]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/filemap.c | |
3 | * | |
4 | * Copyright (C) 1994-1999 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * This file handles the generic file mmap semantics used by | |
9 | * most "normal" filesystems (but you don't /have/ to use this: | |
10 | * the NFS filesystem used to do this differently, for example) | |
11 | */ | |
b95f1b31 | 12 | #include <linux/export.h> |
1da177e4 LT |
13 | #include <linux/compiler.h> |
14 | #include <linux/fs.h> | |
c22ce143 | 15 | #include <linux/uaccess.h> |
1da177e4 | 16 | #include <linux/aio.h> |
c59ede7b | 17 | #include <linux/capability.h> |
1da177e4 | 18 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 19 | #include <linux/gfp.h> |
1da177e4 LT |
20 | #include <linux/mm.h> |
21 | #include <linux/swap.h> | |
22 | #include <linux/mman.h> | |
23 | #include <linux/pagemap.h> | |
24 | #include <linux/file.h> | |
25 | #include <linux/uio.h> | |
26 | #include <linux/hash.h> | |
27 | #include <linux/writeback.h> | |
53253383 | 28 | #include <linux/backing-dev.h> |
1da177e4 LT |
29 | #include <linux/pagevec.h> |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/security.h> | |
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> |
c515e1fd | 36 | #include <linux/cleancache.h> |
0f8053a5 NP |
37 | #include "internal.h" |
38 | ||
1da177e4 | 39 | /* |
1da177e4 LT |
40 | * FIXME: remove all knowledge of the buffer layer from the core VM |
41 | */ | |
148f948b | 42 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 43 | |
1da177e4 LT |
44 | #include <asm/mman.h> |
45 | ||
46 | /* | |
47 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
48 | * though. | |
49 | * | |
50 | * Shared mappings now work. 15.8.1995 Bruno. | |
51 | * | |
52 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
53 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
54 | * | |
55 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
56 | */ | |
57 | ||
58 | /* | |
59 | * Lock ordering: | |
60 | * | |
3d48ae45 | 61 | * ->i_mmap_mutex (truncate_pagecache) |
1da177e4 | 62 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 HD |
63 | * ->swap_lock (exclusive_swap_page, others) |
64 | * ->mapping->tree_lock | |
1da177e4 | 65 | * |
1b1dcc1b | 66 | * ->i_mutex |
3d48ae45 | 67 | * ->i_mmap_mutex (truncate->unmap_mapping_range) |
1da177e4 LT |
68 | * |
69 | * ->mmap_sem | |
3d48ae45 | 70 | * ->i_mmap_mutex |
b8072f09 | 71 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
72 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
73 | * | |
74 | * ->mmap_sem | |
75 | * ->lock_page (access_process_vm) | |
76 | * | |
82591e6e NP |
77 | * ->i_mutex (generic_file_buffered_write) |
78 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) | |
1da177e4 | 79 | * |
f758eeab | 80 | * bdi->wb.list_lock |
a66979ab | 81 | * sb_lock (fs/fs-writeback.c) |
1da177e4 LT |
82 | * ->mapping->tree_lock (__sync_single_inode) |
83 | * | |
3d48ae45 | 84 | * ->i_mmap_mutex |
1da177e4 LT |
85 | * ->anon_vma.lock (vma_adjust) |
86 | * | |
87 | * ->anon_vma.lock | |
b8072f09 | 88 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 89 | * |
b8072f09 | 90 | * ->page_table_lock or pte_lock |
5d337b91 | 91 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
92 | * ->private_lock (try_to_unmap_one) |
93 | * ->tree_lock (try_to_unmap_one) | |
94 | * ->zone.lru_lock (follow_page->mark_page_accessed) | |
053837fc | 95 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
1da177e4 LT |
96 | * ->private_lock (page_remove_rmap->set_page_dirty) |
97 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
f758eeab | 98 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 99 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
f758eeab | 100 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 101 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
102 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
103 | * | |
9a3c531d AK |
104 | * ->i_mmap_mutex |
105 | * ->tasklist_lock (memory_failure, collect_procs_ao) | |
1da177e4 LT |
106 | */ |
107 | ||
108 | /* | |
e64a782f | 109 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 110 | * sure the page is locked and that nobody else uses it - or that usage |
19fd6231 | 111 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 112 | */ |
e64a782f | 113 | void __delete_from_page_cache(struct page *page) |
1da177e4 LT |
114 | { |
115 | struct address_space *mapping = page->mapping; | |
116 | ||
c515e1fd DM |
117 | /* |
118 | * if we're uptodate, flush out into the cleancache, otherwise | |
119 | * invalidate any existing cleancache entries. We can't leave | |
120 | * stale data around in the cleancache once our page is gone | |
121 | */ | |
122 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
123 | cleancache_put_page(page); | |
124 | else | |
125 | cleancache_flush_page(mapping, page); | |
126 | ||
1da177e4 LT |
127 | radix_tree_delete(&mapping->page_tree, page->index); |
128 | page->mapping = NULL; | |
b85e0eff | 129 | /* Leave page->index set: truncation lookup relies upon it */ |
1da177e4 | 130 | mapping->nrpages--; |
347ce434 | 131 | __dec_zone_page_state(page, NR_FILE_PAGES); |
4b02108a KM |
132 | if (PageSwapBacked(page)) |
133 | __dec_zone_page_state(page, NR_SHMEM); | |
45426812 | 134 | BUG_ON(page_mapped(page)); |
3a692790 LT |
135 | |
136 | /* | |
137 | * Some filesystems seem to re-dirty the page even after | |
138 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
139 | * | |
140 | * Fix it up by doing a final dirty accounting check after | |
141 | * having removed the page entirely. | |
142 | */ | |
143 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { | |
144 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
145 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
146 | } | |
1da177e4 LT |
147 | } |
148 | ||
702cfbf9 MK |
149 | /** |
150 | * delete_from_page_cache - delete page from page cache | |
151 | * @page: the page which the kernel is trying to remove from page cache | |
152 | * | |
153 | * This must be called only on pages that have been verified to be in the page | |
154 | * cache and locked. It will never put the page into the free list, the caller | |
155 | * has a reference on the page. | |
156 | */ | |
157 | void delete_from_page_cache(struct page *page) | |
1da177e4 LT |
158 | { |
159 | struct address_space *mapping = page->mapping; | |
6072d13c | 160 | void (*freepage)(struct page *); |
1da177e4 | 161 | |
cd7619d6 | 162 | BUG_ON(!PageLocked(page)); |
1da177e4 | 163 | |
6072d13c | 164 | freepage = mapping->a_ops->freepage; |
19fd6231 | 165 | spin_lock_irq(&mapping->tree_lock); |
e64a782f | 166 | __delete_from_page_cache(page); |
19fd6231 | 167 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 168 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
169 | |
170 | if (freepage) | |
171 | freepage(page); | |
97cecb5a MK |
172 | page_cache_release(page); |
173 | } | |
174 | EXPORT_SYMBOL(delete_from_page_cache); | |
175 | ||
7eaceacc | 176 | static int sleep_on_page(void *word) |
1da177e4 | 177 | { |
1da177e4 LT |
178 | io_schedule(); |
179 | return 0; | |
180 | } | |
181 | ||
7eaceacc | 182 | static int sleep_on_page_killable(void *word) |
2687a356 | 183 | { |
7eaceacc | 184 | sleep_on_page(word); |
2687a356 MW |
185 | return fatal_signal_pending(current) ? -EINTR : 0; |
186 | } | |
187 | ||
1da177e4 | 188 | /** |
485bb99b | 189 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
190 | * @mapping: address space structure to write |
191 | * @start: offset in bytes where the range starts | |
469eb4d0 | 192 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 193 | * @sync_mode: enable synchronous operation |
1da177e4 | 194 | * |
485bb99b RD |
195 | * Start writeback against all of a mapping's dirty pages that lie |
196 | * within the byte offsets <start, end> inclusive. | |
197 | * | |
1da177e4 | 198 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 199 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
200 | * these two operations is that if a dirty page/buffer is encountered, it must |
201 | * be waited upon, and not just skipped over. | |
202 | */ | |
ebcf28e1 AM |
203 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
204 | loff_t end, int sync_mode) | |
1da177e4 LT |
205 | { |
206 | int ret; | |
207 | struct writeback_control wbc = { | |
208 | .sync_mode = sync_mode, | |
05fe478d | 209 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
210 | .range_start = start, |
211 | .range_end = end, | |
1da177e4 LT |
212 | }; |
213 | ||
214 | if (!mapping_cap_writeback_dirty(mapping)) | |
215 | return 0; | |
216 | ||
217 | ret = do_writepages(mapping, &wbc); | |
218 | return ret; | |
219 | } | |
220 | ||
221 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
222 | int sync_mode) | |
223 | { | |
111ebb6e | 224 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
225 | } |
226 | ||
227 | int filemap_fdatawrite(struct address_space *mapping) | |
228 | { | |
229 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
230 | } | |
231 | EXPORT_SYMBOL(filemap_fdatawrite); | |
232 | ||
f4c0a0fd | 233 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 234 | loff_t end) |
1da177e4 LT |
235 | { |
236 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
237 | } | |
f4c0a0fd | 238 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 239 | |
485bb99b RD |
240 | /** |
241 | * filemap_flush - mostly a non-blocking flush | |
242 | * @mapping: target address_space | |
243 | * | |
1da177e4 LT |
244 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
245 | * purposes - I/O may not be started against all dirty pages. | |
246 | */ | |
247 | int filemap_flush(struct address_space *mapping) | |
248 | { | |
249 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
250 | } | |
251 | EXPORT_SYMBOL(filemap_flush); | |
252 | ||
485bb99b | 253 | /** |
94004ed7 CH |
254 | * filemap_fdatawait_range - wait for writeback to complete |
255 | * @mapping: address space structure to wait for | |
256 | * @start_byte: offset in bytes where the range starts | |
257 | * @end_byte: offset in bytes where the range ends (inclusive) | |
485bb99b | 258 | * |
94004ed7 CH |
259 | * Walk the list of under-writeback pages of the given address space |
260 | * in the given range and wait for all of them. | |
1da177e4 | 261 | */ |
94004ed7 CH |
262 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
263 | loff_t end_byte) | |
1da177e4 | 264 | { |
94004ed7 CH |
265 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
266 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
267 | struct pagevec pvec; |
268 | int nr_pages; | |
269 | int ret = 0; | |
1da177e4 | 270 | |
94004ed7 | 271 | if (end_byte < start_byte) |
1da177e4 LT |
272 | return 0; |
273 | ||
274 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
275 | while ((index <= end) && |
276 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
277 | PAGECACHE_TAG_WRITEBACK, | |
278 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
279 | unsigned i; | |
280 | ||
281 | for (i = 0; i < nr_pages; i++) { | |
282 | struct page *page = pvec.pages[i]; | |
283 | ||
284 | /* until radix tree lookup accepts end_index */ | |
285 | if (page->index > end) | |
286 | continue; | |
287 | ||
288 | wait_on_page_writeback(page); | |
212260aa | 289 | if (TestClearPageError(page)) |
1da177e4 LT |
290 | ret = -EIO; |
291 | } | |
292 | pagevec_release(&pvec); | |
293 | cond_resched(); | |
294 | } | |
295 | ||
296 | /* Check for outstanding write errors */ | |
297 | if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
298 | ret = -ENOSPC; | |
299 | if (test_and_clear_bit(AS_EIO, &mapping->flags)) | |
300 | ret = -EIO; | |
301 | ||
302 | return ret; | |
303 | } | |
d3bccb6f JK |
304 | EXPORT_SYMBOL(filemap_fdatawait_range); |
305 | ||
1da177e4 | 306 | /** |
485bb99b | 307 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 308 | * @mapping: address space structure to wait for |
485bb99b RD |
309 | * |
310 | * Walk the list of under-writeback pages of the given address space | |
311 | * and wait for all of them. | |
1da177e4 LT |
312 | */ |
313 | int filemap_fdatawait(struct address_space *mapping) | |
314 | { | |
315 | loff_t i_size = i_size_read(mapping->host); | |
316 | ||
317 | if (i_size == 0) | |
318 | return 0; | |
319 | ||
94004ed7 | 320 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
321 | } |
322 | EXPORT_SYMBOL(filemap_fdatawait); | |
323 | ||
324 | int filemap_write_and_wait(struct address_space *mapping) | |
325 | { | |
28fd1298 | 326 | int err = 0; |
1da177e4 LT |
327 | |
328 | if (mapping->nrpages) { | |
28fd1298 OH |
329 | err = filemap_fdatawrite(mapping); |
330 | /* | |
331 | * Even if the above returned error, the pages may be | |
332 | * written partially (e.g. -ENOSPC), so we wait for it. | |
333 | * But the -EIO is special case, it may indicate the worst | |
334 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
335 | */ | |
336 | if (err != -EIO) { | |
337 | int err2 = filemap_fdatawait(mapping); | |
338 | if (!err) | |
339 | err = err2; | |
340 | } | |
1da177e4 | 341 | } |
28fd1298 | 342 | return err; |
1da177e4 | 343 | } |
28fd1298 | 344 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 345 | |
485bb99b RD |
346 | /** |
347 | * filemap_write_and_wait_range - write out & wait on a file range | |
348 | * @mapping: the address_space for the pages | |
349 | * @lstart: offset in bytes where the range starts | |
350 | * @lend: offset in bytes where the range ends (inclusive) | |
351 | * | |
469eb4d0 AM |
352 | * Write out and wait upon file offsets lstart->lend, inclusive. |
353 | * | |
354 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
355 | * that this function can be used to write to the very end-of-file (end = -1). | |
356 | */ | |
1da177e4 LT |
357 | int filemap_write_and_wait_range(struct address_space *mapping, |
358 | loff_t lstart, loff_t lend) | |
359 | { | |
28fd1298 | 360 | int err = 0; |
1da177e4 LT |
361 | |
362 | if (mapping->nrpages) { | |
28fd1298 OH |
363 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
364 | WB_SYNC_ALL); | |
365 | /* See comment of filemap_write_and_wait() */ | |
366 | if (err != -EIO) { | |
94004ed7 CH |
367 | int err2 = filemap_fdatawait_range(mapping, |
368 | lstart, lend); | |
28fd1298 OH |
369 | if (!err) |
370 | err = err2; | |
371 | } | |
1da177e4 | 372 | } |
28fd1298 | 373 | return err; |
1da177e4 | 374 | } |
f6995585 | 375 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 376 | |
ef6a3c63 MS |
377 | /** |
378 | * replace_page_cache_page - replace a pagecache page with a new one | |
379 | * @old: page to be replaced | |
380 | * @new: page to replace with | |
381 | * @gfp_mask: allocation mode | |
382 | * | |
383 | * This function replaces a page in the pagecache with a new one. On | |
384 | * success it acquires the pagecache reference for the new page and | |
385 | * drops it for the old page. Both the old and new pages must be | |
386 | * locked. This function does not add the new page to the LRU, the | |
387 | * caller must do that. | |
388 | * | |
389 | * The remove + add is atomic. The only way this function can fail is | |
390 | * memory allocation failure. | |
391 | */ | |
392 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
393 | { | |
394 | int error; | |
ef6a3c63 MS |
395 | |
396 | VM_BUG_ON(!PageLocked(old)); | |
397 | VM_BUG_ON(!PageLocked(new)); | |
398 | VM_BUG_ON(new->mapping); | |
399 | ||
ef6a3c63 MS |
400 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
401 | if (!error) { | |
402 | struct address_space *mapping = old->mapping; | |
403 | void (*freepage)(struct page *); | |
404 | ||
405 | pgoff_t offset = old->index; | |
406 | freepage = mapping->a_ops->freepage; | |
407 | ||
408 | page_cache_get(new); | |
409 | new->mapping = mapping; | |
410 | new->index = offset; | |
411 | ||
412 | spin_lock_irq(&mapping->tree_lock); | |
e64a782f | 413 | __delete_from_page_cache(old); |
ef6a3c63 MS |
414 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
415 | BUG_ON(error); | |
416 | mapping->nrpages++; | |
417 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
418 | if (PageSwapBacked(new)) | |
419 | __inc_zone_page_state(new, NR_SHMEM); | |
420 | spin_unlock_irq(&mapping->tree_lock); | |
ab936cbc KH |
421 | /* mem_cgroup codes must not be called under tree_lock */ |
422 | mem_cgroup_replace_page_cache(old, new); | |
ef6a3c63 MS |
423 | radix_tree_preload_end(); |
424 | if (freepage) | |
425 | freepage(old); | |
426 | page_cache_release(old); | |
ef6a3c63 MS |
427 | } |
428 | ||
429 | return error; | |
430 | } | |
431 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
432 | ||
485bb99b | 433 | /** |
e286781d | 434 | * add_to_page_cache_locked - add a locked page to the pagecache |
485bb99b RD |
435 | * @page: page to add |
436 | * @mapping: the page's address_space | |
437 | * @offset: page index | |
438 | * @gfp_mask: page allocation mode | |
439 | * | |
e286781d | 440 | * This function is used to add a page to the pagecache. It must be locked. |
1da177e4 LT |
441 | * This function does not add the page to the LRU. The caller must do that. |
442 | */ | |
e286781d | 443 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, |
6daa0e28 | 444 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 445 | { |
e286781d NP |
446 | int error; |
447 | ||
448 | VM_BUG_ON(!PageLocked(page)); | |
31475dd6 | 449 | VM_BUG_ON(PageSwapBacked(page)); |
e286781d NP |
450 | |
451 | error = mem_cgroup_cache_charge(page, current->mm, | |
2c26fdd7 | 452 | gfp_mask & GFP_RECLAIM_MASK); |
35c754d7 BS |
453 | if (error) |
454 | goto out; | |
1da177e4 | 455 | |
35c754d7 | 456 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
1da177e4 | 457 | if (error == 0) { |
e286781d NP |
458 | page_cache_get(page); |
459 | page->mapping = mapping; | |
460 | page->index = offset; | |
461 | ||
19fd6231 | 462 | spin_lock_irq(&mapping->tree_lock); |
1da177e4 | 463 | error = radix_tree_insert(&mapping->page_tree, offset, page); |
e286781d | 464 | if (likely(!error)) { |
1da177e4 | 465 | mapping->nrpages++; |
347ce434 | 466 | __inc_zone_page_state(page, NR_FILE_PAGES); |
e767e056 | 467 | spin_unlock_irq(&mapping->tree_lock); |
e286781d NP |
468 | } else { |
469 | page->mapping = NULL; | |
b85e0eff | 470 | /* Leave page->index set: truncation relies upon it */ |
e767e056 | 471 | spin_unlock_irq(&mapping->tree_lock); |
69029cd5 | 472 | mem_cgroup_uncharge_cache_page(page); |
e286781d NP |
473 | page_cache_release(page); |
474 | } | |
1da177e4 | 475 | radix_tree_preload_end(); |
35c754d7 | 476 | } else |
69029cd5 | 477 | mem_cgroup_uncharge_cache_page(page); |
8a9f3ccd | 478 | out: |
1da177e4 LT |
479 | return error; |
480 | } | |
e286781d | 481 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
482 | |
483 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 484 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 485 | { |
4f98a2fe RR |
486 | int ret; |
487 | ||
4f98a2fe | 488 | ret = add_to_page_cache(page, mapping, offset, gfp_mask); |
31475dd6 HD |
489 | if (ret == 0) |
490 | lru_cache_add_file(page); | |
1da177e4 LT |
491 | return ret; |
492 | } | |
18bc0bbd | 493 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 494 | |
44110fe3 | 495 | #ifdef CONFIG_NUMA |
2ae88149 | 496 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 497 | { |
c0ff7453 MX |
498 | int n; |
499 | struct page *page; | |
500 | ||
44110fe3 | 501 | if (cpuset_do_page_mem_spread()) { |
c0ff7453 MX |
502 | get_mems_allowed(); |
503 | n = cpuset_mem_spread_node(); | |
504 | page = alloc_pages_exact_node(n, gfp, 0); | |
505 | put_mems_allowed(); | |
506 | return page; | |
44110fe3 | 507 | } |
2ae88149 | 508 | return alloc_pages(gfp, 0); |
44110fe3 | 509 | } |
2ae88149 | 510 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
511 | #endif |
512 | ||
1da177e4 LT |
513 | /* |
514 | * In order to wait for pages to become available there must be | |
515 | * waitqueues associated with pages. By using a hash table of | |
516 | * waitqueues where the bucket discipline is to maintain all | |
517 | * waiters on the same queue and wake all when any of the pages | |
518 | * become available, and for the woken contexts to check to be | |
519 | * sure the appropriate page became available, this saves space | |
520 | * at a cost of "thundering herd" phenomena during rare hash | |
521 | * collisions. | |
522 | */ | |
523 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
524 | { | |
525 | const struct zone *zone = page_zone(page); | |
526 | ||
527 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
528 | } | |
529 | ||
530 | static inline void wake_up_page(struct page *page, int bit) | |
531 | { | |
532 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); | |
533 | } | |
534 | ||
920c7a5d | 535 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
536 | { |
537 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
538 | ||
539 | if (test_bit(bit_nr, &page->flags)) | |
7eaceacc | 540 | __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
541 | TASK_UNINTERRUPTIBLE); |
542 | } | |
543 | EXPORT_SYMBOL(wait_on_page_bit); | |
544 | ||
f62e00cc KM |
545 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
546 | { | |
547 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
548 | ||
549 | if (!test_bit(bit_nr, &page->flags)) | |
550 | return 0; | |
551 | ||
552 | return __wait_on_bit(page_waitqueue(page), &wait, | |
553 | sleep_on_page_killable, TASK_KILLABLE); | |
554 | } | |
555 | ||
385e1ca5 DH |
556 | /** |
557 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
558 | * @page: Page defining the wait queue of interest |
559 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
560 | * |
561 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
562 | */ | |
563 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
564 | { | |
565 | wait_queue_head_t *q = page_waitqueue(page); | |
566 | unsigned long flags; | |
567 | ||
568 | spin_lock_irqsave(&q->lock, flags); | |
569 | __add_wait_queue(q, waiter); | |
570 | spin_unlock_irqrestore(&q->lock, flags); | |
571 | } | |
572 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
573 | ||
1da177e4 | 574 | /** |
485bb99b | 575 | * unlock_page - unlock a locked page |
1da177e4 LT |
576 | * @page: the page |
577 | * | |
578 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
579 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
580 | * mechananism between PageLocked pages and PageWriteback pages is shared. | |
581 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. | |
582 | * | |
8413ac9d NP |
583 | * The mb is necessary to enforce ordering between the clear_bit and the read |
584 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 585 | */ |
920c7a5d | 586 | void unlock_page(struct page *page) |
1da177e4 | 587 | { |
8413ac9d NP |
588 | VM_BUG_ON(!PageLocked(page)); |
589 | clear_bit_unlock(PG_locked, &page->flags); | |
590 | smp_mb__after_clear_bit(); | |
1da177e4 LT |
591 | wake_up_page(page, PG_locked); |
592 | } | |
593 | EXPORT_SYMBOL(unlock_page); | |
594 | ||
485bb99b RD |
595 | /** |
596 | * end_page_writeback - end writeback against a page | |
597 | * @page: the page | |
1da177e4 LT |
598 | */ |
599 | void end_page_writeback(struct page *page) | |
600 | { | |
ac6aadb2 MS |
601 | if (TestClearPageReclaim(page)) |
602 | rotate_reclaimable_page(page); | |
603 | ||
604 | if (!test_clear_page_writeback(page)) | |
605 | BUG(); | |
606 | ||
1da177e4 LT |
607 | smp_mb__after_clear_bit(); |
608 | wake_up_page(page, PG_writeback); | |
609 | } | |
610 | EXPORT_SYMBOL(end_page_writeback); | |
611 | ||
485bb99b RD |
612 | /** |
613 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
614 | * @page: the page to lock | |
1da177e4 | 615 | */ |
920c7a5d | 616 | void __lock_page(struct page *page) |
1da177e4 LT |
617 | { |
618 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
619 | ||
7eaceacc | 620 | __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
621 | TASK_UNINTERRUPTIBLE); |
622 | } | |
623 | EXPORT_SYMBOL(__lock_page); | |
624 | ||
b5606c2d | 625 | int __lock_page_killable(struct page *page) |
2687a356 MW |
626 | { |
627 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
628 | ||
629 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
7eaceacc | 630 | sleep_on_page_killable, TASK_KILLABLE); |
2687a356 | 631 | } |
18bc0bbd | 632 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 633 | |
d065bd81 ML |
634 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
635 | unsigned int flags) | |
636 | { | |
37b23e05 KM |
637 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
638 | /* | |
639 | * CAUTION! In this case, mmap_sem is not released | |
640 | * even though return 0. | |
641 | */ | |
642 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
643 | return 0; | |
644 | ||
645 | up_read(&mm->mmap_sem); | |
646 | if (flags & FAULT_FLAG_KILLABLE) | |
647 | wait_on_page_locked_killable(page); | |
648 | else | |
318b275f | 649 | wait_on_page_locked(page); |
d065bd81 | 650 | return 0; |
37b23e05 KM |
651 | } else { |
652 | if (flags & FAULT_FLAG_KILLABLE) { | |
653 | int ret; | |
654 | ||
655 | ret = __lock_page_killable(page); | |
656 | if (ret) { | |
657 | up_read(&mm->mmap_sem); | |
658 | return 0; | |
659 | } | |
660 | } else | |
661 | __lock_page(page); | |
662 | return 1; | |
d065bd81 ML |
663 | } |
664 | } | |
665 | ||
485bb99b RD |
666 | /** |
667 | * find_get_page - find and get a page reference | |
668 | * @mapping: the address_space to search | |
669 | * @offset: the page index | |
670 | * | |
da6052f7 NP |
671 | * Is there a pagecache struct page at the given (mapping, offset) tuple? |
672 | * If yes, increment its refcount and return it; if no, return NULL. | |
1da177e4 | 673 | */ |
a60637c8 | 674 | struct page *find_get_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 675 | { |
a60637c8 | 676 | void **pagep; |
1da177e4 LT |
677 | struct page *page; |
678 | ||
a60637c8 NP |
679 | rcu_read_lock(); |
680 | repeat: | |
681 | page = NULL; | |
682 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
683 | if (pagep) { | |
684 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
685 | if (unlikely(!page)) |
686 | goto out; | |
a2c16d6c | 687 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
688 | if (radix_tree_deref_retry(page)) |
689 | goto repeat; | |
690 | /* | |
691 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
692 | * here as an exceptional entry: so return it without | |
693 | * attempting to raise page count. | |
694 | */ | |
695 | goto out; | |
a2c16d6c | 696 | } |
a60637c8 NP |
697 | if (!page_cache_get_speculative(page)) |
698 | goto repeat; | |
699 | ||
700 | /* | |
701 | * Has the page moved? | |
702 | * This is part of the lockless pagecache protocol. See | |
703 | * include/linux/pagemap.h for details. | |
704 | */ | |
705 | if (unlikely(page != *pagep)) { | |
706 | page_cache_release(page); | |
707 | goto repeat; | |
708 | } | |
709 | } | |
27d20fdd | 710 | out: |
a60637c8 NP |
711 | rcu_read_unlock(); |
712 | ||
1da177e4 LT |
713 | return page; |
714 | } | |
1da177e4 LT |
715 | EXPORT_SYMBOL(find_get_page); |
716 | ||
1da177e4 LT |
717 | /** |
718 | * find_lock_page - locate, pin and lock a pagecache page | |
67be2dd1 MW |
719 | * @mapping: the address_space to search |
720 | * @offset: the page index | |
1da177e4 LT |
721 | * |
722 | * Locates the desired pagecache page, locks it, increments its reference | |
723 | * count and returns its address. | |
724 | * | |
725 | * Returns zero if the page was not present. find_lock_page() may sleep. | |
726 | */ | |
a60637c8 | 727 | struct page *find_lock_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 LT |
728 | { |
729 | struct page *page; | |
730 | ||
1da177e4 | 731 | repeat: |
a60637c8 | 732 | page = find_get_page(mapping, offset); |
a2c16d6c | 733 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
734 | lock_page(page); |
735 | /* Has the page been truncated? */ | |
736 | if (unlikely(page->mapping != mapping)) { | |
737 | unlock_page(page); | |
738 | page_cache_release(page); | |
739 | goto repeat; | |
1da177e4 | 740 | } |
a60637c8 | 741 | VM_BUG_ON(page->index != offset); |
1da177e4 | 742 | } |
1da177e4 LT |
743 | return page; |
744 | } | |
1da177e4 LT |
745 | EXPORT_SYMBOL(find_lock_page); |
746 | ||
747 | /** | |
748 | * find_or_create_page - locate or add a pagecache page | |
67be2dd1 MW |
749 | * @mapping: the page's address_space |
750 | * @index: the page's index into the mapping | |
751 | * @gfp_mask: page allocation mode | |
1da177e4 LT |
752 | * |
753 | * Locates a page in the pagecache. If the page is not present, a new page | |
754 | * is allocated using @gfp_mask and is added to the pagecache and to the VM's | |
755 | * LRU list. The returned page is locked and has its reference count | |
756 | * incremented. | |
757 | * | |
758 | * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic | |
759 | * allocation! | |
760 | * | |
761 | * find_or_create_page() returns the desired page's address, or zero on | |
762 | * memory exhaustion. | |
763 | */ | |
764 | struct page *find_or_create_page(struct address_space *mapping, | |
57f6b96c | 765 | pgoff_t index, gfp_t gfp_mask) |
1da177e4 | 766 | { |
eb2be189 | 767 | struct page *page; |
1da177e4 LT |
768 | int err; |
769 | repeat: | |
770 | page = find_lock_page(mapping, index); | |
771 | if (!page) { | |
eb2be189 NP |
772 | page = __page_cache_alloc(gfp_mask); |
773 | if (!page) | |
774 | return NULL; | |
67d58ac4 NP |
775 | /* |
776 | * We want a regular kernel memory (not highmem or DMA etc) | |
777 | * allocation for the radix tree nodes, but we need to honour | |
778 | * the context-specific requirements the caller has asked for. | |
779 | * GFP_RECLAIM_MASK collects those requirements. | |
780 | */ | |
781 | err = add_to_page_cache_lru(page, mapping, index, | |
782 | (gfp_mask & GFP_RECLAIM_MASK)); | |
eb2be189 NP |
783 | if (unlikely(err)) { |
784 | page_cache_release(page); | |
785 | page = NULL; | |
786 | if (err == -EEXIST) | |
787 | goto repeat; | |
1da177e4 | 788 | } |
1da177e4 | 789 | } |
1da177e4 LT |
790 | return page; |
791 | } | |
1da177e4 LT |
792 | EXPORT_SYMBOL(find_or_create_page); |
793 | ||
794 | /** | |
795 | * find_get_pages - gang pagecache lookup | |
796 | * @mapping: The address_space to search | |
797 | * @start: The starting page index | |
798 | * @nr_pages: The maximum number of pages | |
799 | * @pages: Where the resulting pages are placed | |
800 | * | |
801 | * find_get_pages() will search for and return a group of up to | |
802 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
803 | * find_get_pages() takes a reference against the returned pages. | |
804 | * | |
805 | * The search returns a group of mapping-contiguous pages with ascending | |
806 | * indexes. There may be holes in the indices due to not-present pages. | |
807 | * | |
808 | * find_get_pages() returns the number of pages which were found. | |
809 | */ | |
810 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
811 | unsigned int nr_pages, struct page **pages) | |
812 | { | |
813 | unsigned int i; | |
814 | unsigned int ret; | |
cc39c6a9 | 815 | unsigned int nr_found, nr_skip; |
a60637c8 NP |
816 | |
817 | rcu_read_lock(); | |
818 | restart: | |
819 | nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, | |
6328650b | 820 | (void ***)pages, NULL, start, nr_pages); |
a60637c8 | 821 | ret = 0; |
cc39c6a9 | 822 | nr_skip = 0; |
a60637c8 NP |
823 | for (i = 0; i < nr_found; i++) { |
824 | struct page *page; | |
825 | repeat: | |
826 | page = radix_tree_deref_slot((void **)pages[i]); | |
827 | if (unlikely(!page)) | |
828 | continue; | |
9d8aa4ea | 829 | |
a2c16d6c | 830 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
831 | if (radix_tree_deref_retry(page)) { |
832 | /* | |
833 | * Transient condition which can only trigger | |
834 | * when entry at index 0 moves out of or back | |
835 | * to root: none yet gotten, safe to restart. | |
836 | */ | |
837 | WARN_ON(start | i); | |
838 | goto restart; | |
839 | } | |
a2c16d6c | 840 | /* |
8079b1c8 HD |
841 | * Otherwise, shmem/tmpfs must be storing a swap entry |
842 | * here as an exceptional entry: so skip over it - | |
843 | * we only reach this from invalidate_mapping_pages(). | |
a2c16d6c | 844 | */ |
cc39c6a9 | 845 | nr_skip++; |
8079b1c8 | 846 | continue; |
27d20fdd | 847 | } |
a60637c8 NP |
848 | |
849 | if (!page_cache_get_speculative(page)) | |
850 | goto repeat; | |
851 | ||
852 | /* Has the page moved? */ | |
853 | if (unlikely(page != *((void **)pages[i]))) { | |
854 | page_cache_release(page); | |
855 | goto repeat; | |
856 | } | |
1da177e4 | 857 | |
a60637c8 NP |
858 | pages[ret] = page; |
859 | ret++; | |
860 | } | |
5b280c0c HD |
861 | |
862 | /* | |
863 | * If all entries were removed before we could secure them, | |
864 | * try again, because callers stop trying once 0 is returned. | |
865 | */ | |
cc39c6a9 | 866 | if (unlikely(!ret && nr_found > nr_skip)) |
5b280c0c | 867 | goto restart; |
a60637c8 | 868 | rcu_read_unlock(); |
1da177e4 LT |
869 | return ret; |
870 | } | |
871 | ||
ebf43500 JA |
872 | /** |
873 | * find_get_pages_contig - gang contiguous pagecache lookup | |
874 | * @mapping: The address_space to search | |
875 | * @index: The starting page index | |
876 | * @nr_pages: The maximum number of pages | |
877 | * @pages: Where the resulting pages are placed | |
878 | * | |
879 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
880 | * that the returned number of pages are guaranteed to be contiguous. | |
881 | * | |
882 | * find_get_pages_contig() returns the number of pages which were found. | |
883 | */ | |
884 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
885 | unsigned int nr_pages, struct page **pages) | |
886 | { | |
887 | unsigned int i; | |
888 | unsigned int ret; | |
a60637c8 NP |
889 | unsigned int nr_found; |
890 | ||
891 | rcu_read_lock(); | |
892 | restart: | |
893 | nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, | |
6328650b | 894 | (void ***)pages, NULL, index, nr_pages); |
a60637c8 NP |
895 | ret = 0; |
896 | for (i = 0; i < nr_found; i++) { | |
897 | struct page *page; | |
898 | repeat: | |
899 | page = radix_tree_deref_slot((void **)pages[i]); | |
900 | if (unlikely(!page)) | |
901 | continue; | |
9d8aa4ea | 902 | |
a2c16d6c | 903 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
904 | if (radix_tree_deref_retry(page)) { |
905 | /* | |
906 | * Transient condition which can only trigger | |
907 | * when entry at index 0 moves out of or back | |
908 | * to root: none yet gotten, safe to restart. | |
909 | */ | |
910 | goto restart; | |
911 | } | |
a2c16d6c | 912 | /* |
8079b1c8 HD |
913 | * Otherwise, shmem/tmpfs must be storing a swap entry |
914 | * here as an exceptional entry: so stop looking for | |
915 | * contiguous pages. | |
a2c16d6c | 916 | */ |
8079b1c8 | 917 | break; |
a2c16d6c | 918 | } |
ebf43500 | 919 | |
a60637c8 NP |
920 | if (!page_cache_get_speculative(page)) |
921 | goto repeat; | |
922 | ||
923 | /* Has the page moved? */ | |
924 | if (unlikely(page != *((void **)pages[i]))) { | |
925 | page_cache_release(page); | |
926 | goto repeat; | |
927 | } | |
928 | ||
9cbb4cb2 NP |
929 | /* |
930 | * must check mapping and index after taking the ref. | |
931 | * otherwise we can get both false positives and false | |
932 | * negatives, which is just confusing to the caller. | |
933 | */ | |
934 | if (page->mapping == NULL || page->index != index) { | |
935 | page_cache_release(page); | |
936 | break; | |
937 | } | |
938 | ||
a60637c8 NP |
939 | pages[ret] = page; |
940 | ret++; | |
ebf43500 JA |
941 | index++; |
942 | } | |
a60637c8 NP |
943 | rcu_read_unlock(); |
944 | return ret; | |
ebf43500 | 945 | } |
ef71c15c | 946 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 947 | |
485bb99b RD |
948 | /** |
949 | * find_get_pages_tag - find and return pages that match @tag | |
950 | * @mapping: the address_space to search | |
951 | * @index: the starting page index | |
952 | * @tag: the tag index | |
953 | * @nr_pages: the maximum number of pages | |
954 | * @pages: where the resulting pages are placed | |
955 | * | |
1da177e4 | 956 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 957 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
958 | */ |
959 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
960 | int tag, unsigned int nr_pages, struct page **pages) | |
961 | { | |
962 | unsigned int i; | |
963 | unsigned int ret; | |
a60637c8 NP |
964 | unsigned int nr_found; |
965 | ||
966 | rcu_read_lock(); | |
967 | restart: | |
968 | nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree, | |
969 | (void ***)pages, *index, nr_pages, tag); | |
970 | ret = 0; | |
971 | for (i = 0; i < nr_found; i++) { | |
972 | struct page *page; | |
973 | repeat: | |
974 | page = radix_tree_deref_slot((void **)pages[i]); | |
975 | if (unlikely(!page)) | |
976 | continue; | |
9d8aa4ea | 977 | |
a2c16d6c | 978 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
979 | if (radix_tree_deref_retry(page)) { |
980 | /* | |
981 | * Transient condition which can only trigger | |
982 | * when entry at index 0 moves out of or back | |
983 | * to root: none yet gotten, safe to restart. | |
984 | */ | |
985 | goto restart; | |
986 | } | |
a2c16d6c | 987 | /* |
8079b1c8 HD |
988 | * This function is never used on a shmem/tmpfs |
989 | * mapping, so a swap entry won't be found here. | |
a2c16d6c | 990 | */ |
8079b1c8 | 991 | BUG(); |
a2c16d6c | 992 | } |
a60637c8 NP |
993 | |
994 | if (!page_cache_get_speculative(page)) | |
995 | goto repeat; | |
996 | ||
997 | /* Has the page moved? */ | |
998 | if (unlikely(page != *((void **)pages[i]))) { | |
999 | page_cache_release(page); | |
1000 | goto repeat; | |
1001 | } | |
1002 | ||
1003 | pages[ret] = page; | |
1004 | ret++; | |
1005 | } | |
5b280c0c HD |
1006 | |
1007 | /* | |
1008 | * If all entries were removed before we could secure them, | |
1009 | * try again, because callers stop trying once 0 is returned. | |
1010 | */ | |
1011 | if (unlikely(!ret && nr_found)) | |
1012 | goto restart; | |
a60637c8 | 1013 | rcu_read_unlock(); |
1da177e4 | 1014 | |
1da177e4 LT |
1015 | if (ret) |
1016 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1017 | |
1da177e4 LT |
1018 | return ret; |
1019 | } | |
ef71c15c | 1020 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1021 | |
485bb99b RD |
1022 | /** |
1023 | * grab_cache_page_nowait - returns locked page at given index in given cache | |
1024 | * @mapping: target address_space | |
1025 | * @index: the page index | |
1026 | * | |
72fd4a35 | 1027 | * Same as grab_cache_page(), but do not wait if the page is unavailable. |
1da177e4 LT |
1028 | * This is intended for speculative data generators, where the data can |
1029 | * be regenerated if the page couldn't be grabbed. This routine should | |
1030 | * be safe to call while holding the lock for another page. | |
1031 | * | |
1032 | * Clear __GFP_FS when allocating the page to avoid recursion into the fs | |
1033 | * and deadlock against the caller's locked page. | |
1034 | */ | |
1035 | struct page * | |
57f6b96c | 1036 | grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) |
1da177e4 LT |
1037 | { |
1038 | struct page *page = find_get_page(mapping, index); | |
1da177e4 LT |
1039 | |
1040 | if (page) { | |
529ae9aa | 1041 | if (trylock_page(page)) |
1da177e4 LT |
1042 | return page; |
1043 | page_cache_release(page); | |
1044 | return NULL; | |
1045 | } | |
2ae88149 | 1046 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); |
67d58ac4 | 1047 | if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) { |
1da177e4 LT |
1048 | page_cache_release(page); |
1049 | page = NULL; | |
1050 | } | |
1051 | return page; | |
1052 | } | |
1da177e4 LT |
1053 | EXPORT_SYMBOL(grab_cache_page_nowait); |
1054 | ||
76d42bd9 WF |
1055 | /* |
1056 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1057 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1058 | * | |
1059 | * ---R__________________________________________B__________ | |
1060 | * ^ reading here ^ bad block(assume 4k) | |
1061 | * | |
1062 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1063 | * => failing the whole request => read(R) => read(R+1) => | |
1064 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1065 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1066 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1067 | * | |
1068 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1069 | */ | |
1070 | static void shrink_readahead_size_eio(struct file *filp, | |
1071 | struct file_ra_state *ra) | |
1072 | { | |
76d42bd9 | 1073 | ra->ra_pages /= 4; |
76d42bd9 WF |
1074 | } |
1075 | ||
485bb99b | 1076 | /** |
36e78914 | 1077 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1078 | * @filp: the file to read |
1079 | * @ppos: current file position | |
1080 | * @desc: read_descriptor | |
1081 | * @actor: read method | |
1082 | * | |
1da177e4 | 1083 | * This is a generic file read routine, and uses the |
485bb99b | 1084 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1085 | * |
1086 | * This is really ugly. But the goto's actually try to clarify some | |
1087 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1088 | */ |
36e78914 CH |
1089 | static void do_generic_file_read(struct file *filp, loff_t *ppos, |
1090 | read_descriptor_t *desc, read_actor_t actor) | |
1da177e4 | 1091 | { |
36e78914 | 1092 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1093 | struct inode *inode = mapping->host; |
36e78914 | 1094 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1095 | pgoff_t index; |
1096 | pgoff_t last_index; | |
1097 | pgoff_t prev_index; | |
1098 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1099 | unsigned int prev_offset; |
1da177e4 | 1100 | int error; |
1da177e4 | 1101 | |
1da177e4 | 1102 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1103 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1104 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
1da177e4 LT |
1105 | last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1106 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1107 | ||
1da177e4 LT |
1108 | for (;;) { |
1109 | struct page *page; | |
57f6b96c | 1110 | pgoff_t end_index; |
a32ea1e1 | 1111 | loff_t isize; |
1da177e4 LT |
1112 | unsigned long nr, ret; |
1113 | ||
1da177e4 | 1114 | cond_resched(); |
1da177e4 LT |
1115 | find_page: |
1116 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1117 | if (!page) { |
cf914a7d | 1118 | page_cache_sync_readahead(mapping, |
7ff81078 | 1119 | ra, filp, |
3ea89ee8 FW |
1120 | index, last_index - index); |
1121 | page = find_get_page(mapping, index); | |
1122 | if (unlikely(page == NULL)) | |
1123 | goto no_cached_page; | |
1124 | } | |
1125 | if (PageReadahead(page)) { | |
cf914a7d | 1126 | page_cache_async_readahead(mapping, |
7ff81078 | 1127 | ra, filp, page, |
3ea89ee8 | 1128 | index, last_index - index); |
1da177e4 | 1129 | } |
8ab22b9a HH |
1130 | if (!PageUptodate(page)) { |
1131 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1132 | !mapping->a_ops->is_partially_uptodate) | |
1133 | goto page_not_up_to_date; | |
529ae9aa | 1134 | if (!trylock_page(page)) |
8ab22b9a | 1135 | goto page_not_up_to_date; |
8d056cb9 DH |
1136 | /* Did it get truncated before we got the lock? */ |
1137 | if (!page->mapping) | |
1138 | goto page_not_up_to_date_locked; | |
8ab22b9a HH |
1139 | if (!mapping->a_ops->is_partially_uptodate(page, |
1140 | desc, offset)) | |
1141 | goto page_not_up_to_date_locked; | |
1142 | unlock_page(page); | |
1143 | } | |
1da177e4 | 1144 | page_ok: |
a32ea1e1 N |
1145 | /* |
1146 | * i_size must be checked after we know the page is Uptodate. | |
1147 | * | |
1148 | * Checking i_size after the check allows us to calculate | |
1149 | * the correct value for "nr", which means the zero-filled | |
1150 | * part of the page is not copied back to userspace (unless | |
1151 | * another truncate extends the file - this is desired though). | |
1152 | */ | |
1153 | ||
1154 | isize = i_size_read(inode); | |
1155 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1156 | if (unlikely(!isize || index > end_index)) { | |
1157 | page_cache_release(page); | |
1158 | goto out; | |
1159 | } | |
1160 | ||
1161 | /* nr is the maximum number of bytes to copy from this page */ | |
1162 | nr = PAGE_CACHE_SIZE; | |
1163 | if (index == end_index) { | |
1164 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1165 | if (nr <= offset) { | |
1166 | page_cache_release(page); | |
1167 | goto out; | |
1168 | } | |
1169 | } | |
1170 | nr = nr - offset; | |
1da177e4 LT |
1171 | |
1172 | /* If users can be writing to this page using arbitrary | |
1173 | * virtual addresses, take care about potential aliasing | |
1174 | * before reading the page on the kernel side. | |
1175 | */ | |
1176 | if (mapping_writably_mapped(mapping)) | |
1177 | flush_dcache_page(page); | |
1178 | ||
1179 | /* | |
ec0f1637 JK |
1180 | * When a sequential read accesses a page several times, |
1181 | * only mark it as accessed the first time. | |
1da177e4 | 1182 | */ |
ec0f1637 | 1183 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1184 | mark_page_accessed(page); |
1185 | prev_index = index; | |
1186 | ||
1187 | /* | |
1188 | * Ok, we have the page, and it's up-to-date, so | |
1189 | * now we can copy it to user space... | |
1190 | * | |
1191 | * The actor routine returns how many bytes were actually used.. | |
1192 | * NOTE! This may not be the same as how much of a user buffer | |
1193 | * we filled up (we may be padding etc), so we can only update | |
1194 | * "pos" here (the actor routine has to update the user buffer | |
1195 | * pointers and the remaining count). | |
1196 | */ | |
1197 | ret = actor(desc, page, offset, nr); | |
1198 | offset += ret; | |
1199 | index += offset >> PAGE_CACHE_SHIFT; | |
1200 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1201 | prev_offset = offset; |
1da177e4 LT |
1202 | |
1203 | page_cache_release(page); | |
1204 | if (ret == nr && desc->count) | |
1205 | continue; | |
1206 | goto out; | |
1207 | ||
1208 | page_not_up_to_date: | |
1209 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1210 | error = lock_page_killable(page); |
1211 | if (unlikely(error)) | |
1212 | goto readpage_error; | |
1da177e4 | 1213 | |
8ab22b9a | 1214 | page_not_up_to_date_locked: |
da6052f7 | 1215 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1216 | if (!page->mapping) { |
1217 | unlock_page(page); | |
1218 | page_cache_release(page); | |
1219 | continue; | |
1220 | } | |
1221 | ||
1222 | /* Did somebody else fill it already? */ | |
1223 | if (PageUptodate(page)) { | |
1224 | unlock_page(page); | |
1225 | goto page_ok; | |
1226 | } | |
1227 | ||
1228 | readpage: | |
91803b49 JM |
1229 | /* |
1230 | * A previous I/O error may have been due to temporary | |
1231 | * failures, eg. multipath errors. | |
1232 | * PG_error will be set again if readpage fails. | |
1233 | */ | |
1234 | ClearPageError(page); | |
1da177e4 LT |
1235 | /* Start the actual read. The read will unlock the page. */ |
1236 | error = mapping->a_ops->readpage(filp, page); | |
1237 | ||
994fc28c ZB |
1238 | if (unlikely(error)) { |
1239 | if (error == AOP_TRUNCATED_PAGE) { | |
1240 | page_cache_release(page); | |
1241 | goto find_page; | |
1242 | } | |
1da177e4 | 1243 | goto readpage_error; |
994fc28c | 1244 | } |
1da177e4 LT |
1245 | |
1246 | if (!PageUptodate(page)) { | |
85462323 ON |
1247 | error = lock_page_killable(page); |
1248 | if (unlikely(error)) | |
1249 | goto readpage_error; | |
1da177e4 LT |
1250 | if (!PageUptodate(page)) { |
1251 | if (page->mapping == NULL) { | |
1252 | /* | |
2ecdc82e | 1253 | * invalidate_mapping_pages got it |
1da177e4 LT |
1254 | */ |
1255 | unlock_page(page); | |
1256 | page_cache_release(page); | |
1257 | goto find_page; | |
1258 | } | |
1259 | unlock_page(page); | |
7ff81078 | 1260 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1261 | error = -EIO; |
1262 | goto readpage_error; | |
1da177e4 LT |
1263 | } |
1264 | unlock_page(page); | |
1265 | } | |
1266 | ||
1da177e4 LT |
1267 | goto page_ok; |
1268 | ||
1269 | readpage_error: | |
1270 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1271 | desc->error = error; | |
1272 | page_cache_release(page); | |
1273 | goto out; | |
1274 | ||
1275 | no_cached_page: | |
1276 | /* | |
1277 | * Ok, it wasn't cached, so we need to create a new | |
1278 | * page.. | |
1279 | */ | |
eb2be189 NP |
1280 | page = page_cache_alloc_cold(mapping); |
1281 | if (!page) { | |
1282 | desc->error = -ENOMEM; | |
1283 | goto out; | |
1da177e4 | 1284 | } |
eb2be189 | 1285 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1286 | index, GFP_KERNEL); |
1287 | if (error) { | |
eb2be189 | 1288 | page_cache_release(page); |
1da177e4 LT |
1289 | if (error == -EEXIST) |
1290 | goto find_page; | |
1291 | desc->error = error; | |
1292 | goto out; | |
1293 | } | |
1da177e4 LT |
1294 | goto readpage; |
1295 | } | |
1296 | ||
1297 | out: | |
7ff81078 FW |
1298 | ra->prev_pos = prev_index; |
1299 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1300 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1301 | |
f4e6b498 | 1302 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1303 | file_accessed(filp); |
1da177e4 | 1304 | } |
1da177e4 LT |
1305 | |
1306 | int file_read_actor(read_descriptor_t *desc, struct page *page, | |
1307 | unsigned long offset, unsigned long size) | |
1308 | { | |
1309 | char *kaddr; | |
1310 | unsigned long left, count = desc->count; | |
1311 | ||
1312 | if (size > count) | |
1313 | size = count; | |
1314 | ||
1315 | /* | |
1316 | * Faults on the destination of a read are common, so do it before | |
1317 | * taking the kmap. | |
1318 | */ | |
1319 | if (!fault_in_pages_writeable(desc->arg.buf, size)) { | |
9b04c5fe | 1320 | kaddr = kmap_atomic(page); |
1da177e4 LT |
1321 | left = __copy_to_user_inatomic(desc->arg.buf, |
1322 | kaddr + offset, size); | |
9b04c5fe | 1323 | kunmap_atomic(kaddr); |
1da177e4 LT |
1324 | if (left == 0) |
1325 | goto success; | |
1326 | } | |
1327 | ||
1328 | /* Do it the slow way */ | |
1329 | kaddr = kmap(page); | |
1330 | left = __copy_to_user(desc->arg.buf, kaddr + offset, size); | |
1331 | kunmap(page); | |
1332 | ||
1333 | if (left) { | |
1334 | size -= left; | |
1335 | desc->error = -EFAULT; | |
1336 | } | |
1337 | success: | |
1338 | desc->count = count - size; | |
1339 | desc->written += size; | |
1340 | desc->arg.buf += size; | |
1341 | return size; | |
1342 | } | |
1343 | ||
0ceb3314 DM |
1344 | /* |
1345 | * Performs necessary checks before doing a write | |
1346 | * @iov: io vector request | |
1347 | * @nr_segs: number of segments in the iovec | |
1348 | * @count: number of bytes to write | |
1349 | * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE | |
1350 | * | |
1351 | * Adjust number of segments and amount of bytes to write (nr_segs should be | |
1352 | * properly initialized first). Returns appropriate error code that caller | |
1353 | * should return or zero in case that write should be allowed. | |
1354 | */ | |
1355 | int generic_segment_checks(const struct iovec *iov, | |
1356 | unsigned long *nr_segs, size_t *count, int access_flags) | |
1357 | { | |
1358 | unsigned long seg; | |
1359 | size_t cnt = 0; | |
1360 | for (seg = 0; seg < *nr_segs; seg++) { | |
1361 | const struct iovec *iv = &iov[seg]; | |
1362 | ||
1363 | /* | |
1364 | * If any segment has a negative length, or the cumulative | |
1365 | * length ever wraps negative then return -EINVAL. | |
1366 | */ | |
1367 | cnt += iv->iov_len; | |
1368 | if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) | |
1369 | return -EINVAL; | |
1370 | if (access_ok(access_flags, iv->iov_base, iv->iov_len)) | |
1371 | continue; | |
1372 | if (seg == 0) | |
1373 | return -EFAULT; | |
1374 | *nr_segs = seg; | |
1375 | cnt -= iv->iov_len; /* This segment is no good */ | |
1376 | break; | |
1377 | } | |
1378 | *count = cnt; | |
1379 | return 0; | |
1380 | } | |
1381 | EXPORT_SYMBOL(generic_segment_checks); | |
1382 | ||
485bb99b | 1383 | /** |
b2abacf3 | 1384 | * generic_file_aio_read - generic filesystem read routine |
485bb99b RD |
1385 | * @iocb: kernel I/O control block |
1386 | * @iov: io vector request | |
1387 | * @nr_segs: number of segments in the iovec | |
b2abacf3 | 1388 | * @pos: current file position |
485bb99b | 1389 | * |
1da177e4 LT |
1390 | * This is the "read()" routine for all filesystems |
1391 | * that can use the page cache directly. | |
1392 | */ | |
1393 | ssize_t | |
543ade1f BP |
1394 | generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
1395 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
1396 | { |
1397 | struct file *filp = iocb->ki_filp; | |
1398 | ssize_t retval; | |
66f998f6 | 1399 | unsigned long seg = 0; |
1da177e4 | 1400 | size_t count; |
543ade1f | 1401 | loff_t *ppos = &iocb->ki_pos; |
1da177e4 LT |
1402 | |
1403 | count = 0; | |
0ceb3314 DM |
1404 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); |
1405 | if (retval) | |
1406 | return retval; | |
1da177e4 LT |
1407 | |
1408 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
1409 | if (filp->f_flags & O_DIRECT) { | |
543ade1f | 1410 | loff_t size; |
1da177e4 LT |
1411 | struct address_space *mapping; |
1412 | struct inode *inode; | |
1413 | ||
1414 | mapping = filp->f_mapping; | |
1415 | inode = mapping->host; | |
1da177e4 LT |
1416 | if (!count) |
1417 | goto out; /* skip atime */ | |
1418 | size = i_size_read(inode); | |
1419 | if (pos < size) { | |
48b47c56 NP |
1420 | retval = filemap_write_and_wait_range(mapping, pos, |
1421 | pos + iov_length(iov, nr_segs) - 1); | |
a969e903 | 1422 | if (!retval) { |
3deaa719 SL |
1423 | struct blk_plug plug; |
1424 | ||
1425 | blk_start_plug(&plug); | |
a969e903 CH |
1426 | retval = mapping->a_ops->direct_IO(READ, iocb, |
1427 | iov, pos, nr_segs); | |
3deaa719 | 1428 | blk_finish_plug(&plug); |
a969e903 | 1429 | } |
66f998f6 | 1430 | if (retval > 0) { |
1da177e4 | 1431 | *ppos = pos + retval; |
66f998f6 JB |
1432 | count -= retval; |
1433 | } | |
1434 | ||
1435 | /* | |
1436 | * Btrfs can have a short DIO read if we encounter | |
1437 | * compressed extents, so if there was an error, or if | |
1438 | * we've already read everything we wanted to, or if | |
1439 | * there was a short read because we hit EOF, go ahead | |
1440 | * and return. Otherwise fallthrough to buffered io for | |
1441 | * the rest of the read. | |
1442 | */ | |
1443 | if (retval < 0 || !count || *ppos >= size) { | |
11fa977e HD |
1444 | file_accessed(filp); |
1445 | goto out; | |
1446 | } | |
0e0bcae3 | 1447 | } |
1da177e4 LT |
1448 | } |
1449 | ||
66f998f6 | 1450 | count = retval; |
11fa977e HD |
1451 | for (seg = 0; seg < nr_segs; seg++) { |
1452 | read_descriptor_t desc; | |
66f998f6 JB |
1453 | loff_t offset = 0; |
1454 | ||
1455 | /* | |
1456 | * If we did a short DIO read we need to skip the section of the | |
1457 | * iov that we've already read data into. | |
1458 | */ | |
1459 | if (count) { | |
1460 | if (count > iov[seg].iov_len) { | |
1461 | count -= iov[seg].iov_len; | |
1462 | continue; | |
1463 | } | |
1464 | offset = count; | |
1465 | count = 0; | |
1466 | } | |
1da177e4 | 1467 | |
11fa977e | 1468 | desc.written = 0; |
66f998f6 JB |
1469 | desc.arg.buf = iov[seg].iov_base + offset; |
1470 | desc.count = iov[seg].iov_len - offset; | |
11fa977e HD |
1471 | if (desc.count == 0) |
1472 | continue; | |
1473 | desc.error = 0; | |
1474 | do_generic_file_read(filp, ppos, &desc, file_read_actor); | |
1475 | retval += desc.written; | |
1476 | if (desc.error) { | |
1477 | retval = retval ?: desc.error; | |
1478 | break; | |
1da177e4 | 1479 | } |
11fa977e HD |
1480 | if (desc.count > 0) |
1481 | break; | |
1da177e4 LT |
1482 | } |
1483 | out: | |
1484 | return retval; | |
1485 | } | |
1da177e4 LT |
1486 | EXPORT_SYMBOL(generic_file_aio_read); |
1487 | ||
1da177e4 LT |
1488 | static ssize_t |
1489 | do_readahead(struct address_space *mapping, struct file *filp, | |
57f6b96c | 1490 | pgoff_t index, unsigned long nr) |
1da177e4 LT |
1491 | { |
1492 | if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage) | |
1493 | return -EINVAL; | |
1494 | ||
f7e839dd | 1495 | force_page_cache_readahead(mapping, filp, index, nr); |
1da177e4 LT |
1496 | return 0; |
1497 | } | |
1498 | ||
6673e0c3 | 1499 | SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count) |
1da177e4 LT |
1500 | { |
1501 | ssize_t ret; | |
1502 | struct file *file; | |
1503 | ||
1504 | ret = -EBADF; | |
1505 | file = fget(fd); | |
1506 | if (file) { | |
1507 | if (file->f_mode & FMODE_READ) { | |
1508 | struct address_space *mapping = file->f_mapping; | |
57f6b96c FW |
1509 | pgoff_t start = offset >> PAGE_CACHE_SHIFT; |
1510 | pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
1511 | unsigned long len = end - start + 1; |
1512 | ret = do_readahead(mapping, file, start, len); | |
1513 | } | |
1514 | fput(file); | |
1515 | } | |
1516 | return ret; | |
1517 | } | |
6673e0c3 HC |
1518 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
1519 | asmlinkage long SyS_readahead(long fd, loff_t offset, long count) | |
1520 | { | |
1521 | return SYSC_readahead((int) fd, offset, (size_t) count); | |
1522 | } | |
1523 | SYSCALL_ALIAS(sys_readahead, SyS_readahead); | |
1524 | #endif | |
1da177e4 LT |
1525 | |
1526 | #ifdef CONFIG_MMU | |
485bb99b RD |
1527 | /** |
1528 | * page_cache_read - adds requested page to the page cache if not already there | |
1529 | * @file: file to read | |
1530 | * @offset: page index | |
1531 | * | |
1da177e4 LT |
1532 | * This adds the requested page to the page cache if it isn't already there, |
1533 | * and schedules an I/O to read in its contents from disk. | |
1534 | */ | |
920c7a5d | 1535 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1536 | { |
1537 | struct address_space *mapping = file->f_mapping; | |
1538 | struct page *page; | |
994fc28c | 1539 | int ret; |
1da177e4 | 1540 | |
994fc28c ZB |
1541 | do { |
1542 | page = page_cache_alloc_cold(mapping); | |
1543 | if (!page) | |
1544 | return -ENOMEM; | |
1545 | ||
1546 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); | |
1547 | if (ret == 0) | |
1548 | ret = mapping->a_ops->readpage(file, page); | |
1549 | else if (ret == -EEXIST) | |
1550 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1551 | |
1da177e4 | 1552 | page_cache_release(page); |
1da177e4 | 1553 | |
994fc28c ZB |
1554 | } while (ret == AOP_TRUNCATED_PAGE); |
1555 | ||
1556 | return ret; | |
1da177e4 LT |
1557 | } |
1558 | ||
1559 | #define MMAP_LOTSAMISS (100) | |
1560 | ||
ef00e08e LT |
1561 | /* |
1562 | * Synchronous readahead happens when we don't even find | |
1563 | * a page in the page cache at all. | |
1564 | */ | |
1565 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1566 | struct file_ra_state *ra, | |
1567 | struct file *file, | |
1568 | pgoff_t offset) | |
1569 | { | |
1570 | unsigned long ra_pages; | |
1571 | struct address_space *mapping = file->f_mapping; | |
1572 | ||
1573 | /* If we don't want any read-ahead, don't bother */ | |
1574 | if (VM_RandomReadHint(vma)) | |
1575 | return; | |
275b12bf WF |
1576 | if (!ra->ra_pages) |
1577 | return; | |
ef00e08e | 1578 | |
2cbea1d3 | 1579 | if (VM_SequentialReadHint(vma)) { |
7ffc59b4 WF |
1580 | page_cache_sync_readahead(mapping, ra, file, offset, |
1581 | ra->ra_pages); | |
ef00e08e LT |
1582 | return; |
1583 | } | |
1584 | ||
207d04ba AK |
1585 | /* Avoid banging the cache line if not needed */ |
1586 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1587 | ra->mmap_miss++; |
1588 | ||
1589 | /* | |
1590 | * Do we miss much more than hit in this file? If so, | |
1591 | * stop bothering with read-ahead. It will only hurt. | |
1592 | */ | |
1593 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1594 | return; | |
1595 | ||
d30a1100 WF |
1596 | /* |
1597 | * mmap read-around | |
1598 | */ | |
ef00e08e | 1599 | ra_pages = max_sane_readahead(ra->ra_pages); |
275b12bf WF |
1600 | ra->start = max_t(long, 0, offset - ra_pages / 2); |
1601 | ra->size = ra_pages; | |
2cbea1d3 | 1602 | ra->async_size = ra_pages / 4; |
275b12bf | 1603 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1604 | } |
1605 | ||
1606 | /* | |
1607 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1608 | * so we want to possibly extend the readahead further.. | |
1609 | */ | |
1610 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1611 | struct file_ra_state *ra, | |
1612 | struct file *file, | |
1613 | struct page *page, | |
1614 | pgoff_t offset) | |
1615 | { | |
1616 | struct address_space *mapping = file->f_mapping; | |
1617 | ||
1618 | /* If we don't want any read-ahead, don't bother */ | |
1619 | if (VM_RandomReadHint(vma)) | |
1620 | return; | |
1621 | if (ra->mmap_miss > 0) | |
1622 | ra->mmap_miss--; | |
1623 | if (PageReadahead(page)) | |
2fad6f5d WF |
1624 | page_cache_async_readahead(mapping, ra, file, |
1625 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1626 | } |
1627 | ||
485bb99b | 1628 | /** |
54cb8821 | 1629 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1630 | * @vma: vma in which the fault was taken |
1631 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1632 | * |
54cb8821 | 1633 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1634 | * mapped memory region to read in file data during a page fault. |
1635 | * | |
1636 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1637 | * it in the page cache, and handles the special cases reasonably without | |
1638 | * having a lot of duplicated code. | |
1639 | */ | |
d0217ac0 | 1640 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1641 | { |
1642 | int error; | |
54cb8821 | 1643 | struct file *file = vma->vm_file; |
1da177e4 LT |
1644 | struct address_space *mapping = file->f_mapping; |
1645 | struct file_ra_state *ra = &file->f_ra; | |
1646 | struct inode *inode = mapping->host; | |
ef00e08e | 1647 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 1648 | struct page *page; |
2004dc8e | 1649 | pgoff_t size; |
83c54070 | 1650 | int ret = 0; |
1da177e4 | 1651 | |
1da177e4 | 1652 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1653 | if (offset >= size) |
5307cc1a | 1654 | return VM_FAULT_SIGBUS; |
1da177e4 | 1655 | |
1da177e4 LT |
1656 | /* |
1657 | * Do we have something in the page cache already? | |
1658 | */ | |
ef00e08e LT |
1659 | page = find_get_page(mapping, offset); |
1660 | if (likely(page)) { | |
1da177e4 | 1661 | /* |
ef00e08e LT |
1662 | * We found the page, so try async readahead before |
1663 | * waiting for the lock. | |
1da177e4 | 1664 | */ |
ef00e08e | 1665 | do_async_mmap_readahead(vma, ra, file, page, offset); |
ef00e08e LT |
1666 | } else { |
1667 | /* No page in the page cache at all */ | |
1668 | do_sync_mmap_readahead(vma, ra, file, offset); | |
1669 | count_vm_event(PGMAJFAULT); | |
456f998e | 1670 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
1671 | ret = VM_FAULT_MAJOR; |
1672 | retry_find: | |
b522c94d | 1673 | page = find_get_page(mapping, offset); |
1da177e4 LT |
1674 | if (!page) |
1675 | goto no_cached_page; | |
1676 | } | |
1677 | ||
d88c0922 ML |
1678 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
1679 | page_cache_release(page); | |
d065bd81 | 1680 | return ret | VM_FAULT_RETRY; |
d88c0922 | 1681 | } |
b522c94d ML |
1682 | |
1683 | /* Did it get truncated? */ | |
1684 | if (unlikely(page->mapping != mapping)) { | |
1685 | unlock_page(page); | |
1686 | put_page(page); | |
1687 | goto retry_find; | |
1688 | } | |
1689 | VM_BUG_ON(page->index != offset); | |
1690 | ||
1da177e4 | 1691 | /* |
d00806b1 NP |
1692 | * We have a locked page in the page cache, now we need to check |
1693 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1694 | */ |
d00806b1 | 1695 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1696 | goto page_not_uptodate; |
1697 | ||
ef00e08e LT |
1698 | /* |
1699 | * Found the page and have a reference on it. | |
1700 | * We must recheck i_size under page lock. | |
1701 | */ | |
d00806b1 | 1702 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1703 | if (unlikely(offset >= size)) { |
d00806b1 | 1704 | unlock_page(page); |
745ad48e | 1705 | page_cache_release(page); |
5307cc1a | 1706 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1707 | } |
1708 | ||
d0217ac0 | 1709 | vmf->page = page; |
83c54070 | 1710 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 1711 | |
1da177e4 LT |
1712 | no_cached_page: |
1713 | /* | |
1714 | * We're only likely to ever get here if MADV_RANDOM is in | |
1715 | * effect. | |
1716 | */ | |
ef00e08e | 1717 | error = page_cache_read(file, offset); |
1da177e4 LT |
1718 | |
1719 | /* | |
1720 | * The page we want has now been added to the page cache. | |
1721 | * In the unlikely event that someone removed it in the | |
1722 | * meantime, we'll just come back here and read it again. | |
1723 | */ | |
1724 | if (error >= 0) | |
1725 | goto retry_find; | |
1726 | ||
1727 | /* | |
1728 | * An error return from page_cache_read can result if the | |
1729 | * system is low on memory, or a problem occurs while trying | |
1730 | * to schedule I/O. | |
1731 | */ | |
1732 | if (error == -ENOMEM) | |
d0217ac0 NP |
1733 | return VM_FAULT_OOM; |
1734 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1735 | |
1736 | page_not_uptodate: | |
1da177e4 LT |
1737 | /* |
1738 | * Umm, take care of errors if the page isn't up-to-date. | |
1739 | * Try to re-read it _once_. We do this synchronously, | |
1740 | * because there really aren't any performance issues here | |
1741 | * and we need to check for errors. | |
1742 | */ | |
1da177e4 | 1743 | ClearPageError(page); |
994fc28c | 1744 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
1745 | if (!error) { |
1746 | wait_on_page_locked(page); | |
1747 | if (!PageUptodate(page)) | |
1748 | error = -EIO; | |
1749 | } | |
d00806b1 NP |
1750 | page_cache_release(page); |
1751 | ||
1752 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 1753 | goto retry_find; |
1da177e4 | 1754 | |
d00806b1 | 1755 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 1756 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 1757 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
1758 | } |
1759 | EXPORT_SYMBOL(filemap_fault); | |
1760 | ||
f0f37e2f | 1761 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 1762 | .fault = filemap_fault, |
1da177e4 LT |
1763 | }; |
1764 | ||
1765 | /* This is used for a general mmap of a disk file */ | |
1766 | ||
1767 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1768 | { | |
1769 | struct address_space *mapping = file->f_mapping; | |
1770 | ||
1771 | if (!mapping->a_ops->readpage) | |
1772 | return -ENOEXEC; | |
1773 | file_accessed(file); | |
1774 | vma->vm_ops = &generic_file_vm_ops; | |
d0217ac0 | 1775 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
1776 | return 0; |
1777 | } | |
1da177e4 LT |
1778 | |
1779 | /* | |
1780 | * This is for filesystems which do not implement ->writepage. | |
1781 | */ | |
1782 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
1783 | { | |
1784 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
1785 | return -EINVAL; | |
1786 | return generic_file_mmap(file, vma); | |
1787 | } | |
1788 | #else | |
1789 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1790 | { | |
1791 | return -ENOSYS; | |
1792 | } | |
1793 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
1794 | { | |
1795 | return -ENOSYS; | |
1796 | } | |
1797 | #endif /* CONFIG_MMU */ | |
1798 | ||
1799 | EXPORT_SYMBOL(generic_file_mmap); | |
1800 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
1801 | ||
6fe6900e | 1802 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 1803 | pgoff_t index, |
5e5358e7 | 1804 | int (*filler)(void *, struct page *), |
0531b2aa LT |
1805 | void *data, |
1806 | gfp_t gfp) | |
1da177e4 | 1807 | { |
eb2be189 | 1808 | struct page *page; |
1da177e4 LT |
1809 | int err; |
1810 | repeat: | |
1811 | page = find_get_page(mapping, index); | |
1812 | if (!page) { | |
0531b2aa | 1813 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
1814 | if (!page) |
1815 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 1816 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 NP |
1817 | if (unlikely(err)) { |
1818 | page_cache_release(page); | |
1819 | if (err == -EEXIST) | |
1820 | goto repeat; | |
1da177e4 | 1821 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
1822 | return ERR_PTR(err); |
1823 | } | |
1da177e4 LT |
1824 | err = filler(data, page); |
1825 | if (err < 0) { | |
1826 | page_cache_release(page); | |
1827 | page = ERR_PTR(err); | |
1828 | } | |
1829 | } | |
1da177e4 LT |
1830 | return page; |
1831 | } | |
1832 | ||
0531b2aa | 1833 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 1834 | pgoff_t index, |
5e5358e7 | 1835 | int (*filler)(void *, struct page *), |
0531b2aa LT |
1836 | void *data, |
1837 | gfp_t gfp) | |
1838 | ||
1da177e4 LT |
1839 | { |
1840 | struct page *page; | |
1841 | int err; | |
1842 | ||
1843 | retry: | |
0531b2aa | 1844 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 1845 | if (IS_ERR(page)) |
c855ff37 | 1846 | return page; |
1da177e4 LT |
1847 | if (PageUptodate(page)) |
1848 | goto out; | |
1849 | ||
1850 | lock_page(page); | |
1851 | if (!page->mapping) { | |
1852 | unlock_page(page); | |
1853 | page_cache_release(page); | |
1854 | goto retry; | |
1855 | } | |
1856 | if (PageUptodate(page)) { | |
1857 | unlock_page(page); | |
1858 | goto out; | |
1859 | } | |
1860 | err = filler(data, page); | |
1861 | if (err < 0) { | |
1862 | page_cache_release(page); | |
c855ff37 | 1863 | return ERR_PTR(err); |
1da177e4 | 1864 | } |
c855ff37 | 1865 | out: |
6fe6900e NP |
1866 | mark_page_accessed(page); |
1867 | return page; | |
1868 | } | |
0531b2aa LT |
1869 | |
1870 | /** | |
1871 | * read_cache_page_async - read into page cache, fill it if needed | |
1872 | * @mapping: the page's address_space | |
1873 | * @index: the page index | |
1874 | * @filler: function to perform the read | |
5e5358e7 | 1875 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa LT |
1876 | * |
1877 | * Same as read_cache_page, but don't wait for page to become unlocked | |
1878 | * after submitting it to the filler. | |
1879 | * | |
1880 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1881 | * not set, try to fill the page but don't wait for it to become unlocked. | |
1882 | * | |
1883 | * If the page does not get brought uptodate, return -EIO. | |
1884 | */ | |
1885 | struct page *read_cache_page_async(struct address_space *mapping, | |
1886 | pgoff_t index, | |
5e5358e7 | 1887 | int (*filler)(void *, struct page *), |
0531b2aa LT |
1888 | void *data) |
1889 | { | |
1890 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
1891 | } | |
6fe6900e NP |
1892 | EXPORT_SYMBOL(read_cache_page_async); |
1893 | ||
0531b2aa LT |
1894 | static struct page *wait_on_page_read(struct page *page) |
1895 | { | |
1896 | if (!IS_ERR(page)) { | |
1897 | wait_on_page_locked(page); | |
1898 | if (!PageUptodate(page)) { | |
1899 | page_cache_release(page); | |
1900 | page = ERR_PTR(-EIO); | |
1901 | } | |
1902 | } | |
1903 | return page; | |
1904 | } | |
1905 | ||
1906 | /** | |
1907 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
1908 | * @mapping: the page's address_space | |
1909 | * @index: the page index | |
1910 | * @gfp: the page allocator flags to use if allocating | |
1911 | * | |
1912 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 1913 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
1914 | * |
1915 | * If the page does not get brought uptodate, return -EIO. | |
1916 | */ | |
1917 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
1918 | pgoff_t index, | |
1919 | gfp_t gfp) | |
1920 | { | |
1921 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
1922 | ||
1923 | return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp)); | |
1924 | } | |
1925 | EXPORT_SYMBOL(read_cache_page_gfp); | |
1926 | ||
6fe6900e NP |
1927 | /** |
1928 | * read_cache_page - read into page cache, fill it if needed | |
1929 | * @mapping: the page's address_space | |
1930 | * @index: the page index | |
1931 | * @filler: function to perform the read | |
5e5358e7 | 1932 | * @data: first arg to filler(data, page) function, often left as NULL |
6fe6900e NP |
1933 | * |
1934 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1935 | * not set, try to fill the page then wait for it to become unlocked. | |
1936 | * | |
1937 | * If the page does not get brought uptodate, return -EIO. | |
1938 | */ | |
1939 | struct page *read_cache_page(struct address_space *mapping, | |
57f6b96c | 1940 | pgoff_t index, |
5e5358e7 | 1941 | int (*filler)(void *, struct page *), |
6fe6900e NP |
1942 | void *data) |
1943 | { | |
0531b2aa | 1944 | return wait_on_page_read(read_cache_page_async(mapping, index, filler, data)); |
1da177e4 | 1945 | } |
1da177e4 LT |
1946 | EXPORT_SYMBOL(read_cache_page); |
1947 | ||
1da177e4 LT |
1948 | /* |
1949 | * The logic we want is | |
1950 | * | |
1951 | * if suid or (sgid and xgrp) | |
1952 | * remove privs | |
1953 | */ | |
01de85e0 | 1954 | int should_remove_suid(struct dentry *dentry) |
1da177e4 | 1955 | { |
649fc7b1 | 1956 | umode_t mode = dentry->d_inode->i_mode; |
1da177e4 | 1957 | int kill = 0; |
1da177e4 LT |
1958 | |
1959 | /* suid always must be killed */ | |
1960 | if (unlikely(mode & S_ISUID)) | |
1961 | kill = ATTR_KILL_SUID; | |
1962 | ||
1963 | /* | |
1964 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1965 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1966 | */ | |
1967 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1968 | kill |= ATTR_KILL_SGID; | |
1969 | ||
7f5ff766 | 1970 | if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) |
01de85e0 | 1971 | return kill; |
1da177e4 | 1972 | |
01de85e0 JA |
1973 | return 0; |
1974 | } | |
d23a147b | 1975 | EXPORT_SYMBOL(should_remove_suid); |
01de85e0 | 1976 | |
7f3d4ee1 | 1977 | static int __remove_suid(struct dentry *dentry, int kill) |
01de85e0 JA |
1978 | { |
1979 | struct iattr newattrs; | |
1980 | ||
1981 | newattrs.ia_valid = ATTR_FORCE | kill; | |
1982 | return notify_change(dentry, &newattrs); | |
1983 | } | |
1984 | ||
2f1936b8 | 1985 | int file_remove_suid(struct file *file) |
01de85e0 | 1986 | { |
2f1936b8 | 1987 | struct dentry *dentry = file->f_path.dentry; |
69b45732 AK |
1988 | struct inode *inode = dentry->d_inode; |
1989 | int killsuid; | |
1990 | int killpriv; | |
b5376771 | 1991 | int error = 0; |
01de85e0 | 1992 | |
69b45732 AK |
1993 | /* Fast path for nothing security related */ |
1994 | if (IS_NOSEC(inode)) | |
1995 | return 0; | |
1996 | ||
1997 | killsuid = should_remove_suid(dentry); | |
1998 | killpriv = security_inode_need_killpriv(dentry); | |
1999 | ||
b5376771 SH |
2000 | if (killpriv < 0) |
2001 | return killpriv; | |
2002 | if (killpriv) | |
2003 | error = security_inode_killpriv(dentry); | |
2004 | if (!error && killsuid) | |
2005 | error = __remove_suid(dentry, killsuid); | |
9e1f1de0 | 2006 | if (!error && (inode->i_sb->s_flags & MS_NOSEC)) |
69b45732 | 2007 | inode->i_flags |= S_NOSEC; |
01de85e0 | 2008 | |
b5376771 | 2009 | return error; |
1da177e4 | 2010 | } |
2f1936b8 | 2011 | EXPORT_SYMBOL(file_remove_suid); |
1da177e4 | 2012 | |
2f718ffc | 2013 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
1da177e4 LT |
2014 | const struct iovec *iov, size_t base, size_t bytes) |
2015 | { | |
f1800536 | 2016 | size_t copied = 0, left = 0; |
1da177e4 LT |
2017 | |
2018 | while (bytes) { | |
2019 | char __user *buf = iov->iov_base + base; | |
2020 | int copy = min(bytes, iov->iov_len - base); | |
2021 | ||
2022 | base = 0; | |
f1800536 | 2023 | left = __copy_from_user_inatomic(vaddr, buf, copy); |
1da177e4 LT |
2024 | copied += copy; |
2025 | bytes -= copy; | |
2026 | vaddr += copy; | |
2027 | iov++; | |
2028 | ||
01408c49 | 2029 | if (unlikely(left)) |
1da177e4 | 2030 | break; |
1da177e4 LT |
2031 | } |
2032 | return copied - left; | |
2033 | } | |
2034 | ||
2f718ffc NP |
2035 | /* |
2036 | * Copy as much as we can into the page and return the number of bytes which | |
af901ca1 | 2037 | * were successfully copied. If a fault is encountered then return the number of |
2f718ffc NP |
2038 | * bytes which were copied. |
2039 | */ | |
2040 | size_t iov_iter_copy_from_user_atomic(struct page *page, | |
2041 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
2042 | { | |
2043 | char *kaddr; | |
2044 | size_t copied; | |
2045 | ||
2046 | BUG_ON(!in_atomic()); | |
9b04c5fe | 2047 | kaddr = kmap_atomic(page); |
2f718ffc NP |
2048 | if (likely(i->nr_segs == 1)) { |
2049 | int left; | |
2050 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 2051 | left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); |
2f718ffc NP |
2052 | copied = bytes - left; |
2053 | } else { | |
2054 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2055 | i->iov, i->iov_offset, bytes); | |
2056 | } | |
9b04c5fe | 2057 | kunmap_atomic(kaddr); |
2f718ffc NP |
2058 | |
2059 | return copied; | |
2060 | } | |
89e10787 | 2061 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); |
2f718ffc NP |
2062 | |
2063 | /* | |
2064 | * This has the same sideeffects and return value as | |
2065 | * iov_iter_copy_from_user_atomic(). | |
2066 | * The difference is that it attempts to resolve faults. | |
2067 | * Page must not be locked. | |
2068 | */ | |
2069 | size_t iov_iter_copy_from_user(struct page *page, | |
2070 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
2071 | { | |
2072 | char *kaddr; | |
2073 | size_t copied; | |
2074 | ||
2075 | kaddr = kmap(page); | |
2076 | if (likely(i->nr_segs == 1)) { | |
2077 | int left; | |
2078 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 2079 | left = __copy_from_user(kaddr + offset, buf, bytes); |
2f718ffc NP |
2080 | copied = bytes - left; |
2081 | } else { | |
2082 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2083 | i->iov, i->iov_offset, bytes); | |
2084 | } | |
2085 | kunmap(page); | |
2086 | return copied; | |
2087 | } | |
89e10787 | 2088 | EXPORT_SYMBOL(iov_iter_copy_from_user); |
2f718ffc | 2089 | |
f7009264 | 2090 | void iov_iter_advance(struct iov_iter *i, size_t bytes) |
2f718ffc | 2091 | { |
f7009264 NP |
2092 | BUG_ON(i->count < bytes); |
2093 | ||
2f718ffc NP |
2094 | if (likely(i->nr_segs == 1)) { |
2095 | i->iov_offset += bytes; | |
f7009264 | 2096 | i->count -= bytes; |
2f718ffc NP |
2097 | } else { |
2098 | const struct iovec *iov = i->iov; | |
2099 | size_t base = i->iov_offset; | |
39be79c1 | 2100 | unsigned long nr_segs = i->nr_segs; |
2f718ffc | 2101 | |
124d3b70 NP |
2102 | /* |
2103 | * The !iov->iov_len check ensures we skip over unlikely | |
f7009264 | 2104 | * zero-length segments (without overruning the iovec). |
124d3b70 | 2105 | */ |
94ad374a | 2106 | while (bytes || unlikely(i->count && !iov->iov_len)) { |
f7009264 | 2107 | int copy; |
2f718ffc | 2108 | |
f7009264 NP |
2109 | copy = min(bytes, iov->iov_len - base); |
2110 | BUG_ON(!i->count || i->count < copy); | |
2111 | i->count -= copy; | |
2f718ffc NP |
2112 | bytes -= copy; |
2113 | base += copy; | |
2114 | if (iov->iov_len == base) { | |
2115 | iov++; | |
39be79c1 | 2116 | nr_segs--; |
2f718ffc NP |
2117 | base = 0; |
2118 | } | |
2119 | } | |
2120 | i->iov = iov; | |
2121 | i->iov_offset = base; | |
39be79c1 | 2122 | i->nr_segs = nr_segs; |
2f718ffc NP |
2123 | } |
2124 | } | |
89e10787 | 2125 | EXPORT_SYMBOL(iov_iter_advance); |
2f718ffc | 2126 | |
afddba49 NP |
2127 | /* |
2128 | * Fault in the first iovec of the given iov_iter, to a maximum length | |
2129 | * of bytes. Returns 0 on success, or non-zero if the memory could not be | |
2130 | * accessed (ie. because it is an invalid address). | |
2131 | * | |
2132 | * writev-intensive code may want this to prefault several iovecs -- that | |
2133 | * would be possible (callers must not rely on the fact that _only_ the | |
2134 | * first iovec will be faulted with the current implementation). | |
2135 | */ | |
2136 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) | |
2f718ffc | 2137 | { |
2f718ffc | 2138 | char __user *buf = i->iov->iov_base + i->iov_offset; |
afddba49 NP |
2139 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); |
2140 | return fault_in_pages_readable(buf, bytes); | |
2f718ffc | 2141 | } |
89e10787 | 2142 | EXPORT_SYMBOL(iov_iter_fault_in_readable); |
2f718ffc NP |
2143 | |
2144 | /* | |
2145 | * Return the count of just the current iov_iter segment. | |
2146 | */ | |
2147 | size_t iov_iter_single_seg_count(struct iov_iter *i) | |
2148 | { | |
2149 | const struct iovec *iov = i->iov; | |
2150 | if (i->nr_segs == 1) | |
2151 | return i->count; | |
2152 | else | |
2153 | return min(i->count, iov->iov_len - i->iov_offset); | |
2154 | } | |
89e10787 | 2155 | EXPORT_SYMBOL(iov_iter_single_seg_count); |
2f718ffc | 2156 | |
1da177e4 LT |
2157 | /* |
2158 | * Performs necessary checks before doing a write | |
2159 | * | |
485bb99b | 2160 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2161 | * Returns appropriate error code that caller should return or |
2162 | * zero in case that write should be allowed. | |
2163 | */ | |
2164 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) | |
2165 | { | |
2166 | struct inode *inode = file->f_mapping->host; | |
59e99e5b | 2167 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
1da177e4 LT |
2168 | |
2169 | if (unlikely(*pos < 0)) | |
2170 | return -EINVAL; | |
2171 | ||
1da177e4 LT |
2172 | if (!isblk) { |
2173 | /* FIXME: this is for backwards compatibility with 2.4 */ | |
2174 | if (file->f_flags & O_APPEND) | |
2175 | *pos = i_size_read(inode); | |
2176 | ||
2177 | if (limit != RLIM_INFINITY) { | |
2178 | if (*pos >= limit) { | |
2179 | send_sig(SIGXFSZ, current, 0); | |
2180 | return -EFBIG; | |
2181 | } | |
2182 | if (*count > limit - (typeof(limit))*pos) { | |
2183 | *count = limit - (typeof(limit))*pos; | |
2184 | } | |
2185 | } | |
2186 | } | |
2187 | ||
2188 | /* | |
2189 | * LFS rule | |
2190 | */ | |
2191 | if (unlikely(*pos + *count > MAX_NON_LFS && | |
2192 | !(file->f_flags & O_LARGEFILE))) { | |
2193 | if (*pos >= MAX_NON_LFS) { | |
1da177e4 LT |
2194 | return -EFBIG; |
2195 | } | |
2196 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { | |
2197 | *count = MAX_NON_LFS - (unsigned long)*pos; | |
2198 | } | |
2199 | } | |
2200 | ||
2201 | /* | |
2202 | * Are we about to exceed the fs block limit ? | |
2203 | * | |
2204 | * If we have written data it becomes a short write. If we have | |
2205 | * exceeded without writing data we send a signal and return EFBIG. | |
2206 | * Linus frestrict idea will clean these up nicely.. | |
2207 | */ | |
2208 | if (likely(!isblk)) { | |
2209 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { | |
2210 | if (*count || *pos > inode->i_sb->s_maxbytes) { | |
1da177e4 LT |
2211 | return -EFBIG; |
2212 | } | |
2213 | /* zero-length writes at ->s_maxbytes are OK */ | |
2214 | } | |
2215 | ||
2216 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) | |
2217 | *count = inode->i_sb->s_maxbytes - *pos; | |
2218 | } else { | |
9361401e | 2219 | #ifdef CONFIG_BLOCK |
1da177e4 LT |
2220 | loff_t isize; |
2221 | if (bdev_read_only(I_BDEV(inode))) | |
2222 | return -EPERM; | |
2223 | isize = i_size_read(inode); | |
2224 | if (*pos >= isize) { | |
2225 | if (*count || *pos > isize) | |
2226 | return -ENOSPC; | |
2227 | } | |
2228 | ||
2229 | if (*pos + *count > isize) | |
2230 | *count = isize - *pos; | |
9361401e DH |
2231 | #else |
2232 | return -EPERM; | |
2233 | #endif | |
1da177e4 LT |
2234 | } |
2235 | return 0; | |
2236 | } | |
2237 | EXPORT_SYMBOL(generic_write_checks); | |
2238 | ||
afddba49 NP |
2239 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2240 | loff_t pos, unsigned len, unsigned flags, | |
2241 | struct page **pagep, void **fsdata) | |
2242 | { | |
2243 | const struct address_space_operations *aops = mapping->a_ops; | |
2244 | ||
4e02ed4b | 2245 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2246 | pagep, fsdata); |
afddba49 NP |
2247 | } |
2248 | EXPORT_SYMBOL(pagecache_write_begin); | |
2249 | ||
2250 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2251 | loff_t pos, unsigned len, unsigned copied, | |
2252 | struct page *page, void *fsdata) | |
2253 | { | |
2254 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2255 | |
4e02ed4b NP |
2256 | mark_page_accessed(page); |
2257 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); | |
afddba49 NP |
2258 | } |
2259 | EXPORT_SYMBOL(pagecache_write_end); | |
2260 | ||
1da177e4 LT |
2261 | ssize_t |
2262 | generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, | |
2263 | unsigned long *nr_segs, loff_t pos, loff_t *ppos, | |
2264 | size_t count, size_t ocount) | |
2265 | { | |
2266 | struct file *file = iocb->ki_filp; | |
2267 | struct address_space *mapping = file->f_mapping; | |
2268 | struct inode *inode = mapping->host; | |
2269 | ssize_t written; | |
a969e903 CH |
2270 | size_t write_len; |
2271 | pgoff_t end; | |
1da177e4 LT |
2272 | |
2273 | if (count != ocount) | |
2274 | *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); | |
2275 | ||
a969e903 CH |
2276 | write_len = iov_length(iov, *nr_segs); |
2277 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; | |
a969e903 | 2278 | |
48b47c56 | 2279 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2280 | if (written) |
2281 | goto out; | |
2282 | ||
2283 | /* | |
2284 | * After a write we want buffered reads to be sure to go to disk to get | |
2285 | * the new data. We invalidate clean cached page from the region we're | |
2286 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2287 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2288 | */ |
2289 | if (mapping->nrpages) { | |
2290 | written = invalidate_inode_pages2_range(mapping, | |
2291 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2292 | /* |
2293 | * If a page can not be invalidated, return 0 to fall back | |
2294 | * to buffered write. | |
2295 | */ | |
2296 | if (written) { | |
2297 | if (written == -EBUSY) | |
2298 | return 0; | |
a969e903 | 2299 | goto out; |
6ccfa806 | 2300 | } |
a969e903 CH |
2301 | } |
2302 | ||
2303 | written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs); | |
2304 | ||
2305 | /* | |
2306 | * Finally, try again to invalidate clean pages which might have been | |
2307 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2308 | * if the source of the write was an mmap'ed region of the file | |
2309 | * we're writing. Either one is a pretty crazy thing to do, | |
2310 | * so we don't support it 100%. If this invalidation | |
2311 | * fails, tough, the write still worked... | |
2312 | */ | |
2313 | if (mapping->nrpages) { | |
2314 | invalidate_inode_pages2_range(mapping, | |
2315 | pos >> PAGE_CACHE_SHIFT, end); | |
2316 | } | |
2317 | ||
1da177e4 | 2318 | if (written > 0) { |
0116651c NK |
2319 | pos += written; |
2320 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { | |
2321 | i_size_write(inode, pos); | |
1da177e4 LT |
2322 | mark_inode_dirty(inode); |
2323 | } | |
0116651c | 2324 | *ppos = pos; |
1da177e4 | 2325 | } |
a969e903 | 2326 | out: |
1da177e4 LT |
2327 | return written; |
2328 | } | |
2329 | EXPORT_SYMBOL(generic_file_direct_write); | |
2330 | ||
eb2be189 NP |
2331 | /* |
2332 | * Find or create a page at the given pagecache position. Return the locked | |
2333 | * page. This function is specifically for buffered writes. | |
2334 | */ | |
54566b2c NP |
2335 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2336 | pgoff_t index, unsigned flags) | |
eb2be189 NP |
2337 | { |
2338 | int status; | |
0faa70cb | 2339 | gfp_t gfp_mask; |
eb2be189 | 2340 | struct page *page; |
54566b2c | 2341 | gfp_t gfp_notmask = 0; |
0faa70cb | 2342 | |
1010bb1b FW |
2343 | gfp_mask = mapping_gfp_mask(mapping); |
2344 | if (mapping_cap_account_dirty(mapping)) | |
2345 | gfp_mask |= __GFP_WRITE; | |
54566b2c NP |
2346 | if (flags & AOP_FLAG_NOFS) |
2347 | gfp_notmask = __GFP_FS; | |
eb2be189 NP |
2348 | repeat: |
2349 | page = find_lock_page(mapping, index); | |
c585a267 | 2350 | if (page) |
3d08bcc8 | 2351 | goto found; |
eb2be189 | 2352 | |
0faa70cb | 2353 | page = __page_cache_alloc(gfp_mask & ~gfp_notmask); |
eb2be189 NP |
2354 | if (!page) |
2355 | return NULL; | |
54566b2c NP |
2356 | status = add_to_page_cache_lru(page, mapping, index, |
2357 | GFP_KERNEL & ~gfp_notmask); | |
eb2be189 NP |
2358 | if (unlikely(status)) { |
2359 | page_cache_release(page); | |
2360 | if (status == -EEXIST) | |
2361 | goto repeat; | |
2362 | return NULL; | |
2363 | } | |
3d08bcc8 DW |
2364 | found: |
2365 | wait_on_page_writeback(page); | |
eb2be189 NP |
2366 | return page; |
2367 | } | |
54566b2c | 2368 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2369 | |
afddba49 NP |
2370 | static ssize_t generic_perform_write(struct file *file, |
2371 | struct iov_iter *i, loff_t pos) | |
2372 | { | |
2373 | struct address_space *mapping = file->f_mapping; | |
2374 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2375 | long status = 0; | |
2376 | ssize_t written = 0; | |
674b892e NP |
2377 | unsigned int flags = 0; |
2378 | ||
2379 | /* | |
2380 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2381 | */ | |
2382 | if (segment_eq(get_fs(), KERNEL_DS)) | |
2383 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
afddba49 NP |
2384 | |
2385 | do { | |
2386 | struct page *page; | |
afddba49 NP |
2387 | unsigned long offset; /* Offset into pagecache page */ |
2388 | unsigned long bytes; /* Bytes to write to page */ | |
2389 | size_t copied; /* Bytes copied from user */ | |
2390 | void *fsdata; | |
2391 | ||
2392 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2393 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2394 | iov_iter_count(i)); | |
2395 | ||
2396 | again: | |
afddba49 NP |
2397 | /* |
2398 | * Bring in the user page that we will copy from _first_. | |
2399 | * Otherwise there's a nasty deadlock on copying from the | |
2400 | * same page as we're writing to, without it being marked | |
2401 | * up-to-date. | |
2402 | * | |
2403 | * Not only is this an optimisation, but it is also required | |
2404 | * to check that the address is actually valid, when atomic | |
2405 | * usercopies are used, below. | |
2406 | */ | |
2407 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2408 | status = -EFAULT; | |
2409 | break; | |
2410 | } | |
2411 | ||
674b892e | 2412 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 NP |
2413 | &page, &fsdata); |
2414 | if (unlikely(status)) | |
2415 | break; | |
2416 | ||
931e80e4 | 2417 | if (mapping_writably_mapped(mapping)) |
2418 | flush_dcache_page(page); | |
2419 | ||
afddba49 NP |
2420 | pagefault_disable(); |
2421 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); | |
2422 | pagefault_enable(); | |
2423 | flush_dcache_page(page); | |
2424 | ||
c8236db9 | 2425 | mark_page_accessed(page); |
afddba49 NP |
2426 | status = a_ops->write_end(file, mapping, pos, bytes, copied, |
2427 | page, fsdata); | |
2428 | if (unlikely(status < 0)) | |
2429 | break; | |
2430 | copied = status; | |
2431 | ||
2432 | cond_resched(); | |
2433 | ||
124d3b70 | 2434 | iov_iter_advance(i, copied); |
afddba49 NP |
2435 | if (unlikely(copied == 0)) { |
2436 | /* | |
2437 | * If we were unable to copy any data at all, we must | |
2438 | * fall back to a single segment length write. | |
2439 | * | |
2440 | * If we didn't fallback here, we could livelock | |
2441 | * because not all segments in the iov can be copied at | |
2442 | * once without a pagefault. | |
2443 | */ | |
2444 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2445 | iov_iter_single_seg_count(i)); | |
2446 | goto again; | |
2447 | } | |
afddba49 NP |
2448 | pos += copied; |
2449 | written += copied; | |
2450 | ||
2451 | balance_dirty_pages_ratelimited(mapping); | |
a50527b1 JK |
2452 | if (fatal_signal_pending(current)) { |
2453 | status = -EINTR; | |
2454 | break; | |
2455 | } | |
afddba49 NP |
2456 | } while (iov_iter_count(i)); |
2457 | ||
2458 | return written ? written : status; | |
2459 | } | |
2460 | ||
2461 | ssize_t | |
2462 | generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, | |
2463 | unsigned long nr_segs, loff_t pos, loff_t *ppos, | |
2464 | size_t count, ssize_t written) | |
2465 | { | |
2466 | struct file *file = iocb->ki_filp; | |
afddba49 NP |
2467 | ssize_t status; |
2468 | struct iov_iter i; | |
2469 | ||
2470 | iov_iter_init(&i, iov, nr_segs, count, written); | |
4e02ed4b | 2471 | status = generic_perform_write(file, &i, pos); |
1da177e4 | 2472 | |
1da177e4 | 2473 | if (likely(status >= 0)) { |
afddba49 NP |
2474 | written += status; |
2475 | *ppos = pos + status; | |
1da177e4 LT |
2476 | } |
2477 | ||
1da177e4 LT |
2478 | return written ? written : status; |
2479 | } | |
2480 | EXPORT_SYMBOL(generic_file_buffered_write); | |
2481 | ||
e4dd9de3 JK |
2482 | /** |
2483 | * __generic_file_aio_write - write data to a file | |
2484 | * @iocb: IO state structure (file, offset, etc.) | |
2485 | * @iov: vector with data to write | |
2486 | * @nr_segs: number of segments in the vector | |
2487 | * @ppos: position where to write | |
2488 | * | |
2489 | * This function does all the work needed for actually writing data to a | |
2490 | * file. It does all basic checks, removes SUID from the file, updates | |
2491 | * modification times and calls proper subroutines depending on whether we | |
2492 | * do direct IO or a standard buffered write. | |
2493 | * | |
2494 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2495 | * object which does not need locking at all. | |
2496 | * | |
2497 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2498 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2499 | * avoid syncing under i_mutex. | |
2500 | */ | |
2501 | ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
2502 | unsigned long nr_segs, loff_t *ppos) | |
1da177e4 LT |
2503 | { |
2504 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2505 | struct address_space * mapping = file->f_mapping; |
1da177e4 LT |
2506 | size_t ocount; /* original count */ |
2507 | size_t count; /* after file limit checks */ | |
2508 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2509 | loff_t pos; |
2510 | ssize_t written; | |
2511 | ssize_t err; | |
2512 | ||
2513 | ocount = 0; | |
0ceb3314 DM |
2514 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
2515 | if (err) | |
2516 | return err; | |
1da177e4 LT |
2517 | |
2518 | count = ocount; | |
2519 | pos = *ppos; | |
2520 | ||
2521 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
2522 | ||
2523 | /* We can write back this queue in page reclaim */ | |
2524 | current->backing_dev_info = mapping->backing_dev_info; | |
2525 | written = 0; | |
2526 | ||
2527 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
2528 | if (err) | |
2529 | goto out; | |
2530 | ||
2531 | if (count == 0) | |
2532 | goto out; | |
2533 | ||
2f1936b8 | 2534 | err = file_remove_suid(file); |
1da177e4 LT |
2535 | if (err) |
2536 | goto out; | |
2537 | ||
870f4817 | 2538 | file_update_time(file); |
1da177e4 LT |
2539 | |
2540 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
2541 | if (unlikely(file->f_flags & O_DIRECT)) { | |
fb5527e6 JM |
2542 | loff_t endbyte; |
2543 | ssize_t written_buffered; | |
2544 | ||
2545 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, | |
2546 | ppos, count, ocount); | |
1da177e4 LT |
2547 | if (written < 0 || written == count) |
2548 | goto out; | |
2549 | /* | |
2550 | * direct-io write to a hole: fall through to buffered I/O | |
2551 | * for completing the rest of the request. | |
2552 | */ | |
2553 | pos += written; | |
2554 | count -= written; | |
fb5527e6 JM |
2555 | written_buffered = generic_file_buffered_write(iocb, iov, |
2556 | nr_segs, pos, ppos, count, | |
2557 | written); | |
2558 | /* | |
2559 | * If generic_file_buffered_write() retuned a synchronous error | |
2560 | * then we want to return the number of bytes which were | |
2561 | * direct-written, or the error code if that was zero. Note | |
2562 | * that this differs from normal direct-io semantics, which | |
2563 | * will return -EFOO even if some bytes were written. | |
2564 | */ | |
2565 | if (written_buffered < 0) { | |
2566 | err = written_buffered; | |
2567 | goto out; | |
2568 | } | |
1da177e4 | 2569 | |
fb5527e6 JM |
2570 | /* |
2571 | * We need to ensure that the page cache pages are written to | |
2572 | * disk and invalidated to preserve the expected O_DIRECT | |
2573 | * semantics. | |
2574 | */ | |
2575 | endbyte = pos + written_buffered - written - 1; | |
c05c4edd | 2576 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); |
fb5527e6 JM |
2577 | if (err == 0) { |
2578 | written = written_buffered; | |
2579 | invalidate_mapping_pages(mapping, | |
2580 | pos >> PAGE_CACHE_SHIFT, | |
2581 | endbyte >> PAGE_CACHE_SHIFT); | |
2582 | } else { | |
2583 | /* | |
2584 | * We don't know how much we wrote, so just return | |
2585 | * the number of bytes which were direct-written | |
2586 | */ | |
2587 | } | |
2588 | } else { | |
2589 | written = generic_file_buffered_write(iocb, iov, nr_segs, | |
2590 | pos, ppos, count, written); | |
2591 | } | |
1da177e4 LT |
2592 | out: |
2593 | current->backing_dev_info = NULL; | |
2594 | return written ? written : err; | |
2595 | } | |
e4dd9de3 JK |
2596 | EXPORT_SYMBOL(__generic_file_aio_write); |
2597 | ||
e4dd9de3 JK |
2598 | /** |
2599 | * generic_file_aio_write - write data to a file | |
2600 | * @iocb: IO state structure | |
2601 | * @iov: vector with data to write | |
2602 | * @nr_segs: number of segments in the vector | |
2603 | * @pos: position in file where to write | |
2604 | * | |
2605 | * This is a wrapper around __generic_file_aio_write() to be used by most | |
2606 | * filesystems. It takes care of syncing the file in case of O_SYNC file | |
2607 | * and acquires i_mutex as needed. | |
2608 | */ | |
027445c3 BP |
2609 | ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
2610 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
2611 | { |
2612 | struct file *file = iocb->ki_filp; | |
148f948b | 2613 | struct inode *inode = file->f_mapping->host; |
55602dd6 | 2614 | struct blk_plug plug; |
1da177e4 | 2615 | ssize_t ret; |
1da177e4 LT |
2616 | |
2617 | BUG_ON(iocb->ki_pos != pos); | |
2618 | ||
1b1dcc1b | 2619 | mutex_lock(&inode->i_mutex); |
55602dd6 | 2620 | blk_start_plug(&plug); |
e4dd9de3 | 2621 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); |
1b1dcc1b | 2622 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 2623 | |
148f948b | 2624 | if (ret > 0 || ret == -EIOCBQUEUED) { |
1da177e4 LT |
2625 | ssize_t err; |
2626 | ||
148f948b | 2627 | err = generic_write_sync(file, pos, ret); |
c7b50db2 | 2628 | if (err < 0 && ret > 0) |
1da177e4 LT |
2629 | ret = err; |
2630 | } | |
55602dd6 | 2631 | blk_finish_plug(&plug); |
1da177e4 LT |
2632 | return ret; |
2633 | } | |
2634 | EXPORT_SYMBOL(generic_file_aio_write); | |
2635 | ||
cf9a2ae8 DH |
2636 | /** |
2637 | * try_to_release_page() - release old fs-specific metadata on a page | |
2638 | * | |
2639 | * @page: the page which the kernel is trying to free | |
2640 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2641 | * | |
2642 | * The address_space is to try to release any data against the page | |
2643 | * (presumably at page->private). If the release was successful, return `1'. | |
2644 | * Otherwise return zero. | |
2645 | * | |
266cf658 DH |
2646 | * This may also be called if PG_fscache is set on a page, indicating that the |
2647 | * page is known to the local caching routines. | |
2648 | * | |
cf9a2ae8 | 2649 | * The @gfp_mask argument specifies whether I/O may be performed to release |
3f31fddf | 2650 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
cf9a2ae8 | 2651 | * |
cf9a2ae8 DH |
2652 | */ |
2653 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2654 | { | |
2655 | struct address_space * const mapping = page->mapping; | |
2656 | ||
2657 | BUG_ON(!PageLocked(page)); | |
2658 | if (PageWriteback(page)) | |
2659 | return 0; | |
2660 | ||
2661 | if (mapping && mapping->a_ops->releasepage) | |
2662 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2663 | return try_to_free_buffers(page); | |
2664 | } | |
2665 | ||
2666 | EXPORT_SYMBOL(try_to_release_page); |