]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/filemap.c | |
3 | * | |
4 | * Copyright (C) 1994-1999 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * This file handles the generic file mmap semantics used by | |
9 | * most "normal" filesystems (but you don't /have/ to use this: | |
10 | * the NFS filesystem used to do this differently, for example) | |
11 | */ | |
b95f1b31 | 12 | #include <linux/export.h> |
1da177e4 LT |
13 | #include <linux/compiler.h> |
14 | #include <linux/fs.h> | |
c22ce143 | 15 | #include <linux/uaccess.h> |
1da177e4 | 16 | #include <linux/aio.h> |
c59ede7b | 17 | #include <linux/capability.h> |
1da177e4 | 18 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 19 | #include <linux/gfp.h> |
1da177e4 LT |
20 | #include <linux/mm.h> |
21 | #include <linux/swap.h> | |
22 | #include <linux/mman.h> | |
23 | #include <linux/pagemap.h> | |
24 | #include <linux/file.h> | |
25 | #include <linux/uio.h> | |
26 | #include <linux/hash.h> | |
27 | #include <linux/writeback.h> | |
53253383 | 28 | #include <linux/backing-dev.h> |
1da177e4 LT |
29 | #include <linux/pagevec.h> |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/security.h> | |
44110fe3 | 32 | #include <linux/cpuset.h> |
2f718ffc | 33 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
8a9f3ccd | 34 | #include <linux/memcontrol.h> |
c515e1fd | 35 | #include <linux/cleancache.h> |
f1820361 | 36 | #include <linux/rmap.h> |
0f8053a5 NP |
37 | #include "internal.h" |
38 | ||
fe0bfaaf RJ |
39 | #define CREATE_TRACE_POINTS |
40 | #include <trace/events/filemap.h> | |
41 | ||
1da177e4 | 42 | /* |
1da177e4 LT |
43 | * FIXME: remove all knowledge of the buffer layer from the core VM |
44 | */ | |
148f948b | 45 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 46 | |
1da177e4 LT |
47 | #include <asm/mman.h> |
48 | ||
49 | /* | |
50 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
51 | * though. | |
52 | * | |
53 | * Shared mappings now work. 15.8.1995 Bruno. | |
54 | * | |
55 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
56 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
57 | * | |
58 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
59 | */ | |
60 | ||
61 | /* | |
62 | * Lock ordering: | |
63 | * | |
3d48ae45 | 64 | * ->i_mmap_mutex (truncate_pagecache) |
1da177e4 | 65 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 HD |
66 | * ->swap_lock (exclusive_swap_page, others) |
67 | * ->mapping->tree_lock | |
1da177e4 | 68 | * |
1b1dcc1b | 69 | * ->i_mutex |
3d48ae45 | 70 | * ->i_mmap_mutex (truncate->unmap_mapping_range) |
1da177e4 LT |
71 | * |
72 | * ->mmap_sem | |
3d48ae45 | 73 | * ->i_mmap_mutex |
b8072f09 | 74 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
75 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
76 | * | |
77 | * ->mmap_sem | |
78 | * ->lock_page (access_process_vm) | |
79 | * | |
ccad2365 | 80 | * ->i_mutex (generic_perform_write) |
82591e6e | 81 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
1da177e4 | 82 | * |
f758eeab | 83 | * bdi->wb.list_lock |
a66979ab | 84 | * sb_lock (fs/fs-writeback.c) |
1da177e4 LT |
85 | * ->mapping->tree_lock (__sync_single_inode) |
86 | * | |
3d48ae45 | 87 | * ->i_mmap_mutex |
1da177e4 LT |
88 | * ->anon_vma.lock (vma_adjust) |
89 | * | |
90 | * ->anon_vma.lock | |
b8072f09 | 91 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 92 | * |
b8072f09 | 93 | * ->page_table_lock or pte_lock |
5d337b91 | 94 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
95 | * ->private_lock (try_to_unmap_one) |
96 | * ->tree_lock (try_to_unmap_one) | |
97 | * ->zone.lru_lock (follow_page->mark_page_accessed) | |
053837fc | 98 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
1da177e4 LT |
99 | * ->private_lock (page_remove_rmap->set_page_dirty) |
100 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
f758eeab | 101 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 102 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
f758eeab | 103 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 104 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
105 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
106 | * | |
9a3c531d AK |
107 | * ->i_mmap_mutex |
108 | * ->tasklist_lock (memory_failure, collect_procs_ao) | |
1da177e4 LT |
109 | */ |
110 | ||
91b0abe3 JW |
111 | static void page_cache_tree_delete(struct address_space *mapping, |
112 | struct page *page, void *shadow) | |
113 | { | |
449dd698 JW |
114 | struct radix_tree_node *node; |
115 | unsigned long index; | |
116 | unsigned int offset; | |
117 | unsigned int tag; | |
118 | void **slot; | |
91b0abe3 | 119 | |
449dd698 JW |
120 | VM_BUG_ON(!PageLocked(page)); |
121 | ||
122 | __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot); | |
123 | ||
124 | if (shadow) { | |
91b0abe3 JW |
125 | mapping->nrshadows++; |
126 | /* | |
127 | * Make sure the nrshadows update is committed before | |
128 | * the nrpages update so that final truncate racing | |
129 | * with reclaim does not see both counters 0 at the | |
130 | * same time and miss a shadow entry. | |
131 | */ | |
132 | smp_wmb(); | |
449dd698 | 133 | } |
91b0abe3 | 134 | mapping->nrpages--; |
449dd698 JW |
135 | |
136 | if (!node) { | |
137 | /* Clear direct pointer tags in root node */ | |
138 | mapping->page_tree.gfp_mask &= __GFP_BITS_MASK; | |
139 | radix_tree_replace_slot(slot, shadow); | |
140 | return; | |
141 | } | |
142 | ||
143 | /* Clear tree tags for the removed page */ | |
144 | index = page->index; | |
145 | offset = index & RADIX_TREE_MAP_MASK; | |
146 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { | |
147 | if (test_bit(offset, node->tags[tag])) | |
148 | radix_tree_tag_clear(&mapping->page_tree, index, tag); | |
149 | } | |
150 | ||
151 | /* Delete page, swap shadow entry */ | |
152 | radix_tree_replace_slot(slot, shadow); | |
153 | workingset_node_pages_dec(node); | |
154 | if (shadow) | |
155 | workingset_node_shadows_inc(node); | |
156 | else | |
157 | if (__radix_tree_delete_node(&mapping->page_tree, node)) | |
158 | return; | |
159 | ||
160 | /* | |
161 | * Track node that only contains shadow entries. | |
162 | * | |
163 | * Avoid acquiring the list_lru lock if already tracked. The | |
164 | * list_empty() test is safe as node->private_list is | |
165 | * protected by mapping->tree_lock. | |
166 | */ | |
167 | if (!workingset_node_pages(node) && | |
168 | list_empty(&node->private_list)) { | |
169 | node->private_data = mapping; | |
170 | list_lru_add(&workingset_shadow_nodes, &node->private_list); | |
171 | } | |
91b0abe3 JW |
172 | } |
173 | ||
1da177e4 | 174 | /* |
e64a782f | 175 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 176 | * sure the page is locked and that nobody else uses it - or that usage |
19fd6231 | 177 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 178 | */ |
91b0abe3 | 179 | void __delete_from_page_cache(struct page *page, void *shadow) |
1da177e4 LT |
180 | { |
181 | struct address_space *mapping = page->mapping; | |
182 | ||
fe0bfaaf | 183 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
184 | /* |
185 | * if we're uptodate, flush out into the cleancache, otherwise | |
186 | * invalidate any existing cleancache entries. We can't leave | |
187 | * stale data around in the cleancache once our page is gone | |
188 | */ | |
189 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
190 | cleancache_put_page(page); | |
191 | else | |
3167760f | 192 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 193 | |
91b0abe3 JW |
194 | page_cache_tree_delete(mapping, page, shadow); |
195 | ||
1da177e4 | 196 | page->mapping = NULL; |
b85e0eff | 197 | /* Leave page->index set: truncation lookup relies upon it */ |
91b0abe3 | 198 | |
347ce434 | 199 | __dec_zone_page_state(page, NR_FILE_PAGES); |
4b02108a KM |
200 | if (PageSwapBacked(page)) |
201 | __dec_zone_page_state(page, NR_SHMEM); | |
45426812 | 202 | BUG_ON(page_mapped(page)); |
3a692790 LT |
203 | |
204 | /* | |
205 | * Some filesystems seem to re-dirty the page even after | |
206 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
207 | * | |
208 | * Fix it up by doing a final dirty accounting check after | |
209 | * having removed the page entirely. | |
210 | */ | |
211 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { | |
212 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
213 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
214 | } | |
1da177e4 LT |
215 | } |
216 | ||
702cfbf9 MK |
217 | /** |
218 | * delete_from_page_cache - delete page from page cache | |
219 | * @page: the page which the kernel is trying to remove from page cache | |
220 | * | |
221 | * This must be called only on pages that have been verified to be in the page | |
222 | * cache and locked. It will never put the page into the free list, the caller | |
223 | * has a reference on the page. | |
224 | */ | |
225 | void delete_from_page_cache(struct page *page) | |
1da177e4 LT |
226 | { |
227 | struct address_space *mapping = page->mapping; | |
6072d13c | 228 | void (*freepage)(struct page *); |
1da177e4 | 229 | |
cd7619d6 | 230 | BUG_ON(!PageLocked(page)); |
1da177e4 | 231 | |
6072d13c | 232 | freepage = mapping->a_ops->freepage; |
19fd6231 | 233 | spin_lock_irq(&mapping->tree_lock); |
91b0abe3 | 234 | __delete_from_page_cache(page, NULL); |
19fd6231 | 235 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 236 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
237 | |
238 | if (freepage) | |
239 | freepage(page); | |
97cecb5a MK |
240 | page_cache_release(page); |
241 | } | |
242 | EXPORT_SYMBOL(delete_from_page_cache); | |
243 | ||
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++; | |
488 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
489 | if (PageSwapBacked(new)) | |
490 | __inc_zone_page_state(new, NR_SHMEM); | |
491 | spin_unlock_irq(&mapping->tree_lock); | |
ab936cbc KH |
492 | /* mem_cgroup codes must not be called under tree_lock */ |
493 | mem_cgroup_replace_page_cache(old, new); | |
ef6a3c63 MS |
494 | radix_tree_preload_end(); |
495 | if (freepage) | |
496 | freepage(old); | |
497 | page_cache_release(old); | |
ef6a3c63 MS |
498 | } |
499 | ||
500 | return error; | |
501 | } | |
502 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
503 | ||
0cd6144a | 504 | static int page_cache_tree_insert(struct address_space *mapping, |
a528910e | 505 | struct page *page, void **shadowp) |
0cd6144a | 506 | { |
449dd698 | 507 | struct radix_tree_node *node; |
0cd6144a JW |
508 | void **slot; |
509 | int error; | |
510 | ||
449dd698 JW |
511 | error = __radix_tree_create(&mapping->page_tree, page->index, |
512 | &node, &slot); | |
513 | if (error) | |
514 | return error; | |
515 | if (*slot) { | |
0cd6144a JW |
516 | void *p; |
517 | ||
518 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
519 | if (!radix_tree_exceptional_entry(p)) | |
520 | return -EEXIST; | |
a528910e JW |
521 | if (shadowp) |
522 | *shadowp = p; | |
449dd698 JW |
523 | mapping->nrshadows--; |
524 | if (node) | |
525 | workingset_node_shadows_dec(node); | |
0cd6144a | 526 | } |
449dd698 JW |
527 | radix_tree_replace_slot(slot, page); |
528 | mapping->nrpages++; | |
529 | if (node) { | |
530 | workingset_node_pages_inc(node); | |
531 | /* | |
532 | * Don't track node that contains actual pages. | |
533 | * | |
534 | * Avoid acquiring the list_lru lock if already | |
535 | * untracked. The list_empty() test is safe as | |
536 | * node->private_list is protected by | |
537 | * mapping->tree_lock. | |
538 | */ | |
539 | if (!list_empty(&node->private_list)) | |
540 | list_lru_del(&workingset_shadow_nodes, | |
541 | &node->private_list); | |
542 | } | |
543 | return 0; | |
0cd6144a JW |
544 | } |
545 | ||
a528910e JW |
546 | static int __add_to_page_cache_locked(struct page *page, |
547 | struct address_space *mapping, | |
548 | pgoff_t offset, gfp_t gfp_mask, | |
549 | void **shadowp) | |
1da177e4 | 550 | { |
e286781d NP |
551 | int error; |
552 | ||
309381fe SL |
553 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
554 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d | 555 | |
d715ae08 | 556 | error = mem_cgroup_charge_file(page, current->mm, |
2c26fdd7 | 557 | gfp_mask & GFP_RECLAIM_MASK); |
35c754d7 | 558 | if (error) |
66a0c8ee | 559 | return error; |
1da177e4 | 560 | |
5e4c0d97 | 561 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 562 | if (error) { |
69029cd5 | 563 | mem_cgroup_uncharge_cache_page(page); |
66a0c8ee KS |
564 | return error; |
565 | } | |
566 | ||
567 | page_cache_get(page); | |
568 | page->mapping = mapping; | |
569 | page->index = offset; | |
570 | ||
571 | spin_lock_irq(&mapping->tree_lock); | |
a528910e | 572 | error = page_cache_tree_insert(mapping, page, shadowp); |
66a0c8ee KS |
573 | radix_tree_preload_end(); |
574 | if (unlikely(error)) | |
575 | goto err_insert; | |
66a0c8ee KS |
576 | __inc_zone_page_state(page, NR_FILE_PAGES); |
577 | spin_unlock_irq(&mapping->tree_lock); | |
578 | trace_mm_filemap_add_to_page_cache(page); | |
579 | return 0; | |
580 | err_insert: | |
581 | page->mapping = NULL; | |
582 | /* Leave page->index set: truncation relies upon it */ | |
583 | spin_unlock_irq(&mapping->tree_lock); | |
584 | mem_cgroup_uncharge_cache_page(page); | |
585 | page_cache_release(page); | |
1da177e4 LT |
586 | return error; |
587 | } | |
a528910e JW |
588 | |
589 | /** | |
590 | * add_to_page_cache_locked - add a locked page to the pagecache | |
591 | * @page: page to add | |
592 | * @mapping: the page's address_space | |
593 | * @offset: page index | |
594 | * @gfp_mask: page allocation mode | |
595 | * | |
596 | * This function is used to add a page to the pagecache. It must be locked. | |
597 | * This function does not add the page to the LRU. The caller must do that. | |
598 | */ | |
599 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
600 | pgoff_t offset, gfp_t gfp_mask) | |
601 | { | |
602 | return __add_to_page_cache_locked(page, mapping, offset, | |
603 | gfp_mask, NULL); | |
604 | } | |
e286781d | 605 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
606 | |
607 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 608 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 609 | { |
a528910e | 610 | void *shadow = NULL; |
4f98a2fe RR |
611 | int ret; |
612 | ||
a528910e JW |
613 | __set_page_locked(page); |
614 | ret = __add_to_page_cache_locked(page, mapping, offset, | |
615 | gfp_mask, &shadow); | |
616 | if (unlikely(ret)) | |
617 | __clear_page_locked(page); | |
618 | else { | |
619 | /* | |
620 | * The page might have been evicted from cache only | |
621 | * recently, in which case it should be activated like | |
622 | * any other repeatedly accessed page. | |
623 | */ | |
624 | if (shadow && workingset_refault(shadow)) { | |
625 | SetPageActive(page); | |
626 | workingset_activation(page); | |
627 | } else | |
628 | ClearPageActive(page); | |
629 | lru_cache_add(page); | |
630 | } | |
1da177e4 LT |
631 | return ret; |
632 | } | |
18bc0bbd | 633 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 634 | |
44110fe3 | 635 | #ifdef CONFIG_NUMA |
2ae88149 | 636 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 637 | { |
c0ff7453 MX |
638 | int n; |
639 | struct page *page; | |
640 | ||
44110fe3 | 641 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
642 | unsigned int cpuset_mems_cookie; |
643 | do { | |
d26914d1 | 644 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 MG |
645 | n = cpuset_mem_spread_node(); |
646 | page = alloc_pages_exact_node(n, gfp, 0); | |
d26914d1 | 647 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 648 | |
c0ff7453 | 649 | return page; |
44110fe3 | 650 | } |
2ae88149 | 651 | return alloc_pages(gfp, 0); |
44110fe3 | 652 | } |
2ae88149 | 653 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
654 | #endif |
655 | ||
1da177e4 LT |
656 | /* |
657 | * In order to wait for pages to become available there must be | |
658 | * waitqueues associated with pages. By using a hash table of | |
659 | * waitqueues where the bucket discipline is to maintain all | |
660 | * waiters on the same queue and wake all when any of the pages | |
661 | * become available, and for the woken contexts to check to be | |
662 | * sure the appropriate page became available, this saves space | |
663 | * at a cost of "thundering herd" phenomena during rare hash | |
664 | * collisions. | |
665 | */ | |
666 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
667 | { | |
668 | const struct zone *zone = page_zone(page); | |
669 | ||
670 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
671 | } | |
672 | ||
673 | static inline void wake_up_page(struct page *page, int bit) | |
674 | { | |
675 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); | |
676 | } | |
677 | ||
920c7a5d | 678 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
679 | { |
680 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
681 | ||
682 | if (test_bit(bit_nr, &page->flags)) | |
74316201 | 683 | __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
684 | TASK_UNINTERRUPTIBLE); |
685 | } | |
686 | EXPORT_SYMBOL(wait_on_page_bit); | |
687 | ||
f62e00cc KM |
688 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
689 | { | |
690 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
691 | ||
692 | if (!test_bit(bit_nr, &page->flags)) | |
693 | return 0; | |
694 | ||
695 | return __wait_on_bit(page_waitqueue(page), &wait, | |
74316201 | 696 | bit_wait_io, TASK_KILLABLE); |
f62e00cc KM |
697 | } |
698 | ||
385e1ca5 DH |
699 | /** |
700 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
701 | * @page: Page defining the wait queue of interest |
702 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
703 | * |
704 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
705 | */ | |
706 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
707 | { | |
708 | wait_queue_head_t *q = page_waitqueue(page); | |
709 | unsigned long flags; | |
710 | ||
711 | spin_lock_irqsave(&q->lock, flags); | |
712 | __add_wait_queue(q, waiter); | |
713 | spin_unlock_irqrestore(&q->lock, flags); | |
714 | } | |
715 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
716 | ||
1da177e4 | 717 | /** |
485bb99b | 718 | * unlock_page - unlock a locked page |
1da177e4 LT |
719 | * @page: the page |
720 | * | |
721 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
722 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
723 | * mechananism between PageLocked pages and PageWriteback pages is shared. | |
724 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. | |
725 | * | |
8413ac9d NP |
726 | * The mb is necessary to enforce ordering between the clear_bit and the read |
727 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 728 | */ |
920c7a5d | 729 | void unlock_page(struct page *page) |
1da177e4 | 730 | { |
309381fe | 731 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
8413ac9d | 732 | clear_bit_unlock(PG_locked, &page->flags); |
4e857c58 | 733 | smp_mb__after_atomic(); |
1da177e4 LT |
734 | wake_up_page(page, PG_locked); |
735 | } | |
736 | EXPORT_SYMBOL(unlock_page); | |
737 | ||
485bb99b RD |
738 | /** |
739 | * end_page_writeback - end writeback against a page | |
740 | * @page: the page | |
1da177e4 LT |
741 | */ |
742 | void end_page_writeback(struct page *page) | |
743 | { | |
888cf2db MG |
744 | /* |
745 | * TestClearPageReclaim could be used here but it is an atomic | |
746 | * operation and overkill in this particular case. Failing to | |
747 | * shuffle a page marked for immediate reclaim is too mild to | |
748 | * justify taking an atomic operation penalty at the end of | |
749 | * ever page writeback. | |
750 | */ | |
751 | if (PageReclaim(page)) { | |
752 | ClearPageReclaim(page); | |
ac6aadb2 | 753 | rotate_reclaimable_page(page); |
888cf2db | 754 | } |
ac6aadb2 MS |
755 | |
756 | if (!test_clear_page_writeback(page)) | |
757 | BUG(); | |
758 | ||
4e857c58 | 759 | smp_mb__after_atomic(); |
1da177e4 LT |
760 | wake_up_page(page, PG_writeback); |
761 | } | |
762 | EXPORT_SYMBOL(end_page_writeback); | |
763 | ||
57d99845 MW |
764 | /* |
765 | * After completing I/O on a page, call this routine to update the page | |
766 | * flags appropriately | |
767 | */ | |
768 | void page_endio(struct page *page, int rw, int err) | |
769 | { | |
770 | if (rw == READ) { | |
771 | if (!err) { | |
772 | SetPageUptodate(page); | |
773 | } else { | |
774 | ClearPageUptodate(page); | |
775 | SetPageError(page); | |
776 | } | |
777 | unlock_page(page); | |
778 | } else { /* rw == WRITE */ | |
779 | if (err) { | |
780 | SetPageError(page); | |
781 | if (page->mapping) | |
782 | mapping_set_error(page->mapping, err); | |
783 | } | |
784 | end_page_writeback(page); | |
785 | } | |
786 | } | |
787 | EXPORT_SYMBOL_GPL(page_endio); | |
788 | ||
485bb99b RD |
789 | /** |
790 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
791 | * @page: the page to lock | |
1da177e4 | 792 | */ |
920c7a5d | 793 | void __lock_page(struct page *page) |
1da177e4 LT |
794 | { |
795 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
796 | ||
74316201 | 797 | __wait_on_bit_lock(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
798 | TASK_UNINTERRUPTIBLE); |
799 | } | |
800 | EXPORT_SYMBOL(__lock_page); | |
801 | ||
b5606c2d | 802 | int __lock_page_killable(struct page *page) |
2687a356 MW |
803 | { |
804 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
805 | ||
806 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
74316201 | 807 | bit_wait_io, TASK_KILLABLE); |
2687a356 | 808 | } |
18bc0bbd | 809 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 810 | |
d065bd81 ML |
811 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
812 | unsigned int flags) | |
813 | { | |
37b23e05 KM |
814 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
815 | /* | |
816 | * CAUTION! In this case, mmap_sem is not released | |
817 | * even though return 0. | |
818 | */ | |
819 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
820 | return 0; | |
821 | ||
822 | up_read(&mm->mmap_sem); | |
823 | if (flags & FAULT_FLAG_KILLABLE) | |
824 | wait_on_page_locked_killable(page); | |
825 | else | |
318b275f | 826 | wait_on_page_locked(page); |
d065bd81 | 827 | return 0; |
37b23e05 KM |
828 | } else { |
829 | if (flags & FAULT_FLAG_KILLABLE) { | |
830 | int ret; | |
831 | ||
832 | ret = __lock_page_killable(page); | |
833 | if (ret) { | |
834 | up_read(&mm->mmap_sem); | |
835 | return 0; | |
836 | } | |
837 | } else | |
838 | __lock_page(page); | |
839 | return 1; | |
d065bd81 ML |
840 | } |
841 | } | |
842 | ||
e7b563bb JW |
843 | /** |
844 | * page_cache_next_hole - find the next hole (not-present entry) | |
845 | * @mapping: mapping | |
846 | * @index: index | |
847 | * @max_scan: maximum range to search | |
848 | * | |
849 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
850 | * lowest indexed hole. | |
851 | * | |
852 | * Returns: the index of the hole if found, otherwise returns an index | |
853 | * outside of the set specified (in which case 'return - index >= | |
854 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
855 | * be returned. | |
856 | * | |
857 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
858 | * like radix_tree_gang_lookup, this will not atomically search a | |
859 | * snapshot of the tree at a single point in time. For example, if a | |
860 | * hole is created at index 5, then subsequently a hole is created at | |
861 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
862 | * if called under rcu_read_lock. | |
863 | */ | |
864 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
865 | pgoff_t index, unsigned long max_scan) | |
866 | { | |
867 | unsigned long i; | |
868 | ||
869 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
870 | struct page *page; |
871 | ||
872 | page = radix_tree_lookup(&mapping->page_tree, index); | |
873 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
874 | break; |
875 | index++; | |
876 | if (index == 0) | |
877 | break; | |
878 | } | |
879 | ||
880 | return index; | |
881 | } | |
882 | EXPORT_SYMBOL(page_cache_next_hole); | |
883 | ||
884 | /** | |
885 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
886 | * @mapping: mapping | |
887 | * @index: index | |
888 | * @max_scan: maximum range to search | |
889 | * | |
890 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
891 | * the first hole. | |
892 | * | |
893 | * Returns: the index of the hole if found, otherwise returns an index | |
894 | * outside of the set specified (in which case 'index - return >= | |
895 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
896 | * will be returned. | |
897 | * | |
898 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
899 | * like radix_tree_gang_lookup, this will not atomically search a | |
900 | * snapshot of the tree at a single point in time. For example, if a | |
901 | * hole is created at index 10, then subsequently a hole is created at | |
902 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
903 | * called under rcu_read_lock. | |
904 | */ | |
905 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
906 | pgoff_t index, unsigned long max_scan) | |
907 | { | |
908 | unsigned long i; | |
909 | ||
910 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
911 | struct page *page; |
912 | ||
913 | page = radix_tree_lookup(&mapping->page_tree, index); | |
914 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
915 | break; |
916 | index--; | |
917 | if (index == ULONG_MAX) | |
918 | break; | |
919 | } | |
920 | ||
921 | return index; | |
922 | } | |
923 | EXPORT_SYMBOL(page_cache_prev_hole); | |
924 | ||
485bb99b | 925 | /** |
0cd6144a | 926 | * find_get_entry - find and get a page cache entry |
485bb99b | 927 | * @mapping: the address_space to search |
0cd6144a JW |
928 | * @offset: the page cache index |
929 | * | |
930 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
931 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 932 | * |
139b6a6f JW |
933 | * If the slot holds a shadow entry of a previously evicted page, or a |
934 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
935 | * |
936 | * Otherwise, %NULL is returned. | |
1da177e4 | 937 | */ |
0cd6144a | 938 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 939 | { |
a60637c8 | 940 | void **pagep; |
1da177e4 LT |
941 | struct page *page; |
942 | ||
a60637c8 NP |
943 | rcu_read_lock(); |
944 | repeat: | |
945 | page = NULL; | |
946 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
947 | if (pagep) { | |
948 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
949 | if (unlikely(!page)) |
950 | goto out; | |
a2c16d6c | 951 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
952 | if (radix_tree_deref_retry(page)) |
953 | goto repeat; | |
954 | /* | |
139b6a6f JW |
955 | * A shadow entry of a recently evicted page, |
956 | * or a swap entry from shmem/tmpfs. Return | |
957 | * it without attempting to raise page count. | |
8079b1c8 HD |
958 | */ |
959 | goto out; | |
a2c16d6c | 960 | } |
a60637c8 NP |
961 | if (!page_cache_get_speculative(page)) |
962 | goto repeat; | |
963 | ||
964 | /* | |
965 | * Has the page moved? | |
966 | * This is part of the lockless pagecache protocol. See | |
967 | * include/linux/pagemap.h for details. | |
968 | */ | |
969 | if (unlikely(page != *pagep)) { | |
970 | page_cache_release(page); | |
971 | goto repeat; | |
972 | } | |
973 | } | |
27d20fdd | 974 | out: |
a60637c8 NP |
975 | rcu_read_unlock(); |
976 | ||
1da177e4 LT |
977 | return page; |
978 | } | |
0cd6144a | 979 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 980 | |
0cd6144a JW |
981 | /** |
982 | * find_lock_entry - locate, pin and lock a page cache entry | |
983 | * @mapping: the address_space to search | |
984 | * @offset: the page cache index | |
985 | * | |
986 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
987 | * page cache page, it is returned locked and with an increased | |
988 | * refcount. | |
989 | * | |
139b6a6f JW |
990 | * If the slot holds a shadow entry of a previously evicted page, or a |
991 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
992 | * |
993 | * Otherwise, %NULL is returned. | |
994 | * | |
995 | * find_lock_entry() may sleep. | |
996 | */ | |
997 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
998 | { |
999 | struct page *page; | |
1000 | ||
1da177e4 | 1001 | repeat: |
0cd6144a | 1002 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1003 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1004 | lock_page(page); |
1005 | /* Has the page been truncated? */ | |
1006 | if (unlikely(page->mapping != mapping)) { | |
1007 | unlock_page(page); | |
1008 | page_cache_release(page); | |
1009 | goto repeat; | |
1da177e4 | 1010 | } |
309381fe | 1011 | VM_BUG_ON_PAGE(page->index != offset, page); |
1da177e4 | 1012 | } |
1da177e4 LT |
1013 | return page; |
1014 | } | |
0cd6144a JW |
1015 | EXPORT_SYMBOL(find_lock_entry); |
1016 | ||
1017 | /** | |
2457aec6 | 1018 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1019 | * @mapping: the address_space to search |
1020 | * @offset: the page index | |
2457aec6 | 1021 | * @fgp_flags: PCG flags |
75325189 RD |
1022 | * @cache_gfp_mask: gfp mask to use for the page cache data page allocation |
1023 | * @radix_gfp_mask: gfp mask to use for radix tree node allocation | |
0cd6144a | 1024 | * |
2457aec6 | 1025 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1026 | * |
75325189 | 1027 | * PCG flags modify how the page is returned. |
0cd6144a | 1028 | * |
2457aec6 MG |
1029 | * FGP_ACCESSED: the page will be marked accessed |
1030 | * FGP_LOCK: Page is return locked | |
1031 | * FGP_CREAT: If page is not present then a new page is allocated using | |
75325189 RD |
1032 | * @cache_gfp_mask and added to the page cache and the VM's LRU |
1033 | * list. If radix tree nodes are allocated during page cache | |
1034 | * insertion then @radix_gfp_mask is used. The page is returned | |
1035 | * locked and with an increased refcount. Otherwise, %NULL is | |
1036 | * returned. | |
1da177e4 | 1037 | * |
2457aec6 MG |
1038 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1039 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1040 | * |
2457aec6 | 1041 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1042 | */ |
2457aec6 MG |
1043 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
1044 | int fgp_flags, gfp_t cache_gfp_mask, gfp_t radix_gfp_mask) | |
1da177e4 | 1045 | { |
eb2be189 | 1046 | struct page *page; |
2457aec6 | 1047 | |
1da177e4 | 1048 | repeat: |
2457aec6 MG |
1049 | page = find_get_entry(mapping, offset); |
1050 | if (radix_tree_exceptional_entry(page)) | |
1051 | page = NULL; | |
1052 | if (!page) | |
1053 | goto no_page; | |
1054 | ||
1055 | if (fgp_flags & FGP_LOCK) { | |
1056 | if (fgp_flags & FGP_NOWAIT) { | |
1057 | if (!trylock_page(page)) { | |
1058 | page_cache_release(page); | |
1059 | return NULL; | |
1060 | } | |
1061 | } else { | |
1062 | lock_page(page); | |
1063 | } | |
1064 | ||
1065 | /* Has the page been truncated? */ | |
1066 | if (unlikely(page->mapping != mapping)) { | |
1067 | unlock_page(page); | |
1068 | page_cache_release(page); | |
1069 | goto repeat; | |
1070 | } | |
1071 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1072 | } | |
1073 | ||
1074 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1075 | mark_page_accessed(page); | |
1076 | ||
1077 | no_page: | |
1078 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1079 | int err; | |
1080 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
1081 | cache_gfp_mask |= __GFP_WRITE; | |
1082 | if (fgp_flags & FGP_NOFS) { | |
1083 | cache_gfp_mask &= ~__GFP_FS; | |
1084 | radix_gfp_mask &= ~__GFP_FS; | |
1085 | } | |
1086 | ||
1087 | page = __page_cache_alloc(cache_gfp_mask); | |
eb2be189 NP |
1088 | if (!page) |
1089 | return NULL; | |
2457aec6 MG |
1090 | |
1091 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1092 | fgp_flags |= FGP_LOCK; | |
1093 | ||
eb39d618 | 1094 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1095 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1096 | __SetPageReferenced(page); |
2457aec6 MG |
1097 | |
1098 | err = add_to_page_cache_lru(page, mapping, offset, radix_gfp_mask); | |
eb2be189 NP |
1099 | if (unlikely(err)) { |
1100 | page_cache_release(page); | |
1101 | page = NULL; | |
1102 | if (err == -EEXIST) | |
1103 | goto repeat; | |
1da177e4 | 1104 | } |
1da177e4 | 1105 | } |
2457aec6 | 1106 | |
1da177e4 LT |
1107 | return page; |
1108 | } | |
2457aec6 | 1109 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1110 | |
0cd6144a JW |
1111 | /** |
1112 | * find_get_entries - gang pagecache lookup | |
1113 | * @mapping: The address_space to search | |
1114 | * @start: The starting page cache index | |
1115 | * @nr_entries: The maximum number of entries | |
1116 | * @entries: Where the resulting entries are placed | |
1117 | * @indices: The cache indices corresponding to the entries in @entries | |
1118 | * | |
1119 | * find_get_entries() will search for and return a group of up to | |
1120 | * @nr_entries entries in the mapping. The entries are placed at | |
1121 | * @entries. find_get_entries() takes a reference against any actual | |
1122 | * pages it returns. | |
1123 | * | |
1124 | * The search returns a group of mapping-contiguous page cache entries | |
1125 | * with ascending indexes. There may be holes in the indices due to | |
1126 | * not-present pages. | |
1127 | * | |
139b6a6f JW |
1128 | * Any shadow entries of evicted pages, or swap entries from |
1129 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1130 | * |
1131 | * find_get_entries() returns the number of pages and shadow entries | |
1132 | * which were found. | |
1133 | */ | |
1134 | unsigned find_get_entries(struct address_space *mapping, | |
1135 | pgoff_t start, unsigned int nr_entries, | |
1136 | struct page **entries, pgoff_t *indices) | |
1137 | { | |
1138 | void **slot; | |
1139 | unsigned int ret = 0; | |
1140 | struct radix_tree_iter iter; | |
1141 | ||
1142 | if (!nr_entries) | |
1143 | return 0; | |
1144 | ||
1145 | rcu_read_lock(); | |
1146 | restart: | |
1147 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { | |
1148 | struct page *page; | |
1149 | repeat: | |
1150 | page = radix_tree_deref_slot(slot); | |
1151 | if (unlikely(!page)) | |
1152 | continue; | |
1153 | if (radix_tree_exception(page)) { | |
1154 | if (radix_tree_deref_retry(page)) | |
1155 | goto restart; | |
1156 | /* | |
139b6a6f JW |
1157 | * A shadow entry of a recently evicted page, |
1158 | * or a swap entry from shmem/tmpfs. Return | |
1159 | * it without attempting to raise page count. | |
0cd6144a JW |
1160 | */ |
1161 | goto export; | |
1162 | } | |
1163 | if (!page_cache_get_speculative(page)) | |
1164 | goto repeat; | |
1165 | ||
1166 | /* Has the page moved? */ | |
1167 | if (unlikely(page != *slot)) { | |
1168 | page_cache_release(page); | |
1169 | goto repeat; | |
1170 | } | |
1171 | export: | |
1172 | indices[ret] = iter.index; | |
1173 | entries[ret] = page; | |
1174 | if (++ret == nr_entries) | |
1175 | break; | |
1176 | } | |
1177 | rcu_read_unlock(); | |
1178 | return ret; | |
1179 | } | |
1180 | ||
1da177e4 LT |
1181 | /** |
1182 | * find_get_pages - gang pagecache lookup | |
1183 | * @mapping: The address_space to search | |
1184 | * @start: The starting page index | |
1185 | * @nr_pages: The maximum number of pages | |
1186 | * @pages: Where the resulting pages are placed | |
1187 | * | |
1188 | * find_get_pages() will search for and return a group of up to | |
1189 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
1190 | * find_get_pages() takes a reference against the returned pages. | |
1191 | * | |
1192 | * The search returns a group of mapping-contiguous pages with ascending | |
1193 | * indexes. There may be holes in the indices due to not-present pages. | |
1194 | * | |
1195 | * find_get_pages() returns the number of pages which were found. | |
1196 | */ | |
1197 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
1198 | unsigned int nr_pages, struct page **pages) | |
1199 | { | |
0fc9d104 KK |
1200 | struct radix_tree_iter iter; |
1201 | void **slot; | |
1202 | unsigned ret = 0; | |
1203 | ||
1204 | if (unlikely(!nr_pages)) | |
1205 | return 0; | |
a60637c8 NP |
1206 | |
1207 | rcu_read_lock(); | |
1208 | restart: | |
0fc9d104 | 1209 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
a60637c8 NP |
1210 | struct page *page; |
1211 | repeat: | |
0fc9d104 | 1212 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1213 | if (unlikely(!page)) |
1214 | continue; | |
9d8aa4ea | 1215 | |
a2c16d6c | 1216 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1217 | if (radix_tree_deref_retry(page)) { |
1218 | /* | |
1219 | * Transient condition which can only trigger | |
1220 | * when entry at index 0 moves out of or back | |
1221 | * to root: none yet gotten, safe to restart. | |
1222 | */ | |
0fc9d104 | 1223 | WARN_ON(iter.index); |
8079b1c8 HD |
1224 | goto restart; |
1225 | } | |
a2c16d6c | 1226 | /* |
139b6a6f JW |
1227 | * A shadow entry of a recently evicted page, |
1228 | * or a swap entry from shmem/tmpfs. Skip | |
1229 | * over it. | |
a2c16d6c | 1230 | */ |
8079b1c8 | 1231 | continue; |
27d20fdd | 1232 | } |
a60637c8 NP |
1233 | |
1234 | if (!page_cache_get_speculative(page)) | |
1235 | goto repeat; | |
1236 | ||
1237 | /* Has the page moved? */ | |
0fc9d104 | 1238 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1239 | page_cache_release(page); |
1240 | goto repeat; | |
1241 | } | |
1da177e4 | 1242 | |
a60637c8 | 1243 | pages[ret] = page; |
0fc9d104 KK |
1244 | if (++ret == nr_pages) |
1245 | break; | |
a60637c8 | 1246 | } |
5b280c0c | 1247 | |
a60637c8 | 1248 | rcu_read_unlock(); |
1da177e4 LT |
1249 | return ret; |
1250 | } | |
1251 | ||
ebf43500 JA |
1252 | /** |
1253 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1254 | * @mapping: The address_space to search | |
1255 | * @index: The starting page index | |
1256 | * @nr_pages: The maximum number of pages | |
1257 | * @pages: Where the resulting pages are placed | |
1258 | * | |
1259 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1260 | * that the returned number of pages are guaranteed to be contiguous. | |
1261 | * | |
1262 | * find_get_pages_contig() returns the number of pages which were found. | |
1263 | */ | |
1264 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1265 | unsigned int nr_pages, struct page **pages) | |
1266 | { | |
0fc9d104 KK |
1267 | struct radix_tree_iter iter; |
1268 | void **slot; | |
1269 | unsigned int ret = 0; | |
1270 | ||
1271 | if (unlikely(!nr_pages)) | |
1272 | return 0; | |
a60637c8 NP |
1273 | |
1274 | rcu_read_lock(); | |
1275 | restart: | |
0fc9d104 | 1276 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
a60637c8 NP |
1277 | struct page *page; |
1278 | repeat: | |
0fc9d104 KK |
1279 | page = radix_tree_deref_slot(slot); |
1280 | /* The hole, there no reason to continue */ | |
a60637c8 | 1281 | if (unlikely(!page)) |
0fc9d104 | 1282 | break; |
9d8aa4ea | 1283 | |
a2c16d6c | 1284 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1285 | if (radix_tree_deref_retry(page)) { |
1286 | /* | |
1287 | * Transient condition which can only trigger | |
1288 | * when entry at index 0 moves out of or back | |
1289 | * to root: none yet gotten, safe to restart. | |
1290 | */ | |
1291 | goto restart; | |
1292 | } | |
a2c16d6c | 1293 | /* |
139b6a6f JW |
1294 | * A shadow entry of a recently evicted page, |
1295 | * or a swap entry from shmem/tmpfs. Stop | |
1296 | * looking for contiguous pages. | |
a2c16d6c | 1297 | */ |
8079b1c8 | 1298 | break; |
a2c16d6c | 1299 | } |
ebf43500 | 1300 | |
a60637c8 NP |
1301 | if (!page_cache_get_speculative(page)) |
1302 | goto repeat; | |
1303 | ||
1304 | /* Has the page moved? */ | |
0fc9d104 | 1305 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1306 | page_cache_release(page); |
1307 | goto repeat; | |
1308 | } | |
1309 | ||
9cbb4cb2 NP |
1310 | /* |
1311 | * must check mapping and index after taking the ref. | |
1312 | * otherwise we can get both false positives and false | |
1313 | * negatives, which is just confusing to the caller. | |
1314 | */ | |
0fc9d104 | 1315 | if (page->mapping == NULL || page->index != iter.index) { |
9cbb4cb2 NP |
1316 | page_cache_release(page); |
1317 | break; | |
1318 | } | |
1319 | ||
a60637c8 | 1320 | pages[ret] = page; |
0fc9d104 KK |
1321 | if (++ret == nr_pages) |
1322 | break; | |
ebf43500 | 1323 | } |
a60637c8 NP |
1324 | rcu_read_unlock(); |
1325 | return ret; | |
ebf43500 | 1326 | } |
ef71c15c | 1327 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1328 | |
485bb99b RD |
1329 | /** |
1330 | * find_get_pages_tag - find and return pages that match @tag | |
1331 | * @mapping: the address_space to search | |
1332 | * @index: the starting page index | |
1333 | * @tag: the tag index | |
1334 | * @nr_pages: the maximum number of pages | |
1335 | * @pages: where the resulting pages are placed | |
1336 | * | |
1da177e4 | 1337 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1338 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
1339 | */ |
1340 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
1341 | int tag, unsigned int nr_pages, struct page **pages) | |
1342 | { | |
0fc9d104 KK |
1343 | struct radix_tree_iter iter; |
1344 | void **slot; | |
1345 | unsigned ret = 0; | |
1346 | ||
1347 | if (unlikely(!nr_pages)) | |
1348 | return 0; | |
a60637c8 NP |
1349 | |
1350 | rcu_read_lock(); | |
1351 | restart: | |
0fc9d104 KK |
1352 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1353 | &iter, *index, tag) { | |
a60637c8 NP |
1354 | struct page *page; |
1355 | repeat: | |
0fc9d104 | 1356 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1357 | if (unlikely(!page)) |
1358 | continue; | |
9d8aa4ea | 1359 | |
a2c16d6c | 1360 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1361 | if (radix_tree_deref_retry(page)) { |
1362 | /* | |
1363 | * Transient condition which can only trigger | |
1364 | * when entry at index 0 moves out of or back | |
1365 | * to root: none yet gotten, safe to restart. | |
1366 | */ | |
1367 | goto restart; | |
1368 | } | |
a2c16d6c | 1369 | /* |
139b6a6f JW |
1370 | * A shadow entry of a recently evicted page. |
1371 | * | |
1372 | * Those entries should never be tagged, but | |
1373 | * this tree walk is lockless and the tags are | |
1374 | * looked up in bulk, one radix tree node at a | |
1375 | * time, so there is a sizable window for page | |
1376 | * reclaim to evict a page we saw tagged. | |
1377 | * | |
1378 | * Skip over it. | |
a2c16d6c | 1379 | */ |
139b6a6f | 1380 | continue; |
a2c16d6c | 1381 | } |
a60637c8 NP |
1382 | |
1383 | if (!page_cache_get_speculative(page)) | |
1384 | goto repeat; | |
1385 | ||
1386 | /* Has the page moved? */ | |
0fc9d104 | 1387 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1388 | page_cache_release(page); |
1389 | goto repeat; | |
1390 | } | |
1391 | ||
1392 | pages[ret] = page; | |
0fc9d104 KK |
1393 | if (++ret == nr_pages) |
1394 | break; | |
a60637c8 | 1395 | } |
5b280c0c | 1396 | |
a60637c8 | 1397 | rcu_read_unlock(); |
1da177e4 | 1398 | |
1da177e4 LT |
1399 | if (ret) |
1400 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1401 | |
1da177e4 LT |
1402 | return ret; |
1403 | } | |
ef71c15c | 1404 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1405 | |
76d42bd9 WF |
1406 | /* |
1407 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1408 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1409 | * | |
1410 | * ---R__________________________________________B__________ | |
1411 | * ^ reading here ^ bad block(assume 4k) | |
1412 | * | |
1413 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1414 | * => failing the whole request => read(R) => read(R+1) => | |
1415 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1416 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1417 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1418 | * | |
1419 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1420 | */ | |
1421 | static void shrink_readahead_size_eio(struct file *filp, | |
1422 | struct file_ra_state *ra) | |
1423 | { | |
76d42bd9 | 1424 | ra->ra_pages /= 4; |
76d42bd9 WF |
1425 | } |
1426 | ||
485bb99b | 1427 | /** |
36e78914 | 1428 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1429 | * @filp: the file to read |
1430 | * @ppos: current file position | |
6e58e79d AV |
1431 | * @iter: data destination |
1432 | * @written: already copied | |
485bb99b | 1433 | * |
1da177e4 | 1434 | * This is a generic file read routine, and uses the |
485bb99b | 1435 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1436 | * |
1437 | * This is really ugly. But the goto's actually try to clarify some | |
1438 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1439 | */ |
6e58e79d AV |
1440 | static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos, |
1441 | struct iov_iter *iter, ssize_t written) | |
1da177e4 | 1442 | { |
36e78914 | 1443 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1444 | struct inode *inode = mapping->host; |
36e78914 | 1445 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1446 | pgoff_t index; |
1447 | pgoff_t last_index; | |
1448 | pgoff_t prev_index; | |
1449 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1450 | unsigned int prev_offset; |
6e58e79d | 1451 | int error = 0; |
1da177e4 | 1452 | |
1da177e4 | 1453 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1454 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1455 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
6e58e79d | 1456 | last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1da177e4 LT |
1457 | offset = *ppos & ~PAGE_CACHE_MASK; |
1458 | ||
1da177e4 LT |
1459 | for (;;) { |
1460 | struct page *page; | |
57f6b96c | 1461 | pgoff_t end_index; |
a32ea1e1 | 1462 | loff_t isize; |
1da177e4 LT |
1463 | unsigned long nr, ret; |
1464 | ||
1da177e4 | 1465 | cond_resched(); |
1da177e4 LT |
1466 | find_page: |
1467 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1468 | if (!page) { |
cf914a7d | 1469 | page_cache_sync_readahead(mapping, |
7ff81078 | 1470 | ra, filp, |
3ea89ee8 FW |
1471 | index, last_index - index); |
1472 | page = find_get_page(mapping, index); | |
1473 | if (unlikely(page == NULL)) | |
1474 | goto no_cached_page; | |
1475 | } | |
1476 | if (PageReadahead(page)) { | |
cf914a7d | 1477 | page_cache_async_readahead(mapping, |
7ff81078 | 1478 | ra, filp, page, |
3ea89ee8 | 1479 | index, last_index - index); |
1da177e4 | 1480 | } |
8ab22b9a HH |
1481 | if (!PageUptodate(page)) { |
1482 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1483 | !mapping->a_ops->is_partially_uptodate) | |
1484 | goto page_not_up_to_date; | |
529ae9aa | 1485 | if (!trylock_page(page)) |
8ab22b9a | 1486 | goto page_not_up_to_date; |
8d056cb9 DH |
1487 | /* Did it get truncated before we got the lock? */ |
1488 | if (!page->mapping) | |
1489 | goto page_not_up_to_date_locked; | |
8ab22b9a | 1490 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 1491 | offset, iter->count)) |
8ab22b9a HH |
1492 | goto page_not_up_to_date_locked; |
1493 | unlock_page(page); | |
1494 | } | |
1da177e4 | 1495 | page_ok: |
a32ea1e1 N |
1496 | /* |
1497 | * i_size must be checked after we know the page is Uptodate. | |
1498 | * | |
1499 | * Checking i_size after the check allows us to calculate | |
1500 | * the correct value for "nr", which means the zero-filled | |
1501 | * part of the page is not copied back to userspace (unless | |
1502 | * another truncate extends the file - this is desired though). | |
1503 | */ | |
1504 | ||
1505 | isize = i_size_read(inode); | |
1506 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1507 | if (unlikely(!isize || index > end_index)) { | |
1508 | page_cache_release(page); | |
1509 | goto out; | |
1510 | } | |
1511 | ||
1512 | /* nr is the maximum number of bytes to copy from this page */ | |
1513 | nr = PAGE_CACHE_SIZE; | |
1514 | if (index == end_index) { | |
1515 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1516 | if (nr <= offset) { | |
1517 | page_cache_release(page); | |
1518 | goto out; | |
1519 | } | |
1520 | } | |
1521 | nr = nr - offset; | |
1da177e4 LT |
1522 | |
1523 | /* If users can be writing to this page using arbitrary | |
1524 | * virtual addresses, take care about potential aliasing | |
1525 | * before reading the page on the kernel side. | |
1526 | */ | |
1527 | if (mapping_writably_mapped(mapping)) | |
1528 | flush_dcache_page(page); | |
1529 | ||
1530 | /* | |
ec0f1637 JK |
1531 | * When a sequential read accesses a page several times, |
1532 | * only mark it as accessed the first time. | |
1da177e4 | 1533 | */ |
ec0f1637 | 1534 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1535 | mark_page_accessed(page); |
1536 | prev_index = index; | |
1537 | ||
1538 | /* | |
1539 | * Ok, we have the page, and it's up-to-date, so | |
1540 | * now we can copy it to user space... | |
1da177e4 | 1541 | */ |
6e58e79d AV |
1542 | |
1543 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 LT |
1544 | offset += ret; |
1545 | index += offset >> PAGE_CACHE_SHIFT; | |
1546 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1547 | prev_offset = offset; |
1da177e4 LT |
1548 | |
1549 | page_cache_release(page); | |
6e58e79d AV |
1550 | written += ret; |
1551 | if (!iov_iter_count(iter)) | |
1552 | goto out; | |
1553 | if (ret < nr) { | |
1554 | error = -EFAULT; | |
1555 | goto out; | |
1556 | } | |
1557 | continue; | |
1da177e4 LT |
1558 | |
1559 | page_not_up_to_date: | |
1560 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1561 | error = lock_page_killable(page); |
1562 | if (unlikely(error)) | |
1563 | goto readpage_error; | |
1da177e4 | 1564 | |
8ab22b9a | 1565 | page_not_up_to_date_locked: |
da6052f7 | 1566 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1567 | if (!page->mapping) { |
1568 | unlock_page(page); | |
1569 | page_cache_release(page); | |
1570 | continue; | |
1571 | } | |
1572 | ||
1573 | /* Did somebody else fill it already? */ | |
1574 | if (PageUptodate(page)) { | |
1575 | unlock_page(page); | |
1576 | goto page_ok; | |
1577 | } | |
1578 | ||
1579 | readpage: | |
91803b49 JM |
1580 | /* |
1581 | * A previous I/O error may have been due to temporary | |
1582 | * failures, eg. multipath errors. | |
1583 | * PG_error will be set again if readpage fails. | |
1584 | */ | |
1585 | ClearPageError(page); | |
1da177e4 LT |
1586 | /* Start the actual read. The read will unlock the page. */ |
1587 | error = mapping->a_ops->readpage(filp, page); | |
1588 | ||
994fc28c ZB |
1589 | if (unlikely(error)) { |
1590 | if (error == AOP_TRUNCATED_PAGE) { | |
1591 | page_cache_release(page); | |
6e58e79d | 1592 | error = 0; |
994fc28c ZB |
1593 | goto find_page; |
1594 | } | |
1da177e4 | 1595 | goto readpage_error; |
994fc28c | 1596 | } |
1da177e4 LT |
1597 | |
1598 | if (!PageUptodate(page)) { | |
85462323 ON |
1599 | error = lock_page_killable(page); |
1600 | if (unlikely(error)) | |
1601 | goto readpage_error; | |
1da177e4 LT |
1602 | if (!PageUptodate(page)) { |
1603 | if (page->mapping == NULL) { | |
1604 | /* | |
2ecdc82e | 1605 | * invalidate_mapping_pages got it |
1da177e4 LT |
1606 | */ |
1607 | unlock_page(page); | |
1608 | page_cache_release(page); | |
1609 | goto find_page; | |
1610 | } | |
1611 | unlock_page(page); | |
7ff81078 | 1612 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1613 | error = -EIO; |
1614 | goto readpage_error; | |
1da177e4 LT |
1615 | } |
1616 | unlock_page(page); | |
1617 | } | |
1618 | ||
1da177e4 LT |
1619 | goto page_ok; |
1620 | ||
1621 | readpage_error: | |
1622 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1da177e4 LT |
1623 | page_cache_release(page); |
1624 | goto out; | |
1625 | ||
1626 | no_cached_page: | |
1627 | /* | |
1628 | * Ok, it wasn't cached, so we need to create a new | |
1629 | * page.. | |
1630 | */ | |
eb2be189 NP |
1631 | page = page_cache_alloc_cold(mapping); |
1632 | if (!page) { | |
6e58e79d | 1633 | error = -ENOMEM; |
eb2be189 | 1634 | goto out; |
1da177e4 | 1635 | } |
eb2be189 | 1636 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1637 | index, GFP_KERNEL); |
1638 | if (error) { | |
eb2be189 | 1639 | page_cache_release(page); |
6e58e79d AV |
1640 | if (error == -EEXIST) { |
1641 | error = 0; | |
1da177e4 | 1642 | goto find_page; |
6e58e79d | 1643 | } |
1da177e4 LT |
1644 | goto out; |
1645 | } | |
1da177e4 LT |
1646 | goto readpage; |
1647 | } | |
1648 | ||
1649 | out: | |
7ff81078 FW |
1650 | ra->prev_pos = prev_index; |
1651 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1652 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1653 | |
f4e6b498 | 1654 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1655 | file_accessed(filp); |
6e58e79d | 1656 | return written ? written : error; |
1da177e4 LT |
1657 | } |
1658 | ||
485bb99b | 1659 | /** |
6abd2322 | 1660 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 1661 | * @iocb: kernel I/O control block |
6abd2322 | 1662 | * @iter: destination for the data read |
485bb99b | 1663 | * |
6abd2322 | 1664 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
1665 | * that can use the page cache directly. |
1666 | */ | |
1667 | ssize_t | |
ed978a81 | 1668 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 1669 | { |
ed978a81 | 1670 | struct file *file = iocb->ki_filp; |
cb66a7a1 | 1671 | ssize_t retval = 0; |
543ade1f | 1672 | loff_t *ppos = &iocb->ki_pos; |
ed978a81 | 1673 | loff_t pos = *ppos; |
1da177e4 LT |
1674 | |
1675 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
ed978a81 AV |
1676 | if (file->f_flags & O_DIRECT) { |
1677 | struct address_space *mapping = file->f_mapping; | |
1678 | struct inode *inode = mapping->host; | |
1679 | size_t count = iov_iter_count(iter); | |
543ade1f | 1680 | loff_t size; |
1da177e4 | 1681 | |
1da177e4 LT |
1682 | if (!count) |
1683 | goto out; /* skip atime */ | |
1684 | size = i_size_read(inode); | |
9fe55eea | 1685 | retval = filemap_write_and_wait_range(mapping, pos, |
a6cbcd4a | 1686 | pos + count - 1); |
9fe55eea | 1687 | if (!retval) { |
ed978a81 | 1688 | struct iov_iter data = *iter; |
26978b8b | 1689 | retval = mapping->a_ops->direct_IO(READ, iocb, &data, pos); |
9fe55eea | 1690 | } |
d8d3d94b | 1691 | |
9fe55eea SW |
1692 | if (retval > 0) { |
1693 | *ppos = pos + retval; | |
ed978a81 | 1694 | iov_iter_advance(iter, retval); |
9fe55eea | 1695 | } |
66f998f6 | 1696 | |
9fe55eea SW |
1697 | /* |
1698 | * Btrfs can have a short DIO read if we encounter | |
1699 | * compressed extents, so if there was an error, or if | |
1700 | * we've already read everything we wanted to, or if | |
1701 | * there was a short read because we hit EOF, go ahead | |
1702 | * and return. Otherwise fallthrough to buffered io for | |
1703 | * the rest of the read. | |
1704 | */ | |
ed978a81 AV |
1705 | if (retval < 0 || !iov_iter_count(iter) || *ppos >= size) { |
1706 | file_accessed(file); | |
9fe55eea | 1707 | goto out; |
0e0bcae3 | 1708 | } |
1da177e4 LT |
1709 | } |
1710 | ||
ed978a81 | 1711 | retval = do_generic_file_read(file, ppos, iter, retval); |
1da177e4 LT |
1712 | out: |
1713 | return retval; | |
1714 | } | |
ed978a81 | 1715 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 1716 | |
1da177e4 | 1717 | #ifdef CONFIG_MMU |
485bb99b RD |
1718 | /** |
1719 | * page_cache_read - adds requested page to the page cache if not already there | |
1720 | * @file: file to read | |
1721 | * @offset: page index | |
1722 | * | |
1da177e4 LT |
1723 | * This adds the requested page to the page cache if it isn't already there, |
1724 | * and schedules an I/O to read in its contents from disk. | |
1725 | */ | |
920c7a5d | 1726 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1727 | { |
1728 | struct address_space *mapping = file->f_mapping; | |
1729 | struct page *page; | |
994fc28c | 1730 | int ret; |
1da177e4 | 1731 | |
994fc28c ZB |
1732 | do { |
1733 | page = page_cache_alloc_cold(mapping); | |
1734 | if (!page) | |
1735 | return -ENOMEM; | |
1736 | ||
1737 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); | |
1738 | if (ret == 0) | |
1739 | ret = mapping->a_ops->readpage(file, page); | |
1740 | else if (ret == -EEXIST) | |
1741 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1742 | |
1da177e4 | 1743 | page_cache_release(page); |
1da177e4 | 1744 | |
994fc28c ZB |
1745 | } while (ret == AOP_TRUNCATED_PAGE); |
1746 | ||
1747 | return ret; | |
1da177e4 LT |
1748 | } |
1749 | ||
1750 | #define MMAP_LOTSAMISS (100) | |
1751 | ||
ef00e08e LT |
1752 | /* |
1753 | * Synchronous readahead happens when we don't even find | |
1754 | * a page in the page cache at all. | |
1755 | */ | |
1756 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1757 | struct file_ra_state *ra, | |
1758 | struct file *file, | |
1759 | pgoff_t offset) | |
1760 | { | |
1761 | unsigned long ra_pages; | |
1762 | struct address_space *mapping = file->f_mapping; | |
1763 | ||
1764 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1765 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 1766 | return; |
275b12bf WF |
1767 | if (!ra->ra_pages) |
1768 | return; | |
ef00e08e | 1769 | |
64363aad | 1770 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
1771 | page_cache_sync_readahead(mapping, ra, file, offset, |
1772 | ra->ra_pages); | |
ef00e08e LT |
1773 | return; |
1774 | } | |
1775 | ||
207d04ba AK |
1776 | /* Avoid banging the cache line if not needed */ |
1777 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1778 | ra->mmap_miss++; |
1779 | ||
1780 | /* | |
1781 | * Do we miss much more than hit in this file? If so, | |
1782 | * stop bothering with read-ahead. It will only hurt. | |
1783 | */ | |
1784 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1785 | return; | |
1786 | ||
d30a1100 WF |
1787 | /* |
1788 | * mmap read-around | |
1789 | */ | |
ef00e08e | 1790 | ra_pages = max_sane_readahead(ra->ra_pages); |
275b12bf WF |
1791 | ra->start = max_t(long, 0, offset - ra_pages / 2); |
1792 | ra->size = ra_pages; | |
2cbea1d3 | 1793 | ra->async_size = ra_pages / 4; |
275b12bf | 1794 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1795 | } |
1796 | ||
1797 | /* | |
1798 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1799 | * so we want to possibly extend the readahead further.. | |
1800 | */ | |
1801 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1802 | struct file_ra_state *ra, | |
1803 | struct file *file, | |
1804 | struct page *page, | |
1805 | pgoff_t offset) | |
1806 | { | |
1807 | struct address_space *mapping = file->f_mapping; | |
1808 | ||
1809 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1810 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
1811 | return; |
1812 | if (ra->mmap_miss > 0) | |
1813 | ra->mmap_miss--; | |
1814 | if (PageReadahead(page)) | |
2fad6f5d WF |
1815 | page_cache_async_readahead(mapping, ra, file, |
1816 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1817 | } |
1818 | ||
485bb99b | 1819 | /** |
54cb8821 | 1820 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1821 | * @vma: vma in which the fault was taken |
1822 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1823 | * |
54cb8821 | 1824 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1825 | * mapped memory region to read in file data during a page fault. |
1826 | * | |
1827 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1828 | * it in the page cache, and handles the special cases reasonably without | |
1829 | * having a lot of duplicated code. | |
1830 | */ | |
d0217ac0 | 1831 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1832 | { |
1833 | int error; | |
54cb8821 | 1834 | struct file *file = vma->vm_file; |
1da177e4 LT |
1835 | struct address_space *mapping = file->f_mapping; |
1836 | struct file_ra_state *ra = &file->f_ra; | |
1837 | struct inode *inode = mapping->host; | |
ef00e08e | 1838 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 1839 | struct page *page; |
99e3e53f | 1840 | loff_t size; |
83c54070 | 1841 | int ret = 0; |
1da177e4 | 1842 | |
99e3e53f KS |
1843 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
1844 | if (offset >= size >> PAGE_CACHE_SHIFT) | |
5307cc1a | 1845 | return VM_FAULT_SIGBUS; |
1da177e4 | 1846 | |
1da177e4 | 1847 | /* |
49426420 | 1848 | * Do we have something in the page cache already? |
1da177e4 | 1849 | */ |
ef00e08e | 1850 | page = find_get_page(mapping, offset); |
45cac65b | 1851 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 1852 | /* |
ef00e08e LT |
1853 | * We found the page, so try async readahead before |
1854 | * waiting for the lock. | |
1da177e4 | 1855 | */ |
ef00e08e | 1856 | do_async_mmap_readahead(vma, ra, file, page, offset); |
45cac65b | 1857 | } else if (!page) { |
ef00e08e LT |
1858 | /* No page in the page cache at all */ |
1859 | do_sync_mmap_readahead(vma, ra, file, offset); | |
1860 | count_vm_event(PGMAJFAULT); | |
456f998e | 1861 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
1862 | ret = VM_FAULT_MAJOR; |
1863 | retry_find: | |
b522c94d | 1864 | page = find_get_page(mapping, offset); |
1da177e4 LT |
1865 | if (!page) |
1866 | goto no_cached_page; | |
1867 | } | |
1868 | ||
d88c0922 ML |
1869 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
1870 | page_cache_release(page); | |
d065bd81 | 1871 | return ret | VM_FAULT_RETRY; |
d88c0922 | 1872 | } |
b522c94d ML |
1873 | |
1874 | /* Did it get truncated? */ | |
1875 | if (unlikely(page->mapping != mapping)) { | |
1876 | unlock_page(page); | |
1877 | put_page(page); | |
1878 | goto retry_find; | |
1879 | } | |
309381fe | 1880 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 1881 | |
1da177e4 | 1882 | /* |
d00806b1 NP |
1883 | * We have a locked page in the page cache, now we need to check |
1884 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1885 | */ |
d00806b1 | 1886 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1887 | goto page_not_uptodate; |
1888 | ||
ef00e08e LT |
1889 | /* |
1890 | * Found the page and have a reference on it. | |
1891 | * We must recheck i_size under page lock. | |
1892 | */ | |
99e3e53f KS |
1893 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
1894 | if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) { | |
d00806b1 | 1895 | unlock_page(page); |
745ad48e | 1896 | page_cache_release(page); |
5307cc1a | 1897 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1898 | } |
1899 | ||
d0217ac0 | 1900 | vmf->page = page; |
83c54070 | 1901 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 1902 | |
1da177e4 LT |
1903 | no_cached_page: |
1904 | /* | |
1905 | * We're only likely to ever get here if MADV_RANDOM is in | |
1906 | * effect. | |
1907 | */ | |
ef00e08e | 1908 | error = page_cache_read(file, offset); |
1da177e4 LT |
1909 | |
1910 | /* | |
1911 | * The page we want has now been added to the page cache. | |
1912 | * In the unlikely event that someone removed it in the | |
1913 | * meantime, we'll just come back here and read it again. | |
1914 | */ | |
1915 | if (error >= 0) | |
1916 | goto retry_find; | |
1917 | ||
1918 | /* | |
1919 | * An error return from page_cache_read can result if the | |
1920 | * system is low on memory, or a problem occurs while trying | |
1921 | * to schedule I/O. | |
1922 | */ | |
1923 | if (error == -ENOMEM) | |
d0217ac0 NP |
1924 | return VM_FAULT_OOM; |
1925 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1926 | |
1927 | page_not_uptodate: | |
1da177e4 LT |
1928 | /* |
1929 | * Umm, take care of errors if the page isn't up-to-date. | |
1930 | * Try to re-read it _once_. We do this synchronously, | |
1931 | * because there really aren't any performance issues here | |
1932 | * and we need to check for errors. | |
1933 | */ | |
1da177e4 | 1934 | ClearPageError(page); |
994fc28c | 1935 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
1936 | if (!error) { |
1937 | wait_on_page_locked(page); | |
1938 | if (!PageUptodate(page)) | |
1939 | error = -EIO; | |
1940 | } | |
d00806b1 NP |
1941 | page_cache_release(page); |
1942 | ||
1943 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 1944 | goto retry_find; |
1da177e4 | 1945 | |
d00806b1 | 1946 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 1947 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 1948 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
1949 | } |
1950 | EXPORT_SYMBOL(filemap_fault); | |
1951 | ||
f1820361 KS |
1952 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
1953 | { | |
1954 | struct radix_tree_iter iter; | |
1955 | void **slot; | |
1956 | struct file *file = vma->vm_file; | |
1957 | struct address_space *mapping = file->f_mapping; | |
1958 | loff_t size; | |
1959 | struct page *page; | |
1960 | unsigned long address = (unsigned long) vmf->virtual_address; | |
1961 | unsigned long addr; | |
1962 | pte_t *pte; | |
1963 | ||
1964 | rcu_read_lock(); | |
1965 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) { | |
1966 | if (iter.index > vmf->max_pgoff) | |
1967 | break; | |
1968 | repeat: | |
1969 | page = radix_tree_deref_slot(slot); | |
1970 | if (unlikely(!page)) | |
1971 | goto next; | |
1972 | if (radix_tree_exception(page)) { | |
1973 | if (radix_tree_deref_retry(page)) | |
1974 | break; | |
1975 | else | |
1976 | goto next; | |
1977 | } | |
1978 | ||
1979 | if (!page_cache_get_speculative(page)) | |
1980 | goto repeat; | |
1981 | ||
1982 | /* Has the page moved? */ | |
1983 | if (unlikely(page != *slot)) { | |
1984 | page_cache_release(page); | |
1985 | goto repeat; | |
1986 | } | |
1987 | ||
1988 | if (!PageUptodate(page) || | |
1989 | PageReadahead(page) || | |
1990 | PageHWPoison(page)) | |
1991 | goto skip; | |
1992 | if (!trylock_page(page)) | |
1993 | goto skip; | |
1994 | ||
1995 | if (page->mapping != mapping || !PageUptodate(page)) | |
1996 | goto unlock; | |
1997 | ||
99e3e53f KS |
1998 | size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE); |
1999 | if (page->index >= size >> PAGE_CACHE_SHIFT) | |
f1820361 KS |
2000 | goto unlock; |
2001 | ||
2002 | pte = vmf->pte + page->index - vmf->pgoff; | |
2003 | if (!pte_none(*pte)) | |
2004 | goto unlock; | |
2005 | ||
2006 | if (file->f_ra.mmap_miss > 0) | |
2007 | file->f_ra.mmap_miss--; | |
2008 | addr = address + (page->index - vmf->pgoff) * PAGE_SIZE; | |
2009 | do_set_pte(vma, addr, page, pte, false, false); | |
2010 | unlock_page(page); | |
2011 | goto next; | |
2012 | unlock: | |
2013 | unlock_page(page); | |
2014 | skip: | |
2015 | page_cache_release(page); | |
2016 | next: | |
2017 | if (iter.index == vmf->max_pgoff) | |
2018 | break; | |
2019 | } | |
2020 | rcu_read_unlock(); | |
2021 | } | |
2022 | EXPORT_SYMBOL(filemap_map_pages); | |
2023 | ||
4fcf1c62 JK |
2024 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2025 | { | |
2026 | struct page *page = vmf->page; | |
496ad9aa | 2027 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
2028 | int ret = VM_FAULT_LOCKED; |
2029 | ||
14da9200 | 2030 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
2031 | file_update_time(vma->vm_file); |
2032 | lock_page(page); | |
2033 | if (page->mapping != inode->i_mapping) { | |
2034 | unlock_page(page); | |
2035 | ret = VM_FAULT_NOPAGE; | |
2036 | goto out; | |
2037 | } | |
14da9200 JK |
2038 | /* |
2039 | * We mark the page dirty already here so that when freeze is in | |
2040 | * progress, we are guaranteed that writeback during freezing will | |
2041 | * see the dirty page and writeprotect it again. | |
2042 | */ | |
2043 | set_page_dirty(page); | |
1d1d1a76 | 2044 | wait_for_stable_page(page); |
4fcf1c62 | 2045 | out: |
14da9200 | 2046 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2047 | return ret; |
2048 | } | |
2049 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2050 | ||
f0f37e2f | 2051 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2052 | .fault = filemap_fault, |
f1820361 | 2053 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2054 | .page_mkwrite = filemap_page_mkwrite, |
0b173bc4 | 2055 | .remap_pages = generic_file_remap_pages, |
1da177e4 LT |
2056 | }; |
2057 | ||
2058 | /* This is used for a general mmap of a disk file */ | |
2059 | ||
2060 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2061 | { | |
2062 | struct address_space *mapping = file->f_mapping; | |
2063 | ||
2064 | if (!mapping->a_ops->readpage) | |
2065 | return -ENOEXEC; | |
2066 | file_accessed(file); | |
2067 | vma->vm_ops = &generic_file_vm_ops; | |
2068 | return 0; | |
2069 | } | |
1da177e4 LT |
2070 | |
2071 | /* | |
2072 | * This is for filesystems which do not implement ->writepage. | |
2073 | */ | |
2074 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2075 | { | |
2076 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2077 | return -EINVAL; | |
2078 | return generic_file_mmap(file, vma); | |
2079 | } | |
2080 | #else | |
2081 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2082 | { | |
2083 | return -ENOSYS; | |
2084 | } | |
2085 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2086 | { | |
2087 | return -ENOSYS; | |
2088 | } | |
2089 | #endif /* CONFIG_MMU */ | |
2090 | ||
2091 | EXPORT_SYMBOL(generic_file_mmap); | |
2092 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2093 | ||
67f9fd91 SL |
2094 | static struct page *wait_on_page_read(struct page *page) |
2095 | { | |
2096 | if (!IS_ERR(page)) { | |
2097 | wait_on_page_locked(page); | |
2098 | if (!PageUptodate(page)) { | |
2099 | page_cache_release(page); | |
2100 | page = ERR_PTR(-EIO); | |
2101 | } | |
2102 | } | |
2103 | return page; | |
2104 | } | |
2105 | ||
6fe6900e | 2106 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 2107 | pgoff_t index, |
5e5358e7 | 2108 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2109 | void *data, |
2110 | gfp_t gfp) | |
1da177e4 | 2111 | { |
eb2be189 | 2112 | struct page *page; |
1da177e4 LT |
2113 | int err; |
2114 | repeat: | |
2115 | page = find_get_page(mapping, index); | |
2116 | if (!page) { | |
0531b2aa | 2117 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
2118 | if (!page) |
2119 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2120 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 NP |
2121 | if (unlikely(err)) { |
2122 | page_cache_release(page); | |
2123 | if (err == -EEXIST) | |
2124 | goto repeat; | |
1da177e4 | 2125 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2126 | return ERR_PTR(err); |
2127 | } | |
1da177e4 LT |
2128 | err = filler(data, page); |
2129 | if (err < 0) { | |
2130 | page_cache_release(page); | |
2131 | page = ERR_PTR(err); | |
67f9fd91 SL |
2132 | } else { |
2133 | page = wait_on_page_read(page); | |
1da177e4 LT |
2134 | } |
2135 | } | |
1da177e4 LT |
2136 | return page; |
2137 | } | |
2138 | ||
0531b2aa | 2139 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2140 | pgoff_t index, |
5e5358e7 | 2141 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2142 | void *data, |
2143 | gfp_t gfp) | |
2144 | ||
1da177e4 LT |
2145 | { |
2146 | struct page *page; | |
2147 | int err; | |
2148 | ||
2149 | retry: | |
0531b2aa | 2150 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 2151 | if (IS_ERR(page)) |
c855ff37 | 2152 | return page; |
1da177e4 LT |
2153 | if (PageUptodate(page)) |
2154 | goto out; | |
2155 | ||
2156 | lock_page(page); | |
2157 | if (!page->mapping) { | |
2158 | unlock_page(page); | |
2159 | page_cache_release(page); | |
2160 | goto retry; | |
2161 | } | |
2162 | if (PageUptodate(page)) { | |
2163 | unlock_page(page); | |
2164 | goto out; | |
2165 | } | |
2166 | err = filler(data, page); | |
2167 | if (err < 0) { | |
2168 | page_cache_release(page); | |
c855ff37 | 2169 | return ERR_PTR(err); |
67f9fd91 SL |
2170 | } else { |
2171 | page = wait_on_page_read(page); | |
2172 | if (IS_ERR(page)) | |
2173 | return page; | |
1da177e4 | 2174 | } |
c855ff37 | 2175 | out: |
6fe6900e NP |
2176 | mark_page_accessed(page); |
2177 | return page; | |
2178 | } | |
0531b2aa LT |
2179 | |
2180 | /** | |
67f9fd91 | 2181 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2182 | * @mapping: the page's address_space |
2183 | * @index: the page index | |
2184 | * @filler: function to perform the read | |
5e5358e7 | 2185 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2186 | * |
0531b2aa | 2187 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2188 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2189 | * |
2190 | * If the page does not get brought uptodate, return -EIO. | |
2191 | */ | |
67f9fd91 | 2192 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2193 | pgoff_t index, |
5e5358e7 | 2194 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2195 | void *data) |
2196 | { | |
2197 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2198 | } | |
67f9fd91 | 2199 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2200 | |
2201 | /** | |
2202 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2203 | * @mapping: the page's address_space | |
2204 | * @index: the page index | |
2205 | * @gfp: the page allocator flags to use if allocating | |
2206 | * | |
2207 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2208 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2209 | * |
2210 | * If the page does not get brought uptodate, return -EIO. | |
2211 | */ | |
2212 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2213 | pgoff_t index, | |
2214 | gfp_t gfp) | |
2215 | { | |
2216 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2217 | ||
67f9fd91 | 2218 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2219 | } |
2220 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2221 | ||
1da177e4 LT |
2222 | /* |
2223 | * Performs necessary checks before doing a write | |
2224 | * | |
485bb99b | 2225 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2226 | * Returns appropriate error code that caller should return or |
2227 | * zero in case that write should be allowed. | |
2228 | */ | |
2229 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) | |
2230 | { | |
2231 | struct inode *inode = file->f_mapping->host; | |
59e99e5b | 2232 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
1da177e4 LT |
2233 | |
2234 | if (unlikely(*pos < 0)) | |
2235 | return -EINVAL; | |
2236 | ||
1da177e4 LT |
2237 | if (!isblk) { |
2238 | /* FIXME: this is for backwards compatibility with 2.4 */ | |
2239 | if (file->f_flags & O_APPEND) | |
2240 | *pos = i_size_read(inode); | |
2241 | ||
2242 | if (limit != RLIM_INFINITY) { | |
2243 | if (*pos >= limit) { | |
2244 | send_sig(SIGXFSZ, current, 0); | |
2245 | return -EFBIG; | |
2246 | } | |
2247 | if (*count > limit - (typeof(limit))*pos) { | |
2248 | *count = limit - (typeof(limit))*pos; | |
2249 | } | |
2250 | } | |
2251 | } | |
2252 | ||
2253 | /* | |
2254 | * LFS rule | |
2255 | */ | |
2256 | if (unlikely(*pos + *count > MAX_NON_LFS && | |
2257 | !(file->f_flags & O_LARGEFILE))) { | |
2258 | if (*pos >= MAX_NON_LFS) { | |
1da177e4 LT |
2259 | return -EFBIG; |
2260 | } | |
2261 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { | |
2262 | *count = MAX_NON_LFS - (unsigned long)*pos; | |
2263 | } | |
2264 | } | |
2265 | ||
2266 | /* | |
2267 | * Are we about to exceed the fs block limit ? | |
2268 | * | |
2269 | * If we have written data it becomes a short write. If we have | |
2270 | * exceeded without writing data we send a signal and return EFBIG. | |
2271 | * Linus frestrict idea will clean these up nicely.. | |
2272 | */ | |
2273 | if (likely(!isblk)) { | |
2274 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { | |
2275 | if (*count || *pos > inode->i_sb->s_maxbytes) { | |
1da177e4 LT |
2276 | return -EFBIG; |
2277 | } | |
2278 | /* zero-length writes at ->s_maxbytes are OK */ | |
2279 | } | |
2280 | ||
2281 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) | |
2282 | *count = inode->i_sb->s_maxbytes - *pos; | |
2283 | } else { | |
9361401e | 2284 | #ifdef CONFIG_BLOCK |
1da177e4 LT |
2285 | loff_t isize; |
2286 | if (bdev_read_only(I_BDEV(inode))) | |
2287 | return -EPERM; | |
2288 | isize = i_size_read(inode); | |
2289 | if (*pos >= isize) { | |
2290 | if (*count || *pos > isize) | |
2291 | return -ENOSPC; | |
2292 | } | |
2293 | ||
2294 | if (*pos + *count > isize) | |
2295 | *count = isize - *pos; | |
9361401e DH |
2296 | #else |
2297 | return -EPERM; | |
2298 | #endif | |
1da177e4 LT |
2299 | } |
2300 | return 0; | |
2301 | } | |
2302 | EXPORT_SYMBOL(generic_write_checks); | |
2303 | ||
afddba49 NP |
2304 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2305 | loff_t pos, unsigned len, unsigned flags, | |
2306 | struct page **pagep, void **fsdata) | |
2307 | { | |
2308 | const struct address_space_operations *aops = mapping->a_ops; | |
2309 | ||
4e02ed4b | 2310 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2311 | pagep, fsdata); |
afddba49 NP |
2312 | } |
2313 | EXPORT_SYMBOL(pagecache_write_begin); | |
2314 | ||
2315 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2316 | loff_t pos, unsigned len, unsigned copied, | |
2317 | struct page *page, void *fsdata) | |
2318 | { | |
2319 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2320 | |
4e02ed4b | 2321 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2322 | } |
2323 | EXPORT_SYMBOL(pagecache_write_end); | |
2324 | ||
1da177e4 | 2325 | ssize_t |
0c949334 | 2326 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos) |
1da177e4 LT |
2327 | { |
2328 | struct file *file = iocb->ki_filp; | |
2329 | struct address_space *mapping = file->f_mapping; | |
2330 | struct inode *inode = mapping->host; | |
2331 | ssize_t written; | |
a969e903 CH |
2332 | size_t write_len; |
2333 | pgoff_t end; | |
26978b8b | 2334 | struct iov_iter data; |
1da177e4 | 2335 | |
0c949334 | 2336 | write_len = iov_iter_count(from); |
a969e903 | 2337 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; |
a969e903 | 2338 | |
48b47c56 | 2339 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2340 | if (written) |
2341 | goto out; | |
2342 | ||
2343 | /* | |
2344 | * After a write we want buffered reads to be sure to go to disk to get | |
2345 | * the new data. We invalidate clean cached page from the region we're | |
2346 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2347 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2348 | */ |
2349 | if (mapping->nrpages) { | |
2350 | written = invalidate_inode_pages2_range(mapping, | |
2351 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2352 | /* |
2353 | * If a page can not be invalidated, return 0 to fall back | |
2354 | * to buffered write. | |
2355 | */ | |
2356 | if (written) { | |
2357 | if (written == -EBUSY) | |
2358 | return 0; | |
a969e903 | 2359 | goto out; |
6ccfa806 | 2360 | } |
a969e903 CH |
2361 | } |
2362 | ||
26978b8b AV |
2363 | data = *from; |
2364 | written = mapping->a_ops->direct_IO(WRITE, iocb, &data, pos); | |
a969e903 CH |
2365 | |
2366 | /* | |
2367 | * Finally, try again to invalidate clean pages which might have been | |
2368 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2369 | * if the source of the write was an mmap'ed region of the file | |
2370 | * we're writing. Either one is a pretty crazy thing to do, | |
2371 | * so we don't support it 100%. If this invalidation | |
2372 | * fails, tough, the write still worked... | |
2373 | */ | |
2374 | if (mapping->nrpages) { | |
2375 | invalidate_inode_pages2_range(mapping, | |
2376 | pos >> PAGE_CACHE_SHIFT, end); | |
2377 | } | |
2378 | ||
1da177e4 | 2379 | if (written > 0) { |
0116651c | 2380 | pos += written; |
f8579f86 | 2381 | iov_iter_advance(from, written); |
0116651c NK |
2382 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
2383 | i_size_write(inode, pos); | |
1da177e4 LT |
2384 | mark_inode_dirty(inode); |
2385 | } | |
5cb6c6c7 | 2386 | iocb->ki_pos = pos; |
1da177e4 | 2387 | } |
a969e903 | 2388 | out: |
1da177e4 LT |
2389 | return written; |
2390 | } | |
2391 | EXPORT_SYMBOL(generic_file_direct_write); | |
2392 | ||
eb2be189 NP |
2393 | /* |
2394 | * Find or create a page at the given pagecache position. Return the locked | |
2395 | * page. This function is specifically for buffered writes. | |
2396 | */ | |
54566b2c NP |
2397 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2398 | pgoff_t index, unsigned flags) | |
eb2be189 | 2399 | { |
eb2be189 | 2400 | struct page *page; |
2457aec6 | 2401 | int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT; |
0faa70cb | 2402 | |
54566b2c | 2403 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
2404 | fgp_flags |= FGP_NOFS; |
2405 | ||
2406 | page = pagecache_get_page(mapping, index, fgp_flags, | |
2407 | mapping_gfp_mask(mapping), | |
2408 | GFP_KERNEL); | |
c585a267 | 2409 | if (page) |
2457aec6 | 2410 | wait_for_stable_page(page); |
eb2be189 | 2411 | |
eb2be189 NP |
2412 | return page; |
2413 | } | |
54566b2c | 2414 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2415 | |
3b93f911 | 2416 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
2417 | struct iov_iter *i, loff_t pos) |
2418 | { | |
2419 | struct address_space *mapping = file->f_mapping; | |
2420 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2421 | long status = 0; | |
2422 | ssize_t written = 0; | |
674b892e NP |
2423 | unsigned int flags = 0; |
2424 | ||
2425 | /* | |
2426 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2427 | */ | |
2428 | if (segment_eq(get_fs(), KERNEL_DS)) | |
2429 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
afddba49 NP |
2430 | |
2431 | do { | |
2432 | struct page *page; | |
afddba49 NP |
2433 | unsigned long offset; /* Offset into pagecache page */ |
2434 | unsigned long bytes; /* Bytes to write to page */ | |
2435 | size_t copied; /* Bytes copied from user */ | |
2436 | void *fsdata; | |
2437 | ||
2438 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2439 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2440 | iov_iter_count(i)); | |
2441 | ||
2442 | again: | |
afddba49 NP |
2443 | /* |
2444 | * Bring in the user page that we will copy from _first_. | |
2445 | * Otherwise there's a nasty deadlock on copying from the | |
2446 | * same page as we're writing to, without it being marked | |
2447 | * up-to-date. | |
2448 | * | |
2449 | * Not only is this an optimisation, but it is also required | |
2450 | * to check that the address is actually valid, when atomic | |
2451 | * usercopies are used, below. | |
2452 | */ | |
2453 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2454 | status = -EFAULT; | |
2455 | break; | |
2456 | } | |
2457 | ||
674b892e | 2458 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 2459 | &page, &fsdata); |
2457aec6 | 2460 | if (unlikely(status < 0)) |
afddba49 NP |
2461 | break; |
2462 | ||
931e80e4 | 2463 | if (mapping_writably_mapped(mapping)) |
2464 | flush_dcache_page(page); | |
2465 | ||
afddba49 | 2466 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
2467 | flush_dcache_page(page); |
2468 | ||
2469 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
2470 | page, fsdata); | |
2471 | if (unlikely(status < 0)) | |
2472 | break; | |
2473 | copied = status; | |
2474 | ||
2475 | cond_resched(); | |
2476 | ||
124d3b70 | 2477 | iov_iter_advance(i, copied); |
afddba49 NP |
2478 | if (unlikely(copied == 0)) { |
2479 | /* | |
2480 | * If we were unable to copy any data at all, we must | |
2481 | * fall back to a single segment length write. | |
2482 | * | |
2483 | * If we didn't fallback here, we could livelock | |
2484 | * because not all segments in the iov can be copied at | |
2485 | * once without a pagefault. | |
2486 | */ | |
2487 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2488 | iov_iter_single_seg_count(i)); | |
2489 | goto again; | |
2490 | } | |
afddba49 NP |
2491 | pos += copied; |
2492 | written += copied; | |
2493 | ||
2494 | balance_dirty_pages_ratelimited(mapping); | |
a50527b1 JK |
2495 | if (fatal_signal_pending(current)) { |
2496 | status = -EINTR; | |
2497 | break; | |
2498 | } | |
afddba49 NP |
2499 | } while (iov_iter_count(i)); |
2500 | ||
2501 | return written ? written : status; | |
2502 | } | |
3b93f911 | 2503 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 2504 | |
e4dd9de3 | 2505 | /** |
8174202b | 2506 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 2507 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 2508 | * @from: iov_iter with data to write |
e4dd9de3 JK |
2509 | * |
2510 | * This function does all the work needed for actually writing data to a | |
2511 | * file. It does all basic checks, removes SUID from the file, updates | |
2512 | * modification times and calls proper subroutines depending on whether we | |
2513 | * do direct IO or a standard buffered write. | |
2514 | * | |
2515 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2516 | * object which does not need locking at all. | |
2517 | * | |
2518 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2519 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2520 | * avoid syncing under i_mutex. | |
2521 | */ | |
8174202b | 2522 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2523 | { |
2524 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2525 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 2526 | struct inode *inode = mapping->host; |
41fc56d5 | 2527 | loff_t pos = iocb->ki_pos; |
3b93f911 | 2528 | ssize_t written = 0; |
1da177e4 | 2529 | ssize_t err; |
3b93f911 | 2530 | ssize_t status; |
8174202b | 2531 | size_t count = iov_iter_count(from); |
1da177e4 | 2532 | |
1da177e4 LT |
2533 | /* We can write back this queue in page reclaim */ |
2534 | current->backing_dev_info = mapping->backing_dev_info; | |
1da177e4 LT |
2535 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); |
2536 | if (err) | |
2537 | goto out; | |
2538 | ||
2539 | if (count == 0) | |
2540 | goto out; | |
2541 | ||
8174202b | 2542 | iov_iter_truncate(from, count); |
0c949334 | 2543 | |
2f1936b8 | 2544 | err = file_remove_suid(file); |
1da177e4 LT |
2545 | if (err) |
2546 | goto out; | |
2547 | ||
c3b2da31 JB |
2548 | err = file_update_time(file); |
2549 | if (err) | |
2550 | goto out; | |
1da177e4 LT |
2551 | |
2552 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
2553 | if (unlikely(file->f_flags & O_DIRECT)) { | |
fb5527e6 | 2554 | loff_t endbyte; |
fb5527e6 | 2555 | |
8174202b | 2556 | written = generic_file_direct_write(iocb, from, pos); |
1da177e4 LT |
2557 | if (written < 0 || written == count) |
2558 | goto out; | |
3b93f911 | 2559 | |
1da177e4 LT |
2560 | /* |
2561 | * direct-io write to a hole: fall through to buffered I/O | |
2562 | * for completing the rest of the request. | |
2563 | */ | |
2564 | pos += written; | |
2565 | count -= written; | |
3b93f911 | 2566 | |
8174202b | 2567 | status = generic_perform_write(file, from, pos); |
fb5527e6 | 2568 | /* |
3b93f911 | 2569 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
2570 | * then we want to return the number of bytes which were |
2571 | * direct-written, or the error code if that was zero. Note | |
2572 | * that this differs from normal direct-io semantics, which | |
2573 | * will return -EFOO even if some bytes were written. | |
2574 | */ | |
3b93f911 AV |
2575 | if (unlikely(status < 0) && !written) { |
2576 | err = status; | |
fb5527e6 JM |
2577 | goto out; |
2578 | } | |
3b93f911 | 2579 | iocb->ki_pos = pos + status; |
fb5527e6 JM |
2580 | /* |
2581 | * We need to ensure that the page cache pages are written to | |
2582 | * disk and invalidated to preserve the expected O_DIRECT | |
2583 | * semantics. | |
2584 | */ | |
3b93f911 | 2585 | endbyte = pos + status - 1; |
c05c4edd | 2586 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); |
fb5527e6 | 2587 | if (err == 0) { |
3b93f911 | 2588 | written += status; |
fb5527e6 JM |
2589 | invalidate_mapping_pages(mapping, |
2590 | pos >> PAGE_CACHE_SHIFT, | |
2591 | endbyte >> PAGE_CACHE_SHIFT); | |
2592 | } else { | |
2593 | /* | |
2594 | * We don't know how much we wrote, so just return | |
2595 | * the number of bytes which were direct-written | |
2596 | */ | |
2597 | } | |
2598 | } else { | |
8174202b | 2599 | written = generic_perform_write(file, from, pos); |
3b93f911 AV |
2600 | if (likely(written >= 0)) |
2601 | iocb->ki_pos = pos + written; | |
fb5527e6 | 2602 | } |
1da177e4 LT |
2603 | out: |
2604 | current->backing_dev_info = NULL; | |
2605 | return written ? written : err; | |
2606 | } | |
8174202b | 2607 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 2608 | |
e4dd9de3 | 2609 | /** |
8174202b | 2610 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 2611 | * @iocb: IO state structure |
8174202b | 2612 | * @from: iov_iter with data to write |
e4dd9de3 | 2613 | * |
8174202b | 2614 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
2615 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
2616 | * and acquires i_mutex as needed. | |
2617 | */ | |
8174202b | 2618 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2619 | { |
2620 | struct file *file = iocb->ki_filp; | |
148f948b | 2621 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2622 | ssize_t ret; |
1da177e4 | 2623 | |
1b1dcc1b | 2624 | mutex_lock(&inode->i_mutex); |
8174202b | 2625 | ret = __generic_file_write_iter(iocb, from); |
1b1dcc1b | 2626 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 2627 | |
02afc27f | 2628 | if (ret > 0) { |
1da177e4 LT |
2629 | ssize_t err; |
2630 | ||
d311d79d AV |
2631 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
2632 | if (err < 0) | |
1da177e4 LT |
2633 | ret = err; |
2634 | } | |
2635 | return ret; | |
2636 | } | |
8174202b | 2637 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 2638 | |
cf9a2ae8 DH |
2639 | /** |
2640 | * try_to_release_page() - release old fs-specific metadata on a page | |
2641 | * | |
2642 | * @page: the page which the kernel is trying to free | |
2643 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2644 | * | |
2645 | * The address_space is to try to release any data against the page | |
2646 | * (presumably at page->private). If the release was successful, return `1'. | |
2647 | * Otherwise return zero. | |
2648 | * | |
266cf658 DH |
2649 | * This may also be called if PG_fscache is set on a page, indicating that the |
2650 | * page is known to the local caching routines. | |
2651 | * | |
cf9a2ae8 | 2652 | * The @gfp_mask argument specifies whether I/O may be performed to release |
3f31fddf | 2653 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
cf9a2ae8 | 2654 | * |
cf9a2ae8 DH |
2655 | */ |
2656 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2657 | { | |
2658 | struct address_space * const mapping = page->mapping; | |
2659 | ||
2660 | BUG_ON(!PageLocked(page)); | |
2661 | if (PageWriteback(page)) | |
2662 | return 0; | |
2663 | ||
2664 | if (mapping && mapping->a_ops->releasepage) | |
2665 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2666 | return try_to_free_buffers(page); | |
2667 | } | |
2668 | ||
2669 | EXPORT_SYMBOL(try_to_release_page); |