]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/filemap.c | |
3 | * | |
4 | * Copyright (C) 1994-1999 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * This file handles the generic file mmap semantics used by | |
9 | * most "normal" filesystems (but you don't /have/ to use this: | |
10 | * the NFS filesystem used to do this differently, for example) | |
11 | */ | |
b95f1b31 | 12 | #include <linux/export.h> |
1da177e4 LT |
13 | #include <linux/compiler.h> |
14 | #include <linux/fs.h> | |
c22ce143 | 15 | #include <linux/uaccess.h> |
c59ede7b | 16 | #include <linux/capability.h> |
1da177e4 | 17 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 18 | #include <linux/gfp.h> |
1da177e4 LT |
19 | #include <linux/mm.h> |
20 | #include <linux/swap.h> | |
21 | #include <linux/mman.h> | |
22 | #include <linux/pagemap.h> | |
23 | #include <linux/file.h> | |
24 | #include <linux/uio.h> | |
25 | #include <linux/hash.h> | |
26 | #include <linux/writeback.h> | |
53253383 | 27 | #include <linux/backing-dev.h> |
1da177e4 LT |
28 | #include <linux/pagevec.h> |
29 | #include <linux/blkdev.h> | |
30 | #include <linux/security.h> | |
44110fe3 | 31 | #include <linux/cpuset.h> |
2f718ffc | 32 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
00501b53 | 33 | #include <linux/hugetlb.h> |
8a9f3ccd | 34 | #include <linux/memcontrol.h> |
c515e1fd | 35 | #include <linux/cleancache.h> |
f1820361 | 36 | #include <linux/rmap.h> |
0f8053a5 NP |
37 | #include "internal.h" |
38 | ||
fe0bfaaf RJ |
39 | #define CREATE_TRACE_POINTS |
40 | #include <trace/events/filemap.h> | |
41 | ||
1da177e4 | 42 | /* |
1da177e4 LT |
43 | * FIXME: remove all knowledge of the buffer layer from the core VM |
44 | */ | |
148f948b | 45 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 46 | |
1da177e4 LT |
47 | #include <asm/mman.h> |
48 | ||
49 | /* | |
50 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
51 | * though. | |
52 | * | |
53 | * Shared mappings now work. 15.8.1995 Bruno. | |
54 | * | |
55 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
56 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
57 | * | |
58 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
59 | */ | |
60 | ||
61 | /* | |
62 | * Lock ordering: | |
63 | * | |
c8c06efa | 64 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 65 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 HD |
66 | * ->swap_lock (exclusive_swap_page, others) |
67 | * ->mapping->tree_lock | |
1da177e4 | 68 | * |
1b1dcc1b | 69 | * ->i_mutex |
c8c06efa | 70 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
1da177e4 LT |
71 | * |
72 | * ->mmap_sem | |
c8c06efa | 73 | * ->i_mmap_rwsem |
b8072f09 | 74 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
75 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
76 | * | |
77 | * ->mmap_sem | |
78 | * ->lock_page (access_process_vm) | |
79 | * | |
ccad2365 | 80 | * ->i_mutex (generic_perform_write) |
82591e6e | 81 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
1da177e4 | 82 | * |
f758eeab | 83 | * bdi->wb.list_lock |
a66979ab | 84 | * sb_lock (fs/fs-writeback.c) |
1da177e4 LT |
85 | * ->mapping->tree_lock (__sync_single_inode) |
86 | * | |
c8c06efa | 87 | * ->i_mmap_rwsem |
1da177e4 LT |
88 | * ->anon_vma.lock (vma_adjust) |
89 | * | |
90 | * ->anon_vma.lock | |
b8072f09 | 91 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 92 | * |
b8072f09 | 93 | * ->page_table_lock or pte_lock |
5d337b91 | 94 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
95 | * ->private_lock (try_to_unmap_one) |
96 | * ->tree_lock (try_to_unmap_one) | |
97 | * ->zone.lru_lock (follow_page->mark_page_accessed) | |
053837fc | 98 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
1da177e4 LT |
99 | * ->private_lock (page_remove_rmap->set_page_dirty) |
100 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
f758eeab | 101 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 102 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
f758eeab | 103 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 104 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
105 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
106 | * | |
c8c06efa | 107 | * ->i_mmap_rwsem |
9a3c531d | 108 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
109 | */ |
110 | ||
91b0abe3 JW |
111 | static void page_cache_tree_delete(struct address_space *mapping, |
112 | struct page *page, void *shadow) | |
113 | { | |
449dd698 JW |
114 | struct radix_tree_node *node; |
115 | unsigned long index; | |
116 | unsigned int offset; | |
117 | unsigned int tag; | |
118 | void **slot; | |
91b0abe3 | 119 | |
449dd698 JW |
120 | VM_BUG_ON(!PageLocked(page)); |
121 | ||
122 | __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot); | |
123 | ||
124 | if (shadow) { | |
91b0abe3 JW |
125 | mapping->nrshadows++; |
126 | /* | |
127 | * Make sure the nrshadows update is committed before | |
128 | * the nrpages update so that final truncate racing | |
129 | * with reclaim does not see both counters 0 at the | |
130 | * same time and miss a shadow entry. | |
131 | */ | |
132 | smp_wmb(); | |
449dd698 | 133 | } |
91b0abe3 | 134 | mapping->nrpages--; |
449dd698 JW |
135 | |
136 | if (!node) { | |
137 | /* Clear direct pointer tags in root node */ | |
138 | mapping->page_tree.gfp_mask &= __GFP_BITS_MASK; | |
139 | radix_tree_replace_slot(slot, shadow); | |
140 | return; | |
141 | } | |
142 | ||
143 | /* Clear tree tags for the removed page */ | |
144 | index = page->index; | |
145 | offset = index & RADIX_TREE_MAP_MASK; | |
146 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { | |
147 | if (test_bit(offset, node->tags[tag])) | |
148 | radix_tree_tag_clear(&mapping->page_tree, index, tag); | |
149 | } | |
150 | ||
151 | /* Delete page, swap shadow entry */ | |
152 | radix_tree_replace_slot(slot, shadow); | |
153 | workingset_node_pages_dec(node); | |
154 | if (shadow) | |
155 | workingset_node_shadows_inc(node); | |
156 | else | |
157 | if (__radix_tree_delete_node(&mapping->page_tree, node)) | |
158 | return; | |
159 | ||
160 | /* | |
161 | * Track node that only contains shadow entries. | |
162 | * | |
163 | * Avoid acquiring the list_lru lock if already tracked. The | |
164 | * list_empty() test is safe as node->private_list is | |
165 | * protected by mapping->tree_lock. | |
166 | */ | |
167 | if (!workingset_node_pages(node) && | |
168 | list_empty(&node->private_list)) { | |
169 | node->private_data = mapping; | |
170 | list_lru_add(&workingset_shadow_nodes, &node->private_list); | |
171 | } | |
91b0abe3 JW |
172 | } |
173 | ||
1da177e4 | 174 | /* |
e64a782f | 175 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 176 | * sure the page is locked and that nobody else uses it - or that usage |
19fd6231 | 177 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 178 | */ |
91b0abe3 | 179 | void __delete_from_page_cache(struct page *page, void *shadow) |
1da177e4 LT |
180 | { |
181 | struct address_space *mapping = page->mapping; | |
182 | ||
fe0bfaaf | 183 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
184 | /* |
185 | * if we're uptodate, flush out into the cleancache, otherwise | |
186 | * invalidate any existing cleancache entries. We can't leave | |
187 | * stale data around in the cleancache once our page is gone | |
188 | */ | |
189 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
190 | cleancache_put_page(page); | |
191 | else | |
3167760f | 192 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 193 | |
91b0abe3 JW |
194 | page_cache_tree_delete(mapping, page, shadow); |
195 | ||
1da177e4 | 196 | page->mapping = NULL; |
b85e0eff | 197 | /* Leave page->index set: truncation lookup relies upon it */ |
91b0abe3 | 198 | |
4165b9b4 MH |
199 | /* hugetlb pages do not participate in page cache accounting. */ |
200 | if (!PageHuge(page)) | |
201 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
4b02108a KM |
202 | if (PageSwapBacked(page)) |
203 | __dec_zone_page_state(page, NR_SHMEM); | |
45426812 | 204 | BUG_ON(page_mapped(page)); |
3a692790 LT |
205 | |
206 | /* | |
b9ea2515 KK |
207 | * At this point page must be either written or cleaned by truncate. |
208 | * Dirty page here signals a bug and loss of unwritten data. | |
3a692790 | 209 | * |
b9ea2515 KK |
210 | * This fixes dirty accounting after removing the page entirely but |
211 | * leaves PageDirty set: it has no effect for truncated page and | |
212 | * anyway will be cleared before returning page into buddy allocator. | |
3a692790 | 213 | */ |
b9ea2515 KK |
214 | if (WARN_ON_ONCE(PageDirty(page))) |
215 | account_page_cleaned(page, mapping); | |
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); |
6072d13c LT |
237 | |
238 | if (freepage) | |
239 | freepage(page); | |
97cecb5a MK |
240 | page_cache_release(page); |
241 | } | |
242 | EXPORT_SYMBOL(delete_from_page_cache); | |
243 | ||
865ffef3 DM |
244 | static int filemap_check_errors(struct address_space *mapping) |
245 | { | |
246 | int ret = 0; | |
247 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
248 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
249 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 250 | ret = -ENOSPC; |
7fcbbaf1 JA |
251 | if (test_bit(AS_EIO, &mapping->flags) && |
252 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
253 | ret = -EIO; |
254 | return ret; | |
255 | } | |
256 | ||
1da177e4 | 257 | /** |
485bb99b | 258 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
259 | * @mapping: address space structure to write |
260 | * @start: offset in bytes where the range starts | |
469eb4d0 | 261 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 262 | * @sync_mode: enable synchronous operation |
1da177e4 | 263 | * |
485bb99b RD |
264 | * Start writeback against all of a mapping's dirty pages that lie |
265 | * within the byte offsets <start, end> inclusive. | |
266 | * | |
1da177e4 | 267 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 268 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
269 | * these two operations is that if a dirty page/buffer is encountered, it must |
270 | * be waited upon, and not just skipped over. | |
271 | */ | |
ebcf28e1 AM |
272 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
273 | loff_t end, int sync_mode) | |
1da177e4 LT |
274 | { |
275 | int ret; | |
276 | struct writeback_control wbc = { | |
277 | .sync_mode = sync_mode, | |
05fe478d | 278 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
279 | .range_start = start, |
280 | .range_end = end, | |
1da177e4 LT |
281 | }; |
282 | ||
283 | if (!mapping_cap_writeback_dirty(mapping)) | |
284 | return 0; | |
285 | ||
286 | ret = do_writepages(mapping, &wbc); | |
287 | return ret; | |
288 | } | |
289 | ||
290 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
291 | int sync_mode) | |
292 | { | |
111ebb6e | 293 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
294 | } |
295 | ||
296 | int filemap_fdatawrite(struct address_space *mapping) | |
297 | { | |
298 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
299 | } | |
300 | EXPORT_SYMBOL(filemap_fdatawrite); | |
301 | ||
f4c0a0fd | 302 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 303 | loff_t end) |
1da177e4 LT |
304 | { |
305 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
306 | } | |
f4c0a0fd | 307 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 308 | |
485bb99b RD |
309 | /** |
310 | * filemap_flush - mostly a non-blocking flush | |
311 | * @mapping: target address_space | |
312 | * | |
1da177e4 LT |
313 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
314 | * purposes - I/O may not be started against all dirty pages. | |
315 | */ | |
316 | int filemap_flush(struct address_space *mapping) | |
317 | { | |
318 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
319 | } | |
320 | EXPORT_SYMBOL(filemap_flush); | |
321 | ||
485bb99b | 322 | /** |
94004ed7 CH |
323 | * filemap_fdatawait_range - wait for writeback to complete |
324 | * @mapping: address space structure to wait for | |
325 | * @start_byte: offset in bytes where the range starts | |
326 | * @end_byte: offset in bytes where the range ends (inclusive) | |
485bb99b | 327 | * |
94004ed7 CH |
328 | * Walk the list of under-writeback pages of the given address space |
329 | * in the given range and wait for all of them. | |
1da177e4 | 330 | */ |
94004ed7 CH |
331 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
332 | loff_t end_byte) | |
1da177e4 | 333 | { |
94004ed7 CH |
334 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
335 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
336 | struct pagevec pvec; |
337 | int nr_pages; | |
865ffef3 | 338 | int ret2, ret = 0; |
1da177e4 | 339 | |
94004ed7 | 340 | if (end_byte < start_byte) |
865ffef3 | 341 | goto out; |
1da177e4 LT |
342 | |
343 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
344 | while ((index <= end) && |
345 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
346 | PAGECACHE_TAG_WRITEBACK, | |
347 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
348 | unsigned i; | |
349 | ||
350 | for (i = 0; i < nr_pages; i++) { | |
351 | struct page *page = pvec.pages[i]; | |
352 | ||
353 | /* until radix tree lookup accepts end_index */ | |
354 | if (page->index > end) | |
355 | continue; | |
356 | ||
357 | wait_on_page_writeback(page); | |
212260aa | 358 | if (TestClearPageError(page)) |
1da177e4 LT |
359 | ret = -EIO; |
360 | } | |
361 | pagevec_release(&pvec); | |
362 | cond_resched(); | |
363 | } | |
865ffef3 DM |
364 | out: |
365 | ret2 = filemap_check_errors(mapping); | |
366 | if (!ret) | |
367 | ret = ret2; | |
1da177e4 LT |
368 | |
369 | return ret; | |
370 | } | |
d3bccb6f JK |
371 | EXPORT_SYMBOL(filemap_fdatawait_range); |
372 | ||
1da177e4 | 373 | /** |
485bb99b | 374 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 375 | * @mapping: address space structure to wait for |
485bb99b RD |
376 | * |
377 | * Walk the list of under-writeback pages of the given address space | |
378 | * and wait for all of them. | |
1da177e4 LT |
379 | */ |
380 | int filemap_fdatawait(struct address_space *mapping) | |
381 | { | |
382 | loff_t i_size = i_size_read(mapping->host); | |
383 | ||
384 | if (i_size == 0) | |
385 | return 0; | |
386 | ||
94004ed7 | 387 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
388 | } |
389 | EXPORT_SYMBOL(filemap_fdatawait); | |
390 | ||
391 | int filemap_write_and_wait(struct address_space *mapping) | |
392 | { | |
28fd1298 | 393 | int err = 0; |
1da177e4 LT |
394 | |
395 | if (mapping->nrpages) { | |
28fd1298 OH |
396 | err = filemap_fdatawrite(mapping); |
397 | /* | |
398 | * Even if the above returned error, the pages may be | |
399 | * written partially (e.g. -ENOSPC), so we wait for it. | |
400 | * But the -EIO is special case, it may indicate the worst | |
401 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
402 | */ | |
403 | if (err != -EIO) { | |
404 | int err2 = filemap_fdatawait(mapping); | |
405 | if (!err) | |
406 | err = err2; | |
407 | } | |
865ffef3 DM |
408 | } else { |
409 | err = filemap_check_errors(mapping); | |
1da177e4 | 410 | } |
28fd1298 | 411 | return err; |
1da177e4 | 412 | } |
28fd1298 | 413 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 414 | |
485bb99b RD |
415 | /** |
416 | * filemap_write_and_wait_range - write out & wait on a file range | |
417 | * @mapping: the address_space for the pages | |
418 | * @lstart: offset in bytes where the range starts | |
419 | * @lend: offset in bytes where the range ends (inclusive) | |
420 | * | |
469eb4d0 AM |
421 | * Write out and wait upon file offsets lstart->lend, inclusive. |
422 | * | |
423 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
424 | * that this function can be used to write to the very end-of-file (end = -1). | |
425 | */ | |
1da177e4 LT |
426 | int filemap_write_and_wait_range(struct address_space *mapping, |
427 | loff_t lstart, loff_t lend) | |
428 | { | |
28fd1298 | 429 | int err = 0; |
1da177e4 LT |
430 | |
431 | if (mapping->nrpages) { | |
28fd1298 OH |
432 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
433 | WB_SYNC_ALL); | |
434 | /* See comment of filemap_write_and_wait() */ | |
435 | if (err != -EIO) { | |
94004ed7 CH |
436 | int err2 = filemap_fdatawait_range(mapping, |
437 | lstart, lend); | |
28fd1298 OH |
438 | if (!err) |
439 | err = err2; | |
440 | } | |
865ffef3 DM |
441 | } else { |
442 | err = filemap_check_errors(mapping); | |
1da177e4 | 443 | } |
28fd1298 | 444 | return err; |
1da177e4 | 445 | } |
f6995585 | 446 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 447 | |
ef6a3c63 MS |
448 | /** |
449 | * replace_page_cache_page - replace a pagecache page with a new one | |
450 | * @old: page to be replaced | |
451 | * @new: page to replace with | |
452 | * @gfp_mask: allocation mode | |
453 | * | |
454 | * This function replaces a page in the pagecache with a new one. On | |
455 | * success it acquires the pagecache reference for the new page and | |
456 | * drops it for the old page. Both the old and new pages must be | |
457 | * locked. This function does not add the new page to the LRU, the | |
458 | * caller must do that. | |
459 | * | |
460 | * The remove + add is atomic. The only way this function can fail is | |
461 | * memory allocation failure. | |
462 | */ | |
463 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
464 | { | |
465 | int error; | |
ef6a3c63 | 466 | |
309381fe SL |
467 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
468 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
469 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 470 | |
ef6a3c63 MS |
471 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
472 | if (!error) { | |
473 | struct address_space *mapping = old->mapping; | |
474 | void (*freepage)(struct page *); | |
475 | ||
476 | pgoff_t offset = old->index; | |
477 | freepage = mapping->a_ops->freepage; | |
478 | ||
479 | page_cache_get(new); | |
480 | new->mapping = mapping; | |
481 | new->index = offset; | |
482 | ||
483 | spin_lock_irq(&mapping->tree_lock); | |
91b0abe3 | 484 | __delete_from_page_cache(old, NULL); |
ef6a3c63 MS |
485 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
486 | BUG_ON(error); | |
487 | mapping->nrpages++; | |
4165b9b4 MH |
488 | |
489 | /* | |
490 | * hugetlb pages do not participate in page cache accounting. | |
491 | */ | |
492 | if (!PageHuge(new)) | |
493 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
ef6a3c63 MS |
494 | if (PageSwapBacked(new)) |
495 | __inc_zone_page_state(new, NR_SHMEM); | |
496 | spin_unlock_irq(&mapping->tree_lock); | |
0a31bc97 | 497 | mem_cgroup_migrate(old, new, true); |
ef6a3c63 MS |
498 | radix_tree_preload_end(); |
499 | if (freepage) | |
500 | freepage(old); | |
501 | page_cache_release(old); | |
ef6a3c63 MS |
502 | } |
503 | ||
504 | return error; | |
505 | } | |
506 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
507 | ||
0cd6144a | 508 | static int page_cache_tree_insert(struct address_space *mapping, |
a528910e | 509 | struct page *page, void **shadowp) |
0cd6144a | 510 | { |
449dd698 | 511 | struct radix_tree_node *node; |
0cd6144a JW |
512 | void **slot; |
513 | int error; | |
514 | ||
449dd698 JW |
515 | error = __radix_tree_create(&mapping->page_tree, page->index, |
516 | &node, &slot); | |
517 | if (error) | |
518 | return error; | |
519 | if (*slot) { | |
0cd6144a JW |
520 | void *p; |
521 | ||
522 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
523 | if (!radix_tree_exceptional_entry(p)) | |
524 | return -EEXIST; | |
a528910e JW |
525 | if (shadowp) |
526 | *shadowp = p; | |
449dd698 JW |
527 | mapping->nrshadows--; |
528 | if (node) | |
529 | workingset_node_shadows_dec(node); | |
0cd6144a | 530 | } |
449dd698 JW |
531 | radix_tree_replace_slot(slot, page); |
532 | mapping->nrpages++; | |
533 | if (node) { | |
534 | workingset_node_pages_inc(node); | |
535 | /* | |
536 | * Don't track node that contains actual pages. | |
537 | * | |
538 | * Avoid acquiring the list_lru lock if already | |
539 | * untracked. The list_empty() test is safe as | |
540 | * node->private_list is protected by | |
541 | * mapping->tree_lock. | |
542 | */ | |
543 | if (!list_empty(&node->private_list)) | |
544 | list_lru_del(&workingset_shadow_nodes, | |
545 | &node->private_list); | |
546 | } | |
547 | return 0; | |
0cd6144a JW |
548 | } |
549 | ||
a528910e JW |
550 | static int __add_to_page_cache_locked(struct page *page, |
551 | struct address_space *mapping, | |
552 | pgoff_t offset, gfp_t gfp_mask, | |
553 | void **shadowp) | |
1da177e4 | 554 | { |
00501b53 JW |
555 | int huge = PageHuge(page); |
556 | struct mem_cgroup *memcg; | |
e286781d NP |
557 | int error; |
558 | ||
309381fe SL |
559 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
560 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d | 561 | |
00501b53 JW |
562 | if (!huge) { |
563 | error = mem_cgroup_try_charge(page, current->mm, | |
564 | gfp_mask, &memcg); | |
565 | if (error) | |
566 | return error; | |
567 | } | |
1da177e4 | 568 | |
5e4c0d97 | 569 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 570 | if (error) { |
00501b53 JW |
571 | if (!huge) |
572 | mem_cgroup_cancel_charge(page, memcg); | |
66a0c8ee KS |
573 | return error; |
574 | } | |
575 | ||
576 | page_cache_get(page); | |
577 | page->mapping = mapping; | |
578 | page->index = offset; | |
579 | ||
580 | spin_lock_irq(&mapping->tree_lock); | |
a528910e | 581 | error = page_cache_tree_insert(mapping, page, shadowp); |
66a0c8ee KS |
582 | radix_tree_preload_end(); |
583 | if (unlikely(error)) | |
584 | goto err_insert; | |
4165b9b4 MH |
585 | |
586 | /* hugetlb pages do not participate in page cache accounting. */ | |
587 | if (!huge) | |
588 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
66a0c8ee | 589 | spin_unlock_irq(&mapping->tree_lock); |
00501b53 JW |
590 | if (!huge) |
591 | mem_cgroup_commit_charge(page, memcg, false); | |
66a0c8ee KS |
592 | trace_mm_filemap_add_to_page_cache(page); |
593 | return 0; | |
594 | err_insert: | |
595 | page->mapping = NULL; | |
596 | /* Leave page->index set: truncation relies upon it */ | |
597 | spin_unlock_irq(&mapping->tree_lock); | |
00501b53 JW |
598 | if (!huge) |
599 | mem_cgroup_cancel_charge(page, memcg); | |
66a0c8ee | 600 | page_cache_release(page); |
1da177e4 LT |
601 | return error; |
602 | } | |
a528910e JW |
603 | |
604 | /** | |
605 | * add_to_page_cache_locked - add a locked page to the pagecache | |
606 | * @page: page to add | |
607 | * @mapping: the page's address_space | |
608 | * @offset: page index | |
609 | * @gfp_mask: page allocation mode | |
610 | * | |
611 | * This function is used to add a page to the pagecache. It must be locked. | |
612 | * This function does not add the page to the LRU. The caller must do that. | |
613 | */ | |
614 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
615 | pgoff_t offset, gfp_t gfp_mask) | |
616 | { | |
617 | return __add_to_page_cache_locked(page, mapping, offset, | |
618 | gfp_mask, NULL); | |
619 | } | |
e286781d | 620 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
621 | |
622 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 623 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 624 | { |
a528910e | 625 | void *shadow = NULL; |
4f98a2fe RR |
626 | int ret; |
627 | ||
a528910e JW |
628 | __set_page_locked(page); |
629 | ret = __add_to_page_cache_locked(page, mapping, offset, | |
630 | gfp_mask, &shadow); | |
631 | if (unlikely(ret)) | |
632 | __clear_page_locked(page); | |
633 | else { | |
634 | /* | |
635 | * The page might have been evicted from cache only | |
636 | * recently, in which case it should be activated like | |
637 | * any other repeatedly accessed page. | |
638 | */ | |
639 | if (shadow && workingset_refault(shadow)) { | |
640 | SetPageActive(page); | |
641 | workingset_activation(page); | |
642 | } else | |
643 | ClearPageActive(page); | |
644 | lru_cache_add(page); | |
645 | } | |
1da177e4 LT |
646 | return ret; |
647 | } | |
18bc0bbd | 648 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 649 | |
44110fe3 | 650 | #ifdef CONFIG_NUMA |
2ae88149 | 651 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 652 | { |
c0ff7453 MX |
653 | int n; |
654 | struct page *page; | |
655 | ||
44110fe3 | 656 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
657 | unsigned int cpuset_mems_cookie; |
658 | do { | |
d26914d1 | 659 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 MG |
660 | n = cpuset_mem_spread_node(); |
661 | page = alloc_pages_exact_node(n, gfp, 0); | |
d26914d1 | 662 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 663 | |
c0ff7453 | 664 | return page; |
44110fe3 | 665 | } |
2ae88149 | 666 | return alloc_pages(gfp, 0); |
44110fe3 | 667 | } |
2ae88149 | 668 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
669 | #endif |
670 | ||
1da177e4 LT |
671 | /* |
672 | * In order to wait for pages to become available there must be | |
673 | * waitqueues associated with pages. By using a hash table of | |
674 | * waitqueues where the bucket discipline is to maintain all | |
675 | * waiters on the same queue and wake all when any of the pages | |
676 | * become available, and for the woken contexts to check to be | |
677 | * sure the appropriate page became available, this saves space | |
678 | * at a cost of "thundering herd" phenomena during rare hash | |
679 | * collisions. | |
680 | */ | |
a4796e37 | 681 | wait_queue_head_t *page_waitqueue(struct page *page) |
1da177e4 LT |
682 | { |
683 | const struct zone *zone = page_zone(page); | |
684 | ||
685 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
686 | } | |
a4796e37 | 687 | EXPORT_SYMBOL(page_waitqueue); |
1da177e4 | 688 | |
920c7a5d | 689 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
690 | { |
691 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
692 | ||
693 | if (test_bit(bit_nr, &page->flags)) | |
74316201 | 694 | __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
695 | TASK_UNINTERRUPTIBLE); |
696 | } | |
697 | EXPORT_SYMBOL(wait_on_page_bit); | |
698 | ||
f62e00cc KM |
699 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
700 | { | |
701 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
702 | ||
703 | if (!test_bit(bit_nr, &page->flags)) | |
704 | return 0; | |
705 | ||
706 | return __wait_on_bit(page_waitqueue(page), &wait, | |
74316201 | 707 | bit_wait_io, TASK_KILLABLE); |
f62e00cc KM |
708 | } |
709 | ||
cbbce822 N |
710 | int wait_on_page_bit_killable_timeout(struct page *page, |
711 | int bit_nr, unsigned long timeout) | |
712 | { | |
713 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
714 | ||
715 | wait.key.timeout = jiffies + timeout; | |
716 | if (!test_bit(bit_nr, &page->flags)) | |
717 | return 0; | |
718 | return __wait_on_bit(page_waitqueue(page), &wait, | |
719 | bit_wait_io_timeout, TASK_KILLABLE); | |
720 | } | |
721 | EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout); | |
722 | ||
385e1ca5 DH |
723 | /** |
724 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
725 | * @page: Page defining the wait queue of interest |
726 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
727 | * |
728 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
729 | */ | |
730 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
731 | { | |
732 | wait_queue_head_t *q = page_waitqueue(page); | |
733 | unsigned long flags; | |
734 | ||
735 | spin_lock_irqsave(&q->lock, flags); | |
736 | __add_wait_queue(q, waiter); | |
737 | spin_unlock_irqrestore(&q->lock, flags); | |
738 | } | |
739 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
740 | ||
1da177e4 | 741 | /** |
485bb99b | 742 | * unlock_page - unlock a locked page |
1da177e4 LT |
743 | * @page: the page |
744 | * | |
745 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
746 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 747 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
748 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
749 | * | |
8413ac9d NP |
750 | * The mb is necessary to enforce ordering between the clear_bit and the read |
751 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 752 | */ |
920c7a5d | 753 | void unlock_page(struct page *page) |
1da177e4 | 754 | { |
309381fe | 755 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
8413ac9d | 756 | clear_bit_unlock(PG_locked, &page->flags); |
4e857c58 | 757 | smp_mb__after_atomic(); |
1da177e4 LT |
758 | wake_up_page(page, PG_locked); |
759 | } | |
760 | EXPORT_SYMBOL(unlock_page); | |
761 | ||
485bb99b RD |
762 | /** |
763 | * end_page_writeback - end writeback against a page | |
764 | * @page: the page | |
1da177e4 LT |
765 | */ |
766 | void end_page_writeback(struct page *page) | |
767 | { | |
888cf2db MG |
768 | /* |
769 | * TestClearPageReclaim could be used here but it is an atomic | |
770 | * operation and overkill in this particular case. Failing to | |
771 | * shuffle a page marked for immediate reclaim is too mild to | |
772 | * justify taking an atomic operation penalty at the end of | |
773 | * ever page writeback. | |
774 | */ | |
775 | if (PageReclaim(page)) { | |
776 | ClearPageReclaim(page); | |
ac6aadb2 | 777 | rotate_reclaimable_page(page); |
888cf2db | 778 | } |
ac6aadb2 MS |
779 | |
780 | if (!test_clear_page_writeback(page)) | |
781 | BUG(); | |
782 | ||
4e857c58 | 783 | smp_mb__after_atomic(); |
1da177e4 LT |
784 | wake_up_page(page, PG_writeback); |
785 | } | |
786 | EXPORT_SYMBOL(end_page_writeback); | |
787 | ||
57d99845 MW |
788 | /* |
789 | * After completing I/O on a page, call this routine to update the page | |
790 | * flags appropriately | |
791 | */ | |
792 | void page_endio(struct page *page, int rw, int err) | |
793 | { | |
794 | if (rw == READ) { | |
795 | if (!err) { | |
796 | SetPageUptodate(page); | |
797 | } else { | |
798 | ClearPageUptodate(page); | |
799 | SetPageError(page); | |
800 | } | |
801 | unlock_page(page); | |
802 | } else { /* rw == WRITE */ | |
803 | if (err) { | |
804 | SetPageError(page); | |
805 | if (page->mapping) | |
806 | mapping_set_error(page->mapping, err); | |
807 | } | |
808 | end_page_writeback(page); | |
809 | } | |
810 | } | |
811 | EXPORT_SYMBOL_GPL(page_endio); | |
812 | ||
485bb99b RD |
813 | /** |
814 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
815 | * @page: the page to lock | |
1da177e4 | 816 | */ |
920c7a5d | 817 | void __lock_page(struct page *page) |
1da177e4 LT |
818 | { |
819 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
820 | ||
74316201 | 821 | __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
822 | TASK_UNINTERRUPTIBLE); |
823 | } | |
824 | EXPORT_SYMBOL(__lock_page); | |
825 | ||
b5606c2d | 826 | int __lock_page_killable(struct page *page) |
2687a356 MW |
827 | { |
828 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
829 | ||
830 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
74316201 | 831 | bit_wait_io, TASK_KILLABLE); |
2687a356 | 832 | } |
18bc0bbd | 833 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 834 | |
9a95f3cf PC |
835 | /* |
836 | * Return values: | |
837 | * 1 - page is locked; mmap_sem is still held. | |
838 | * 0 - page is not locked. | |
839 | * mmap_sem has been released (up_read()), unless flags had both | |
840 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
841 | * which case mmap_sem is still held. | |
842 | * | |
843 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
844 | * with the page locked and the mmap_sem unperturbed. | |
845 | */ | |
d065bd81 ML |
846 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
847 | unsigned int flags) | |
848 | { | |
37b23e05 KM |
849 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
850 | /* | |
851 | * CAUTION! In this case, mmap_sem is not released | |
852 | * even though return 0. | |
853 | */ | |
854 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
855 | return 0; | |
856 | ||
857 | up_read(&mm->mmap_sem); | |
858 | if (flags & FAULT_FLAG_KILLABLE) | |
859 | wait_on_page_locked_killable(page); | |
860 | else | |
318b275f | 861 | wait_on_page_locked(page); |
d065bd81 | 862 | return 0; |
37b23e05 KM |
863 | } else { |
864 | if (flags & FAULT_FLAG_KILLABLE) { | |
865 | int ret; | |
866 | ||
867 | ret = __lock_page_killable(page); | |
868 | if (ret) { | |
869 | up_read(&mm->mmap_sem); | |
870 | return 0; | |
871 | } | |
872 | } else | |
873 | __lock_page(page); | |
874 | return 1; | |
d065bd81 ML |
875 | } |
876 | } | |
877 | ||
e7b563bb JW |
878 | /** |
879 | * page_cache_next_hole - find the next hole (not-present entry) | |
880 | * @mapping: mapping | |
881 | * @index: index | |
882 | * @max_scan: maximum range to search | |
883 | * | |
884 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
885 | * lowest indexed hole. | |
886 | * | |
887 | * Returns: the index of the hole if found, otherwise returns an index | |
888 | * outside of the set specified (in which case 'return - index >= | |
889 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
890 | * be returned. | |
891 | * | |
892 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
893 | * like radix_tree_gang_lookup, this will not atomically search a | |
894 | * snapshot of the tree at a single point in time. For example, if a | |
895 | * hole is created at index 5, then subsequently a hole is created at | |
896 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
897 | * if called under rcu_read_lock. | |
898 | */ | |
899 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
900 | pgoff_t index, unsigned long max_scan) | |
901 | { | |
902 | unsigned long i; | |
903 | ||
904 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
905 | struct page *page; |
906 | ||
907 | page = radix_tree_lookup(&mapping->page_tree, index); | |
908 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
909 | break; |
910 | index++; | |
911 | if (index == 0) | |
912 | break; | |
913 | } | |
914 | ||
915 | return index; | |
916 | } | |
917 | EXPORT_SYMBOL(page_cache_next_hole); | |
918 | ||
919 | /** | |
920 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
921 | * @mapping: mapping | |
922 | * @index: index | |
923 | * @max_scan: maximum range to search | |
924 | * | |
925 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
926 | * the first hole. | |
927 | * | |
928 | * Returns: the index of the hole if found, otherwise returns an index | |
929 | * outside of the set specified (in which case 'index - return >= | |
930 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
931 | * will be returned. | |
932 | * | |
933 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
934 | * like radix_tree_gang_lookup, this will not atomically search a | |
935 | * snapshot of the tree at a single point in time. For example, if a | |
936 | * hole is created at index 10, then subsequently a hole is created at | |
937 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
938 | * called under rcu_read_lock. | |
939 | */ | |
940 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
941 | pgoff_t index, unsigned long max_scan) | |
942 | { | |
943 | unsigned long i; | |
944 | ||
945 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
946 | struct page *page; |
947 | ||
948 | page = radix_tree_lookup(&mapping->page_tree, index); | |
949 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
950 | break; |
951 | index--; | |
952 | if (index == ULONG_MAX) | |
953 | break; | |
954 | } | |
955 | ||
956 | return index; | |
957 | } | |
958 | EXPORT_SYMBOL(page_cache_prev_hole); | |
959 | ||
485bb99b | 960 | /** |
0cd6144a | 961 | * find_get_entry - find and get a page cache entry |
485bb99b | 962 | * @mapping: the address_space to search |
0cd6144a JW |
963 | * @offset: the page cache index |
964 | * | |
965 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
966 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 967 | * |
139b6a6f JW |
968 | * If the slot holds a shadow entry of a previously evicted page, or a |
969 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
970 | * |
971 | * Otherwise, %NULL is returned. | |
1da177e4 | 972 | */ |
0cd6144a | 973 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 974 | { |
a60637c8 | 975 | void **pagep; |
1da177e4 LT |
976 | struct page *page; |
977 | ||
a60637c8 NP |
978 | rcu_read_lock(); |
979 | repeat: | |
980 | page = NULL; | |
981 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
982 | if (pagep) { | |
983 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
984 | if (unlikely(!page)) |
985 | goto out; | |
a2c16d6c | 986 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
987 | if (radix_tree_deref_retry(page)) |
988 | goto repeat; | |
989 | /* | |
139b6a6f JW |
990 | * A shadow entry of a recently evicted page, |
991 | * or a swap entry from shmem/tmpfs. Return | |
992 | * it without attempting to raise page count. | |
8079b1c8 HD |
993 | */ |
994 | goto out; | |
a2c16d6c | 995 | } |
a60637c8 NP |
996 | if (!page_cache_get_speculative(page)) |
997 | goto repeat; | |
998 | ||
999 | /* | |
1000 | * Has the page moved? | |
1001 | * This is part of the lockless pagecache protocol. See | |
1002 | * include/linux/pagemap.h for details. | |
1003 | */ | |
1004 | if (unlikely(page != *pagep)) { | |
1005 | page_cache_release(page); | |
1006 | goto repeat; | |
1007 | } | |
1008 | } | |
27d20fdd | 1009 | out: |
a60637c8 NP |
1010 | rcu_read_unlock(); |
1011 | ||
1da177e4 LT |
1012 | return page; |
1013 | } | |
0cd6144a | 1014 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1015 | |
0cd6144a JW |
1016 | /** |
1017 | * find_lock_entry - locate, pin and lock a page cache entry | |
1018 | * @mapping: the address_space to search | |
1019 | * @offset: the page cache index | |
1020 | * | |
1021 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1022 | * page cache page, it is returned locked and with an increased | |
1023 | * refcount. | |
1024 | * | |
139b6a6f JW |
1025 | * If the slot holds a shadow entry of a previously evicted page, or a |
1026 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1027 | * |
1028 | * Otherwise, %NULL is returned. | |
1029 | * | |
1030 | * find_lock_entry() may sleep. | |
1031 | */ | |
1032 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1033 | { |
1034 | struct page *page; | |
1035 | ||
1da177e4 | 1036 | repeat: |
0cd6144a | 1037 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1038 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1039 | lock_page(page); |
1040 | /* Has the page been truncated? */ | |
1041 | if (unlikely(page->mapping != mapping)) { | |
1042 | unlock_page(page); | |
1043 | page_cache_release(page); | |
1044 | goto repeat; | |
1da177e4 | 1045 | } |
309381fe | 1046 | VM_BUG_ON_PAGE(page->index != offset, page); |
1da177e4 | 1047 | } |
1da177e4 LT |
1048 | return page; |
1049 | } | |
0cd6144a JW |
1050 | EXPORT_SYMBOL(find_lock_entry); |
1051 | ||
1052 | /** | |
2457aec6 | 1053 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1054 | * @mapping: the address_space to search |
1055 | * @offset: the page index | |
2457aec6 | 1056 | * @fgp_flags: PCG flags |
45f87de5 | 1057 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
0cd6144a | 1058 | * |
2457aec6 | 1059 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1060 | * |
75325189 | 1061 | * PCG flags modify how the page is returned. |
0cd6144a | 1062 | * |
2457aec6 MG |
1063 | * FGP_ACCESSED: the page will be marked accessed |
1064 | * FGP_LOCK: Page is return locked | |
1065 | * FGP_CREAT: If page is not present then a new page is allocated using | |
45f87de5 MH |
1066 | * @gfp_mask and added to the page cache and the VM's LRU |
1067 | * list. The page is returned locked and with an increased | |
1068 | * refcount. Otherwise, %NULL is returned. | |
1da177e4 | 1069 | * |
2457aec6 MG |
1070 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1071 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1072 | * |
2457aec6 | 1073 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1074 | */ |
2457aec6 | 1075 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
45f87de5 | 1076 | int fgp_flags, gfp_t gfp_mask) |
1da177e4 | 1077 | { |
eb2be189 | 1078 | struct page *page; |
2457aec6 | 1079 | |
1da177e4 | 1080 | repeat: |
2457aec6 MG |
1081 | page = find_get_entry(mapping, offset); |
1082 | if (radix_tree_exceptional_entry(page)) | |
1083 | page = NULL; | |
1084 | if (!page) | |
1085 | goto no_page; | |
1086 | ||
1087 | if (fgp_flags & FGP_LOCK) { | |
1088 | if (fgp_flags & FGP_NOWAIT) { | |
1089 | if (!trylock_page(page)) { | |
1090 | page_cache_release(page); | |
1091 | return NULL; | |
1092 | } | |
1093 | } else { | |
1094 | lock_page(page); | |
1095 | } | |
1096 | ||
1097 | /* Has the page been truncated? */ | |
1098 | if (unlikely(page->mapping != mapping)) { | |
1099 | unlock_page(page); | |
1100 | page_cache_release(page); | |
1101 | goto repeat; | |
1102 | } | |
1103 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1104 | } | |
1105 | ||
1106 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1107 | mark_page_accessed(page); | |
1108 | ||
1109 | no_page: | |
1110 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1111 | int err; | |
1112 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1113 | gfp_mask |= __GFP_WRITE; |
1114 | if (fgp_flags & FGP_NOFS) | |
1115 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1116 | |
45f87de5 | 1117 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1118 | if (!page) |
1119 | return NULL; | |
2457aec6 MG |
1120 | |
1121 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1122 | fgp_flags |= FGP_LOCK; | |
1123 | ||
eb39d618 | 1124 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1125 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1126 | __SetPageReferenced(page); |
2457aec6 | 1127 | |
45f87de5 MH |
1128 | err = add_to_page_cache_lru(page, mapping, offset, |
1129 | gfp_mask & GFP_RECLAIM_MASK); | |
eb2be189 NP |
1130 | if (unlikely(err)) { |
1131 | page_cache_release(page); | |
1132 | page = NULL; | |
1133 | if (err == -EEXIST) | |
1134 | goto repeat; | |
1da177e4 | 1135 | } |
1da177e4 | 1136 | } |
2457aec6 | 1137 | |
1da177e4 LT |
1138 | return page; |
1139 | } | |
2457aec6 | 1140 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1141 | |
0cd6144a JW |
1142 | /** |
1143 | * find_get_entries - gang pagecache lookup | |
1144 | * @mapping: The address_space to search | |
1145 | * @start: The starting page cache index | |
1146 | * @nr_entries: The maximum number of entries | |
1147 | * @entries: Where the resulting entries are placed | |
1148 | * @indices: The cache indices corresponding to the entries in @entries | |
1149 | * | |
1150 | * find_get_entries() will search for and return a group of up to | |
1151 | * @nr_entries entries in the mapping. The entries are placed at | |
1152 | * @entries. find_get_entries() takes a reference against any actual | |
1153 | * pages it returns. | |
1154 | * | |
1155 | * The search returns a group of mapping-contiguous page cache entries | |
1156 | * with ascending indexes. There may be holes in the indices due to | |
1157 | * not-present pages. | |
1158 | * | |
139b6a6f JW |
1159 | * Any shadow entries of evicted pages, or swap entries from |
1160 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1161 | * |
1162 | * find_get_entries() returns the number of pages and shadow entries | |
1163 | * which were found. | |
1164 | */ | |
1165 | unsigned find_get_entries(struct address_space *mapping, | |
1166 | pgoff_t start, unsigned int nr_entries, | |
1167 | struct page **entries, pgoff_t *indices) | |
1168 | { | |
1169 | void **slot; | |
1170 | unsigned int ret = 0; | |
1171 | struct radix_tree_iter iter; | |
1172 | ||
1173 | if (!nr_entries) | |
1174 | return 0; | |
1175 | ||
1176 | rcu_read_lock(); | |
1177 | restart: | |
1178 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { | |
1179 | struct page *page; | |
1180 | repeat: | |
1181 | page = radix_tree_deref_slot(slot); | |
1182 | if (unlikely(!page)) | |
1183 | continue; | |
1184 | if (radix_tree_exception(page)) { | |
1185 | if (radix_tree_deref_retry(page)) | |
1186 | goto restart; | |
1187 | /* | |
139b6a6f JW |
1188 | * A shadow entry of a recently evicted page, |
1189 | * or a swap entry from shmem/tmpfs. Return | |
1190 | * it without attempting to raise page count. | |
0cd6144a JW |
1191 | */ |
1192 | goto export; | |
1193 | } | |
1194 | if (!page_cache_get_speculative(page)) | |
1195 | goto repeat; | |
1196 | ||
1197 | /* Has the page moved? */ | |
1198 | if (unlikely(page != *slot)) { | |
1199 | page_cache_release(page); | |
1200 | goto repeat; | |
1201 | } | |
1202 | export: | |
1203 | indices[ret] = iter.index; | |
1204 | entries[ret] = page; | |
1205 | if (++ret == nr_entries) | |
1206 | break; | |
1207 | } | |
1208 | rcu_read_unlock(); | |
1209 | return ret; | |
1210 | } | |
1211 | ||
1da177e4 LT |
1212 | /** |
1213 | * find_get_pages - gang pagecache lookup | |
1214 | * @mapping: The address_space to search | |
1215 | * @start: The starting page index | |
1216 | * @nr_pages: The maximum number of pages | |
1217 | * @pages: Where the resulting pages are placed | |
1218 | * | |
1219 | * find_get_pages() will search for and return a group of up to | |
1220 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
1221 | * find_get_pages() takes a reference against the returned pages. | |
1222 | * | |
1223 | * The search returns a group of mapping-contiguous pages with ascending | |
1224 | * indexes. There may be holes in the indices due to not-present pages. | |
1225 | * | |
1226 | * find_get_pages() returns the number of pages which were found. | |
1227 | */ | |
1228 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
1229 | unsigned int nr_pages, struct page **pages) | |
1230 | { | |
0fc9d104 KK |
1231 | struct radix_tree_iter iter; |
1232 | void **slot; | |
1233 | unsigned ret = 0; | |
1234 | ||
1235 | if (unlikely(!nr_pages)) | |
1236 | return 0; | |
a60637c8 NP |
1237 | |
1238 | rcu_read_lock(); | |
1239 | restart: | |
0fc9d104 | 1240 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
a60637c8 NP |
1241 | struct page *page; |
1242 | repeat: | |
0fc9d104 | 1243 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1244 | if (unlikely(!page)) |
1245 | continue; | |
9d8aa4ea | 1246 | |
a2c16d6c | 1247 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1248 | if (radix_tree_deref_retry(page)) { |
1249 | /* | |
1250 | * Transient condition which can only trigger | |
1251 | * when entry at index 0 moves out of or back | |
1252 | * to root: none yet gotten, safe to restart. | |
1253 | */ | |
0fc9d104 | 1254 | WARN_ON(iter.index); |
8079b1c8 HD |
1255 | goto restart; |
1256 | } | |
a2c16d6c | 1257 | /* |
139b6a6f JW |
1258 | * A shadow entry of a recently evicted page, |
1259 | * or a swap entry from shmem/tmpfs. Skip | |
1260 | * over it. | |
a2c16d6c | 1261 | */ |
8079b1c8 | 1262 | continue; |
27d20fdd | 1263 | } |
a60637c8 NP |
1264 | |
1265 | if (!page_cache_get_speculative(page)) | |
1266 | goto repeat; | |
1267 | ||
1268 | /* Has the page moved? */ | |
0fc9d104 | 1269 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1270 | page_cache_release(page); |
1271 | goto repeat; | |
1272 | } | |
1da177e4 | 1273 | |
a60637c8 | 1274 | pages[ret] = page; |
0fc9d104 KK |
1275 | if (++ret == nr_pages) |
1276 | break; | |
a60637c8 | 1277 | } |
5b280c0c | 1278 | |
a60637c8 | 1279 | rcu_read_unlock(); |
1da177e4 LT |
1280 | return ret; |
1281 | } | |
1282 | ||
ebf43500 JA |
1283 | /** |
1284 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1285 | * @mapping: The address_space to search | |
1286 | * @index: The starting page index | |
1287 | * @nr_pages: The maximum number of pages | |
1288 | * @pages: Where the resulting pages are placed | |
1289 | * | |
1290 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1291 | * that the returned number of pages are guaranteed to be contiguous. | |
1292 | * | |
1293 | * find_get_pages_contig() returns the number of pages which were found. | |
1294 | */ | |
1295 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1296 | unsigned int nr_pages, struct page **pages) | |
1297 | { | |
0fc9d104 KK |
1298 | struct radix_tree_iter iter; |
1299 | void **slot; | |
1300 | unsigned int ret = 0; | |
1301 | ||
1302 | if (unlikely(!nr_pages)) | |
1303 | return 0; | |
a60637c8 NP |
1304 | |
1305 | rcu_read_lock(); | |
1306 | restart: | |
0fc9d104 | 1307 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
a60637c8 NP |
1308 | struct page *page; |
1309 | repeat: | |
0fc9d104 KK |
1310 | page = radix_tree_deref_slot(slot); |
1311 | /* The hole, there no reason to continue */ | |
a60637c8 | 1312 | if (unlikely(!page)) |
0fc9d104 | 1313 | break; |
9d8aa4ea | 1314 | |
a2c16d6c | 1315 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1316 | if (radix_tree_deref_retry(page)) { |
1317 | /* | |
1318 | * Transient condition which can only trigger | |
1319 | * when entry at index 0 moves out of or back | |
1320 | * to root: none yet gotten, safe to restart. | |
1321 | */ | |
1322 | goto restart; | |
1323 | } | |
a2c16d6c | 1324 | /* |
139b6a6f JW |
1325 | * A shadow entry of a recently evicted page, |
1326 | * or a swap entry from shmem/tmpfs. Stop | |
1327 | * looking for contiguous pages. | |
a2c16d6c | 1328 | */ |
8079b1c8 | 1329 | break; |
a2c16d6c | 1330 | } |
ebf43500 | 1331 | |
a60637c8 NP |
1332 | if (!page_cache_get_speculative(page)) |
1333 | goto repeat; | |
1334 | ||
1335 | /* Has the page moved? */ | |
0fc9d104 | 1336 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1337 | page_cache_release(page); |
1338 | goto repeat; | |
1339 | } | |
1340 | ||
9cbb4cb2 NP |
1341 | /* |
1342 | * must check mapping and index after taking the ref. | |
1343 | * otherwise we can get both false positives and false | |
1344 | * negatives, which is just confusing to the caller. | |
1345 | */ | |
0fc9d104 | 1346 | if (page->mapping == NULL || page->index != iter.index) { |
9cbb4cb2 NP |
1347 | page_cache_release(page); |
1348 | break; | |
1349 | } | |
1350 | ||
a60637c8 | 1351 | pages[ret] = page; |
0fc9d104 KK |
1352 | if (++ret == nr_pages) |
1353 | break; | |
ebf43500 | 1354 | } |
a60637c8 NP |
1355 | rcu_read_unlock(); |
1356 | return ret; | |
ebf43500 | 1357 | } |
ef71c15c | 1358 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1359 | |
485bb99b RD |
1360 | /** |
1361 | * find_get_pages_tag - find and return pages that match @tag | |
1362 | * @mapping: the address_space to search | |
1363 | * @index: the starting page index | |
1364 | * @tag: the tag index | |
1365 | * @nr_pages: the maximum number of pages | |
1366 | * @pages: where the resulting pages are placed | |
1367 | * | |
1da177e4 | 1368 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1369 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
1370 | */ |
1371 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
1372 | int tag, unsigned int nr_pages, struct page **pages) | |
1373 | { | |
0fc9d104 KK |
1374 | struct radix_tree_iter iter; |
1375 | void **slot; | |
1376 | unsigned ret = 0; | |
1377 | ||
1378 | if (unlikely(!nr_pages)) | |
1379 | return 0; | |
a60637c8 NP |
1380 | |
1381 | rcu_read_lock(); | |
1382 | restart: | |
0fc9d104 KK |
1383 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1384 | &iter, *index, tag) { | |
a60637c8 NP |
1385 | struct page *page; |
1386 | repeat: | |
0fc9d104 | 1387 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1388 | if (unlikely(!page)) |
1389 | continue; | |
9d8aa4ea | 1390 | |
a2c16d6c | 1391 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1392 | if (radix_tree_deref_retry(page)) { |
1393 | /* | |
1394 | * Transient condition which can only trigger | |
1395 | * when entry at index 0 moves out of or back | |
1396 | * to root: none yet gotten, safe to restart. | |
1397 | */ | |
1398 | goto restart; | |
1399 | } | |
a2c16d6c | 1400 | /* |
139b6a6f JW |
1401 | * A shadow entry of a recently evicted page. |
1402 | * | |
1403 | * Those entries should never be tagged, but | |
1404 | * this tree walk is lockless and the tags are | |
1405 | * looked up in bulk, one radix tree node at a | |
1406 | * time, so there is a sizable window for page | |
1407 | * reclaim to evict a page we saw tagged. | |
1408 | * | |
1409 | * Skip over it. | |
a2c16d6c | 1410 | */ |
139b6a6f | 1411 | continue; |
a2c16d6c | 1412 | } |
a60637c8 NP |
1413 | |
1414 | if (!page_cache_get_speculative(page)) | |
1415 | goto repeat; | |
1416 | ||
1417 | /* Has the page moved? */ | |
0fc9d104 | 1418 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1419 | page_cache_release(page); |
1420 | goto repeat; | |
1421 | } | |
1422 | ||
1423 | pages[ret] = page; | |
0fc9d104 KK |
1424 | if (++ret == nr_pages) |
1425 | break; | |
a60637c8 | 1426 | } |
5b280c0c | 1427 | |
a60637c8 | 1428 | rcu_read_unlock(); |
1da177e4 | 1429 | |
1da177e4 LT |
1430 | if (ret) |
1431 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1432 | |
1da177e4 LT |
1433 | return ret; |
1434 | } | |
ef71c15c | 1435 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1436 | |
76d42bd9 WF |
1437 | /* |
1438 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1439 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1440 | * | |
1441 | * ---R__________________________________________B__________ | |
1442 | * ^ reading here ^ bad block(assume 4k) | |
1443 | * | |
1444 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1445 | * => failing the whole request => read(R) => read(R+1) => | |
1446 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1447 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1448 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1449 | * | |
1450 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1451 | */ | |
1452 | static void shrink_readahead_size_eio(struct file *filp, | |
1453 | struct file_ra_state *ra) | |
1454 | { | |
76d42bd9 | 1455 | ra->ra_pages /= 4; |
76d42bd9 WF |
1456 | } |
1457 | ||
485bb99b | 1458 | /** |
36e78914 | 1459 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1460 | * @filp: the file to read |
1461 | * @ppos: current file position | |
6e58e79d AV |
1462 | * @iter: data destination |
1463 | * @written: already copied | |
485bb99b | 1464 | * |
1da177e4 | 1465 | * This is a generic file read routine, and uses the |
485bb99b | 1466 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1467 | * |
1468 | * This is really ugly. But the goto's actually try to clarify some | |
1469 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1470 | */ |
6e58e79d AV |
1471 | static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos, |
1472 | struct iov_iter *iter, ssize_t written) | |
1da177e4 | 1473 | { |
36e78914 | 1474 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1475 | struct inode *inode = mapping->host; |
36e78914 | 1476 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1477 | pgoff_t index; |
1478 | pgoff_t last_index; | |
1479 | pgoff_t prev_index; | |
1480 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1481 | unsigned int prev_offset; |
6e58e79d | 1482 | int error = 0; |
1da177e4 | 1483 | |
1da177e4 | 1484 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1485 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1486 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
6e58e79d | 1487 | last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1da177e4 LT |
1488 | offset = *ppos & ~PAGE_CACHE_MASK; |
1489 | ||
1da177e4 LT |
1490 | for (;;) { |
1491 | struct page *page; | |
57f6b96c | 1492 | pgoff_t end_index; |
a32ea1e1 | 1493 | loff_t isize; |
1da177e4 LT |
1494 | unsigned long nr, ret; |
1495 | ||
1da177e4 | 1496 | cond_resched(); |
1da177e4 LT |
1497 | find_page: |
1498 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1499 | if (!page) { |
cf914a7d | 1500 | page_cache_sync_readahead(mapping, |
7ff81078 | 1501 | ra, filp, |
3ea89ee8 FW |
1502 | index, last_index - index); |
1503 | page = find_get_page(mapping, index); | |
1504 | if (unlikely(page == NULL)) | |
1505 | goto no_cached_page; | |
1506 | } | |
1507 | if (PageReadahead(page)) { | |
cf914a7d | 1508 | page_cache_async_readahead(mapping, |
7ff81078 | 1509 | ra, filp, page, |
3ea89ee8 | 1510 | index, last_index - index); |
1da177e4 | 1511 | } |
8ab22b9a HH |
1512 | if (!PageUptodate(page)) { |
1513 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1514 | !mapping->a_ops->is_partially_uptodate) | |
1515 | goto page_not_up_to_date; | |
529ae9aa | 1516 | if (!trylock_page(page)) |
8ab22b9a | 1517 | goto page_not_up_to_date; |
8d056cb9 DH |
1518 | /* Did it get truncated before we got the lock? */ |
1519 | if (!page->mapping) | |
1520 | goto page_not_up_to_date_locked; | |
8ab22b9a | 1521 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 1522 | offset, iter->count)) |
8ab22b9a HH |
1523 | goto page_not_up_to_date_locked; |
1524 | unlock_page(page); | |
1525 | } | |
1da177e4 | 1526 | page_ok: |
a32ea1e1 N |
1527 | /* |
1528 | * i_size must be checked after we know the page is Uptodate. | |
1529 | * | |
1530 | * Checking i_size after the check allows us to calculate | |
1531 | * the correct value for "nr", which means the zero-filled | |
1532 | * part of the page is not copied back to userspace (unless | |
1533 | * another truncate extends the file - this is desired though). | |
1534 | */ | |
1535 | ||
1536 | isize = i_size_read(inode); | |
1537 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1538 | if (unlikely(!isize || index > end_index)) { | |
1539 | page_cache_release(page); | |
1540 | goto out; | |
1541 | } | |
1542 | ||
1543 | /* nr is the maximum number of bytes to copy from this page */ | |
1544 | nr = PAGE_CACHE_SIZE; | |
1545 | if (index == end_index) { | |
1546 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1547 | if (nr <= offset) { | |
1548 | page_cache_release(page); | |
1549 | goto out; | |
1550 | } | |
1551 | } | |
1552 | nr = nr - offset; | |
1da177e4 LT |
1553 | |
1554 | /* If users can be writing to this page using arbitrary | |
1555 | * virtual addresses, take care about potential aliasing | |
1556 | * before reading the page on the kernel side. | |
1557 | */ | |
1558 | if (mapping_writably_mapped(mapping)) | |
1559 | flush_dcache_page(page); | |
1560 | ||
1561 | /* | |
ec0f1637 JK |
1562 | * When a sequential read accesses a page several times, |
1563 | * only mark it as accessed the first time. | |
1da177e4 | 1564 | */ |
ec0f1637 | 1565 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1566 | mark_page_accessed(page); |
1567 | prev_index = index; | |
1568 | ||
1569 | /* | |
1570 | * Ok, we have the page, and it's up-to-date, so | |
1571 | * now we can copy it to user space... | |
1da177e4 | 1572 | */ |
6e58e79d AV |
1573 | |
1574 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 LT |
1575 | offset += ret; |
1576 | index += offset >> PAGE_CACHE_SHIFT; | |
1577 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1578 | prev_offset = offset; |
1da177e4 LT |
1579 | |
1580 | page_cache_release(page); | |
6e58e79d AV |
1581 | written += ret; |
1582 | if (!iov_iter_count(iter)) | |
1583 | goto out; | |
1584 | if (ret < nr) { | |
1585 | error = -EFAULT; | |
1586 | goto out; | |
1587 | } | |
1588 | continue; | |
1da177e4 LT |
1589 | |
1590 | page_not_up_to_date: | |
1591 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1592 | error = lock_page_killable(page); |
1593 | if (unlikely(error)) | |
1594 | goto readpage_error; | |
1da177e4 | 1595 | |
8ab22b9a | 1596 | page_not_up_to_date_locked: |
da6052f7 | 1597 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1598 | if (!page->mapping) { |
1599 | unlock_page(page); | |
1600 | page_cache_release(page); | |
1601 | continue; | |
1602 | } | |
1603 | ||
1604 | /* Did somebody else fill it already? */ | |
1605 | if (PageUptodate(page)) { | |
1606 | unlock_page(page); | |
1607 | goto page_ok; | |
1608 | } | |
1609 | ||
1610 | readpage: | |
91803b49 JM |
1611 | /* |
1612 | * A previous I/O error may have been due to temporary | |
1613 | * failures, eg. multipath errors. | |
1614 | * PG_error will be set again if readpage fails. | |
1615 | */ | |
1616 | ClearPageError(page); | |
1da177e4 LT |
1617 | /* Start the actual read. The read will unlock the page. */ |
1618 | error = mapping->a_ops->readpage(filp, page); | |
1619 | ||
994fc28c ZB |
1620 | if (unlikely(error)) { |
1621 | if (error == AOP_TRUNCATED_PAGE) { | |
1622 | page_cache_release(page); | |
6e58e79d | 1623 | error = 0; |
994fc28c ZB |
1624 | goto find_page; |
1625 | } | |
1da177e4 | 1626 | goto readpage_error; |
994fc28c | 1627 | } |
1da177e4 LT |
1628 | |
1629 | if (!PageUptodate(page)) { | |
85462323 ON |
1630 | error = lock_page_killable(page); |
1631 | if (unlikely(error)) | |
1632 | goto readpage_error; | |
1da177e4 LT |
1633 | if (!PageUptodate(page)) { |
1634 | if (page->mapping == NULL) { | |
1635 | /* | |
2ecdc82e | 1636 | * invalidate_mapping_pages got it |
1da177e4 LT |
1637 | */ |
1638 | unlock_page(page); | |
1639 | page_cache_release(page); | |
1640 | goto find_page; | |
1641 | } | |
1642 | unlock_page(page); | |
7ff81078 | 1643 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1644 | error = -EIO; |
1645 | goto readpage_error; | |
1da177e4 LT |
1646 | } |
1647 | unlock_page(page); | |
1648 | } | |
1649 | ||
1da177e4 LT |
1650 | goto page_ok; |
1651 | ||
1652 | readpage_error: | |
1653 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1da177e4 LT |
1654 | page_cache_release(page); |
1655 | goto out; | |
1656 | ||
1657 | no_cached_page: | |
1658 | /* | |
1659 | * Ok, it wasn't cached, so we need to create a new | |
1660 | * page.. | |
1661 | */ | |
eb2be189 NP |
1662 | page = page_cache_alloc_cold(mapping); |
1663 | if (!page) { | |
6e58e79d | 1664 | error = -ENOMEM; |
eb2be189 | 1665 | goto out; |
1da177e4 | 1666 | } |
eb2be189 | 1667 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1668 | index, GFP_KERNEL); |
1669 | if (error) { | |
eb2be189 | 1670 | page_cache_release(page); |
6e58e79d AV |
1671 | if (error == -EEXIST) { |
1672 | error = 0; | |
1da177e4 | 1673 | goto find_page; |
6e58e79d | 1674 | } |
1da177e4 LT |
1675 | goto out; |
1676 | } | |
1da177e4 LT |
1677 | goto readpage; |
1678 | } | |
1679 | ||
1680 | out: | |
7ff81078 FW |
1681 | ra->prev_pos = prev_index; |
1682 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1683 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1684 | |
f4e6b498 | 1685 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1686 | file_accessed(filp); |
6e58e79d | 1687 | return written ? written : error; |
1da177e4 LT |
1688 | } |
1689 | ||
485bb99b | 1690 | /** |
6abd2322 | 1691 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 1692 | * @iocb: kernel I/O control block |
6abd2322 | 1693 | * @iter: destination for the data read |
485bb99b | 1694 | * |
6abd2322 | 1695 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
1696 | * that can use the page cache directly. |
1697 | */ | |
1698 | ssize_t | |
ed978a81 | 1699 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 1700 | { |
ed978a81 | 1701 | struct file *file = iocb->ki_filp; |
cb66a7a1 | 1702 | ssize_t retval = 0; |
543ade1f | 1703 | loff_t *ppos = &iocb->ki_pos; |
ed978a81 | 1704 | loff_t pos = *ppos; |
1da177e4 | 1705 | |
2ba48ce5 | 1706 | if (iocb->ki_flags & IOCB_DIRECT) { |
ed978a81 AV |
1707 | struct address_space *mapping = file->f_mapping; |
1708 | struct inode *inode = mapping->host; | |
1709 | size_t count = iov_iter_count(iter); | |
543ade1f | 1710 | loff_t size; |
1da177e4 | 1711 | |
1da177e4 LT |
1712 | if (!count) |
1713 | goto out; /* skip atime */ | |
1714 | size = i_size_read(inode); | |
9fe55eea | 1715 | retval = filemap_write_and_wait_range(mapping, pos, |
a6cbcd4a | 1716 | pos + count - 1); |
9fe55eea | 1717 | if (!retval) { |
ed978a81 | 1718 | struct iov_iter data = *iter; |
22c6186e | 1719 | retval = mapping->a_ops->direct_IO(iocb, &data, pos); |
9fe55eea | 1720 | } |
d8d3d94b | 1721 | |
9fe55eea SW |
1722 | if (retval > 0) { |
1723 | *ppos = pos + retval; | |
ed978a81 | 1724 | iov_iter_advance(iter, retval); |
9fe55eea | 1725 | } |
66f998f6 | 1726 | |
9fe55eea SW |
1727 | /* |
1728 | * Btrfs can have a short DIO read if we encounter | |
1729 | * compressed extents, so if there was an error, or if | |
1730 | * we've already read everything we wanted to, or if | |
1731 | * there was a short read because we hit EOF, go ahead | |
1732 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
1733 | * the rest of the read. Buffered reads will not work for |
1734 | * DAX files, so don't bother trying. | |
9fe55eea | 1735 | */ |
fbbbad4b MW |
1736 | if (retval < 0 || !iov_iter_count(iter) || *ppos >= size || |
1737 | IS_DAX(inode)) { | |
ed978a81 | 1738 | file_accessed(file); |
9fe55eea | 1739 | goto out; |
0e0bcae3 | 1740 | } |
1da177e4 LT |
1741 | } |
1742 | ||
ed978a81 | 1743 | retval = do_generic_file_read(file, ppos, iter, retval); |
1da177e4 LT |
1744 | out: |
1745 | return retval; | |
1746 | } | |
ed978a81 | 1747 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 1748 | |
1da177e4 | 1749 | #ifdef CONFIG_MMU |
485bb99b RD |
1750 | /** |
1751 | * page_cache_read - adds requested page to the page cache if not already there | |
1752 | * @file: file to read | |
1753 | * @offset: page index | |
1754 | * | |
1da177e4 LT |
1755 | * This adds the requested page to the page cache if it isn't already there, |
1756 | * and schedules an I/O to read in its contents from disk. | |
1757 | */ | |
920c7a5d | 1758 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1759 | { |
1760 | struct address_space *mapping = file->f_mapping; | |
99dadfdd | 1761 | struct page *page; |
994fc28c | 1762 | int ret; |
1da177e4 | 1763 | |
994fc28c ZB |
1764 | do { |
1765 | page = page_cache_alloc_cold(mapping); | |
1766 | if (!page) | |
1767 | return -ENOMEM; | |
1768 | ||
1769 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); | |
1770 | if (ret == 0) | |
1771 | ret = mapping->a_ops->readpage(file, page); | |
1772 | else if (ret == -EEXIST) | |
1773 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1774 | |
1da177e4 | 1775 | page_cache_release(page); |
1da177e4 | 1776 | |
994fc28c | 1777 | } while (ret == AOP_TRUNCATED_PAGE); |
99dadfdd | 1778 | |
994fc28c | 1779 | return ret; |
1da177e4 LT |
1780 | } |
1781 | ||
1782 | #define MMAP_LOTSAMISS (100) | |
1783 | ||
ef00e08e LT |
1784 | /* |
1785 | * Synchronous readahead happens when we don't even find | |
1786 | * a page in the page cache at all. | |
1787 | */ | |
1788 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1789 | struct file_ra_state *ra, | |
1790 | struct file *file, | |
1791 | pgoff_t offset) | |
1792 | { | |
1793 | unsigned long ra_pages; | |
1794 | struct address_space *mapping = file->f_mapping; | |
1795 | ||
1796 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1797 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 1798 | return; |
275b12bf WF |
1799 | if (!ra->ra_pages) |
1800 | return; | |
ef00e08e | 1801 | |
64363aad | 1802 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
1803 | page_cache_sync_readahead(mapping, ra, file, offset, |
1804 | ra->ra_pages); | |
ef00e08e LT |
1805 | return; |
1806 | } | |
1807 | ||
207d04ba AK |
1808 | /* Avoid banging the cache line if not needed */ |
1809 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1810 | ra->mmap_miss++; |
1811 | ||
1812 | /* | |
1813 | * Do we miss much more than hit in this file? If so, | |
1814 | * stop bothering with read-ahead. It will only hurt. | |
1815 | */ | |
1816 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1817 | return; | |
1818 | ||
d30a1100 WF |
1819 | /* |
1820 | * mmap read-around | |
1821 | */ | |
ef00e08e | 1822 | ra_pages = max_sane_readahead(ra->ra_pages); |
275b12bf WF |
1823 | ra->start = max_t(long, 0, offset - ra_pages / 2); |
1824 | ra->size = ra_pages; | |
2cbea1d3 | 1825 | ra->async_size = ra_pages / 4; |
275b12bf | 1826 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1827 | } |
1828 | ||
1829 | /* | |
1830 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1831 | * so we want to possibly extend the readahead further.. | |
1832 | */ | |
1833 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1834 | struct file_ra_state *ra, | |
1835 | struct file *file, | |
1836 | struct page *page, | |
1837 | pgoff_t offset) | |
1838 | { | |
1839 | struct address_space *mapping = file->f_mapping; | |
1840 | ||
1841 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1842 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
1843 | return; |
1844 | if (ra->mmap_miss > 0) | |
1845 | ra->mmap_miss--; | |
1846 | if (PageReadahead(page)) | |
2fad6f5d WF |
1847 | page_cache_async_readahead(mapping, ra, file, |
1848 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1849 | } |
1850 | ||
485bb99b | 1851 | /** |
54cb8821 | 1852 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1853 | * @vma: vma in which the fault was taken |
1854 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1855 | * |
54cb8821 | 1856 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1857 | * mapped memory region to read in file data during a page fault. |
1858 | * | |
1859 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1860 | * it in the page cache, and handles the special cases reasonably without | |
1861 | * having a lot of duplicated code. | |
9a95f3cf PC |
1862 | * |
1863 | * vma->vm_mm->mmap_sem must be held on entry. | |
1864 | * | |
1865 | * If our return value has VM_FAULT_RETRY set, it's because | |
1866 | * lock_page_or_retry() returned 0. | |
1867 | * The mmap_sem has usually been released in this case. | |
1868 | * See __lock_page_or_retry() for the exception. | |
1869 | * | |
1870 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
1871 | * has not been released. | |
1872 | * | |
1873 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
1da177e4 | 1874 | */ |
d0217ac0 | 1875 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1876 | { |
1877 | int error; | |
54cb8821 | 1878 | struct file *file = vma->vm_file; |
1da177e4 LT |
1879 | struct address_space *mapping = file->f_mapping; |
1880 | struct file_ra_state *ra = &file->f_ra; | |
1881 | struct inode *inode = mapping->host; | |
ef00e08e | 1882 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 1883 | struct page *page; |
99e3e53f | 1884 | loff_t size; |
83c54070 | 1885 | int ret = 0; |
1da177e4 | 1886 | |
99e3e53f KS |
1887 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
1888 | if (offset >= size >> PAGE_CACHE_SHIFT) | |
5307cc1a | 1889 | return VM_FAULT_SIGBUS; |
1da177e4 | 1890 | |
1da177e4 | 1891 | /* |
49426420 | 1892 | * Do we have something in the page cache already? |
1da177e4 | 1893 | */ |
ef00e08e | 1894 | page = find_get_page(mapping, offset); |
45cac65b | 1895 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 1896 | /* |
ef00e08e LT |
1897 | * We found the page, so try async readahead before |
1898 | * waiting for the lock. | |
1da177e4 | 1899 | */ |
ef00e08e | 1900 | do_async_mmap_readahead(vma, ra, file, page, offset); |
45cac65b | 1901 | } else if (!page) { |
ef00e08e LT |
1902 | /* No page in the page cache at all */ |
1903 | do_sync_mmap_readahead(vma, ra, file, offset); | |
1904 | count_vm_event(PGMAJFAULT); | |
456f998e | 1905 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
1906 | ret = VM_FAULT_MAJOR; |
1907 | retry_find: | |
b522c94d | 1908 | page = find_get_page(mapping, offset); |
1da177e4 LT |
1909 | if (!page) |
1910 | goto no_cached_page; | |
1911 | } | |
1912 | ||
d88c0922 ML |
1913 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
1914 | page_cache_release(page); | |
d065bd81 | 1915 | return ret | VM_FAULT_RETRY; |
d88c0922 | 1916 | } |
b522c94d ML |
1917 | |
1918 | /* Did it get truncated? */ | |
1919 | if (unlikely(page->mapping != mapping)) { | |
1920 | unlock_page(page); | |
1921 | put_page(page); | |
1922 | goto retry_find; | |
1923 | } | |
309381fe | 1924 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 1925 | |
1da177e4 | 1926 | /* |
d00806b1 NP |
1927 | * We have a locked page in the page cache, now we need to check |
1928 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1929 | */ |
d00806b1 | 1930 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1931 | goto page_not_uptodate; |
1932 | ||
ef00e08e LT |
1933 | /* |
1934 | * Found the page and have a reference on it. | |
1935 | * We must recheck i_size under page lock. | |
1936 | */ | |
99e3e53f KS |
1937 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
1938 | if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) { | |
d00806b1 | 1939 | unlock_page(page); |
745ad48e | 1940 | page_cache_release(page); |
5307cc1a | 1941 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1942 | } |
1943 | ||
d0217ac0 | 1944 | vmf->page = page; |
83c54070 | 1945 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 1946 | |
1da177e4 LT |
1947 | no_cached_page: |
1948 | /* | |
1949 | * We're only likely to ever get here if MADV_RANDOM is in | |
1950 | * effect. | |
1951 | */ | |
ef00e08e | 1952 | error = page_cache_read(file, offset); |
1da177e4 LT |
1953 | |
1954 | /* | |
1955 | * The page we want has now been added to the page cache. | |
1956 | * In the unlikely event that someone removed it in the | |
1957 | * meantime, we'll just come back here and read it again. | |
1958 | */ | |
1959 | if (error >= 0) | |
1960 | goto retry_find; | |
1961 | ||
1962 | /* | |
1963 | * An error return from page_cache_read can result if the | |
1964 | * system is low on memory, or a problem occurs while trying | |
1965 | * to schedule I/O. | |
1966 | */ | |
1967 | if (error == -ENOMEM) | |
d0217ac0 NP |
1968 | return VM_FAULT_OOM; |
1969 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1970 | |
1971 | page_not_uptodate: | |
1da177e4 LT |
1972 | /* |
1973 | * Umm, take care of errors if the page isn't up-to-date. | |
1974 | * Try to re-read it _once_. We do this synchronously, | |
1975 | * because there really aren't any performance issues here | |
1976 | * and we need to check for errors. | |
1977 | */ | |
1da177e4 | 1978 | ClearPageError(page); |
994fc28c | 1979 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
1980 | if (!error) { |
1981 | wait_on_page_locked(page); | |
1982 | if (!PageUptodate(page)) | |
1983 | error = -EIO; | |
1984 | } | |
d00806b1 NP |
1985 | page_cache_release(page); |
1986 | ||
1987 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 1988 | goto retry_find; |
1da177e4 | 1989 | |
d00806b1 | 1990 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 1991 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 1992 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
1993 | } |
1994 | EXPORT_SYMBOL(filemap_fault); | |
1995 | ||
f1820361 KS |
1996 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
1997 | { | |
1998 | struct radix_tree_iter iter; | |
1999 | void **slot; | |
2000 | struct file *file = vma->vm_file; | |
2001 | struct address_space *mapping = file->f_mapping; | |
2002 | loff_t size; | |
2003 | struct page *page; | |
2004 | unsigned long address = (unsigned long) vmf->virtual_address; | |
2005 | unsigned long addr; | |
2006 | pte_t *pte; | |
2007 | ||
2008 | rcu_read_lock(); | |
2009 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) { | |
2010 | if (iter.index > vmf->max_pgoff) | |
2011 | break; | |
2012 | repeat: | |
2013 | page = radix_tree_deref_slot(slot); | |
2014 | if (unlikely(!page)) | |
2015 | goto next; | |
2016 | if (radix_tree_exception(page)) { | |
2017 | if (radix_tree_deref_retry(page)) | |
2018 | break; | |
2019 | else | |
2020 | goto next; | |
2021 | } | |
2022 | ||
2023 | if (!page_cache_get_speculative(page)) | |
2024 | goto repeat; | |
2025 | ||
2026 | /* Has the page moved? */ | |
2027 | if (unlikely(page != *slot)) { | |
2028 | page_cache_release(page); | |
2029 | goto repeat; | |
2030 | } | |
2031 | ||
2032 | if (!PageUptodate(page) || | |
2033 | PageReadahead(page) || | |
2034 | PageHWPoison(page)) | |
2035 | goto skip; | |
2036 | if (!trylock_page(page)) | |
2037 | goto skip; | |
2038 | ||
2039 | if (page->mapping != mapping || !PageUptodate(page)) | |
2040 | goto unlock; | |
2041 | ||
99e3e53f KS |
2042 | size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE); |
2043 | if (page->index >= size >> PAGE_CACHE_SHIFT) | |
f1820361 KS |
2044 | goto unlock; |
2045 | ||
2046 | pte = vmf->pte + page->index - vmf->pgoff; | |
2047 | if (!pte_none(*pte)) | |
2048 | goto unlock; | |
2049 | ||
2050 | if (file->f_ra.mmap_miss > 0) | |
2051 | file->f_ra.mmap_miss--; | |
2052 | addr = address + (page->index - vmf->pgoff) * PAGE_SIZE; | |
2053 | do_set_pte(vma, addr, page, pte, false, false); | |
2054 | unlock_page(page); | |
2055 | goto next; | |
2056 | unlock: | |
2057 | unlock_page(page); | |
2058 | skip: | |
2059 | page_cache_release(page); | |
2060 | next: | |
2061 | if (iter.index == vmf->max_pgoff) | |
2062 | break; | |
2063 | } | |
2064 | rcu_read_unlock(); | |
2065 | } | |
2066 | EXPORT_SYMBOL(filemap_map_pages); | |
2067 | ||
4fcf1c62 JK |
2068 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2069 | { | |
2070 | struct page *page = vmf->page; | |
496ad9aa | 2071 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
2072 | int ret = VM_FAULT_LOCKED; |
2073 | ||
14da9200 | 2074 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
2075 | file_update_time(vma->vm_file); |
2076 | lock_page(page); | |
2077 | if (page->mapping != inode->i_mapping) { | |
2078 | unlock_page(page); | |
2079 | ret = VM_FAULT_NOPAGE; | |
2080 | goto out; | |
2081 | } | |
14da9200 JK |
2082 | /* |
2083 | * We mark the page dirty already here so that when freeze is in | |
2084 | * progress, we are guaranteed that writeback during freezing will | |
2085 | * see the dirty page and writeprotect it again. | |
2086 | */ | |
2087 | set_page_dirty(page); | |
1d1d1a76 | 2088 | wait_for_stable_page(page); |
4fcf1c62 | 2089 | out: |
14da9200 | 2090 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2091 | return ret; |
2092 | } | |
2093 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2094 | ||
f0f37e2f | 2095 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2096 | .fault = filemap_fault, |
f1820361 | 2097 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2098 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2099 | }; |
2100 | ||
2101 | /* This is used for a general mmap of a disk file */ | |
2102 | ||
2103 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2104 | { | |
2105 | struct address_space *mapping = file->f_mapping; | |
2106 | ||
2107 | if (!mapping->a_ops->readpage) | |
2108 | return -ENOEXEC; | |
2109 | file_accessed(file); | |
2110 | vma->vm_ops = &generic_file_vm_ops; | |
2111 | return 0; | |
2112 | } | |
1da177e4 LT |
2113 | |
2114 | /* | |
2115 | * This is for filesystems which do not implement ->writepage. | |
2116 | */ | |
2117 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2118 | { | |
2119 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2120 | return -EINVAL; | |
2121 | return generic_file_mmap(file, vma); | |
2122 | } | |
2123 | #else | |
2124 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2125 | { | |
2126 | return -ENOSYS; | |
2127 | } | |
2128 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2129 | { | |
2130 | return -ENOSYS; | |
2131 | } | |
2132 | #endif /* CONFIG_MMU */ | |
2133 | ||
2134 | EXPORT_SYMBOL(generic_file_mmap); | |
2135 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2136 | ||
67f9fd91 SL |
2137 | static struct page *wait_on_page_read(struct page *page) |
2138 | { | |
2139 | if (!IS_ERR(page)) { | |
2140 | wait_on_page_locked(page); | |
2141 | if (!PageUptodate(page)) { | |
2142 | page_cache_release(page); | |
2143 | page = ERR_PTR(-EIO); | |
2144 | } | |
2145 | } | |
2146 | return page; | |
2147 | } | |
2148 | ||
6fe6900e | 2149 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 2150 | pgoff_t index, |
5e5358e7 | 2151 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2152 | void *data, |
2153 | gfp_t gfp) | |
1da177e4 | 2154 | { |
eb2be189 | 2155 | struct page *page; |
1da177e4 LT |
2156 | int err; |
2157 | repeat: | |
2158 | page = find_get_page(mapping, index); | |
2159 | if (!page) { | |
0531b2aa | 2160 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
2161 | if (!page) |
2162 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2163 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 NP |
2164 | if (unlikely(err)) { |
2165 | page_cache_release(page); | |
2166 | if (err == -EEXIST) | |
2167 | goto repeat; | |
1da177e4 | 2168 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2169 | return ERR_PTR(err); |
2170 | } | |
1da177e4 LT |
2171 | err = filler(data, page); |
2172 | if (err < 0) { | |
2173 | page_cache_release(page); | |
2174 | page = ERR_PTR(err); | |
67f9fd91 SL |
2175 | } else { |
2176 | page = wait_on_page_read(page); | |
1da177e4 LT |
2177 | } |
2178 | } | |
1da177e4 LT |
2179 | return page; |
2180 | } | |
2181 | ||
0531b2aa | 2182 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2183 | pgoff_t index, |
5e5358e7 | 2184 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2185 | void *data, |
2186 | gfp_t gfp) | |
2187 | ||
1da177e4 LT |
2188 | { |
2189 | struct page *page; | |
2190 | int err; | |
2191 | ||
2192 | retry: | |
0531b2aa | 2193 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 2194 | if (IS_ERR(page)) |
c855ff37 | 2195 | return page; |
1da177e4 LT |
2196 | if (PageUptodate(page)) |
2197 | goto out; | |
2198 | ||
2199 | lock_page(page); | |
2200 | if (!page->mapping) { | |
2201 | unlock_page(page); | |
2202 | page_cache_release(page); | |
2203 | goto retry; | |
2204 | } | |
2205 | if (PageUptodate(page)) { | |
2206 | unlock_page(page); | |
2207 | goto out; | |
2208 | } | |
2209 | err = filler(data, page); | |
2210 | if (err < 0) { | |
2211 | page_cache_release(page); | |
c855ff37 | 2212 | return ERR_PTR(err); |
67f9fd91 SL |
2213 | } else { |
2214 | page = wait_on_page_read(page); | |
2215 | if (IS_ERR(page)) | |
2216 | return page; | |
1da177e4 | 2217 | } |
c855ff37 | 2218 | out: |
6fe6900e NP |
2219 | mark_page_accessed(page); |
2220 | return page; | |
2221 | } | |
0531b2aa LT |
2222 | |
2223 | /** | |
67f9fd91 | 2224 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2225 | * @mapping: the page's address_space |
2226 | * @index: the page index | |
2227 | * @filler: function to perform the read | |
5e5358e7 | 2228 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2229 | * |
0531b2aa | 2230 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2231 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2232 | * |
2233 | * If the page does not get brought uptodate, return -EIO. | |
2234 | */ | |
67f9fd91 | 2235 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2236 | pgoff_t index, |
5e5358e7 | 2237 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2238 | void *data) |
2239 | { | |
2240 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2241 | } | |
67f9fd91 | 2242 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2243 | |
2244 | /** | |
2245 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2246 | * @mapping: the page's address_space | |
2247 | * @index: the page index | |
2248 | * @gfp: the page allocator flags to use if allocating | |
2249 | * | |
2250 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2251 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2252 | * |
2253 | * If the page does not get brought uptodate, return -EIO. | |
2254 | */ | |
2255 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2256 | pgoff_t index, | |
2257 | gfp_t gfp) | |
2258 | { | |
2259 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2260 | ||
67f9fd91 | 2261 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2262 | } |
2263 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2264 | ||
1da177e4 LT |
2265 | /* |
2266 | * Performs necessary checks before doing a write | |
2267 | * | |
485bb99b | 2268 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2269 | * Returns appropriate error code that caller should return or |
2270 | * zero in case that write should be allowed. | |
2271 | */ | |
3309dd04 | 2272 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2273 | { |
3309dd04 | 2274 | struct file *file = iocb->ki_filp; |
1da177e4 | 2275 | struct inode *inode = file->f_mapping->host; |
59e99e5b | 2276 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
3309dd04 | 2277 | loff_t pos; |
1da177e4 | 2278 | |
3309dd04 AV |
2279 | if (!iov_iter_count(from)) |
2280 | return 0; | |
1da177e4 | 2281 | |
0fa6b005 | 2282 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2283 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2284 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2285 | |
3309dd04 | 2286 | pos = iocb->ki_pos; |
1da177e4 | 2287 | |
0fa6b005 | 2288 | if (limit != RLIM_INFINITY) { |
3309dd04 | 2289 | if (iocb->ki_pos >= limit) { |
0fa6b005 AV |
2290 | send_sig(SIGXFSZ, current, 0); |
2291 | return -EFBIG; | |
1da177e4 | 2292 | } |
3309dd04 | 2293 | iov_iter_truncate(from, limit - (unsigned long)pos); |
1da177e4 LT |
2294 | } |
2295 | ||
2296 | /* | |
2297 | * LFS rule | |
2298 | */ | |
3309dd04 | 2299 | if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS && |
1da177e4 | 2300 | !(file->f_flags & O_LARGEFILE))) { |
3309dd04 | 2301 | if (pos >= MAX_NON_LFS) |
1da177e4 | 2302 | return -EFBIG; |
3309dd04 | 2303 | iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos); |
1da177e4 LT |
2304 | } |
2305 | ||
2306 | /* | |
2307 | * Are we about to exceed the fs block limit ? | |
2308 | * | |
2309 | * If we have written data it becomes a short write. If we have | |
2310 | * exceeded without writing data we send a signal and return EFBIG. | |
2311 | * Linus frestrict idea will clean these up nicely.. | |
2312 | */ | |
3309dd04 AV |
2313 | if (unlikely(pos >= inode->i_sb->s_maxbytes)) |
2314 | return -EFBIG; | |
1da177e4 | 2315 | |
3309dd04 AV |
2316 | iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos); |
2317 | return iov_iter_count(from); | |
1da177e4 LT |
2318 | } |
2319 | EXPORT_SYMBOL(generic_write_checks); | |
2320 | ||
afddba49 NP |
2321 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2322 | loff_t pos, unsigned len, unsigned flags, | |
2323 | struct page **pagep, void **fsdata) | |
2324 | { | |
2325 | const struct address_space_operations *aops = mapping->a_ops; | |
2326 | ||
4e02ed4b | 2327 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2328 | pagep, fsdata); |
afddba49 NP |
2329 | } |
2330 | EXPORT_SYMBOL(pagecache_write_begin); | |
2331 | ||
2332 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2333 | loff_t pos, unsigned len, unsigned copied, | |
2334 | struct page *page, void *fsdata) | |
2335 | { | |
2336 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2337 | |
4e02ed4b | 2338 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2339 | } |
2340 | EXPORT_SYMBOL(pagecache_write_end); | |
2341 | ||
1da177e4 | 2342 | ssize_t |
0c949334 | 2343 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos) |
1da177e4 LT |
2344 | { |
2345 | struct file *file = iocb->ki_filp; | |
2346 | struct address_space *mapping = file->f_mapping; | |
2347 | struct inode *inode = mapping->host; | |
2348 | ssize_t written; | |
a969e903 CH |
2349 | size_t write_len; |
2350 | pgoff_t end; | |
26978b8b | 2351 | struct iov_iter data; |
1da177e4 | 2352 | |
0c949334 | 2353 | write_len = iov_iter_count(from); |
a969e903 | 2354 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; |
a969e903 | 2355 | |
48b47c56 | 2356 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2357 | if (written) |
2358 | goto out; | |
2359 | ||
2360 | /* | |
2361 | * After a write we want buffered reads to be sure to go to disk to get | |
2362 | * the new data. We invalidate clean cached page from the region we're | |
2363 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2364 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2365 | */ |
2366 | if (mapping->nrpages) { | |
2367 | written = invalidate_inode_pages2_range(mapping, | |
2368 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2369 | /* |
2370 | * If a page can not be invalidated, return 0 to fall back | |
2371 | * to buffered write. | |
2372 | */ | |
2373 | if (written) { | |
2374 | if (written == -EBUSY) | |
2375 | return 0; | |
a969e903 | 2376 | goto out; |
6ccfa806 | 2377 | } |
a969e903 CH |
2378 | } |
2379 | ||
26978b8b | 2380 | data = *from; |
22c6186e | 2381 | written = mapping->a_ops->direct_IO(iocb, &data, pos); |
a969e903 CH |
2382 | |
2383 | /* | |
2384 | * Finally, try again to invalidate clean pages which might have been | |
2385 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2386 | * if the source of the write was an mmap'ed region of the file | |
2387 | * we're writing. Either one is a pretty crazy thing to do, | |
2388 | * so we don't support it 100%. If this invalidation | |
2389 | * fails, tough, the write still worked... | |
2390 | */ | |
2391 | if (mapping->nrpages) { | |
2392 | invalidate_inode_pages2_range(mapping, | |
2393 | pos >> PAGE_CACHE_SHIFT, end); | |
2394 | } | |
2395 | ||
1da177e4 | 2396 | if (written > 0) { |
0116651c | 2397 | pos += written; |
f8579f86 | 2398 | iov_iter_advance(from, written); |
0116651c NK |
2399 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
2400 | i_size_write(inode, pos); | |
1da177e4 LT |
2401 | mark_inode_dirty(inode); |
2402 | } | |
5cb6c6c7 | 2403 | iocb->ki_pos = pos; |
1da177e4 | 2404 | } |
a969e903 | 2405 | out: |
1da177e4 LT |
2406 | return written; |
2407 | } | |
2408 | EXPORT_SYMBOL(generic_file_direct_write); | |
2409 | ||
eb2be189 NP |
2410 | /* |
2411 | * Find or create a page at the given pagecache position. Return the locked | |
2412 | * page. This function is specifically for buffered writes. | |
2413 | */ | |
54566b2c NP |
2414 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2415 | pgoff_t index, unsigned flags) | |
eb2be189 | 2416 | { |
eb2be189 | 2417 | struct page *page; |
2457aec6 | 2418 | int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT; |
0faa70cb | 2419 | |
54566b2c | 2420 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
2421 | fgp_flags |= FGP_NOFS; |
2422 | ||
2423 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 2424 | mapping_gfp_mask(mapping)); |
c585a267 | 2425 | if (page) |
2457aec6 | 2426 | wait_for_stable_page(page); |
eb2be189 | 2427 | |
eb2be189 NP |
2428 | return page; |
2429 | } | |
54566b2c | 2430 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2431 | |
3b93f911 | 2432 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
2433 | struct iov_iter *i, loff_t pos) |
2434 | { | |
2435 | struct address_space *mapping = file->f_mapping; | |
2436 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2437 | long status = 0; | |
2438 | ssize_t written = 0; | |
674b892e NP |
2439 | unsigned int flags = 0; |
2440 | ||
2441 | /* | |
2442 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2443 | */ | |
777eda2c | 2444 | if (!iter_is_iovec(i)) |
674b892e | 2445 | flags |= AOP_FLAG_UNINTERRUPTIBLE; |
afddba49 NP |
2446 | |
2447 | do { | |
2448 | struct page *page; | |
afddba49 NP |
2449 | unsigned long offset; /* Offset into pagecache page */ |
2450 | unsigned long bytes; /* Bytes to write to page */ | |
2451 | size_t copied; /* Bytes copied from user */ | |
2452 | void *fsdata; | |
2453 | ||
2454 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2455 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2456 | iov_iter_count(i)); | |
2457 | ||
2458 | again: | |
afddba49 NP |
2459 | /* |
2460 | * Bring in the user page that we will copy from _first_. | |
2461 | * Otherwise there's a nasty deadlock on copying from the | |
2462 | * same page as we're writing to, without it being marked | |
2463 | * up-to-date. | |
2464 | * | |
2465 | * Not only is this an optimisation, but it is also required | |
2466 | * to check that the address is actually valid, when atomic | |
2467 | * usercopies are used, below. | |
2468 | */ | |
2469 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2470 | status = -EFAULT; | |
2471 | break; | |
2472 | } | |
2473 | ||
674b892e | 2474 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 2475 | &page, &fsdata); |
2457aec6 | 2476 | if (unlikely(status < 0)) |
afddba49 NP |
2477 | break; |
2478 | ||
931e80e4 | 2479 | if (mapping_writably_mapped(mapping)) |
2480 | flush_dcache_page(page); | |
2481 | ||
afddba49 | 2482 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
2483 | flush_dcache_page(page); |
2484 | ||
2485 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
2486 | page, fsdata); | |
2487 | if (unlikely(status < 0)) | |
2488 | break; | |
2489 | copied = status; | |
2490 | ||
2491 | cond_resched(); | |
2492 | ||
124d3b70 | 2493 | iov_iter_advance(i, copied); |
afddba49 NP |
2494 | if (unlikely(copied == 0)) { |
2495 | /* | |
2496 | * If we were unable to copy any data at all, we must | |
2497 | * fall back to a single segment length write. | |
2498 | * | |
2499 | * If we didn't fallback here, we could livelock | |
2500 | * because not all segments in the iov can be copied at | |
2501 | * once without a pagefault. | |
2502 | */ | |
2503 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2504 | iov_iter_single_seg_count(i)); | |
2505 | goto again; | |
2506 | } | |
afddba49 NP |
2507 | pos += copied; |
2508 | written += copied; | |
2509 | ||
2510 | balance_dirty_pages_ratelimited(mapping); | |
a50527b1 JK |
2511 | if (fatal_signal_pending(current)) { |
2512 | status = -EINTR; | |
2513 | break; | |
2514 | } | |
afddba49 NP |
2515 | } while (iov_iter_count(i)); |
2516 | ||
2517 | return written ? written : status; | |
2518 | } | |
3b93f911 | 2519 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 2520 | |
e4dd9de3 | 2521 | /** |
8174202b | 2522 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 2523 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 2524 | * @from: iov_iter with data to write |
e4dd9de3 JK |
2525 | * |
2526 | * This function does all the work needed for actually writing data to a | |
2527 | * file. It does all basic checks, removes SUID from the file, updates | |
2528 | * modification times and calls proper subroutines depending on whether we | |
2529 | * do direct IO or a standard buffered write. | |
2530 | * | |
2531 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2532 | * object which does not need locking at all. | |
2533 | * | |
2534 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2535 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2536 | * avoid syncing under i_mutex. | |
2537 | */ | |
8174202b | 2538 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2539 | { |
2540 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2541 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 2542 | struct inode *inode = mapping->host; |
3b93f911 | 2543 | ssize_t written = 0; |
1da177e4 | 2544 | ssize_t err; |
3b93f911 | 2545 | ssize_t status; |
1da177e4 | 2546 | |
1da177e4 | 2547 | /* We can write back this queue in page reclaim */ |
de1414a6 | 2548 | current->backing_dev_info = inode_to_bdi(inode); |
2f1936b8 | 2549 | err = file_remove_suid(file); |
1da177e4 LT |
2550 | if (err) |
2551 | goto out; | |
2552 | ||
c3b2da31 JB |
2553 | err = file_update_time(file); |
2554 | if (err) | |
2555 | goto out; | |
1da177e4 | 2556 | |
2ba48ce5 | 2557 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 2558 | loff_t pos, endbyte; |
fb5527e6 | 2559 | |
0b8def9d | 2560 | written = generic_file_direct_write(iocb, from, iocb->ki_pos); |
1da177e4 | 2561 | /* |
fbbbad4b MW |
2562 | * If the write stopped short of completing, fall back to |
2563 | * buffered writes. Some filesystems do this for writes to | |
2564 | * holes, for example. For DAX files, a buffered write will | |
2565 | * not succeed (even if it did, DAX does not handle dirty | |
2566 | * page-cache pages correctly). | |
1da177e4 | 2567 | */ |
0b8def9d | 2568 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
2569 | goto out; |
2570 | ||
0b8def9d | 2571 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 2572 | /* |
3b93f911 | 2573 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
2574 | * then we want to return the number of bytes which were |
2575 | * direct-written, or the error code if that was zero. Note | |
2576 | * that this differs from normal direct-io semantics, which | |
2577 | * will return -EFOO even if some bytes were written. | |
2578 | */ | |
60bb4529 | 2579 | if (unlikely(status < 0)) { |
3b93f911 | 2580 | err = status; |
fb5527e6 JM |
2581 | goto out; |
2582 | } | |
fb5527e6 JM |
2583 | /* |
2584 | * We need to ensure that the page cache pages are written to | |
2585 | * disk and invalidated to preserve the expected O_DIRECT | |
2586 | * semantics. | |
2587 | */ | |
3b93f911 | 2588 | endbyte = pos + status - 1; |
0b8def9d | 2589 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 2590 | if (err == 0) { |
0b8def9d | 2591 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 2592 | written += status; |
fb5527e6 JM |
2593 | invalidate_mapping_pages(mapping, |
2594 | pos >> PAGE_CACHE_SHIFT, | |
2595 | endbyte >> PAGE_CACHE_SHIFT); | |
2596 | } else { | |
2597 | /* | |
2598 | * We don't know how much we wrote, so just return | |
2599 | * the number of bytes which were direct-written | |
2600 | */ | |
2601 | } | |
2602 | } else { | |
0b8def9d AV |
2603 | written = generic_perform_write(file, from, iocb->ki_pos); |
2604 | if (likely(written > 0)) | |
2605 | iocb->ki_pos += written; | |
fb5527e6 | 2606 | } |
1da177e4 LT |
2607 | out: |
2608 | current->backing_dev_info = NULL; | |
2609 | return written ? written : err; | |
2610 | } | |
8174202b | 2611 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 2612 | |
e4dd9de3 | 2613 | /** |
8174202b | 2614 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 2615 | * @iocb: IO state structure |
8174202b | 2616 | * @from: iov_iter with data to write |
e4dd9de3 | 2617 | * |
8174202b | 2618 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
2619 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
2620 | * and acquires i_mutex as needed. | |
2621 | */ | |
8174202b | 2622 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2623 | { |
2624 | struct file *file = iocb->ki_filp; | |
148f948b | 2625 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2626 | ssize_t ret; |
1da177e4 | 2627 | |
1b1dcc1b | 2628 | mutex_lock(&inode->i_mutex); |
3309dd04 AV |
2629 | ret = generic_write_checks(iocb, from); |
2630 | if (ret > 0) | |
5f380c7f | 2631 | ret = __generic_file_write_iter(iocb, from); |
1b1dcc1b | 2632 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 2633 | |
02afc27f | 2634 | if (ret > 0) { |
1da177e4 LT |
2635 | ssize_t err; |
2636 | ||
d311d79d AV |
2637 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
2638 | if (err < 0) | |
1da177e4 LT |
2639 | ret = err; |
2640 | } | |
2641 | return ret; | |
2642 | } | |
8174202b | 2643 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 2644 | |
cf9a2ae8 DH |
2645 | /** |
2646 | * try_to_release_page() - release old fs-specific metadata on a page | |
2647 | * | |
2648 | * @page: the page which the kernel is trying to free | |
2649 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2650 | * | |
2651 | * The address_space is to try to release any data against the page | |
2652 | * (presumably at page->private). If the release was successful, return `1'. | |
2653 | * Otherwise return zero. | |
2654 | * | |
266cf658 DH |
2655 | * This may also be called if PG_fscache is set on a page, indicating that the |
2656 | * page is known to the local caching routines. | |
2657 | * | |
cf9a2ae8 | 2658 | * The @gfp_mask argument specifies whether I/O may be performed to release |
3f31fddf | 2659 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
cf9a2ae8 | 2660 | * |
cf9a2ae8 DH |
2661 | */ |
2662 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2663 | { | |
2664 | struct address_space * const mapping = page->mapping; | |
2665 | ||
2666 | BUG_ON(!PageLocked(page)); | |
2667 | if (PageWriteback(page)) | |
2668 | return 0; | |
2669 | ||
2670 | if (mapping && mapping->a_ops->releasepage) | |
2671 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2672 | return try_to_free_buffers(page); | |
2673 | } | |
2674 | ||
2675 | EXPORT_SYMBOL(try_to_release_page); |