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 | |
3167760f | 125 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 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()) { |
cc9a6c87 MG |
502 | unsigned int cpuset_mems_cookie; |
503 | do { | |
504 | cpuset_mems_cookie = get_mems_allowed(); | |
505 | n = cpuset_mem_spread_node(); | |
506 | page = alloc_pages_exact_node(n, gfp, 0); | |
507 | } while (!put_mems_allowed(cpuset_mems_cookie) && !page); | |
508 | ||
c0ff7453 | 509 | return page; |
44110fe3 | 510 | } |
2ae88149 | 511 | return alloc_pages(gfp, 0); |
44110fe3 | 512 | } |
2ae88149 | 513 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
514 | #endif |
515 | ||
1da177e4 LT |
516 | /* |
517 | * In order to wait for pages to become available there must be | |
518 | * waitqueues associated with pages. By using a hash table of | |
519 | * waitqueues where the bucket discipline is to maintain all | |
520 | * waiters on the same queue and wake all when any of the pages | |
521 | * become available, and for the woken contexts to check to be | |
522 | * sure the appropriate page became available, this saves space | |
523 | * at a cost of "thundering herd" phenomena during rare hash | |
524 | * collisions. | |
525 | */ | |
526 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
527 | { | |
528 | const struct zone *zone = page_zone(page); | |
529 | ||
530 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
531 | } | |
532 | ||
533 | static inline void wake_up_page(struct page *page, int bit) | |
534 | { | |
535 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); | |
536 | } | |
537 | ||
920c7a5d | 538 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
539 | { |
540 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
541 | ||
542 | if (test_bit(bit_nr, &page->flags)) | |
7eaceacc | 543 | __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
544 | TASK_UNINTERRUPTIBLE); |
545 | } | |
546 | EXPORT_SYMBOL(wait_on_page_bit); | |
547 | ||
f62e00cc KM |
548 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
549 | { | |
550 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
551 | ||
552 | if (!test_bit(bit_nr, &page->flags)) | |
553 | return 0; | |
554 | ||
555 | return __wait_on_bit(page_waitqueue(page), &wait, | |
556 | sleep_on_page_killable, TASK_KILLABLE); | |
557 | } | |
558 | ||
385e1ca5 DH |
559 | /** |
560 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
561 | * @page: Page defining the wait queue of interest |
562 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
563 | * |
564 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
565 | */ | |
566 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
567 | { | |
568 | wait_queue_head_t *q = page_waitqueue(page); | |
569 | unsigned long flags; | |
570 | ||
571 | spin_lock_irqsave(&q->lock, flags); | |
572 | __add_wait_queue(q, waiter); | |
573 | spin_unlock_irqrestore(&q->lock, flags); | |
574 | } | |
575 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
576 | ||
1da177e4 | 577 | /** |
485bb99b | 578 | * unlock_page - unlock a locked page |
1da177e4 LT |
579 | * @page: the page |
580 | * | |
581 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
582 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
583 | * mechananism between PageLocked pages and PageWriteback pages is shared. | |
584 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. | |
585 | * | |
8413ac9d NP |
586 | * The mb is necessary to enforce ordering between the clear_bit and the read |
587 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 588 | */ |
920c7a5d | 589 | void unlock_page(struct page *page) |
1da177e4 | 590 | { |
8413ac9d NP |
591 | VM_BUG_ON(!PageLocked(page)); |
592 | clear_bit_unlock(PG_locked, &page->flags); | |
593 | smp_mb__after_clear_bit(); | |
1da177e4 LT |
594 | wake_up_page(page, PG_locked); |
595 | } | |
596 | EXPORT_SYMBOL(unlock_page); | |
597 | ||
485bb99b RD |
598 | /** |
599 | * end_page_writeback - end writeback against a page | |
600 | * @page: the page | |
1da177e4 LT |
601 | */ |
602 | void end_page_writeback(struct page *page) | |
603 | { | |
ac6aadb2 MS |
604 | if (TestClearPageReclaim(page)) |
605 | rotate_reclaimable_page(page); | |
606 | ||
607 | if (!test_clear_page_writeback(page)) | |
608 | BUG(); | |
609 | ||
1da177e4 LT |
610 | smp_mb__after_clear_bit(); |
611 | wake_up_page(page, PG_writeback); | |
612 | } | |
613 | EXPORT_SYMBOL(end_page_writeback); | |
614 | ||
485bb99b RD |
615 | /** |
616 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
617 | * @page: the page to lock | |
1da177e4 | 618 | */ |
920c7a5d | 619 | void __lock_page(struct page *page) |
1da177e4 LT |
620 | { |
621 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
622 | ||
7eaceacc | 623 | __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
624 | TASK_UNINTERRUPTIBLE); |
625 | } | |
626 | EXPORT_SYMBOL(__lock_page); | |
627 | ||
b5606c2d | 628 | int __lock_page_killable(struct page *page) |
2687a356 MW |
629 | { |
630 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
631 | ||
632 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
7eaceacc | 633 | sleep_on_page_killable, TASK_KILLABLE); |
2687a356 | 634 | } |
18bc0bbd | 635 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 636 | |
d065bd81 ML |
637 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
638 | unsigned int flags) | |
639 | { | |
37b23e05 KM |
640 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
641 | /* | |
642 | * CAUTION! In this case, mmap_sem is not released | |
643 | * even though return 0. | |
644 | */ | |
645 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
646 | return 0; | |
647 | ||
648 | up_read(&mm->mmap_sem); | |
649 | if (flags & FAULT_FLAG_KILLABLE) | |
650 | wait_on_page_locked_killable(page); | |
651 | else | |
318b275f | 652 | wait_on_page_locked(page); |
d065bd81 | 653 | return 0; |
37b23e05 KM |
654 | } else { |
655 | if (flags & FAULT_FLAG_KILLABLE) { | |
656 | int ret; | |
657 | ||
658 | ret = __lock_page_killable(page); | |
659 | if (ret) { | |
660 | up_read(&mm->mmap_sem); | |
661 | return 0; | |
662 | } | |
663 | } else | |
664 | __lock_page(page); | |
665 | return 1; | |
d065bd81 ML |
666 | } |
667 | } | |
668 | ||
485bb99b RD |
669 | /** |
670 | * find_get_page - find and get a page reference | |
671 | * @mapping: the address_space to search | |
672 | * @offset: the page index | |
673 | * | |
da6052f7 NP |
674 | * Is there a pagecache struct page at the given (mapping, offset) tuple? |
675 | * If yes, increment its refcount and return it; if no, return NULL. | |
1da177e4 | 676 | */ |
a60637c8 | 677 | struct page *find_get_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 678 | { |
a60637c8 | 679 | void **pagep; |
1da177e4 LT |
680 | struct page *page; |
681 | ||
a60637c8 NP |
682 | rcu_read_lock(); |
683 | repeat: | |
684 | page = NULL; | |
685 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
686 | if (pagep) { | |
687 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
688 | if (unlikely(!page)) |
689 | goto out; | |
a2c16d6c | 690 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
691 | if (radix_tree_deref_retry(page)) |
692 | goto repeat; | |
693 | /* | |
694 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
695 | * here as an exceptional entry: so return it without | |
696 | * attempting to raise page count. | |
697 | */ | |
698 | goto out; | |
a2c16d6c | 699 | } |
a60637c8 NP |
700 | if (!page_cache_get_speculative(page)) |
701 | goto repeat; | |
702 | ||
703 | /* | |
704 | * Has the page moved? | |
705 | * This is part of the lockless pagecache protocol. See | |
706 | * include/linux/pagemap.h for details. | |
707 | */ | |
708 | if (unlikely(page != *pagep)) { | |
709 | page_cache_release(page); | |
710 | goto repeat; | |
711 | } | |
712 | } | |
27d20fdd | 713 | out: |
a60637c8 NP |
714 | rcu_read_unlock(); |
715 | ||
1da177e4 LT |
716 | return page; |
717 | } | |
1da177e4 LT |
718 | EXPORT_SYMBOL(find_get_page); |
719 | ||
1da177e4 LT |
720 | /** |
721 | * find_lock_page - locate, pin and lock a pagecache page | |
67be2dd1 MW |
722 | * @mapping: the address_space to search |
723 | * @offset: the page index | |
1da177e4 LT |
724 | * |
725 | * Locates the desired pagecache page, locks it, increments its reference | |
726 | * count and returns its address. | |
727 | * | |
728 | * Returns zero if the page was not present. find_lock_page() may sleep. | |
729 | */ | |
a60637c8 | 730 | struct page *find_lock_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 LT |
731 | { |
732 | struct page *page; | |
733 | ||
1da177e4 | 734 | repeat: |
a60637c8 | 735 | page = find_get_page(mapping, offset); |
a2c16d6c | 736 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
737 | lock_page(page); |
738 | /* Has the page been truncated? */ | |
739 | if (unlikely(page->mapping != mapping)) { | |
740 | unlock_page(page); | |
741 | page_cache_release(page); | |
742 | goto repeat; | |
1da177e4 | 743 | } |
a60637c8 | 744 | VM_BUG_ON(page->index != offset); |
1da177e4 | 745 | } |
1da177e4 LT |
746 | return page; |
747 | } | |
1da177e4 LT |
748 | EXPORT_SYMBOL(find_lock_page); |
749 | ||
750 | /** | |
751 | * find_or_create_page - locate or add a pagecache page | |
67be2dd1 MW |
752 | * @mapping: the page's address_space |
753 | * @index: the page's index into the mapping | |
754 | * @gfp_mask: page allocation mode | |
1da177e4 LT |
755 | * |
756 | * Locates a page in the pagecache. If the page is not present, a new page | |
757 | * is allocated using @gfp_mask and is added to the pagecache and to the VM's | |
758 | * LRU list. The returned page is locked and has its reference count | |
759 | * incremented. | |
760 | * | |
761 | * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic | |
762 | * allocation! | |
763 | * | |
764 | * find_or_create_page() returns the desired page's address, or zero on | |
765 | * memory exhaustion. | |
766 | */ | |
767 | struct page *find_or_create_page(struct address_space *mapping, | |
57f6b96c | 768 | pgoff_t index, gfp_t gfp_mask) |
1da177e4 | 769 | { |
eb2be189 | 770 | struct page *page; |
1da177e4 LT |
771 | int err; |
772 | repeat: | |
773 | page = find_lock_page(mapping, index); | |
774 | if (!page) { | |
eb2be189 NP |
775 | page = __page_cache_alloc(gfp_mask); |
776 | if (!page) | |
777 | return NULL; | |
67d58ac4 NP |
778 | /* |
779 | * We want a regular kernel memory (not highmem or DMA etc) | |
780 | * allocation for the radix tree nodes, but we need to honour | |
781 | * the context-specific requirements the caller has asked for. | |
782 | * GFP_RECLAIM_MASK collects those requirements. | |
783 | */ | |
784 | err = add_to_page_cache_lru(page, mapping, index, | |
785 | (gfp_mask & GFP_RECLAIM_MASK)); | |
eb2be189 NP |
786 | if (unlikely(err)) { |
787 | page_cache_release(page); | |
788 | page = NULL; | |
789 | if (err == -EEXIST) | |
790 | goto repeat; | |
1da177e4 | 791 | } |
1da177e4 | 792 | } |
1da177e4 LT |
793 | return page; |
794 | } | |
1da177e4 LT |
795 | EXPORT_SYMBOL(find_or_create_page); |
796 | ||
797 | /** | |
798 | * find_get_pages - gang pagecache lookup | |
799 | * @mapping: The address_space to search | |
800 | * @start: The starting page index | |
801 | * @nr_pages: The maximum number of pages | |
802 | * @pages: Where the resulting pages are placed | |
803 | * | |
804 | * find_get_pages() will search for and return a group of up to | |
805 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
806 | * find_get_pages() takes a reference against the returned pages. | |
807 | * | |
808 | * The search returns a group of mapping-contiguous pages with ascending | |
809 | * indexes. There may be holes in the indices due to not-present pages. | |
810 | * | |
811 | * find_get_pages() returns the number of pages which were found. | |
812 | */ | |
813 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
814 | unsigned int nr_pages, struct page **pages) | |
815 | { | |
0fc9d104 KK |
816 | struct radix_tree_iter iter; |
817 | void **slot; | |
818 | unsigned ret = 0; | |
819 | ||
820 | if (unlikely(!nr_pages)) | |
821 | return 0; | |
a60637c8 NP |
822 | |
823 | rcu_read_lock(); | |
824 | restart: | |
0fc9d104 | 825 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
a60637c8 NP |
826 | struct page *page; |
827 | repeat: | |
0fc9d104 | 828 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
829 | if (unlikely(!page)) |
830 | continue; | |
9d8aa4ea | 831 | |
a2c16d6c | 832 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
833 | if (radix_tree_deref_retry(page)) { |
834 | /* | |
835 | * Transient condition which can only trigger | |
836 | * when entry at index 0 moves out of or back | |
837 | * to root: none yet gotten, safe to restart. | |
838 | */ | |
0fc9d104 | 839 | WARN_ON(iter.index); |
8079b1c8 HD |
840 | goto restart; |
841 | } | |
a2c16d6c | 842 | /* |
8079b1c8 HD |
843 | * Otherwise, shmem/tmpfs must be storing a swap entry |
844 | * here as an exceptional entry: so skip over it - | |
845 | * we only reach this from invalidate_mapping_pages(). | |
a2c16d6c | 846 | */ |
8079b1c8 | 847 | continue; |
27d20fdd | 848 | } |
a60637c8 NP |
849 | |
850 | if (!page_cache_get_speculative(page)) | |
851 | goto repeat; | |
852 | ||
853 | /* Has the page moved? */ | |
0fc9d104 | 854 | if (unlikely(page != *slot)) { |
a60637c8 NP |
855 | page_cache_release(page); |
856 | goto repeat; | |
857 | } | |
1da177e4 | 858 | |
a60637c8 | 859 | pages[ret] = page; |
0fc9d104 KK |
860 | if (++ret == nr_pages) |
861 | break; | |
a60637c8 | 862 | } |
5b280c0c | 863 | |
a60637c8 | 864 | rcu_read_unlock(); |
1da177e4 LT |
865 | return ret; |
866 | } | |
867 | ||
ebf43500 JA |
868 | /** |
869 | * find_get_pages_contig - gang contiguous pagecache lookup | |
870 | * @mapping: The address_space to search | |
871 | * @index: The starting page index | |
872 | * @nr_pages: The maximum number of pages | |
873 | * @pages: Where the resulting pages are placed | |
874 | * | |
875 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
876 | * that the returned number of pages are guaranteed to be contiguous. | |
877 | * | |
878 | * find_get_pages_contig() returns the number of pages which were found. | |
879 | */ | |
880 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
881 | unsigned int nr_pages, struct page **pages) | |
882 | { | |
0fc9d104 KK |
883 | struct radix_tree_iter iter; |
884 | void **slot; | |
885 | unsigned int ret = 0; | |
886 | ||
887 | if (unlikely(!nr_pages)) | |
888 | return 0; | |
a60637c8 NP |
889 | |
890 | rcu_read_lock(); | |
891 | restart: | |
0fc9d104 | 892 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
a60637c8 NP |
893 | struct page *page; |
894 | repeat: | |
0fc9d104 KK |
895 | page = radix_tree_deref_slot(slot); |
896 | /* The hole, there no reason to continue */ | |
a60637c8 | 897 | if (unlikely(!page)) |
0fc9d104 | 898 | break; |
9d8aa4ea | 899 | |
a2c16d6c | 900 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
901 | if (radix_tree_deref_retry(page)) { |
902 | /* | |
903 | * Transient condition which can only trigger | |
904 | * when entry at index 0 moves out of or back | |
905 | * to root: none yet gotten, safe to restart. | |
906 | */ | |
907 | goto restart; | |
908 | } | |
a2c16d6c | 909 | /* |
8079b1c8 HD |
910 | * Otherwise, shmem/tmpfs must be storing a swap entry |
911 | * here as an exceptional entry: so stop looking for | |
912 | * contiguous pages. | |
a2c16d6c | 913 | */ |
8079b1c8 | 914 | break; |
a2c16d6c | 915 | } |
ebf43500 | 916 | |
a60637c8 NP |
917 | if (!page_cache_get_speculative(page)) |
918 | goto repeat; | |
919 | ||
920 | /* Has the page moved? */ | |
0fc9d104 | 921 | if (unlikely(page != *slot)) { |
a60637c8 NP |
922 | page_cache_release(page); |
923 | goto repeat; | |
924 | } | |
925 | ||
9cbb4cb2 NP |
926 | /* |
927 | * must check mapping and index after taking the ref. | |
928 | * otherwise we can get both false positives and false | |
929 | * negatives, which is just confusing to the caller. | |
930 | */ | |
0fc9d104 | 931 | if (page->mapping == NULL || page->index != iter.index) { |
9cbb4cb2 NP |
932 | page_cache_release(page); |
933 | break; | |
934 | } | |
935 | ||
a60637c8 | 936 | pages[ret] = page; |
0fc9d104 KK |
937 | if (++ret == nr_pages) |
938 | break; | |
ebf43500 | 939 | } |
a60637c8 NP |
940 | rcu_read_unlock(); |
941 | return ret; | |
ebf43500 | 942 | } |
ef71c15c | 943 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 944 | |
485bb99b RD |
945 | /** |
946 | * find_get_pages_tag - find and return pages that match @tag | |
947 | * @mapping: the address_space to search | |
948 | * @index: the starting page index | |
949 | * @tag: the tag index | |
950 | * @nr_pages: the maximum number of pages | |
951 | * @pages: where the resulting pages are placed | |
952 | * | |
1da177e4 | 953 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 954 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
955 | */ |
956 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
957 | int tag, unsigned int nr_pages, struct page **pages) | |
958 | { | |
0fc9d104 KK |
959 | struct radix_tree_iter iter; |
960 | void **slot; | |
961 | unsigned ret = 0; | |
962 | ||
963 | if (unlikely(!nr_pages)) | |
964 | return 0; | |
a60637c8 NP |
965 | |
966 | rcu_read_lock(); | |
967 | restart: | |
0fc9d104 KK |
968 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
969 | &iter, *index, tag) { | |
a60637c8 NP |
970 | struct page *page; |
971 | repeat: | |
0fc9d104 | 972 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
973 | if (unlikely(!page)) |
974 | continue; | |
9d8aa4ea | 975 | |
a2c16d6c | 976 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
977 | if (radix_tree_deref_retry(page)) { |
978 | /* | |
979 | * Transient condition which can only trigger | |
980 | * when entry at index 0 moves out of or back | |
981 | * to root: none yet gotten, safe to restart. | |
982 | */ | |
983 | goto restart; | |
984 | } | |
a2c16d6c | 985 | /* |
8079b1c8 HD |
986 | * This function is never used on a shmem/tmpfs |
987 | * mapping, so a swap entry won't be found here. | |
a2c16d6c | 988 | */ |
8079b1c8 | 989 | BUG(); |
a2c16d6c | 990 | } |
a60637c8 NP |
991 | |
992 | if (!page_cache_get_speculative(page)) | |
993 | goto repeat; | |
994 | ||
995 | /* Has the page moved? */ | |
0fc9d104 | 996 | if (unlikely(page != *slot)) { |
a60637c8 NP |
997 | page_cache_release(page); |
998 | goto repeat; | |
999 | } | |
1000 | ||
1001 | pages[ret] = page; | |
0fc9d104 KK |
1002 | if (++ret == nr_pages) |
1003 | break; | |
a60637c8 | 1004 | } |
5b280c0c | 1005 | |
a60637c8 | 1006 | rcu_read_unlock(); |
1da177e4 | 1007 | |
1da177e4 LT |
1008 | if (ret) |
1009 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1010 | |
1da177e4 LT |
1011 | return ret; |
1012 | } | |
ef71c15c | 1013 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1014 | |
485bb99b RD |
1015 | /** |
1016 | * grab_cache_page_nowait - returns locked page at given index in given cache | |
1017 | * @mapping: target address_space | |
1018 | * @index: the page index | |
1019 | * | |
72fd4a35 | 1020 | * Same as grab_cache_page(), but do not wait if the page is unavailable. |
1da177e4 LT |
1021 | * This is intended for speculative data generators, where the data can |
1022 | * be regenerated if the page couldn't be grabbed. This routine should | |
1023 | * be safe to call while holding the lock for another page. | |
1024 | * | |
1025 | * Clear __GFP_FS when allocating the page to avoid recursion into the fs | |
1026 | * and deadlock against the caller's locked page. | |
1027 | */ | |
1028 | struct page * | |
57f6b96c | 1029 | grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) |
1da177e4 LT |
1030 | { |
1031 | struct page *page = find_get_page(mapping, index); | |
1da177e4 LT |
1032 | |
1033 | if (page) { | |
529ae9aa | 1034 | if (trylock_page(page)) |
1da177e4 LT |
1035 | return page; |
1036 | page_cache_release(page); | |
1037 | return NULL; | |
1038 | } | |
2ae88149 | 1039 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); |
67d58ac4 | 1040 | if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) { |
1da177e4 LT |
1041 | page_cache_release(page); |
1042 | page = NULL; | |
1043 | } | |
1044 | return page; | |
1045 | } | |
1da177e4 LT |
1046 | EXPORT_SYMBOL(grab_cache_page_nowait); |
1047 | ||
76d42bd9 WF |
1048 | /* |
1049 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1050 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1051 | * | |
1052 | * ---R__________________________________________B__________ | |
1053 | * ^ reading here ^ bad block(assume 4k) | |
1054 | * | |
1055 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1056 | * => failing the whole request => read(R) => read(R+1) => | |
1057 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1058 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1059 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1060 | * | |
1061 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1062 | */ | |
1063 | static void shrink_readahead_size_eio(struct file *filp, | |
1064 | struct file_ra_state *ra) | |
1065 | { | |
76d42bd9 | 1066 | ra->ra_pages /= 4; |
76d42bd9 WF |
1067 | } |
1068 | ||
485bb99b | 1069 | /** |
36e78914 | 1070 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1071 | * @filp: the file to read |
1072 | * @ppos: current file position | |
1073 | * @desc: read_descriptor | |
1074 | * @actor: read method | |
1075 | * | |
1da177e4 | 1076 | * This is a generic file read routine, and uses the |
485bb99b | 1077 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1078 | * |
1079 | * This is really ugly. But the goto's actually try to clarify some | |
1080 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1081 | */ |
36e78914 CH |
1082 | static void do_generic_file_read(struct file *filp, loff_t *ppos, |
1083 | read_descriptor_t *desc, read_actor_t actor) | |
1da177e4 | 1084 | { |
36e78914 | 1085 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1086 | struct inode *inode = mapping->host; |
36e78914 | 1087 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1088 | pgoff_t index; |
1089 | pgoff_t last_index; | |
1090 | pgoff_t prev_index; | |
1091 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1092 | unsigned int prev_offset; |
1da177e4 | 1093 | int error; |
1da177e4 | 1094 | |
1da177e4 | 1095 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1096 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1097 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
1da177e4 LT |
1098 | last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1099 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1100 | ||
1da177e4 LT |
1101 | for (;;) { |
1102 | struct page *page; | |
57f6b96c | 1103 | pgoff_t end_index; |
a32ea1e1 | 1104 | loff_t isize; |
1da177e4 LT |
1105 | unsigned long nr, ret; |
1106 | ||
1da177e4 | 1107 | cond_resched(); |
1da177e4 LT |
1108 | find_page: |
1109 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1110 | if (!page) { |
cf914a7d | 1111 | page_cache_sync_readahead(mapping, |
7ff81078 | 1112 | ra, filp, |
3ea89ee8 FW |
1113 | index, last_index - index); |
1114 | page = find_get_page(mapping, index); | |
1115 | if (unlikely(page == NULL)) | |
1116 | goto no_cached_page; | |
1117 | } | |
1118 | if (PageReadahead(page)) { | |
cf914a7d | 1119 | page_cache_async_readahead(mapping, |
7ff81078 | 1120 | ra, filp, page, |
3ea89ee8 | 1121 | index, last_index - index); |
1da177e4 | 1122 | } |
8ab22b9a HH |
1123 | if (!PageUptodate(page)) { |
1124 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1125 | !mapping->a_ops->is_partially_uptodate) | |
1126 | goto page_not_up_to_date; | |
529ae9aa | 1127 | if (!trylock_page(page)) |
8ab22b9a | 1128 | goto page_not_up_to_date; |
8d056cb9 DH |
1129 | /* Did it get truncated before we got the lock? */ |
1130 | if (!page->mapping) | |
1131 | goto page_not_up_to_date_locked; | |
8ab22b9a HH |
1132 | if (!mapping->a_ops->is_partially_uptodate(page, |
1133 | desc, offset)) | |
1134 | goto page_not_up_to_date_locked; | |
1135 | unlock_page(page); | |
1136 | } | |
1da177e4 | 1137 | page_ok: |
a32ea1e1 N |
1138 | /* |
1139 | * i_size must be checked after we know the page is Uptodate. | |
1140 | * | |
1141 | * Checking i_size after the check allows us to calculate | |
1142 | * the correct value for "nr", which means the zero-filled | |
1143 | * part of the page is not copied back to userspace (unless | |
1144 | * another truncate extends the file - this is desired though). | |
1145 | */ | |
1146 | ||
1147 | isize = i_size_read(inode); | |
1148 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1149 | if (unlikely(!isize || index > end_index)) { | |
1150 | page_cache_release(page); | |
1151 | goto out; | |
1152 | } | |
1153 | ||
1154 | /* nr is the maximum number of bytes to copy from this page */ | |
1155 | nr = PAGE_CACHE_SIZE; | |
1156 | if (index == end_index) { | |
1157 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1158 | if (nr <= offset) { | |
1159 | page_cache_release(page); | |
1160 | goto out; | |
1161 | } | |
1162 | } | |
1163 | nr = nr - offset; | |
1da177e4 LT |
1164 | |
1165 | /* If users can be writing to this page using arbitrary | |
1166 | * virtual addresses, take care about potential aliasing | |
1167 | * before reading the page on the kernel side. | |
1168 | */ | |
1169 | if (mapping_writably_mapped(mapping)) | |
1170 | flush_dcache_page(page); | |
1171 | ||
1172 | /* | |
ec0f1637 JK |
1173 | * When a sequential read accesses a page several times, |
1174 | * only mark it as accessed the first time. | |
1da177e4 | 1175 | */ |
ec0f1637 | 1176 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1177 | mark_page_accessed(page); |
1178 | prev_index = index; | |
1179 | ||
1180 | /* | |
1181 | * Ok, we have the page, and it's up-to-date, so | |
1182 | * now we can copy it to user space... | |
1183 | * | |
1184 | * The actor routine returns how many bytes were actually used.. | |
1185 | * NOTE! This may not be the same as how much of a user buffer | |
1186 | * we filled up (we may be padding etc), so we can only update | |
1187 | * "pos" here (the actor routine has to update the user buffer | |
1188 | * pointers and the remaining count). | |
1189 | */ | |
1190 | ret = actor(desc, page, offset, nr); | |
1191 | offset += ret; | |
1192 | index += offset >> PAGE_CACHE_SHIFT; | |
1193 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1194 | prev_offset = offset; |
1da177e4 LT |
1195 | |
1196 | page_cache_release(page); | |
1197 | if (ret == nr && desc->count) | |
1198 | continue; | |
1199 | goto out; | |
1200 | ||
1201 | page_not_up_to_date: | |
1202 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1203 | error = lock_page_killable(page); |
1204 | if (unlikely(error)) | |
1205 | goto readpage_error; | |
1da177e4 | 1206 | |
8ab22b9a | 1207 | page_not_up_to_date_locked: |
da6052f7 | 1208 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1209 | if (!page->mapping) { |
1210 | unlock_page(page); | |
1211 | page_cache_release(page); | |
1212 | continue; | |
1213 | } | |
1214 | ||
1215 | /* Did somebody else fill it already? */ | |
1216 | if (PageUptodate(page)) { | |
1217 | unlock_page(page); | |
1218 | goto page_ok; | |
1219 | } | |
1220 | ||
1221 | readpage: | |
91803b49 JM |
1222 | /* |
1223 | * A previous I/O error may have been due to temporary | |
1224 | * failures, eg. multipath errors. | |
1225 | * PG_error will be set again if readpage fails. | |
1226 | */ | |
1227 | ClearPageError(page); | |
1da177e4 LT |
1228 | /* Start the actual read. The read will unlock the page. */ |
1229 | error = mapping->a_ops->readpage(filp, page); | |
1230 | ||
994fc28c ZB |
1231 | if (unlikely(error)) { |
1232 | if (error == AOP_TRUNCATED_PAGE) { | |
1233 | page_cache_release(page); | |
1234 | goto find_page; | |
1235 | } | |
1da177e4 | 1236 | goto readpage_error; |
994fc28c | 1237 | } |
1da177e4 LT |
1238 | |
1239 | if (!PageUptodate(page)) { | |
85462323 ON |
1240 | error = lock_page_killable(page); |
1241 | if (unlikely(error)) | |
1242 | goto readpage_error; | |
1da177e4 LT |
1243 | if (!PageUptodate(page)) { |
1244 | if (page->mapping == NULL) { | |
1245 | /* | |
2ecdc82e | 1246 | * invalidate_mapping_pages got it |
1da177e4 LT |
1247 | */ |
1248 | unlock_page(page); | |
1249 | page_cache_release(page); | |
1250 | goto find_page; | |
1251 | } | |
1252 | unlock_page(page); | |
7ff81078 | 1253 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1254 | error = -EIO; |
1255 | goto readpage_error; | |
1da177e4 LT |
1256 | } |
1257 | unlock_page(page); | |
1258 | } | |
1259 | ||
1da177e4 LT |
1260 | goto page_ok; |
1261 | ||
1262 | readpage_error: | |
1263 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1264 | desc->error = error; | |
1265 | page_cache_release(page); | |
1266 | goto out; | |
1267 | ||
1268 | no_cached_page: | |
1269 | /* | |
1270 | * Ok, it wasn't cached, so we need to create a new | |
1271 | * page.. | |
1272 | */ | |
eb2be189 NP |
1273 | page = page_cache_alloc_cold(mapping); |
1274 | if (!page) { | |
1275 | desc->error = -ENOMEM; | |
1276 | goto out; | |
1da177e4 | 1277 | } |
eb2be189 | 1278 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1279 | index, GFP_KERNEL); |
1280 | if (error) { | |
eb2be189 | 1281 | page_cache_release(page); |
1da177e4 LT |
1282 | if (error == -EEXIST) |
1283 | goto find_page; | |
1284 | desc->error = error; | |
1285 | goto out; | |
1286 | } | |
1da177e4 LT |
1287 | goto readpage; |
1288 | } | |
1289 | ||
1290 | out: | |
7ff81078 FW |
1291 | ra->prev_pos = prev_index; |
1292 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1293 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1294 | |
f4e6b498 | 1295 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1296 | file_accessed(filp); |
1da177e4 | 1297 | } |
1da177e4 LT |
1298 | |
1299 | int file_read_actor(read_descriptor_t *desc, struct page *page, | |
1300 | unsigned long offset, unsigned long size) | |
1301 | { | |
1302 | char *kaddr; | |
1303 | unsigned long left, count = desc->count; | |
1304 | ||
1305 | if (size > count) | |
1306 | size = count; | |
1307 | ||
1308 | /* | |
1309 | * Faults on the destination of a read are common, so do it before | |
1310 | * taking the kmap. | |
1311 | */ | |
1312 | if (!fault_in_pages_writeable(desc->arg.buf, size)) { | |
9b04c5fe | 1313 | kaddr = kmap_atomic(page); |
1da177e4 LT |
1314 | left = __copy_to_user_inatomic(desc->arg.buf, |
1315 | kaddr + offset, size); | |
9b04c5fe | 1316 | kunmap_atomic(kaddr); |
1da177e4 LT |
1317 | if (left == 0) |
1318 | goto success; | |
1319 | } | |
1320 | ||
1321 | /* Do it the slow way */ | |
1322 | kaddr = kmap(page); | |
1323 | left = __copy_to_user(desc->arg.buf, kaddr + offset, size); | |
1324 | kunmap(page); | |
1325 | ||
1326 | if (left) { | |
1327 | size -= left; | |
1328 | desc->error = -EFAULT; | |
1329 | } | |
1330 | success: | |
1331 | desc->count = count - size; | |
1332 | desc->written += size; | |
1333 | desc->arg.buf += size; | |
1334 | return size; | |
1335 | } | |
1336 | ||
0ceb3314 DM |
1337 | /* |
1338 | * Performs necessary checks before doing a write | |
1339 | * @iov: io vector request | |
1340 | * @nr_segs: number of segments in the iovec | |
1341 | * @count: number of bytes to write | |
1342 | * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE | |
1343 | * | |
1344 | * Adjust number of segments and amount of bytes to write (nr_segs should be | |
1345 | * properly initialized first). Returns appropriate error code that caller | |
1346 | * should return or zero in case that write should be allowed. | |
1347 | */ | |
1348 | int generic_segment_checks(const struct iovec *iov, | |
1349 | unsigned long *nr_segs, size_t *count, int access_flags) | |
1350 | { | |
1351 | unsigned long seg; | |
1352 | size_t cnt = 0; | |
1353 | for (seg = 0; seg < *nr_segs; seg++) { | |
1354 | const struct iovec *iv = &iov[seg]; | |
1355 | ||
1356 | /* | |
1357 | * If any segment has a negative length, or the cumulative | |
1358 | * length ever wraps negative then return -EINVAL. | |
1359 | */ | |
1360 | cnt += iv->iov_len; | |
1361 | if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) | |
1362 | return -EINVAL; | |
1363 | if (access_ok(access_flags, iv->iov_base, iv->iov_len)) | |
1364 | continue; | |
1365 | if (seg == 0) | |
1366 | return -EFAULT; | |
1367 | *nr_segs = seg; | |
1368 | cnt -= iv->iov_len; /* This segment is no good */ | |
1369 | break; | |
1370 | } | |
1371 | *count = cnt; | |
1372 | return 0; | |
1373 | } | |
1374 | EXPORT_SYMBOL(generic_segment_checks); | |
1375 | ||
485bb99b | 1376 | /** |
b2abacf3 | 1377 | * generic_file_aio_read - generic filesystem read routine |
485bb99b RD |
1378 | * @iocb: kernel I/O control block |
1379 | * @iov: io vector request | |
1380 | * @nr_segs: number of segments in the iovec | |
b2abacf3 | 1381 | * @pos: current file position |
485bb99b | 1382 | * |
1da177e4 LT |
1383 | * This is the "read()" routine for all filesystems |
1384 | * that can use the page cache directly. | |
1385 | */ | |
1386 | ssize_t | |
543ade1f BP |
1387 | generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
1388 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
1389 | { |
1390 | struct file *filp = iocb->ki_filp; | |
1391 | ssize_t retval; | |
66f998f6 | 1392 | unsigned long seg = 0; |
1da177e4 | 1393 | size_t count; |
543ade1f | 1394 | loff_t *ppos = &iocb->ki_pos; |
1da177e4 LT |
1395 | |
1396 | count = 0; | |
0ceb3314 DM |
1397 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); |
1398 | if (retval) | |
1399 | return retval; | |
1da177e4 LT |
1400 | |
1401 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
1402 | if (filp->f_flags & O_DIRECT) { | |
543ade1f | 1403 | loff_t size; |
1da177e4 LT |
1404 | struct address_space *mapping; |
1405 | struct inode *inode; | |
1406 | ||
1407 | mapping = filp->f_mapping; | |
1408 | inode = mapping->host; | |
1da177e4 LT |
1409 | if (!count) |
1410 | goto out; /* skip atime */ | |
1411 | size = i_size_read(inode); | |
1412 | if (pos < size) { | |
48b47c56 NP |
1413 | retval = filemap_write_and_wait_range(mapping, pos, |
1414 | pos + iov_length(iov, nr_segs) - 1); | |
a969e903 | 1415 | if (!retval) { |
3deaa719 SL |
1416 | struct blk_plug plug; |
1417 | ||
1418 | blk_start_plug(&plug); | |
a969e903 CH |
1419 | retval = mapping->a_ops->direct_IO(READ, iocb, |
1420 | iov, pos, nr_segs); | |
3deaa719 | 1421 | blk_finish_plug(&plug); |
a969e903 | 1422 | } |
66f998f6 | 1423 | if (retval > 0) { |
1da177e4 | 1424 | *ppos = pos + retval; |
66f998f6 JB |
1425 | count -= retval; |
1426 | } | |
1427 | ||
1428 | /* | |
1429 | * Btrfs can have a short DIO read if we encounter | |
1430 | * compressed extents, so if there was an error, or if | |
1431 | * we've already read everything we wanted to, or if | |
1432 | * there was a short read because we hit EOF, go ahead | |
1433 | * and return. Otherwise fallthrough to buffered io for | |
1434 | * the rest of the read. | |
1435 | */ | |
1436 | if (retval < 0 || !count || *ppos >= size) { | |
11fa977e HD |
1437 | file_accessed(filp); |
1438 | goto out; | |
1439 | } | |
0e0bcae3 | 1440 | } |
1da177e4 LT |
1441 | } |
1442 | ||
66f998f6 | 1443 | count = retval; |
11fa977e HD |
1444 | for (seg = 0; seg < nr_segs; seg++) { |
1445 | read_descriptor_t desc; | |
66f998f6 JB |
1446 | loff_t offset = 0; |
1447 | ||
1448 | /* | |
1449 | * If we did a short DIO read we need to skip the section of the | |
1450 | * iov that we've already read data into. | |
1451 | */ | |
1452 | if (count) { | |
1453 | if (count > iov[seg].iov_len) { | |
1454 | count -= iov[seg].iov_len; | |
1455 | continue; | |
1456 | } | |
1457 | offset = count; | |
1458 | count = 0; | |
1459 | } | |
1da177e4 | 1460 | |
11fa977e | 1461 | desc.written = 0; |
66f998f6 JB |
1462 | desc.arg.buf = iov[seg].iov_base + offset; |
1463 | desc.count = iov[seg].iov_len - offset; | |
11fa977e HD |
1464 | if (desc.count == 0) |
1465 | continue; | |
1466 | desc.error = 0; | |
1467 | do_generic_file_read(filp, ppos, &desc, file_read_actor); | |
1468 | retval += desc.written; | |
1469 | if (desc.error) { | |
1470 | retval = retval ?: desc.error; | |
1471 | break; | |
1da177e4 | 1472 | } |
11fa977e HD |
1473 | if (desc.count > 0) |
1474 | break; | |
1da177e4 LT |
1475 | } |
1476 | out: | |
1477 | return retval; | |
1478 | } | |
1da177e4 LT |
1479 | EXPORT_SYMBOL(generic_file_aio_read); |
1480 | ||
1da177e4 LT |
1481 | static ssize_t |
1482 | do_readahead(struct address_space *mapping, struct file *filp, | |
57f6b96c | 1483 | pgoff_t index, unsigned long nr) |
1da177e4 LT |
1484 | { |
1485 | if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage) | |
1486 | return -EINVAL; | |
1487 | ||
f7e839dd | 1488 | force_page_cache_readahead(mapping, filp, index, nr); |
1da177e4 LT |
1489 | return 0; |
1490 | } | |
1491 | ||
6673e0c3 | 1492 | SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count) |
1da177e4 LT |
1493 | { |
1494 | ssize_t ret; | |
1495 | struct file *file; | |
1496 | ||
1497 | ret = -EBADF; | |
1498 | file = fget(fd); | |
1499 | if (file) { | |
1500 | if (file->f_mode & FMODE_READ) { | |
1501 | struct address_space *mapping = file->f_mapping; | |
57f6b96c FW |
1502 | pgoff_t start = offset >> PAGE_CACHE_SHIFT; |
1503 | pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
1504 | unsigned long len = end - start + 1; |
1505 | ret = do_readahead(mapping, file, start, len); | |
1506 | } | |
1507 | fput(file); | |
1508 | } | |
1509 | return ret; | |
1510 | } | |
6673e0c3 HC |
1511 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
1512 | asmlinkage long SyS_readahead(long fd, loff_t offset, long count) | |
1513 | { | |
1514 | return SYSC_readahead((int) fd, offset, (size_t) count); | |
1515 | } | |
1516 | SYSCALL_ALIAS(sys_readahead, SyS_readahead); | |
1517 | #endif | |
1da177e4 LT |
1518 | |
1519 | #ifdef CONFIG_MMU | |
485bb99b RD |
1520 | /** |
1521 | * page_cache_read - adds requested page to the page cache if not already there | |
1522 | * @file: file to read | |
1523 | * @offset: page index | |
1524 | * | |
1da177e4 LT |
1525 | * This adds the requested page to the page cache if it isn't already there, |
1526 | * and schedules an I/O to read in its contents from disk. | |
1527 | */ | |
920c7a5d | 1528 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1529 | { |
1530 | struct address_space *mapping = file->f_mapping; | |
1531 | struct page *page; | |
994fc28c | 1532 | int ret; |
1da177e4 | 1533 | |
994fc28c ZB |
1534 | do { |
1535 | page = page_cache_alloc_cold(mapping); | |
1536 | if (!page) | |
1537 | return -ENOMEM; | |
1538 | ||
1539 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); | |
1540 | if (ret == 0) | |
1541 | ret = mapping->a_ops->readpage(file, page); | |
1542 | else if (ret == -EEXIST) | |
1543 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1544 | |
1da177e4 | 1545 | page_cache_release(page); |
1da177e4 | 1546 | |
994fc28c ZB |
1547 | } while (ret == AOP_TRUNCATED_PAGE); |
1548 | ||
1549 | return ret; | |
1da177e4 LT |
1550 | } |
1551 | ||
1552 | #define MMAP_LOTSAMISS (100) | |
1553 | ||
ef00e08e LT |
1554 | /* |
1555 | * Synchronous readahead happens when we don't even find | |
1556 | * a page in the page cache at all. | |
1557 | */ | |
1558 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1559 | struct file_ra_state *ra, | |
1560 | struct file *file, | |
1561 | pgoff_t offset) | |
1562 | { | |
1563 | unsigned long ra_pages; | |
1564 | struct address_space *mapping = file->f_mapping; | |
1565 | ||
1566 | /* If we don't want any read-ahead, don't bother */ | |
1567 | if (VM_RandomReadHint(vma)) | |
1568 | return; | |
275b12bf WF |
1569 | if (!ra->ra_pages) |
1570 | return; | |
ef00e08e | 1571 | |
2cbea1d3 | 1572 | if (VM_SequentialReadHint(vma)) { |
7ffc59b4 WF |
1573 | page_cache_sync_readahead(mapping, ra, file, offset, |
1574 | ra->ra_pages); | |
ef00e08e LT |
1575 | return; |
1576 | } | |
1577 | ||
207d04ba AK |
1578 | /* Avoid banging the cache line if not needed */ |
1579 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1580 | ra->mmap_miss++; |
1581 | ||
1582 | /* | |
1583 | * Do we miss much more than hit in this file? If so, | |
1584 | * stop bothering with read-ahead. It will only hurt. | |
1585 | */ | |
1586 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1587 | return; | |
1588 | ||
d30a1100 WF |
1589 | /* |
1590 | * mmap read-around | |
1591 | */ | |
ef00e08e | 1592 | ra_pages = max_sane_readahead(ra->ra_pages); |
275b12bf WF |
1593 | ra->start = max_t(long, 0, offset - ra_pages / 2); |
1594 | ra->size = ra_pages; | |
2cbea1d3 | 1595 | ra->async_size = ra_pages / 4; |
275b12bf | 1596 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1597 | } |
1598 | ||
1599 | /* | |
1600 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1601 | * so we want to possibly extend the readahead further.. | |
1602 | */ | |
1603 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1604 | struct file_ra_state *ra, | |
1605 | struct file *file, | |
1606 | struct page *page, | |
1607 | pgoff_t offset) | |
1608 | { | |
1609 | struct address_space *mapping = file->f_mapping; | |
1610 | ||
1611 | /* If we don't want any read-ahead, don't bother */ | |
1612 | if (VM_RandomReadHint(vma)) | |
1613 | return; | |
1614 | if (ra->mmap_miss > 0) | |
1615 | ra->mmap_miss--; | |
1616 | if (PageReadahead(page)) | |
2fad6f5d WF |
1617 | page_cache_async_readahead(mapping, ra, file, |
1618 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1619 | } |
1620 | ||
485bb99b | 1621 | /** |
54cb8821 | 1622 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1623 | * @vma: vma in which the fault was taken |
1624 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1625 | * |
54cb8821 | 1626 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1627 | * mapped memory region to read in file data during a page fault. |
1628 | * | |
1629 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1630 | * it in the page cache, and handles the special cases reasonably without | |
1631 | * having a lot of duplicated code. | |
1632 | */ | |
d0217ac0 | 1633 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1634 | { |
1635 | int error; | |
54cb8821 | 1636 | struct file *file = vma->vm_file; |
1da177e4 LT |
1637 | struct address_space *mapping = file->f_mapping; |
1638 | struct file_ra_state *ra = &file->f_ra; | |
1639 | struct inode *inode = mapping->host; | |
ef00e08e | 1640 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 1641 | struct page *page; |
2004dc8e | 1642 | pgoff_t size; |
83c54070 | 1643 | int ret = 0; |
1da177e4 | 1644 | |
1da177e4 | 1645 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1646 | if (offset >= size) |
5307cc1a | 1647 | return VM_FAULT_SIGBUS; |
1da177e4 | 1648 | |
1da177e4 LT |
1649 | /* |
1650 | * Do we have something in the page cache already? | |
1651 | */ | |
ef00e08e LT |
1652 | page = find_get_page(mapping, offset); |
1653 | if (likely(page)) { | |
1da177e4 | 1654 | /* |
ef00e08e LT |
1655 | * We found the page, so try async readahead before |
1656 | * waiting for the lock. | |
1da177e4 | 1657 | */ |
ef00e08e | 1658 | do_async_mmap_readahead(vma, ra, file, page, offset); |
ef00e08e LT |
1659 | } else { |
1660 | /* No page in the page cache at all */ | |
1661 | do_sync_mmap_readahead(vma, ra, file, offset); | |
1662 | count_vm_event(PGMAJFAULT); | |
456f998e | 1663 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
1664 | ret = VM_FAULT_MAJOR; |
1665 | retry_find: | |
b522c94d | 1666 | page = find_get_page(mapping, offset); |
1da177e4 LT |
1667 | if (!page) |
1668 | goto no_cached_page; | |
1669 | } | |
1670 | ||
d88c0922 ML |
1671 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
1672 | page_cache_release(page); | |
d065bd81 | 1673 | return ret | VM_FAULT_RETRY; |
d88c0922 | 1674 | } |
b522c94d ML |
1675 | |
1676 | /* Did it get truncated? */ | |
1677 | if (unlikely(page->mapping != mapping)) { | |
1678 | unlock_page(page); | |
1679 | put_page(page); | |
1680 | goto retry_find; | |
1681 | } | |
1682 | VM_BUG_ON(page->index != offset); | |
1683 | ||
1da177e4 | 1684 | /* |
d00806b1 NP |
1685 | * We have a locked page in the page cache, now we need to check |
1686 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1687 | */ |
d00806b1 | 1688 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1689 | goto page_not_uptodate; |
1690 | ||
ef00e08e LT |
1691 | /* |
1692 | * Found the page and have a reference on it. | |
1693 | * We must recheck i_size under page lock. | |
1694 | */ | |
d00806b1 | 1695 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1696 | if (unlikely(offset >= size)) { |
d00806b1 | 1697 | unlock_page(page); |
745ad48e | 1698 | page_cache_release(page); |
5307cc1a | 1699 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1700 | } |
1701 | ||
d0217ac0 | 1702 | vmf->page = page; |
83c54070 | 1703 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 1704 | |
1da177e4 LT |
1705 | no_cached_page: |
1706 | /* | |
1707 | * We're only likely to ever get here if MADV_RANDOM is in | |
1708 | * effect. | |
1709 | */ | |
ef00e08e | 1710 | error = page_cache_read(file, offset); |
1da177e4 LT |
1711 | |
1712 | /* | |
1713 | * The page we want has now been added to the page cache. | |
1714 | * In the unlikely event that someone removed it in the | |
1715 | * meantime, we'll just come back here and read it again. | |
1716 | */ | |
1717 | if (error >= 0) | |
1718 | goto retry_find; | |
1719 | ||
1720 | /* | |
1721 | * An error return from page_cache_read can result if the | |
1722 | * system is low on memory, or a problem occurs while trying | |
1723 | * to schedule I/O. | |
1724 | */ | |
1725 | if (error == -ENOMEM) | |
d0217ac0 NP |
1726 | return VM_FAULT_OOM; |
1727 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1728 | |
1729 | page_not_uptodate: | |
1da177e4 LT |
1730 | /* |
1731 | * Umm, take care of errors if the page isn't up-to-date. | |
1732 | * Try to re-read it _once_. We do this synchronously, | |
1733 | * because there really aren't any performance issues here | |
1734 | * and we need to check for errors. | |
1735 | */ | |
1da177e4 | 1736 | ClearPageError(page); |
994fc28c | 1737 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
1738 | if (!error) { |
1739 | wait_on_page_locked(page); | |
1740 | if (!PageUptodate(page)) | |
1741 | error = -EIO; | |
1742 | } | |
d00806b1 NP |
1743 | page_cache_release(page); |
1744 | ||
1745 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 1746 | goto retry_find; |
1da177e4 | 1747 | |
d00806b1 | 1748 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 1749 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 1750 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
1751 | } |
1752 | EXPORT_SYMBOL(filemap_fault); | |
1753 | ||
f0f37e2f | 1754 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 1755 | .fault = filemap_fault, |
1da177e4 LT |
1756 | }; |
1757 | ||
1758 | /* This is used for a general mmap of a disk file */ | |
1759 | ||
1760 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1761 | { | |
1762 | struct address_space *mapping = file->f_mapping; | |
1763 | ||
1764 | if (!mapping->a_ops->readpage) | |
1765 | return -ENOEXEC; | |
1766 | file_accessed(file); | |
1767 | vma->vm_ops = &generic_file_vm_ops; | |
d0217ac0 | 1768 | vma->vm_flags |= VM_CAN_NONLINEAR; |
1da177e4 LT |
1769 | return 0; |
1770 | } | |
1da177e4 LT |
1771 | |
1772 | /* | |
1773 | * This is for filesystems which do not implement ->writepage. | |
1774 | */ | |
1775 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
1776 | { | |
1777 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
1778 | return -EINVAL; | |
1779 | return generic_file_mmap(file, vma); | |
1780 | } | |
1781 | #else | |
1782 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1783 | { | |
1784 | return -ENOSYS; | |
1785 | } | |
1786 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
1787 | { | |
1788 | return -ENOSYS; | |
1789 | } | |
1790 | #endif /* CONFIG_MMU */ | |
1791 | ||
1792 | EXPORT_SYMBOL(generic_file_mmap); | |
1793 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
1794 | ||
6fe6900e | 1795 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 1796 | pgoff_t index, |
5e5358e7 | 1797 | int (*filler)(void *, struct page *), |
0531b2aa LT |
1798 | void *data, |
1799 | gfp_t gfp) | |
1da177e4 | 1800 | { |
eb2be189 | 1801 | struct page *page; |
1da177e4 LT |
1802 | int err; |
1803 | repeat: | |
1804 | page = find_get_page(mapping, index); | |
1805 | if (!page) { | |
0531b2aa | 1806 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
1807 | if (!page) |
1808 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 1809 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 NP |
1810 | if (unlikely(err)) { |
1811 | page_cache_release(page); | |
1812 | if (err == -EEXIST) | |
1813 | goto repeat; | |
1da177e4 | 1814 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
1815 | return ERR_PTR(err); |
1816 | } | |
1da177e4 LT |
1817 | err = filler(data, page); |
1818 | if (err < 0) { | |
1819 | page_cache_release(page); | |
1820 | page = ERR_PTR(err); | |
1821 | } | |
1822 | } | |
1da177e4 LT |
1823 | return page; |
1824 | } | |
1825 | ||
0531b2aa | 1826 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 1827 | pgoff_t index, |
5e5358e7 | 1828 | int (*filler)(void *, struct page *), |
0531b2aa LT |
1829 | void *data, |
1830 | gfp_t gfp) | |
1831 | ||
1da177e4 LT |
1832 | { |
1833 | struct page *page; | |
1834 | int err; | |
1835 | ||
1836 | retry: | |
0531b2aa | 1837 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 1838 | if (IS_ERR(page)) |
c855ff37 | 1839 | return page; |
1da177e4 LT |
1840 | if (PageUptodate(page)) |
1841 | goto out; | |
1842 | ||
1843 | lock_page(page); | |
1844 | if (!page->mapping) { | |
1845 | unlock_page(page); | |
1846 | page_cache_release(page); | |
1847 | goto retry; | |
1848 | } | |
1849 | if (PageUptodate(page)) { | |
1850 | unlock_page(page); | |
1851 | goto out; | |
1852 | } | |
1853 | err = filler(data, page); | |
1854 | if (err < 0) { | |
1855 | page_cache_release(page); | |
c855ff37 | 1856 | return ERR_PTR(err); |
1da177e4 | 1857 | } |
c855ff37 | 1858 | out: |
6fe6900e NP |
1859 | mark_page_accessed(page); |
1860 | return page; | |
1861 | } | |
0531b2aa LT |
1862 | |
1863 | /** | |
1864 | * read_cache_page_async - read into page cache, fill it if needed | |
1865 | * @mapping: the page's address_space | |
1866 | * @index: the page index | |
1867 | * @filler: function to perform the read | |
5e5358e7 | 1868 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa LT |
1869 | * |
1870 | * Same as read_cache_page, but don't wait for page to become unlocked | |
1871 | * after submitting it to the filler. | |
1872 | * | |
1873 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1874 | * not set, try to fill the page but don't wait for it to become unlocked. | |
1875 | * | |
1876 | * If the page does not get brought uptodate, return -EIO. | |
1877 | */ | |
1878 | struct page *read_cache_page_async(struct address_space *mapping, | |
1879 | pgoff_t index, | |
5e5358e7 | 1880 | int (*filler)(void *, struct page *), |
0531b2aa LT |
1881 | void *data) |
1882 | { | |
1883 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
1884 | } | |
6fe6900e NP |
1885 | EXPORT_SYMBOL(read_cache_page_async); |
1886 | ||
0531b2aa LT |
1887 | static struct page *wait_on_page_read(struct page *page) |
1888 | { | |
1889 | if (!IS_ERR(page)) { | |
1890 | wait_on_page_locked(page); | |
1891 | if (!PageUptodate(page)) { | |
1892 | page_cache_release(page); | |
1893 | page = ERR_PTR(-EIO); | |
1894 | } | |
1895 | } | |
1896 | return page; | |
1897 | } | |
1898 | ||
1899 | /** | |
1900 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
1901 | * @mapping: the page's address_space | |
1902 | * @index: the page index | |
1903 | * @gfp: the page allocator flags to use if allocating | |
1904 | * | |
1905 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 1906 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
1907 | * |
1908 | * If the page does not get brought uptodate, return -EIO. | |
1909 | */ | |
1910 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
1911 | pgoff_t index, | |
1912 | gfp_t gfp) | |
1913 | { | |
1914 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
1915 | ||
1916 | return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp)); | |
1917 | } | |
1918 | EXPORT_SYMBOL(read_cache_page_gfp); | |
1919 | ||
6fe6900e NP |
1920 | /** |
1921 | * read_cache_page - read into page cache, fill it if needed | |
1922 | * @mapping: the page's address_space | |
1923 | * @index: the page index | |
1924 | * @filler: function to perform the read | |
5e5358e7 | 1925 | * @data: first arg to filler(data, page) function, often left as NULL |
6fe6900e NP |
1926 | * |
1927 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1928 | * not set, try to fill the page then wait for it to become unlocked. | |
1929 | * | |
1930 | * If the page does not get brought uptodate, return -EIO. | |
1931 | */ | |
1932 | struct page *read_cache_page(struct address_space *mapping, | |
57f6b96c | 1933 | pgoff_t index, |
5e5358e7 | 1934 | int (*filler)(void *, struct page *), |
6fe6900e NP |
1935 | void *data) |
1936 | { | |
0531b2aa | 1937 | return wait_on_page_read(read_cache_page_async(mapping, index, filler, data)); |
1da177e4 | 1938 | } |
1da177e4 LT |
1939 | EXPORT_SYMBOL(read_cache_page); |
1940 | ||
1da177e4 LT |
1941 | /* |
1942 | * The logic we want is | |
1943 | * | |
1944 | * if suid or (sgid and xgrp) | |
1945 | * remove privs | |
1946 | */ | |
01de85e0 | 1947 | int should_remove_suid(struct dentry *dentry) |
1da177e4 | 1948 | { |
649fc7b1 | 1949 | umode_t mode = dentry->d_inode->i_mode; |
1da177e4 | 1950 | int kill = 0; |
1da177e4 LT |
1951 | |
1952 | /* suid always must be killed */ | |
1953 | if (unlikely(mode & S_ISUID)) | |
1954 | kill = ATTR_KILL_SUID; | |
1955 | ||
1956 | /* | |
1957 | * sgid without any exec bits is just a mandatory locking mark; leave | |
1958 | * it alone. If some exec bits are set, it's a real sgid; kill it. | |
1959 | */ | |
1960 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
1961 | kill |= ATTR_KILL_SGID; | |
1962 | ||
7f5ff766 | 1963 | if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) |
01de85e0 | 1964 | return kill; |
1da177e4 | 1965 | |
01de85e0 JA |
1966 | return 0; |
1967 | } | |
d23a147b | 1968 | EXPORT_SYMBOL(should_remove_suid); |
01de85e0 | 1969 | |
7f3d4ee1 | 1970 | static int __remove_suid(struct dentry *dentry, int kill) |
01de85e0 JA |
1971 | { |
1972 | struct iattr newattrs; | |
1973 | ||
1974 | newattrs.ia_valid = ATTR_FORCE | kill; | |
1975 | return notify_change(dentry, &newattrs); | |
1976 | } | |
1977 | ||
2f1936b8 | 1978 | int file_remove_suid(struct file *file) |
01de85e0 | 1979 | { |
2f1936b8 | 1980 | struct dentry *dentry = file->f_path.dentry; |
69b45732 AK |
1981 | struct inode *inode = dentry->d_inode; |
1982 | int killsuid; | |
1983 | int killpriv; | |
b5376771 | 1984 | int error = 0; |
01de85e0 | 1985 | |
69b45732 AK |
1986 | /* Fast path for nothing security related */ |
1987 | if (IS_NOSEC(inode)) | |
1988 | return 0; | |
1989 | ||
1990 | killsuid = should_remove_suid(dentry); | |
1991 | killpriv = security_inode_need_killpriv(dentry); | |
1992 | ||
b5376771 SH |
1993 | if (killpriv < 0) |
1994 | return killpriv; | |
1995 | if (killpriv) | |
1996 | error = security_inode_killpriv(dentry); | |
1997 | if (!error && killsuid) | |
1998 | error = __remove_suid(dentry, killsuid); | |
9e1f1de0 | 1999 | if (!error && (inode->i_sb->s_flags & MS_NOSEC)) |
69b45732 | 2000 | inode->i_flags |= S_NOSEC; |
01de85e0 | 2001 | |
b5376771 | 2002 | return error; |
1da177e4 | 2003 | } |
2f1936b8 | 2004 | EXPORT_SYMBOL(file_remove_suid); |
1da177e4 | 2005 | |
2f718ffc | 2006 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
1da177e4 LT |
2007 | const struct iovec *iov, size_t base, size_t bytes) |
2008 | { | |
f1800536 | 2009 | size_t copied = 0, left = 0; |
1da177e4 LT |
2010 | |
2011 | while (bytes) { | |
2012 | char __user *buf = iov->iov_base + base; | |
2013 | int copy = min(bytes, iov->iov_len - base); | |
2014 | ||
2015 | base = 0; | |
f1800536 | 2016 | left = __copy_from_user_inatomic(vaddr, buf, copy); |
1da177e4 LT |
2017 | copied += copy; |
2018 | bytes -= copy; | |
2019 | vaddr += copy; | |
2020 | iov++; | |
2021 | ||
01408c49 | 2022 | if (unlikely(left)) |
1da177e4 | 2023 | break; |
1da177e4 LT |
2024 | } |
2025 | return copied - left; | |
2026 | } | |
2027 | ||
2f718ffc NP |
2028 | /* |
2029 | * Copy as much as we can into the page and return the number of bytes which | |
af901ca1 | 2030 | * were successfully copied. If a fault is encountered then return the number of |
2f718ffc NP |
2031 | * bytes which were copied. |
2032 | */ | |
2033 | size_t iov_iter_copy_from_user_atomic(struct page *page, | |
2034 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
2035 | { | |
2036 | char *kaddr; | |
2037 | size_t copied; | |
2038 | ||
2039 | BUG_ON(!in_atomic()); | |
9b04c5fe | 2040 | kaddr = kmap_atomic(page); |
2f718ffc NP |
2041 | if (likely(i->nr_segs == 1)) { |
2042 | int left; | |
2043 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 2044 | left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); |
2f718ffc NP |
2045 | copied = bytes - left; |
2046 | } else { | |
2047 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2048 | i->iov, i->iov_offset, bytes); | |
2049 | } | |
9b04c5fe | 2050 | kunmap_atomic(kaddr); |
2f718ffc NP |
2051 | |
2052 | return copied; | |
2053 | } | |
89e10787 | 2054 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); |
2f718ffc NP |
2055 | |
2056 | /* | |
2057 | * This has the same sideeffects and return value as | |
2058 | * iov_iter_copy_from_user_atomic(). | |
2059 | * The difference is that it attempts to resolve faults. | |
2060 | * Page must not be locked. | |
2061 | */ | |
2062 | size_t iov_iter_copy_from_user(struct page *page, | |
2063 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
2064 | { | |
2065 | char *kaddr; | |
2066 | size_t copied; | |
2067 | ||
2068 | kaddr = kmap(page); | |
2069 | if (likely(i->nr_segs == 1)) { | |
2070 | int left; | |
2071 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 2072 | left = __copy_from_user(kaddr + offset, buf, bytes); |
2f718ffc NP |
2073 | copied = bytes - left; |
2074 | } else { | |
2075 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2076 | i->iov, i->iov_offset, bytes); | |
2077 | } | |
2078 | kunmap(page); | |
2079 | return copied; | |
2080 | } | |
89e10787 | 2081 | EXPORT_SYMBOL(iov_iter_copy_from_user); |
2f718ffc | 2082 | |
f7009264 | 2083 | void iov_iter_advance(struct iov_iter *i, size_t bytes) |
2f718ffc | 2084 | { |
f7009264 NP |
2085 | BUG_ON(i->count < bytes); |
2086 | ||
2f718ffc NP |
2087 | if (likely(i->nr_segs == 1)) { |
2088 | i->iov_offset += bytes; | |
f7009264 | 2089 | i->count -= bytes; |
2f718ffc NP |
2090 | } else { |
2091 | const struct iovec *iov = i->iov; | |
2092 | size_t base = i->iov_offset; | |
39be79c1 | 2093 | unsigned long nr_segs = i->nr_segs; |
2f718ffc | 2094 | |
124d3b70 NP |
2095 | /* |
2096 | * The !iov->iov_len check ensures we skip over unlikely | |
f7009264 | 2097 | * zero-length segments (without overruning the iovec). |
124d3b70 | 2098 | */ |
94ad374a | 2099 | while (bytes || unlikely(i->count && !iov->iov_len)) { |
f7009264 | 2100 | int copy; |
2f718ffc | 2101 | |
f7009264 NP |
2102 | copy = min(bytes, iov->iov_len - base); |
2103 | BUG_ON(!i->count || i->count < copy); | |
2104 | i->count -= copy; | |
2f718ffc NP |
2105 | bytes -= copy; |
2106 | base += copy; | |
2107 | if (iov->iov_len == base) { | |
2108 | iov++; | |
39be79c1 | 2109 | nr_segs--; |
2f718ffc NP |
2110 | base = 0; |
2111 | } | |
2112 | } | |
2113 | i->iov = iov; | |
2114 | i->iov_offset = base; | |
39be79c1 | 2115 | i->nr_segs = nr_segs; |
2f718ffc NP |
2116 | } |
2117 | } | |
89e10787 | 2118 | EXPORT_SYMBOL(iov_iter_advance); |
2f718ffc | 2119 | |
afddba49 NP |
2120 | /* |
2121 | * Fault in the first iovec of the given iov_iter, to a maximum length | |
2122 | * of bytes. Returns 0 on success, or non-zero if the memory could not be | |
2123 | * accessed (ie. because it is an invalid address). | |
2124 | * | |
2125 | * writev-intensive code may want this to prefault several iovecs -- that | |
2126 | * would be possible (callers must not rely on the fact that _only_ the | |
2127 | * first iovec will be faulted with the current implementation). | |
2128 | */ | |
2129 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) | |
2f718ffc | 2130 | { |
2f718ffc | 2131 | char __user *buf = i->iov->iov_base + i->iov_offset; |
afddba49 NP |
2132 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); |
2133 | return fault_in_pages_readable(buf, bytes); | |
2f718ffc | 2134 | } |
89e10787 | 2135 | EXPORT_SYMBOL(iov_iter_fault_in_readable); |
2f718ffc NP |
2136 | |
2137 | /* | |
2138 | * Return the count of just the current iov_iter segment. | |
2139 | */ | |
2140 | size_t iov_iter_single_seg_count(struct iov_iter *i) | |
2141 | { | |
2142 | const struct iovec *iov = i->iov; | |
2143 | if (i->nr_segs == 1) | |
2144 | return i->count; | |
2145 | else | |
2146 | return min(i->count, iov->iov_len - i->iov_offset); | |
2147 | } | |
89e10787 | 2148 | EXPORT_SYMBOL(iov_iter_single_seg_count); |
2f718ffc | 2149 | |
1da177e4 LT |
2150 | /* |
2151 | * Performs necessary checks before doing a write | |
2152 | * | |
485bb99b | 2153 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2154 | * Returns appropriate error code that caller should return or |
2155 | * zero in case that write should be allowed. | |
2156 | */ | |
2157 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) | |
2158 | { | |
2159 | struct inode *inode = file->f_mapping->host; | |
59e99e5b | 2160 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
1da177e4 LT |
2161 | |
2162 | if (unlikely(*pos < 0)) | |
2163 | return -EINVAL; | |
2164 | ||
1da177e4 LT |
2165 | if (!isblk) { |
2166 | /* FIXME: this is for backwards compatibility with 2.4 */ | |
2167 | if (file->f_flags & O_APPEND) | |
2168 | *pos = i_size_read(inode); | |
2169 | ||
2170 | if (limit != RLIM_INFINITY) { | |
2171 | if (*pos >= limit) { | |
2172 | send_sig(SIGXFSZ, current, 0); | |
2173 | return -EFBIG; | |
2174 | } | |
2175 | if (*count > limit - (typeof(limit))*pos) { | |
2176 | *count = limit - (typeof(limit))*pos; | |
2177 | } | |
2178 | } | |
2179 | } | |
2180 | ||
2181 | /* | |
2182 | * LFS rule | |
2183 | */ | |
2184 | if (unlikely(*pos + *count > MAX_NON_LFS && | |
2185 | !(file->f_flags & O_LARGEFILE))) { | |
2186 | if (*pos >= MAX_NON_LFS) { | |
1da177e4 LT |
2187 | return -EFBIG; |
2188 | } | |
2189 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { | |
2190 | *count = MAX_NON_LFS - (unsigned long)*pos; | |
2191 | } | |
2192 | } | |
2193 | ||
2194 | /* | |
2195 | * Are we about to exceed the fs block limit ? | |
2196 | * | |
2197 | * If we have written data it becomes a short write. If we have | |
2198 | * exceeded without writing data we send a signal and return EFBIG. | |
2199 | * Linus frestrict idea will clean these up nicely.. | |
2200 | */ | |
2201 | if (likely(!isblk)) { | |
2202 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { | |
2203 | if (*count || *pos > inode->i_sb->s_maxbytes) { | |
1da177e4 LT |
2204 | return -EFBIG; |
2205 | } | |
2206 | /* zero-length writes at ->s_maxbytes are OK */ | |
2207 | } | |
2208 | ||
2209 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) | |
2210 | *count = inode->i_sb->s_maxbytes - *pos; | |
2211 | } else { | |
9361401e | 2212 | #ifdef CONFIG_BLOCK |
1da177e4 LT |
2213 | loff_t isize; |
2214 | if (bdev_read_only(I_BDEV(inode))) | |
2215 | return -EPERM; | |
2216 | isize = i_size_read(inode); | |
2217 | if (*pos >= isize) { | |
2218 | if (*count || *pos > isize) | |
2219 | return -ENOSPC; | |
2220 | } | |
2221 | ||
2222 | if (*pos + *count > isize) | |
2223 | *count = isize - *pos; | |
9361401e DH |
2224 | #else |
2225 | return -EPERM; | |
2226 | #endif | |
1da177e4 LT |
2227 | } |
2228 | return 0; | |
2229 | } | |
2230 | EXPORT_SYMBOL(generic_write_checks); | |
2231 | ||
afddba49 NP |
2232 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2233 | loff_t pos, unsigned len, unsigned flags, | |
2234 | struct page **pagep, void **fsdata) | |
2235 | { | |
2236 | const struct address_space_operations *aops = mapping->a_ops; | |
2237 | ||
4e02ed4b | 2238 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2239 | pagep, fsdata); |
afddba49 NP |
2240 | } |
2241 | EXPORT_SYMBOL(pagecache_write_begin); | |
2242 | ||
2243 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2244 | loff_t pos, unsigned len, unsigned copied, | |
2245 | struct page *page, void *fsdata) | |
2246 | { | |
2247 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2248 | |
4e02ed4b NP |
2249 | mark_page_accessed(page); |
2250 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); | |
afddba49 NP |
2251 | } |
2252 | EXPORT_SYMBOL(pagecache_write_end); | |
2253 | ||
1da177e4 LT |
2254 | ssize_t |
2255 | generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, | |
2256 | unsigned long *nr_segs, loff_t pos, loff_t *ppos, | |
2257 | size_t count, size_t ocount) | |
2258 | { | |
2259 | struct file *file = iocb->ki_filp; | |
2260 | struct address_space *mapping = file->f_mapping; | |
2261 | struct inode *inode = mapping->host; | |
2262 | ssize_t written; | |
a969e903 CH |
2263 | size_t write_len; |
2264 | pgoff_t end; | |
1da177e4 LT |
2265 | |
2266 | if (count != ocount) | |
2267 | *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); | |
2268 | ||
a969e903 CH |
2269 | write_len = iov_length(iov, *nr_segs); |
2270 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; | |
a969e903 | 2271 | |
48b47c56 | 2272 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2273 | if (written) |
2274 | goto out; | |
2275 | ||
2276 | /* | |
2277 | * After a write we want buffered reads to be sure to go to disk to get | |
2278 | * the new data. We invalidate clean cached page from the region we're | |
2279 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2280 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2281 | */ |
2282 | if (mapping->nrpages) { | |
2283 | written = invalidate_inode_pages2_range(mapping, | |
2284 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2285 | /* |
2286 | * If a page can not be invalidated, return 0 to fall back | |
2287 | * to buffered write. | |
2288 | */ | |
2289 | if (written) { | |
2290 | if (written == -EBUSY) | |
2291 | return 0; | |
a969e903 | 2292 | goto out; |
6ccfa806 | 2293 | } |
a969e903 CH |
2294 | } |
2295 | ||
2296 | written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs); | |
2297 | ||
2298 | /* | |
2299 | * Finally, try again to invalidate clean pages which might have been | |
2300 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2301 | * if the source of the write was an mmap'ed region of the file | |
2302 | * we're writing. Either one is a pretty crazy thing to do, | |
2303 | * so we don't support it 100%. If this invalidation | |
2304 | * fails, tough, the write still worked... | |
2305 | */ | |
2306 | if (mapping->nrpages) { | |
2307 | invalidate_inode_pages2_range(mapping, | |
2308 | pos >> PAGE_CACHE_SHIFT, end); | |
2309 | } | |
2310 | ||
1da177e4 | 2311 | if (written > 0) { |
0116651c NK |
2312 | pos += written; |
2313 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { | |
2314 | i_size_write(inode, pos); | |
1da177e4 LT |
2315 | mark_inode_dirty(inode); |
2316 | } | |
0116651c | 2317 | *ppos = pos; |
1da177e4 | 2318 | } |
a969e903 | 2319 | out: |
1da177e4 LT |
2320 | return written; |
2321 | } | |
2322 | EXPORT_SYMBOL(generic_file_direct_write); | |
2323 | ||
eb2be189 NP |
2324 | /* |
2325 | * Find or create a page at the given pagecache position. Return the locked | |
2326 | * page. This function is specifically for buffered writes. | |
2327 | */ | |
54566b2c NP |
2328 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2329 | pgoff_t index, unsigned flags) | |
eb2be189 NP |
2330 | { |
2331 | int status; | |
0faa70cb | 2332 | gfp_t gfp_mask; |
eb2be189 | 2333 | struct page *page; |
54566b2c | 2334 | gfp_t gfp_notmask = 0; |
0faa70cb | 2335 | |
1010bb1b FW |
2336 | gfp_mask = mapping_gfp_mask(mapping); |
2337 | if (mapping_cap_account_dirty(mapping)) | |
2338 | gfp_mask |= __GFP_WRITE; | |
54566b2c NP |
2339 | if (flags & AOP_FLAG_NOFS) |
2340 | gfp_notmask = __GFP_FS; | |
eb2be189 NP |
2341 | repeat: |
2342 | page = find_lock_page(mapping, index); | |
c585a267 | 2343 | if (page) |
3d08bcc8 | 2344 | goto found; |
eb2be189 | 2345 | |
0faa70cb | 2346 | page = __page_cache_alloc(gfp_mask & ~gfp_notmask); |
eb2be189 NP |
2347 | if (!page) |
2348 | return NULL; | |
54566b2c NP |
2349 | status = add_to_page_cache_lru(page, mapping, index, |
2350 | GFP_KERNEL & ~gfp_notmask); | |
eb2be189 NP |
2351 | if (unlikely(status)) { |
2352 | page_cache_release(page); | |
2353 | if (status == -EEXIST) | |
2354 | goto repeat; | |
2355 | return NULL; | |
2356 | } | |
3d08bcc8 DW |
2357 | found: |
2358 | wait_on_page_writeback(page); | |
eb2be189 NP |
2359 | return page; |
2360 | } | |
54566b2c | 2361 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2362 | |
afddba49 NP |
2363 | static ssize_t generic_perform_write(struct file *file, |
2364 | struct iov_iter *i, loff_t pos) | |
2365 | { | |
2366 | struct address_space *mapping = file->f_mapping; | |
2367 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2368 | long status = 0; | |
2369 | ssize_t written = 0; | |
674b892e NP |
2370 | unsigned int flags = 0; |
2371 | ||
2372 | /* | |
2373 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2374 | */ | |
2375 | if (segment_eq(get_fs(), KERNEL_DS)) | |
2376 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
afddba49 NP |
2377 | |
2378 | do { | |
2379 | struct page *page; | |
afddba49 NP |
2380 | unsigned long offset; /* Offset into pagecache page */ |
2381 | unsigned long bytes; /* Bytes to write to page */ | |
2382 | size_t copied; /* Bytes copied from user */ | |
2383 | void *fsdata; | |
2384 | ||
2385 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2386 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2387 | iov_iter_count(i)); | |
2388 | ||
2389 | again: | |
afddba49 NP |
2390 | /* |
2391 | * Bring in the user page that we will copy from _first_. | |
2392 | * Otherwise there's a nasty deadlock on copying from the | |
2393 | * same page as we're writing to, without it being marked | |
2394 | * up-to-date. | |
2395 | * | |
2396 | * Not only is this an optimisation, but it is also required | |
2397 | * to check that the address is actually valid, when atomic | |
2398 | * usercopies are used, below. | |
2399 | */ | |
2400 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2401 | status = -EFAULT; | |
2402 | break; | |
2403 | } | |
2404 | ||
674b892e | 2405 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 NP |
2406 | &page, &fsdata); |
2407 | if (unlikely(status)) | |
2408 | break; | |
2409 | ||
931e80e4 | 2410 | if (mapping_writably_mapped(mapping)) |
2411 | flush_dcache_page(page); | |
2412 | ||
afddba49 NP |
2413 | pagefault_disable(); |
2414 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); | |
2415 | pagefault_enable(); | |
2416 | flush_dcache_page(page); | |
2417 | ||
c8236db9 | 2418 | mark_page_accessed(page); |
afddba49 NP |
2419 | status = a_ops->write_end(file, mapping, pos, bytes, copied, |
2420 | page, fsdata); | |
2421 | if (unlikely(status < 0)) | |
2422 | break; | |
2423 | copied = status; | |
2424 | ||
2425 | cond_resched(); | |
2426 | ||
124d3b70 | 2427 | iov_iter_advance(i, copied); |
afddba49 NP |
2428 | if (unlikely(copied == 0)) { |
2429 | /* | |
2430 | * If we were unable to copy any data at all, we must | |
2431 | * fall back to a single segment length write. | |
2432 | * | |
2433 | * If we didn't fallback here, we could livelock | |
2434 | * because not all segments in the iov can be copied at | |
2435 | * once without a pagefault. | |
2436 | */ | |
2437 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2438 | iov_iter_single_seg_count(i)); | |
2439 | goto again; | |
2440 | } | |
afddba49 NP |
2441 | pos += copied; |
2442 | written += copied; | |
2443 | ||
2444 | balance_dirty_pages_ratelimited(mapping); | |
a50527b1 JK |
2445 | if (fatal_signal_pending(current)) { |
2446 | status = -EINTR; | |
2447 | break; | |
2448 | } | |
afddba49 NP |
2449 | } while (iov_iter_count(i)); |
2450 | ||
2451 | return written ? written : status; | |
2452 | } | |
2453 | ||
2454 | ssize_t | |
2455 | generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, | |
2456 | unsigned long nr_segs, loff_t pos, loff_t *ppos, | |
2457 | size_t count, ssize_t written) | |
2458 | { | |
2459 | struct file *file = iocb->ki_filp; | |
afddba49 NP |
2460 | ssize_t status; |
2461 | struct iov_iter i; | |
2462 | ||
2463 | iov_iter_init(&i, iov, nr_segs, count, written); | |
4e02ed4b | 2464 | status = generic_perform_write(file, &i, pos); |
1da177e4 | 2465 | |
1da177e4 | 2466 | if (likely(status >= 0)) { |
afddba49 NP |
2467 | written += status; |
2468 | *ppos = pos + status; | |
1da177e4 LT |
2469 | } |
2470 | ||
1da177e4 LT |
2471 | return written ? written : status; |
2472 | } | |
2473 | EXPORT_SYMBOL(generic_file_buffered_write); | |
2474 | ||
e4dd9de3 JK |
2475 | /** |
2476 | * __generic_file_aio_write - write data to a file | |
2477 | * @iocb: IO state structure (file, offset, etc.) | |
2478 | * @iov: vector with data to write | |
2479 | * @nr_segs: number of segments in the vector | |
2480 | * @ppos: position where to write | |
2481 | * | |
2482 | * This function does all the work needed for actually writing data to a | |
2483 | * file. It does all basic checks, removes SUID from the file, updates | |
2484 | * modification times and calls proper subroutines depending on whether we | |
2485 | * do direct IO or a standard buffered write. | |
2486 | * | |
2487 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2488 | * object which does not need locking at all. | |
2489 | * | |
2490 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2491 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2492 | * avoid syncing under i_mutex. | |
2493 | */ | |
2494 | ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
2495 | unsigned long nr_segs, loff_t *ppos) | |
1da177e4 LT |
2496 | { |
2497 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2498 | struct address_space * mapping = file->f_mapping; |
1da177e4 LT |
2499 | size_t ocount; /* original count */ |
2500 | size_t count; /* after file limit checks */ | |
2501 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2502 | loff_t pos; |
2503 | ssize_t written; | |
2504 | ssize_t err; | |
2505 | ||
2506 | ocount = 0; | |
0ceb3314 DM |
2507 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
2508 | if (err) | |
2509 | return err; | |
1da177e4 LT |
2510 | |
2511 | count = ocount; | |
2512 | pos = *ppos; | |
2513 | ||
2514 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
2515 | ||
2516 | /* We can write back this queue in page reclaim */ | |
2517 | current->backing_dev_info = mapping->backing_dev_info; | |
2518 | written = 0; | |
2519 | ||
2520 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
2521 | if (err) | |
2522 | goto out; | |
2523 | ||
2524 | if (count == 0) | |
2525 | goto out; | |
2526 | ||
2f1936b8 | 2527 | err = file_remove_suid(file); |
1da177e4 LT |
2528 | if (err) |
2529 | goto out; | |
2530 | ||
870f4817 | 2531 | file_update_time(file); |
1da177e4 LT |
2532 | |
2533 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
2534 | if (unlikely(file->f_flags & O_DIRECT)) { | |
fb5527e6 JM |
2535 | loff_t endbyte; |
2536 | ssize_t written_buffered; | |
2537 | ||
2538 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, | |
2539 | ppos, count, ocount); | |
1da177e4 LT |
2540 | if (written < 0 || written == count) |
2541 | goto out; | |
2542 | /* | |
2543 | * direct-io write to a hole: fall through to buffered I/O | |
2544 | * for completing the rest of the request. | |
2545 | */ | |
2546 | pos += written; | |
2547 | count -= written; | |
fb5527e6 JM |
2548 | written_buffered = generic_file_buffered_write(iocb, iov, |
2549 | nr_segs, pos, ppos, count, | |
2550 | written); | |
2551 | /* | |
2552 | * If generic_file_buffered_write() retuned a synchronous error | |
2553 | * then we want to return the number of bytes which were | |
2554 | * direct-written, or the error code if that was zero. Note | |
2555 | * that this differs from normal direct-io semantics, which | |
2556 | * will return -EFOO even if some bytes were written. | |
2557 | */ | |
2558 | if (written_buffered < 0) { | |
2559 | err = written_buffered; | |
2560 | goto out; | |
2561 | } | |
1da177e4 | 2562 | |
fb5527e6 JM |
2563 | /* |
2564 | * We need to ensure that the page cache pages are written to | |
2565 | * disk and invalidated to preserve the expected O_DIRECT | |
2566 | * semantics. | |
2567 | */ | |
2568 | endbyte = pos + written_buffered - written - 1; | |
c05c4edd | 2569 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); |
fb5527e6 JM |
2570 | if (err == 0) { |
2571 | written = written_buffered; | |
2572 | invalidate_mapping_pages(mapping, | |
2573 | pos >> PAGE_CACHE_SHIFT, | |
2574 | endbyte >> PAGE_CACHE_SHIFT); | |
2575 | } else { | |
2576 | /* | |
2577 | * We don't know how much we wrote, so just return | |
2578 | * the number of bytes which were direct-written | |
2579 | */ | |
2580 | } | |
2581 | } else { | |
2582 | written = generic_file_buffered_write(iocb, iov, nr_segs, | |
2583 | pos, ppos, count, written); | |
2584 | } | |
1da177e4 LT |
2585 | out: |
2586 | current->backing_dev_info = NULL; | |
2587 | return written ? written : err; | |
2588 | } | |
e4dd9de3 JK |
2589 | EXPORT_SYMBOL(__generic_file_aio_write); |
2590 | ||
e4dd9de3 JK |
2591 | /** |
2592 | * generic_file_aio_write - write data to a file | |
2593 | * @iocb: IO state structure | |
2594 | * @iov: vector with data to write | |
2595 | * @nr_segs: number of segments in the vector | |
2596 | * @pos: position in file where to write | |
2597 | * | |
2598 | * This is a wrapper around __generic_file_aio_write() to be used by most | |
2599 | * filesystems. It takes care of syncing the file in case of O_SYNC file | |
2600 | * and acquires i_mutex as needed. | |
2601 | */ | |
027445c3 BP |
2602 | ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
2603 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
2604 | { |
2605 | struct file *file = iocb->ki_filp; | |
148f948b | 2606 | struct inode *inode = file->f_mapping->host; |
55602dd6 | 2607 | struct blk_plug plug; |
1da177e4 | 2608 | ssize_t ret; |
1da177e4 LT |
2609 | |
2610 | BUG_ON(iocb->ki_pos != pos); | |
2611 | ||
1b1dcc1b | 2612 | mutex_lock(&inode->i_mutex); |
55602dd6 | 2613 | blk_start_plug(&plug); |
e4dd9de3 | 2614 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); |
1b1dcc1b | 2615 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 2616 | |
148f948b | 2617 | if (ret > 0 || ret == -EIOCBQUEUED) { |
1da177e4 LT |
2618 | ssize_t err; |
2619 | ||
148f948b | 2620 | err = generic_write_sync(file, pos, ret); |
c7b50db2 | 2621 | if (err < 0 && ret > 0) |
1da177e4 LT |
2622 | ret = err; |
2623 | } | |
55602dd6 | 2624 | blk_finish_plug(&plug); |
1da177e4 LT |
2625 | return ret; |
2626 | } | |
2627 | EXPORT_SYMBOL(generic_file_aio_write); | |
2628 | ||
cf9a2ae8 DH |
2629 | /** |
2630 | * try_to_release_page() - release old fs-specific metadata on a page | |
2631 | * | |
2632 | * @page: the page which the kernel is trying to free | |
2633 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2634 | * | |
2635 | * The address_space is to try to release any data against the page | |
2636 | * (presumably at page->private). If the release was successful, return `1'. | |
2637 | * Otherwise return zero. | |
2638 | * | |
266cf658 DH |
2639 | * This may also be called if PG_fscache is set on a page, indicating that the |
2640 | * page is known to the local caching routines. | |
2641 | * | |
cf9a2ae8 | 2642 | * The @gfp_mask argument specifies whether I/O may be performed to release |
3f31fddf | 2643 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
cf9a2ae8 | 2644 | * |
cf9a2ae8 DH |
2645 | */ |
2646 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2647 | { | |
2648 | struct address_space * const mapping = page->mapping; | |
2649 | ||
2650 | BUG_ON(!PageLocked(page)); | |
2651 | if (PageWriteback(page)) | |
2652 | return 0; | |
2653 | ||
2654 | if (mapping && mapping->a_ops->releasepage) | |
2655 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2656 | return try_to_free_buffers(page); | |
2657 | } | |
2658 | ||
2659 | EXPORT_SYMBOL(try_to_release_page); |