344 | rcu_read_unlock(); |
1345 | return ret; | |
ebf43500 | 1346 | } |
ef71c15c | 1347 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1348 | |
485bb99b RD |
1349 | /** |
1350 | * find_get_pages_tag - find and return pages that match @tag | |
1351 | * @mapping: the address_space to search | |
1352 | * @index: the starting page index | |
1353 | * @tag: the tag index | |
1354 | * @nr_pages: the maximum number of pages | |
1355 | * @pages: where the resulting pages are placed | |
1356 | * | |
1da177e4 | 1357 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1358 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
1359 | */ |
1360 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
1361 | int tag, unsigned int nr_pages, struct page **pages) | |
1362 | { | |
0fc9d104 KK |
1363 | struct radix_tree_iter iter; |
1364 | void **slot; | |
1365 | unsigned ret = 0; | |
1366 | ||
1367 | if (unlikely(!nr_pages)) | |
1368 | return 0; | |
a60637c8 NP |
1369 | |
1370 | rcu_read_lock(); | |
1371 | restart: | |
0fc9d104 KK |
1372 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1373 | &iter, *index, tag) { | |
a60637c8 NP |
1374 | struct page *page; |
1375 | repeat: | |
0fc9d104 | 1376 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1377 | if (unlikely(!page)) |
1378 | continue; | |
9d8aa4ea | 1379 | |
a2c16d6c | 1380 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1381 | if (radix_tree_deref_retry(page)) { |
1382 | /* | |
1383 | * Transient condition which can only trigger | |
1384 | * when entry at index 0 moves out of or back | |
1385 | * to root: none yet gotten, safe to restart. | |
1386 | */ | |
1387 | goto restart; | |
1388 | } | |
a2c16d6c | 1389 | /* |
139b6a6f JW |
1390 | * A shadow entry of a recently evicted page. |
1391 | * | |
1392 | * Those entries should never be tagged, but | |
1393 | * this tree walk is lockless and the tags are | |
1394 | * looked up in bulk, one radix tree node at a | |
1395 | * time, so there is a sizable window for page | |
1396 | * reclaim to evict a page we saw tagged. | |
1397 | * | |
1398 | * Skip over it. | |
a2c16d6c | 1399 | */ |
139b6a6f | 1400 | continue; |
a2c16d6c | 1401 | } |
a60637c8 NP |
1402 | |
1403 | if (!page_cache_get_speculative(page)) | |
1404 | goto repeat; | |
1405 | ||
1406 | /* Has the page moved? */ | |
0fc9d104 | 1407 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1408 | page_cache_release(page); |
1409 | goto repeat; | |
1410 | } | |
1411 | ||
1412 | pages[ret] = page; | |
0fc9d104 KK |
1413 | if (++ret == nr_pages) |
1414 | break; | |
a60637c8 | 1415 | } |
5b280c0c | 1416 | |
a60637c8 | 1417 | rcu_read_unlock(); |
1da177e4 | 1418 | |
1da177e4 LT |
1419 | if (ret) |
1420 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1421 | |
1da177e4 LT |
1422 | return ret; |
1423 | } | |
ef71c15c | 1424 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1425 | |
76d42bd9 WF |
1426 | /* |
1427 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1428 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1429 | * | |
1430 | * ---R__________________________________________B__________ | |
1431 | * ^ reading here ^ bad block(assume 4k) | |
1432 | * | |
1433 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1434 | * => failing the whole request => read(R) => read(R+1) => | |
1435 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1436 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1437 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1438 | * | |
1439 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1440 | */ | |
1441 | static void shrink_readahead_size_eio(struct file *filp, | |
1442 | struct file_ra_state *ra) | |
1443 | { | |
76d42bd9 | 1444 | ra->ra_pages /= 4; |
76d42bd9 WF |
1445 | } |
1446 | ||
485bb99b | 1447 | /** |
36e78914 | 1448 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1449 | * @filp: the file to read |
1450 | * @ppos: current file position | |
6e58e79d AV |
1451 | * @iter: data destination |
1452 | * @written: already copied | |
485bb99b | 1453 | * |
1da177e4 | 1454 | * This is a generic file read routine, and uses the |
485bb99b | 1455 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1456 | * |
1457 | * This is really ugly. But the goto's actually try to clarify some | |
1458 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1459 | */ |
6e58e79d AV |
1460 | static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos, |
1461 | struct iov_iter *iter, ssize_t written) | |
1da177e4 | 1462 | { |
36e78914 | 1463 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1464 | struct inode *inode = mapping->host; |
36e78914 | 1465 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1466 | pgoff_t index; |
1467 | pgoff_t last_index; | |
1468 | pgoff_t prev_index; | |
1469 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1470 | unsigned int prev_offset; |
6e58e79d | 1471 | int error = 0; |
1da177e4 | 1472 | |
1da177e4 | 1473 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1474 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1475 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
6e58e79d | 1476 | last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1da177e4 LT |
1477 | offset = *ppos & ~PAGE_CACHE_MASK; |
1478 | ||
1da177e4 LT |
1479 | for (;;) { |
1480 | struct page *page; | |
57f6b96c | 1481 | pgoff_t end_index; |
a32ea1e1 | 1482 | loff_t isize; |
1da177e4 LT |
1483 | unsigned long nr, ret; |
1484 | ||
1da177e4 | 1485 | cond_resched(); |
1da177e4 LT |
1486 | find_page: |
1487 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1488 | if (!page) { |
cf914a7d | 1489 | page_cache_sync_readahead(mapping, |
7ff81078 | 1490 | ra, filp, |
3ea89ee8 FW |
1491 | index, last_index - index); |
1492 | page = find_get_page(mapping, index); | |
1493 | if (unlikely(page == NULL)) | |
1494 | goto no_cached_page; | |
1495 | } | |
1496 | if (PageReadahead(page)) { | |
cf914a7d | 1497 | page_cache_async_readahead(mapping, |
7ff81078 | 1498 | ra, filp, page, |
3ea89ee8 | 1499 | index, last_index - index); |
1da177e4 | 1500 | } |
8ab22b9a HH |
1501 | if (!PageUptodate(page)) { |
1502 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1503 | !mapping->a_ops->is_partially_uptodate) | |
1504 | goto page_not_up_to_date; | |
529ae9aa | 1505 | if (!trylock_page(page)) |
8ab22b9a | 1506 | goto page_not_up_to_date; |
8d056cb9 DH |
1507 | /* Did it get truncated before we got the lock? */ |
1508 | if (!page->mapping) | |
1509 | goto page_not_up_to_date_locked; | |
8ab22b9a | 1510 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 1511 | offset, iter->count)) |
8ab22b9a HH |
1512 | goto page_not_up_to_date_locked; |
1513 | unlock_page(page); | |
1514 | } | |
1da177e4 | 1515 | page_ok: |
a32ea1e1 N |
1516 | /* |
1517 | * i_size must be checked after we know the page is Uptodate. | |
1518 | * | |
1519 | * Checking i_size after the check allows us to calculate | |
1520 | * the correct value for "nr", which means the zero-filled | |
1521 | * part of the page is not copied back to userspace (unless | |
1522 | * another truncate extends the file - this is desired though). | |
1523 | */ | |
1524 | ||
1525 | isize = i_size_read(inode); | |
1526 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1527 | if (unlikely(!isize || index > end_index)) { | |
1528 | page_cache_release(page); | |
1529 | goto out; | |
1530 | } | |
1531 | ||
1532 | /* nr is the maximum number of bytes to copy from this page */ | |
1533 | nr = PAGE_CACHE_SIZE; | |
1534 | if (index == end_index) { | |
1535 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1536 | if (nr <= offset) { | |
1537 | page_cache_release(page); | |
1538 | goto out; | |
1539 | } | |
1540 | } | |
1541 | nr = nr - offset; | |
1da177e4 LT |
1542 | |
1543 | /* If users can be writing to this page using arbitrary | |
1544 | * virtual addresses, take care about potential aliasing | |
1545 | * before reading the page on the kernel side. | |
1546 | */ | |
1547 | if (mapping_writably_mapped(mapping)) | |
1548 | flush_dcache_page(page); | |
1549 | ||
1550 | /* | |
ec0f1637 JK |
1551 | * When a sequential read accesses a page several times, |
1552 | * only mark it as accessed the first time. | |
1da177e4 | 1553 | */ |
ec0f1637 | 1554 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1555 | mark_page_accessed(page); |
1556 | prev_index = index; | |
1557 | ||
1558 | /* | |
1559 | * Ok, we have the page, and it's up-to-date, so | |
1560 | * now we can copy it to user space... | |
1da177e4 | 1561 | */ |
6e58e79d AV |
1562 | |
1563 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 LT |
1564 | offset += ret; |
1565 | index += offset >> PAGE_CACHE_SHIFT; | |
1566 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1567 | prev_offset = offset; |
1da177e4 LT |
1568 | |
1569 | page_cache_release(page); | |
6e58e79d AV |
1570 | written += ret; |
1571 | if (!iov_iter_count(iter)) | |
1572 | goto out; | |
1573 | if (ret < nr) { | |
1574 | error = -EFAULT; | |
1575 | goto out; | |
1576 | } | |
1577 | continue; | |
1da177e4 LT |
1578 | |
1579 | page_not_up_to_date: | |
1580 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1581 | error = lock_page_killable(page); |
1582 | if (unlikely(error)) | |
1583 | goto readpage_error; | |
1da177e4 | 1584 | |
8ab22b9a | 1585 | page_not_up_to_date_locked: |
da6052f7 | 1586 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1587 | if (!page->mapping) { |
1588 | unlock_page(page); | |
1589 | page_cache_release(page); | |
1590 | continue; | |
1591 | } | |
1592 | ||
1593 | /* Did somebody else fill it already? */ | |
1594 | if (PageUptodate(page)) { | |
1595 | unlock_page(page); | |
1596 | goto page_ok; | |
1597 | } | |
1598 | ||
1599 | readpage: | |
91803b49 JM |
1600 | /* |
1601 | * A previous I/O error may have been due to temporary | |
1602 | * failures, eg. multipath errors. | |
1603 | * PG_error will be set again if readpage fails. | |
1604 | */ | |
1605 | ClearPageError(page); | |
1da177e4 LT |
1606 | /* Start the actual read. The read will unlock the page. */ |
1607 | error = mapping->a_ops->readpage(filp, page); | |
1608 | ||
994fc28c ZB |
1609 | if (unlikely(error)) { |
1610 | if (error == AOP_TRUNCATED_PAGE) { | |
1611 | page_cache_release(page); | |
6e58e79d | 1612 | error = 0; |
994fc28c ZB |
1613 | goto find_page; |
1614 | } | |
1da177e4 | 1615 | goto readpage_error; |
994fc28c | 1616 | } |
1da177e4 LT |
1617 | |
1618 | if (!PageUptodate(page)) { | |
85462323 ON |
1619 | error = lock_page_killable(page); |
1620 | if (unlikely(error)) | |
1621 | goto readpage_error; | |
1da177e4 LT |
1622 | if (!PageUptodate(page)) { |
1623 | if (page->mapping == NULL) { | |
1624 | /* | |
2ecdc82e | 1625 | * invalidate_mapping_pages got it |
1da177e4 LT |
1626 | */ |
1627 | unlock_page(page); | |
1628 | page_cache_release(page); | |
1629 | goto find_page; | |
1630 | } | |
1631 | unlock_page(page); | |
7ff81078 | 1632 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1633 | error = -EIO; |
1634 | goto readpage_error; | |
1da177e4 LT |
1635 | } |
1636 | unlock_page(page); | |
1637 | } | |
1638 | ||
1da177e4 LT |
1639 | goto page_ok; |
1640 | ||
1641 | readpage_error: | |
1642 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1da177e4 LT |
1643 | page_cache_release(page); |
1644 | goto out; | |
1645 | ||
1646 | no_cached_page: | |
1647 | /* | |
1648 | * Ok, it wasn't cached, so we need to create a new | |
1649 | * page.. | |
1650 | */ | |
eb2be189 NP |
1651 | page = page_cache_alloc_cold(mapping); |
1652 | if (!page) { | |
6e58e79d | 1653 | error = -ENOMEM; |
eb2be189 | 1654 | goto out; |
1da177e4 | 1655 | } |
eb2be189 | 1656 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1657 | index, GFP_KERNEL); |
1658 | if (error) { | |
eb2be189 | 1659 | page_cache_release(page); |
6e58e79d AV |
1660 | if (error == -EEXIST) { |
1661 | error = 0; | |
1da177e4 | 1662 | goto find_page; |
6e58e79d | 1663 | } |
1da177e4 LT |
1664 | goto out; |
1665 | } | |
1da177e4 LT |
1666 | goto readpage; |
1667 | } | |
1668 | ||
1669 | out: | |
7ff81078 FW |
1670 | ra->prev_pos = prev_index; |
1671 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1672 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1673 | |
f4e6b498 | 1674 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1675 | file_accessed(filp); |
6e58e79d | 1676 | return written ? written : error; |
1da177e4 LT |
1677 | } |
1678 | ||
485bb99b | 1679 | /** |
6abd2322 | 1680 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 1681 | * @iocb: kernel I/O control block |
6abd2322 | 1682 | * @iter: destination for the data read |
485bb99b | 1683 | * |
6abd2322 | 1684 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
1685 | * that can use the page cache directly. |
1686 | */ | |
1687 | ssize_t | |
ed978a81 | 1688 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 1689 | { |
ed978a81 | 1690 | struct file *file = iocb->ki_filp; |
cb66a7a1 | 1691 | ssize_t retval = 0; |
543ade1f | 1692 | loff_t *ppos = &iocb->ki_pos; |
ed978a81 | 1693 | loff_t pos = *ppos; |
1da177e4 LT |
1694 | |
1695 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
ed978a81 AV |
1696 | if (file->f_flags & O_DIRECT) { |
1697 | struct address_space *mapping = file->f_mapping; | |
1698 | struct inode *inode = mapping->host; | |
1699 | size_t count = iov_iter_count(iter); | |
543ade1f | 1700 | loff_t size; |
1da177e4 | 1701 | |
1da177e4 LT |
1702 | if (!count) |
1703 | goto out; /* skip atime */ | |
1704 | size = i_size_read(inode); | |
9fe55eea | 1705 | retval = filemap_write_and_wait_range(mapping, pos, |
a6cbcd4a | 1706 | pos + count - 1); |
9fe55eea | 1707 | if (!retval) { |
ed978a81 | 1708 | struct iov_iter data = *iter; |
26978b8b | 1709 | retval = mapping->a_ops->direct_IO(READ, iocb, &data, pos); |
9fe55eea | 1710 | } |
d8d3d94b | 1711 | |
9fe55eea SW |
1712 | if (retval > 0) { |
1713 | *ppos = pos + retval; | |
ed978a81 | 1714 | iov_iter_advance(iter, retval); |
9fe55eea | 1715 | } |
66f998f6 | 1716 | |
9fe55eea SW |
1717 | /* |
1718 | * Btrfs can have a short DIO read if we encounter | |
1719 | * compressed extents, so if there was an error, or if | |
1720 | * we've already read everything we wanted to, or if | |
1721 | * there was a short read because we hit EOF, go ahead | |
1722 | * and return. Otherwise fallthrough to buffered io for | |
1723 | * the rest of the read. | |
1724 | */ | |
ed978a81 AV |
1725 | if (retval < 0 || !iov_iter_count(iter) || *ppos >= size) { |
1726 | file_accessed(file); | |
9fe55eea | 1727 | goto out; |
0e0bcae3 | 1728 | } |
1da177e4 LT |
1729 | } |
1730 | ||
ed978a81 | 1731 | retval = do_generic_file_read(file, ppos, iter, retval); |
1da177e4 LT |
1732 | out: |
1733 | return retval; | |
1734 | } | |
ed978a81 | 1735 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 1736 | |
1da177e4 | 1737 | #ifdef CONFIG_MMU |
485bb99b RD |
1738 | /** |
1739 | * page_cache_read - adds requested page to the page cache if not already there | |
1740 | * @file: file to read | |
1741 | * @offset: page index | |
1742 | * | |
1da177e4 LT |
1743 | * This adds the requested page to the page cache if it isn't already there, |
1744 | * and schedules an I/O to read in its contents from disk. | |
1745 | */ | |
920c7a5d | 1746 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1747 | { |
1748 | struct address_space *mapping = file->f_mapping; | |
1749 | struct page *page; | |
994fc28c | 1750 | int ret; |
1da177e4 | 1751 | |
994fc28c ZB |
1752 | do { |
1753 | page = page_cache_alloc_cold(mapping); | |
1754 | if (!page) | |
1755 | return -ENOMEM; | |
1756 | ||
1757 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); | |
1758 | if (ret == 0) | |
1759 | ret = mapping->a_ops->readpage(file, page); | |
1760 | else if (ret == -EEXIST) | |
1761 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1762 | |
1da177e4 | 1763 | page_cache_release(page); |
1da177e4 | 1764 | |
994fc28c ZB |
1765 | } while (ret == AOP_TRUNCATED_PAGE); |
1766 | ||
1767 | return ret; | |
1da177e4 LT |
1768 | } |
1769 | ||
1770 | #define MMAP_LOTSAMISS (100) | |
1771 | ||
ef00e08e LT |
1772 | /* |
1773 | * Synchronous readahead happens when we don't even find | |
1774 | * a page in the page cache at all. | |
1775 | */ | |
1776 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1777 | struct file_ra_state *ra, | |
1778 | struct file *file, | |
1779 | pgoff_t offset) | |
1780 | { | |
1781 | unsigned long ra_pages; | |
1782 | struct address_space *mapping = file->f_mapping; | |
1783 | ||
1784 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1785 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 1786 | return; |
275b12bf WF |
1787 | if (!ra->ra_pages) |
1788 | return; | |
ef00e08e | 1789 | |
64363aad | 1790 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
1791 | page_cache_sync_readahead(mapping, ra, file, offset, |
1792 | ra->ra_pages); | |
ef00e08e LT |
1793 | return; |
1794 | } | |
1795 | ||
207d04ba AK |
1796 | /* Avoid banging the cache line if not needed */ |
1797 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1798 | ra->mmap_miss++; |
1799 | ||
1800 | /* | |
1801 | * Do we miss much more than hit in this file? If so, | |
1802 | * stop bothering with read-ahead. It will only hurt. | |
1803 | */ | |
1804 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1805 | return; | |
1806 | ||
d30a1100 WF |
1807 | /* |
1808 | * mmap read-around | |
1809 | */ | |
ef00e08e | 1810 | ra_pages = max_sane_readahead(ra->ra_pages); |
275b12bf WF |
1811 | ra->start = max_t(long, 0, offset - ra_pages / 2); |
1812 | ra->size = ra_pages; | |
2cbea1d3 | 1813 | ra->async_size = ra_pages / 4; |
275b12bf | 1814 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1815 | } |
1816 | ||
1817 | /* | |
1818 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1819 | * so we want to possibly extend the readahead further.. | |
1820 | */ | |
1821 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1822 | struct file_ra_state *ra, | |
1823 | struct file *file, | |
1824 | struct page *page, | |
1825 | pgoff_t offset) | |
1826 | { | |
1827 | struct address_space *mapping = file->f_mapping; | |
1828 | ||
1829 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1830 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
1831 | return; |
1832 | if (ra->mmap_miss > 0) | |
1833 | ra->mmap_miss--; | |
1834 | if (PageReadahead(page)) | |
2fad6f5d WF |
1835 | page_cache_async_readahead(mapping, ra, file, |
1836 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1837 | } |
1838 | ||
485bb99b | 1839 | /** |
54cb8821 | 1840 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1841 | * @vma: vma in which the fault was taken |
1842 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1843 | * |
54cb8821 | 1844 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1845 | * mapped memory region to read in file data during a page fault. |
1846 | * | |
1847 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1848 | * it in the page cache, and handles the special cases reasonably without | |
1849 | * having a lot of duplicated code. | |
9a95f3cf PC |
1850 | * |
1851 | * vma->vm_mm->mmap_sem must be held on entry. | |
1852 | * | |
1853 | * If our return value has VM_FAULT_RETRY set, it's because | |
1854 | * lock_page_or_retry() returned 0. | |
1855 | * The mmap_sem has usually been released in this case. | |
1856 | * See __lock_page_or_retry() for the exception. | |
1857 | * | |
1858 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
1859 | * has not been released. | |
1860 | * | |
1861 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
1da177e4 | 1862 | */ |
d0217ac0 | 1863 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1864 | { |
1865 | int error; | |
54cb8821 | 1866 | struct file *file = vma->vm_file; |
1da177e4 LT |
1867 | struct address_space *mapping = file->f_mapping; |
1868 | struct file_ra_state *ra = &file->f_ra; | |
1869 | struct inode *inode = mapping->host; | |
ef00e08e | 1870 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 1871 | struct page *page; |
99e3e53f | 1872 | loff_t size; |
83c54070 | 1873 | int ret = 0; |
1da177e4 | 1874 | |
99e3e53f KS |
1875 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
1876 | if (offset >= size >> PAGE_CACHE_SHIFT) | |
5307cc1a | 1877 | return VM_FAULT_SIGBUS; |
1da177e4 | 1878 | |
1da177e4 | 1879 | /* |
49426420 | 1880 | * Do we have something in the page cache already? |
1da177e4 | 1881 | */ |
ef00e08e | 1882 | page = find_get_page(mapping, offset); |
45cac65b | 1883 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 1884 | /* |
ef00e08e LT |
1885 | * We found the page, so try async readahead before |
1886 | * waiting for the lock. | |
1da177e4 | 1887 | */ |
ef00e08e | 1888 | do_async_mmap_readahead(vma, ra, file, page, offset); |
45cac65b | 1889 | } else if (!page) { |
ef00e08e LT |
1890 | /* No page in the page cache at all */ |
1891 | do_sync_mmap_readahead(vma, ra, file, offset); | |
1892 | count_vm_event(PGMAJFAULT); | |
456f998e | 1893 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
1894 | ret = VM_FAULT_MAJOR; |
1895 | retry_find: | |
b522c94d | 1896 | page = find_get_page(mapping, offset); |
1da177e4 LT |
1897 | if (!page) |
1898 | goto no_cached_page; | |
1899 | } | |
1900 | ||
d88c0922 ML |
1901 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
1902 | page_cache_release(page); | |
d065bd81 | 1903 | return ret | VM_FAULT_RETRY; |
d88c0922 | 1904 | } |
b522c94d ML |
1905 | |
1906 | /* Did it get truncated? */ | |
1907 | if (unlikely(page->mapping != mapping)) { | |
1908 | unlock_page(page); | |
1909 | put_page(page); | |
1910 | goto retry_find; | |
1911 | } | |
309381fe | 1912 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 1913 | |
1da177e4 | 1914 | /* |
d00806b1 NP |
1915 | * We have a locked page in the page cache, now we need to check |
1916 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1917 | */ |
d00806b1 | 1918 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1919 | goto page_not_uptodate; |
1920 | ||
ef00e08e LT |
1921 | /* |
1922 | * Found the page and have a reference on it. | |
1923 | * We must recheck i_size under page lock. | |
1924 | */ | |
99e3e53f KS |
1925 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
1926 | if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) { | |
d00806b1 | 1927 | unlock_page(page); |
745ad48e | 1928 | page_cache_release(page); |
5307cc1a | 1929 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1930 | } |
1931 | ||
d0217ac0 | 1932 | vmf->page = page; |
83c54070 | 1933 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 1934 | |
1da177e4 LT |
1935 | no_cached_page: |
1936 | /* | |
1937 | * We're only likely to ever get here if MADV_RANDOM is in | |
1938 | * effect. | |
1939 | */ | |
ef00e08e | 1940 | error = page_cache_read(file, offset); |
1da177e4 LT |
1941 | |
1942 | /* | |
1943 | * The page we want has now been added to the page cache. | |
1944 | * In the unlikely event that someone removed it in the | |
1945 | * meantime, we'll just come back here and read it again. | |
1946 | */ | |
1947 | if (error >= 0) | |
1948 | goto retry_find; | |
1949 | ||
1950 | /* | |
1951 | * An error return from page_cache_read can result if the | |
1952 | * system is low on memory, or a problem occurs while trying | |
1953 | * to schedule I/O. | |
1954 | */ | |
1955 | if (error == -ENOMEM) | |
d0217ac0 NP |
1956 | return VM_FAULT_OOM; |
1957 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1958 | |
1959 | page_not_uptodate: | |
1da177e4 LT |
1960 | /* |
1961 | * Umm, take care of errors if the page isn't up-to-date. | |
1962 | * Try to re-read it _once_. We do this synchronously, | |
1963 | * because there really aren't any performance issues here | |
1964 | * and we need to check for errors. | |
1965 | */ | |
1da177e4 | 1966 | ClearPageError(page); |
994fc28c | 1967 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
1968 | if (!error) { |
1969 | wait_on_page_locked(page); | |
1970 | if (!PageUptodate(page)) | |
1971 | error = -EIO; | |
1972 | } | |
d00806b1 NP |
1973 | page_cache_release(page); |
1974 | ||
1975 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 1976 | goto retry_find; |
1da177e4 | 1977 | |
d00806b1 | 1978 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 1979 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 1980 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
1981 | } |
1982 | EXPORT_SYMBOL(filemap_fault); | |
1983 | ||
f1820361 KS |
1984 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
1985 | { | |
1986 | struct radix_tree_iter iter; | |
1987 | void **slot; | |
1988 | struct file *file = vma->vm_file; | |
1989 | struct address_space *mapping = file->f_mapping; | |
1990 | loff_t size; | |
1991 | struct page *page; | |
1992 | unsigned long address = (unsigned long) vmf->virtual_address; | |
1993 | unsigned long addr; | |
1994 | pte_t *pte; | |
1995 | ||
1996 | rcu_read_lock(); | |
1997 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) { | |
1998 | if (iter.index > vmf->max_pgoff) | |
1999 | break; | |
2000 | repeat: | |
2001 | page = radix_tree_deref_slot(slot); | |
2002 | if (unlikely(!page)) | |
2003 | goto next; | |
2004 | if (radix_tree_exception(page)) { | |
2005 | if (radix_tree_deref_retry(page)) | |
2006 | break; | |
2007 | else | |
2008 | goto next; | |
2009 | } | |
2010 | ||
2011 | if (!page_cache_get_speculative(page)) | |
2012 | goto repeat; | |
2013 | ||
2014 | /* Has the page moved? */ | |
2015 | if (unlikely(page != *slot)) { | |
2016 | page_cache_release(page); | |
2017 | goto repeat; | |
2018 | } | |
2019 | ||
2020 | if (!PageUptodate(page) || | |
2021 | PageReadahead(page) || | |
2022 | PageHWPoison(page)) | |
2023 | goto skip; | |
2024 | if (!trylock_page(page)) | |
2025 | goto skip; | |
2026 | ||
2027 | if (page->mapping != mapping || !PageUptodate(page)) | |
2028 | goto unlock; | |
2029 | ||
99e3e53f KS |
2030 | size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE); |
2031 | if (page->index >= size >> PAGE_CACHE_SHIFT) | |
f1820361 KS |
2032 | goto unlock; |
2033 | ||
2034 | pte = vmf->pte + page->index - vmf->pgoff; | |
2035 | if (!pte_none(*pte)) | |
2036 | goto unlock; | |
2037 | ||
2038 | if (file->f_ra.mmap_miss > 0) | |
2039 | file->f_ra.mmap_miss--; | |
2040 | addr = address + (page->index - vmf->pgoff) * PAGE_SIZE; | |
2041 | do_set_pte(vma, addr, page, pte, false, false); | |
2042 | unlock_page(page); | |
2043 | goto next; | |
2044 | unlock: | |
2045 | unlock_page(page); | |
2046 | skip: | |
2047 | page_cache_release(page); | |
2048 | next: | |
2049 | if (iter.index == vmf->max_pgoff) | |
2050 | break; | |
2051 | } | |
2052 | rcu_read_unlock(); | |
2053 | } | |
2054 | EXPORT_SYMBOL(filemap_map_pages); | |
2055 | ||
4fcf1c62 JK |
2056 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2057 | { | |
2058 | struct page *page = vmf->page; | |
496ad9aa | 2059 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
2060 | int ret = VM_FAULT_LOCKED; |
2061 | ||
14da9200 | 2062 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
2063 | file_update_time(vma->vm_file); |
2064 | lock_page(page); | |
2065 | if (page->mapping != inode->i_mapping) { | |
2066 | unlock_page(page); | |
2067 | ret = VM_FAULT_NOPAGE; | |
2068 | goto out; | |
2069 | } | |
14da9200 JK |
2070 | /* |
2071 | * We mark the page dirty already here so that when freeze is in | |
2072 | * progress, we are guaranteed that writeback during freezing will | |
2073 | * see the dirty page and writeprotect it again. | |
2074 | */ | |
2075 | set_page_dirty(page); | |
1d1d1a76 | 2076 | wait_for_stable_page(page); |
4fcf1c62 | 2077 | out: |
14da9200 | 2078 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2079 | return ret; |
2080 | } | |
2081 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2082 | ||
f0f37e2f | 2083 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2084 | .fault = filemap_fault, |
f1820361 | 2085 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2086 | .page_mkwrite = filemap_page_mkwrite, |
0b173bc4 | 2087 | .remap_pages = generic_file_remap_pages, |
1da177e4 LT |
2088 | }; |
2089 | ||
2090 | /* This is used for a general mmap of a disk file */ | |
2091 | ||
2092 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2093 | { | |
2094 | struct address_space *mapping = file->f_mapping; | |
2095 | ||
2096 | if (!mapping->a_ops->readpage) | |
2097 | return -ENOEXEC; | |
2098 | file_accessed(file); | |
2099 | vma->vm_ops = &generic_file_vm_ops; | |
2100 | return 0; | |
2101 | } | |
1da177e4 LT |
2102 | |
2103 | /* | |
2104 | * This is for filesystems which do not implement ->writepage. | |
2105 | */ | |
2106 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2107 | { | |
2108 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2109 | return -EINVAL; | |
2110 | return generic_file_mmap(file, vma); | |
2111 | } | |
2112 | #else | |
2113 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2114 | { | |
2115 | return -ENOSYS; | |
2116 | } | |
2117 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2118 | { | |
2119 | return -ENOSYS; | |
2120 | } | |
2121 | #endif /* CONFIG_MMU */ | |
2122 | ||
2123 | EXPORT_SYMBOL(generic_file_mmap); | |
2124 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2125 | ||
67f9fd91 SL |
2126 | static struct page *wait_on_page_read(struct page *page) |
2127 | { | |
2128 | if (!IS_ERR(page)) { | |
2129 | wait_on_page_locked(page); | |
2130 | if (!PageUptodate(page)) { | |
2131 | page_cache_release(page); | |
2132 | page = ERR_PTR(-EIO); | |
2133 | } | |
2134 | } | |
2135 | return page; | |
2136 | } | |
2137 | ||
6fe6900e | 2138 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 2139 | pgoff_t index, |
5e5358e7 | 2140 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2141 | void *data, |
2142 | gfp_t gfp) | |
1da177e4 | 2143 | { |
eb2be189 | 2144 | struct page *page; |
1da177e4 LT |
2145 | int err; |
2146 | repeat: | |
2147 | page = find_get_page(mapping, index); | |
2148 | if (!page) { | |
0531b2aa | 2149 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
2150 | if (!page) |
2151 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2152 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 NP |
2153 | if (unlikely(err)) { |
2154 | page_cache_release(page); | |
2155 | if (err == -EEXIST) | |
2156 | goto repeat; | |
1da177e4 | 2157 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2158 | return ERR_PTR(err); |
2159 | } | |
1da177e4 LT |
2160 | err = filler(data, page); |
2161 | if (err < 0) { | |
2162 | page_cache_release(page); | |
2163 | page = ERR_PTR(err); | |
67f9fd91 SL |
2164 | } else { |
2165 | page = wait_on_page_read(page); | |
1da177e4 LT |
2166 | } |
2167 | } | |
1da177e4 LT |
2168 | return page; |
2169 | } | |
2170 | ||
0531b2aa | 2171 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2172 | pgoff_t index, |
5e5358e7 | 2173 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2174 | void *data, |
2175 | gfp_t gfp) | |
2176 | ||
1da177e4 LT |
2177 | { |
2178 | struct page *page; | |
2179 | int err; | |
2180 | ||
2181 | retry: | |
0531b2aa | 2182 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 2183 | if (IS_ERR(page)) |
c855ff37 | 2184 | return page; |
1da177e4 LT |
2185 | if (PageUptodate(page)) |
2186 | goto out; | |
2187 | ||
2188 | lock_page(page); | |
2189 | if (!page->mapping) { | |
2190 | unlock_page(page); | |
2191 | page_cache_release(page); | |
2192 | goto retry; | |
2193 | } | |
2194 | if (PageUptodate(page)) { | |
2195 | unlock_page(page); | |
2196 | goto out; | |
2197 | } | |
2198 | err = filler(data, page); | |
2199 | if (err < 0) { | |
2200 | page_cache_release(page); | |
c855ff37 | 2201 | return ERR_PTR(err); |
67f9fd91 SL |
2202 | } else { |
2203 | page = wait_on_page_read(page); | |
2204 | if (IS_ERR(page)) | |
2205 | return page; | |
1da177e4 | 2206 | } |
c855ff37 | 2207 | out: |
6fe6900e NP |
2208 | mark_page_accessed(page); |
2209 | return page; | |
2210 | } | |
0531b2aa LT |
2211 | |
2212 | /** | |
67f9fd91 | 2213 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2214 | * @mapping: the page's address_space |
2215 | * @index: the page index | |
2216 | * @filler: function to perform the read | |
5e5358e7 | 2217 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2218 | * |
0531b2aa | 2219 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2220 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2221 | * |
2222 | * If the page does not get brought uptodate, return -EIO. | |
2223 | */ | |
67f9fd91 | 2224 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2225 | pgoff_t index, |
5e5358e7 | 2226 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2227 | void *data) |
2228 | { | |
2229 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2230 | } | |
67f9fd91 | 2231 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2232 | |
2233 | /** | |
2234 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2235 | * @mapping: the page's address_space | |
2236 | * @index: the page index | |
2237 | * @gfp: the page allocator flags to use if allocating | |
2238 | * | |
2239 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2240 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2241 | * |
2242 | * If the page does not get brought uptodate, return -EIO. | |
2243 | */ | |
2244 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2245 | pgoff_t index, | |
2246 | gfp_t gfp) | |
2247 | { | |
2248 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2249 | ||
67f9fd91 | 2250 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2251 | } |
2252 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2253 | ||
1da177e4 LT |
2254 | /* |
2255 | * Performs necessary checks before doing a write | |
2256 | * | |
485bb99b | 2257 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2258 | * Returns appropriate error code that caller should return or |
2259 | * zero in case that write should be allowed. | |
2260 | */ | |
2261 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) | |
2262 | { | |
2263 | struct inode *inode = file->f_mapping->host; | |
59e99e5b | 2264 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
1da177e4 LT |
2265 | |
2266 | if (unlikely(*pos < 0)) | |
2267 | return -EINVAL; | |
2268 | ||
1da177e4 LT |
2269 | if (!isblk) { |
2270 | /* FIXME: this is for backwards compatibility with 2.4 */ | |
2271 | if (file->f_flags & O_APPEND) | |
2272 | *pos = i_size_read(inode); | |
2273 | ||
2274 | if (limit != RLIM_INFINITY) { | |
2275 | if (*pos >= limit) { | |
2276 | send_sig(SIGXFSZ, current, 0); | |
2277 | return -EFBIG; | |
2278 | } | |
2279 | if (*count > limit - (typeof(limit))*pos) { | |
2280 | *count = limit - (typeof(limit))*pos; | |
2281 | } | |
2282 | } | |
2283 | } | |
2284 | ||
2285 | /* | |
2286 | * LFS rule | |
2287 | */ | |
2288 | if (unlikely(*pos + *count > MAX_NON_LFS && | |
2289 | !(file->f_flags & O_LARGEFILE))) { | |
2290 | if (*pos >= MAX_NON_LFS) { | |
1da177e4 LT |
2291 | return -EFBIG; |
2292 | } | |
2293 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { | |
2294 | *count = MAX_NON_LFS - (unsigned long)*pos; | |
2295 | } | |
2296 | } | |
2297 | ||
2298 | /* | |
2299 | * Are we about to exceed the fs block limit ? | |
2300 | * | |
2301 | * If we have written data it becomes a short write. If we have | |
2302 | * exceeded without writing data we send a signal and return EFBIG. | |
2303 | * Linus frestrict idea will clean these up nicely.. | |
2304 | */ | |
2305 | if (likely(!isblk)) { | |
2306 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { | |
2307 | if (*count || *pos > inode->i_sb->s_maxbytes) { | |
1da177e4 LT |
2308 | return -EFBIG; |
2309 | } | |
2310 | /* zero-length writes at ->s_maxbytes are OK */ | |
2311 | } | |
2312 | ||
2313 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) | |
2314 | *count = inode->i_sb->s_maxbytes - *pos; | |
2315 | } else { | |
9361401e | 2316 | #ifdef CONFIG_BLOCK |
1da177e4 LT |
2317 | loff_t isize; |
2318 | if (bdev_read_only(I_BDEV(inode))) | |
2319 | return -EPERM; | |
2320 | isize = i_size_read(inode); | |
2321 | if (*pos >= isize) { | |
2322 | if (*count || *pos > isize) | |
2323 | return -ENOSPC; | |
2324 | } | |
2325 | ||
2326 | if (*pos + *count > isize) | |
2327 | *count = isize - *pos; | |
9361401e DH |
2328 | #else |
2329 | return -EPERM; | |
2330 | #endif | |
1da177e4 LT |
2331 | } |
2332 | return 0; | |
2333 | } | |
2334 | EXPORT_SYMBOL(generic_write_checks); | |
2335 | ||
afddba49 NP |
2336 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2337 | loff_t pos, unsigned len, unsigned flags, | |
2338 | struct page **pagep, void **fsdata) | |
2339 | { | |
2340 | const struct address_space_operations *aops = mapping->a_ops; | |
2341 | ||
4e02ed4b | 2342 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2343 | pagep, fsdata); |
afddba49 NP |
2344 | } |
2345 | EXPORT_SYMBOL(pagecache_write_begin); | |
2346 | ||
2347 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2348 | loff_t pos, unsigned len, unsigned copied, | |
2349 | struct page *page, void *fsdata) | |
2350 | { | |
2351 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2352 | |
4e02ed4b | 2353 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2354 | } |
2355 | EXPORT_SYMBOL(pagecache_write_end); | |
2356 | ||
1da177e4 | 2357 | ssize_t |
0c949334 | 2358 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos) |
1da177e4 LT |
2359 | { |
2360 | struct file *file = iocb->ki_filp; | |
2361 | struct address_space *mapping = file->f_mapping; | |
2362 | struct inode *inode = mapping->host; | |
2363 | ssize_t written; | |
a969e903 CH |
2364 | size_t write_len; |
2365 | pgoff_t end; | |
26978b8b | 2366 | struct iov_iter data; |
1da177e4 | 2367 | |
0c949334 | 2368 | write_len = iov_iter_count(from); |
a969e903 | 2369 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; |
a969e903 | 2370 | |
48b47c56 | 2371 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2372 | if (written) |
2373 | goto out; | |
2374 | ||
2375 | /* | |
2376 | * After a write we want buffered reads to be sure to go to disk to get | |
2377 | * the new data. We invalidate clean cached page from the region we're | |
2378 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2379 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2380 | */ |
2381 | if (mapping->nrpages) { | |
2382 | written = invalidate_inode_pages2_range(mapping, | |
2383 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2384 | /* |
2385 | * If a page can not be invalidated, return 0 to fall back | |
2386 | * to buffered write. | |
2387 | */ | |
2388 | if (written) { | |
2389 | if (written == -EBUSY) | |
2390 | return 0; | |
a969e903 | 2391 | goto out; |
6ccfa806 | 2392 | } |
a969e903 CH |
2393 | } |
2394 | ||
26978b8b AV |
2395 | data = *from; |
2396 | written = mapping->a_ops->direct_IO(WRITE, iocb, &data, pos); | |
a969e903 CH |
2397 | |
2398 | /* | |
2399 | * Finally, try again to invalidate clean pages which might have been | |
2400 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2401 | * if the source of the write was an mmap'ed region of the file | |
2402 | * we're writing. Either one is a pretty crazy thing to do, | |
2403 | * so we don't support it 100%. If this invalidation | |
2404 | * fails, tough, the write still worked... | |
2405 | */ | |
2406 | if (mapping->nrpages) { | |
2407 | invalidate_inode_pages2_range(mapping, | |
2408 | pos >> PAGE_CACHE_SHIFT, end); | |
2409 | } | |
2410 | ||
1da177e4 | 2411 | if (written > 0) { |
0116651c | 2412 | pos += written; |
f8579f86 | 2413 | iov_iter_advance(from, written); |
0116651c NK |
2414 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
2415 | i_size_write(inode, pos); | |
1da177e4 LT |
2416 | mark_inode_dirty(inode); |
2417 | } | |
5cb6c6c7 | 2418 | iocb->ki_pos = pos; |
1da177e4 | 2419 | } |
a969e903 | 2420 | out: |
1da177e4 LT |
2421 | return written; |
2422 | } | |
2423 | EXPORT_SYMBOL(generic_file_direct_write); | |
2424 | ||
eb2be189 NP |
2425 | /* |
2426 | * Find or create a page at the given pagecache position. Return the locked | |
2427 | * page. This function is specifically for buffered writes. | |
2428 | */ | |
54566b2c NP |
2429 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2430 | pgoff_t index, unsigned flags) | |
eb2be189 | 2431 | { |
eb2be189 | 2432 | struct page *page; |
2457aec6 | 2433 | int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT; |
0faa70cb | 2434 | |
54566b2c | 2435 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
2436 | fgp_flags |= FGP_NOFS; |
2437 | ||
2438 | page = pagecache_get_page(mapping, index, fgp_flags, | |
2439 | mapping_gfp_mask(mapping), | |
2440 | GFP_KERNEL); | |
c585a267 | 2441 | if (page) |
2457aec6 | 2442 | wait_for_stable_page(page); |
eb2be189 | 2443 | |
eb2be189 NP |
2444 | return page; |
2445 | } | |
54566b2c | 2446 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2447 | |
3b93f911 | 2448 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
2449 | struct iov_iter *i, loff_t pos) |
2450 | { | |
2451 | struct address_space *mapping = file->f_mapping; | |
2452 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2453 | long status = 0; | |
2454 | ssize_t written = 0; | |
674b892e NP |
2455 | unsigned int flags = 0; |
2456 | ||
2457 | /* | |
2458 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2459 | */ | |
2460 | if (segment_eq(get_fs(), KERNEL_DS)) | |
2461 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
afddba49 NP |
2462 | |
2463 | do { | |
2464 | struct page *page; | |
afddba49 NP |
2465 | unsigned long offset; /* Offset into pagecache page */ |
2466 | unsigned long bytes; /* Bytes to write to page */ | |
2467 | size_t copied; /* Bytes copied from user */ | |
2468 | void *fsdata; | |
2469 | ||
2470 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2471 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2472 | iov_iter_count(i)); | |
2473 | ||
2474 | again: | |
afddba49 NP |
2475 | /* |
2476 | * Bring in the user page that we will copy from _first_. | |
2477 | * Otherwise there's a nasty deadlock on copying from the | |
2478 | * same page as we're writing to, without it being marked | |
2479 | * up-to-date. | |
2480 | * | |
2481 | * Not only is this an optimisation, but it is also required | |
2482 | * to check that the address is actually valid, when atomic | |
2483 | * usercopies are used, below. | |
2484 | */ | |
2485 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2486 | status = -EFAULT; | |
2487 | break; | |
2488 | } | |
2489 | ||
674b892e | 2490 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 2491 | &page, &fsdata); |
2457aec6 | 2492 | if (unlikely(status < 0)) |
afddba49 NP |
2493 | break; |
2494 | ||
931e80e4 | 2495 | if (mapping_writably_mapped(mapping)) |
2496 | flush_dcache_page(page); | |
2497 | ||
afddba49 | 2498 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
2499 | flush_dcache_page(page); |
2500 | ||
2501 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
2502 | page, fsdata); | |
2503 | if (unlikely(status < 0)) | |
2504 | break; | |
2505 | copied = status; | |
2506 | ||
2507 | cond_resched(); | |
2508 | ||
124d3b70 | 2509 | iov_iter_advance(i, copied); |
afddba49 NP |
2510 | if (unlikely(copied == 0)) { |
2511 | /* | |
2512 | * If we were unable to copy any data at all, we must | |
2513 | * fall back to a single segment length write. | |
2514 | * | |
2515 | * If we didn't fallback here, we could livelock | |
2516 | * because not all segments in the iov can be copied at | |
2517 | * once without a pagefault. | |
2518 | */ | |
2519 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2520 | iov_iter_single_seg_count(i)); | |
2521 | goto again; | |
2522 | } | |
afddba49 NP |
2523 | pos += copied; |
2524 | written += copied; | |
2525 | ||
2526 | balance_dirty_pages_ratelimited(mapping); | |
a50527b1 JK |
2527 | if (fatal_signal_pending(current)) { |
2528 | status = -EINTR; | |
2529 | break; | |
2530 | } | |
afddba49 NP |
2531 | } while (iov_iter_count(i)); |
2532 | ||
2533 | return written ? written : status; | |
2534 | } | |
3b93f911 | 2535 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 2536 | |
e4dd9de3 | 2537 | /** |
8174202b | 2538 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 2539 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 2540 | * @from: iov_iter with data to write |
e4dd9de3 JK |
2541 | * |
2542 | * This function does all the work needed for actually writing data to a | |
2543 | * file. It does all basic checks, removes SUID from the file, updates | |
2544 | * modification times and calls proper subroutines depending on whether we | |
2545 | * do direct IO or a standard buffered write. | |
2546 | * | |
2547 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2548 | * object which does not need locking at all. | |
2549 | * | |
2550 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2551 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2552 | * avoid syncing under i_mutex. | |
2553 | */ | |
8174202b | 2554 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2555 | { |
2556 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2557 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 2558 | struct inode *inode = mapping->host; |
41fc56d5 | 2559 | loff_t pos = iocb->ki_pos; |
3b93f911 | 2560 | ssize_t written = 0; |
1da177e4 | 2561 | ssize_t err; |
3b93f911 | 2562 | ssize_t status; |
8174202b | 2563 | size_t count = iov_iter_count(from); |
1da177e4 | 2564 | |
1da177e4 LT |
2565 | /* We can write back this queue in page reclaim */ |
2566 | current->backing_dev_info = mapping->backing_dev_info; | |
1da177e4 LT |
2567 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); |
2568 | if (err) | |
2569 | goto out; | |
2570 | ||
2571 | if (count == 0) | |
2572 | goto out; | |
2573 | ||
8174202b | 2574 | iov_iter_truncate(from, count); |
0c949334 | 2575 | |
2f1936b8 | 2576 | err = file_remove_suid(file); |
1da177e4 LT |
2577 | if (err) |
2578 | goto out; | |
2579 | ||
c3b2da31 JB |
2580 | err = file_update_time(file); |
2581 | if (err) | |
2582 | goto out; | |
1da177e4 LT |
2583 | |
2584 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
2585 | if (unlikely(file->f_flags & O_DIRECT)) { | |
fb5527e6 | 2586 | loff_t endbyte; |
fb5527e6 | 2587 | |
8174202b | 2588 | written = generic_file_direct_write(iocb, from, pos); |
1da177e4 LT |
2589 | if (written < 0 || written == count) |
2590 | goto out; | |
3b93f911 | 2591 | |
1da177e4 LT |
2592 | /* |
2593 | * direct-io write to a hole: fall through to buffered I/O | |
2594 | * for completing the rest of the request. | |
2595 | */ | |
2596 | pos += written; | |
2597 | count -= written; | |
3b93f911 | 2598 | |
8174202b | 2599 | status = generic_perform_write(file, from, pos); |
fb5527e6 | 2600 | /* |
3b93f911 | 2601 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
2602 | * then we want to return the number of bytes which were |
2603 | * direct-written, or the error code if that was zero. Note | |
2604 | * that this differs from normal direct-io semantics, which | |
2605 | * will return -EFOO even if some bytes were written. | |
2606 | */ | |
3b93f911 AV |
2607 | if (unlikely(status < 0) && !written) { |
2608 | err = status; | |
fb5527e6 JM |
2609 | goto out; |
2610 | } | |
3b93f911 | 2611 | iocb->ki_pos = pos + status; |
fb5527e6 JM |
2612 | /* |
2613 | * We need to ensure that the page cache pages are written to | |
2614 | * disk and invalidated to preserve the expected O_DIRECT | |
2615 | * semantics. | |
2616 | */ | |
3b93f911 | 2617 | endbyte = pos + status - 1; |
c05c4edd | 2618 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); |
fb5527e6 | 2619 | if (err == 0) { |
3b93f911 | 2620 | written += status; |
fb5527e6 JM |
2621 | invalidate_mapping_pages(mapping, |
2622 | pos >> PAGE_CACHE_SHIFT, | |
2623 | endbyte >> PAGE_CACHE_SHIFT); | |
2624 | } else { | |
2625 | /* | |
2626 | * We don't know how much we wrote, so just return | |
2627 | * the number of bytes which were direct-written | |
2628 | */ | |
2629 | } | |
2630 | } else { | |
8174202b | 2631 | written = generic_perform_write(file, from, pos); |
3b93f911 AV |
2632 | if (likely(written >= 0)) |
2633 | iocb->ki_pos = pos + written; | |
fb5527e6 | 2634 | } |
1da177e4 LT |
2635 | out: |
2636 | current->backing_dev_info = NULL; | |
2637 | return written ? written : err; | |
2638 | } | |
8174202b | 2639 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 2640 | |
e4dd9de3 | 2641 | /** |
8174202b | 2642 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 2643 | * @iocb: IO state structure |
8174202b | 2644 | * @from: iov_iter with data to write |
e4dd9de3 | 2645 | * |
8174202b | 2646 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
2647 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
2648 | * and acquires i_mutex as needed. | |
2649 | */ | |
8174202b | 2650 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2651 | { |
2652 | struct file *file = iocb->ki_filp; | |
148f948b | 2653 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2654 | ssize_t ret; |
1da177e4 | 2655 | |
1b1dcc1b | 2656 | mutex_lock(&inode->i_mutex); |
8174202b | 2657 | ret = __generic_file_write_iter(iocb, from); |
1b1dcc1b | 2658 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 2659 | |
02afc27f | 2660 | if (ret > 0) { |
1da177e4 LT |
2661 | ssize_t err; |
2662 | ||
d311d79d AV |
2663 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
2664 | if (err < 0) | |
1da177e4 LT |
2665 | ret = err; |
2666 | } | |
2667 | return ret; | |
2668 | } | |
8174202b | 2669 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 2670 | |
cf9a2ae8 DH |
2671 | /** |
2672 | * try_to_release_page() - release old fs-specific metadata on a page | |
2673 | * | |
2674 | * @page: the page which the kernel is trying to free | |
2675 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2676 | * | |
2677 | * The address_space is to try to release any data against the page | |
2678 | * (presumably at page->private). If the release was successful, return `1'. | |
2679 | * Otherwise return zero. | |
2680 | * | |
266cf658 DH |
2681 | * This may also be called if PG_fscache is set on a page, indicating that the |
2682 | * page is known to the local caching routines. | |
2683 | * | |
cf9a2ae8 | 2684 | * The @gfp_mask argument specifies whether I/O may be performed to release |
3f31fddf | 2685 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
cf9a2ae8 | 2686 | * |
cf9a2ae8 DH |
2687 | */ |
2688 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2689 | { | |
2690 | struct address_space * const mapping = page->mapping; | |
2691 | ||
2692 | BUG_ON(!PageLocked(page)); | |
2693 | if (PageWriteback(page)) | |
2694 | return 0; | |
2695 | ||
2696 | if (mapping && mapping->a_ops->releasepage) | |
2697 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2698 | return try_to_free_buffers(page); | |
2699 | } | |
2700 | ||
2701 | EXPORT_SYMBOL(try_to_release_page); |