]>
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 | */ | |
1da177e4 LT |
12 | #include <linux/module.h> |
13 | #include <linux/slab.h> | |
14 | #include <linux/compiler.h> | |
15 | #include <linux/fs.h> | |
c22ce143 | 16 | #include <linux/uaccess.h> |
1da177e4 | 17 | #include <linux/aio.h> |
c59ede7b | 18 | #include <linux/capability.h> |
1da177e4 LT |
19 | #include <linux/kernel_stat.h> |
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> | |
32 | #include <linux/syscalls.h> | |
44110fe3 | 33 | #include <linux/cpuset.h> |
2f718ffc | 34 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
8a9f3ccd | 35 | #include <linux/memcontrol.h> |
0f8053a5 NP |
36 | #include "internal.h" |
37 | ||
1da177e4 | 38 | /* |
1da177e4 LT |
39 | * FIXME: remove all knowledge of the buffer layer from the core VM |
40 | */ | |
41 | #include <linux/buffer_head.h> /* for generic_osync_inode */ | |
42 | ||
1da177e4 LT |
43 | #include <asm/mman.h> |
44 | ||
5ce7852c AB |
45 | static ssize_t |
46 | generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, | |
47 | loff_t offset, unsigned long nr_segs); | |
48 | ||
1da177e4 LT |
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 | * | |
64 | * ->i_mmap_lock (vmtruncate) | |
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 |
1da177e4 LT |
70 | * ->i_mmap_lock (truncate->unmap_mapping_range) |
71 | * | |
72 | * ->mmap_sem | |
73 | * ->i_mmap_lock | |
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 | * | |
82591e6e NP |
80 | * ->i_mutex (generic_file_buffered_write) |
81 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) | |
1da177e4 | 82 | * |
1b1dcc1b | 83 | * ->i_mutex |
1da177e4 LT |
84 | * ->i_alloc_sem (various) |
85 | * | |
86 | * ->inode_lock | |
87 | * ->sb_lock (fs/fs-writeback.c) | |
88 | * ->mapping->tree_lock (__sync_single_inode) | |
89 | * | |
90 | * ->i_mmap_lock | |
91 | * ->anon_vma.lock (vma_adjust) | |
92 | * | |
93 | * ->anon_vma.lock | |
b8072f09 | 94 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 95 | * |
b8072f09 | 96 | * ->page_table_lock or pte_lock |
5d337b91 | 97 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
98 | * ->private_lock (try_to_unmap_one) |
99 | * ->tree_lock (try_to_unmap_one) | |
100 | * ->zone.lru_lock (follow_page->mark_page_accessed) | |
053837fc | 101 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
1da177e4 LT |
102 | * ->private_lock (page_remove_rmap->set_page_dirty) |
103 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
104 | * ->inode_lock (page_remove_rmap->set_page_dirty) | |
105 | * ->inode_lock (zap_pte_range->set_page_dirty) | |
106 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) | |
107 | * | |
108 | * ->task->proc_lock | |
109 | * ->dcache_lock (proc_pid_lookup) | |
110 | */ | |
111 | ||
112 | /* | |
113 | * Remove a page from the page cache and free it. Caller has to make | |
114 | * sure the page is locked and that nobody else uses it - or that usage | |
115 | * is safe. The caller must hold a write_lock on the mapping's tree_lock. | |
116 | */ | |
117 | void __remove_from_page_cache(struct page *page) | |
118 | { | |
119 | struct address_space *mapping = page->mapping; | |
120 | ||
8a9f3ccd | 121 | mem_cgroup_uncharge_page(page); |
1da177e4 LT |
122 | radix_tree_delete(&mapping->page_tree, page->index); |
123 | page->mapping = NULL; | |
124 | mapping->nrpages--; | |
347ce434 | 125 | __dec_zone_page_state(page, NR_FILE_PAGES); |
45426812 | 126 | BUG_ON(page_mapped(page)); |
3a692790 LT |
127 | |
128 | /* | |
129 | * Some filesystems seem to re-dirty the page even after | |
130 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
131 | * | |
132 | * Fix it up by doing a final dirty accounting check after | |
133 | * having removed the page entirely. | |
134 | */ | |
135 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { | |
136 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
137 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
138 | } | |
1da177e4 LT |
139 | } |
140 | ||
141 | void remove_from_page_cache(struct page *page) | |
142 | { | |
143 | struct address_space *mapping = page->mapping; | |
144 | ||
cd7619d6 | 145 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
146 | |
147 | write_lock_irq(&mapping->tree_lock); | |
148 | __remove_from_page_cache(page); | |
149 | write_unlock_irq(&mapping->tree_lock); | |
150 | } | |
151 | ||
152 | static int sync_page(void *word) | |
153 | { | |
154 | struct address_space *mapping; | |
155 | struct page *page; | |
156 | ||
07808b74 | 157 | page = container_of((unsigned long *)word, struct page, flags); |
1da177e4 LT |
158 | |
159 | /* | |
dd1d5afc WLII |
160 | * page_mapping() is being called without PG_locked held. |
161 | * Some knowledge of the state and use of the page is used to | |
162 | * reduce the requirements down to a memory barrier. | |
163 | * The danger here is of a stale page_mapping() return value | |
164 | * indicating a struct address_space different from the one it's | |
165 | * associated with when it is associated with one. | |
166 | * After smp_mb(), it's either the correct page_mapping() for | |
167 | * the page, or an old page_mapping() and the page's own | |
168 | * page_mapping() has gone NULL. | |
169 | * The ->sync_page() address_space operation must tolerate | |
170 | * page_mapping() going NULL. By an amazing coincidence, | |
171 | * this comes about because none of the users of the page | |
172 | * in the ->sync_page() methods make essential use of the | |
173 | * page_mapping(), merely passing the page down to the backing | |
174 | * device's unplug functions when it's non-NULL, which in turn | |
4c21e2f2 | 175 | * ignore it for all cases but swap, where only page_private(page) is |
dd1d5afc WLII |
176 | * of interest. When page_mapping() does go NULL, the entire |
177 | * call stack gracefully ignores the page and returns. | |
178 | * -- wli | |
1da177e4 LT |
179 | */ |
180 | smp_mb(); | |
181 | mapping = page_mapping(page); | |
182 | if (mapping && mapping->a_ops && mapping->a_ops->sync_page) | |
183 | mapping->a_ops->sync_page(page); | |
184 | io_schedule(); | |
185 | return 0; | |
186 | } | |
187 | ||
2687a356 MW |
188 | static int sync_page_killable(void *word) |
189 | { | |
190 | sync_page(word); | |
191 | return fatal_signal_pending(current) ? -EINTR : 0; | |
192 | } | |
193 | ||
1da177e4 | 194 | /** |
485bb99b | 195 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
196 | * @mapping: address space structure to write |
197 | * @start: offset in bytes where the range starts | |
469eb4d0 | 198 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 199 | * @sync_mode: enable synchronous operation |
1da177e4 | 200 | * |
485bb99b RD |
201 | * Start writeback against all of a mapping's dirty pages that lie |
202 | * within the byte offsets <start, end> inclusive. | |
203 | * | |
1da177e4 | 204 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 205 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
206 | * these two operations is that if a dirty page/buffer is encountered, it must |
207 | * be waited upon, and not just skipped over. | |
208 | */ | |
ebcf28e1 AM |
209 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
210 | loff_t end, int sync_mode) | |
1da177e4 LT |
211 | { |
212 | int ret; | |
213 | struct writeback_control wbc = { | |
214 | .sync_mode = sync_mode, | |
215 | .nr_to_write = mapping->nrpages * 2, | |
111ebb6e OH |
216 | .range_start = start, |
217 | .range_end = end, | |
1da177e4 LT |
218 | }; |
219 | ||
220 | if (!mapping_cap_writeback_dirty(mapping)) | |
221 | return 0; | |
222 | ||
223 | ret = do_writepages(mapping, &wbc); | |
224 | return ret; | |
225 | } | |
226 | ||
227 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
228 | int sync_mode) | |
229 | { | |
111ebb6e | 230 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
231 | } |
232 | ||
233 | int filemap_fdatawrite(struct address_space *mapping) | |
234 | { | |
235 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
236 | } | |
237 | EXPORT_SYMBOL(filemap_fdatawrite); | |
238 | ||
ebcf28e1 AM |
239 | static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
240 | loff_t end) | |
1da177e4 LT |
241 | { |
242 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
243 | } | |
244 | ||
485bb99b RD |
245 | /** |
246 | * filemap_flush - mostly a non-blocking flush | |
247 | * @mapping: target address_space | |
248 | * | |
1da177e4 LT |
249 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
250 | * purposes - I/O may not be started against all dirty pages. | |
251 | */ | |
252 | int filemap_flush(struct address_space *mapping) | |
253 | { | |
254 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
255 | } | |
256 | EXPORT_SYMBOL(filemap_flush); | |
257 | ||
485bb99b RD |
258 | /** |
259 | * wait_on_page_writeback_range - wait for writeback to complete | |
260 | * @mapping: target address_space | |
261 | * @start: beginning page index | |
262 | * @end: ending page index | |
263 | * | |
1da177e4 LT |
264 | * Wait for writeback to complete against pages indexed by start->end |
265 | * inclusive | |
266 | */ | |
ebcf28e1 | 267 | int wait_on_page_writeback_range(struct address_space *mapping, |
1da177e4 LT |
268 | pgoff_t start, pgoff_t end) |
269 | { | |
270 | struct pagevec pvec; | |
271 | int nr_pages; | |
272 | int ret = 0; | |
273 | pgoff_t index; | |
274 | ||
275 | if (end < start) | |
276 | return 0; | |
277 | ||
278 | pagevec_init(&pvec, 0); | |
279 | index = start; | |
280 | while ((index <= end) && | |
281 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
282 | PAGECACHE_TAG_WRITEBACK, | |
283 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
284 | unsigned i; | |
285 | ||
286 | for (i = 0; i < nr_pages; i++) { | |
287 | struct page *page = pvec.pages[i]; | |
288 | ||
289 | /* until radix tree lookup accepts end_index */ | |
290 | if (page->index > end) | |
291 | continue; | |
292 | ||
293 | wait_on_page_writeback(page); | |
294 | if (PageError(page)) | |
295 | ret = -EIO; | |
296 | } | |
297 | pagevec_release(&pvec); | |
298 | cond_resched(); | |
299 | } | |
300 | ||
301 | /* Check for outstanding write errors */ | |
302 | if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
303 | ret = -ENOSPC; | |
304 | if (test_and_clear_bit(AS_EIO, &mapping->flags)) | |
305 | ret = -EIO; | |
306 | ||
307 | return ret; | |
308 | } | |
309 | ||
485bb99b RD |
310 | /** |
311 | * sync_page_range - write and wait on all pages in the passed range | |
312 | * @inode: target inode | |
313 | * @mapping: target address_space | |
314 | * @pos: beginning offset in pages to write | |
315 | * @count: number of bytes to write | |
316 | * | |
1da177e4 LT |
317 | * Write and wait upon all the pages in the passed range. This is a "data |
318 | * integrity" operation. It waits upon in-flight writeout before starting and | |
319 | * waiting upon new writeout. If there was an IO error, return it. | |
320 | * | |
1b1dcc1b | 321 | * We need to re-take i_mutex during the generic_osync_inode list walk because |
1da177e4 LT |
322 | * it is otherwise livelockable. |
323 | */ | |
324 | int sync_page_range(struct inode *inode, struct address_space *mapping, | |
268fc16e | 325 | loff_t pos, loff_t count) |
1da177e4 LT |
326 | { |
327 | pgoff_t start = pos >> PAGE_CACHE_SHIFT; | |
328 | pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT; | |
329 | int ret; | |
330 | ||
331 | if (!mapping_cap_writeback_dirty(mapping) || !count) | |
332 | return 0; | |
333 | ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1); | |
334 | if (ret == 0) { | |
1b1dcc1b | 335 | mutex_lock(&inode->i_mutex); |
1da177e4 | 336 | ret = generic_osync_inode(inode, mapping, OSYNC_METADATA); |
1b1dcc1b | 337 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
338 | } |
339 | if (ret == 0) | |
340 | ret = wait_on_page_writeback_range(mapping, start, end); | |
341 | return ret; | |
342 | } | |
343 | EXPORT_SYMBOL(sync_page_range); | |
344 | ||
485bb99b RD |
345 | /** |
346 | * sync_page_range_nolock | |
347 | * @inode: target inode | |
348 | * @mapping: target address_space | |
349 | * @pos: beginning offset in pages to write | |
350 | * @count: number of bytes to write | |
351 | * | |
72fd4a35 | 352 | * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea |
1da177e4 LT |
353 | * as it forces O_SYNC writers to different parts of the same file |
354 | * to be serialised right until io completion. | |
355 | */ | |
268fc16e OH |
356 | int sync_page_range_nolock(struct inode *inode, struct address_space *mapping, |
357 | loff_t pos, loff_t count) | |
1da177e4 LT |
358 | { |
359 | pgoff_t start = pos >> PAGE_CACHE_SHIFT; | |
360 | pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT; | |
361 | int ret; | |
362 | ||
363 | if (!mapping_cap_writeback_dirty(mapping) || !count) | |
364 | return 0; | |
365 | ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1); | |
366 | if (ret == 0) | |
367 | ret = generic_osync_inode(inode, mapping, OSYNC_METADATA); | |
368 | if (ret == 0) | |
369 | ret = wait_on_page_writeback_range(mapping, start, end); | |
370 | return ret; | |
371 | } | |
268fc16e | 372 | EXPORT_SYMBOL(sync_page_range_nolock); |
1da177e4 LT |
373 | |
374 | /** | |
485bb99b | 375 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 376 | * @mapping: address space structure to wait for |
485bb99b RD |
377 | * |
378 | * Walk the list of under-writeback pages of the given address space | |
379 | * and wait for all of them. | |
1da177e4 LT |
380 | */ |
381 | int filemap_fdatawait(struct address_space *mapping) | |
382 | { | |
383 | loff_t i_size = i_size_read(mapping->host); | |
384 | ||
385 | if (i_size == 0) | |
386 | return 0; | |
387 | ||
388 | return wait_on_page_writeback_range(mapping, 0, | |
389 | (i_size - 1) >> PAGE_CACHE_SHIFT); | |
390 | } | |
391 | EXPORT_SYMBOL(filemap_fdatawait); | |
392 | ||
393 | int filemap_write_and_wait(struct address_space *mapping) | |
394 | { | |
28fd1298 | 395 | int err = 0; |
1da177e4 LT |
396 | |
397 | if (mapping->nrpages) { | |
28fd1298 OH |
398 | err = filemap_fdatawrite(mapping); |
399 | /* | |
400 | * Even if the above returned error, the pages may be | |
401 | * written partially (e.g. -ENOSPC), so we wait for it. | |
402 | * But the -EIO is special case, it may indicate the worst | |
403 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
404 | */ | |
405 | if (err != -EIO) { | |
406 | int err2 = filemap_fdatawait(mapping); | |
407 | if (!err) | |
408 | err = err2; | |
409 | } | |
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) { | |
436 | int err2 = wait_on_page_writeback_range(mapping, | |
437 | lstart >> PAGE_CACHE_SHIFT, | |
438 | lend >> PAGE_CACHE_SHIFT); | |
439 | if (!err) | |
440 | err = err2; | |
441 | } | |
1da177e4 | 442 | } |
28fd1298 | 443 | return err; |
1da177e4 LT |
444 | } |
445 | ||
485bb99b RD |
446 | /** |
447 | * add_to_page_cache - add newly allocated pagecache pages | |
448 | * @page: page to add | |
449 | * @mapping: the page's address_space | |
450 | * @offset: page index | |
451 | * @gfp_mask: page allocation mode | |
452 | * | |
453 | * This function is used to add newly allocated pagecache pages; | |
1da177e4 LT |
454 | * the page is new, so we can just run SetPageLocked() against it. |
455 | * The other page state flags were set by rmqueue(). | |
456 | * | |
457 | * This function does not add the page to the LRU. The caller must do that. | |
458 | */ | |
459 | int add_to_page_cache(struct page *page, struct address_space *mapping, | |
6daa0e28 | 460 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 461 | { |
4c6bc8dd BP |
462 | int error = mem_cgroup_cache_charge(page, current->mm, |
463 | gfp_mask & ~__GFP_HIGHMEM); | |
35c754d7 BS |
464 | if (error) |
465 | goto out; | |
1da177e4 | 466 | |
35c754d7 | 467 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
1da177e4 LT |
468 | if (error == 0) { |
469 | write_lock_irq(&mapping->tree_lock); | |
470 | error = radix_tree_insert(&mapping->page_tree, offset, page); | |
471 | if (!error) { | |
472 | page_cache_get(page); | |
473 | SetPageLocked(page); | |
474 | page->mapping = mapping; | |
475 | page->index = offset; | |
476 | mapping->nrpages++; | |
347ce434 | 477 | __inc_zone_page_state(page, NR_FILE_PAGES); |
8a9f3ccd BS |
478 | } else |
479 | mem_cgroup_uncharge_page(page); | |
480 | ||
1da177e4 LT |
481 | write_unlock_irq(&mapping->tree_lock); |
482 | radix_tree_preload_end(); | |
35c754d7 BS |
483 | } else |
484 | mem_cgroup_uncharge_page(page); | |
8a9f3ccd | 485 | out: |
1da177e4 LT |
486 | return error; |
487 | } | |
1da177e4 LT |
488 | EXPORT_SYMBOL(add_to_page_cache); |
489 | ||
490 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 491 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 LT |
492 | { |
493 | int ret = add_to_page_cache(page, mapping, offset, gfp_mask); | |
494 | if (ret == 0) | |
495 | lru_cache_add(page); | |
496 | return ret; | |
497 | } | |
498 | ||
44110fe3 | 499 | #ifdef CONFIG_NUMA |
2ae88149 | 500 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 PJ |
501 | { |
502 | if (cpuset_do_page_mem_spread()) { | |
503 | int n = cpuset_mem_spread_node(); | |
2ae88149 | 504 | return alloc_pages_node(n, gfp, 0); |
44110fe3 | 505 | } |
2ae88149 | 506 | return alloc_pages(gfp, 0); |
44110fe3 | 507 | } |
2ae88149 | 508 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
509 | #endif |
510 | ||
db37648c NP |
511 | static int __sleep_on_page_lock(void *word) |
512 | { | |
513 | io_schedule(); | |
514 | return 0; | |
515 | } | |
516 | ||
1da177e4 LT |
517 | /* |
518 | * In order to wait for pages to become available there must be | |
519 | * waitqueues associated with pages. By using a hash table of | |
520 | * waitqueues where the bucket discipline is to maintain all | |
521 | * waiters on the same queue and wake all when any of the pages | |
522 | * become available, and for the woken contexts to check to be | |
523 | * sure the appropriate page became available, this saves space | |
524 | * at a cost of "thundering herd" phenomena during rare hash | |
525 | * collisions. | |
526 | */ | |
527 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
528 | { | |
529 | const struct zone *zone = page_zone(page); | |
530 | ||
531 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
532 | } | |
533 | ||
534 | static inline void wake_up_page(struct page *page, int bit) | |
535 | { | |
536 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); | |
537 | } | |
538 | ||
920c7a5d | 539 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
540 | { |
541 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
542 | ||
543 | if (test_bit(bit_nr, &page->flags)) | |
544 | __wait_on_bit(page_waitqueue(page), &wait, sync_page, | |
545 | TASK_UNINTERRUPTIBLE); | |
546 | } | |
547 | EXPORT_SYMBOL(wait_on_page_bit); | |
548 | ||
549 | /** | |
485bb99b | 550 | * unlock_page - unlock a locked page |
1da177e4 LT |
551 | * @page: the page |
552 | * | |
553 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
554 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
555 | * mechananism between PageLocked pages and PageWriteback pages is shared. | |
556 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. | |
557 | * | |
558 | * The first mb is necessary to safely close the critical section opened by the | |
559 | * TestSetPageLocked(), the second mb is necessary to enforce ordering between | |
560 | * the clear_bit and the read of the waitqueue (to avoid SMP races with a | |
561 | * parallel wait_on_page_locked()). | |
562 | */ | |
920c7a5d | 563 | void unlock_page(struct page *page) |
1da177e4 LT |
564 | { |
565 | smp_mb__before_clear_bit(); | |
566 | if (!TestClearPageLocked(page)) | |
567 | BUG(); | |
568 | smp_mb__after_clear_bit(); | |
569 | wake_up_page(page, PG_locked); | |
570 | } | |
571 | EXPORT_SYMBOL(unlock_page); | |
572 | ||
485bb99b RD |
573 | /** |
574 | * end_page_writeback - end writeback against a page | |
575 | * @page: the page | |
1da177e4 LT |
576 | */ |
577 | void end_page_writeback(struct page *page) | |
578 | { | |
579 | if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) { | |
580 | if (!test_clear_page_writeback(page)) | |
581 | BUG(); | |
582 | } | |
583 | smp_mb__after_clear_bit(); | |
584 | wake_up_page(page, PG_writeback); | |
585 | } | |
586 | EXPORT_SYMBOL(end_page_writeback); | |
587 | ||
485bb99b RD |
588 | /** |
589 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
590 | * @page: the page to lock | |
1da177e4 | 591 | * |
485bb99b | 592 | * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some |
1da177e4 LT |
593 | * random driver's requestfn sets TASK_RUNNING, we could busywait. However |
594 | * chances are that on the second loop, the block layer's plug list is empty, | |
595 | * so sync_page() will then return in state TASK_UNINTERRUPTIBLE. | |
596 | */ | |
920c7a5d | 597 | void __lock_page(struct page *page) |
1da177e4 LT |
598 | { |
599 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
600 | ||
601 | __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page, | |
602 | TASK_UNINTERRUPTIBLE); | |
603 | } | |
604 | EXPORT_SYMBOL(__lock_page); | |
605 | ||
b5606c2d | 606 | int __lock_page_killable(struct page *page) |
2687a356 MW |
607 | { |
608 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
609 | ||
610 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
611 | sync_page_killable, TASK_KILLABLE); | |
612 | } | |
613 | ||
db37648c NP |
614 | /* |
615 | * Variant of lock_page that does not require the caller to hold a reference | |
616 | * on the page's mapping. | |
617 | */ | |
920c7a5d | 618 | void __lock_page_nosync(struct page *page) |
db37648c NP |
619 | { |
620 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
621 | __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock, | |
622 | TASK_UNINTERRUPTIBLE); | |
623 | } | |
624 | ||
485bb99b RD |
625 | /** |
626 | * find_get_page - find and get a page reference | |
627 | * @mapping: the address_space to search | |
628 | * @offset: the page index | |
629 | * | |
da6052f7 NP |
630 | * Is there a pagecache struct page at the given (mapping, offset) tuple? |
631 | * If yes, increment its refcount and return it; if no, return NULL. | |
1da177e4 | 632 | */ |
57f6b96c | 633 | struct page * find_get_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 LT |
634 | { |
635 | struct page *page; | |
636 | ||
637 | read_lock_irq(&mapping->tree_lock); | |
638 | page = radix_tree_lookup(&mapping->page_tree, offset); | |
639 | if (page) | |
640 | page_cache_get(page); | |
641 | read_unlock_irq(&mapping->tree_lock); | |
642 | return page; | |
643 | } | |
1da177e4 LT |
644 | EXPORT_SYMBOL(find_get_page); |
645 | ||
1da177e4 LT |
646 | /** |
647 | * find_lock_page - locate, pin and lock a pagecache page | |
67be2dd1 MW |
648 | * @mapping: the address_space to search |
649 | * @offset: the page index | |
1da177e4 LT |
650 | * |
651 | * Locates the desired pagecache page, locks it, increments its reference | |
652 | * count and returns its address. | |
653 | * | |
654 | * Returns zero if the page was not present. find_lock_page() may sleep. | |
655 | */ | |
656 | struct page *find_lock_page(struct address_space *mapping, | |
57f6b96c | 657 | pgoff_t offset) |
1da177e4 LT |
658 | { |
659 | struct page *page; | |
660 | ||
1da177e4 | 661 | repeat: |
45726cb4 | 662 | read_lock_irq(&mapping->tree_lock); |
1da177e4 LT |
663 | page = radix_tree_lookup(&mapping->page_tree, offset); |
664 | if (page) { | |
665 | page_cache_get(page); | |
666 | if (TestSetPageLocked(page)) { | |
667 | read_unlock_irq(&mapping->tree_lock); | |
bbfbb7ce | 668 | __lock_page(page); |
1da177e4 LT |
669 | |
670 | /* Has the page been truncated while we slept? */ | |
45726cb4 | 671 | if (unlikely(page->mapping != mapping)) { |
1da177e4 LT |
672 | unlock_page(page); |
673 | page_cache_release(page); | |
674 | goto repeat; | |
675 | } | |
45726cb4 NP |
676 | VM_BUG_ON(page->index != offset); |
677 | goto out; | |
1da177e4 LT |
678 | } |
679 | } | |
680 | read_unlock_irq(&mapping->tree_lock); | |
45726cb4 | 681 | out: |
1da177e4 LT |
682 | return page; |
683 | } | |
1da177e4 LT |
684 | EXPORT_SYMBOL(find_lock_page); |
685 | ||
686 | /** | |
687 | * find_or_create_page - locate or add a pagecache page | |
67be2dd1 MW |
688 | * @mapping: the page's address_space |
689 | * @index: the page's index into the mapping | |
690 | * @gfp_mask: page allocation mode | |
1da177e4 LT |
691 | * |
692 | * Locates a page in the pagecache. If the page is not present, a new page | |
693 | * is allocated using @gfp_mask and is added to the pagecache and to the VM's | |
694 | * LRU list. The returned page is locked and has its reference count | |
695 | * incremented. | |
696 | * | |
697 | * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic | |
698 | * allocation! | |
699 | * | |
700 | * find_or_create_page() returns the desired page's address, or zero on | |
701 | * memory exhaustion. | |
702 | */ | |
703 | struct page *find_or_create_page(struct address_space *mapping, | |
57f6b96c | 704 | pgoff_t index, gfp_t gfp_mask) |
1da177e4 | 705 | { |
eb2be189 | 706 | struct page *page; |
1da177e4 LT |
707 | int err; |
708 | repeat: | |
709 | page = find_lock_page(mapping, index); | |
710 | if (!page) { | |
eb2be189 NP |
711 | page = __page_cache_alloc(gfp_mask); |
712 | if (!page) | |
713 | return NULL; | |
714 | err = add_to_page_cache_lru(page, mapping, index, gfp_mask); | |
715 | if (unlikely(err)) { | |
716 | page_cache_release(page); | |
717 | page = NULL; | |
718 | if (err == -EEXIST) | |
719 | goto repeat; | |
1da177e4 | 720 | } |
1da177e4 | 721 | } |
1da177e4 LT |
722 | return page; |
723 | } | |
1da177e4 LT |
724 | EXPORT_SYMBOL(find_or_create_page); |
725 | ||
726 | /** | |
727 | * find_get_pages - gang pagecache lookup | |
728 | * @mapping: The address_space to search | |
729 | * @start: The starting page index | |
730 | * @nr_pages: The maximum number of pages | |
731 | * @pages: Where the resulting pages are placed | |
732 | * | |
733 | * find_get_pages() will search for and return a group of up to | |
734 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
735 | * find_get_pages() takes a reference against the returned pages. | |
736 | * | |
737 | * The search returns a group of mapping-contiguous pages with ascending | |
738 | * indexes. There may be holes in the indices due to not-present pages. | |
739 | * | |
740 | * find_get_pages() returns the number of pages which were found. | |
741 | */ | |
742 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
743 | unsigned int nr_pages, struct page **pages) | |
744 | { | |
745 | unsigned int i; | |
746 | unsigned int ret; | |
747 | ||
748 | read_lock_irq(&mapping->tree_lock); | |
749 | ret = radix_tree_gang_lookup(&mapping->page_tree, | |
750 | (void **)pages, start, nr_pages); | |
751 | for (i = 0; i < ret; i++) | |
752 | page_cache_get(pages[i]); | |
753 | read_unlock_irq(&mapping->tree_lock); | |
754 | return ret; | |
755 | } | |
756 | ||
ebf43500 JA |
757 | /** |
758 | * find_get_pages_contig - gang contiguous pagecache lookup | |
759 | * @mapping: The address_space to search | |
760 | * @index: The starting page index | |
761 | * @nr_pages: The maximum number of pages | |
762 | * @pages: Where the resulting pages are placed | |
763 | * | |
764 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
765 | * that the returned number of pages are guaranteed to be contiguous. | |
766 | * | |
767 | * find_get_pages_contig() returns the number of pages which were found. | |
768 | */ | |
769 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
770 | unsigned int nr_pages, struct page **pages) | |
771 | { | |
772 | unsigned int i; | |
773 | unsigned int ret; | |
774 | ||
775 | read_lock_irq(&mapping->tree_lock); | |
776 | ret = radix_tree_gang_lookup(&mapping->page_tree, | |
777 | (void **)pages, index, nr_pages); | |
778 | for (i = 0; i < ret; i++) { | |
779 | if (pages[i]->mapping == NULL || pages[i]->index != index) | |
780 | break; | |
781 | ||
782 | page_cache_get(pages[i]); | |
783 | index++; | |
784 | } | |
785 | read_unlock_irq(&mapping->tree_lock); | |
786 | return i; | |
787 | } | |
ef71c15c | 788 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 789 | |
485bb99b RD |
790 | /** |
791 | * find_get_pages_tag - find and return pages that match @tag | |
792 | * @mapping: the address_space to search | |
793 | * @index: the starting page index | |
794 | * @tag: the tag index | |
795 | * @nr_pages: the maximum number of pages | |
796 | * @pages: where the resulting pages are placed | |
797 | * | |
1da177e4 | 798 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 799 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
800 | */ |
801 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
802 | int tag, unsigned int nr_pages, struct page **pages) | |
803 | { | |
804 | unsigned int i; | |
805 | unsigned int ret; | |
806 | ||
807 | read_lock_irq(&mapping->tree_lock); | |
808 | ret = radix_tree_gang_lookup_tag(&mapping->page_tree, | |
809 | (void **)pages, *index, nr_pages, tag); | |
810 | for (i = 0; i < ret; i++) | |
811 | page_cache_get(pages[i]); | |
812 | if (ret) | |
813 | *index = pages[ret - 1]->index + 1; | |
814 | read_unlock_irq(&mapping->tree_lock); | |
815 | return ret; | |
816 | } | |
ef71c15c | 817 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 818 | |
485bb99b RD |
819 | /** |
820 | * grab_cache_page_nowait - returns locked page at given index in given cache | |
821 | * @mapping: target address_space | |
822 | * @index: the page index | |
823 | * | |
72fd4a35 | 824 | * Same as grab_cache_page(), but do not wait if the page is unavailable. |
1da177e4 LT |
825 | * This is intended for speculative data generators, where the data can |
826 | * be regenerated if the page couldn't be grabbed. This routine should | |
827 | * be safe to call while holding the lock for another page. | |
828 | * | |
829 | * Clear __GFP_FS when allocating the page to avoid recursion into the fs | |
830 | * and deadlock against the caller's locked page. | |
831 | */ | |
832 | struct page * | |
57f6b96c | 833 | grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) |
1da177e4 LT |
834 | { |
835 | struct page *page = find_get_page(mapping, index); | |
1da177e4 LT |
836 | |
837 | if (page) { | |
838 | if (!TestSetPageLocked(page)) | |
839 | return page; | |
840 | page_cache_release(page); | |
841 | return NULL; | |
842 | } | |
2ae88149 NP |
843 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); |
844 | if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) { | |
1da177e4 LT |
845 | page_cache_release(page); |
846 | page = NULL; | |
847 | } | |
848 | return page; | |
849 | } | |
1da177e4 LT |
850 | EXPORT_SYMBOL(grab_cache_page_nowait); |
851 | ||
76d42bd9 WF |
852 | /* |
853 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
854 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
855 | * | |
856 | * ---R__________________________________________B__________ | |
857 | * ^ reading here ^ bad block(assume 4k) | |
858 | * | |
859 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
860 | * => failing the whole request => read(R) => read(R+1) => | |
861 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
862 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
863 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
864 | * | |
865 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
866 | */ | |
867 | static void shrink_readahead_size_eio(struct file *filp, | |
868 | struct file_ra_state *ra) | |
869 | { | |
870 | if (!ra->ra_pages) | |
871 | return; | |
872 | ||
873 | ra->ra_pages /= 4; | |
76d42bd9 WF |
874 | } |
875 | ||
485bb99b | 876 | /** |
36e78914 | 877 | * do_generic_file_read - generic file read routine |
485bb99b RD |
878 | * @filp: the file to read |
879 | * @ppos: current file position | |
880 | * @desc: read_descriptor | |
881 | * @actor: read method | |
882 | * | |
1da177e4 | 883 | * This is a generic file read routine, and uses the |
485bb99b | 884 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
885 | * |
886 | * This is really ugly. But the goto's actually try to clarify some | |
887 | * of the logic when it comes to error handling etc. | |
1da177e4 | 888 | */ |
36e78914 CH |
889 | static void do_generic_file_read(struct file *filp, loff_t *ppos, |
890 | read_descriptor_t *desc, read_actor_t actor) | |
1da177e4 | 891 | { |
36e78914 | 892 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 893 | struct inode *inode = mapping->host; |
36e78914 | 894 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
895 | pgoff_t index; |
896 | pgoff_t last_index; | |
897 | pgoff_t prev_index; | |
898 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 899 | unsigned int prev_offset; |
1da177e4 | 900 | int error; |
1da177e4 | 901 | |
1da177e4 | 902 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
903 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
904 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
1da177e4 LT |
905 | last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
906 | offset = *ppos & ~PAGE_CACHE_MASK; | |
907 | ||
1da177e4 LT |
908 | for (;;) { |
909 | struct page *page; | |
57f6b96c | 910 | pgoff_t end_index; |
a32ea1e1 | 911 | loff_t isize; |
1da177e4 LT |
912 | unsigned long nr, ret; |
913 | ||
1da177e4 | 914 | cond_resched(); |
1da177e4 LT |
915 | find_page: |
916 | page = find_get_page(mapping, index); | |
3ea89ee8 | 917 | if (!page) { |
cf914a7d | 918 | page_cache_sync_readahead(mapping, |
7ff81078 | 919 | ra, filp, |
3ea89ee8 FW |
920 | index, last_index - index); |
921 | page = find_get_page(mapping, index); | |
922 | if (unlikely(page == NULL)) | |
923 | goto no_cached_page; | |
924 | } | |
925 | if (PageReadahead(page)) { | |
cf914a7d | 926 | page_cache_async_readahead(mapping, |
7ff81078 | 927 | ra, filp, page, |
3ea89ee8 | 928 | index, last_index - index); |
1da177e4 LT |
929 | } |
930 | if (!PageUptodate(page)) | |
931 | goto page_not_up_to_date; | |
932 | page_ok: | |
a32ea1e1 N |
933 | /* |
934 | * i_size must be checked after we know the page is Uptodate. | |
935 | * | |
936 | * Checking i_size after the check allows us to calculate | |
937 | * the correct value for "nr", which means the zero-filled | |
938 | * part of the page is not copied back to userspace (unless | |
939 | * another truncate extends the file - this is desired though). | |
940 | */ | |
941 | ||
942 | isize = i_size_read(inode); | |
943 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
944 | if (unlikely(!isize || index > end_index)) { | |
945 | page_cache_release(page); | |
946 | goto out; | |
947 | } | |
948 | ||
949 | /* nr is the maximum number of bytes to copy from this page */ | |
950 | nr = PAGE_CACHE_SIZE; | |
951 | if (index == end_index) { | |
952 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
953 | if (nr <= offset) { | |
954 | page_cache_release(page); | |
955 | goto out; | |
956 | } | |
957 | } | |
958 | nr = nr - offset; | |
1da177e4 LT |
959 | |
960 | /* If users can be writing to this page using arbitrary | |
961 | * virtual addresses, take care about potential aliasing | |
962 | * before reading the page on the kernel side. | |
963 | */ | |
964 | if (mapping_writably_mapped(mapping)) | |
965 | flush_dcache_page(page); | |
966 | ||
967 | /* | |
ec0f1637 JK |
968 | * When a sequential read accesses a page several times, |
969 | * only mark it as accessed the first time. | |
1da177e4 | 970 | */ |
ec0f1637 | 971 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
972 | mark_page_accessed(page); |
973 | prev_index = index; | |
974 | ||
975 | /* | |
976 | * Ok, we have the page, and it's up-to-date, so | |
977 | * now we can copy it to user space... | |
978 | * | |
979 | * The actor routine returns how many bytes were actually used.. | |
980 | * NOTE! This may not be the same as how much of a user buffer | |
981 | * we filled up (we may be padding etc), so we can only update | |
982 | * "pos" here (the actor routine has to update the user buffer | |
983 | * pointers and the remaining count). | |
984 | */ | |
985 | ret = actor(desc, page, offset, nr); | |
986 | offset += ret; | |
987 | index += offset >> PAGE_CACHE_SHIFT; | |
988 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 989 | prev_offset = offset; |
1da177e4 LT |
990 | |
991 | page_cache_release(page); | |
992 | if (ret == nr && desc->count) | |
993 | continue; | |
994 | goto out; | |
995 | ||
996 | page_not_up_to_date: | |
997 | /* Get exclusive access to the page ... */ | |
0b94e97a MW |
998 | if (lock_page_killable(page)) |
999 | goto readpage_eio; | |
1da177e4 | 1000 | |
da6052f7 | 1001 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1002 | if (!page->mapping) { |
1003 | unlock_page(page); | |
1004 | page_cache_release(page); | |
1005 | continue; | |
1006 | } | |
1007 | ||
1008 | /* Did somebody else fill it already? */ | |
1009 | if (PageUptodate(page)) { | |
1010 | unlock_page(page); | |
1011 | goto page_ok; | |
1012 | } | |
1013 | ||
1014 | readpage: | |
1015 | /* Start the actual read. The read will unlock the page. */ | |
1016 | error = mapping->a_ops->readpage(filp, page); | |
1017 | ||
994fc28c ZB |
1018 | if (unlikely(error)) { |
1019 | if (error == AOP_TRUNCATED_PAGE) { | |
1020 | page_cache_release(page); | |
1021 | goto find_page; | |
1022 | } | |
1da177e4 | 1023 | goto readpage_error; |
994fc28c | 1024 | } |
1da177e4 LT |
1025 | |
1026 | if (!PageUptodate(page)) { | |
0b94e97a MW |
1027 | if (lock_page_killable(page)) |
1028 | goto readpage_eio; | |
1da177e4 LT |
1029 | if (!PageUptodate(page)) { |
1030 | if (page->mapping == NULL) { | |
1031 | /* | |
1032 | * invalidate_inode_pages got it | |
1033 | */ | |
1034 | unlock_page(page); | |
1035 | page_cache_release(page); | |
1036 | goto find_page; | |
1037 | } | |
1038 | unlock_page(page); | |
7ff81078 | 1039 | shrink_readahead_size_eio(filp, ra); |
0b94e97a | 1040 | goto readpage_eio; |
1da177e4 LT |
1041 | } |
1042 | unlock_page(page); | |
1043 | } | |
1044 | ||
1da177e4 LT |
1045 | goto page_ok; |
1046 | ||
0b94e97a MW |
1047 | readpage_eio: |
1048 | error = -EIO; | |
1da177e4 LT |
1049 | readpage_error: |
1050 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1051 | desc->error = error; | |
1052 | page_cache_release(page); | |
1053 | goto out; | |
1054 | ||
1055 | no_cached_page: | |
1056 | /* | |
1057 | * Ok, it wasn't cached, so we need to create a new | |
1058 | * page.. | |
1059 | */ | |
eb2be189 NP |
1060 | page = page_cache_alloc_cold(mapping); |
1061 | if (!page) { | |
1062 | desc->error = -ENOMEM; | |
1063 | goto out; | |
1da177e4 | 1064 | } |
eb2be189 | 1065 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1066 | index, GFP_KERNEL); |
1067 | if (error) { | |
eb2be189 | 1068 | page_cache_release(page); |
1da177e4 LT |
1069 | if (error == -EEXIST) |
1070 | goto find_page; | |
1071 | desc->error = error; | |
1072 | goto out; | |
1073 | } | |
1da177e4 LT |
1074 | goto readpage; |
1075 | } | |
1076 | ||
1077 | out: | |
7ff81078 FW |
1078 | ra->prev_pos = prev_index; |
1079 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1080 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1081 | |
f4e6b498 | 1082 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
1da177e4 LT |
1083 | if (filp) |
1084 | file_accessed(filp); | |
1085 | } | |
1da177e4 LT |
1086 | |
1087 | int file_read_actor(read_descriptor_t *desc, struct page *page, | |
1088 | unsigned long offset, unsigned long size) | |
1089 | { | |
1090 | char *kaddr; | |
1091 | unsigned long left, count = desc->count; | |
1092 | ||
1093 | if (size > count) | |
1094 | size = count; | |
1095 | ||
1096 | /* | |
1097 | * Faults on the destination of a read are common, so do it before | |
1098 | * taking the kmap. | |
1099 | */ | |
1100 | if (!fault_in_pages_writeable(desc->arg.buf, size)) { | |
1101 | kaddr = kmap_atomic(page, KM_USER0); | |
1102 | left = __copy_to_user_inatomic(desc->arg.buf, | |
1103 | kaddr + offset, size); | |
1104 | kunmap_atomic(kaddr, KM_USER0); | |
1105 | if (left == 0) | |
1106 | goto success; | |
1107 | } | |
1108 | ||
1109 | /* Do it the slow way */ | |
1110 | kaddr = kmap(page); | |
1111 | left = __copy_to_user(desc->arg.buf, kaddr + offset, size); | |
1112 | kunmap(page); | |
1113 | ||
1114 | if (left) { | |
1115 | size -= left; | |
1116 | desc->error = -EFAULT; | |
1117 | } | |
1118 | success: | |
1119 | desc->count = count - size; | |
1120 | desc->written += size; | |
1121 | desc->arg.buf += size; | |
1122 | return size; | |
1123 | } | |
1124 | ||
0ceb3314 DM |
1125 | /* |
1126 | * Performs necessary checks before doing a write | |
1127 | * @iov: io vector request | |
1128 | * @nr_segs: number of segments in the iovec | |
1129 | * @count: number of bytes to write | |
1130 | * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE | |
1131 | * | |
1132 | * Adjust number of segments and amount of bytes to write (nr_segs should be | |
1133 | * properly initialized first). Returns appropriate error code that caller | |
1134 | * should return or zero in case that write should be allowed. | |
1135 | */ | |
1136 | int generic_segment_checks(const struct iovec *iov, | |
1137 | unsigned long *nr_segs, size_t *count, int access_flags) | |
1138 | { | |
1139 | unsigned long seg; | |
1140 | size_t cnt = 0; | |
1141 | for (seg = 0; seg < *nr_segs; seg++) { | |
1142 | const struct iovec *iv = &iov[seg]; | |
1143 | ||
1144 | /* | |
1145 | * If any segment has a negative length, or the cumulative | |
1146 | * length ever wraps negative then return -EINVAL. | |
1147 | */ | |
1148 | cnt += iv->iov_len; | |
1149 | if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) | |
1150 | return -EINVAL; | |
1151 | if (access_ok(access_flags, iv->iov_base, iv->iov_len)) | |
1152 | continue; | |
1153 | if (seg == 0) | |
1154 | return -EFAULT; | |
1155 | *nr_segs = seg; | |
1156 | cnt -= iv->iov_len; /* This segment is no good */ | |
1157 | break; | |
1158 | } | |
1159 | *count = cnt; | |
1160 | return 0; | |
1161 | } | |
1162 | EXPORT_SYMBOL(generic_segment_checks); | |
1163 | ||
485bb99b | 1164 | /** |
b2abacf3 | 1165 | * generic_file_aio_read - generic filesystem read routine |
485bb99b RD |
1166 | * @iocb: kernel I/O control block |
1167 | * @iov: io vector request | |
1168 | * @nr_segs: number of segments in the iovec | |
b2abacf3 | 1169 | * @pos: current file position |
485bb99b | 1170 | * |
1da177e4 LT |
1171 | * This is the "read()" routine for all filesystems |
1172 | * that can use the page cache directly. | |
1173 | */ | |
1174 | ssize_t | |
543ade1f BP |
1175 | generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
1176 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
1177 | { |
1178 | struct file *filp = iocb->ki_filp; | |
1179 | ssize_t retval; | |
1180 | unsigned long seg; | |
1181 | size_t count; | |
543ade1f | 1182 | loff_t *ppos = &iocb->ki_pos; |
1da177e4 LT |
1183 | |
1184 | count = 0; | |
0ceb3314 DM |
1185 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); |
1186 | if (retval) | |
1187 | return retval; | |
1da177e4 LT |
1188 | |
1189 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
1190 | if (filp->f_flags & O_DIRECT) { | |
543ade1f | 1191 | loff_t size; |
1da177e4 LT |
1192 | struct address_space *mapping; |
1193 | struct inode *inode; | |
1194 | ||
1195 | mapping = filp->f_mapping; | |
1196 | inode = mapping->host; | |
1197 | retval = 0; | |
1198 | if (!count) | |
1199 | goto out; /* skip atime */ | |
1200 | size = i_size_read(inode); | |
1201 | if (pos < size) { | |
1202 | retval = generic_file_direct_IO(READ, iocb, | |
1203 | iov, pos, nr_segs); | |
1da177e4 LT |
1204 | if (retval > 0) |
1205 | *ppos = pos + retval; | |
1206 | } | |
0e0bcae3 | 1207 | if (likely(retval != 0)) { |
3f1a9aae | 1208 | file_accessed(filp); |
a9e5f4d0 | 1209 | goto out; |
0e0bcae3 | 1210 | } |
1da177e4 LT |
1211 | } |
1212 | ||
1213 | retval = 0; | |
1214 | if (count) { | |
1215 | for (seg = 0; seg < nr_segs; seg++) { | |
1216 | read_descriptor_t desc; | |
1217 | ||
1218 | desc.written = 0; | |
1219 | desc.arg.buf = iov[seg].iov_base; | |
1220 | desc.count = iov[seg].iov_len; | |
1221 | if (desc.count == 0) | |
1222 | continue; | |
1223 | desc.error = 0; | |
1224 | do_generic_file_read(filp,ppos,&desc,file_read_actor); | |
1225 | retval += desc.written; | |
39e88ca2 TH |
1226 | if (desc.error) { |
1227 | retval = retval ?: desc.error; | |
1da177e4 LT |
1228 | break; |
1229 | } | |
c44939ec AM |
1230 | if (desc.count > 0) |
1231 | break; | |
1da177e4 LT |
1232 | } |
1233 | } | |
1234 | out: | |
1235 | return retval; | |
1236 | } | |
1da177e4 LT |
1237 | EXPORT_SYMBOL(generic_file_aio_read); |
1238 | ||
1da177e4 LT |
1239 | static ssize_t |
1240 | do_readahead(struct address_space *mapping, struct file *filp, | |
57f6b96c | 1241 | pgoff_t index, unsigned long nr) |
1da177e4 LT |
1242 | { |
1243 | if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage) | |
1244 | return -EINVAL; | |
1245 | ||
1246 | force_page_cache_readahead(mapping, filp, index, | |
1247 | max_sane_readahead(nr)); | |
1248 | return 0; | |
1249 | } | |
1250 | ||
1251 | asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count) | |
1252 | { | |
1253 | ssize_t ret; | |
1254 | struct file *file; | |
1255 | ||
1256 | ret = -EBADF; | |
1257 | file = fget(fd); | |
1258 | if (file) { | |
1259 | if (file->f_mode & FMODE_READ) { | |
1260 | struct address_space *mapping = file->f_mapping; | |
57f6b96c FW |
1261 | pgoff_t start = offset >> PAGE_CACHE_SHIFT; |
1262 | pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
1263 | unsigned long len = end - start + 1; |
1264 | ret = do_readahead(mapping, file, start, len); | |
1265 | } | |
1266 | fput(file); | |
1267 | } | |
1268 | return ret; | |
1269 | } | |
1270 | ||
1271 | #ifdef CONFIG_MMU | |
485bb99b RD |
1272 | /** |
1273 | * page_cache_read - adds requested page to the page cache if not already there | |
1274 | * @file: file to read | |
1275 | * @offset: page index | |
1276 | * | |
1da177e4 LT |
1277 | * This adds the requested page to the page cache if it isn't already there, |
1278 | * and schedules an I/O to read in its contents from disk. | |
1279 | */ | |
920c7a5d | 1280 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1281 | { |
1282 | struct address_space *mapping = file->f_mapping; | |
1283 | struct page *page; | |
994fc28c | 1284 | int ret; |
1da177e4 | 1285 | |
994fc28c ZB |
1286 | do { |
1287 | page = page_cache_alloc_cold(mapping); | |
1288 | if (!page) | |
1289 | return -ENOMEM; | |
1290 | ||
1291 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); | |
1292 | if (ret == 0) | |
1293 | ret = mapping->a_ops->readpage(file, page); | |
1294 | else if (ret == -EEXIST) | |
1295 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1296 | |
1da177e4 | 1297 | page_cache_release(page); |
1da177e4 | 1298 | |
994fc28c ZB |
1299 | } while (ret == AOP_TRUNCATED_PAGE); |
1300 | ||
1301 | return ret; | |
1da177e4 LT |
1302 | } |
1303 | ||
1304 | #define MMAP_LOTSAMISS (100) | |
1305 | ||
485bb99b | 1306 | /** |
54cb8821 | 1307 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1308 | * @vma: vma in which the fault was taken |
1309 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1310 | * |
54cb8821 | 1311 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1312 | * mapped memory region to read in file data during a page fault. |
1313 | * | |
1314 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1315 | * it in the page cache, and handles the special cases reasonably without | |
1316 | * having a lot of duplicated code. | |
1317 | */ | |
d0217ac0 | 1318 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1319 | { |
1320 | int error; | |
54cb8821 | 1321 | struct file *file = vma->vm_file; |
1da177e4 LT |
1322 | struct address_space *mapping = file->f_mapping; |
1323 | struct file_ra_state *ra = &file->f_ra; | |
1324 | struct inode *inode = mapping->host; | |
1325 | struct page *page; | |
2004dc8e | 1326 | pgoff_t size; |
54cb8821 | 1327 | int did_readaround = 0; |
83c54070 | 1328 | int ret = 0; |
1da177e4 | 1329 | |
1da177e4 | 1330 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
d0217ac0 | 1331 | if (vmf->pgoff >= size) |
5307cc1a | 1332 | return VM_FAULT_SIGBUS; |
1da177e4 LT |
1333 | |
1334 | /* If we don't want any read-ahead, don't bother */ | |
54cb8821 | 1335 | if (VM_RandomReadHint(vma)) |
1da177e4 LT |
1336 | goto no_cached_page; |
1337 | ||
1da177e4 LT |
1338 | /* |
1339 | * Do we have something in the page cache already? | |
1340 | */ | |
1341 | retry_find: | |
d0217ac0 | 1342 | page = find_lock_page(mapping, vmf->pgoff); |
3ea89ee8 FW |
1343 | /* |
1344 | * For sequential accesses, we use the generic readahead logic. | |
1345 | */ | |
1346 | if (VM_SequentialReadHint(vma)) { | |
1347 | if (!page) { | |
cf914a7d | 1348 | page_cache_sync_readahead(mapping, ra, file, |
3ea89ee8 FW |
1349 | vmf->pgoff, 1); |
1350 | page = find_lock_page(mapping, vmf->pgoff); | |
1351 | if (!page) | |
1352 | goto no_cached_page; | |
1353 | } | |
1354 | if (PageReadahead(page)) { | |
cf914a7d | 1355 | page_cache_async_readahead(mapping, ra, file, page, |
3ea89ee8 FW |
1356 | vmf->pgoff, 1); |
1357 | } | |
1358 | } | |
1359 | ||
1da177e4 LT |
1360 | if (!page) { |
1361 | unsigned long ra_pages; | |
1362 | ||
1da177e4 LT |
1363 | ra->mmap_miss++; |
1364 | ||
1365 | /* | |
1366 | * Do we miss much more than hit in this file? If so, | |
1367 | * stop bothering with read-ahead. It will only hurt. | |
1368 | */ | |
0bb7ba6b | 1369 | if (ra->mmap_miss > MMAP_LOTSAMISS) |
1da177e4 LT |
1370 | goto no_cached_page; |
1371 | ||
1372 | /* | |
1373 | * To keep the pgmajfault counter straight, we need to | |
1374 | * check did_readaround, as this is an inner loop. | |
1375 | */ | |
1376 | if (!did_readaround) { | |
d0217ac0 | 1377 | ret = VM_FAULT_MAJOR; |
f8891e5e | 1378 | count_vm_event(PGMAJFAULT); |
1da177e4 LT |
1379 | } |
1380 | did_readaround = 1; | |
1381 | ra_pages = max_sane_readahead(file->f_ra.ra_pages); | |
1382 | if (ra_pages) { | |
1383 | pgoff_t start = 0; | |
1384 | ||
d0217ac0 NP |
1385 | if (vmf->pgoff > ra_pages / 2) |
1386 | start = vmf->pgoff - ra_pages / 2; | |
1da177e4 LT |
1387 | do_page_cache_readahead(mapping, file, start, ra_pages); |
1388 | } | |
d0217ac0 | 1389 | page = find_lock_page(mapping, vmf->pgoff); |
1da177e4 LT |
1390 | if (!page) |
1391 | goto no_cached_page; | |
1392 | } | |
1393 | ||
1394 | if (!did_readaround) | |
0bb7ba6b | 1395 | ra->mmap_miss--; |
1da177e4 LT |
1396 | |
1397 | /* | |
d00806b1 NP |
1398 | * We have a locked page in the page cache, now we need to check |
1399 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1400 | */ |
d00806b1 | 1401 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1402 | goto page_not_uptodate; |
1403 | ||
d00806b1 NP |
1404 | /* Must recheck i_size under page lock */ |
1405 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
d0217ac0 | 1406 | if (unlikely(vmf->pgoff >= size)) { |
d00806b1 | 1407 | unlock_page(page); |
745ad48e | 1408 | page_cache_release(page); |
5307cc1a | 1409 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1410 | } |
1411 | ||
1da177e4 LT |
1412 | /* |
1413 | * Found the page and have a reference on it. | |
1414 | */ | |
1415 | mark_page_accessed(page); | |
f4e6b498 | 1416 | ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT; |
d0217ac0 | 1417 | vmf->page = page; |
83c54070 | 1418 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 1419 | |
1da177e4 LT |
1420 | no_cached_page: |
1421 | /* | |
1422 | * We're only likely to ever get here if MADV_RANDOM is in | |
1423 | * effect. | |
1424 | */ | |
d0217ac0 | 1425 | error = page_cache_read(file, vmf->pgoff); |
1da177e4 LT |
1426 | |
1427 | /* | |
1428 | * The page we want has now been added to the page cache. | |
1429 | * In the unlikely event that someone removed it in the | |
1430 | * meantime, we'll just come back here and read it again. | |
1431 | */ | |
1432 | if (error >= 0) | |
1433 | goto retry_find; | |
1434 | ||
1435 | /* | |
1436 | * An error return from page_cache_read can result if the | |
1437 | * system is low on memory, or a problem occurs while trying | |
1438 | * to schedule I/O. | |
1439 | */ | |
1440 | if (error == -ENOMEM) | |
d0217ac0 NP |
1441 | return VM_FAULT_OOM; |
1442 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1443 | |
1444 | page_not_uptodate: | |
d00806b1 | 1445 | /* IO error path */ |
1da177e4 | 1446 | if (!did_readaround) { |
d0217ac0 | 1447 | ret = VM_FAULT_MAJOR; |
f8891e5e | 1448 | count_vm_event(PGMAJFAULT); |
1da177e4 | 1449 | } |
1da177e4 LT |
1450 | |
1451 | /* | |
1452 | * Umm, take care of errors if the page isn't up-to-date. | |
1453 | * Try to re-read it _once_. We do this synchronously, | |
1454 | * because there really aren't any performance issues here | |
1455 | * and we need to check for errors. | |
1456 | */ | |
1da177e4 | 1457 | ClearPageError(page); |
994fc28c | 1458 | error = mapping->a_ops->readpage(file, page); |
d00806b1 NP |
1459 | page_cache_release(page); |
1460 | ||
1461 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 1462 | goto retry_find; |
1da177e4 | 1463 | |
d00806b1 | 1464 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 1465 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 1466 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
1467 | } |
1468 | EXPORT_SYMBOL(filemap_fault); | |
1469 | ||
1da177e4 | 1470 | struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 1471 | .fault = filemap_fault, |
1da177e4 LT |
1472 | }; |
1473 | ||
1474 | /* This is used for a general mmap of a disk file */ | |
1475 | ||
1476 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1477 | { | |
1478 | struct address_space *mapping = file->f_mapping; | |
1479 | ||
1480 | if (!mapping->a_ops->readpage) | |
1481 | return -ENOEXEC; | |
1482 | file_accessed(file); | |
1483 | vma->vm_ops = &generic_file_vm_ops; | |
d0217ac0 | 1484 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
1485 | return 0; |
1486 | } | |
1da177e4 LT |
1487 | |
1488 | /* | |
1489 | * This is for filesystems which do not implement ->writepage. | |
1490 | */ | |
1491 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
1492 | { | |
1493 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
1494 | return -EINVAL; | |
1495 | return generic_file_mmap(file, vma); | |
1496 | } | |
1497 | #else | |
1498 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1499 | { | |
1500 | return -ENOSYS; | |
1501 | } | |
1502 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
1503 | { | |
1504 | return -ENOSYS; | |
1505 | } | |
1506 | #endif /* CONFIG_MMU */ | |
1507 | ||
1508 | EXPORT_SYMBOL(generic_file_mmap); | |
1509 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
1510 | ||
6fe6900e | 1511 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 1512 | pgoff_t index, |
1da177e4 LT |
1513 | int (*filler)(void *,struct page*), |
1514 | void *data) | |
1515 | { | |
eb2be189 | 1516 | struct page *page; |
1da177e4 LT |
1517 | int err; |
1518 | repeat: | |
1519 | page = find_get_page(mapping, index); | |
1520 | if (!page) { | |
eb2be189 NP |
1521 | page = page_cache_alloc_cold(mapping); |
1522 | if (!page) | |
1523 | return ERR_PTR(-ENOMEM); | |
1524 | err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL); | |
1525 | if (unlikely(err)) { | |
1526 | page_cache_release(page); | |
1527 | if (err == -EEXIST) | |
1528 | goto repeat; | |
1da177e4 | 1529 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
1530 | return ERR_PTR(err); |
1531 | } | |
1da177e4 LT |
1532 | err = filler(data, page); |
1533 | if (err < 0) { | |
1534 | page_cache_release(page); | |
1535 | page = ERR_PTR(err); | |
1536 | } | |
1537 | } | |
1da177e4 LT |
1538 | return page; |
1539 | } | |
1540 | ||
6fe6900e NP |
1541 | /* |
1542 | * Same as read_cache_page, but don't wait for page to become unlocked | |
1543 | * after submitting it to the filler. | |
1da177e4 | 1544 | */ |
6fe6900e | 1545 | struct page *read_cache_page_async(struct address_space *mapping, |
57f6b96c | 1546 | pgoff_t index, |
1da177e4 LT |
1547 | int (*filler)(void *,struct page*), |
1548 | void *data) | |
1549 | { | |
1550 | struct page *page; | |
1551 | int err; | |
1552 | ||
1553 | retry: | |
1554 | page = __read_cache_page(mapping, index, filler, data); | |
1555 | if (IS_ERR(page)) | |
c855ff37 | 1556 | return page; |
1da177e4 LT |
1557 | if (PageUptodate(page)) |
1558 | goto out; | |
1559 | ||
1560 | lock_page(page); | |
1561 | if (!page->mapping) { | |
1562 | unlock_page(page); | |
1563 | page_cache_release(page); | |
1564 | goto retry; | |
1565 | } | |
1566 | if (PageUptodate(page)) { | |
1567 | unlock_page(page); | |
1568 | goto out; | |
1569 | } | |
1570 | err = filler(data, page); | |
1571 | if (err < 0) { | |
1572 | page_cache_release(page); | |
c855ff37 | 1573 | return ERR_PTR(err); |
1da177e4 | 1574 | } |
c855ff37 | 1575 | out: |
6fe6900e NP |
1576 | mark_page_accessed(page); |
1577 | return page; | |
1578 | } | |
1579 | EXPORT_SYMBOL(read_cache_page_async); | |
1580 | ||
1581 | /** | |
1582 | * read_cache_page - read into page cache, fill it if needed | |
1583 | * @mapping: the page's address_space | |
1584 | * @index: the page index | |
1585 | * @filler: function to perform the read | |
1586 | * @data: destination for read data | |
1587 | * | |
1588 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1589 | * not set, try to fill the page then wait for it to become unlocked. | |
1590 | * | |
1591 | * If the page does not get brought uptodate, return -EIO. | |
1592 | */ | |
1593 | struct page *read_cache_page(struct address_space *mapping, | |
57f6b96c | 1594 | pgoff_t index, |
6fe6900e NP |
1595 | int (*filler)(void *,struct page*), |
1596 | void *data) | |
1597 | { | |
1598 | struct page *page; | |
1599 | ||
1600 | page = read_cache_page_async(mapping, index, filler, data); | |
1601 | if (IS_ERR(page)) | |
1602 | goto out; | |
1603 | wait_on_page_locked(page); | |
1604 | if (!PageUptodate(page)) { | |
1605 | page_cache_release(page); | |
1606 | page = ERR_PTR(-EIO); | |
1607 | } | |
1da177e4 LT |
1608 | out: |
1609 | return page; | |
1610 | } | |
1da177e4 LT |
1611 | EXPORT_SYMBOL(read_cache_page); |
1612 | ||
1da177e4 LT |
1613 | /* |
1614 | * The logic we want is | |
1615 | * | |
1616 | * if suid or (sgid and xgrp) | |
1617 | * remove privs | |
1618 | */ | |
01de85e0 | 1619 | int should_remove_suid(struct dentry *dentry) |
1da177e4 LT |
1620 | { |
1621 | mode_t mode = dentry->d_inode->i_mode; | |
1622 | int kill = 0; | |
1da177e4 LT |
1623 | |
1624 | /* suid always must be killed */ | |
1625 | if (unlikely(mode & S_ISUID)) | |
1626 | kill = ATTR_KILL_SUID; | |
1627 | ||
1628 | /* | |
1629 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1630 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1631 | */ | |
1632 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1633 | kill |= ATTR_KILL_SGID; | |
1634 | ||
01de85e0 JA |
1635 | if (unlikely(kill && !capable(CAP_FSETID))) |
1636 | return kill; | |
1da177e4 | 1637 | |
01de85e0 JA |
1638 | return 0; |
1639 | } | |
d23a147b | 1640 | EXPORT_SYMBOL(should_remove_suid); |
01de85e0 JA |
1641 | |
1642 | int __remove_suid(struct dentry *dentry, int kill) | |
1643 | { | |
1644 | struct iattr newattrs; | |
1645 | ||
1646 | newattrs.ia_valid = ATTR_FORCE | kill; | |
1647 | return notify_change(dentry, &newattrs); | |
1648 | } | |
1649 | ||
1650 | int remove_suid(struct dentry *dentry) | |
1651 | { | |
b5376771 SH |
1652 | int killsuid = should_remove_suid(dentry); |
1653 | int killpriv = security_inode_need_killpriv(dentry); | |
1654 | int error = 0; | |
01de85e0 | 1655 | |
b5376771 SH |
1656 | if (killpriv < 0) |
1657 | return killpriv; | |
1658 | if (killpriv) | |
1659 | error = security_inode_killpriv(dentry); | |
1660 | if (!error && killsuid) | |
1661 | error = __remove_suid(dentry, killsuid); | |
01de85e0 | 1662 | |
b5376771 | 1663 | return error; |
1da177e4 LT |
1664 | } |
1665 | EXPORT_SYMBOL(remove_suid); | |
1666 | ||
2f718ffc | 1667 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
1da177e4 LT |
1668 | const struct iovec *iov, size_t base, size_t bytes) |
1669 | { | |
1670 | size_t copied = 0, left = 0; | |
1671 | ||
1672 | while (bytes) { | |
1673 | char __user *buf = iov->iov_base + base; | |
1674 | int copy = min(bytes, iov->iov_len - base); | |
1675 | ||
1676 | base = 0; | |
c22ce143 | 1677 | left = __copy_from_user_inatomic_nocache(vaddr, buf, copy); |
1da177e4 LT |
1678 | copied += copy; |
1679 | bytes -= copy; | |
1680 | vaddr += copy; | |
1681 | iov++; | |
1682 | ||
01408c49 | 1683 | if (unlikely(left)) |
1da177e4 | 1684 | break; |
1da177e4 LT |
1685 | } |
1686 | return copied - left; | |
1687 | } | |
1688 | ||
2f718ffc NP |
1689 | /* |
1690 | * Copy as much as we can into the page and return the number of bytes which | |
1691 | * were sucessfully copied. If a fault is encountered then return the number of | |
1692 | * bytes which were copied. | |
1693 | */ | |
1694 | size_t iov_iter_copy_from_user_atomic(struct page *page, | |
1695 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
1696 | { | |
1697 | char *kaddr; | |
1698 | size_t copied; | |
1699 | ||
1700 | BUG_ON(!in_atomic()); | |
1701 | kaddr = kmap_atomic(page, KM_USER0); | |
1702 | if (likely(i->nr_segs == 1)) { | |
1703 | int left; | |
1704 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
1705 | left = __copy_from_user_inatomic_nocache(kaddr + offset, | |
1706 | buf, bytes); | |
1707 | copied = bytes - left; | |
1708 | } else { | |
1709 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
1710 | i->iov, i->iov_offset, bytes); | |
1711 | } | |
1712 | kunmap_atomic(kaddr, KM_USER0); | |
1713 | ||
1714 | return copied; | |
1715 | } | |
89e10787 | 1716 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); |
2f718ffc NP |
1717 | |
1718 | /* | |
1719 | * This has the same sideeffects and return value as | |
1720 | * iov_iter_copy_from_user_atomic(). | |
1721 | * The difference is that it attempts to resolve faults. | |
1722 | * Page must not be locked. | |
1723 | */ | |
1724 | size_t iov_iter_copy_from_user(struct page *page, | |
1725 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
1726 | { | |
1727 | char *kaddr; | |
1728 | size_t copied; | |
1729 | ||
1730 | kaddr = kmap(page); | |
1731 | if (likely(i->nr_segs == 1)) { | |
1732 | int left; | |
1733 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
1734 | left = __copy_from_user_nocache(kaddr + offset, buf, bytes); | |
1735 | copied = bytes - left; | |
1736 | } else { | |
1737 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
1738 | i->iov, i->iov_offset, bytes); | |
1739 | } | |
1740 | kunmap(page); | |
1741 | return copied; | |
1742 | } | |
89e10787 | 1743 | EXPORT_SYMBOL(iov_iter_copy_from_user); |
2f718ffc | 1744 | |
f7009264 | 1745 | void iov_iter_advance(struct iov_iter *i, size_t bytes) |
2f718ffc | 1746 | { |
f7009264 NP |
1747 | BUG_ON(i->count < bytes); |
1748 | ||
2f718ffc NP |
1749 | if (likely(i->nr_segs == 1)) { |
1750 | i->iov_offset += bytes; | |
f7009264 | 1751 | i->count -= bytes; |
2f718ffc NP |
1752 | } else { |
1753 | const struct iovec *iov = i->iov; | |
1754 | size_t base = i->iov_offset; | |
1755 | ||
124d3b70 NP |
1756 | /* |
1757 | * The !iov->iov_len check ensures we skip over unlikely | |
f7009264 | 1758 | * zero-length segments (without overruning the iovec). |
124d3b70 | 1759 | */ |
f7009264 NP |
1760 | while (bytes || unlikely(!iov->iov_len && i->count)) { |
1761 | int copy; | |
2f718ffc | 1762 | |
f7009264 NP |
1763 | copy = min(bytes, iov->iov_len - base); |
1764 | BUG_ON(!i->count || i->count < copy); | |
1765 | i->count -= copy; | |
2f718ffc NP |
1766 | bytes -= copy; |
1767 | base += copy; | |
1768 | if (iov->iov_len == base) { | |
1769 | iov++; | |
1770 | base = 0; | |
1771 | } | |
1772 | } | |
1773 | i->iov = iov; | |
1774 | i->iov_offset = base; | |
1775 | } | |
1776 | } | |
89e10787 | 1777 | EXPORT_SYMBOL(iov_iter_advance); |
2f718ffc | 1778 | |
afddba49 NP |
1779 | /* |
1780 | * Fault in the first iovec of the given iov_iter, to a maximum length | |
1781 | * of bytes. Returns 0 on success, or non-zero if the memory could not be | |
1782 | * accessed (ie. because it is an invalid address). | |
1783 | * | |
1784 | * writev-intensive code may want this to prefault several iovecs -- that | |
1785 | * would be possible (callers must not rely on the fact that _only_ the | |
1786 | * first iovec will be faulted with the current implementation). | |
1787 | */ | |
1788 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) | |
2f718ffc | 1789 | { |
2f718ffc | 1790 | char __user *buf = i->iov->iov_base + i->iov_offset; |
afddba49 NP |
1791 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); |
1792 | return fault_in_pages_readable(buf, bytes); | |
2f718ffc | 1793 | } |
89e10787 | 1794 | EXPORT_SYMBOL(iov_iter_fault_in_readable); |
2f718ffc NP |
1795 | |
1796 | /* | |
1797 | * Return the count of just the current iov_iter segment. | |
1798 | */ | |
1799 | size_t iov_iter_single_seg_count(struct iov_iter *i) | |
1800 | { | |
1801 | const struct iovec *iov = i->iov; | |
1802 | if (i->nr_segs == 1) | |
1803 | return i->count; | |
1804 | else | |
1805 | return min(i->count, iov->iov_len - i->iov_offset); | |
1806 | } | |
89e10787 | 1807 | EXPORT_SYMBOL(iov_iter_single_seg_count); |
2f718ffc | 1808 | |
1da177e4 LT |
1809 | /* |
1810 | * Performs necessary checks before doing a write | |
1811 | * | |
485bb99b | 1812 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
1813 | * Returns appropriate error code that caller should return or |
1814 | * zero in case that write should be allowed. | |
1815 | */ | |
1816 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) | |
1817 | { | |
1818 | struct inode *inode = file->f_mapping->host; | |
1819 | unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
1820 | ||
1821 | if (unlikely(*pos < 0)) | |
1822 | return -EINVAL; | |
1823 | ||
1da177e4 LT |
1824 | if (!isblk) { |
1825 | /* FIXME: this is for backwards compatibility with 2.4 */ | |
1826 | if (file->f_flags & O_APPEND) | |
1827 | *pos = i_size_read(inode); | |
1828 | ||
1829 | if (limit != RLIM_INFINITY) { | |
1830 | if (*pos >= limit) { | |
1831 | send_sig(SIGXFSZ, current, 0); | |
1832 | return -EFBIG; | |
1833 | } | |
1834 | if (*count > limit - (typeof(limit))*pos) { | |
1835 | *count = limit - (typeof(limit))*pos; | |
1836 | } | |
1837 | } | |
1838 | } | |
1839 | ||
1840 | /* | |
1841 | * LFS rule | |
1842 | */ | |
1843 | if (unlikely(*pos + *count > MAX_NON_LFS && | |
1844 | !(file->f_flags & O_LARGEFILE))) { | |
1845 | if (*pos >= MAX_NON_LFS) { | |
1da177e4 LT |
1846 | return -EFBIG; |
1847 | } | |
1848 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { | |
1849 | *count = MAX_NON_LFS - (unsigned long)*pos; | |
1850 | } | |
1851 | } | |
1852 | ||
1853 | /* | |
1854 | * Are we about to exceed the fs block limit ? | |
1855 | * | |
1856 | * If we have written data it becomes a short write. If we have | |
1857 | * exceeded without writing data we send a signal and return EFBIG. | |
1858 | * Linus frestrict idea will clean these up nicely.. | |
1859 | */ | |
1860 | if (likely(!isblk)) { | |
1861 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { | |
1862 | if (*count || *pos > inode->i_sb->s_maxbytes) { | |
1da177e4 LT |
1863 | return -EFBIG; |
1864 | } | |
1865 | /* zero-length writes at ->s_maxbytes are OK */ | |
1866 | } | |
1867 | ||
1868 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) | |
1869 | *count = inode->i_sb->s_maxbytes - *pos; | |
1870 | } else { | |
9361401e | 1871 | #ifdef CONFIG_BLOCK |
1da177e4 LT |
1872 | loff_t isize; |
1873 | if (bdev_read_only(I_BDEV(inode))) | |
1874 | return -EPERM; | |
1875 | isize = i_size_read(inode); | |
1876 | if (*pos >= isize) { | |
1877 | if (*count || *pos > isize) | |
1878 | return -ENOSPC; | |
1879 | } | |
1880 | ||
1881 | if (*pos + *count > isize) | |
1882 | *count = isize - *pos; | |
9361401e DH |
1883 | #else |
1884 | return -EPERM; | |
1885 | #endif | |
1da177e4 LT |
1886 | } |
1887 | return 0; | |
1888 | } | |
1889 | EXPORT_SYMBOL(generic_write_checks); | |
1890 | ||
afddba49 NP |
1891 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
1892 | loff_t pos, unsigned len, unsigned flags, | |
1893 | struct page **pagep, void **fsdata) | |
1894 | { | |
1895 | const struct address_space_operations *aops = mapping->a_ops; | |
1896 | ||
1897 | if (aops->write_begin) { | |
1898 | return aops->write_begin(file, mapping, pos, len, flags, | |
1899 | pagep, fsdata); | |
1900 | } else { | |
1901 | int ret; | |
1902 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
1903 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); | |
1904 | struct inode *inode = mapping->host; | |
1905 | struct page *page; | |
1906 | again: | |
1907 | page = __grab_cache_page(mapping, index); | |
1908 | *pagep = page; | |
1909 | if (!page) | |
1910 | return -ENOMEM; | |
1911 | ||
1912 | if (flags & AOP_FLAG_UNINTERRUPTIBLE && !PageUptodate(page)) { | |
1913 | /* | |
1914 | * There is no way to resolve a short write situation | |
1915 | * for a !Uptodate page (except by double copying in | |
1916 | * the caller done by generic_perform_write_2copy). | |
1917 | * | |
1918 | * Instead, we have to bring it uptodate here. | |
1919 | */ | |
1920 | ret = aops->readpage(file, page); | |
1921 | page_cache_release(page); | |
1922 | if (ret) { | |
1923 | if (ret == AOP_TRUNCATED_PAGE) | |
1924 | goto again; | |
1925 | return ret; | |
1926 | } | |
1927 | goto again; | |
1928 | } | |
1929 | ||
1930 | ret = aops->prepare_write(file, page, offset, offset+len); | |
1931 | if (ret) { | |
55144768 | 1932 | unlock_page(page); |
afddba49 NP |
1933 | page_cache_release(page); |
1934 | if (pos + len > inode->i_size) | |
1935 | vmtruncate(inode, inode->i_size); | |
afddba49 NP |
1936 | } |
1937 | return ret; | |
1938 | } | |
1939 | } | |
1940 | EXPORT_SYMBOL(pagecache_write_begin); | |
1941 | ||
1942 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
1943 | loff_t pos, unsigned len, unsigned copied, | |
1944 | struct page *page, void *fsdata) | |
1945 | { | |
1946 | const struct address_space_operations *aops = mapping->a_ops; | |
1947 | int ret; | |
1948 | ||
1949 | if (aops->write_end) { | |
1950 | mark_page_accessed(page); | |
1951 | ret = aops->write_end(file, mapping, pos, len, copied, | |
1952 | page, fsdata); | |
1953 | } else { | |
1954 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); | |
1955 | struct inode *inode = mapping->host; | |
1956 | ||
1957 | flush_dcache_page(page); | |
1958 | ret = aops->commit_write(file, page, offset, offset+len); | |
1959 | unlock_page(page); | |
1960 | mark_page_accessed(page); | |
1961 | page_cache_release(page); | |
afddba49 NP |
1962 | |
1963 | if (ret < 0) { | |
1964 | if (pos + len > inode->i_size) | |
1965 | vmtruncate(inode, inode->i_size); | |
1966 | } else if (ret > 0) | |
1967 | ret = min_t(size_t, copied, ret); | |
1968 | else | |
1969 | ret = copied; | |
1970 | } | |
1971 | ||
1972 | return ret; | |
1973 | } | |
1974 | EXPORT_SYMBOL(pagecache_write_end); | |
1975 | ||
1da177e4 LT |
1976 | ssize_t |
1977 | generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, | |
1978 | unsigned long *nr_segs, loff_t pos, loff_t *ppos, | |
1979 | size_t count, size_t ocount) | |
1980 | { | |
1981 | struct file *file = iocb->ki_filp; | |
1982 | struct address_space *mapping = file->f_mapping; | |
1983 | struct inode *inode = mapping->host; | |
1984 | ssize_t written; | |
1985 | ||
1986 | if (count != ocount) | |
1987 | *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); | |
1988 | ||
1989 | written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs); | |
1990 | if (written > 0) { | |
1991 | loff_t end = pos + written; | |
1992 | if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { | |
1993 | i_size_write(inode, end); | |
1994 | mark_inode_dirty(inode); | |
1995 | } | |
1996 | *ppos = end; | |
1997 | } | |
1998 | ||
1999 | /* | |
2000 | * Sync the fs metadata but not the minor inode changes and | |
2001 | * of course not the data as we did direct DMA for the IO. | |
1b1dcc1b | 2002 | * i_mutex is held, which protects generic_osync_inode() from |
8459d86a | 2003 | * livelocking. AIO O_DIRECT ops attempt to sync metadata here. |
1da177e4 | 2004 | */ |
8459d86a ZB |
2005 | if ((written >= 0 || written == -EIOCBQUEUED) && |
2006 | ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
1e8a81c5 HH |
2007 | int err = generic_osync_inode(inode, mapping, OSYNC_METADATA); |
2008 | if (err < 0) | |
2009 | written = err; | |
2010 | } | |
1da177e4 LT |
2011 | return written; |
2012 | } | |
2013 | EXPORT_SYMBOL(generic_file_direct_write); | |
2014 | ||
eb2be189 NP |
2015 | /* |
2016 | * Find or create a page at the given pagecache position. Return the locked | |
2017 | * page. This function is specifically for buffered writes. | |
2018 | */ | |
afddba49 | 2019 | struct page *__grab_cache_page(struct address_space *mapping, pgoff_t index) |
eb2be189 NP |
2020 | { |
2021 | int status; | |
2022 | struct page *page; | |
2023 | repeat: | |
2024 | page = find_lock_page(mapping, index); | |
2025 | if (likely(page)) | |
2026 | return page; | |
2027 | ||
2028 | page = page_cache_alloc(mapping); | |
2029 | if (!page) | |
2030 | return NULL; | |
2031 | status = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL); | |
2032 | if (unlikely(status)) { | |
2033 | page_cache_release(page); | |
2034 | if (status == -EEXIST) | |
2035 | goto repeat; | |
2036 | return NULL; | |
2037 | } | |
2038 | return page; | |
2039 | } | |
afddba49 | 2040 | EXPORT_SYMBOL(__grab_cache_page); |
eb2be189 | 2041 | |
afddba49 NP |
2042 | static ssize_t generic_perform_write_2copy(struct file *file, |
2043 | struct iov_iter *i, loff_t pos) | |
1da177e4 | 2044 | { |
ae37461c | 2045 | struct address_space *mapping = file->f_mapping; |
f5e54d6e | 2046 | const struct address_space_operations *a_ops = mapping->a_ops; |
afddba49 NP |
2047 | struct inode *inode = mapping->host; |
2048 | long status = 0; | |
2049 | ssize_t written = 0; | |
1da177e4 LT |
2050 | |
2051 | do { | |
08291429 | 2052 | struct page *src_page; |
eb2be189 | 2053 | struct page *page; |
ae37461c AM |
2054 | pgoff_t index; /* Pagecache index for current page */ |
2055 | unsigned long offset; /* Offset into pagecache page */ | |
08291429 | 2056 | unsigned long bytes; /* Bytes to write to page */ |
ae37461c | 2057 | size_t copied; /* Bytes copied from user */ |
1da177e4 | 2058 | |
ae37461c | 2059 | offset = (pos & (PAGE_CACHE_SIZE - 1)); |
1da177e4 | 2060 | index = pos >> PAGE_CACHE_SHIFT; |
2f718ffc | 2061 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
afddba49 | 2062 | iov_iter_count(i)); |
41cb8ac0 | 2063 | |
08291429 NP |
2064 | /* |
2065 | * a non-NULL src_page indicates that we're doing the | |
2066 | * copy via get_user_pages and kmap. | |
2067 | */ | |
2068 | src_page = NULL; | |
2069 | ||
41cb8ac0 NP |
2070 | /* |
2071 | * Bring in the user page that we will copy from _first_. | |
2072 | * Otherwise there's a nasty deadlock on copying from the | |
2073 | * same page as we're writing to, without it being marked | |
2074 | * up-to-date. | |
08291429 NP |
2075 | * |
2076 | * Not only is this an optimisation, but it is also required | |
2077 | * to check that the address is actually valid, when atomic | |
2078 | * usercopies are used, below. | |
41cb8ac0 | 2079 | */ |
afddba49 | 2080 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { |
08291429 NP |
2081 | status = -EFAULT; |
2082 | break; | |
2083 | } | |
eb2be189 NP |
2084 | |
2085 | page = __grab_cache_page(mapping, index); | |
1da177e4 LT |
2086 | if (!page) { |
2087 | status = -ENOMEM; | |
2088 | break; | |
2089 | } | |
2090 | ||
08291429 NP |
2091 | /* |
2092 | * non-uptodate pages cannot cope with short copies, and we | |
2093 | * cannot take a pagefault with the destination page locked. | |
2094 | * So pin the source page to copy it. | |
2095 | */ | |
674b892e | 2096 | if (!PageUptodate(page) && !segment_eq(get_fs(), KERNEL_DS)) { |
08291429 NP |
2097 | unlock_page(page); |
2098 | ||
2099 | src_page = alloc_page(GFP_KERNEL); | |
2100 | if (!src_page) { | |
2101 | page_cache_release(page); | |
2102 | status = -ENOMEM; | |
2103 | break; | |
2104 | } | |
2105 | ||
2106 | /* | |
2107 | * Cannot get_user_pages with a page locked for the | |
2108 | * same reason as we can't take a page fault with a | |
2109 | * page locked (as explained below). | |
2110 | */ | |
afddba49 | 2111 | copied = iov_iter_copy_from_user(src_page, i, |
2f718ffc | 2112 | offset, bytes); |
08291429 NP |
2113 | if (unlikely(copied == 0)) { |
2114 | status = -EFAULT; | |
2115 | page_cache_release(page); | |
2116 | page_cache_release(src_page); | |
2117 | break; | |
2118 | } | |
2119 | bytes = copied; | |
2120 | ||
2121 | lock_page(page); | |
2122 | /* | |
2123 | * Can't handle the page going uptodate here, because | |
2124 | * that means we would use non-atomic usercopies, which | |
2125 | * zero out the tail of the page, which can cause | |
2126 | * zeroes to become transiently visible. We could just | |
2127 | * use a non-zeroing copy, but the APIs aren't too | |
2128 | * consistent. | |
2129 | */ | |
2130 | if (unlikely(!page->mapping || PageUptodate(page))) { | |
2131 | unlock_page(page); | |
2132 | page_cache_release(page); | |
2133 | page_cache_release(src_page); | |
2134 | continue; | |
2135 | } | |
08291429 NP |
2136 | } |
2137 | ||
1da177e4 | 2138 | status = a_ops->prepare_write(file, page, offset, offset+bytes); |
64649a58 NP |
2139 | if (unlikely(status)) |
2140 | goto fs_write_aop_error; | |
994fc28c | 2141 | |
08291429 NP |
2142 | if (!src_page) { |
2143 | /* | |
2144 | * Must not enter the pagefault handler here, because | |
2145 | * we hold the page lock, so we might recursively | |
2146 | * deadlock on the same lock, or get an ABBA deadlock | |
2147 | * against a different lock, or against the mmap_sem | |
2148 | * (which nests outside the page lock). So increment | |
2149 | * preempt count, and use _atomic usercopies. | |
2150 | * | |
2151 | * The page is uptodate so we are OK to encounter a | |
2152 | * short copy: if unmodified parts of the page are | |
2153 | * marked dirty and written out to disk, it doesn't | |
2154 | * really matter. | |
2155 | */ | |
2156 | pagefault_disable(); | |
afddba49 | 2157 | copied = iov_iter_copy_from_user_atomic(page, i, |
2f718ffc | 2158 | offset, bytes); |
08291429 NP |
2159 | pagefault_enable(); |
2160 | } else { | |
2161 | void *src, *dst; | |
2162 | src = kmap_atomic(src_page, KM_USER0); | |
2163 | dst = kmap_atomic(page, KM_USER1); | |
2164 | memcpy(dst + offset, src + offset, bytes); | |
2165 | kunmap_atomic(dst, KM_USER1); | |
2166 | kunmap_atomic(src, KM_USER0); | |
2167 | copied = bytes; | |
2168 | } | |
1da177e4 | 2169 | flush_dcache_page(page); |
4a9e5ef1 | 2170 | |
1da177e4 | 2171 | status = a_ops->commit_write(file, page, offset, offset+bytes); |
55144768 | 2172 | if (unlikely(status < 0)) |
64649a58 | 2173 | goto fs_write_aop_error; |
64649a58 | 2174 | if (unlikely(status > 0)) /* filesystem did partial write */ |
08291429 NP |
2175 | copied = min_t(size_t, copied, status); |
2176 | ||
2177 | unlock_page(page); | |
2178 | mark_page_accessed(page); | |
2179 | page_cache_release(page); | |
2180 | if (src_page) | |
2181 | page_cache_release(src_page); | |
64649a58 | 2182 | |
afddba49 | 2183 | iov_iter_advance(i, copied); |
4a9e5ef1 | 2184 | pos += copied; |
afddba49 | 2185 | written += copied; |
4a9e5ef1 | 2186 | |
1da177e4 LT |
2187 | balance_dirty_pages_ratelimited(mapping); |
2188 | cond_resched(); | |
64649a58 NP |
2189 | continue; |
2190 | ||
2191 | fs_write_aop_error: | |
55144768 | 2192 | unlock_page(page); |
64649a58 | 2193 | page_cache_release(page); |
08291429 NP |
2194 | if (src_page) |
2195 | page_cache_release(src_page); | |
64649a58 NP |
2196 | |
2197 | /* | |
2198 | * prepare_write() may have instantiated a few blocks | |
2199 | * outside i_size. Trim these off again. Don't need | |
2200 | * i_size_read because we hold i_mutex. | |
2201 | */ | |
2202 | if (pos + bytes > inode->i_size) | |
2203 | vmtruncate(inode, inode->i_size); | |
55144768 | 2204 | break; |
afddba49 NP |
2205 | } while (iov_iter_count(i)); |
2206 | ||
2207 | return written ? written : status; | |
2208 | } | |
2209 | ||
2210 | static ssize_t generic_perform_write(struct file *file, | |
2211 | struct iov_iter *i, loff_t pos) | |
2212 | { | |
2213 | struct address_space *mapping = file->f_mapping; | |
2214 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2215 | long status = 0; | |
2216 | ssize_t written = 0; | |
674b892e NP |
2217 | unsigned int flags = 0; |
2218 | ||
2219 | /* | |
2220 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2221 | */ | |
2222 | if (segment_eq(get_fs(), KERNEL_DS)) | |
2223 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
afddba49 NP |
2224 | |
2225 | do { | |
2226 | struct page *page; | |
2227 | pgoff_t index; /* Pagecache index for current page */ | |
2228 | unsigned long offset; /* Offset into pagecache page */ | |
2229 | unsigned long bytes; /* Bytes to write to page */ | |
2230 | size_t copied; /* Bytes copied from user */ | |
2231 | void *fsdata; | |
2232 | ||
2233 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
2234 | index = pos >> PAGE_CACHE_SHIFT; | |
2235 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2236 | iov_iter_count(i)); | |
2237 | ||
2238 | again: | |
2239 | ||
2240 | /* | |
2241 | * Bring in the user page that we will copy from _first_. | |
2242 | * Otherwise there's a nasty deadlock on copying from the | |
2243 | * same page as we're writing to, without it being marked | |
2244 | * up-to-date. | |
2245 | * | |
2246 | * Not only is this an optimisation, but it is also required | |
2247 | * to check that the address is actually valid, when atomic | |
2248 | * usercopies are used, below. | |
2249 | */ | |
2250 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2251 | status = -EFAULT; | |
2252 | break; | |
2253 | } | |
2254 | ||
674b892e | 2255 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 NP |
2256 | &page, &fsdata); |
2257 | if (unlikely(status)) | |
2258 | break; | |
2259 | ||
2260 | pagefault_disable(); | |
2261 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); | |
2262 | pagefault_enable(); | |
2263 | flush_dcache_page(page); | |
2264 | ||
2265 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
2266 | page, fsdata); | |
2267 | if (unlikely(status < 0)) | |
2268 | break; | |
2269 | copied = status; | |
2270 | ||
2271 | cond_resched(); | |
2272 | ||
124d3b70 | 2273 | iov_iter_advance(i, copied); |
afddba49 NP |
2274 | if (unlikely(copied == 0)) { |
2275 | /* | |
2276 | * If we were unable to copy any data at all, we must | |
2277 | * fall back to a single segment length write. | |
2278 | * | |
2279 | * If we didn't fallback here, we could livelock | |
2280 | * because not all segments in the iov can be copied at | |
2281 | * once without a pagefault. | |
2282 | */ | |
2283 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2284 | iov_iter_single_seg_count(i)); | |
2285 | goto again; | |
2286 | } | |
afddba49 NP |
2287 | pos += copied; |
2288 | written += copied; | |
2289 | ||
2290 | balance_dirty_pages_ratelimited(mapping); | |
2291 | ||
2292 | } while (iov_iter_count(i)); | |
2293 | ||
2294 | return written ? written : status; | |
2295 | } | |
2296 | ||
2297 | ssize_t | |
2298 | generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, | |
2299 | unsigned long nr_segs, loff_t pos, loff_t *ppos, | |
2300 | size_t count, ssize_t written) | |
2301 | { | |
2302 | struct file *file = iocb->ki_filp; | |
2303 | struct address_space *mapping = file->f_mapping; | |
2304 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2305 | struct inode *inode = mapping->host; | |
2306 | ssize_t status; | |
2307 | struct iov_iter i; | |
2308 | ||
2309 | iov_iter_init(&i, iov, nr_segs, count, written); | |
2310 | if (a_ops->write_begin) | |
2311 | status = generic_perform_write(file, &i, pos); | |
2312 | else | |
2313 | status = generic_perform_write_2copy(file, &i, pos); | |
1da177e4 | 2314 | |
1da177e4 | 2315 | if (likely(status >= 0)) { |
afddba49 NP |
2316 | written += status; |
2317 | *ppos = pos + status; | |
2318 | ||
2319 | /* | |
2320 | * For now, when the user asks for O_SYNC, we'll actually give | |
2321 | * O_DSYNC | |
2322 | */ | |
1da177e4 LT |
2323 | if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) { |
2324 | if (!a_ops->writepage || !is_sync_kiocb(iocb)) | |
2325 | status = generic_osync_inode(inode, mapping, | |
2326 | OSYNC_METADATA|OSYNC_DATA); | |
2327 | } | |
2328 | } | |
2329 | ||
2330 | /* | |
2331 | * If we get here for O_DIRECT writes then we must have fallen through | |
2332 | * to buffered writes (block instantiation inside i_size). So we sync | |
2333 | * the file data here, to try to honour O_DIRECT expectations. | |
2334 | */ | |
2335 | if (unlikely(file->f_flags & O_DIRECT) && written) | |
2336 | status = filemap_write_and_wait(mapping); | |
2337 | ||
1da177e4 LT |
2338 | return written ? written : status; |
2339 | } | |
2340 | EXPORT_SYMBOL(generic_file_buffered_write); | |
2341 | ||
5ce7852c | 2342 | static ssize_t |
1da177e4 LT |
2343 | __generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov, |
2344 | unsigned long nr_segs, loff_t *ppos) | |
2345 | { | |
2346 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2347 | struct address_space * mapping = file->f_mapping; |
1da177e4 LT |
2348 | size_t ocount; /* original count */ |
2349 | size_t count; /* after file limit checks */ | |
2350 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2351 | loff_t pos; |
2352 | ssize_t written; | |
2353 | ssize_t err; | |
2354 | ||
2355 | ocount = 0; | |
0ceb3314 DM |
2356 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
2357 | if (err) | |
2358 | return err; | |
1da177e4 LT |
2359 | |
2360 | count = ocount; | |
2361 | pos = *ppos; | |
2362 | ||
2363 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
2364 | ||
2365 | /* We can write back this queue in page reclaim */ | |
2366 | current->backing_dev_info = mapping->backing_dev_info; | |
2367 | written = 0; | |
2368 | ||
2369 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
2370 | if (err) | |
2371 | goto out; | |
2372 | ||
2373 | if (count == 0) | |
2374 | goto out; | |
2375 | ||
d3ac7f89 | 2376 | err = remove_suid(file->f_path.dentry); |
1da177e4 LT |
2377 | if (err) |
2378 | goto out; | |
2379 | ||
870f4817 | 2380 | file_update_time(file); |
1da177e4 LT |
2381 | |
2382 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
2383 | if (unlikely(file->f_flags & O_DIRECT)) { | |
fb5527e6 JM |
2384 | loff_t endbyte; |
2385 | ssize_t written_buffered; | |
2386 | ||
2387 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, | |
2388 | ppos, count, ocount); | |
1da177e4 LT |
2389 | if (written < 0 || written == count) |
2390 | goto out; | |
2391 | /* | |
2392 | * direct-io write to a hole: fall through to buffered I/O | |
2393 | * for completing the rest of the request. | |
2394 | */ | |
2395 | pos += written; | |
2396 | count -= written; | |
fb5527e6 JM |
2397 | written_buffered = generic_file_buffered_write(iocb, iov, |
2398 | nr_segs, pos, ppos, count, | |
2399 | written); | |
2400 | /* | |
2401 | * If generic_file_buffered_write() retuned a synchronous error | |
2402 | * then we want to return the number of bytes which were | |
2403 | * direct-written, or the error code if that was zero. Note | |
2404 | * that this differs from normal direct-io semantics, which | |
2405 | * will return -EFOO even if some bytes were written. | |
2406 | */ | |
2407 | if (written_buffered < 0) { | |
2408 | err = written_buffered; | |
2409 | goto out; | |
2410 | } | |
1da177e4 | 2411 | |
fb5527e6 JM |
2412 | /* |
2413 | * We need to ensure that the page cache pages are written to | |
2414 | * disk and invalidated to preserve the expected O_DIRECT | |
2415 | * semantics. | |
2416 | */ | |
2417 | endbyte = pos + written_buffered - written - 1; | |
ef51c976 MF |
2418 | err = do_sync_mapping_range(file->f_mapping, pos, endbyte, |
2419 | SYNC_FILE_RANGE_WAIT_BEFORE| | |
2420 | SYNC_FILE_RANGE_WRITE| | |
2421 | SYNC_FILE_RANGE_WAIT_AFTER); | |
fb5527e6 JM |
2422 | if (err == 0) { |
2423 | written = written_buffered; | |
2424 | invalidate_mapping_pages(mapping, | |
2425 | pos >> PAGE_CACHE_SHIFT, | |
2426 | endbyte >> PAGE_CACHE_SHIFT); | |
2427 | } else { | |
2428 | /* | |
2429 | * We don't know how much we wrote, so just return | |
2430 | * the number of bytes which were direct-written | |
2431 | */ | |
2432 | } | |
2433 | } else { | |
2434 | written = generic_file_buffered_write(iocb, iov, nr_segs, | |
2435 | pos, ppos, count, written); | |
2436 | } | |
1da177e4 LT |
2437 | out: |
2438 | current->backing_dev_info = NULL; | |
2439 | return written ? written : err; | |
2440 | } | |
1da177e4 | 2441 | |
027445c3 BP |
2442 | ssize_t generic_file_aio_write_nolock(struct kiocb *iocb, |
2443 | const struct iovec *iov, unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
2444 | { |
2445 | struct file *file = iocb->ki_filp; | |
2446 | struct address_space *mapping = file->f_mapping; | |
2447 | struct inode *inode = mapping->host; | |
2448 | ssize_t ret; | |
1da177e4 | 2449 | |
027445c3 BP |
2450 | BUG_ON(iocb->ki_pos != pos); |
2451 | ||
2452 | ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, | |
2453 | &iocb->ki_pos); | |
1da177e4 LT |
2454 | |
2455 | if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
027445c3 | 2456 | ssize_t err; |
1da177e4 LT |
2457 | |
2458 | err = sync_page_range_nolock(inode, mapping, pos, ret); | |
2459 | if (err < 0) | |
2460 | ret = err; | |
2461 | } | |
2462 | return ret; | |
2463 | } | |
027445c3 | 2464 | EXPORT_SYMBOL(generic_file_aio_write_nolock); |
1da177e4 | 2465 | |
027445c3 BP |
2466 | ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
2467 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
2468 | { |
2469 | struct file *file = iocb->ki_filp; | |
2470 | struct address_space *mapping = file->f_mapping; | |
2471 | struct inode *inode = mapping->host; | |
2472 | ssize_t ret; | |
1da177e4 LT |
2473 | |
2474 | BUG_ON(iocb->ki_pos != pos); | |
2475 | ||
1b1dcc1b | 2476 | mutex_lock(&inode->i_mutex); |
027445c3 BP |
2477 | ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, |
2478 | &iocb->ki_pos); | |
1b1dcc1b | 2479 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
2480 | |
2481 | if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
2482 | ssize_t err; | |
2483 | ||
2484 | err = sync_page_range(inode, mapping, pos, ret); | |
2485 | if (err < 0) | |
2486 | ret = err; | |
2487 | } | |
2488 | return ret; | |
2489 | } | |
2490 | EXPORT_SYMBOL(generic_file_aio_write); | |
2491 | ||
1da177e4 | 2492 | /* |
1b1dcc1b | 2493 | * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something |
1da177e4 LT |
2494 | * went wrong during pagecache shootdown. |
2495 | */ | |
5ce7852c | 2496 | static ssize_t |
1da177e4 LT |
2497 | generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, |
2498 | loff_t offset, unsigned long nr_segs) | |
2499 | { | |
2500 | struct file *file = iocb->ki_filp; | |
2501 | struct address_space *mapping = file->f_mapping; | |
2502 | ssize_t retval; | |
65b8291c ZB |
2503 | size_t write_len; |
2504 | pgoff_t end = 0; /* silence gcc */ | |
1da177e4 LT |
2505 | |
2506 | /* | |
2507 | * If it's a write, unmap all mmappings of the file up-front. This | |
2508 | * will cause any pte dirty bits to be propagated into the pageframes | |
2509 | * for the subsequent filemap_write_and_wait(). | |
2510 | */ | |
2511 | if (rw == WRITE) { | |
2512 | write_len = iov_length(iov, nr_segs); | |
65b8291c | 2513 | end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT; |
1da177e4 LT |
2514 | if (mapping_mapped(mapping)) |
2515 | unmap_mapping_range(mapping, offset, write_len, 0); | |
2516 | } | |
2517 | ||
2518 | retval = filemap_write_and_wait(mapping); | |
65b8291c ZB |
2519 | if (retval) |
2520 | goto out; | |
2521 | ||
2522 | /* | |
2523 | * After a write we want buffered reads to be sure to go to disk to get | |
2524 | * the new data. We invalidate clean cached page from the region we're | |
2525 | * about to write. We do this *before* the write so that we can return | |
2526 | * -EIO without clobbering -EIOCBQUEUED from ->direct_IO(). | |
2527 | */ | |
2528 | if (rw == WRITE && mapping->nrpages) { | |
2529 | retval = invalidate_inode_pages2_range(mapping, | |
1da177e4 | 2530 | offset >> PAGE_CACHE_SHIFT, end); |
65b8291c ZB |
2531 | if (retval) |
2532 | goto out; | |
1da177e4 | 2533 | } |
65b8291c ZB |
2534 | |
2535 | retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs); | |
65b8291c ZB |
2536 | |
2537 | /* | |
2538 | * Finally, try again to invalidate clean pages which might have been | |
bdb76ef5 ZB |
2539 | * cached by non-direct readahead, or faulted in by get_user_pages() |
2540 | * if the source of the write was an mmap'ed region of the file | |
2541 | * we're writing. Either one is a pretty crazy thing to do, | |
2542 | * so we don't support it 100%. If this invalidation | |
2543 | * fails, tough, the write still worked... | |
65b8291c ZB |
2544 | */ |
2545 | if (rw == WRITE && mapping->nrpages) { | |
bdb76ef5 | 2546 | invalidate_inode_pages2_range(mapping, offset >> PAGE_CACHE_SHIFT, end); |
65b8291c ZB |
2547 | } |
2548 | out: | |
1da177e4 LT |
2549 | return retval; |
2550 | } | |
cf9a2ae8 DH |
2551 | |
2552 | /** | |
2553 | * try_to_release_page() - release old fs-specific metadata on a page | |
2554 | * | |
2555 | * @page: the page which the kernel is trying to free | |
2556 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2557 | * | |
2558 | * The address_space is to try to release any data against the page | |
2559 | * (presumably at page->private). If the release was successful, return `1'. | |
2560 | * Otherwise return zero. | |
2561 | * | |
2562 | * The @gfp_mask argument specifies whether I/O may be performed to release | |
2563 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT). | |
2564 | * | |
2565 | * NOTE: @gfp_mask may go away, and this function may become non-blocking. | |
2566 | */ | |
2567 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2568 | { | |
2569 | struct address_space * const mapping = page->mapping; | |
2570 | ||
2571 | BUG_ON(!PageLocked(page)); | |
2572 | if (PageWriteback(page)) | |
2573 | return 0; | |
2574 | ||
2575 | if (mapping && mapping->a_ops->releasepage) | |
2576 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2577 | return try_to_free_buffers(page); | |
2578 | } | |
2579 | ||
2580 | EXPORT_SYMBOL(try_to_release_page); |