Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/filemap.c | |
3 | * | |
4 | * Copyright (C) 1994-1999 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * This file handles the generic file mmap semantics used by | |
9 | * most "normal" filesystems (but you don't /have/ to use this: | |
10 | * the NFS filesystem used to do this differently, for example) | |
11 | */ | |
b95f1b31 | 12 | #include <linux/export.h> |
1da177e4 LT |
13 | #include <linux/compiler.h> |
14 | #include <linux/fs.h> | |
c22ce143 | 15 | #include <linux/uaccess.h> |
1da177e4 | 16 | #include <linux/aio.h> |
c59ede7b | 17 | #include <linux/capability.h> |
1da177e4 | 18 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 19 | #include <linux/gfp.h> |
1da177e4 LT |
20 | #include <linux/mm.h> |
21 | #include <linux/swap.h> | |
22 | #include <linux/mman.h> | |
23 | #include <linux/pagemap.h> | |
24 | #include <linux/file.h> | |
25 | #include <linux/uio.h> | |
26 | #include <linux/hash.h> | |
27 | #include <linux/writeback.h> | |
53253383 | 28 | #include <linux/backing-dev.h> |
1da177e4 LT |
29 | #include <linux/pagevec.h> |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/security.h> | |
44110fe3 | 32 | #include <linux/cpuset.h> |
2f718ffc | 33 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
8a9f3ccd | 34 | #include <linux/memcontrol.h> |
c515e1fd | 35 | #include <linux/cleancache.h> |
0f8053a5 NP |
36 | #include "internal.h" |
37 | ||
fe0bfaaf RJ |
38 | #define CREATE_TRACE_POINTS |
39 | #include <trace/events/filemap.h> | |
40 | ||
1da177e4 | 41 | /* |
1da177e4 LT |
42 | * FIXME: remove all knowledge of the buffer layer from the core VM |
43 | */ | |
148f948b | 44 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 45 | |
1da177e4 LT |
46 | #include <asm/mman.h> |
47 | ||
48 | /* | |
49 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
50 | * though. | |
51 | * | |
52 | * Shared mappings now work. 15.8.1995 Bruno. | |
53 | * | |
54 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
55 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
56 | * | |
57 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
58 | */ | |
59 | ||
60 | /* | |
61 | * Lock ordering: | |
62 | * | |
3d48ae45 | 63 | * ->i_mmap_mutex (truncate_pagecache) |
1da177e4 | 64 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 HD |
65 | * ->swap_lock (exclusive_swap_page, others) |
66 | * ->mapping->tree_lock | |
1da177e4 | 67 | * |
1b1dcc1b | 68 | * ->i_mutex |
3d48ae45 | 69 | * ->i_mmap_mutex (truncate->unmap_mapping_range) |
1da177e4 LT |
70 | * |
71 | * ->mmap_sem | |
3d48ae45 | 72 | * ->i_mmap_mutex |
b8072f09 | 73 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
74 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
75 | * | |
76 | * ->mmap_sem | |
77 | * ->lock_page (access_process_vm) | |
78 | * | |
82591e6e NP |
79 | * ->i_mutex (generic_file_buffered_write) |
80 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) | |
1da177e4 | 81 | * |
f758eeab | 82 | * bdi->wb.list_lock |
a66979ab | 83 | * sb_lock (fs/fs-writeback.c) |
1da177e4 LT |
84 | * ->mapping->tree_lock (__sync_single_inode) |
85 | * | |
3d48ae45 | 86 | * ->i_mmap_mutex |
1da177e4 LT |
87 | * ->anon_vma.lock (vma_adjust) |
88 | * | |
89 | * ->anon_vma.lock | |
b8072f09 | 90 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 91 | * |
b8072f09 | 92 | * ->page_table_lock or pte_lock |
5d337b91 | 93 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
94 | * ->private_lock (try_to_unmap_one) |
95 | * ->tree_lock (try_to_unmap_one) | |
96 | * ->zone.lru_lock (follow_page->mark_page_accessed) | |
053837fc | 97 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
1da177e4 LT |
98 | * ->private_lock (page_remove_rmap->set_page_dirty) |
99 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
f758eeab | 100 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 101 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
f758eeab | 102 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 103 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
104 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
105 | * | |
9a3c531d AK |
106 | * ->i_mmap_mutex |
107 | * ->tasklist_lock (memory_failure, collect_procs_ao) | |
1da177e4 LT |
108 | */ |
109 | ||
110 | /* | |
e64a782f | 111 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 112 | * sure the page is locked and that nobody else uses it - or that usage |
19fd6231 | 113 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 114 | */ |
e64a782f | 115 | void __delete_from_page_cache(struct page *page) |
1da177e4 LT |
116 | { |
117 | struct address_space *mapping = page->mapping; | |
118 | ||
fe0bfaaf | 119 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
120 | /* |
121 | * if we're uptodate, flush out into the cleancache, otherwise | |
122 | * invalidate any existing cleancache entries. We can't leave | |
123 | * stale data around in the cleancache once our page is gone | |
124 | */ | |
125 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
126 | cleancache_put_page(page); | |
127 | else | |
3167760f | 128 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 129 | |
1da177e4 LT |
130 | radix_tree_delete(&mapping->page_tree, page->index); |
131 | page->mapping = NULL; | |
b85e0eff | 132 | /* Leave page->index set: truncation lookup relies upon it */ |
1da177e4 | 133 | mapping->nrpages--; |
347ce434 | 134 | __dec_zone_page_state(page, NR_FILE_PAGES); |
4b02108a KM |
135 | if (PageSwapBacked(page)) |
136 | __dec_zone_page_state(page, NR_SHMEM); | |
45426812 | 137 | BUG_ON(page_mapped(page)); |
3a692790 LT |
138 | |
139 | /* | |
140 | * Some filesystems seem to re-dirty the page even after | |
141 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
142 | * | |
143 | * Fix it up by doing a final dirty accounting check after | |
144 | * having removed the page entirely. | |
145 | */ | |
146 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { | |
147 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
148 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
149 | } | |
1da177e4 LT |
150 | } |
151 | ||
702cfbf9 MK |
152 | /** |
153 | * delete_from_page_cache - delete page from page cache | |
154 | * @page: the page which the kernel is trying to remove from page cache | |
155 | * | |
156 | * This must be called only on pages that have been verified to be in the page | |
157 | * cache and locked. It will never put the page into the free list, the caller | |
158 | * has a reference on the page. | |
159 | */ | |
160 | void delete_from_page_cache(struct page *page) | |
1da177e4 LT |
161 | { |
162 | struct address_space *mapping = page->mapping; | |
6072d13c | 163 | void (*freepage)(struct page *); |
1da177e4 | 164 | |
cd7619d6 | 165 | BUG_ON(!PageLocked(page)); |
1da177e4 | 166 | |
6072d13c | 167 | freepage = mapping->a_ops->freepage; |
19fd6231 | 168 | spin_lock_irq(&mapping->tree_lock); |
e64a782f | 169 | __delete_from_page_cache(page); |
19fd6231 | 170 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 171 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
172 | |
173 | if (freepage) | |
174 | freepage(page); | |
97cecb5a MK |
175 | page_cache_release(page); |
176 | } | |
177 | EXPORT_SYMBOL(delete_from_page_cache); | |
178 | ||
7eaceacc | 179 | static int sleep_on_page(void *word) |
1da177e4 | 180 | { |
1da177e4 LT |
181 | io_schedule(); |
182 | return 0; | |
183 | } | |
184 | ||
7eaceacc | 185 | static int sleep_on_page_killable(void *word) |
2687a356 | 186 | { |
7eaceacc | 187 | sleep_on_page(word); |
2687a356 MW |
188 | return fatal_signal_pending(current) ? -EINTR : 0; |
189 | } | |
190 | ||
865ffef3 DM |
191 | static int filemap_check_errors(struct address_space *mapping) |
192 | { | |
193 | int ret = 0; | |
194 | /* Check for outstanding write errors */ | |
195 | if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
196 | ret = -ENOSPC; | |
197 | if (test_and_clear_bit(AS_EIO, &mapping->flags)) | |
198 | ret = -EIO; | |
199 | return ret; | |
200 | } | |
201 | ||
1da177e4 | 202 | /** |
485bb99b | 203 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
204 | * @mapping: address space structure to write |
205 | * @start: offset in bytes where the range starts | |
469eb4d0 | 206 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 207 | * @sync_mode: enable synchronous operation |
1da177e4 | 208 | * |
485bb99b RD |
209 | * Start writeback against all of a mapping's dirty pages that lie |
210 | * within the byte offsets <start, end> inclusive. | |
211 | * | |
1da177e4 | 212 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 213 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
214 | * these two operations is that if a dirty page/buffer is encountered, it must |
215 | * be waited upon, and not just skipped over. | |
216 | */ | |
ebcf28e1 AM |
217 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
218 | loff_t end, int sync_mode) | |
1da177e4 LT |
219 | { |
220 | int ret; | |
221 | struct writeback_control wbc = { | |
222 | .sync_mode = sync_mode, | |
05fe478d | 223 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
224 | .range_start = start, |
225 | .range_end = end, | |
1da177e4 LT |
226 | }; |
227 | ||
228 | if (!mapping_cap_writeback_dirty(mapping)) | |
229 | return 0; | |
230 | ||
231 | ret = do_writepages(mapping, &wbc); | |
232 | return ret; | |
233 | } | |
234 | ||
235 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
236 | int sync_mode) | |
237 | { | |
111ebb6e | 238 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
239 | } |
240 | ||
241 | int filemap_fdatawrite(struct address_space *mapping) | |
242 | { | |
243 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
244 | } | |
245 | EXPORT_SYMBOL(filemap_fdatawrite); | |
246 | ||
f4c0a0fd | 247 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 248 | loff_t end) |
1da177e4 LT |
249 | { |
250 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
251 | } | |
f4c0a0fd | 252 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 253 | |
485bb99b RD |
254 | /** |
255 | * filemap_flush - mostly a non-blocking flush | |
256 | * @mapping: target address_space | |
257 | * | |
1da177e4 LT |
258 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
259 | * purposes - I/O may not be started against all dirty pages. | |
260 | */ | |
261 | int filemap_flush(struct address_space *mapping) | |
262 | { | |
263 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
264 | } | |
265 | EXPORT_SYMBOL(filemap_flush); | |
266 | ||
485bb99b | 267 | /** |
94004ed7 CH |
268 | * filemap_fdatawait_range - wait for writeback to complete |
269 | * @mapping: address space structure to wait for | |
270 | * @start_byte: offset in bytes where the range starts | |
271 | * @end_byte: offset in bytes where the range ends (inclusive) | |
485bb99b | 272 | * |
94004ed7 CH |
273 | * Walk the list of under-writeback pages of the given address space |
274 | * in the given range and wait for all of them. | |
1da177e4 | 275 | */ |
94004ed7 CH |
276 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
277 | loff_t end_byte) | |
1da177e4 | 278 | { |
94004ed7 CH |
279 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
280 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
281 | struct pagevec pvec; |
282 | int nr_pages; | |
865ffef3 | 283 | int ret2, ret = 0; |
1da177e4 | 284 | |
94004ed7 | 285 | if (end_byte < start_byte) |
865ffef3 | 286 | goto out; |
1da177e4 LT |
287 | |
288 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
289 | while ((index <= end) && |
290 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
291 | PAGECACHE_TAG_WRITEBACK, | |
292 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
293 | unsigned i; | |
294 | ||
295 | for (i = 0; i < nr_pages; i++) { | |
296 | struct page *page = pvec.pages[i]; | |
297 | ||
298 | /* until radix tree lookup accepts end_index */ | |
299 | if (page->index > end) | |
300 | continue; | |
301 | ||
302 | wait_on_page_writeback(page); | |
212260aa | 303 | if (TestClearPageError(page)) |
1da177e4 LT |
304 | ret = -EIO; |
305 | } | |
306 | pagevec_release(&pvec); | |
307 | cond_resched(); | |
308 | } | |
865ffef3 DM |
309 | out: |
310 | ret2 = filemap_check_errors(mapping); | |
311 | if (!ret) | |
312 | ret = ret2; | |
1da177e4 LT |
313 | |
314 | return ret; | |
315 | } | |
d3bccb6f JK |
316 | EXPORT_SYMBOL(filemap_fdatawait_range); |
317 | ||
1da177e4 | 318 | /** |
485bb99b | 319 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 320 | * @mapping: address space structure to wait for |
485bb99b RD |
321 | * |
322 | * Walk the list of under-writeback pages of the given address space | |
323 | * and wait for all of them. | |
1da177e4 LT |
324 | */ |
325 | int filemap_fdatawait(struct address_space *mapping) | |
326 | { | |
327 | loff_t i_size = i_size_read(mapping->host); | |
328 | ||
329 | if (i_size == 0) | |
330 | return 0; | |
331 | ||
94004ed7 | 332 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
333 | } |
334 | EXPORT_SYMBOL(filemap_fdatawait); | |
335 | ||
336 | int filemap_write_and_wait(struct address_space *mapping) | |
337 | { | |
28fd1298 | 338 | int err = 0; |
1da177e4 LT |
339 | |
340 | if (mapping->nrpages) { | |
28fd1298 OH |
341 | err = filemap_fdatawrite(mapping); |
342 | /* | |
343 | * Even if the above returned error, the pages may be | |
344 | * written partially (e.g. -ENOSPC), so we wait for it. | |
345 | * But the -EIO is special case, it may indicate the worst | |
346 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
347 | */ | |
348 | if (err != -EIO) { | |
349 | int err2 = filemap_fdatawait(mapping); | |
350 | if (!err) | |
351 | err = err2; | |
352 | } | |
865ffef3 DM |
353 | } else { |
354 | err = filemap_check_errors(mapping); | |
1da177e4 | 355 | } |
28fd1298 | 356 | return err; |
1da177e4 | 357 | } |
28fd1298 | 358 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 359 | |
485bb99b RD |
360 | /** |
361 | * filemap_write_and_wait_range - write out & wait on a file range | |
362 | * @mapping: the address_space for the pages | |
363 | * @lstart: offset in bytes where the range starts | |
364 | * @lend: offset in bytes where the range ends (inclusive) | |
365 | * | |
469eb4d0 AM |
366 | * Write out and wait upon file offsets lstart->lend, inclusive. |
367 | * | |
368 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
369 | * that this function can be used to write to the very end-of-file (end = -1). | |
370 | */ | |
1da177e4 LT |
371 | int filemap_write_and_wait_range(struct address_space *mapping, |
372 | loff_t lstart, loff_t lend) | |
373 | { | |
28fd1298 | 374 | int err = 0; |
1da177e4 LT |
375 | |
376 | if (mapping->nrpages) { | |
28fd1298 OH |
377 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
378 | WB_SYNC_ALL); | |
379 | /* See comment of filemap_write_and_wait() */ | |
380 | if (err != -EIO) { | |
94004ed7 CH |
381 | int err2 = filemap_fdatawait_range(mapping, |
382 | lstart, lend); | |
28fd1298 OH |
383 | if (!err) |
384 | err = err2; | |
385 | } | |
865ffef3 DM |
386 | } else { |
387 | err = filemap_check_errors(mapping); | |
1da177e4 | 388 | } |
28fd1298 | 389 | return err; |
1da177e4 | 390 | } |
f6995585 | 391 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 392 | |
ef6a3c63 MS |
393 | /** |
394 | * replace_page_cache_page - replace a pagecache page with a new one | |
395 | * @old: page to be replaced | |
396 | * @new: page to replace with | |
397 | * @gfp_mask: allocation mode | |
398 | * | |
399 | * This function replaces a page in the pagecache with a new one. On | |
400 | * success it acquires the pagecache reference for the new page and | |
401 | * drops it for the old page. Both the old and new pages must be | |
402 | * locked. This function does not add the new page to the LRU, the | |
403 | * caller must do that. | |
404 | * | |
405 | * The remove + add is atomic. The only way this function can fail is | |
406 | * memory allocation failure. | |
407 | */ | |
408 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
409 | { | |
410 | int error; | |
ef6a3c63 | 411 | |
309381fe SL |
412 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
413 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
414 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 415 | |
ef6a3c63 MS |
416 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
417 | if (!error) { | |
418 | struct address_space *mapping = old->mapping; | |
419 | void (*freepage)(struct page *); | |
420 | ||
421 | pgoff_t offset = old->index; | |
422 | freepage = mapping->a_ops->freepage; | |
423 | ||
424 | page_cache_get(new); | |
425 | new->mapping = mapping; | |
426 | new->index = offset; | |
427 | ||
428 | spin_lock_irq(&mapping->tree_lock); | |
e64a782f | 429 | __delete_from_page_cache(old); |
ef6a3c63 MS |
430 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
431 | BUG_ON(error); | |
432 | mapping->nrpages++; | |
433 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
434 | if (PageSwapBacked(new)) | |
435 | __inc_zone_page_state(new, NR_SHMEM); | |
436 | spin_unlock_irq(&mapping->tree_lock); | |
ab936cbc KH |
437 | /* mem_cgroup codes must not be called under tree_lock */ |
438 | mem_cgroup_replace_page_cache(old, new); | |
ef6a3c63 MS |
439 | radix_tree_preload_end(); |
440 | if (freepage) | |
441 | freepage(old); | |
442 | page_cache_release(old); | |
ef6a3c63 MS |
443 | } |
444 | ||
445 | return error; | |
446 | } | |
447 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
448 | ||
485bb99b | 449 | /** |
e286781d | 450 | * add_to_page_cache_locked - add a locked page to the pagecache |
485bb99b RD |
451 | * @page: page to add |
452 | * @mapping: the page's address_space | |
453 | * @offset: page index | |
454 | * @gfp_mask: page allocation mode | |
455 | * | |
e286781d | 456 | * This function is used to add a page to the pagecache. It must be locked. |
1da177e4 LT |
457 | * This function does not add the page to the LRU. The caller must do that. |
458 | */ | |
e286781d | 459 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, |
6daa0e28 | 460 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 461 | { |
e286781d NP |
462 | int error; |
463 | ||
309381fe SL |
464 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
465 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d NP |
466 | |
467 | error = mem_cgroup_cache_charge(page, current->mm, | |
2c26fdd7 | 468 | gfp_mask & GFP_RECLAIM_MASK); |
35c754d7 | 469 | if (error) |
66a0c8ee | 470 | return error; |
1da177e4 | 471 | |
5e4c0d97 | 472 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 473 | if (error) { |
69029cd5 | 474 | mem_cgroup_uncharge_cache_page(page); |
66a0c8ee KS |
475 | return error; |
476 | } | |
477 | ||
478 | page_cache_get(page); | |
479 | page->mapping = mapping; | |
480 | page->index = offset; | |
481 | ||
482 | spin_lock_irq(&mapping->tree_lock); | |
483 | error = radix_tree_insert(&mapping->page_tree, offset, page); | |
484 | radix_tree_preload_end(); | |
485 | if (unlikely(error)) | |
486 | goto err_insert; | |
487 | mapping->nrpages++; | |
488 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
489 | spin_unlock_irq(&mapping->tree_lock); | |
490 | trace_mm_filemap_add_to_page_cache(page); | |
491 | return 0; | |
492 | err_insert: | |
493 | page->mapping = NULL; | |
494 | /* Leave page->index set: truncation relies upon it */ | |
495 | spin_unlock_irq(&mapping->tree_lock); | |
496 | mem_cgroup_uncharge_cache_page(page); | |
497 | page_cache_release(page); | |
1da177e4 LT |
498 | return error; |
499 | } | |
e286781d | 500 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
501 | |
502 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 503 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 504 | { |
4f98a2fe RR |
505 | int ret; |
506 | ||
4f98a2fe | 507 | ret = add_to_page_cache(page, mapping, offset, gfp_mask); |
31475dd6 HD |
508 | if (ret == 0) |
509 | lru_cache_add_file(page); | |
1da177e4 LT |
510 | return ret; |
511 | } | |
18bc0bbd | 512 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 513 | |
44110fe3 | 514 | #ifdef CONFIG_NUMA |
2ae88149 | 515 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 516 | { |
c0ff7453 MX |
517 | int n; |
518 | struct page *page; | |
519 | ||
44110fe3 | 520 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
521 | unsigned int cpuset_mems_cookie; |
522 | do { | |
d26914d1 | 523 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 MG |
524 | n = cpuset_mem_spread_node(); |
525 | page = alloc_pages_exact_node(n, gfp, 0); | |
d26914d1 | 526 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 527 | |
c0ff7453 | 528 | return page; |
44110fe3 | 529 | } |
2ae88149 | 530 | return alloc_pages(gfp, 0); |
44110fe3 | 531 | } |
2ae88149 | 532 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
533 | #endif |
534 | ||
1da177e4 LT |
535 | /* |
536 | * In order to wait for pages to become available there must be | |
537 | * waitqueues associated with pages. By using a hash table of | |
538 | * waitqueues where the bucket discipline is to maintain all | |
539 | * waiters on the same queue and wake all when any of the pages | |
540 | * become available, and for the woken contexts to check to be | |
541 | * sure the appropriate page became available, this saves space | |
542 | * at a cost of "thundering herd" phenomena during rare hash | |
543 | * collisions. | |
544 | */ | |
545 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
546 | { | |
547 | const struct zone *zone = page_zone(page); | |
548 | ||
549 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
550 | } | |
551 | ||
552 | static inline void wake_up_page(struct page *page, int bit) | |
553 | { | |
554 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); | |
555 | } | |
556 | ||
920c7a5d | 557 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
558 | { |
559 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
560 | ||
561 | if (test_bit(bit_nr, &page->flags)) | |
7eaceacc | 562 | __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
563 | TASK_UNINTERRUPTIBLE); |
564 | } | |
565 | EXPORT_SYMBOL(wait_on_page_bit); | |
566 | ||
f62e00cc KM |
567 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
568 | { | |
569 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
570 | ||
571 | if (!test_bit(bit_nr, &page->flags)) | |
572 | return 0; | |
573 | ||
574 | return __wait_on_bit(page_waitqueue(page), &wait, | |
575 | sleep_on_page_killable, TASK_KILLABLE); | |
576 | } | |
577 | ||
385e1ca5 DH |
578 | /** |
579 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
580 | * @page: Page defining the wait queue of interest |
581 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
582 | * |
583 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
584 | */ | |
585 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
586 | { | |
587 | wait_queue_head_t *q = page_waitqueue(page); | |
588 | unsigned long flags; | |
589 | ||
590 | spin_lock_irqsave(&q->lock, flags); | |
591 | __add_wait_queue(q, waiter); | |
592 | spin_unlock_irqrestore(&q->lock, flags); | |
593 | } | |
594 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
595 | ||
1da177e4 | 596 | /** |
485bb99b | 597 | * unlock_page - unlock a locked page |
1da177e4 LT |
598 | * @page: the page |
599 | * | |
600 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
601 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
602 | * mechananism between PageLocked pages and PageWriteback pages is shared. | |
603 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. | |
604 | * | |
8413ac9d NP |
605 | * The mb is necessary to enforce ordering between the clear_bit and the read |
606 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 607 | */ |
920c7a5d | 608 | void unlock_page(struct page *page) |
1da177e4 | 609 | { |
309381fe | 610 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
8413ac9d NP |
611 | clear_bit_unlock(PG_locked, &page->flags); |
612 | smp_mb__after_clear_bit(); | |
1da177e4 LT |
613 | wake_up_page(page, PG_locked); |
614 | } | |
615 | EXPORT_SYMBOL(unlock_page); | |
616 | ||
485bb99b RD |
617 | /** |
618 | * end_page_writeback - end writeback against a page | |
619 | * @page: the page | |
1da177e4 LT |
620 | */ |
621 | void end_page_writeback(struct page *page) | |
622 | { | |
ac6aadb2 MS |
623 | if (TestClearPageReclaim(page)) |
624 | rotate_reclaimable_page(page); | |
625 | ||
626 | if (!test_clear_page_writeback(page)) | |
627 | BUG(); | |
628 | ||
1da177e4 LT |
629 | smp_mb__after_clear_bit(); |
630 | wake_up_page(page, PG_writeback); | |
631 | } | |
632 | EXPORT_SYMBOL(end_page_writeback); | |
633 | ||
485bb99b RD |
634 | /** |
635 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
636 | * @page: the page to lock | |
1da177e4 | 637 | */ |
920c7a5d | 638 | void __lock_page(struct page *page) |
1da177e4 LT |
639 | { |
640 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
641 | ||
7eaceacc | 642 | __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
643 | TASK_UNINTERRUPTIBLE); |
644 | } | |
645 | EXPORT_SYMBOL(__lock_page); | |
646 | ||
b5606c2d | 647 | int __lock_page_killable(struct page *page) |
2687a356 MW |
648 | { |
649 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
650 | ||
651 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
7eaceacc | 652 | sleep_on_page_killable, TASK_KILLABLE); |
2687a356 | 653 | } |
18bc0bbd | 654 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 655 | |
d065bd81 ML |
656 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
657 | unsigned int flags) | |
658 | { | |
37b23e05 KM |
659 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
660 | /* | |
661 | * CAUTION! In this case, mmap_sem is not released | |
662 | * even though return 0. | |
663 | */ | |
664 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
665 | return 0; | |
666 | ||
667 | up_read(&mm->mmap_sem); | |
668 | if (flags & FAULT_FLAG_KILLABLE) | |
669 | wait_on_page_locked_killable(page); | |
670 | else | |
318b275f | 671 | wait_on_page_locked(page); |
d065bd81 | 672 | return 0; |
37b23e05 KM |
673 | } else { |
674 | if (flags & FAULT_FLAG_KILLABLE) { | |
675 | int ret; | |
676 | ||
677 | ret = __lock_page_killable(page); | |
678 | if (ret) { | |
679 | up_read(&mm->mmap_sem); | |
680 | return 0; | |
681 | } | |
682 | } else | |
683 | __lock_page(page); | |
684 | return 1; | |
d065bd81 ML |
685 | } |
686 | } | |
687 | ||
e7b563bb JW |
688 | /** |
689 | * page_cache_next_hole - find the next hole (not-present entry) | |
690 | * @mapping: mapping | |
691 | * @index: index | |
692 | * @max_scan: maximum range to search | |
693 | * | |
694 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
695 | * lowest indexed hole. | |
696 | * | |
697 | * Returns: the index of the hole if found, otherwise returns an index | |
698 | * outside of the set specified (in which case 'return - index >= | |
699 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
700 | * be returned. | |
701 | * | |
702 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
703 | * like radix_tree_gang_lookup, this will not atomically search a | |
704 | * snapshot of the tree at a single point in time. For example, if a | |
705 | * hole is created at index 5, then subsequently a hole is created at | |
706 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
707 | * if called under rcu_read_lock. | |
708 | */ | |
709 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
710 | pgoff_t index, unsigned long max_scan) | |
711 | { | |
712 | unsigned long i; | |
713 | ||
714 | for (i = 0; i < max_scan; i++) { | |
715 | if (!radix_tree_lookup(&mapping->page_tree, index)) | |
716 | break; | |
717 | index++; | |
718 | if (index == 0) | |
719 | break; | |
720 | } | |
721 | ||
722 | return index; | |
723 | } | |
724 | EXPORT_SYMBOL(page_cache_next_hole); | |
725 | ||
726 | /** | |
727 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
728 | * @mapping: mapping | |
729 | * @index: index | |
730 | * @max_scan: maximum range to search | |
731 | * | |
732 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
733 | * the first hole. | |
734 | * | |
735 | * Returns: the index of the hole if found, otherwise returns an index | |
736 | * outside of the set specified (in which case 'index - return >= | |
737 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
738 | * will be returned. | |
739 | * | |
740 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
741 | * like radix_tree_gang_lookup, this will not atomically search a | |
742 | * snapshot of the tree at a single point in time. For example, if a | |
743 | * hole is created at index 10, then subsequently a hole is created at | |
744 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
745 | * called under rcu_read_lock. | |
746 | */ | |
747 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
748 | pgoff_t index, unsigned long max_scan) | |
749 | { | |
750 | unsigned long i; | |
751 | ||
752 | for (i = 0; i < max_scan; i++) { | |
753 | if (!radix_tree_lookup(&mapping->page_tree, index)) | |
754 | break; | |
755 | index--; | |
756 | if (index == ULONG_MAX) | |
757 | break; | |
758 | } | |
759 | ||
760 | return index; | |
761 | } | |
762 | EXPORT_SYMBOL(page_cache_prev_hole); | |
763 | ||
485bb99b RD |
764 | /** |
765 | * find_get_page - find and get a page reference | |
766 | * @mapping: the address_space to search | |
767 | * @offset: the page index | |
768 | * | |
da6052f7 NP |
769 | * Is there a pagecache struct page at the given (mapping, offset) tuple? |
770 | * If yes, increment its refcount and return it; if no, return NULL. | |
1da177e4 | 771 | */ |
a60637c8 | 772 | struct page *find_get_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 773 | { |
a60637c8 | 774 | void **pagep; |
1da177e4 LT |
775 | struct page *page; |
776 | ||
a60637c8 NP |
777 | rcu_read_lock(); |
778 | repeat: | |
779 | page = NULL; | |
780 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
781 | if (pagep) { | |
782 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
783 | if (unlikely(!page)) |
784 | goto out; | |
a2c16d6c | 785 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
786 | if (radix_tree_deref_retry(page)) |
787 | goto repeat; | |
788 | /* | |
789 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
790 | * here as an exceptional entry: so return it without | |
791 | * attempting to raise page count. | |
792 | */ | |
793 | goto out; | |
a2c16d6c | 794 | } |
a60637c8 NP |
795 | if (!page_cache_get_speculative(page)) |
796 | goto repeat; | |
797 | ||
798 | /* | |
799 | * Has the page moved? | |
800 | * This is part of the lockless pagecache protocol. See | |
801 | * include/linux/pagemap.h for details. | |
802 | */ | |
803 | if (unlikely(page != *pagep)) { | |
804 | page_cache_release(page); | |
805 | goto repeat; | |
806 | } | |
807 | } | |
27d20fdd | 808 | out: |
a60637c8 NP |
809 | rcu_read_unlock(); |
810 | ||
1da177e4 LT |
811 | return page; |
812 | } | |
1da177e4 LT |
813 | EXPORT_SYMBOL(find_get_page); |
814 | ||
1da177e4 LT |
815 | /** |
816 | * find_lock_page - locate, pin and lock a pagecache page | |
67be2dd1 MW |
817 | * @mapping: the address_space to search |
818 | * @offset: the page index | |
1da177e4 LT |
819 | * |
820 | * Locates the desired pagecache page, locks it, increments its reference | |
821 | * count and returns its address. | |
822 | * | |
823 | * Returns zero if the page was not present. find_lock_page() may sleep. | |
824 | */ | |
a60637c8 | 825 | struct page *find_lock_page(struct address_space *mapping, pgoff_t offset) |
1da177e4 LT |
826 | { |
827 | struct page *page; | |
828 | ||
1da177e4 | 829 | repeat: |
a60637c8 | 830 | page = find_get_page(mapping, offset); |
a2c16d6c | 831 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
832 | lock_page(page); |
833 | /* Has the page been truncated? */ | |
834 | if (unlikely(page->mapping != mapping)) { | |
835 | unlock_page(page); | |
836 | page_cache_release(page); | |
837 | goto repeat; | |
1da177e4 | 838 | } |
309381fe | 839 | VM_BUG_ON_PAGE(page->index != offset, page); |
1da177e4 | 840 | } |
1da177e4 LT |
841 | return page; |
842 | } | |
1da177e4 LT |
843 | EXPORT_SYMBOL(find_lock_page); |
844 | ||
845 | /** | |
846 | * find_or_create_page - locate or add a pagecache page | |
67be2dd1 MW |
847 | * @mapping: the page's address_space |
848 | * @index: the page's index into the mapping | |
849 | * @gfp_mask: page allocation mode | |
1da177e4 LT |
850 | * |
851 | * Locates a page in the pagecache. If the page is not present, a new page | |
852 | * is allocated using @gfp_mask and is added to the pagecache and to the VM's | |
853 | * LRU list. The returned page is locked and has its reference count | |
854 | * incremented. | |
855 | * | |
856 | * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic | |
857 | * allocation! | |
858 | * | |
859 | * find_or_create_page() returns the desired page's address, or zero on | |
860 | * memory exhaustion. | |
861 | */ | |
862 | struct page *find_or_create_page(struct address_space *mapping, | |
57f6b96c | 863 | pgoff_t index, gfp_t gfp_mask) |
1da177e4 | 864 | { |
eb2be189 | 865 | struct page *page; |
1da177e4 LT |
866 | int err; |
867 | repeat: | |
868 | page = find_lock_page(mapping, index); | |
869 | if (!page) { | |
eb2be189 NP |
870 | page = __page_cache_alloc(gfp_mask); |
871 | if (!page) | |
872 | return NULL; | |
67d58ac4 NP |
873 | /* |
874 | * We want a regular kernel memory (not highmem or DMA etc) | |
875 | * allocation for the radix tree nodes, but we need to honour | |
876 | * the context-specific requirements the caller has asked for. | |
877 | * GFP_RECLAIM_MASK collects those requirements. | |
878 | */ | |
879 | err = add_to_page_cache_lru(page, mapping, index, | |
880 | (gfp_mask & GFP_RECLAIM_MASK)); | |
eb2be189 NP |
881 | if (unlikely(err)) { |
882 | page_cache_release(page); | |
883 | page = NULL; | |
884 | if (err == -EEXIST) | |
885 | goto repeat; | |
1da177e4 | 886 | } |
1da177e4 | 887 | } |
1da177e4 LT |
888 | return page; |
889 | } | |
1da177e4 LT |
890 | EXPORT_SYMBOL(find_or_create_page); |
891 | ||
892 | /** | |
893 | * find_get_pages - gang pagecache lookup | |
894 | * @mapping: The address_space to search | |
895 | * @start: The starting page index | |
896 | * @nr_pages: The maximum number of pages | |
897 | * @pages: Where the resulting pages are placed | |
898 | * | |
899 | * find_get_pages() will search for and return a group of up to | |
900 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
901 | * find_get_pages() takes a reference against the returned pages. | |
902 | * | |
903 | * The search returns a group of mapping-contiguous pages with ascending | |
904 | * indexes. There may be holes in the indices due to not-present pages. | |
905 | * | |
906 | * find_get_pages() returns the number of pages which were found. | |
907 | */ | |
908 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
909 | unsigned int nr_pages, struct page **pages) | |
910 | { | |
0fc9d104 KK |
911 | struct radix_tree_iter iter; |
912 | void **slot; | |
913 | unsigned ret = 0; | |
914 | ||
915 | if (unlikely(!nr_pages)) | |
916 | return 0; | |
a60637c8 NP |
917 | |
918 | rcu_read_lock(); | |
919 | restart: | |
0fc9d104 | 920 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
a60637c8 NP |
921 | struct page *page; |
922 | repeat: | |
0fc9d104 | 923 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
924 | if (unlikely(!page)) |
925 | continue; | |
9d8aa4ea | 926 | |
a2c16d6c | 927 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
928 | if (radix_tree_deref_retry(page)) { |
929 | /* | |
930 | * Transient condition which can only trigger | |
931 | * when entry at index 0 moves out of or back | |
932 | * to root: none yet gotten, safe to restart. | |
933 | */ | |
0fc9d104 | 934 | WARN_ON(iter.index); |
8079b1c8 HD |
935 | goto restart; |
936 | } | |
a2c16d6c | 937 | /* |
8079b1c8 HD |
938 | * Otherwise, shmem/tmpfs must be storing a swap entry |
939 | * here as an exceptional entry: so skip over it - | |
940 | * we only reach this from invalidate_mapping_pages(). | |
a2c16d6c | 941 | */ |
8079b1c8 | 942 | continue; |
27d20fdd | 943 | } |
a60637c8 NP |
944 | |
945 | if (!page_cache_get_speculative(page)) | |
946 | goto repeat; | |
947 | ||
948 | /* Has the page moved? */ | |
0fc9d104 | 949 | if (unlikely(page != *slot)) { |
a60637c8 NP |
950 | page_cache_release(page); |
951 | goto repeat; | |
952 | } | |
1da177e4 | 953 | |
a60637c8 | 954 | pages[ret] = page; |
0fc9d104 KK |
955 | if (++ret == nr_pages) |
956 | break; | |
a60637c8 | 957 | } |
5b280c0c | 958 | |
a60637c8 | 959 | rcu_read_unlock(); |
1da177e4 LT |
960 | return ret; |
961 | } | |
962 | ||
ebf43500 JA |
963 | /** |
964 | * find_get_pages_contig - gang contiguous pagecache lookup | |
965 | * @mapping: The address_space to search | |
966 | * @index: The starting page index | |
967 | * @nr_pages: The maximum number of pages | |
968 | * @pages: Where the resulting pages are placed | |
969 | * | |
970 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
971 | * that the returned number of pages are guaranteed to be contiguous. | |
972 | * | |
973 | * find_get_pages_contig() returns the number of pages which were found. | |
974 | */ | |
975 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
976 | unsigned int nr_pages, struct page **pages) | |
977 | { | |
0fc9d104 KK |
978 | struct radix_tree_iter iter; |
979 | void **slot; | |
980 | unsigned int ret = 0; | |
981 | ||
982 | if (unlikely(!nr_pages)) | |
983 | return 0; | |
a60637c8 NP |
984 | |
985 | rcu_read_lock(); | |
986 | restart: | |
0fc9d104 | 987 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
a60637c8 NP |
988 | struct page *page; |
989 | repeat: | |
0fc9d104 KK |
990 | page = radix_tree_deref_slot(slot); |
991 | /* The hole, there no reason to continue */ | |
a60637c8 | 992 | if (unlikely(!page)) |
0fc9d104 | 993 | break; |
9d8aa4ea | 994 | |
a2c16d6c | 995 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
996 | if (radix_tree_deref_retry(page)) { |
997 | /* | |
998 | * Transient condition which can only trigger | |
999 | * when entry at index 0 moves out of or back | |
1000 | * to root: none yet gotten, safe to restart. | |
1001 | */ | |
1002 | goto restart; | |
1003 | } | |
a2c16d6c | 1004 | /* |
8079b1c8 HD |
1005 | * Otherwise, shmem/tmpfs must be storing a swap entry |
1006 | * here as an exceptional entry: so stop looking for | |
1007 | * contiguous pages. | |
a2c16d6c | 1008 | */ |
8079b1c8 | 1009 | break; |
a2c16d6c | 1010 | } |
ebf43500 | 1011 | |
a60637c8 NP |
1012 | if (!page_cache_get_speculative(page)) |
1013 | goto repeat; | |
1014 | ||
1015 | /* Has the page moved? */ | |
0fc9d104 | 1016 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1017 | page_cache_release(page); |
1018 | goto repeat; | |
1019 | } | |
1020 | ||
9cbb4cb2 NP |
1021 | /* |
1022 | * must check mapping and index after taking the ref. | |
1023 | * otherwise we can get both false positives and false | |
1024 | * negatives, which is just confusing to the caller. | |
1025 | */ | |
0fc9d104 | 1026 | if (page->mapping == NULL || page->index != iter.index) { |
9cbb4cb2 NP |
1027 | page_cache_release(page); |
1028 | break; | |
1029 | } | |
1030 | ||
a60637c8 | 1031 | pages[ret] = page; |
0fc9d104 KK |
1032 | if (++ret == nr_pages) |
1033 | break; | |
ebf43500 | 1034 | } |
a60637c8 NP |
1035 | rcu_read_unlock(); |
1036 | return ret; | |
ebf43500 | 1037 | } |
ef71c15c | 1038 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1039 | |
485bb99b RD |
1040 | /** |
1041 | * find_get_pages_tag - find and return pages that match @tag | |
1042 | * @mapping: the address_space to search | |
1043 | * @index: the starting page index | |
1044 | * @tag: the tag index | |
1045 | * @nr_pages: the maximum number of pages | |
1046 | * @pages: where the resulting pages are placed | |
1047 | * | |
1da177e4 | 1048 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1049 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
1050 | */ |
1051 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
1052 | int tag, unsigned int nr_pages, struct page **pages) | |
1053 | { | |
0fc9d104 KK |
1054 | struct radix_tree_iter iter; |
1055 | void **slot; | |
1056 | unsigned ret = 0; | |
1057 | ||
1058 | if (unlikely(!nr_pages)) | |
1059 | return 0; | |
a60637c8 NP |
1060 | |
1061 | rcu_read_lock(); | |
1062 | restart: | |
0fc9d104 KK |
1063 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1064 | &iter, *index, tag) { | |
a60637c8 NP |
1065 | struct page *page; |
1066 | repeat: | |
0fc9d104 | 1067 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1068 | if (unlikely(!page)) |
1069 | continue; | |
9d8aa4ea | 1070 | |
a2c16d6c | 1071 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1072 | if (radix_tree_deref_retry(page)) { |
1073 | /* | |
1074 | * Transient condition which can only trigger | |
1075 | * when entry at index 0 moves out of or back | |
1076 | * to root: none yet gotten, safe to restart. | |
1077 | */ | |
1078 | goto restart; | |
1079 | } | |
a2c16d6c | 1080 | /* |
8079b1c8 HD |
1081 | * This function is never used on a shmem/tmpfs |
1082 | * mapping, so a swap entry won't be found here. | |
a2c16d6c | 1083 | */ |
8079b1c8 | 1084 | BUG(); |
a2c16d6c | 1085 | } |
a60637c8 NP |
1086 | |
1087 | if (!page_cache_get_speculative(page)) | |
1088 | goto repeat; | |
1089 | ||
1090 | /* Has the page moved? */ | |
0fc9d104 | 1091 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1092 | page_cache_release(page); |
1093 | goto repeat; | |
1094 | } | |
1095 | ||
1096 | pages[ret] = page; | |
0fc9d104 KK |
1097 | if (++ret == nr_pages) |
1098 | break; | |
a60637c8 | 1099 | } |
5b280c0c | 1100 | |
a60637c8 | 1101 | rcu_read_unlock(); |
1da177e4 | 1102 | |
1da177e4 LT |
1103 | if (ret) |
1104 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1105 | |
1da177e4 LT |
1106 | return ret; |
1107 | } | |
ef71c15c | 1108 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1109 | |
485bb99b RD |
1110 | /** |
1111 | * grab_cache_page_nowait - returns locked page at given index in given cache | |
1112 | * @mapping: target address_space | |
1113 | * @index: the page index | |
1114 | * | |
72fd4a35 | 1115 | * Same as grab_cache_page(), but do not wait if the page is unavailable. |
1da177e4 LT |
1116 | * This is intended for speculative data generators, where the data can |
1117 | * be regenerated if the page couldn't be grabbed. This routine should | |
1118 | * be safe to call while holding the lock for another page. | |
1119 | * | |
1120 | * Clear __GFP_FS when allocating the page to avoid recursion into the fs | |
1121 | * and deadlock against the caller's locked page. | |
1122 | */ | |
1123 | struct page * | |
57f6b96c | 1124 | grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) |
1da177e4 LT |
1125 | { |
1126 | struct page *page = find_get_page(mapping, index); | |
1da177e4 LT |
1127 | |
1128 | if (page) { | |
529ae9aa | 1129 | if (trylock_page(page)) |
1da177e4 LT |
1130 | return page; |
1131 | page_cache_release(page); | |
1132 | return NULL; | |
1133 | } | |
2ae88149 | 1134 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); |
67d58ac4 | 1135 | if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) { |
1da177e4 LT |
1136 | page_cache_release(page); |
1137 | page = NULL; | |
1138 | } | |
1139 | return page; | |
1140 | } | |
1da177e4 LT |
1141 | EXPORT_SYMBOL(grab_cache_page_nowait); |
1142 | ||
76d42bd9 WF |
1143 | /* |
1144 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1145 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1146 | * | |
1147 | * ---R__________________________________________B__________ | |
1148 | * ^ reading here ^ bad block(assume 4k) | |
1149 | * | |
1150 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1151 | * => failing the whole request => read(R) => read(R+1) => | |
1152 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1153 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1154 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1155 | * | |
1156 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1157 | */ | |
1158 | static void shrink_readahead_size_eio(struct file *filp, | |
1159 | struct file_ra_state *ra) | |
1160 | { | |
76d42bd9 | 1161 | ra->ra_pages /= 4; |
76d42bd9 WF |
1162 | } |
1163 | ||
485bb99b | 1164 | /** |
36e78914 | 1165 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1166 | * @filp: the file to read |
1167 | * @ppos: current file position | |
1168 | * @desc: read_descriptor | |
485bb99b | 1169 | * |
1da177e4 | 1170 | * This is a generic file read routine, and uses the |
485bb99b | 1171 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1172 | * |
1173 | * This is really ugly. But the goto's actually try to clarify some | |
1174 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1175 | */ |
36e78914 | 1176 | static void do_generic_file_read(struct file *filp, loff_t *ppos, |
b77d88d4 | 1177 | read_descriptor_t *desc) |
1da177e4 | 1178 | { |
36e78914 | 1179 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1180 | struct inode *inode = mapping->host; |
36e78914 | 1181 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1182 | pgoff_t index; |
1183 | pgoff_t last_index; | |
1184 | pgoff_t prev_index; | |
1185 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1186 | unsigned int prev_offset; |
1da177e4 | 1187 | int error; |
1da177e4 | 1188 | |
1da177e4 | 1189 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1190 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1191 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
1da177e4 LT |
1192 | last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1193 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1194 | ||
1da177e4 LT |
1195 | for (;;) { |
1196 | struct page *page; | |
57f6b96c | 1197 | pgoff_t end_index; |
a32ea1e1 | 1198 | loff_t isize; |
1da177e4 LT |
1199 | unsigned long nr, ret; |
1200 | ||
1da177e4 | 1201 | cond_resched(); |
1da177e4 LT |
1202 | find_page: |
1203 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1204 | if (!page) { |
cf914a7d | 1205 | page_cache_sync_readahead(mapping, |
7ff81078 | 1206 | ra, filp, |
3ea89ee8 FW |
1207 | index, last_index - index); |
1208 | page = find_get_page(mapping, index); | |
1209 | if (unlikely(page == NULL)) | |
1210 | goto no_cached_page; | |
1211 | } | |
1212 | if (PageReadahead(page)) { | |
cf914a7d | 1213 | page_cache_async_readahead(mapping, |
7ff81078 | 1214 | ra, filp, page, |
3ea89ee8 | 1215 | index, last_index - index); |
1da177e4 | 1216 | } |
8ab22b9a HH |
1217 | if (!PageUptodate(page)) { |
1218 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1219 | !mapping->a_ops->is_partially_uptodate) | |
1220 | goto page_not_up_to_date; | |
529ae9aa | 1221 | if (!trylock_page(page)) |
8ab22b9a | 1222 | goto page_not_up_to_date; |
8d056cb9 DH |
1223 | /* Did it get truncated before we got the lock? */ |
1224 | if (!page->mapping) | |
1225 | goto page_not_up_to_date_locked; | |
8ab22b9a HH |
1226 | if (!mapping->a_ops->is_partially_uptodate(page, |
1227 | desc, offset)) | |
1228 | goto page_not_up_to_date_locked; | |
1229 | unlock_page(page); | |
1230 | } | |
1da177e4 | 1231 | page_ok: |
a32ea1e1 N |
1232 | /* |
1233 | * i_size must be checked after we know the page is Uptodate. | |
1234 | * | |
1235 | * Checking i_size after the check allows us to calculate | |
1236 | * the correct value for "nr", which means the zero-filled | |
1237 | * part of the page is not copied back to userspace (unless | |
1238 | * another truncate extends the file - this is desired though). | |
1239 | */ | |
1240 | ||
1241 | isize = i_size_read(inode); | |
1242 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1243 | if (unlikely(!isize || index > end_index)) { | |
1244 | page_cache_release(page); | |
1245 | goto out; | |
1246 | } | |
1247 | ||
1248 | /* nr is the maximum number of bytes to copy from this page */ | |
1249 | nr = PAGE_CACHE_SIZE; | |
1250 | if (index == end_index) { | |
1251 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1252 | if (nr <= offset) { | |
1253 | page_cache_release(page); | |
1254 | goto out; | |
1255 | } | |
1256 | } | |
1257 | nr = nr - offset; | |
1da177e4 LT |
1258 | |
1259 | /* If users can be writing to this page using arbitrary | |
1260 | * virtual addresses, take care about potential aliasing | |
1261 | * before reading the page on the kernel side. | |
1262 | */ | |
1263 | if (mapping_writably_mapped(mapping)) | |
1264 | flush_dcache_page(page); | |
1265 | ||
1266 | /* | |
ec0f1637 JK |
1267 | * When a sequential read accesses a page several times, |
1268 | * only mark it as accessed the first time. | |
1da177e4 | 1269 | */ |
ec0f1637 | 1270 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1271 | mark_page_accessed(page); |
1272 | prev_index = index; | |
1273 | ||
1274 | /* | |
1275 | * Ok, we have the page, and it's up-to-date, so | |
1276 | * now we can copy it to user space... | |
1277 | * | |
b77d88d4 KS |
1278 | * The file_read_actor routine returns how many bytes were |
1279 | * actually used.. | |
1da177e4 LT |
1280 | * NOTE! This may not be the same as how much of a user buffer |
1281 | * we filled up (we may be padding etc), so we can only update | |
1282 | * "pos" here (the actor routine has to update the user buffer | |
1283 | * pointers and the remaining count). | |
1284 | */ | |
b77d88d4 | 1285 | ret = file_read_actor(desc, page, offset, nr); |
1da177e4 LT |
1286 | offset += ret; |
1287 | index += offset >> PAGE_CACHE_SHIFT; | |
1288 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1289 | prev_offset = offset; |
1da177e4 LT |
1290 | |
1291 | page_cache_release(page); | |
1292 | if (ret == nr && desc->count) | |
1293 | continue; | |
1294 | goto out; | |
1295 | ||
1296 | page_not_up_to_date: | |
1297 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1298 | error = lock_page_killable(page); |
1299 | if (unlikely(error)) | |
1300 | goto readpage_error; | |
1da177e4 | 1301 | |
8ab22b9a | 1302 | page_not_up_to_date_locked: |
da6052f7 | 1303 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1304 | if (!page->mapping) { |
1305 | unlock_page(page); | |
1306 | page_cache_release(page); | |
1307 | continue; | |
1308 | } | |
1309 | ||
1310 | /* Did somebody else fill it already? */ | |
1311 | if (PageUptodate(page)) { | |
1312 | unlock_page(page); | |
1313 | goto page_ok; | |
1314 | } | |
1315 | ||
1316 | readpage: | |
91803b49 JM |
1317 | /* |
1318 | * A previous I/O error may have been due to temporary | |
1319 | * failures, eg. multipath errors. | |
1320 | * PG_error will be set again if readpage fails. | |
1321 | */ | |
1322 | ClearPageError(page); | |
1da177e4 LT |
1323 | /* Start the actual read. The read will unlock the page. */ |
1324 | error = mapping->a_ops->readpage(filp, page); | |
1325 | ||
994fc28c ZB |
1326 | if (unlikely(error)) { |
1327 | if (error == AOP_TRUNCATED_PAGE) { | |
1328 | page_cache_release(page); | |
1329 | goto find_page; | |
1330 | } | |
1da177e4 | 1331 | goto readpage_error; |
994fc28c | 1332 | } |
1da177e4 LT |
1333 | |
1334 | if (!PageUptodate(page)) { | |
85462323 ON |
1335 | error = lock_page_killable(page); |
1336 | if (unlikely(error)) | |
1337 | goto readpage_error; | |
1da177e4 LT |
1338 | if (!PageUptodate(page)) { |
1339 | if (page->mapping == NULL) { | |
1340 | /* | |
2ecdc82e | 1341 | * invalidate_mapping_pages got it |
1da177e4 LT |
1342 | */ |
1343 | unlock_page(page); | |
1344 | page_cache_release(page); | |
1345 | goto find_page; | |
1346 | } | |
1347 | unlock_page(page); | |
7ff81078 | 1348 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1349 | error = -EIO; |
1350 | goto readpage_error; | |
1da177e4 LT |
1351 | } |
1352 | unlock_page(page); | |
1353 | } | |
1354 | ||
1da177e4 LT |
1355 | goto page_ok; |
1356 | ||
1357 | readpage_error: | |
1358 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1359 | desc->error = error; | |
1360 | page_cache_release(page); | |
1361 | goto out; | |
1362 | ||
1363 | no_cached_page: | |
1364 | /* | |
1365 | * Ok, it wasn't cached, so we need to create a new | |
1366 | * page.. | |
1367 | */ | |
eb2be189 NP |
1368 | page = page_cache_alloc_cold(mapping); |
1369 | if (!page) { | |
1370 | desc->error = -ENOMEM; | |
1371 | goto out; | |
1da177e4 | 1372 | } |
eb2be189 | 1373 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1374 | index, GFP_KERNEL); |
1375 | if (error) { | |
eb2be189 | 1376 | page_cache_release(page); |
1da177e4 LT |
1377 | if (error == -EEXIST) |
1378 | goto find_page; | |
1379 | desc->error = error; | |
1380 | goto out; | |
1381 | } | |
1da177e4 LT |
1382 | goto readpage; |
1383 | } | |
1384 | ||
1385 | out: | |
7ff81078 FW |
1386 | ra->prev_pos = prev_index; |
1387 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1388 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1389 | |
f4e6b498 | 1390 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1391 | file_accessed(filp); |
1da177e4 | 1392 | } |
1da177e4 LT |
1393 | |
1394 | int file_read_actor(read_descriptor_t *desc, struct page *page, | |
1395 | unsigned long offset, unsigned long size) | |
1396 | { | |
1397 | char *kaddr; | |
1398 | unsigned long left, count = desc->count; | |
1399 | ||
1400 | if (size > count) | |
1401 | size = count; | |
1402 | ||
1403 | /* | |
1404 | * Faults on the destination of a read are common, so do it before | |
1405 | * taking the kmap. | |
1406 | */ | |
1407 | if (!fault_in_pages_writeable(desc->arg.buf, size)) { | |
9b04c5fe | 1408 | kaddr = kmap_atomic(page); |
1da177e4 LT |
1409 | left = __copy_to_user_inatomic(desc->arg.buf, |
1410 | kaddr + offset, size); | |
9b04c5fe | 1411 | kunmap_atomic(kaddr); |
1da177e4 LT |
1412 | if (left == 0) |
1413 | goto success; | |
1414 | } | |
1415 | ||
1416 | /* Do it the slow way */ | |
1417 | kaddr = kmap(page); | |
1418 | left = __copy_to_user(desc->arg.buf, kaddr + offset, size); | |
1419 | kunmap(page); | |
1420 | ||
1421 | if (left) { | |
1422 | size -= left; | |
1423 | desc->error = -EFAULT; | |
1424 | } | |
1425 | success: | |
1426 | desc->count = count - size; | |
1427 | desc->written += size; | |
1428 | desc->arg.buf += size; | |
1429 | return size; | |
1430 | } | |
1431 | ||
0ceb3314 DM |
1432 | /* |
1433 | * Performs necessary checks before doing a write | |
1434 | * @iov: io vector request | |
1435 | * @nr_segs: number of segments in the iovec | |
1436 | * @count: number of bytes to write | |
1437 | * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE | |
1438 | * | |
1439 | * Adjust number of segments and amount of bytes to write (nr_segs should be | |
1440 | * properly initialized first). Returns appropriate error code that caller | |
1441 | * should return or zero in case that write should be allowed. | |
1442 | */ | |
1443 | int generic_segment_checks(const struct iovec *iov, | |
1444 | unsigned long *nr_segs, size_t *count, int access_flags) | |
1445 | { | |
1446 | unsigned long seg; | |
1447 | size_t cnt = 0; | |
1448 | for (seg = 0; seg < *nr_segs; seg++) { | |
1449 | const struct iovec *iv = &iov[seg]; | |
1450 | ||
1451 | /* | |
1452 | * If any segment has a negative length, or the cumulative | |
1453 | * length ever wraps negative then return -EINVAL. | |
1454 | */ | |
1455 | cnt += iv->iov_len; | |
1456 | if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) | |
1457 | return -EINVAL; | |
1458 | if (access_ok(access_flags, iv->iov_base, iv->iov_len)) | |
1459 | continue; | |
1460 | if (seg == 0) | |
1461 | return -EFAULT; | |
1462 | *nr_segs = seg; | |
1463 | cnt -= iv->iov_len; /* This segment is no good */ | |
1464 | break; | |
1465 | } | |
1466 | *count = cnt; | |
1467 | return 0; | |
1468 | } | |
1469 | EXPORT_SYMBOL(generic_segment_checks); | |
1470 | ||
485bb99b | 1471 | /** |
b2abacf3 | 1472 | * generic_file_aio_read - generic filesystem read routine |
485bb99b RD |
1473 | * @iocb: kernel I/O control block |
1474 | * @iov: io vector request | |
1475 | * @nr_segs: number of segments in the iovec | |
b2abacf3 | 1476 | * @pos: current file position |
485bb99b | 1477 | * |
1da177e4 LT |
1478 | * This is the "read()" routine for all filesystems |
1479 | * that can use the page cache directly. | |
1480 | */ | |
1481 | ssize_t | |
543ade1f BP |
1482 | generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
1483 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
1484 | { |
1485 | struct file *filp = iocb->ki_filp; | |
1486 | ssize_t retval; | |
66f998f6 | 1487 | unsigned long seg = 0; |
1da177e4 | 1488 | size_t count; |
543ade1f | 1489 | loff_t *ppos = &iocb->ki_pos; |
1da177e4 LT |
1490 | |
1491 | count = 0; | |
0ceb3314 DM |
1492 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); |
1493 | if (retval) | |
1494 | return retval; | |
1da177e4 LT |
1495 | |
1496 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
1497 | if (filp->f_flags & O_DIRECT) { | |
543ade1f | 1498 | loff_t size; |
1da177e4 LT |
1499 | struct address_space *mapping; |
1500 | struct inode *inode; | |
1501 | ||
1502 | mapping = filp->f_mapping; | |
1503 | inode = mapping->host; | |
1da177e4 LT |
1504 | if (!count) |
1505 | goto out; /* skip atime */ | |
1506 | size = i_size_read(inode); | |
9fe55eea | 1507 | retval = filemap_write_and_wait_range(mapping, pos, |
48b47c56 | 1508 | pos + iov_length(iov, nr_segs) - 1); |
9fe55eea SW |
1509 | if (!retval) { |
1510 | retval = mapping->a_ops->direct_IO(READ, iocb, | |
1511 | iov, pos, nr_segs); | |
1512 | } | |
1513 | if (retval > 0) { | |
1514 | *ppos = pos + retval; | |
1515 | count -= retval; | |
1516 | } | |
66f998f6 | 1517 | |
9fe55eea SW |
1518 | /* |
1519 | * Btrfs can have a short DIO read if we encounter | |
1520 | * compressed extents, so if there was an error, or if | |
1521 | * we've already read everything we wanted to, or if | |
1522 | * there was a short read because we hit EOF, go ahead | |
1523 | * and return. Otherwise fallthrough to buffered io for | |
1524 | * the rest of the read. | |
1525 | */ | |
1526 | if (retval < 0 || !count || *ppos >= size) { | |
1527 | file_accessed(filp); | |
1528 | goto out; | |
0e0bcae3 | 1529 | } |
1da177e4 LT |
1530 | } |
1531 | ||
66f998f6 | 1532 | count = retval; |
11fa977e HD |
1533 | for (seg = 0; seg < nr_segs; seg++) { |
1534 | read_descriptor_t desc; | |
66f998f6 JB |
1535 | loff_t offset = 0; |
1536 | ||
1537 | /* | |
1538 | * If we did a short DIO read we need to skip the section of the | |
1539 | * iov that we've already read data into. | |
1540 | */ | |
1541 | if (count) { | |
1542 | if (count > iov[seg].iov_len) { | |
1543 | count -= iov[seg].iov_len; | |
1544 | continue; | |
1545 | } | |
1546 | offset = count; | |
1547 | count = 0; | |
1548 | } | |
1da177e4 | 1549 | |
11fa977e | 1550 | desc.written = 0; |
66f998f6 JB |
1551 | desc.arg.buf = iov[seg].iov_base + offset; |
1552 | desc.count = iov[seg].iov_len - offset; | |
11fa977e HD |
1553 | if (desc.count == 0) |
1554 | continue; | |
1555 | desc.error = 0; | |
b77d88d4 | 1556 | do_generic_file_read(filp, ppos, &desc); |
11fa977e HD |
1557 | retval += desc.written; |
1558 | if (desc.error) { | |
1559 | retval = retval ?: desc.error; | |
1560 | break; | |
1da177e4 | 1561 | } |
11fa977e HD |
1562 | if (desc.count > 0) |
1563 | break; | |
1da177e4 LT |
1564 | } |
1565 | out: | |
1566 | return retval; | |
1567 | } | |
1da177e4 LT |
1568 | EXPORT_SYMBOL(generic_file_aio_read); |
1569 | ||
1da177e4 | 1570 | #ifdef CONFIG_MMU |
485bb99b RD |
1571 | /** |
1572 | * page_cache_read - adds requested page to the page cache if not already there | |
1573 | * @file: file to read | |
1574 | * @offset: page index | |
1575 | * | |
1da177e4 LT |
1576 | * This adds the requested page to the page cache if it isn't already there, |
1577 | * and schedules an I/O to read in its contents from disk. | |
1578 | */ | |
920c7a5d | 1579 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1580 | { |
1581 | struct address_space *mapping = file->f_mapping; | |
1582 | struct page *page; | |
994fc28c | 1583 | int ret; |
1da177e4 | 1584 | |
994fc28c ZB |
1585 | do { |
1586 | page = page_cache_alloc_cold(mapping); | |
1587 | if (!page) | |
1588 | return -ENOMEM; | |
1589 | ||
1590 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); | |
1591 | if (ret == 0) | |
1592 | ret = mapping->a_ops->readpage(file, page); | |
1593 | else if (ret == -EEXIST) | |
1594 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1595 | |
1da177e4 | 1596 | page_cache_release(page); |
1da177e4 | 1597 | |
994fc28c ZB |
1598 | } while (ret == AOP_TRUNCATED_PAGE); |
1599 | ||
1600 | return ret; | |
1da177e4 LT |
1601 | } |
1602 | ||
1603 | #define MMAP_LOTSAMISS (100) | |
1604 | ||
ef00e08e LT |
1605 | /* |
1606 | * Synchronous readahead happens when we don't even find | |
1607 | * a page in the page cache at all. | |
1608 | */ | |
1609 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1610 | struct file_ra_state *ra, | |
1611 | struct file *file, | |
1612 | pgoff_t offset) | |
1613 | { | |
1614 | unsigned long ra_pages; | |
1615 | struct address_space *mapping = file->f_mapping; | |
1616 | ||
1617 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1618 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 1619 | return; |
275b12bf WF |
1620 | if (!ra->ra_pages) |
1621 | return; | |
ef00e08e | 1622 | |
64363aad | 1623 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
1624 | page_cache_sync_readahead(mapping, ra, file, offset, |
1625 | ra->ra_pages); | |
ef00e08e LT |
1626 | return; |
1627 | } | |
1628 | ||
207d04ba AK |
1629 | /* Avoid banging the cache line if not needed */ |
1630 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1631 | ra->mmap_miss++; |
1632 | ||
1633 | /* | |
1634 | * Do we miss much more than hit in this file? If so, | |
1635 | * stop bothering with read-ahead. It will only hurt. | |
1636 | */ | |
1637 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1638 | return; | |
1639 | ||
d30a1100 WF |
1640 | /* |
1641 | * mmap read-around | |
1642 | */ | |
ef00e08e | 1643 | ra_pages = max_sane_readahead(ra->ra_pages); |
275b12bf WF |
1644 | ra->start = max_t(long, 0, offset - ra_pages / 2); |
1645 | ra->size = ra_pages; | |
2cbea1d3 | 1646 | ra->async_size = ra_pages / 4; |
275b12bf | 1647 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1648 | } |
1649 | ||
1650 | /* | |
1651 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1652 | * so we want to possibly extend the readahead further.. | |
1653 | */ | |
1654 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1655 | struct file_ra_state *ra, | |
1656 | struct file *file, | |
1657 | struct page *page, | |
1658 | pgoff_t offset) | |
1659 | { | |
1660 | struct address_space *mapping = file->f_mapping; | |
1661 | ||
1662 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1663 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
1664 | return; |
1665 | if (ra->mmap_miss > 0) | |
1666 | ra->mmap_miss--; | |
1667 | if (PageReadahead(page)) | |
2fad6f5d WF |
1668 | page_cache_async_readahead(mapping, ra, file, |
1669 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1670 | } |
1671 | ||
485bb99b | 1672 | /** |
54cb8821 | 1673 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1674 | * @vma: vma in which the fault was taken |
1675 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1676 | * |
54cb8821 | 1677 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1678 | * mapped memory region to read in file data during a page fault. |
1679 | * | |
1680 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1681 | * it in the page cache, and handles the special cases reasonably without | |
1682 | * having a lot of duplicated code. | |
1683 | */ | |
d0217ac0 | 1684 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1685 | { |
1686 | int error; | |
54cb8821 | 1687 | struct file *file = vma->vm_file; |
1da177e4 LT |
1688 | struct address_space *mapping = file->f_mapping; |
1689 | struct file_ra_state *ra = &file->f_ra; | |
1690 | struct inode *inode = mapping->host; | |
ef00e08e | 1691 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 1692 | struct page *page; |
2004dc8e | 1693 | pgoff_t size; |
83c54070 | 1694 | int ret = 0; |
1da177e4 | 1695 | |
1da177e4 | 1696 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1697 | if (offset >= size) |
5307cc1a | 1698 | return VM_FAULT_SIGBUS; |
1da177e4 | 1699 | |
1da177e4 | 1700 | /* |
49426420 | 1701 | * Do we have something in the page cache already? |
1da177e4 | 1702 | */ |
ef00e08e | 1703 | page = find_get_page(mapping, offset); |
45cac65b | 1704 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 1705 | /* |
ef00e08e LT |
1706 | * We found the page, so try async readahead before |
1707 | * waiting for the lock. | |
1da177e4 | 1708 | */ |
ef00e08e | 1709 | do_async_mmap_readahead(vma, ra, file, page, offset); |
45cac65b | 1710 | } else if (!page) { |
ef00e08e LT |
1711 | /* No page in the page cache at all */ |
1712 | do_sync_mmap_readahead(vma, ra, file, offset); | |
1713 | count_vm_event(PGMAJFAULT); | |
456f998e | 1714 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
1715 | ret = VM_FAULT_MAJOR; |
1716 | retry_find: | |
b522c94d | 1717 | page = find_get_page(mapping, offset); |
1da177e4 LT |
1718 | if (!page) |
1719 | goto no_cached_page; | |
1720 | } | |
1721 | ||
d88c0922 ML |
1722 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
1723 | page_cache_release(page); | |
d065bd81 | 1724 | return ret | VM_FAULT_RETRY; |
d88c0922 | 1725 | } |
b522c94d ML |
1726 | |
1727 | /* Did it get truncated? */ | |
1728 | if (unlikely(page->mapping != mapping)) { | |
1729 | unlock_page(page); | |
1730 | put_page(page); | |
1731 | goto retry_find; | |
1732 | } | |
309381fe | 1733 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 1734 | |
1da177e4 | 1735 | /* |
d00806b1 NP |
1736 | * We have a locked page in the page cache, now we need to check |
1737 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1738 | */ |
d00806b1 | 1739 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1740 | goto page_not_uptodate; |
1741 | ||
ef00e08e LT |
1742 | /* |
1743 | * Found the page and have a reference on it. | |
1744 | * We must recheck i_size under page lock. | |
1745 | */ | |
d00806b1 | 1746 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1747 | if (unlikely(offset >= size)) { |
d00806b1 | 1748 | unlock_page(page); |
745ad48e | 1749 | page_cache_release(page); |
5307cc1a | 1750 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1751 | } |
1752 | ||
d0217ac0 | 1753 | vmf->page = page; |
83c54070 | 1754 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 1755 | |
1da177e4 LT |
1756 | no_cached_page: |
1757 | /* | |
1758 | * We're only likely to ever get here if MADV_RANDOM is in | |
1759 | * effect. | |
1760 | */ | |
ef00e08e | 1761 | error = page_cache_read(file, offset); |
1da177e4 LT |
1762 | |
1763 | /* | |
1764 | * The page we want has now been added to the page cache. | |
1765 | * In the unlikely event that someone removed it in the | |
1766 | * meantime, we'll just come back here and read it again. | |
1767 | */ | |
1768 | if (error >= 0) | |
1769 | goto retry_find; | |
1770 | ||
1771 | /* | |
1772 | * An error return from page_cache_read can result if the | |
1773 | * system is low on memory, or a problem occurs while trying | |
1774 | * to schedule I/O. | |
1775 | */ | |
1776 | if (error == -ENOMEM) | |
d0217ac0 NP |
1777 | return VM_FAULT_OOM; |
1778 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1779 | |
1780 | page_not_uptodate: | |
1da177e4 LT |
1781 | /* |
1782 | * Umm, take care of errors if the page isn't up-to-date. | |
1783 | * Try to re-read it _once_. We do this synchronously, | |
1784 | * because there really aren't any performance issues here | |
1785 | * and we need to check for errors. | |
1786 | */ | |
1da177e4 | 1787 | ClearPageError(page); |
994fc28c | 1788 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
1789 | if (!error) { |
1790 | wait_on_page_locked(page); | |
1791 | if (!PageUptodate(page)) | |
1792 | error = -EIO; | |
1793 | } | |
d00806b1 NP |
1794 | page_cache_release(page); |
1795 | ||
1796 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 1797 | goto retry_find; |
1da177e4 | 1798 | |
d00806b1 | 1799 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 1800 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 1801 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
1802 | } |
1803 | EXPORT_SYMBOL(filemap_fault); | |
1804 | ||
4fcf1c62 JK |
1805 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
1806 | { | |
1807 | struct page *page = vmf->page; | |
496ad9aa | 1808 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
1809 | int ret = VM_FAULT_LOCKED; |
1810 | ||
14da9200 | 1811 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
1812 | file_update_time(vma->vm_file); |
1813 | lock_page(page); | |
1814 | if (page->mapping != inode->i_mapping) { | |
1815 | unlock_page(page); | |
1816 | ret = VM_FAULT_NOPAGE; | |
1817 | goto out; | |
1818 | } | |
14da9200 JK |
1819 | /* |
1820 | * We mark the page dirty already here so that when freeze is in | |
1821 | * progress, we are guaranteed that writeback during freezing will | |
1822 | * see the dirty page and writeprotect it again. | |
1823 | */ | |
1824 | set_page_dirty(page); | |
1d1d1a76 | 1825 | wait_for_stable_page(page); |
4fcf1c62 | 1826 | out: |
14da9200 | 1827 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
1828 | return ret; |
1829 | } | |
1830 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
1831 | ||
f0f37e2f | 1832 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 1833 | .fault = filemap_fault, |
4fcf1c62 | 1834 | .page_mkwrite = filemap_page_mkwrite, |
0b173bc4 | 1835 | .remap_pages = generic_file_remap_pages, |
1da177e4 LT |
1836 | }; |
1837 | ||
1838 | /* This is used for a general mmap of a disk file */ | |
1839 | ||
1840 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1841 | { | |
1842 | struct address_space *mapping = file->f_mapping; | |
1843 | ||
1844 | if (!mapping->a_ops->readpage) | |
1845 | return -ENOEXEC; | |
1846 | file_accessed(file); | |
1847 | vma->vm_ops = &generic_file_vm_ops; | |
1848 | return 0; | |
1849 | } | |
1da177e4 LT |
1850 | |
1851 | /* | |
1852 | * This is for filesystems which do not implement ->writepage. | |
1853 | */ | |
1854 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
1855 | { | |
1856 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
1857 | return -EINVAL; | |
1858 | return generic_file_mmap(file, vma); | |
1859 | } | |
1860 | #else | |
1861 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
1862 | { | |
1863 | return -ENOSYS; | |
1864 | } | |
1865 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
1866 | { | |
1867 | return -ENOSYS; | |
1868 | } | |
1869 | #endif /* CONFIG_MMU */ | |
1870 | ||
1871 | EXPORT_SYMBOL(generic_file_mmap); | |
1872 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
1873 | ||
6fe6900e | 1874 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 1875 | pgoff_t index, |
5e5358e7 | 1876 | int (*filler)(void *, struct page *), |
0531b2aa LT |
1877 | void *data, |
1878 | gfp_t gfp) | |
1da177e4 | 1879 | { |
eb2be189 | 1880 | struct page *page; |
1da177e4 LT |
1881 | int err; |
1882 | repeat: | |
1883 | page = find_get_page(mapping, index); | |
1884 | if (!page) { | |
0531b2aa | 1885 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
1886 | if (!page) |
1887 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 1888 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 NP |
1889 | if (unlikely(err)) { |
1890 | page_cache_release(page); | |
1891 | if (err == -EEXIST) | |
1892 | goto repeat; | |
1da177e4 | 1893 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
1894 | return ERR_PTR(err); |
1895 | } | |
1da177e4 LT |
1896 | err = filler(data, page); |
1897 | if (err < 0) { | |
1898 | page_cache_release(page); | |
1899 | page = ERR_PTR(err); | |
1900 | } | |
1901 | } | |
1da177e4 LT |
1902 | return page; |
1903 | } | |
1904 | ||
0531b2aa | 1905 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 1906 | pgoff_t index, |
5e5358e7 | 1907 | int (*filler)(void *, struct page *), |
0531b2aa LT |
1908 | void *data, |
1909 | gfp_t gfp) | |
1910 | ||
1da177e4 LT |
1911 | { |
1912 | struct page *page; | |
1913 | int err; | |
1914 | ||
1915 | retry: | |
0531b2aa | 1916 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 1917 | if (IS_ERR(page)) |
c855ff37 | 1918 | return page; |
1da177e4 LT |
1919 | if (PageUptodate(page)) |
1920 | goto out; | |
1921 | ||
1922 | lock_page(page); | |
1923 | if (!page->mapping) { | |
1924 | unlock_page(page); | |
1925 | page_cache_release(page); | |
1926 | goto retry; | |
1927 | } | |
1928 | if (PageUptodate(page)) { | |
1929 | unlock_page(page); | |
1930 | goto out; | |
1931 | } | |
1932 | err = filler(data, page); | |
1933 | if (err < 0) { | |
1934 | page_cache_release(page); | |
c855ff37 | 1935 | return ERR_PTR(err); |
1da177e4 | 1936 | } |
c855ff37 | 1937 | out: |
6fe6900e NP |
1938 | mark_page_accessed(page); |
1939 | return page; | |
1940 | } | |
0531b2aa LT |
1941 | |
1942 | /** | |
1943 | * read_cache_page_async - read into page cache, fill it if needed | |
1944 | * @mapping: the page's address_space | |
1945 | * @index: the page index | |
1946 | * @filler: function to perform the read | |
5e5358e7 | 1947 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa LT |
1948 | * |
1949 | * Same as read_cache_page, but don't wait for page to become unlocked | |
1950 | * after submitting it to the filler. | |
1951 | * | |
1952 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
1953 | * not set, try to fill the page but don't wait for it to become unlocked. | |
1954 | * | |
1955 | * If the page does not get brought uptodate, return -EIO. | |
1956 | */ | |
1957 | struct page *read_cache_page_async(struct address_space *mapping, | |
1958 | pgoff_t index, | |
5e5358e7 | 1959 | int (*filler)(void *, struct page *), |
0531b2aa LT |
1960 | void *data) |
1961 | { | |
1962 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
1963 | } | |
6fe6900e NP |
1964 | EXPORT_SYMBOL(read_cache_page_async); |
1965 | ||
0531b2aa LT |
1966 | static struct page *wait_on_page_read(struct page *page) |
1967 | { | |
1968 | if (!IS_ERR(page)) { | |
1969 | wait_on_page_locked(page); | |
1970 | if (!PageUptodate(page)) { | |
1971 | page_cache_release(page); | |
1972 | page = ERR_PTR(-EIO); | |
1973 | } | |
1974 | } | |
1975 | return page; | |
1976 | } | |
1977 | ||
1978 | /** | |
1979 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
1980 | * @mapping: the page's address_space | |
1981 | * @index: the page index | |
1982 | * @gfp: the page allocator flags to use if allocating | |
1983 | * | |
1984 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 1985 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
1986 | * |
1987 | * If the page does not get brought uptodate, return -EIO. | |
1988 | */ | |
1989 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
1990 | pgoff_t index, | |
1991 | gfp_t gfp) | |
1992 | { | |
1993 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
1994 | ||
1995 | return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp)); | |
1996 | } | |
1997 | EXPORT_SYMBOL(read_cache_page_gfp); | |
1998 | ||
6fe6900e NP |
1999 | /** |
2000 | * read_cache_page - read into page cache, fill it if needed | |
2001 | * @mapping: the page's address_space | |
2002 | * @index: the page index | |
2003 | * @filler: function to perform the read | |
5e5358e7 | 2004 | * @data: first arg to filler(data, page) function, often left as NULL |
6fe6900e NP |
2005 | * |
2006 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
2007 | * not set, try to fill the page then wait for it to become unlocked. | |
2008 | * | |
2009 | * If the page does not get brought uptodate, return -EIO. | |
2010 | */ | |
2011 | struct page *read_cache_page(struct address_space *mapping, | |
57f6b96c | 2012 | pgoff_t index, |
5e5358e7 | 2013 | int (*filler)(void *, struct page *), |
6fe6900e NP |
2014 | void *data) |
2015 | { | |
0531b2aa | 2016 | return wait_on_page_read(read_cache_page_async(mapping, index, filler, data)); |
1da177e4 | 2017 | } |
1da177e4 LT |
2018 | EXPORT_SYMBOL(read_cache_page); |
2019 | ||
2f718ffc | 2020 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
1da177e4 LT |
2021 | const struct iovec *iov, size_t base, size_t bytes) |
2022 | { | |
f1800536 | 2023 | size_t copied = 0, left = 0; |
1da177e4 LT |
2024 | |
2025 | while (bytes) { | |
2026 | char __user *buf = iov->iov_base + base; | |
2027 | int copy = min(bytes, iov->iov_len - base); | |
2028 | ||
2029 | base = 0; | |
f1800536 | 2030 | left = __copy_from_user_inatomic(vaddr, buf, copy); |
1da177e4 LT |
2031 | copied += copy; |
2032 | bytes -= copy; | |
2033 | vaddr += copy; | |
2034 | iov++; | |
2035 | ||
01408c49 | 2036 | if (unlikely(left)) |
1da177e4 | 2037 | break; |
1da177e4 LT |
2038 | } |
2039 | return copied - left; | |
2040 | } | |
2041 | ||
2f718ffc NP |
2042 | /* |
2043 | * Copy as much as we can into the page and return the number of bytes which | |
af901ca1 | 2044 | * were successfully copied. If a fault is encountered then return the number of |
2f718ffc NP |
2045 | * bytes which were copied. |
2046 | */ | |
2047 | size_t iov_iter_copy_from_user_atomic(struct page *page, | |
2048 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
2049 | { | |
2050 | char *kaddr; | |
2051 | size_t copied; | |
2052 | ||
2053 | BUG_ON(!in_atomic()); | |
9b04c5fe | 2054 | kaddr = kmap_atomic(page); |
2f718ffc NP |
2055 | if (likely(i->nr_segs == 1)) { |
2056 | int left; | |
2057 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 2058 | left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); |
2f718ffc NP |
2059 | copied = bytes - left; |
2060 | } else { | |
2061 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2062 | i->iov, i->iov_offset, bytes); | |
2063 | } | |
9b04c5fe | 2064 | kunmap_atomic(kaddr); |
2f718ffc NP |
2065 | |
2066 | return copied; | |
2067 | } | |
89e10787 | 2068 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); |
2f718ffc NP |
2069 | |
2070 | /* | |
2071 | * This has the same sideeffects and return value as | |
2072 | * iov_iter_copy_from_user_atomic(). | |
2073 | * The difference is that it attempts to resolve faults. | |
2074 | * Page must not be locked. | |
2075 | */ | |
2076 | size_t iov_iter_copy_from_user(struct page *page, | |
2077 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
2078 | { | |
2079 | char *kaddr; | |
2080 | size_t copied; | |
2081 | ||
2082 | kaddr = kmap(page); | |
2083 | if (likely(i->nr_segs == 1)) { | |
2084 | int left; | |
2085 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 2086 | left = __copy_from_user(kaddr + offset, buf, bytes); |
2f718ffc NP |
2087 | copied = bytes - left; |
2088 | } else { | |
2089 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2090 | i->iov, i->iov_offset, bytes); | |
2091 | } | |
2092 | kunmap(page); | |
2093 | return copied; | |
2094 | } | |
89e10787 | 2095 | EXPORT_SYMBOL(iov_iter_copy_from_user); |
2f718ffc | 2096 | |
f7009264 | 2097 | void iov_iter_advance(struct iov_iter *i, size_t bytes) |
2f718ffc | 2098 | { |
f7009264 NP |
2099 | BUG_ON(i->count < bytes); |
2100 | ||
2f718ffc NP |
2101 | if (likely(i->nr_segs == 1)) { |
2102 | i->iov_offset += bytes; | |
f7009264 | 2103 | i->count -= bytes; |
2f718ffc NP |
2104 | } else { |
2105 | const struct iovec *iov = i->iov; | |
2106 | size_t base = i->iov_offset; | |
39be79c1 | 2107 | unsigned long nr_segs = i->nr_segs; |
2f718ffc | 2108 | |
124d3b70 NP |
2109 | /* |
2110 | * The !iov->iov_len check ensures we skip over unlikely | |
f7009264 | 2111 | * zero-length segments (without overruning the iovec). |
124d3b70 | 2112 | */ |
94ad374a | 2113 | while (bytes || unlikely(i->count && !iov->iov_len)) { |
f7009264 | 2114 | int copy; |
2f718ffc | 2115 | |
f7009264 NP |
2116 | copy = min(bytes, iov->iov_len - base); |
2117 | BUG_ON(!i->count || i->count < copy); | |
2118 | i->count -= copy; | |
2f718ffc NP |
2119 | bytes -= copy; |
2120 | base += copy; | |
2121 | if (iov->iov_len == base) { | |
2122 | iov++; | |
39be79c1 | 2123 | nr_segs--; |
2f718ffc NP |
2124 | base = 0; |
2125 | } | |
2126 | } | |
2127 | i->iov = iov; | |
2128 | i->iov_offset = base; | |
39be79c1 | 2129 | i->nr_segs = nr_segs; |
2f718ffc NP |
2130 | } |
2131 | } | |
89e10787 | 2132 | EXPORT_SYMBOL(iov_iter_advance); |
2f718ffc | 2133 | |
afddba49 NP |
2134 | /* |
2135 | * Fault in the first iovec of the given iov_iter, to a maximum length | |
2136 | * of bytes. Returns 0 on success, or non-zero if the memory could not be | |
2137 | * accessed (ie. because it is an invalid address). | |
2138 | * | |
2139 | * writev-intensive code may want this to prefault several iovecs -- that | |
2140 | * would be possible (callers must not rely on the fact that _only_ the | |
2141 | * first iovec will be faulted with the current implementation). | |
2142 | */ | |
2143 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) | |
2f718ffc | 2144 | { |
2f718ffc | 2145 | char __user *buf = i->iov->iov_base + i->iov_offset; |
afddba49 NP |
2146 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); |
2147 | return fault_in_pages_readable(buf, bytes); | |
2f718ffc | 2148 | } |
89e10787 | 2149 | EXPORT_SYMBOL(iov_iter_fault_in_readable); |
2f718ffc NP |
2150 | |
2151 | /* | |
2152 | * Return the count of just the current iov_iter segment. | |
2153 | */ | |
d28574e0 | 2154 | size_t iov_iter_single_seg_count(const struct iov_iter *i) |
2f718ffc NP |
2155 | { |
2156 | const struct iovec *iov = i->iov; | |
2157 | if (i->nr_segs == 1) | |
2158 | return i->count; | |
2159 | else | |
2160 | return min(i->count, iov->iov_len - i->iov_offset); | |
2161 | } | |
89e10787 | 2162 | EXPORT_SYMBOL(iov_iter_single_seg_count); |
2f718ffc | 2163 | |
1da177e4 LT |
2164 | /* |
2165 | * Performs necessary checks before doing a write | |
2166 | * | |
485bb99b | 2167 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2168 | * Returns appropriate error code that caller should return or |
2169 | * zero in case that write should be allowed. | |
2170 | */ | |
2171 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) | |
2172 | { | |
2173 | struct inode *inode = file->f_mapping->host; | |
59e99e5b | 2174 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
1da177e4 LT |
2175 | |
2176 | if (unlikely(*pos < 0)) | |
2177 | return -EINVAL; | |
2178 | ||
1da177e4 LT |
2179 | if (!isblk) { |
2180 | /* FIXME: this is for backwards compatibility with 2.4 */ | |
2181 | if (file->f_flags & O_APPEND) | |
2182 | *pos = i_size_read(inode); | |
2183 | ||
2184 | if (limit != RLIM_INFINITY) { | |
2185 | if (*pos >= limit) { | |
2186 | send_sig(SIGXFSZ, current, 0); | |
2187 | return -EFBIG; | |
2188 | } | |
2189 | if (*count > limit - (typeof(limit))*pos) { | |
2190 | *count = limit - (typeof(limit))*pos; | |
2191 | } | |
2192 | } | |
2193 | } | |
2194 | ||
2195 | /* | |
2196 | * LFS rule | |
2197 | */ | |
2198 | if (unlikely(*pos + *count > MAX_NON_LFS && | |
2199 | !(file->f_flags & O_LARGEFILE))) { | |
2200 | if (*pos >= MAX_NON_LFS) { | |
1da177e4 LT |
2201 | return -EFBIG; |
2202 | } | |
2203 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { | |
2204 | *count = MAX_NON_LFS - (unsigned long)*pos; | |
2205 | } | |
2206 | } | |
2207 | ||
2208 | /* | |
2209 | * Are we about to exceed the fs block limit ? | |
2210 | * | |
2211 | * If we have written data it becomes a short write. If we have | |
2212 | * exceeded without writing data we send a signal and return EFBIG. | |
2213 | * Linus frestrict idea will clean these up nicely.. | |
2214 | */ | |
2215 | if (likely(!isblk)) { | |
2216 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { | |
2217 | if (*count || *pos > inode->i_sb->s_maxbytes) { | |
1da177e4 LT |
2218 | return -EFBIG; |
2219 | } | |
2220 | /* zero-length writes at ->s_maxbytes are OK */ | |
2221 | } | |
2222 | ||
2223 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) | |
2224 | *count = inode->i_sb->s_maxbytes - *pos; | |
2225 | } else { | |
9361401e | 2226 | #ifdef CONFIG_BLOCK |
1da177e4 LT |
2227 | loff_t isize; |
2228 | if (bdev_read_only(I_BDEV(inode))) | |
2229 | return -EPERM; | |
2230 | isize = i_size_read(inode); | |
2231 | if (*pos >= isize) { | |
2232 | if (*count || *pos > isize) | |
2233 | return -ENOSPC; | |
2234 | } | |
2235 | ||
2236 | if (*pos + *count > isize) | |
2237 | *count = isize - *pos; | |
9361401e DH |
2238 | #else |
2239 | return -EPERM; | |
2240 | #endif | |
1da177e4 LT |
2241 | } |
2242 | return 0; | |
2243 | } | |
2244 | EXPORT_SYMBOL(generic_write_checks); | |
2245 | ||
afddba49 NP |
2246 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2247 | loff_t pos, unsigned len, unsigned flags, | |
2248 | struct page **pagep, void **fsdata) | |
2249 | { | |
2250 | const struct address_space_operations *aops = mapping->a_ops; | |
2251 | ||
4e02ed4b | 2252 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2253 | pagep, fsdata); |
afddba49 NP |
2254 | } |
2255 | EXPORT_SYMBOL(pagecache_write_begin); | |
2256 | ||
2257 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2258 | loff_t pos, unsigned len, unsigned copied, | |
2259 | struct page *page, void *fsdata) | |
2260 | { | |
2261 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2262 | |
4e02ed4b NP |
2263 | mark_page_accessed(page); |
2264 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); | |
afddba49 NP |
2265 | } |
2266 | EXPORT_SYMBOL(pagecache_write_end); | |
2267 | ||
1da177e4 LT |
2268 | ssize_t |
2269 | generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, | |
2270 | unsigned long *nr_segs, loff_t pos, loff_t *ppos, | |
2271 | size_t count, size_t ocount) | |
2272 | { | |
2273 | struct file *file = iocb->ki_filp; | |
2274 | struct address_space *mapping = file->f_mapping; | |
2275 | struct inode *inode = mapping->host; | |
2276 | ssize_t written; | |
a969e903 CH |
2277 | size_t write_len; |
2278 | pgoff_t end; | |
1da177e4 LT |
2279 | |
2280 | if (count != ocount) | |
2281 | *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); | |
2282 | ||
a969e903 CH |
2283 | write_len = iov_length(iov, *nr_segs); |
2284 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; | |
a969e903 | 2285 | |
48b47c56 | 2286 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2287 | if (written) |
2288 | goto out; | |
2289 | ||
2290 | /* | |
2291 | * After a write we want buffered reads to be sure to go to disk to get | |
2292 | * the new data. We invalidate clean cached page from the region we're | |
2293 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2294 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2295 | */ |
2296 | if (mapping->nrpages) { | |
2297 | written = invalidate_inode_pages2_range(mapping, | |
2298 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2299 | /* |
2300 | * If a page can not be invalidated, return 0 to fall back | |
2301 | * to buffered write. | |
2302 | */ | |
2303 | if (written) { | |
2304 | if (written == -EBUSY) | |
2305 | return 0; | |
a969e903 | 2306 | goto out; |
6ccfa806 | 2307 | } |
a969e903 CH |
2308 | } |
2309 | ||
2310 | written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs); | |
2311 | ||
2312 | /* | |
2313 | * Finally, try again to invalidate clean pages which might have been | |
2314 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2315 | * if the source of the write was an mmap'ed region of the file | |
2316 | * we're writing. Either one is a pretty crazy thing to do, | |
2317 | * so we don't support it 100%. If this invalidation | |
2318 | * fails, tough, the write still worked... | |
2319 | */ | |
2320 | if (mapping->nrpages) { | |
2321 | invalidate_inode_pages2_range(mapping, | |
2322 | pos >> PAGE_CACHE_SHIFT, end); | |
2323 | } | |
2324 | ||
1da177e4 | 2325 | if (written > 0) { |
0116651c NK |
2326 | pos += written; |
2327 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { | |
2328 | i_size_write(inode, pos); | |
1da177e4 LT |
2329 | mark_inode_dirty(inode); |
2330 | } | |
0116651c | 2331 | *ppos = pos; |
1da177e4 | 2332 | } |
a969e903 | 2333 | out: |
1da177e4 LT |
2334 | return written; |
2335 | } | |
2336 | EXPORT_SYMBOL(generic_file_direct_write); | |
2337 | ||
eb2be189 NP |
2338 | /* |
2339 | * Find or create a page at the given pagecache position. Return the locked | |
2340 | * page. This function is specifically for buffered writes. | |
2341 | */ | |
54566b2c NP |
2342 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2343 | pgoff_t index, unsigned flags) | |
eb2be189 NP |
2344 | { |
2345 | int status; | |
0faa70cb | 2346 | gfp_t gfp_mask; |
eb2be189 | 2347 | struct page *page; |
54566b2c | 2348 | gfp_t gfp_notmask = 0; |
0faa70cb | 2349 | |
1010bb1b FW |
2350 | gfp_mask = mapping_gfp_mask(mapping); |
2351 | if (mapping_cap_account_dirty(mapping)) | |
2352 | gfp_mask |= __GFP_WRITE; | |
54566b2c NP |
2353 | if (flags & AOP_FLAG_NOFS) |
2354 | gfp_notmask = __GFP_FS; | |
eb2be189 NP |
2355 | repeat: |
2356 | page = find_lock_page(mapping, index); | |
c585a267 | 2357 | if (page) |
3d08bcc8 | 2358 | goto found; |
eb2be189 | 2359 | |
0faa70cb | 2360 | page = __page_cache_alloc(gfp_mask & ~gfp_notmask); |
eb2be189 NP |
2361 | if (!page) |
2362 | return NULL; | |
54566b2c NP |
2363 | status = add_to_page_cache_lru(page, mapping, index, |
2364 | GFP_KERNEL & ~gfp_notmask); | |
eb2be189 NP |
2365 | if (unlikely(status)) { |
2366 | page_cache_release(page); | |
2367 | if (status == -EEXIST) | |
2368 | goto repeat; | |
2369 | return NULL; | |
2370 | } | |
3d08bcc8 | 2371 | found: |
1d1d1a76 | 2372 | wait_for_stable_page(page); |
eb2be189 NP |
2373 | return page; |
2374 | } | |
54566b2c | 2375 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2376 | |
afddba49 NP |
2377 | static ssize_t generic_perform_write(struct file *file, |
2378 | struct iov_iter *i, loff_t pos) | |
2379 | { | |
2380 | struct address_space *mapping = file->f_mapping; | |
2381 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2382 | long status = 0; | |
2383 | ssize_t written = 0; | |
674b892e NP |
2384 | unsigned int flags = 0; |
2385 | ||
2386 | /* | |
2387 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2388 | */ | |
2389 | if (segment_eq(get_fs(), KERNEL_DS)) | |
2390 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
afddba49 NP |
2391 | |
2392 | do { | |
2393 | struct page *page; | |
afddba49 NP |
2394 | unsigned long offset; /* Offset into pagecache page */ |
2395 | unsigned long bytes; /* Bytes to write to page */ | |
2396 | size_t copied; /* Bytes copied from user */ | |
2397 | void *fsdata; | |
2398 | ||
2399 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2400 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2401 | iov_iter_count(i)); | |
2402 | ||
2403 | again: | |
afddba49 NP |
2404 | /* |
2405 | * Bring in the user page that we will copy from _first_. | |
2406 | * Otherwise there's a nasty deadlock on copying from the | |
2407 | * same page as we're writing to, without it being marked | |
2408 | * up-to-date. | |
2409 | * | |
2410 | * Not only is this an optimisation, but it is also required | |
2411 | * to check that the address is actually valid, when atomic | |
2412 | * usercopies are used, below. | |
2413 | */ | |
2414 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2415 | status = -EFAULT; | |
2416 | break; | |
2417 | } | |
2418 | ||
674b892e | 2419 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 NP |
2420 | &page, &fsdata); |
2421 | if (unlikely(status)) | |
2422 | break; | |
2423 | ||
931e80e4 | 2424 | if (mapping_writably_mapped(mapping)) |
2425 | flush_dcache_page(page); | |
2426 | ||
afddba49 NP |
2427 | pagefault_disable(); |
2428 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); | |
2429 | pagefault_enable(); | |
2430 | flush_dcache_page(page); | |
2431 | ||
c8236db9 | 2432 | mark_page_accessed(page); |
afddba49 NP |
2433 | status = a_ops->write_end(file, mapping, pos, bytes, copied, |
2434 | page, fsdata); | |
2435 | if (unlikely(status < 0)) | |
2436 | break; | |
2437 | copied = status; | |
2438 | ||
2439 | cond_resched(); | |
2440 | ||
124d3b70 | 2441 | iov_iter_advance(i, copied); |
afddba49 NP |
2442 | if (unlikely(copied == 0)) { |
2443 | /* | |
2444 | * If we were unable to copy any data at all, we must | |
2445 | * fall back to a single segment length write. | |
2446 | * | |
2447 | * If we didn't fallback here, we could livelock | |
2448 | * because not all segments in the iov can be copied at | |
2449 | * once without a pagefault. | |
2450 | */ | |
2451 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2452 | iov_iter_single_seg_count(i)); | |
2453 | goto again; | |
2454 | } | |
afddba49 NP |
2455 | pos += copied; |
2456 | written += copied; | |
2457 | ||
2458 | balance_dirty_pages_ratelimited(mapping); | |
a50527b1 JK |
2459 | if (fatal_signal_pending(current)) { |
2460 | status = -EINTR; | |
2461 | break; | |
2462 | } | |
afddba49 NP |
2463 | } while (iov_iter_count(i)); |
2464 | ||
2465 | return written ? written : status; | |
2466 | } | |
2467 | ||
2468 | ssize_t | |
2469 | generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, | |
2470 | unsigned long nr_segs, loff_t pos, loff_t *ppos, | |
2471 | size_t count, ssize_t written) | |
2472 | { | |
2473 | struct file *file = iocb->ki_filp; | |
afddba49 NP |
2474 | ssize_t status; |
2475 | struct iov_iter i; | |
2476 | ||
2477 | iov_iter_init(&i, iov, nr_segs, count, written); | |
4e02ed4b | 2478 | status = generic_perform_write(file, &i, pos); |
1da177e4 | 2479 | |
1da177e4 | 2480 | if (likely(status >= 0)) { |
afddba49 NP |
2481 | written += status; |
2482 | *ppos = pos + status; | |
1da177e4 LT |
2483 | } |
2484 | ||
1da177e4 LT |
2485 | return written ? written : status; |
2486 | } | |
2487 | EXPORT_SYMBOL(generic_file_buffered_write); | |
2488 | ||
e4dd9de3 JK |
2489 | /** |
2490 | * __generic_file_aio_write - write data to a file | |
2491 | * @iocb: IO state structure (file, offset, etc.) | |
2492 | * @iov: vector with data to write | |
2493 | * @nr_segs: number of segments in the vector | |
2494 | * @ppos: position where to write | |
2495 | * | |
2496 | * This function does all the work needed for actually writing data to a | |
2497 | * file. It does all basic checks, removes SUID from the file, updates | |
2498 | * modification times and calls proper subroutines depending on whether we | |
2499 | * do direct IO or a standard buffered write. | |
2500 | * | |
2501 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2502 | * object which does not need locking at all. | |
2503 | * | |
2504 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2505 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2506 | * avoid syncing under i_mutex. | |
2507 | */ | |
2508 | ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
2509 | unsigned long nr_segs, loff_t *ppos) | |
1da177e4 LT |
2510 | { |
2511 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2512 | struct address_space * mapping = file->f_mapping; |
1da177e4 LT |
2513 | size_t ocount; /* original count */ |
2514 | size_t count; /* after file limit checks */ | |
2515 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2516 | loff_t pos; |
2517 | ssize_t written; | |
2518 | ssize_t err; | |
2519 | ||
2520 | ocount = 0; | |
0ceb3314 DM |
2521 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
2522 | if (err) | |
2523 | return err; | |
1da177e4 LT |
2524 | |
2525 | count = ocount; | |
2526 | pos = *ppos; | |
2527 | ||
1da177e4 LT |
2528 | /* We can write back this queue in page reclaim */ |
2529 | current->backing_dev_info = mapping->backing_dev_info; | |
2530 | written = 0; | |
2531 | ||
2532 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
2533 | if (err) | |
2534 | goto out; | |
2535 | ||
2536 | if (count == 0) | |
2537 | goto out; | |
2538 | ||
2f1936b8 | 2539 | err = file_remove_suid(file); |
1da177e4 LT |
2540 | if (err) |
2541 | goto out; | |
2542 | ||
c3b2da31 JB |
2543 | err = file_update_time(file); |
2544 | if (err) | |
2545 | goto out; | |
1da177e4 LT |
2546 | |
2547 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
2548 | if (unlikely(file->f_flags & O_DIRECT)) { | |
fb5527e6 JM |
2549 | loff_t endbyte; |
2550 | ssize_t written_buffered; | |
2551 | ||
2552 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, | |
2553 | ppos, count, ocount); | |
1da177e4 LT |
2554 | if (written < 0 || written == count) |
2555 | goto out; | |
2556 | /* | |
2557 | * direct-io write to a hole: fall through to buffered I/O | |
2558 | * for completing the rest of the request. | |
2559 | */ | |
2560 | pos += written; | |
2561 | count -= written; | |
fb5527e6 JM |
2562 | written_buffered = generic_file_buffered_write(iocb, iov, |
2563 | nr_segs, pos, ppos, count, | |
2564 | written); | |
2565 | /* | |
2566 | * If generic_file_buffered_write() retuned a synchronous error | |
2567 | * then we want to return the number of bytes which were | |
2568 | * direct-written, or the error code if that was zero. Note | |
2569 | * that this differs from normal direct-io semantics, which | |
2570 | * will return -EFOO even if some bytes were written. | |
2571 | */ | |
2572 | if (written_buffered < 0) { | |
2573 | err = written_buffered; | |
2574 | goto out; | |
2575 | } | |
1da177e4 | 2576 | |
fb5527e6 JM |
2577 | /* |
2578 | * We need to ensure that the page cache pages are written to | |
2579 | * disk and invalidated to preserve the expected O_DIRECT | |
2580 | * semantics. | |
2581 | */ | |
2582 | endbyte = pos + written_buffered - written - 1; | |
c05c4edd | 2583 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); |
fb5527e6 JM |
2584 | if (err == 0) { |
2585 | written = written_buffered; | |
2586 | invalidate_mapping_pages(mapping, | |
2587 | pos >> PAGE_CACHE_SHIFT, | |
2588 | endbyte >> PAGE_CACHE_SHIFT); | |
2589 | } else { | |
2590 | /* | |
2591 | * We don't know how much we wrote, so just return | |
2592 | * the number of bytes which were direct-written | |
2593 | */ | |
2594 | } | |
2595 | } else { | |
2596 | written = generic_file_buffered_write(iocb, iov, nr_segs, | |
2597 | pos, ppos, count, written); | |
2598 | } | |
1da177e4 LT |
2599 | out: |
2600 | current->backing_dev_info = NULL; | |
2601 | return written ? written : err; | |
2602 | } | |
e4dd9de3 JK |
2603 | EXPORT_SYMBOL(__generic_file_aio_write); |
2604 | ||
e4dd9de3 JK |
2605 | /** |
2606 | * generic_file_aio_write - write data to a file | |
2607 | * @iocb: IO state structure | |
2608 | * @iov: vector with data to write | |
2609 | * @nr_segs: number of segments in the vector | |
2610 | * @pos: position in file where to write | |
2611 | * | |
2612 | * This is a wrapper around __generic_file_aio_write() to be used by most | |
2613 | * filesystems. It takes care of syncing the file in case of O_SYNC file | |
2614 | * and acquires i_mutex as needed. | |
2615 | */ | |
027445c3 BP |
2616 | ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
2617 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
2618 | { |
2619 | struct file *file = iocb->ki_filp; | |
148f948b | 2620 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2621 | ssize_t ret; |
1da177e4 LT |
2622 | |
2623 | BUG_ON(iocb->ki_pos != pos); | |
2624 | ||
1b1dcc1b | 2625 | mutex_lock(&inode->i_mutex); |
e4dd9de3 | 2626 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); |
1b1dcc1b | 2627 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 2628 | |
02afc27f | 2629 | if (ret > 0) { |
1da177e4 LT |
2630 | ssize_t err; |
2631 | ||
d311d79d AV |
2632 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
2633 | if (err < 0) | |
1da177e4 LT |
2634 | ret = err; |
2635 | } | |
2636 | return ret; | |
2637 | } | |
2638 | EXPORT_SYMBOL(generic_file_aio_write); | |
2639 | ||
cf9a2ae8 DH |
2640 | /** |
2641 | * try_to_release_page() - release old fs-specific metadata on a page | |
2642 | * | |
2643 | * @page: the page which the kernel is trying to free | |
2644 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2645 | * | |
2646 | * The address_space is to try to release any data against the page | |
2647 | * (presumably at page->private). If the release was successful, return `1'. | |
2648 | * Otherwise return zero. | |
2649 | * | |
266cf658 DH |
2650 | * This may also be called if PG_fscache is set on a page, indicating that the |
2651 | * page is known to the local caching routines. | |
2652 | * | |
cf9a2ae8 | 2653 | * The @gfp_mask argument specifies whether I/O may be performed to release |
3f31fddf | 2654 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
cf9a2ae8 | 2655 | * |
cf9a2ae8 DH |
2656 | */ |
2657 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2658 | { | |
2659 | struct address_space * const mapping = page->mapping; | |
2660 | ||
2661 | BUG_ON(!PageLocked(page)); | |
2662 | if (PageWriteback(page)) | |
2663 | return 0; | |
2664 | ||
2665 | if (mapping && mapping->a_ops->releasepage) | |
2666 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2667 | return try_to_free_buffers(page); | |
2668 | } | |
2669 | ||
2670 | EXPORT_SYMBOL(try_to_release_page); |