]>
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 | ||
91b0abe3 JW |
110 | static void page_cache_tree_delete(struct address_space *mapping, |
111 | struct page *page, void *shadow) | |
112 | { | |
113 | if (shadow) { | |
114 | void **slot; | |
115 | ||
116 | slot = radix_tree_lookup_slot(&mapping->page_tree, page->index); | |
117 | radix_tree_replace_slot(slot, shadow); | |
118 | mapping->nrshadows++; | |
119 | /* | |
120 | * Make sure the nrshadows update is committed before | |
121 | * the nrpages update so that final truncate racing | |
122 | * with reclaim does not see both counters 0 at the | |
123 | * same time and miss a shadow entry. | |
124 | */ | |
125 | smp_wmb(); | |
126 | } else | |
127 | radix_tree_delete(&mapping->page_tree, page->index); | |
128 | mapping->nrpages--; | |
129 | } | |
130 | ||
1da177e4 | 131 | /* |
e64a782f | 132 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 133 | * sure the page is locked and that nobody else uses it - or that usage |
19fd6231 | 134 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 135 | */ |
91b0abe3 | 136 | void __delete_from_page_cache(struct page *page, void *shadow) |
1da177e4 LT |
137 | { |
138 | struct address_space *mapping = page->mapping; | |
139 | ||
fe0bfaaf | 140 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
141 | /* |
142 | * if we're uptodate, flush out into the cleancache, otherwise | |
143 | * invalidate any existing cleancache entries. We can't leave | |
144 | * stale data around in the cleancache once our page is gone | |
145 | */ | |
146 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
147 | cleancache_put_page(page); | |
148 | else | |
3167760f | 149 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 150 | |
91b0abe3 JW |
151 | page_cache_tree_delete(mapping, page, shadow); |
152 | ||
1da177e4 | 153 | page->mapping = NULL; |
b85e0eff | 154 | /* Leave page->index set: truncation lookup relies upon it */ |
91b0abe3 | 155 | |
347ce434 | 156 | __dec_zone_page_state(page, NR_FILE_PAGES); |
4b02108a KM |
157 | if (PageSwapBacked(page)) |
158 | __dec_zone_page_state(page, NR_SHMEM); | |
45426812 | 159 | BUG_ON(page_mapped(page)); |
3a692790 LT |
160 | |
161 | /* | |
162 | * Some filesystems seem to re-dirty the page even after | |
163 | * the VM has canceled the dirty bit (eg ext3 journaling). | |
164 | * | |
165 | * Fix it up by doing a final dirty accounting check after | |
166 | * having removed the page entirely. | |
167 | */ | |
168 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { | |
169 | dec_zone_page_state(page, NR_FILE_DIRTY); | |
170 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); | |
171 | } | |
1da177e4 LT |
172 | } |
173 | ||
702cfbf9 MK |
174 | /** |
175 | * delete_from_page_cache - delete page from page cache | |
176 | * @page: the page which the kernel is trying to remove from page cache | |
177 | * | |
178 | * This must be called only on pages that have been verified to be in the page | |
179 | * cache and locked. It will never put the page into the free list, the caller | |
180 | * has a reference on the page. | |
181 | */ | |
182 | void delete_from_page_cache(struct page *page) | |
1da177e4 LT |
183 | { |
184 | struct address_space *mapping = page->mapping; | |
6072d13c | 185 | void (*freepage)(struct page *); |
1da177e4 | 186 | |
cd7619d6 | 187 | BUG_ON(!PageLocked(page)); |
1da177e4 | 188 | |
6072d13c | 189 | freepage = mapping->a_ops->freepage; |
19fd6231 | 190 | spin_lock_irq(&mapping->tree_lock); |
91b0abe3 | 191 | __delete_from_page_cache(page, NULL); |
19fd6231 | 192 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 193 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
194 | |
195 | if (freepage) | |
196 | freepage(page); | |
97cecb5a MK |
197 | page_cache_release(page); |
198 | } | |
199 | EXPORT_SYMBOL(delete_from_page_cache); | |
200 | ||
7eaceacc | 201 | static int sleep_on_page(void *word) |
1da177e4 | 202 | { |
1da177e4 LT |
203 | io_schedule(); |
204 | return 0; | |
205 | } | |
206 | ||
7eaceacc | 207 | static int sleep_on_page_killable(void *word) |
2687a356 | 208 | { |
7eaceacc | 209 | sleep_on_page(word); |
2687a356 MW |
210 | return fatal_signal_pending(current) ? -EINTR : 0; |
211 | } | |
212 | ||
865ffef3 DM |
213 | static int filemap_check_errors(struct address_space *mapping) |
214 | { | |
215 | int ret = 0; | |
216 | /* Check for outstanding write errors */ | |
217 | if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
218 | ret = -ENOSPC; | |
219 | if (test_and_clear_bit(AS_EIO, &mapping->flags)) | |
220 | ret = -EIO; | |
221 | return ret; | |
222 | } | |
223 | ||
1da177e4 | 224 | /** |
485bb99b | 225 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
226 | * @mapping: address space structure to write |
227 | * @start: offset in bytes where the range starts | |
469eb4d0 | 228 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 229 | * @sync_mode: enable synchronous operation |
1da177e4 | 230 | * |
485bb99b RD |
231 | * Start writeback against all of a mapping's dirty pages that lie |
232 | * within the byte offsets <start, end> inclusive. | |
233 | * | |
1da177e4 | 234 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 235 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
236 | * these two operations is that if a dirty page/buffer is encountered, it must |
237 | * be waited upon, and not just skipped over. | |
238 | */ | |
ebcf28e1 AM |
239 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
240 | loff_t end, int sync_mode) | |
1da177e4 LT |
241 | { |
242 | int ret; | |
243 | struct writeback_control wbc = { | |
244 | .sync_mode = sync_mode, | |
05fe478d | 245 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
246 | .range_start = start, |
247 | .range_end = end, | |
1da177e4 LT |
248 | }; |
249 | ||
250 | if (!mapping_cap_writeback_dirty(mapping)) | |
251 | return 0; | |
252 | ||
253 | ret = do_writepages(mapping, &wbc); | |
254 | return ret; | |
255 | } | |
256 | ||
257 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
258 | int sync_mode) | |
259 | { | |
111ebb6e | 260 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
261 | } |
262 | ||
263 | int filemap_fdatawrite(struct address_space *mapping) | |
264 | { | |
265 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
266 | } | |
267 | EXPORT_SYMBOL(filemap_fdatawrite); | |
268 | ||
f4c0a0fd | 269 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 270 | loff_t end) |
1da177e4 LT |
271 | { |
272 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
273 | } | |
f4c0a0fd | 274 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 275 | |
485bb99b RD |
276 | /** |
277 | * filemap_flush - mostly a non-blocking flush | |
278 | * @mapping: target address_space | |
279 | * | |
1da177e4 LT |
280 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
281 | * purposes - I/O may not be started against all dirty pages. | |
282 | */ | |
283 | int filemap_flush(struct address_space *mapping) | |
284 | { | |
285 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
286 | } | |
287 | EXPORT_SYMBOL(filemap_flush); | |
288 | ||
485bb99b | 289 | /** |
94004ed7 CH |
290 | * filemap_fdatawait_range - wait for writeback to complete |
291 | * @mapping: address space structure to wait for | |
292 | * @start_byte: offset in bytes where the range starts | |
293 | * @end_byte: offset in bytes where the range ends (inclusive) | |
485bb99b | 294 | * |
94004ed7 CH |
295 | * Walk the list of under-writeback pages of the given address space |
296 | * in the given range and wait for all of them. | |
1da177e4 | 297 | */ |
94004ed7 CH |
298 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
299 | loff_t end_byte) | |
1da177e4 | 300 | { |
94004ed7 CH |
301 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
302 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
303 | struct pagevec pvec; |
304 | int nr_pages; | |
865ffef3 | 305 | int ret2, ret = 0; |
1da177e4 | 306 | |
94004ed7 | 307 | if (end_byte < start_byte) |
865ffef3 | 308 | goto out; |
1da177e4 LT |
309 | |
310 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
311 | while ((index <= end) && |
312 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
313 | PAGECACHE_TAG_WRITEBACK, | |
314 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
315 | unsigned i; | |
316 | ||
317 | for (i = 0; i < nr_pages; i++) { | |
318 | struct page *page = pvec.pages[i]; | |
319 | ||
320 | /* until radix tree lookup accepts end_index */ | |
321 | if (page->index > end) | |
322 | continue; | |
323 | ||
324 | wait_on_page_writeback(page); | |
212260aa | 325 | if (TestClearPageError(page)) |
1da177e4 LT |
326 | ret = -EIO; |
327 | } | |
328 | pagevec_release(&pvec); | |
329 | cond_resched(); | |
330 | } | |
865ffef3 DM |
331 | out: |
332 | ret2 = filemap_check_errors(mapping); | |
333 | if (!ret) | |
334 | ret = ret2; | |
1da177e4 LT |
335 | |
336 | return ret; | |
337 | } | |
d3bccb6f JK |
338 | EXPORT_SYMBOL(filemap_fdatawait_range); |
339 | ||
1da177e4 | 340 | /** |
485bb99b | 341 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 342 | * @mapping: address space structure to wait for |
485bb99b RD |
343 | * |
344 | * Walk the list of under-writeback pages of the given address space | |
345 | * and wait for all of them. | |
1da177e4 LT |
346 | */ |
347 | int filemap_fdatawait(struct address_space *mapping) | |
348 | { | |
349 | loff_t i_size = i_size_read(mapping->host); | |
350 | ||
351 | if (i_size == 0) | |
352 | return 0; | |
353 | ||
94004ed7 | 354 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
355 | } |
356 | EXPORT_SYMBOL(filemap_fdatawait); | |
357 | ||
358 | int filemap_write_and_wait(struct address_space *mapping) | |
359 | { | |
28fd1298 | 360 | int err = 0; |
1da177e4 LT |
361 | |
362 | if (mapping->nrpages) { | |
28fd1298 OH |
363 | err = filemap_fdatawrite(mapping); |
364 | /* | |
365 | * Even if the above returned error, the pages may be | |
366 | * written partially (e.g. -ENOSPC), so we wait for it. | |
367 | * But the -EIO is special case, it may indicate the worst | |
368 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
369 | */ | |
370 | if (err != -EIO) { | |
371 | int err2 = filemap_fdatawait(mapping); | |
372 | if (!err) | |
373 | err = err2; | |
374 | } | |
865ffef3 DM |
375 | } else { |
376 | err = filemap_check_errors(mapping); | |
1da177e4 | 377 | } |
28fd1298 | 378 | return err; |
1da177e4 | 379 | } |
28fd1298 | 380 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 381 | |
485bb99b RD |
382 | /** |
383 | * filemap_write_and_wait_range - write out & wait on a file range | |
384 | * @mapping: the address_space for the pages | |
385 | * @lstart: offset in bytes where the range starts | |
386 | * @lend: offset in bytes where the range ends (inclusive) | |
387 | * | |
469eb4d0 AM |
388 | * Write out and wait upon file offsets lstart->lend, inclusive. |
389 | * | |
390 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
391 | * that this function can be used to write to the very end-of-file (end = -1). | |
392 | */ | |
1da177e4 LT |
393 | int filemap_write_and_wait_range(struct address_space *mapping, |
394 | loff_t lstart, loff_t lend) | |
395 | { | |
28fd1298 | 396 | int err = 0; |
1da177e4 LT |
397 | |
398 | if (mapping->nrpages) { | |
28fd1298 OH |
399 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
400 | WB_SYNC_ALL); | |
401 | /* See comment of filemap_write_and_wait() */ | |
402 | if (err != -EIO) { | |
94004ed7 CH |
403 | int err2 = filemap_fdatawait_range(mapping, |
404 | lstart, lend); | |
28fd1298 OH |
405 | if (!err) |
406 | err = err2; | |
407 | } | |
865ffef3 DM |
408 | } else { |
409 | err = filemap_check_errors(mapping); | |
1da177e4 | 410 | } |
28fd1298 | 411 | return err; |
1da177e4 | 412 | } |
f6995585 | 413 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 414 | |
ef6a3c63 MS |
415 | /** |
416 | * replace_page_cache_page - replace a pagecache page with a new one | |
417 | * @old: page to be replaced | |
418 | * @new: page to replace with | |
419 | * @gfp_mask: allocation mode | |
420 | * | |
421 | * This function replaces a page in the pagecache with a new one. On | |
422 | * success it acquires the pagecache reference for the new page and | |
423 | * drops it for the old page. Both the old and new pages must be | |
424 | * locked. This function does not add the new page to the LRU, the | |
425 | * caller must do that. | |
426 | * | |
427 | * The remove + add is atomic. The only way this function can fail is | |
428 | * memory allocation failure. | |
429 | */ | |
430 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
431 | { | |
432 | int error; | |
ef6a3c63 | 433 | |
309381fe SL |
434 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
435 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
436 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 437 | |
ef6a3c63 MS |
438 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
439 | if (!error) { | |
440 | struct address_space *mapping = old->mapping; | |
441 | void (*freepage)(struct page *); | |
442 | ||
443 | pgoff_t offset = old->index; | |
444 | freepage = mapping->a_ops->freepage; | |
445 | ||
446 | page_cache_get(new); | |
447 | new->mapping = mapping; | |
448 | new->index = offset; | |
449 | ||
450 | spin_lock_irq(&mapping->tree_lock); | |
91b0abe3 | 451 | __delete_from_page_cache(old, NULL); |
ef6a3c63 MS |
452 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
453 | BUG_ON(error); | |
454 | mapping->nrpages++; | |
455 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
456 | if (PageSwapBacked(new)) | |
457 | __inc_zone_page_state(new, NR_SHMEM); | |
458 | spin_unlock_irq(&mapping->tree_lock); | |
ab936cbc KH |
459 | /* mem_cgroup codes must not be called under tree_lock */ |
460 | mem_cgroup_replace_page_cache(old, new); | |
ef6a3c63 MS |
461 | radix_tree_preload_end(); |
462 | if (freepage) | |
463 | freepage(old); | |
464 | page_cache_release(old); | |
ef6a3c63 MS |
465 | } |
466 | ||
467 | return error; | |
468 | } | |
469 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
470 | ||
0cd6144a JW |
471 | static int page_cache_tree_insert(struct address_space *mapping, |
472 | struct page *page) | |
473 | { | |
474 | void **slot; | |
475 | int error; | |
476 | ||
477 | slot = radix_tree_lookup_slot(&mapping->page_tree, page->index); | |
478 | if (slot) { | |
479 | void *p; | |
480 | ||
481 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
482 | if (!radix_tree_exceptional_entry(p)) | |
483 | return -EEXIST; | |
484 | radix_tree_replace_slot(slot, page); | |
91b0abe3 | 485 | mapping->nrshadows--; |
0cd6144a JW |
486 | mapping->nrpages++; |
487 | return 0; | |
488 | } | |
489 | error = radix_tree_insert(&mapping->page_tree, page->index, page); | |
490 | if (!error) | |
491 | mapping->nrpages++; | |
492 | return error; | |
493 | } | |
494 | ||
485bb99b | 495 | /** |
e286781d | 496 | * add_to_page_cache_locked - add a locked page to the pagecache |
485bb99b RD |
497 | * @page: page to add |
498 | * @mapping: the page's address_space | |
499 | * @offset: page index | |
500 | * @gfp_mask: page allocation mode | |
501 | * | |
e286781d | 502 | * This function is used to add a page to the pagecache. It must be locked. |
1da177e4 LT |
503 | * This function does not add the page to the LRU. The caller must do that. |
504 | */ | |
e286781d | 505 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, |
6daa0e28 | 506 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 507 | { |
e286781d NP |
508 | int error; |
509 | ||
309381fe SL |
510 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
511 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d NP |
512 | |
513 | error = mem_cgroup_cache_charge(page, current->mm, | |
2c26fdd7 | 514 | gfp_mask & GFP_RECLAIM_MASK); |
35c754d7 | 515 | if (error) |
66a0c8ee | 516 | return error; |
1da177e4 | 517 | |
5e4c0d97 | 518 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 519 | if (error) { |
69029cd5 | 520 | mem_cgroup_uncharge_cache_page(page); |
66a0c8ee KS |
521 | return error; |
522 | } | |
523 | ||
524 | page_cache_get(page); | |
525 | page->mapping = mapping; | |
526 | page->index = offset; | |
527 | ||
528 | spin_lock_irq(&mapping->tree_lock); | |
0cd6144a | 529 | error = page_cache_tree_insert(mapping, page); |
66a0c8ee KS |
530 | radix_tree_preload_end(); |
531 | if (unlikely(error)) | |
532 | goto err_insert; | |
66a0c8ee KS |
533 | __inc_zone_page_state(page, NR_FILE_PAGES); |
534 | spin_unlock_irq(&mapping->tree_lock); | |
535 | trace_mm_filemap_add_to_page_cache(page); | |
536 | return 0; | |
537 | err_insert: | |
538 | page->mapping = NULL; | |
539 | /* Leave page->index set: truncation relies upon it */ | |
540 | spin_unlock_irq(&mapping->tree_lock); | |
541 | mem_cgroup_uncharge_cache_page(page); | |
542 | page_cache_release(page); | |
1da177e4 LT |
543 | return error; |
544 | } | |
e286781d | 545 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
546 | |
547 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 548 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 549 | { |
4f98a2fe RR |
550 | int ret; |
551 | ||
4f98a2fe | 552 | ret = add_to_page_cache(page, mapping, offset, gfp_mask); |
31475dd6 HD |
553 | if (ret == 0) |
554 | lru_cache_add_file(page); | |
1da177e4 LT |
555 | return ret; |
556 | } | |
18bc0bbd | 557 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 558 | |
44110fe3 | 559 | #ifdef CONFIG_NUMA |
2ae88149 | 560 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 561 | { |
c0ff7453 MX |
562 | int n; |
563 | struct page *page; | |
564 | ||
44110fe3 | 565 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
566 | unsigned int cpuset_mems_cookie; |
567 | do { | |
d26914d1 | 568 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 MG |
569 | n = cpuset_mem_spread_node(); |
570 | page = alloc_pages_exact_node(n, gfp, 0); | |
d26914d1 | 571 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 572 | |
c0ff7453 | 573 | return page; |
44110fe3 | 574 | } |
2ae88149 | 575 | return alloc_pages(gfp, 0); |
44110fe3 | 576 | } |
2ae88149 | 577 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
578 | #endif |
579 | ||
1da177e4 LT |
580 | /* |
581 | * In order to wait for pages to become available there must be | |
582 | * waitqueues associated with pages. By using a hash table of | |
583 | * waitqueues where the bucket discipline is to maintain all | |
584 | * waiters on the same queue and wake all when any of the pages | |
585 | * become available, and for the woken contexts to check to be | |
586 | * sure the appropriate page became available, this saves space | |
587 | * at a cost of "thundering herd" phenomena during rare hash | |
588 | * collisions. | |
589 | */ | |
590 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
591 | { | |
592 | const struct zone *zone = page_zone(page); | |
593 | ||
594 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
595 | } | |
596 | ||
597 | static inline void wake_up_page(struct page *page, int bit) | |
598 | { | |
599 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); | |
600 | } | |
601 | ||
920c7a5d | 602 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
603 | { |
604 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
605 | ||
606 | if (test_bit(bit_nr, &page->flags)) | |
7eaceacc | 607 | __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
608 | TASK_UNINTERRUPTIBLE); |
609 | } | |
610 | EXPORT_SYMBOL(wait_on_page_bit); | |
611 | ||
f62e00cc KM |
612 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
613 | { | |
614 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
615 | ||
616 | if (!test_bit(bit_nr, &page->flags)) | |
617 | return 0; | |
618 | ||
619 | return __wait_on_bit(page_waitqueue(page), &wait, | |
620 | sleep_on_page_killable, TASK_KILLABLE); | |
621 | } | |
622 | ||
385e1ca5 DH |
623 | /** |
624 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
625 | * @page: Page defining the wait queue of interest |
626 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
627 | * |
628 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
629 | */ | |
630 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
631 | { | |
632 | wait_queue_head_t *q = page_waitqueue(page); | |
633 | unsigned long flags; | |
634 | ||
635 | spin_lock_irqsave(&q->lock, flags); | |
636 | __add_wait_queue(q, waiter); | |
637 | spin_unlock_irqrestore(&q->lock, flags); | |
638 | } | |
639 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
640 | ||
1da177e4 | 641 | /** |
485bb99b | 642 | * unlock_page - unlock a locked page |
1da177e4 LT |
643 | * @page: the page |
644 | * | |
645 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
646 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
647 | * mechananism between PageLocked pages and PageWriteback pages is shared. | |
648 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. | |
649 | * | |
8413ac9d NP |
650 | * The mb is necessary to enforce ordering between the clear_bit and the read |
651 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 652 | */ |
920c7a5d | 653 | void unlock_page(struct page *page) |
1da177e4 | 654 | { |
309381fe | 655 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
8413ac9d NP |
656 | clear_bit_unlock(PG_locked, &page->flags); |
657 | smp_mb__after_clear_bit(); | |
1da177e4 LT |
658 | wake_up_page(page, PG_locked); |
659 | } | |
660 | EXPORT_SYMBOL(unlock_page); | |
661 | ||
485bb99b RD |
662 | /** |
663 | * end_page_writeback - end writeback against a page | |
664 | * @page: the page | |
1da177e4 LT |
665 | */ |
666 | void end_page_writeback(struct page *page) | |
667 | { | |
ac6aadb2 MS |
668 | if (TestClearPageReclaim(page)) |
669 | rotate_reclaimable_page(page); | |
670 | ||
671 | if (!test_clear_page_writeback(page)) | |
672 | BUG(); | |
673 | ||
1da177e4 LT |
674 | smp_mb__after_clear_bit(); |
675 | wake_up_page(page, PG_writeback); | |
676 | } | |
677 | EXPORT_SYMBOL(end_page_writeback); | |
678 | ||
485bb99b RD |
679 | /** |
680 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
681 | * @page: the page to lock | |
1da177e4 | 682 | */ |
920c7a5d | 683 | void __lock_page(struct page *page) |
1da177e4 LT |
684 | { |
685 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
686 | ||
7eaceacc | 687 | __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page, |
1da177e4 LT |
688 | TASK_UNINTERRUPTIBLE); |
689 | } | |
690 | EXPORT_SYMBOL(__lock_page); | |
691 | ||
b5606c2d | 692 | int __lock_page_killable(struct page *page) |
2687a356 MW |
693 | { |
694 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); | |
695 | ||
696 | return __wait_on_bit_lock(page_waitqueue(page), &wait, | |
7eaceacc | 697 | sleep_on_page_killable, TASK_KILLABLE); |
2687a356 | 698 | } |
18bc0bbd | 699 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 700 | |
d065bd81 ML |
701 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
702 | unsigned int flags) | |
703 | { | |
37b23e05 KM |
704 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
705 | /* | |
706 | * CAUTION! In this case, mmap_sem is not released | |
707 | * even though return 0. | |
708 | */ | |
709 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
710 | return 0; | |
711 | ||
712 | up_read(&mm->mmap_sem); | |
713 | if (flags & FAULT_FLAG_KILLABLE) | |
714 | wait_on_page_locked_killable(page); | |
715 | else | |
318b275f | 716 | wait_on_page_locked(page); |
d065bd81 | 717 | return 0; |
37b23e05 KM |
718 | } else { |
719 | if (flags & FAULT_FLAG_KILLABLE) { | |
720 | int ret; | |
721 | ||
722 | ret = __lock_page_killable(page); | |
723 | if (ret) { | |
724 | up_read(&mm->mmap_sem); | |
725 | return 0; | |
726 | } | |
727 | } else | |
728 | __lock_page(page); | |
729 | return 1; | |
d065bd81 ML |
730 | } |
731 | } | |
732 | ||
e7b563bb JW |
733 | /** |
734 | * page_cache_next_hole - find the next hole (not-present entry) | |
735 | * @mapping: mapping | |
736 | * @index: index | |
737 | * @max_scan: maximum range to search | |
738 | * | |
739 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
740 | * lowest indexed hole. | |
741 | * | |
742 | * Returns: the index of the hole if found, otherwise returns an index | |
743 | * outside of the set specified (in which case 'return - index >= | |
744 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
745 | * be returned. | |
746 | * | |
747 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
748 | * like radix_tree_gang_lookup, this will not atomically search a | |
749 | * snapshot of the tree at a single point in time. For example, if a | |
750 | * hole is created at index 5, then subsequently a hole is created at | |
751 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
752 | * if called under rcu_read_lock. | |
753 | */ | |
754 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
755 | pgoff_t index, unsigned long max_scan) | |
756 | { | |
757 | unsigned long i; | |
758 | ||
759 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
760 | struct page *page; |
761 | ||
762 | page = radix_tree_lookup(&mapping->page_tree, index); | |
763 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
764 | break; |
765 | index++; | |
766 | if (index == 0) | |
767 | break; | |
768 | } | |
769 | ||
770 | return index; | |
771 | } | |
772 | EXPORT_SYMBOL(page_cache_next_hole); | |
773 | ||
774 | /** | |
775 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
776 | * @mapping: mapping | |
777 | * @index: index | |
778 | * @max_scan: maximum range to search | |
779 | * | |
780 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
781 | * the first hole. | |
782 | * | |
783 | * Returns: the index of the hole if found, otherwise returns an index | |
784 | * outside of the set specified (in which case 'index - return >= | |
785 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
786 | * will be returned. | |
787 | * | |
788 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
789 | * like radix_tree_gang_lookup, this will not atomically search a | |
790 | * snapshot of the tree at a single point in time. For example, if a | |
791 | * hole is created at index 10, then subsequently a hole is created at | |
792 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
793 | * called under rcu_read_lock. | |
794 | */ | |
795 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
796 | pgoff_t index, unsigned long max_scan) | |
797 | { | |
798 | unsigned long i; | |
799 | ||
800 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
801 | struct page *page; |
802 | ||
803 | page = radix_tree_lookup(&mapping->page_tree, index); | |
804 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
805 | break; |
806 | index--; | |
807 | if (index == ULONG_MAX) | |
808 | break; | |
809 | } | |
810 | ||
811 | return index; | |
812 | } | |
813 | EXPORT_SYMBOL(page_cache_prev_hole); | |
814 | ||
485bb99b | 815 | /** |
0cd6144a | 816 | * find_get_entry - find and get a page cache entry |
485bb99b | 817 | * @mapping: the address_space to search |
0cd6144a JW |
818 | * @offset: the page cache index |
819 | * | |
820 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
821 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 822 | * |
0cd6144a JW |
823 | * If the slot holds a shadow entry of a previously evicted page, it |
824 | * is returned. | |
825 | * | |
826 | * Otherwise, %NULL is returned. | |
1da177e4 | 827 | */ |
0cd6144a | 828 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 829 | { |
a60637c8 | 830 | void **pagep; |
1da177e4 LT |
831 | struct page *page; |
832 | ||
a60637c8 NP |
833 | rcu_read_lock(); |
834 | repeat: | |
835 | page = NULL; | |
836 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
837 | if (pagep) { | |
838 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
839 | if (unlikely(!page)) |
840 | goto out; | |
a2c16d6c | 841 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
842 | if (radix_tree_deref_retry(page)) |
843 | goto repeat; | |
844 | /* | |
845 | * Otherwise, shmem/tmpfs must be storing a swap entry | |
846 | * here as an exceptional entry: so return it without | |
847 | * attempting to raise page count. | |
848 | */ | |
849 | goto out; | |
a2c16d6c | 850 | } |
a60637c8 NP |
851 | if (!page_cache_get_speculative(page)) |
852 | goto repeat; | |
853 | ||
854 | /* | |
855 | * Has the page moved? | |
856 | * This is part of the lockless pagecache protocol. See | |
857 | * include/linux/pagemap.h for details. | |
858 | */ | |
859 | if (unlikely(page != *pagep)) { | |
860 | page_cache_release(page); | |
861 | goto repeat; | |
862 | } | |
863 | } | |
27d20fdd | 864 | out: |
a60637c8 NP |
865 | rcu_read_unlock(); |
866 | ||
1da177e4 LT |
867 | return page; |
868 | } | |
0cd6144a | 869 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 870 | |
1da177e4 | 871 | /** |
0cd6144a | 872 | * find_get_page - find and get a page reference |
67be2dd1 MW |
873 | * @mapping: the address_space to search |
874 | * @offset: the page index | |
1da177e4 | 875 | * |
0cd6144a JW |
876 | * Looks up the page cache slot at @mapping & @offset. If there is a |
877 | * page cache page, it is returned with an increased refcount. | |
1da177e4 | 878 | * |
0cd6144a | 879 | * Otherwise, %NULL is returned. |
1da177e4 | 880 | */ |
0cd6144a JW |
881 | struct page *find_get_page(struct address_space *mapping, pgoff_t offset) |
882 | { | |
883 | struct page *page = find_get_entry(mapping, offset); | |
884 | ||
885 | if (radix_tree_exceptional_entry(page)) | |
886 | page = NULL; | |
887 | return page; | |
888 | } | |
889 | EXPORT_SYMBOL(find_get_page); | |
890 | ||
891 | /** | |
892 | * find_lock_entry - locate, pin and lock a page cache entry | |
893 | * @mapping: the address_space to search | |
894 | * @offset: the page cache index | |
895 | * | |
896 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
897 | * page cache page, it is returned locked and with an increased | |
898 | * refcount. | |
899 | * | |
900 | * If the slot holds a shadow entry of a previously evicted page, it | |
901 | * is returned. | |
902 | * | |
903 | * Otherwise, %NULL is returned. | |
904 | * | |
905 | * find_lock_entry() may sleep. | |
906 | */ | |
907 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
908 | { |
909 | struct page *page; | |
910 | ||
1da177e4 | 911 | repeat: |
0cd6144a | 912 | page = find_get_entry(mapping, offset); |
a2c16d6c | 913 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
914 | lock_page(page); |
915 | /* Has the page been truncated? */ | |
916 | if (unlikely(page->mapping != mapping)) { | |
917 | unlock_page(page); | |
918 | page_cache_release(page); | |
919 | goto repeat; | |
1da177e4 | 920 | } |
309381fe | 921 | VM_BUG_ON_PAGE(page->index != offset, page); |
1da177e4 | 922 | } |
1da177e4 LT |
923 | return page; |
924 | } | |
0cd6144a JW |
925 | EXPORT_SYMBOL(find_lock_entry); |
926 | ||
927 | /** | |
928 | * find_lock_page - locate, pin and lock a pagecache page | |
929 | * @mapping: the address_space to search | |
930 | * @offset: the page index | |
931 | * | |
932 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
933 | * page cache page, it is returned locked and with an increased | |
934 | * refcount. | |
935 | * | |
936 | * Otherwise, %NULL is returned. | |
937 | * | |
938 | * find_lock_page() may sleep. | |
939 | */ | |
940 | struct page *find_lock_page(struct address_space *mapping, pgoff_t offset) | |
941 | { | |
942 | struct page *page = find_lock_entry(mapping, offset); | |
943 | ||
944 | if (radix_tree_exceptional_entry(page)) | |
945 | page = NULL; | |
946 | return page; | |
947 | } | |
1da177e4 LT |
948 | EXPORT_SYMBOL(find_lock_page); |
949 | ||
950 | /** | |
951 | * find_or_create_page - locate or add a pagecache page | |
67be2dd1 MW |
952 | * @mapping: the page's address_space |
953 | * @index: the page's index into the mapping | |
954 | * @gfp_mask: page allocation mode | |
1da177e4 | 955 | * |
0cd6144a JW |
956 | * Looks up the page cache slot at @mapping & @offset. If there is a |
957 | * page cache page, it is returned locked and with an increased | |
958 | * refcount. | |
959 | * | |
960 | * If the page is not present, a new page is allocated using @gfp_mask | |
961 | * and added to the page cache and the VM's LRU list. The page is | |
962 | * returned locked and with an increased refcount. | |
1da177e4 | 963 | * |
0cd6144a | 964 | * On memory exhaustion, %NULL is returned. |
1da177e4 | 965 | * |
0cd6144a JW |
966 | * find_or_create_page() may sleep, even if @gfp_flags specifies an |
967 | * atomic allocation! | |
1da177e4 LT |
968 | */ |
969 | struct page *find_or_create_page(struct address_space *mapping, | |
57f6b96c | 970 | pgoff_t index, gfp_t gfp_mask) |
1da177e4 | 971 | { |
eb2be189 | 972 | struct page *page; |
1da177e4 LT |
973 | int err; |
974 | repeat: | |
975 | page = find_lock_page(mapping, index); | |
976 | if (!page) { | |
eb2be189 NP |
977 | page = __page_cache_alloc(gfp_mask); |
978 | if (!page) | |
979 | return NULL; | |
67d58ac4 NP |
980 | /* |
981 | * We want a regular kernel memory (not highmem or DMA etc) | |
982 | * allocation for the radix tree nodes, but we need to honour | |
983 | * the context-specific requirements the caller has asked for. | |
984 | * GFP_RECLAIM_MASK collects those requirements. | |
985 | */ | |
986 | err = add_to_page_cache_lru(page, mapping, index, | |
987 | (gfp_mask & GFP_RECLAIM_MASK)); | |
eb2be189 NP |
988 | if (unlikely(err)) { |
989 | page_cache_release(page); | |
990 | page = NULL; | |
991 | if (err == -EEXIST) | |
992 | goto repeat; | |
1da177e4 | 993 | } |
1da177e4 | 994 | } |
1da177e4 LT |
995 | return page; |
996 | } | |
1da177e4 LT |
997 | EXPORT_SYMBOL(find_or_create_page); |
998 | ||
0cd6144a JW |
999 | /** |
1000 | * find_get_entries - gang pagecache lookup | |
1001 | * @mapping: The address_space to search | |
1002 | * @start: The starting page cache index | |
1003 | * @nr_entries: The maximum number of entries | |
1004 | * @entries: Where the resulting entries are placed | |
1005 | * @indices: The cache indices corresponding to the entries in @entries | |
1006 | * | |
1007 | * find_get_entries() will search for and return a group of up to | |
1008 | * @nr_entries entries in the mapping. The entries are placed at | |
1009 | * @entries. find_get_entries() takes a reference against any actual | |
1010 | * pages it returns. | |
1011 | * | |
1012 | * The search returns a group of mapping-contiguous page cache entries | |
1013 | * with ascending indexes. There may be holes in the indices due to | |
1014 | * not-present pages. | |
1015 | * | |
1016 | * Any shadow entries of evicted pages are included in the returned | |
1017 | * array. | |
1018 | * | |
1019 | * find_get_entries() returns the number of pages and shadow entries | |
1020 | * which were found. | |
1021 | */ | |
1022 | unsigned find_get_entries(struct address_space *mapping, | |
1023 | pgoff_t start, unsigned int nr_entries, | |
1024 | struct page **entries, pgoff_t *indices) | |
1025 | { | |
1026 | void **slot; | |
1027 | unsigned int ret = 0; | |
1028 | struct radix_tree_iter iter; | |
1029 | ||
1030 | if (!nr_entries) | |
1031 | return 0; | |
1032 | ||
1033 | rcu_read_lock(); | |
1034 | restart: | |
1035 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { | |
1036 | struct page *page; | |
1037 | repeat: | |
1038 | page = radix_tree_deref_slot(slot); | |
1039 | if (unlikely(!page)) | |
1040 | continue; | |
1041 | if (radix_tree_exception(page)) { | |
1042 | if (radix_tree_deref_retry(page)) | |
1043 | goto restart; | |
1044 | /* | |
1045 | * Otherwise, we must be storing a swap entry | |
1046 | * here as an exceptional entry: so return it | |
1047 | * without attempting to raise page count. | |
1048 | */ | |
1049 | goto export; | |
1050 | } | |
1051 | if (!page_cache_get_speculative(page)) | |
1052 | goto repeat; | |
1053 | ||
1054 | /* Has the page moved? */ | |
1055 | if (unlikely(page != *slot)) { | |
1056 | page_cache_release(page); | |
1057 | goto repeat; | |
1058 | } | |
1059 | export: | |
1060 | indices[ret] = iter.index; | |
1061 | entries[ret] = page; | |
1062 | if (++ret == nr_entries) | |
1063 | break; | |
1064 | } | |
1065 | rcu_read_unlock(); | |
1066 | return ret; | |
1067 | } | |
1068 | ||
1da177e4 LT |
1069 | /** |
1070 | * find_get_pages - gang pagecache lookup | |
1071 | * @mapping: The address_space to search | |
1072 | * @start: The starting page index | |
1073 | * @nr_pages: The maximum number of pages | |
1074 | * @pages: Where the resulting pages are placed | |
1075 | * | |
1076 | * find_get_pages() will search for and return a group of up to | |
1077 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
1078 | * find_get_pages() takes a reference against the returned pages. | |
1079 | * | |
1080 | * The search returns a group of mapping-contiguous pages with ascending | |
1081 | * indexes. There may be holes in the indices due to not-present pages. | |
1082 | * | |
1083 | * find_get_pages() returns the number of pages which were found. | |
1084 | */ | |
1085 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
1086 | unsigned int nr_pages, struct page **pages) | |
1087 | { | |
0fc9d104 KK |
1088 | struct radix_tree_iter iter; |
1089 | void **slot; | |
1090 | unsigned ret = 0; | |
1091 | ||
1092 | if (unlikely(!nr_pages)) | |
1093 | return 0; | |
a60637c8 NP |
1094 | |
1095 | rcu_read_lock(); | |
1096 | restart: | |
0fc9d104 | 1097 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
a60637c8 NP |
1098 | struct page *page; |
1099 | repeat: | |
0fc9d104 | 1100 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1101 | if (unlikely(!page)) |
1102 | continue; | |
9d8aa4ea | 1103 | |
a2c16d6c | 1104 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1105 | if (radix_tree_deref_retry(page)) { |
1106 | /* | |
1107 | * Transient condition which can only trigger | |
1108 | * when entry at index 0 moves out of or back | |
1109 | * to root: none yet gotten, safe to restart. | |
1110 | */ | |
0fc9d104 | 1111 | WARN_ON(iter.index); |
8079b1c8 HD |
1112 | goto restart; |
1113 | } | |
a2c16d6c | 1114 | /* |
8079b1c8 HD |
1115 | * Otherwise, shmem/tmpfs must be storing a swap entry |
1116 | * here as an exceptional entry: so skip over it - | |
1117 | * we only reach this from invalidate_mapping_pages(). | |
a2c16d6c | 1118 | */ |
8079b1c8 | 1119 | continue; |
27d20fdd | 1120 | } |
a60637c8 NP |
1121 | |
1122 | if (!page_cache_get_speculative(page)) | |
1123 | goto repeat; | |
1124 | ||
1125 | /* Has the page moved? */ | |
0fc9d104 | 1126 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1127 | page_cache_release(page); |
1128 | goto repeat; | |
1129 | } | |
1da177e4 | 1130 | |
a60637c8 | 1131 | pages[ret] = page; |
0fc9d104 KK |
1132 | if (++ret == nr_pages) |
1133 | break; | |
a60637c8 | 1134 | } |
5b280c0c | 1135 | |
a60637c8 | 1136 | rcu_read_unlock(); |
1da177e4 LT |
1137 | return ret; |
1138 | } | |
1139 | ||
ebf43500 JA |
1140 | /** |
1141 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1142 | * @mapping: The address_space to search | |
1143 | * @index: The starting page index | |
1144 | * @nr_pages: The maximum number of pages | |
1145 | * @pages: Where the resulting pages are placed | |
1146 | * | |
1147 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1148 | * that the returned number of pages are guaranteed to be contiguous. | |
1149 | * | |
1150 | * find_get_pages_contig() returns the number of pages which were found. | |
1151 | */ | |
1152 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1153 | unsigned int nr_pages, struct page **pages) | |
1154 | { | |
0fc9d104 KK |
1155 | struct radix_tree_iter iter; |
1156 | void **slot; | |
1157 | unsigned int ret = 0; | |
1158 | ||
1159 | if (unlikely(!nr_pages)) | |
1160 | return 0; | |
a60637c8 NP |
1161 | |
1162 | rcu_read_lock(); | |
1163 | restart: | |
0fc9d104 | 1164 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
a60637c8 NP |
1165 | struct page *page; |
1166 | repeat: | |
0fc9d104 KK |
1167 | page = radix_tree_deref_slot(slot); |
1168 | /* The hole, there no reason to continue */ | |
a60637c8 | 1169 | if (unlikely(!page)) |
0fc9d104 | 1170 | break; |
9d8aa4ea | 1171 | |
a2c16d6c | 1172 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1173 | if (radix_tree_deref_retry(page)) { |
1174 | /* | |
1175 | * Transient condition which can only trigger | |
1176 | * when entry at index 0 moves out of or back | |
1177 | * to root: none yet gotten, safe to restart. | |
1178 | */ | |
1179 | goto restart; | |
1180 | } | |
a2c16d6c | 1181 | /* |
8079b1c8 HD |
1182 | * Otherwise, shmem/tmpfs must be storing a swap entry |
1183 | * here as an exceptional entry: so stop looking for | |
1184 | * contiguous pages. | |
a2c16d6c | 1185 | */ |
8079b1c8 | 1186 | break; |
a2c16d6c | 1187 | } |
ebf43500 | 1188 | |
a60637c8 NP |
1189 | if (!page_cache_get_speculative(page)) |
1190 | goto repeat; | |
1191 | ||
1192 | /* Has the page moved? */ | |
0fc9d104 | 1193 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1194 | page_cache_release(page); |
1195 | goto repeat; | |
1196 | } | |
1197 | ||
9cbb4cb2 NP |
1198 | /* |
1199 | * must check mapping and index after taking the ref. | |
1200 | * otherwise we can get both false positives and false | |
1201 | * negatives, which is just confusing to the caller. | |
1202 | */ | |
0fc9d104 | 1203 | if (page->mapping == NULL || page->index != iter.index) { |
9cbb4cb2 NP |
1204 | page_cache_release(page); |
1205 | break; | |
1206 | } | |
1207 | ||
a60637c8 | 1208 | pages[ret] = page; |
0fc9d104 KK |
1209 | if (++ret == nr_pages) |
1210 | break; | |
ebf43500 | 1211 | } |
a60637c8 NP |
1212 | rcu_read_unlock(); |
1213 | return ret; | |
ebf43500 | 1214 | } |
ef71c15c | 1215 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1216 | |
485bb99b RD |
1217 | /** |
1218 | * find_get_pages_tag - find and return pages that match @tag | |
1219 | * @mapping: the address_space to search | |
1220 | * @index: the starting page index | |
1221 | * @tag: the tag index | |
1222 | * @nr_pages: the maximum number of pages | |
1223 | * @pages: where the resulting pages are placed | |
1224 | * | |
1da177e4 | 1225 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1226 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
1227 | */ |
1228 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
1229 | int tag, unsigned int nr_pages, struct page **pages) | |
1230 | { | |
0fc9d104 KK |
1231 | struct radix_tree_iter iter; |
1232 | void **slot; | |
1233 | unsigned ret = 0; | |
1234 | ||
1235 | if (unlikely(!nr_pages)) | |
1236 | return 0; | |
a60637c8 NP |
1237 | |
1238 | rcu_read_lock(); | |
1239 | restart: | |
0fc9d104 KK |
1240 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1241 | &iter, *index, tag) { | |
a60637c8 NP |
1242 | struct page *page; |
1243 | repeat: | |
0fc9d104 | 1244 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1245 | if (unlikely(!page)) |
1246 | continue; | |
9d8aa4ea | 1247 | |
a2c16d6c | 1248 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1249 | if (radix_tree_deref_retry(page)) { |
1250 | /* | |
1251 | * Transient condition which can only trigger | |
1252 | * when entry at index 0 moves out of or back | |
1253 | * to root: none yet gotten, safe to restart. | |
1254 | */ | |
1255 | goto restart; | |
1256 | } | |
a2c16d6c | 1257 | /* |
8079b1c8 HD |
1258 | * This function is never used on a shmem/tmpfs |
1259 | * mapping, so a swap entry won't be found here. | |
a2c16d6c | 1260 | */ |
8079b1c8 | 1261 | BUG(); |
a2c16d6c | 1262 | } |
a60637c8 NP |
1263 | |
1264 | if (!page_cache_get_speculative(page)) | |
1265 | goto repeat; | |
1266 | ||
1267 | /* Has the page moved? */ | |
0fc9d104 | 1268 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1269 | page_cache_release(page); |
1270 | goto repeat; | |
1271 | } | |
1272 | ||
1273 | pages[ret] = page; | |
0fc9d104 KK |
1274 | if (++ret == nr_pages) |
1275 | break; | |
a60637c8 | 1276 | } |
5b280c0c | 1277 | |
a60637c8 | 1278 | rcu_read_unlock(); |
1da177e4 | 1279 | |
1da177e4 LT |
1280 | if (ret) |
1281 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1282 | |
1da177e4 LT |
1283 | return ret; |
1284 | } | |
ef71c15c | 1285 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1286 | |
485bb99b RD |
1287 | /** |
1288 | * grab_cache_page_nowait - returns locked page at given index in given cache | |
1289 | * @mapping: target address_space | |
1290 | * @index: the page index | |
1291 | * | |
72fd4a35 | 1292 | * Same as grab_cache_page(), but do not wait if the page is unavailable. |
1da177e4 LT |
1293 | * This is intended for speculative data generators, where the data can |
1294 | * be regenerated if the page couldn't be grabbed. This routine should | |
1295 | * be safe to call while holding the lock for another page. | |
1296 | * | |
1297 | * Clear __GFP_FS when allocating the page to avoid recursion into the fs | |
1298 | * and deadlock against the caller's locked page. | |
1299 | */ | |
1300 | struct page * | |
57f6b96c | 1301 | grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) |
1da177e4 LT |
1302 | { |
1303 | struct page *page = find_get_page(mapping, index); | |
1da177e4 LT |
1304 | |
1305 | if (page) { | |
529ae9aa | 1306 | if (trylock_page(page)) |
1da177e4 LT |
1307 | return page; |
1308 | page_cache_release(page); | |
1309 | return NULL; | |
1310 | } | |
2ae88149 | 1311 | page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); |
67d58ac4 | 1312 | if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) { |
1da177e4 LT |
1313 | page_cache_release(page); |
1314 | page = NULL; | |
1315 | } | |
1316 | return page; | |
1317 | } | |
1da177e4 LT |
1318 | EXPORT_SYMBOL(grab_cache_page_nowait); |
1319 | ||
76d42bd9 WF |
1320 | /* |
1321 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1322 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1323 | * | |
1324 | * ---R__________________________________________B__________ | |
1325 | * ^ reading here ^ bad block(assume 4k) | |
1326 | * | |
1327 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1328 | * => failing the whole request => read(R) => read(R+1) => | |
1329 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1330 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1331 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1332 | * | |
1333 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1334 | */ | |
1335 | static void shrink_readahead_size_eio(struct file *filp, | |
1336 | struct file_ra_state *ra) | |
1337 | { | |
76d42bd9 | 1338 | ra->ra_pages /= 4; |
76d42bd9 WF |
1339 | } |
1340 | ||
485bb99b | 1341 | /** |
36e78914 | 1342 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1343 | * @filp: the file to read |
1344 | * @ppos: current file position | |
1345 | * @desc: read_descriptor | |
485bb99b | 1346 | * |
1da177e4 | 1347 | * This is a generic file read routine, and uses the |
485bb99b | 1348 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1349 | * |
1350 | * This is really ugly. But the goto's actually try to clarify some | |
1351 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1352 | */ |
36e78914 | 1353 | static void do_generic_file_read(struct file *filp, loff_t *ppos, |
b77d88d4 | 1354 | read_descriptor_t *desc) |
1da177e4 | 1355 | { |
36e78914 | 1356 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1357 | struct inode *inode = mapping->host; |
36e78914 | 1358 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1359 | pgoff_t index; |
1360 | pgoff_t last_index; | |
1361 | pgoff_t prev_index; | |
1362 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1363 | unsigned int prev_offset; |
1da177e4 | 1364 | int error; |
1da177e4 | 1365 | |
1da177e4 | 1366 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1367 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1368 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
1da177e4 LT |
1369 | last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1370 | offset = *ppos & ~PAGE_CACHE_MASK; | |
1371 | ||
1da177e4 LT |
1372 | for (;;) { |
1373 | struct page *page; | |
57f6b96c | 1374 | pgoff_t end_index; |
a32ea1e1 | 1375 | loff_t isize; |
1da177e4 LT |
1376 | unsigned long nr, ret; |
1377 | ||
1da177e4 | 1378 | cond_resched(); |
1da177e4 LT |
1379 | find_page: |
1380 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1381 | if (!page) { |
cf914a7d | 1382 | page_cache_sync_readahead(mapping, |
7ff81078 | 1383 | ra, filp, |
3ea89ee8 FW |
1384 | index, last_index - index); |
1385 | page = find_get_page(mapping, index); | |
1386 | if (unlikely(page == NULL)) | |
1387 | goto no_cached_page; | |
1388 | } | |
1389 | if (PageReadahead(page)) { | |
cf914a7d | 1390 | page_cache_async_readahead(mapping, |
7ff81078 | 1391 | ra, filp, page, |
3ea89ee8 | 1392 | index, last_index - index); |
1da177e4 | 1393 | } |
8ab22b9a HH |
1394 | if (!PageUptodate(page)) { |
1395 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1396 | !mapping->a_ops->is_partially_uptodate) | |
1397 | goto page_not_up_to_date; | |
529ae9aa | 1398 | if (!trylock_page(page)) |
8ab22b9a | 1399 | goto page_not_up_to_date; |
8d056cb9 DH |
1400 | /* Did it get truncated before we got the lock? */ |
1401 | if (!page->mapping) | |
1402 | goto page_not_up_to_date_locked; | |
8ab22b9a HH |
1403 | if (!mapping->a_ops->is_partially_uptodate(page, |
1404 | desc, offset)) | |
1405 | goto page_not_up_to_date_locked; | |
1406 | unlock_page(page); | |
1407 | } | |
1da177e4 | 1408 | page_ok: |
a32ea1e1 N |
1409 | /* |
1410 | * i_size must be checked after we know the page is Uptodate. | |
1411 | * | |
1412 | * Checking i_size after the check allows us to calculate | |
1413 | * the correct value for "nr", which means the zero-filled | |
1414 | * part of the page is not copied back to userspace (unless | |
1415 | * another truncate extends the file - this is desired though). | |
1416 | */ | |
1417 | ||
1418 | isize = i_size_read(inode); | |
1419 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1420 | if (unlikely(!isize || index > end_index)) { | |
1421 | page_cache_release(page); | |
1422 | goto out; | |
1423 | } | |
1424 | ||
1425 | /* nr is the maximum number of bytes to copy from this page */ | |
1426 | nr = PAGE_CACHE_SIZE; | |
1427 | if (index == end_index) { | |
1428 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1429 | if (nr <= offset) { | |
1430 | page_cache_release(page); | |
1431 | goto out; | |
1432 | } | |
1433 | } | |
1434 | nr = nr - offset; | |
1da177e4 LT |
1435 | |
1436 | /* If users can be writing to this page using arbitrary | |
1437 | * virtual addresses, take care about potential aliasing | |
1438 | * before reading the page on the kernel side. | |
1439 | */ | |
1440 | if (mapping_writably_mapped(mapping)) | |
1441 | flush_dcache_page(page); | |
1442 | ||
1443 | /* | |
ec0f1637 JK |
1444 | * When a sequential read accesses a page several times, |
1445 | * only mark it as accessed the first time. | |
1da177e4 | 1446 | */ |
ec0f1637 | 1447 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1448 | mark_page_accessed(page); |
1449 | prev_index = index; | |
1450 | ||
1451 | /* | |
1452 | * Ok, we have the page, and it's up-to-date, so | |
1453 | * now we can copy it to user space... | |
1454 | * | |
b77d88d4 KS |
1455 | * The file_read_actor routine returns how many bytes were |
1456 | * actually used.. | |
1da177e4 LT |
1457 | * NOTE! This may not be the same as how much of a user buffer |
1458 | * we filled up (we may be padding etc), so we can only update | |
1459 | * "pos" here (the actor routine has to update the user buffer | |
1460 | * pointers and the remaining count). | |
1461 | */ | |
b77d88d4 | 1462 | ret = file_read_actor(desc, page, offset, nr); |
1da177e4 LT |
1463 | offset += ret; |
1464 | index += offset >> PAGE_CACHE_SHIFT; | |
1465 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1466 | prev_offset = offset; |
1da177e4 LT |
1467 | |
1468 | page_cache_release(page); | |
1469 | if (ret == nr && desc->count) | |
1470 | continue; | |
1471 | goto out; | |
1472 | ||
1473 | page_not_up_to_date: | |
1474 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1475 | error = lock_page_killable(page); |
1476 | if (unlikely(error)) | |
1477 | goto readpage_error; | |
1da177e4 | 1478 | |
8ab22b9a | 1479 | page_not_up_to_date_locked: |
da6052f7 | 1480 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1481 | if (!page->mapping) { |
1482 | unlock_page(page); | |
1483 | page_cache_release(page); | |
1484 | continue; | |
1485 | } | |
1486 | ||
1487 | /* Did somebody else fill it already? */ | |
1488 | if (PageUptodate(page)) { | |
1489 | unlock_page(page); | |
1490 | goto page_ok; | |
1491 | } | |
1492 | ||
1493 | readpage: | |
91803b49 JM |
1494 | /* |
1495 | * A previous I/O error may have been due to temporary | |
1496 | * failures, eg. multipath errors. | |
1497 | * PG_error will be set again if readpage fails. | |
1498 | */ | |
1499 | ClearPageError(page); | |
1da177e4 LT |
1500 | /* Start the actual read. The read will unlock the page. */ |
1501 | error = mapping->a_ops->readpage(filp, page); | |
1502 | ||
994fc28c ZB |
1503 | if (unlikely(error)) { |
1504 | if (error == AOP_TRUNCATED_PAGE) { | |
1505 | page_cache_release(page); | |
1506 | goto find_page; | |
1507 | } | |
1da177e4 | 1508 | goto readpage_error; |
994fc28c | 1509 | } |
1da177e4 LT |
1510 | |
1511 | if (!PageUptodate(page)) { | |
85462323 ON |
1512 | error = lock_page_killable(page); |
1513 | if (unlikely(error)) | |
1514 | goto readpage_error; | |
1da177e4 LT |
1515 | if (!PageUptodate(page)) { |
1516 | if (page->mapping == NULL) { | |
1517 | /* | |
2ecdc82e | 1518 | * invalidate_mapping_pages got it |
1da177e4 LT |
1519 | */ |
1520 | unlock_page(page); | |
1521 | page_cache_release(page); | |
1522 | goto find_page; | |
1523 | } | |
1524 | unlock_page(page); | |
7ff81078 | 1525 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1526 | error = -EIO; |
1527 | goto readpage_error; | |
1da177e4 LT |
1528 | } |
1529 | unlock_page(page); | |
1530 | } | |
1531 | ||
1da177e4 LT |
1532 | goto page_ok; |
1533 | ||
1534 | readpage_error: | |
1535 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1536 | desc->error = error; | |
1537 | page_cache_release(page); | |
1538 | goto out; | |
1539 | ||
1540 | no_cached_page: | |
1541 | /* | |
1542 | * Ok, it wasn't cached, so we need to create a new | |
1543 | * page.. | |
1544 | */ | |
eb2be189 NP |
1545 | page = page_cache_alloc_cold(mapping); |
1546 | if (!page) { | |
1547 | desc->error = -ENOMEM; | |
1548 | goto out; | |
1da177e4 | 1549 | } |
eb2be189 | 1550 | error = add_to_page_cache_lru(page, mapping, |
1da177e4 LT |
1551 | index, GFP_KERNEL); |
1552 | if (error) { | |
eb2be189 | 1553 | page_cache_release(page); |
1da177e4 LT |
1554 | if (error == -EEXIST) |
1555 | goto find_page; | |
1556 | desc->error = error; | |
1557 | goto out; | |
1558 | } | |
1da177e4 LT |
1559 | goto readpage; |
1560 | } | |
1561 | ||
1562 | out: | |
7ff81078 FW |
1563 | ra->prev_pos = prev_index; |
1564 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1565 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1566 | |
f4e6b498 | 1567 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1568 | file_accessed(filp); |
1da177e4 | 1569 | } |
1da177e4 LT |
1570 | |
1571 | int file_read_actor(read_descriptor_t *desc, struct page *page, | |
1572 | unsigned long offset, unsigned long size) | |
1573 | { | |
1574 | char *kaddr; | |
1575 | unsigned long left, count = desc->count; | |
1576 | ||
1577 | if (size > count) | |
1578 | size = count; | |
1579 | ||
1580 | /* | |
1581 | * Faults on the destination of a read are common, so do it before | |
1582 | * taking the kmap. | |
1583 | */ | |
1584 | if (!fault_in_pages_writeable(desc->arg.buf, size)) { | |
9b04c5fe | 1585 | kaddr = kmap_atomic(page); |
1da177e4 LT |
1586 | left = __copy_to_user_inatomic(desc->arg.buf, |
1587 | kaddr + offset, size); | |
9b04c5fe | 1588 | kunmap_atomic(kaddr); |
1da177e4 LT |
1589 | if (left == 0) |
1590 | goto success; | |
1591 | } | |
1592 | ||
1593 | /* Do it the slow way */ | |
1594 | kaddr = kmap(page); | |
1595 | left = __copy_to_user(desc->arg.buf, kaddr + offset, size); | |
1596 | kunmap(page); | |
1597 | ||
1598 | if (left) { | |
1599 | size -= left; | |
1600 | desc->error = -EFAULT; | |
1601 | } | |
1602 | success: | |
1603 | desc->count = count - size; | |
1604 | desc->written += size; | |
1605 | desc->arg.buf += size; | |
1606 | return size; | |
1607 | } | |
1608 | ||
0ceb3314 DM |
1609 | /* |
1610 | * Performs necessary checks before doing a write | |
1611 | * @iov: io vector request | |
1612 | * @nr_segs: number of segments in the iovec | |
1613 | * @count: number of bytes to write | |
1614 | * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE | |
1615 | * | |
1616 | * Adjust number of segments and amount of bytes to write (nr_segs should be | |
1617 | * properly initialized first). Returns appropriate error code that caller | |
1618 | * should return or zero in case that write should be allowed. | |
1619 | */ | |
1620 | int generic_segment_checks(const struct iovec *iov, | |
1621 | unsigned long *nr_segs, size_t *count, int access_flags) | |
1622 | { | |
1623 | unsigned long seg; | |
1624 | size_t cnt = 0; | |
1625 | for (seg = 0; seg < *nr_segs; seg++) { | |
1626 | const struct iovec *iv = &iov[seg]; | |
1627 | ||
1628 | /* | |
1629 | * If any segment has a negative length, or the cumulative | |
1630 | * length ever wraps negative then return -EINVAL. | |
1631 | */ | |
1632 | cnt += iv->iov_len; | |
1633 | if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) | |
1634 | return -EINVAL; | |
1635 | if (access_ok(access_flags, iv->iov_base, iv->iov_len)) | |
1636 | continue; | |
1637 | if (seg == 0) | |
1638 | return -EFAULT; | |
1639 | *nr_segs = seg; | |
1640 | cnt -= iv->iov_len; /* This segment is no good */ | |
1641 | break; | |
1642 | } | |
1643 | *count = cnt; | |
1644 | return 0; | |
1645 | } | |
1646 | EXPORT_SYMBOL(generic_segment_checks); | |
1647 | ||
485bb99b | 1648 | /** |
b2abacf3 | 1649 | * generic_file_aio_read - generic filesystem read routine |
485bb99b RD |
1650 | * @iocb: kernel I/O control block |
1651 | * @iov: io vector request | |
1652 | * @nr_segs: number of segments in the iovec | |
b2abacf3 | 1653 | * @pos: current file position |
485bb99b | 1654 | * |
1da177e4 LT |
1655 | * This is the "read()" routine for all filesystems |
1656 | * that can use the page cache directly. | |
1657 | */ | |
1658 | ssize_t | |
543ade1f BP |
1659 | generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
1660 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
1661 | { |
1662 | struct file *filp = iocb->ki_filp; | |
1663 | ssize_t retval; | |
66f998f6 | 1664 | unsigned long seg = 0; |
1da177e4 | 1665 | size_t count; |
543ade1f | 1666 | loff_t *ppos = &iocb->ki_pos; |
1da177e4 LT |
1667 | |
1668 | count = 0; | |
0ceb3314 DM |
1669 | retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); |
1670 | if (retval) | |
1671 | return retval; | |
1da177e4 LT |
1672 | |
1673 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
1674 | if (filp->f_flags & O_DIRECT) { | |
543ade1f | 1675 | loff_t size; |
1da177e4 LT |
1676 | struct address_space *mapping; |
1677 | struct inode *inode; | |
1678 | ||
1679 | mapping = filp->f_mapping; | |
1680 | inode = mapping->host; | |
1da177e4 LT |
1681 | if (!count) |
1682 | goto out; /* skip atime */ | |
1683 | size = i_size_read(inode); | |
9fe55eea | 1684 | retval = filemap_write_and_wait_range(mapping, pos, |
48b47c56 | 1685 | pos + iov_length(iov, nr_segs) - 1); |
9fe55eea SW |
1686 | if (!retval) { |
1687 | retval = mapping->a_ops->direct_IO(READ, iocb, | |
1688 | iov, pos, nr_segs); | |
1689 | } | |
1690 | if (retval > 0) { | |
1691 | *ppos = pos + retval; | |
1692 | count -= retval; | |
1693 | } | |
66f998f6 | 1694 | |
9fe55eea SW |
1695 | /* |
1696 | * Btrfs can have a short DIO read if we encounter | |
1697 | * compressed extents, so if there was an error, or if | |
1698 | * we've already read everything we wanted to, or if | |
1699 | * there was a short read because we hit EOF, go ahead | |
1700 | * and return. Otherwise fallthrough to buffered io for | |
1701 | * the rest of the read. | |
1702 | */ | |
1703 | if (retval < 0 || !count || *ppos >= size) { | |
1704 | file_accessed(filp); | |
1705 | goto out; | |
0e0bcae3 | 1706 | } |
1da177e4 LT |
1707 | } |
1708 | ||
66f998f6 | 1709 | count = retval; |
11fa977e HD |
1710 | for (seg = 0; seg < nr_segs; seg++) { |
1711 | read_descriptor_t desc; | |
66f998f6 JB |
1712 | loff_t offset = 0; |
1713 | ||
1714 | /* | |
1715 | * If we did a short DIO read we need to skip the section of the | |
1716 | * iov that we've already read data into. | |
1717 | */ | |
1718 | if (count) { | |
1719 | if (count > iov[seg].iov_len) { | |
1720 | count -= iov[seg].iov_len; | |
1721 | continue; | |
1722 | } | |
1723 | offset = count; | |
1724 | count = 0; | |
1725 | } | |
1da177e4 | 1726 | |
11fa977e | 1727 | desc.written = 0; |
66f998f6 JB |
1728 | desc.arg.buf = iov[seg].iov_base + offset; |
1729 | desc.count = iov[seg].iov_len - offset; | |
11fa977e HD |
1730 | if (desc.count == 0) |
1731 | continue; | |
1732 | desc.error = 0; | |
b77d88d4 | 1733 | do_generic_file_read(filp, ppos, &desc); |
11fa977e HD |
1734 | retval += desc.written; |
1735 | if (desc.error) { | |
1736 | retval = retval ?: desc.error; | |
1737 | break; | |
1da177e4 | 1738 | } |
11fa977e HD |
1739 | if (desc.count > 0) |
1740 | break; | |
1da177e4 LT |
1741 | } |
1742 | out: | |
1743 | return retval; | |
1744 | } | |
1da177e4 LT |
1745 | EXPORT_SYMBOL(generic_file_aio_read); |
1746 | ||
1da177e4 | 1747 | #ifdef CONFIG_MMU |
485bb99b RD |
1748 | /** |
1749 | * page_cache_read - adds requested page to the page cache if not already there | |
1750 | * @file: file to read | |
1751 | * @offset: page index | |
1752 | * | |
1da177e4 LT |
1753 | * This adds the requested page to the page cache if it isn't already there, |
1754 | * and schedules an I/O to read in its contents from disk. | |
1755 | */ | |
920c7a5d | 1756 | static int page_cache_read(struct file *file, pgoff_t offset) |
1da177e4 LT |
1757 | { |
1758 | struct address_space *mapping = file->f_mapping; | |
1759 | struct page *page; | |
994fc28c | 1760 | int ret; |
1da177e4 | 1761 | |
994fc28c ZB |
1762 | do { |
1763 | page = page_cache_alloc_cold(mapping); | |
1764 | if (!page) | |
1765 | return -ENOMEM; | |
1766 | ||
1767 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); | |
1768 | if (ret == 0) | |
1769 | ret = mapping->a_ops->readpage(file, page); | |
1770 | else if (ret == -EEXIST) | |
1771 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1772 | |
1da177e4 | 1773 | page_cache_release(page); |
1da177e4 | 1774 | |
994fc28c ZB |
1775 | } while (ret == AOP_TRUNCATED_PAGE); |
1776 | ||
1777 | return ret; | |
1da177e4 LT |
1778 | } |
1779 | ||
1780 | #define MMAP_LOTSAMISS (100) | |
1781 | ||
ef00e08e LT |
1782 | /* |
1783 | * Synchronous readahead happens when we don't even find | |
1784 | * a page in the page cache at all. | |
1785 | */ | |
1786 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1787 | struct file_ra_state *ra, | |
1788 | struct file *file, | |
1789 | pgoff_t offset) | |
1790 | { | |
1791 | unsigned long ra_pages; | |
1792 | struct address_space *mapping = file->f_mapping; | |
1793 | ||
1794 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1795 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 1796 | return; |
275b12bf WF |
1797 | if (!ra->ra_pages) |
1798 | return; | |
ef00e08e | 1799 | |
64363aad | 1800 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
1801 | page_cache_sync_readahead(mapping, ra, file, offset, |
1802 | ra->ra_pages); | |
ef00e08e LT |
1803 | return; |
1804 | } | |
1805 | ||
207d04ba AK |
1806 | /* Avoid banging the cache line if not needed */ |
1807 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1808 | ra->mmap_miss++; |
1809 | ||
1810 | /* | |
1811 | * Do we miss much more than hit in this file? If so, | |
1812 | * stop bothering with read-ahead. It will only hurt. | |
1813 | */ | |
1814 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1815 | return; | |
1816 | ||
d30a1100 WF |
1817 | /* |
1818 | * mmap read-around | |
1819 | */ | |
ef00e08e | 1820 | ra_pages = max_sane_readahead(ra->ra_pages); |
275b12bf WF |
1821 | ra->start = max_t(long, 0, offset - ra_pages / 2); |
1822 | ra->size = ra_pages; | |
2cbea1d3 | 1823 | ra->async_size = ra_pages / 4; |
275b12bf | 1824 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1825 | } |
1826 | ||
1827 | /* | |
1828 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1829 | * so we want to possibly extend the readahead further.. | |
1830 | */ | |
1831 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1832 | struct file_ra_state *ra, | |
1833 | struct file *file, | |
1834 | struct page *page, | |
1835 | pgoff_t offset) | |
1836 | { | |
1837 | struct address_space *mapping = file->f_mapping; | |
1838 | ||
1839 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1840 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
1841 | return; |
1842 | if (ra->mmap_miss > 0) | |
1843 | ra->mmap_miss--; | |
1844 | if (PageReadahead(page)) | |
2fad6f5d WF |
1845 | page_cache_async_readahead(mapping, ra, file, |
1846 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1847 | } |
1848 | ||
485bb99b | 1849 | /** |
54cb8821 | 1850 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1851 | * @vma: vma in which the fault was taken |
1852 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1853 | * |
54cb8821 | 1854 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1855 | * mapped memory region to read in file data during a page fault. |
1856 | * | |
1857 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1858 | * it in the page cache, and handles the special cases reasonably without | |
1859 | * having a lot of duplicated code. | |
1860 | */ | |
d0217ac0 | 1861 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
1862 | { |
1863 | int error; | |
54cb8821 | 1864 | struct file *file = vma->vm_file; |
1da177e4 LT |
1865 | struct address_space *mapping = file->f_mapping; |
1866 | struct file_ra_state *ra = &file->f_ra; | |
1867 | struct inode *inode = mapping->host; | |
ef00e08e | 1868 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 1869 | struct page *page; |
2004dc8e | 1870 | pgoff_t size; |
83c54070 | 1871 | int ret = 0; |
1da177e4 | 1872 | |
1da177e4 | 1873 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1874 | if (offset >= size) |
5307cc1a | 1875 | return VM_FAULT_SIGBUS; |
1da177e4 | 1876 | |
1da177e4 | 1877 | /* |
49426420 | 1878 | * Do we have something in the page cache already? |
1da177e4 | 1879 | */ |
ef00e08e | 1880 | page = find_get_page(mapping, offset); |
45cac65b | 1881 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 1882 | /* |
ef00e08e LT |
1883 | * We found the page, so try async readahead before |
1884 | * waiting for the lock. | |
1da177e4 | 1885 | */ |
ef00e08e | 1886 | do_async_mmap_readahead(vma, ra, file, page, offset); |
45cac65b | 1887 | } else if (!page) { |
ef00e08e LT |
1888 | /* No page in the page cache at all */ |
1889 | do_sync_mmap_readahead(vma, ra, file, offset); | |
1890 | count_vm_event(PGMAJFAULT); | |
456f998e | 1891 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
1892 | ret = VM_FAULT_MAJOR; |
1893 | retry_find: | |
b522c94d | 1894 | page = find_get_page(mapping, offset); |
1da177e4 LT |
1895 | if (!page) |
1896 | goto no_cached_page; | |
1897 | } | |
1898 | ||
d88c0922 ML |
1899 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
1900 | page_cache_release(page); | |
d065bd81 | 1901 | return ret | VM_FAULT_RETRY; |
d88c0922 | 1902 | } |
b522c94d ML |
1903 | |
1904 | /* Did it get truncated? */ | |
1905 | if (unlikely(page->mapping != mapping)) { | |
1906 | unlock_page(page); | |
1907 | put_page(page); | |
1908 | goto retry_find; | |
1909 | } | |
309381fe | 1910 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 1911 | |
1da177e4 | 1912 | /* |
d00806b1 NP |
1913 | * We have a locked page in the page cache, now we need to check |
1914 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 1915 | */ |
d00806b1 | 1916 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
1917 | goto page_not_uptodate; |
1918 | ||
ef00e08e LT |
1919 | /* |
1920 | * Found the page and have a reference on it. | |
1921 | * We must recheck i_size under page lock. | |
1922 | */ | |
d00806b1 | 1923 | size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
ef00e08e | 1924 | if (unlikely(offset >= size)) { |
d00806b1 | 1925 | unlock_page(page); |
745ad48e | 1926 | page_cache_release(page); |
5307cc1a | 1927 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
1928 | } |
1929 | ||
d0217ac0 | 1930 | vmf->page = page; |
83c54070 | 1931 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 1932 | |
1da177e4 LT |
1933 | no_cached_page: |
1934 | /* | |
1935 | * We're only likely to ever get here if MADV_RANDOM is in | |
1936 | * effect. | |
1937 | */ | |
ef00e08e | 1938 | error = page_cache_read(file, offset); |
1da177e4 LT |
1939 | |
1940 | /* | |
1941 | * The page we want has now been added to the page cache. | |
1942 | * In the unlikely event that someone removed it in the | |
1943 | * meantime, we'll just come back here and read it again. | |
1944 | */ | |
1945 | if (error >= 0) | |
1946 | goto retry_find; | |
1947 | ||
1948 | /* | |
1949 | * An error return from page_cache_read can result if the | |
1950 | * system is low on memory, or a problem occurs while trying | |
1951 | * to schedule I/O. | |
1952 | */ | |
1953 | if (error == -ENOMEM) | |
d0217ac0 NP |
1954 | return VM_FAULT_OOM; |
1955 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
1956 | |
1957 | page_not_uptodate: | |
1da177e4 LT |
1958 | /* |
1959 | * Umm, take care of errors if the page isn't up-to-date. | |
1960 | * Try to re-read it _once_. We do this synchronously, | |
1961 | * because there really aren't any performance issues here | |
1962 | * and we need to check for errors. | |
1963 | */ | |
1da177e4 | 1964 | ClearPageError(page); |
994fc28c | 1965 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
1966 | if (!error) { |
1967 | wait_on_page_locked(page); | |
1968 | if (!PageUptodate(page)) | |
1969 | error = -EIO; | |
1970 | } | |
d00806b1 NP |
1971 | page_cache_release(page); |
1972 | ||
1973 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 1974 | goto retry_find; |
1da177e4 | 1975 | |
d00806b1 | 1976 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 1977 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 1978 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
1979 | } |
1980 | EXPORT_SYMBOL(filemap_fault); | |
1981 | ||
4fcf1c62 JK |
1982 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
1983 | { | |
1984 | struct page *page = vmf->page; | |
496ad9aa | 1985 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
1986 | int ret = VM_FAULT_LOCKED; |
1987 | ||
14da9200 | 1988 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
1989 | file_update_time(vma->vm_file); |
1990 | lock_page(page); | |
1991 | if (page->mapping != inode->i_mapping) { | |
1992 | unlock_page(page); | |
1993 | ret = VM_FAULT_NOPAGE; | |
1994 | goto out; | |
1995 | } | |
14da9200 JK |
1996 | /* |
1997 | * We mark the page dirty already here so that when freeze is in | |
1998 | * progress, we are guaranteed that writeback during freezing will | |
1999 | * see the dirty page and writeprotect it again. | |
2000 | */ | |
2001 | set_page_dirty(page); | |
1d1d1a76 | 2002 | wait_for_stable_page(page); |
4fcf1c62 | 2003 | out: |
14da9200 | 2004 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2005 | return ret; |
2006 | } | |
2007 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2008 | ||
f0f37e2f | 2009 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2010 | .fault = filemap_fault, |
4fcf1c62 | 2011 | .page_mkwrite = filemap_page_mkwrite, |
0b173bc4 | 2012 | .remap_pages = generic_file_remap_pages, |
1da177e4 LT |
2013 | }; |
2014 | ||
2015 | /* This is used for a general mmap of a disk file */ | |
2016 | ||
2017 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2018 | { | |
2019 | struct address_space *mapping = file->f_mapping; | |
2020 | ||
2021 | if (!mapping->a_ops->readpage) | |
2022 | return -ENOEXEC; | |
2023 | file_accessed(file); | |
2024 | vma->vm_ops = &generic_file_vm_ops; | |
2025 | return 0; | |
2026 | } | |
1da177e4 LT |
2027 | |
2028 | /* | |
2029 | * This is for filesystems which do not implement ->writepage. | |
2030 | */ | |
2031 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2032 | { | |
2033 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2034 | return -EINVAL; | |
2035 | return generic_file_mmap(file, vma); | |
2036 | } | |
2037 | #else | |
2038 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2039 | { | |
2040 | return -ENOSYS; | |
2041 | } | |
2042 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2043 | { | |
2044 | return -ENOSYS; | |
2045 | } | |
2046 | #endif /* CONFIG_MMU */ | |
2047 | ||
2048 | EXPORT_SYMBOL(generic_file_mmap); | |
2049 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2050 | ||
6fe6900e | 2051 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 2052 | pgoff_t index, |
5e5358e7 | 2053 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2054 | void *data, |
2055 | gfp_t gfp) | |
1da177e4 | 2056 | { |
eb2be189 | 2057 | struct page *page; |
1da177e4 LT |
2058 | int err; |
2059 | repeat: | |
2060 | page = find_get_page(mapping, index); | |
2061 | if (!page) { | |
0531b2aa | 2062 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
2063 | if (!page) |
2064 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2065 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 NP |
2066 | if (unlikely(err)) { |
2067 | page_cache_release(page); | |
2068 | if (err == -EEXIST) | |
2069 | goto repeat; | |
1da177e4 | 2070 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2071 | return ERR_PTR(err); |
2072 | } | |
1da177e4 LT |
2073 | err = filler(data, page); |
2074 | if (err < 0) { | |
2075 | page_cache_release(page); | |
2076 | page = ERR_PTR(err); | |
2077 | } | |
2078 | } | |
1da177e4 LT |
2079 | return page; |
2080 | } | |
2081 | ||
0531b2aa | 2082 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2083 | pgoff_t index, |
5e5358e7 | 2084 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2085 | void *data, |
2086 | gfp_t gfp) | |
2087 | ||
1da177e4 LT |
2088 | { |
2089 | struct page *page; | |
2090 | int err; | |
2091 | ||
2092 | retry: | |
0531b2aa | 2093 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 2094 | if (IS_ERR(page)) |
c855ff37 | 2095 | return page; |
1da177e4 LT |
2096 | if (PageUptodate(page)) |
2097 | goto out; | |
2098 | ||
2099 | lock_page(page); | |
2100 | if (!page->mapping) { | |
2101 | unlock_page(page); | |
2102 | page_cache_release(page); | |
2103 | goto retry; | |
2104 | } | |
2105 | if (PageUptodate(page)) { | |
2106 | unlock_page(page); | |
2107 | goto out; | |
2108 | } | |
2109 | err = filler(data, page); | |
2110 | if (err < 0) { | |
2111 | page_cache_release(page); | |
c855ff37 | 2112 | return ERR_PTR(err); |
1da177e4 | 2113 | } |
c855ff37 | 2114 | out: |
6fe6900e NP |
2115 | mark_page_accessed(page); |
2116 | return page; | |
2117 | } | |
0531b2aa LT |
2118 | |
2119 | /** | |
2120 | * read_cache_page_async - read into page cache, fill it if needed | |
2121 | * @mapping: the page's address_space | |
2122 | * @index: the page index | |
2123 | * @filler: function to perform the read | |
5e5358e7 | 2124 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa LT |
2125 | * |
2126 | * Same as read_cache_page, but don't wait for page to become unlocked | |
2127 | * after submitting it to the filler. | |
2128 | * | |
2129 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
2130 | * not set, try to fill the page but don't wait for it to become unlocked. | |
2131 | * | |
2132 | * If the page does not get brought uptodate, return -EIO. | |
2133 | */ | |
2134 | struct page *read_cache_page_async(struct address_space *mapping, | |
2135 | pgoff_t index, | |
5e5358e7 | 2136 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2137 | void *data) |
2138 | { | |
2139 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2140 | } | |
6fe6900e NP |
2141 | EXPORT_SYMBOL(read_cache_page_async); |
2142 | ||
0531b2aa LT |
2143 | static struct page *wait_on_page_read(struct page *page) |
2144 | { | |
2145 | if (!IS_ERR(page)) { | |
2146 | wait_on_page_locked(page); | |
2147 | if (!PageUptodate(page)) { | |
2148 | page_cache_release(page); | |
2149 | page = ERR_PTR(-EIO); | |
2150 | } | |
2151 | } | |
2152 | return page; | |
2153 | } | |
2154 | ||
2155 | /** | |
2156 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2157 | * @mapping: the page's address_space | |
2158 | * @index: the page index | |
2159 | * @gfp: the page allocator flags to use if allocating | |
2160 | * | |
2161 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2162 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2163 | * |
2164 | * If the page does not get brought uptodate, return -EIO. | |
2165 | */ | |
2166 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2167 | pgoff_t index, | |
2168 | gfp_t gfp) | |
2169 | { | |
2170 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2171 | ||
2172 | return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp)); | |
2173 | } | |
2174 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2175 | ||
6fe6900e NP |
2176 | /** |
2177 | * read_cache_page - read into page cache, fill it if needed | |
2178 | * @mapping: the page's address_space | |
2179 | * @index: the page index | |
2180 | * @filler: function to perform the read | |
5e5358e7 | 2181 | * @data: first arg to filler(data, page) function, often left as NULL |
6fe6900e NP |
2182 | * |
2183 | * Read into the page cache. If a page already exists, and PageUptodate() is | |
2184 | * not set, try to fill the page then wait for it to become unlocked. | |
2185 | * | |
2186 | * If the page does not get brought uptodate, return -EIO. | |
2187 | */ | |
2188 | struct page *read_cache_page(struct address_space *mapping, | |
57f6b96c | 2189 | pgoff_t index, |
5e5358e7 | 2190 | int (*filler)(void *, struct page *), |
6fe6900e NP |
2191 | void *data) |
2192 | { | |
0531b2aa | 2193 | return wait_on_page_read(read_cache_page_async(mapping, index, filler, data)); |
1da177e4 | 2194 | } |
1da177e4 LT |
2195 | EXPORT_SYMBOL(read_cache_page); |
2196 | ||
2f718ffc | 2197 | static size_t __iovec_copy_from_user_inatomic(char *vaddr, |
1da177e4 LT |
2198 | const struct iovec *iov, size_t base, size_t bytes) |
2199 | { | |
f1800536 | 2200 | size_t copied = 0, left = 0; |
1da177e4 LT |
2201 | |
2202 | while (bytes) { | |
2203 | char __user *buf = iov->iov_base + base; | |
2204 | int copy = min(bytes, iov->iov_len - base); | |
2205 | ||
2206 | base = 0; | |
f1800536 | 2207 | left = __copy_from_user_inatomic(vaddr, buf, copy); |
1da177e4 LT |
2208 | copied += copy; |
2209 | bytes -= copy; | |
2210 | vaddr += copy; | |
2211 | iov++; | |
2212 | ||
01408c49 | 2213 | if (unlikely(left)) |
1da177e4 | 2214 | break; |
1da177e4 LT |
2215 | } |
2216 | return copied - left; | |
2217 | } | |
2218 | ||
2f718ffc NP |
2219 | /* |
2220 | * Copy as much as we can into the page and return the number of bytes which | |
af901ca1 | 2221 | * were successfully copied. If a fault is encountered then return the number of |
2f718ffc NP |
2222 | * bytes which were copied. |
2223 | */ | |
2224 | size_t iov_iter_copy_from_user_atomic(struct page *page, | |
2225 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
2226 | { | |
2227 | char *kaddr; | |
2228 | size_t copied; | |
2229 | ||
2230 | BUG_ON(!in_atomic()); | |
9b04c5fe | 2231 | kaddr = kmap_atomic(page); |
2f718ffc NP |
2232 | if (likely(i->nr_segs == 1)) { |
2233 | int left; | |
2234 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 2235 | left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); |
2f718ffc NP |
2236 | copied = bytes - left; |
2237 | } else { | |
2238 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2239 | i->iov, i->iov_offset, bytes); | |
2240 | } | |
9b04c5fe | 2241 | kunmap_atomic(kaddr); |
2f718ffc NP |
2242 | |
2243 | return copied; | |
2244 | } | |
89e10787 | 2245 | EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); |
2f718ffc NP |
2246 | |
2247 | /* | |
2248 | * This has the same sideeffects and return value as | |
2249 | * iov_iter_copy_from_user_atomic(). | |
2250 | * The difference is that it attempts to resolve faults. | |
2251 | * Page must not be locked. | |
2252 | */ | |
2253 | size_t iov_iter_copy_from_user(struct page *page, | |
2254 | struct iov_iter *i, unsigned long offset, size_t bytes) | |
2255 | { | |
2256 | char *kaddr; | |
2257 | size_t copied; | |
2258 | ||
2259 | kaddr = kmap(page); | |
2260 | if (likely(i->nr_segs == 1)) { | |
2261 | int left; | |
2262 | char __user *buf = i->iov->iov_base + i->iov_offset; | |
f1800536 | 2263 | left = __copy_from_user(kaddr + offset, buf, bytes); |
2f718ffc NP |
2264 | copied = bytes - left; |
2265 | } else { | |
2266 | copied = __iovec_copy_from_user_inatomic(kaddr + offset, | |
2267 | i->iov, i->iov_offset, bytes); | |
2268 | } | |
2269 | kunmap(page); | |
2270 | return copied; | |
2271 | } | |
89e10787 | 2272 | EXPORT_SYMBOL(iov_iter_copy_from_user); |
2f718ffc | 2273 | |
f7009264 | 2274 | void iov_iter_advance(struct iov_iter *i, size_t bytes) |
2f718ffc | 2275 | { |
f7009264 NP |
2276 | BUG_ON(i->count < bytes); |
2277 | ||
2f718ffc NP |
2278 | if (likely(i->nr_segs == 1)) { |
2279 | i->iov_offset += bytes; | |
f7009264 | 2280 | i->count -= bytes; |
2f718ffc NP |
2281 | } else { |
2282 | const struct iovec *iov = i->iov; | |
2283 | size_t base = i->iov_offset; | |
39be79c1 | 2284 | unsigned long nr_segs = i->nr_segs; |
2f718ffc | 2285 | |
124d3b70 NP |
2286 | /* |
2287 | * The !iov->iov_len check ensures we skip over unlikely | |
f7009264 | 2288 | * zero-length segments (without overruning the iovec). |
124d3b70 | 2289 | */ |
94ad374a | 2290 | while (bytes || unlikely(i->count && !iov->iov_len)) { |
f7009264 | 2291 | int copy; |
2f718ffc | 2292 | |
f7009264 NP |
2293 | copy = min(bytes, iov->iov_len - base); |
2294 | BUG_ON(!i->count || i->count < copy); | |
2295 | i->count -= copy; | |
2f718ffc NP |
2296 | bytes -= copy; |
2297 | base += copy; | |
2298 | if (iov->iov_len == base) { | |
2299 | iov++; | |
39be79c1 | 2300 | nr_segs--; |
2f718ffc NP |
2301 | base = 0; |
2302 | } | |
2303 | } | |
2304 | i->iov = iov; | |
2305 | i->iov_offset = base; | |
39be79c1 | 2306 | i->nr_segs = nr_segs; |
2f718ffc NP |
2307 | } |
2308 | } | |
89e10787 | 2309 | EXPORT_SYMBOL(iov_iter_advance); |
2f718ffc | 2310 | |
afddba49 NP |
2311 | /* |
2312 | * Fault in the first iovec of the given iov_iter, to a maximum length | |
2313 | * of bytes. Returns 0 on success, or non-zero if the memory could not be | |
2314 | * accessed (ie. because it is an invalid address). | |
2315 | * | |
2316 | * writev-intensive code may want this to prefault several iovecs -- that | |
2317 | * would be possible (callers must not rely on the fact that _only_ the | |
2318 | * first iovec will be faulted with the current implementation). | |
2319 | */ | |
2320 | int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) | |
2f718ffc | 2321 | { |
2f718ffc | 2322 | char __user *buf = i->iov->iov_base + i->iov_offset; |
afddba49 NP |
2323 | bytes = min(bytes, i->iov->iov_len - i->iov_offset); |
2324 | return fault_in_pages_readable(buf, bytes); | |
2f718ffc | 2325 | } |
89e10787 | 2326 | EXPORT_SYMBOL(iov_iter_fault_in_readable); |
2f718ffc NP |
2327 | |
2328 | /* | |
2329 | * Return the count of just the current iov_iter segment. | |
2330 | */ | |
d28574e0 | 2331 | size_t iov_iter_single_seg_count(const struct iov_iter *i) |
2f718ffc NP |
2332 | { |
2333 | const struct iovec *iov = i->iov; | |
2334 | if (i->nr_segs == 1) | |
2335 | return i->count; | |
2336 | else | |
2337 | return min(i->count, iov->iov_len - i->iov_offset); | |
2338 | } | |
89e10787 | 2339 | EXPORT_SYMBOL(iov_iter_single_seg_count); |
2f718ffc | 2340 | |
1da177e4 LT |
2341 | /* |
2342 | * Performs necessary checks before doing a write | |
2343 | * | |
485bb99b | 2344 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2345 | * Returns appropriate error code that caller should return or |
2346 | * zero in case that write should be allowed. | |
2347 | */ | |
2348 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) | |
2349 | { | |
2350 | struct inode *inode = file->f_mapping->host; | |
59e99e5b | 2351 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
1da177e4 LT |
2352 | |
2353 | if (unlikely(*pos < 0)) | |
2354 | return -EINVAL; | |
2355 | ||
1da177e4 LT |
2356 | if (!isblk) { |
2357 | /* FIXME: this is for backwards compatibility with 2.4 */ | |
2358 | if (file->f_flags & O_APPEND) | |
2359 | *pos = i_size_read(inode); | |
2360 | ||
2361 | if (limit != RLIM_INFINITY) { | |
2362 | if (*pos >= limit) { | |
2363 | send_sig(SIGXFSZ, current, 0); | |
2364 | return -EFBIG; | |
2365 | } | |
2366 | if (*count > limit - (typeof(limit))*pos) { | |
2367 | *count = limit - (typeof(limit))*pos; | |
2368 | } | |
2369 | } | |
2370 | } | |
2371 | ||
2372 | /* | |
2373 | * LFS rule | |
2374 | */ | |
2375 | if (unlikely(*pos + *count > MAX_NON_LFS && | |
2376 | !(file->f_flags & O_LARGEFILE))) { | |
2377 | if (*pos >= MAX_NON_LFS) { | |
1da177e4 LT |
2378 | return -EFBIG; |
2379 | } | |
2380 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { | |
2381 | *count = MAX_NON_LFS - (unsigned long)*pos; | |
2382 | } | |
2383 | } | |
2384 | ||
2385 | /* | |
2386 | * Are we about to exceed the fs block limit ? | |
2387 | * | |
2388 | * If we have written data it becomes a short write. If we have | |
2389 | * exceeded without writing data we send a signal and return EFBIG. | |
2390 | * Linus frestrict idea will clean these up nicely.. | |
2391 | */ | |
2392 | if (likely(!isblk)) { | |
2393 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { | |
2394 | if (*count || *pos > inode->i_sb->s_maxbytes) { | |
1da177e4 LT |
2395 | return -EFBIG; |
2396 | } | |
2397 | /* zero-length writes at ->s_maxbytes are OK */ | |
2398 | } | |
2399 | ||
2400 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) | |
2401 | *count = inode->i_sb->s_maxbytes - *pos; | |
2402 | } else { | |
9361401e | 2403 | #ifdef CONFIG_BLOCK |
1da177e4 LT |
2404 | loff_t isize; |
2405 | if (bdev_read_only(I_BDEV(inode))) | |
2406 | return -EPERM; | |
2407 | isize = i_size_read(inode); | |
2408 | if (*pos >= isize) { | |
2409 | if (*count || *pos > isize) | |
2410 | return -ENOSPC; | |
2411 | } | |
2412 | ||
2413 | if (*pos + *count > isize) | |
2414 | *count = isize - *pos; | |
9361401e DH |
2415 | #else |
2416 | return -EPERM; | |
2417 | #endif | |
1da177e4 LT |
2418 | } |
2419 | return 0; | |
2420 | } | |
2421 | EXPORT_SYMBOL(generic_write_checks); | |
2422 | ||
afddba49 NP |
2423 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2424 | loff_t pos, unsigned len, unsigned flags, | |
2425 | struct page **pagep, void **fsdata) | |
2426 | { | |
2427 | const struct address_space_operations *aops = mapping->a_ops; | |
2428 | ||
4e02ed4b | 2429 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2430 | pagep, fsdata); |
afddba49 NP |
2431 | } |
2432 | EXPORT_SYMBOL(pagecache_write_begin); | |
2433 | ||
2434 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2435 | loff_t pos, unsigned len, unsigned copied, | |
2436 | struct page *page, void *fsdata) | |
2437 | { | |
2438 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2439 | |
4e02ed4b NP |
2440 | mark_page_accessed(page); |
2441 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); | |
afddba49 NP |
2442 | } |
2443 | EXPORT_SYMBOL(pagecache_write_end); | |
2444 | ||
1da177e4 LT |
2445 | ssize_t |
2446 | generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, | |
2447 | unsigned long *nr_segs, loff_t pos, loff_t *ppos, | |
2448 | size_t count, size_t ocount) | |
2449 | { | |
2450 | struct file *file = iocb->ki_filp; | |
2451 | struct address_space *mapping = file->f_mapping; | |
2452 | struct inode *inode = mapping->host; | |
2453 | ssize_t written; | |
a969e903 CH |
2454 | size_t write_len; |
2455 | pgoff_t end; | |
1da177e4 LT |
2456 | |
2457 | if (count != ocount) | |
2458 | *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); | |
2459 | ||
a969e903 CH |
2460 | write_len = iov_length(iov, *nr_segs); |
2461 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; | |
a969e903 | 2462 | |
48b47c56 | 2463 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2464 | if (written) |
2465 | goto out; | |
2466 | ||
2467 | /* | |
2468 | * After a write we want buffered reads to be sure to go to disk to get | |
2469 | * the new data. We invalidate clean cached page from the region we're | |
2470 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2471 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2472 | */ |
2473 | if (mapping->nrpages) { | |
2474 | written = invalidate_inode_pages2_range(mapping, | |
2475 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2476 | /* |
2477 | * If a page can not be invalidated, return 0 to fall back | |
2478 | * to buffered write. | |
2479 | */ | |
2480 | if (written) { | |
2481 | if (written == -EBUSY) | |
2482 | return 0; | |
a969e903 | 2483 | goto out; |
6ccfa806 | 2484 | } |
a969e903 CH |
2485 | } |
2486 | ||
2487 | written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs); | |
2488 | ||
2489 | /* | |
2490 | * Finally, try again to invalidate clean pages which might have been | |
2491 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2492 | * if the source of the write was an mmap'ed region of the file | |
2493 | * we're writing. Either one is a pretty crazy thing to do, | |
2494 | * so we don't support it 100%. If this invalidation | |
2495 | * fails, tough, the write still worked... | |
2496 | */ | |
2497 | if (mapping->nrpages) { | |
2498 | invalidate_inode_pages2_range(mapping, | |
2499 | pos >> PAGE_CACHE_SHIFT, end); | |
2500 | } | |
2501 | ||
1da177e4 | 2502 | if (written > 0) { |
0116651c NK |
2503 | pos += written; |
2504 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { | |
2505 | i_size_write(inode, pos); | |
1da177e4 LT |
2506 | mark_inode_dirty(inode); |
2507 | } | |
0116651c | 2508 | *ppos = pos; |
1da177e4 | 2509 | } |
a969e903 | 2510 | out: |
1da177e4 LT |
2511 | return written; |
2512 | } | |
2513 | EXPORT_SYMBOL(generic_file_direct_write); | |
2514 | ||
eb2be189 NP |
2515 | /* |
2516 | * Find or create a page at the given pagecache position. Return the locked | |
2517 | * page. This function is specifically for buffered writes. | |
2518 | */ | |
54566b2c NP |
2519 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2520 | pgoff_t index, unsigned flags) | |
eb2be189 NP |
2521 | { |
2522 | int status; | |
0faa70cb | 2523 | gfp_t gfp_mask; |
eb2be189 | 2524 | struct page *page; |
54566b2c | 2525 | gfp_t gfp_notmask = 0; |
0faa70cb | 2526 | |
1010bb1b FW |
2527 | gfp_mask = mapping_gfp_mask(mapping); |
2528 | if (mapping_cap_account_dirty(mapping)) | |
2529 | gfp_mask |= __GFP_WRITE; | |
54566b2c NP |
2530 | if (flags & AOP_FLAG_NOFS) |
2531 | gfp_notmask = __GFP_FS; | |
eb2be189 NP |
2532 | repeat: |
2533 | page = find_lock_page(mapping, index); | |
c585a267 | 2534 | if (page) |
3d08bcc8 | 2535 | goto found; |
eb2be189 | 2536 | |
0faa70cb | 2537 | page = __page_cache_alloc(gfp_mask & ~gfp_notmask); |
eb2be189 NP |
2538 | if (!page) |
2539 | return NULL; | |
54566b2c NP |
2540 | status = add_to_page_cache_lru(page, mapping, index, |
2541 | GFP_KERNEL & ~gfp_notmask); | |
eb2be189 NP |
2542 | if (unlikely(status)) { |
2543 | page_cache_release(page); | |
2544 | if (status == -EEXIST) | |
2545 | goto repeat; | |
2546 | return NULL; | |
2547 | } | |
3d08bcc8 | 2548 | found: |
1d1d1a76 | 2549 | wait_for_stable_page(page); |
eb2be189 NP |
2550 | return page; |
2551 | } | |
54566b2c | 2552 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2553 | |
afddba49 NP |
2554 | static ssize_t generic_perform_write(struct file *file, |
2555 | struct iov_iter *i, loff_t pos) | |
2556 | { | |
2557 | struct address_space *mapping = file->f_mapping; | |
2558 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2559 | long status = 0; | |
2560 | ssize_t written = 0; | |
674b892e NP |
2561 | unsigned int flags = 0; |
2562 | ||
2563 | /* | |
2564 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2565 | */ | |
2566 | if (segment_eq(get_fs(), KERNEL_DS)) | |
2567 | flags |= AOP_FLAG_UNINTERRUPTIBLE; | |
afddba49 NP |
2568 | |
2569 | do { | |
2570 | struct page *page; | |
afddba49 NP |
2571 | unsigned long offset; /* Offset into pagecache page */ |
2572 | unsigned long bytes; /* Bytes to write to page */ | |
2573 | size_t copied; /* Bytes copied from user */ | |
2574 | void *fsdata; | |
2575 | ||
2576 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2577 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2578 | iov_iter_count(i)); | |
2579 | ||
2580 | again: | |
afddba49 NP |
2581 | /* |
2582 | * Bring in the user page that we will copy from _first_. | |
2583 | * Otherwise there's a nasty deadlock on copying from the | |
2584 | * same page as we're writing to, without it being marked | |
2585 | * up-to-date. | |
2586 | * | |
2587 | * Not only is this an optimisation, but it is also required | |
2588 | * to check that the address is actually valid, when atomic | |
2589 | * usercopies are used, below. | |
2590 | */ | |
2591 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2592 | status = -EFAULT; | |
2593 | break; | |
2594 | } | |
2595 | ||
674b892e | 2596 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 NP |
2597 | &page, &fsdata); |
2598 | if (unlikely(status)) | |
2599 | break; | |
2600 | ||
931e80e4 | 2601 | if (mapping_writably_mapped(mapping)) |
2602 | flush_dcache_page(page); | |
2603 | ||
afddba49 NP |
2604 | pagefault_disable(); |
2605 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); | |
2606 | pagefault_enable(); | |
2607 | flush_dcache_page(page); | |
2608 | ||
c8236db9 | 2609 | mark_page_accessed(page); |
afddba49 NP |
2610 | status = a_ops->write_end(file, mapping, pos, bytes, copied, |
2611 | page, fsdata); | |
2612 | if (unlikely(status < 0)) | |
2613 | break; | |
2614 | copied = status; | |
2615 | ||
2616 | cond_resched(); | |
2617 | ||
124d3b70 | 2618 | iov_iter_advance(i, copied); |
afddba49 NP |
2619 | if (unlikely(copied == 0)) { |
2620 | /* | |
2621 | * If we were unable to copy any data at all, we must | |
2622 | * fall back to a single segment length write. | |
2623 | * | |
2624 | * If we didn't fallback here, we could livelock | |
2625 | * because not all segments in the iov can be copied at | |
2626 | * once without a pagefault. | |
2627 | */ | |
2628 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2629 | iov_iter_single_seg_count(i)); | |
2630 | goto again; | |
2631 | } | |
afddba49 NP |
2632 | pos += copied; |
2633 | written += copied; | |
2634 | ||
2635 | balance_dirty_pages_ratelimited(mapping); | |
a50527b1 JK |
2636 | if (fatal_signal_pending(current)) { |
2637 | status = -EINTR; | |
2638 | break; | |
2639 | } | |
afddba49 NP |
2640 | } while (iov_iter_count(i)); |
2641 | ||
2642 | return written ? written : status; | |
2643 | } | |
2644 | ||
2645 | ssize_t | |
2646 | generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, | |
2647 | unsigned long nr_segs, loff_t pos, loff_t *ppos, | |
2648 | size_t count, ssize_t written) | |
2649 | { | |
2650 | struct file *file = iocb->ki_filp; | |
afddba49 NP |
2651 | ssize_t status; |
2652 | struct iov_iter i; | |
2653 | ||
2654 | iov_iter_init(&i, iov, nr_segs, count, written); | |
4e02ed4b | 2655 | status = generic_perform_write(file, &i, pos); |
1da177e4 | 2656 | |
1da177e4 | 2657 | if (likely(status >= 0)) { |
afddba49 NP |
2658 | written += status; |
2659 | *ppos = pos + status; | |
1da177e4 LT |
2660 | } |
2661 | ||
1da177e4 LT |
2662 | return written ? written : status; |
2663 | } | |
2664 | EXPORT_SYMBOL(generic_file_buffered_write); | |
2665 | ||
e4dd9de3 JK |
2666 | /** |
2667 | * __generic_file_aio_write - write data to a file | |
2668 | * @iocb: IO state structure (file, offset, etc.) | |
2669 | * @iov: vector with data to write | |
2670 | * @nr_segs: number of segments in the vector | |
2671 | * @ppos: position where to write | |
2672 | * | |
2673 | * This function does all the work needed for actually writing data to a | |
2674 | * file. It does all basic checks, removes SUID from the file, updates | |
2675 | * modification times and calls proper subroutines depending on whether we | |
2676 | * do direct IO or a standard buffered write. | |
2677 | * | |
2678 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2679 | * object which does not need locking at all. | |
2680 | * | |
2681 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2682 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2683 | * avoid syncing under i_mutex. | |
2684 | */ | |
2685 | ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
2686 | unsigned long nr_segs, loff_t *ppos) | |
1da177e4 LT |
2687 | { |
2688 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2689 | struct address_space * mapping = file->f_mapping; |
1da177e4 LT |
2690 | size_t ocount; /* original count */ |
2691 | size_t count; /* after file limit checks */ | |
2692 | struct inode *inode = mapping->host; | |
1da177e4 LT |
2693 | loff_t pos; |
2694 | ssize_t written; | |
2695 | ssize_t err; | |
2696 | ||
2697 | ocount = 0; | |
0ceb3314 DM |
2698 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
2699 | if (err) | |
2700 | return err; | |
1da177e4 LT |
2701 | |
2702 | count = ocount; | |
2703 | pos = *ppos; | |
2704 | ||
1da177e4 LT |
2705 | /* We can write back this queue in page reclaim */ |
2706 | current->backing_dev_info = mapping->backing_dev_info; | |
2707 | written = 0; | |
2708 | ||
2709 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
2710 | if (err) | |
2711 | goto out; | |
2712 | ||
2713 | if (count == 0) | |
2714 | goto out; | |
2715 | ||
2f1936b8 | 2716 | err = file_remove_suid(file); |
1da177e4 LT |
2717 | if (err) |
2718 | goto out; | |
2719 | ||
c3b2da31 JB |
2720 | err = file_update_time(file); |
2721 | if (err) | |
2722 | goto out; | |
1da177e4 LT |
2723 | |
2724 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ | |
2725 | if (unlikely(file->f_flags & O_DIRECT)) { | |
fb5527e6 JM |
2726 | loff_t endbyte; |
2727 | ssize_t written_buffered; | |
2728 | ||
2729 | written = generic_file_direct_write(iocb, iov, &nr_segs, pos, | |
2730 | ppos, count, ocount); | |
1da177e4 LT |
2731 | if (written < 0 || written == count) |
2732 | goto out; | |
2733 | /* | |
2734 | * direct-io write to a hole: fall through to buffered I/O | |
2735 | * for completing the rest of the request. | |
2736 | */ | |
2737 | pos += written; | |
2738 | count -= written; | |
fb5527e6 JM |
2739 | written_buffered = generic_file_buffered_write(iocb, iov, |
2740 | nr_segs, pos, ppos, count, | |
2741 | written); | |
2742 | /* | |
2743 | * If generic_file_buffered_write() retuned a synchronous error | |
2744 | * then we want to return the number of bytes which were | |
2745 | * direct-written, or the error code if that was zero. Note | |
2746 | * that this differs from normal direct-io semantics, which | |
2747 | * will return -EFOO even if some bytes were written. | |
2748 | */ | |
2749 | if (written_buffered < 0) { | |
2750 | err = written_buffered; | |
2751 | goto out; | |
2752 | } | |
1da177e4 | 2753 | |
fb5527e6 JM |
2754 | /* |
2755 | * We need to ensure that the page cache pages are written to | |
2756 | * disk and invalidated to preserve the expected O_DIRECT | |
2757 | * semantics. | |
2758 | */ | |
2759 | endbyte = pos + written_buffered - written - 1; | |
c05c4edd | 2760 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); |
fb5527e6 JM |
2761 | if (err == 0) { |
2762 | written = written_buffered; | |
2763 | invalidate_mapping_pages(mapping, | |
2764 | pos >> PAGE_CACHE_SHIFT, | |
2765 | endbyte >> PAGE_CACHE_SHIFT); | |
2766 | } else { | |
2767 | /* | |
2768 | * We don't know how much we wrote, so just return | |
2769 | * the number of bytes which were direct-written | |
2770 | */ | |
2771 | } | |
2772 | } else { | |
2773 | written = generic_file_buffered_write(iocb, iov, nr_segs, | |
2774 | pos, ppos, count, written); | |
2775 | } | |
1da177e4 LT |
2776 | out: |
2777 | current->backing_dev_info = NULL; | |
2778 | return written ? written : err; | |
2779 | } | |
e4dd9de3 JK |
2780 | EXPORT_SYMBOL(__generic_file_aio_write); |
2781 | ||
e4dd9de3 JK |
2782 | /** |
2783 | * generic_file_aio_write - write data to a file | |
2784 | * @iocb: IO state structure | |
2785 | * @iov: vector with data to write | |
2786 | * @nr_segs: number of segments in the vector | |
2787 | * @pos: position in file where to write | |
2788 | * | |
2789 | * This is a wrapper around __generic_file_aio_write() to be used by most | |
2790 | * filesystems. It takes care of syncing the file in case of O_SYNC file | |
2791 | * and acquires i_mutex as needed. | |
2792 | */ | |
027445c3 BP |
2793 | ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, |
2794 | unsigned long nr_segs, loff_t pos) | |
1da177e4 LT |
2795 | { |
2796 | struct file *file = iocb->ki_filp; | |
148f948b | 2797 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2798 | ssize_t ret; |
1da177e4 LT |
2799 | |
2800 | BUG_ON(iocb->ki_pos != pos); | |
2801 | ||
1b1dcc1b | 2802 | mutex_lock(&inode->i_mutex); |
e4dd9de3 | 2803 | ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); |
1b1dcc1b | 2804 | mutex_unlock(&inode->i_mutex); |
1da177e4 | 2805 | |
02afc27f | 2806 | if (ret > 0) { |
1da177e4 LT |
2807 | ssize_t err; |
2808 | ||
d311d79d AV |
2809 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
2810 | if (err < 0) | |
1da177e4 LT |
2811 | ret = err; |
2812 | } | |
2813 | return ret; | |
2814 | } | |
2815 | EXPORT_SYMBOL(generic_file_aio_write); | |
2816 | ||
cf9a2ae8 DH |
2817 | /** |
2818 | * try_to_release_page() - release old fs-specific metadata on a page | |
2819 | * | |
2820 | * @page: the page which the kernel is trying to free | |
2821 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2822 | * | |
2823 | * The address_space is to try to release any data against the page | |
2824 | * (presumably at page->private). If the release was successful, return `1'. | |
2825 | * Otherwise return zero. | |
2826 | * | |
266cf658 DH |
2827 | * This may also be called if PG_fscache is set on a page, indicating that the |
2828 | * page is known to the local caching routines. | |
2829 | * | |
cf9a2ae8 | 2830 | * The @gfp_mask argument specifies whether I/O may be performed to release |
3f31fddf | 2831 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
cf9a2ae8 | 2832 | * |
cf9a2ae8 DH |
2833 | */ |
2834 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2835 | { | |
2836 | struct address_space * const mapping = page->mapping; | |
2837 | ||
2838 | BUG_ON(!PageLocked(page)); | |
2839 | if (PageWriteback(page)) | |
2840 | return 0; | |
2841 | ||
2842 | if (mapping && mapping->a_ops->releasepage) | |
2843 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2844 | return try_to_free_buffers(page); | |
2845 | } | |
2846 | ||
2847 | EXPORT_SYMBOL(try_to_release_page); |