Root/
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 | */ |
12 | #include <linux/export.h> |
13 | #include <linux/compiler.h> |
14 | #include <linux/fs.h> |
15 | #include <linux/uaccess.h> |
16 | #include <linux/aio.h> |
17 | #include <linux/capability.h> |
18 | #include <linux/kernel_stat.h> |
19 | #include <linux/gfp.h> |
20 | #include <linux/mm.h> |
21 | #include <linux/swap.h> |
22 | #include <linux/mman.h> |
23 | #include <linux/pagemap.h> |
24 | #include <linux/file.h> |
25 | #include <linux/uio.h> |
26 | #include <linux/hash.h> |
27 | #include <linux/writeback.h> |
28 | #include <linux/backing-dev.h> |
29 | #include <linux/pagevec.h> |
30 | #include <linux/blkdev.h> |
31 | #include <linux/security.h> |
32 | #include <linux/cpuset.h> |
33 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
34 | #include <linux/memcontrol.h> |
35 | #include <linux/cleancache.h> |
36 | #include <linux/rmap.h> |
37 | #include "internal.h" |
38 | |
39 | #define CREATE_TRACE_POINTS |
40 | #include <trace/events/filemap.h> |
41 | |
42 | /* |
43 | * FIXME: remove all knowledge of the buffer layer from the core VM |
44 | */ |
45 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
46 | |
47 | #include <asm/mman.h> |
48 | |
49 | /* |
50 | * Shared mappings implemented 30.11.1994. It's not fully working yet, |
51 | * though. |
52 | * |
53 | * Shared mappings now work. 15.8.1995 Bruno. |
54 | * |
55 | * finished 'unifying' the page and buffer cache and SMP-threaded the |
56 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> |
57 | * |
58 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> |
59 | */ |
60 | |
61 | /* |
62 | * Lock ordering: |
63 | * |
64 | * ->i_mmap_mutex (truncate_pagecache) |
65 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
66 | * ->swap_lock (exclusive_swap_page, others) |
67 | * ->mapping->tree_lock |
68 | * |
69 | * ->i_mutex |
70 | * ->i_mmap_mutex (truncate->unmap_mapping_range) |
71 | * |
72 | * ->mmap_sem |
73 | * ->i_mmap_mutex |
74 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
75 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
76 | * |
77 | * ->mmap_sem |
78 | * ->lock_page (access_process_vm) |
79 | * |
80 | * ->i_mutex (generic_perform_write) |
81 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
82 | * |
83 | * bdi->wb.list_lock |
84 | * sb_lock (fs/fs-writeback.c) |
85 | * ->mapping->tree_lock (__sync_single_inode) |
86 | * |
87 | * ->i_mmap_mutex |
88 | * ->anon_vma.lock (vma_adjust) |
89 | * |
90 | * ->anon_vma.lock |
91 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
92 | * |
93 | * ->page_table_lock or pte_lock |
94 | * ->swap_lock (try_to_unmap_one) |
95 | * ->private_lock (try_to_unmap_one) |
96 | * ->tree_lock (try_to_unmap_one) |
97 | * ->zone.lru_lock (follow_page->mark_page_accessed) |
98 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
99 | * ->private_lock (page_remove_rmap->set_page_dirty) |
100 | * ->tree_lock (page_remove_rmap->set_page_dirty) |
101 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
102 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
103 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
104 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
105 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
106 | * |
107 | * ->i_mmap_mutex |
108 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
109 | */ |
110 | |
111 | static void page_cache_tree_delete(struct address_space *mapping, |
112 | struct page *page, void *shadow) |
113 | { |
114 | struct radix_tree_node *node; |
115 | unsigned long index; |
116 | unsigned int offset; |
117 | unsigned int tag; |
118 | void **slot; |
119 | |
120 | VM_BUG_ON(!PageLocked(page)); |
121 | |
122 | __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot); |
123 | |
124 | if (shadow) { |
125 | mapping->nrshadows++; |
126 | /* |
127 | * Make sure the nrshadows update is committed before |
128 | * the nrpages update so that final truncate racing |
129 | * with reclaim does not see both counters 0 at the |
130 | * same time and miss a shadow entry. |
131 | */ |
132 | smp_wmb(); |
133 | } |
134 | mapping->nrpages--; |
135 | |
136 | if (!node) { |
137 | /* Clear direct pointer tags in root node */ |
138 | mapping->page_tree.gfp_mask &= __GFP_BITS_MASK; |
139 | radix_tree_replace_slot(slot, shadow); |
140 | return; |
141 | } |
142 | |
143 | /* Clear tree tags for the removed page */ |
144 | index = page->index; |
145 | offset = index & RADIX_TREE_MAP_MASK; |
146 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { |
147 | if (test_bit(offset, node->tags[tag])) |
148 | radix_tree_tag_clear(&mapping->page_tree, index, tag); |
149 | } |
150 | |
151 | /* Delete page, swap shadow entry */ |
152 | radix_tree_replace_slot(slot, shadow); |
153 | workingset_node_pages_dec(node); |
154 | if (shadow) |
155 | workingset_node_shadows_inc(node); |
156 | else |
157 | if (__radix_tree_delete_node(&mapping->page_tree, node)) |
158 | return; |
159 | |
160 | /* |
161 | * Track node that only contains shadow entries. |
162 | * |
163 | * Avoid acquiring the list_lru lock if already tracked. The |
164 | * list_empty() test is safe as node->private_list is |
165 | * protected by mapping->tree_lock. |
166 | */ |
167 | if (!workingset_node_pages(node) && |
168 | list_empty(&node->private_list)) { |
169 | node->private_data = mapping; |
170 | list_lru_add(&workingset_shadow_nodes, &node->private_list); |
171 | } |
172 | } |
173 | |
174 | /* |
175 | * Delete a page from the page cache and free it. Caller has to make |
176 | * sure the page is locked and that nobody else uses it - or that usage |
177 | * is safe. The caller must hold the mapping's tree_lock. |
178 | */ |
179 | void __delete_from_page_cache(struct page *page, void *shadow) |
180 | { |
181 | struct address_space *mapping = page->mapping; |
182 | |
183 | trace_mm_filemap_delete_from_page_cache(page); |
184 | /* |
185 | * if we're uptodate, flush out into the cleancache, otherwise |
186 | * invalidate any existing cleancache entries. We can't leave |
187 | * stale data around in the cleancache once our page is gone |
188 | */ |
189 | if (PageUptodate(page) && PageMappedToDisk(page)) |
190 | cleancache_put_page(page); |
191 | else |
192 | cleancache_invalidate_page(mapping, page); |
193 | |
194 | page_cache_tree_delete(mapping, page, shadow); |
195 | |
196 | page->mapping = NULL; |
197 | /* Leave page->index set: truncation lookup relies upon it */ |
198 | |
199 | __dec_zone_page_state(page, NR_FILE_PAGES); |
200 | if (PageSwapBacked(page)) |
201 | __dec_zone_page_state(page, NR_SHMEM); |
202 | BUG_ON(page_mapped(page)); |
203 | |
204 | /* |
205 | * Some filesystems seem to re-dirty the page even after |
206 | * the VM has canceled the dirty bit (eg ext3 journaling). |
207 | * |
208 | * Fix it up by doing a final dirty accounting check after |
209 | * having removed the page entirely. |
210 | */ |
211 | if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { |
212 | dec_zone_page_state(page, NR_FILE_DIRTY); |
213 | dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); |
214 | } |
215 | } |
216 | |
217 | /** |
218 | * delete_from_page_cache - delete page from page cache |
219 | * @page: the page which the kernel is trying to remove from page cache |
220 | * |
221 | * This must be called only on pages that have been verified to be in the page |
222 | * cache and locked. It will never put the page into the free list, the caller |
223 | * has a reference on the page. |
224 | */ |
225 | void delete_from_page_cache(struct page *page) |
226 | { |
227 | struct address_space *mapping = page->mapping; |
228 | void (*freepage)(struct page *); |
229 | |
230 | BUG_ON(!PageLocked(page)); |
231 | |
232 | freepage = mapping->a_ops->freepage; |
233 | spin_lock_irq(&mapping->tree_lock); |
234 | __delete_from_page_cache(page, NULL); |
235 | spin_unlock_irq(&mapping->tree_lock); |
236 | mem_cgroup_uncharge_cache_page(page); |
237 | |
238 | if (freepage) |
239 | freepage(page); |
240 | page_cache_release(page); |
241 | } |
242 | EXPORT_SYMBOL(delete_from_page_cache); |
243 | |
244 | static int sleep_on_page(void *word) |
245 | { |
246 | io_schedule(); |
247 | return 0; |
248 | } |
249 | |
250 | static int sleep_on_page_killable(void *word) |
251 | { |
252 | sleep_on_page(word); |
253 | return fatal_signal_pending(current) ? -EINTR : 0; |
254 | } |
255 | |
256 | static int filemap_check_errors(struct address_space *mapping) |
257 | { |
258 | int ret = 0; |
259 | /* Check for outstanding write errors */ |
260 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
261 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) |
262 | ret = -ENOSPC; |
263 | if (test_bit(AS_EIO, &mapping->flags) && |
264 | test_and_clear_bit(AS_EIO, &mapping->flags)) |
265 | ret = -EIO; |
266 | return ret; |
267 | } |
268 | |
269 | /** |
270 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
271 | * @mapping: address space structure to write |
272 | * @start: offset in bytes where the range starts |
273 | * @end: offset in bytes where the range ends (inclusive) |
274 | * @sync_mode: enable synchronous operation |
275 | * |
276 | * Start writeback against all of a mapping's dirty pages that lie |
277 | * within the byte offsets <start, end> inclusive. |
278 | * |
279 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
280 | * opposed to a regular memory cleansing writeback. The difference between |
281 | * these two operations is that if a dirty page/buffer is encountered, it must |
282 | * be waited upon, and not just skipped over. |
283 | */ |
284 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
285 | loff_t end, int sync_mode) |
286 | { |
287 | int ret; |
288 | struct writeback_control wbc = { |
289 | .sync_mode = sync_mode, |
290 | .nr_to_write = LONG_MAX, |
291 | .range_start = start, |
292 | .range_end = end, |
293 | }; |
294 | |
295 | if (!mapping_cap_writeback_dirty(mapping)) |
296 | return 0; |
297 | |
298 | ret = do_writepages(mapping, &wbc); |
299 | return ret; |
300 | } |
301 | |
302 | static inline int __filemap_fdatawrite(struct address_space *mapping, |
303 | int sync_mode) |
304 | { |
305 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
306 | } |
307 | |
308 | int filemap_fdatawrite(struct address_space *mapping) |
309 | { |
310 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); |
311 | } |
312 | EXPORT_SYMBOL(filemap_fdatawrite); |
313 | |
314 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
315 | loff_t end) |
316 | { |
317 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); |
318 | } |
319 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
320 | |
321 | /** |
322 | * filemap_flush - mostly a non-blocking flush |
323 | * @mapping: target address_space |
324 | * |
325 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
326 | * purposes - I/O may not be started against all dirty pages. |
327 | */ |
328 | int filemap_flush(struct address_space *mapping) |
329 | { |
330 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); |
331 | } |
332 | EXPORT_SYMBOL(filemap_flush); |
333 | |
334 | /** |
335 | * filemap_fdatawait_range - wait for writeback to complete |
336 | * @mapping: address space structure to wait for |
337 | * @start_byte: offset in bytes where the range starts |
338 | * @end_byte: offset in bytes where the range ends (inclusive) |
339 | * |
340 | * Walk the list of under-writeback pages of the given address space |
341 | * in the given range and wait for all of them. |
342 | */ |
343 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, |
344 | loff_t end_byte) |
345 | { |
346 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
347 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; |
348 | struct pagevec pvec; |
349 | int nr_pages; |
350 | int ret2, ret = 0; |
351 | |
352 | if (end_byte < start_byte) |
353 | goto out; |
354 | |
355 | pagevec_init(&pvec, 0); |
356 | while ((index <= end) && |
357 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, |
358 | PAGECACHE_TAG_WRITEBACK, |
359 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { |
360 | unsigned i; |
361 | |
362 | for (i = 0; i < nr_pages; i++) { |
363 | struct page *page = pvec.pages[i]; |
364 | |
365 | /* until radix tree lookup accepts end_index */ |
366 | if (page->index > end) |
367 | continue; |
368 | |
369 | wait_on_page_writeback(page); |
370 | if (TestClearPageError(page)) |
371 | ret = -EIO; |
372 | } |
373 | pagevec_release(&pvec); |
374 | cond_resched(); |
375 | } |
376 | out: |
377 | ret2 = filemap_check_errors(mapping); |
378 | if (!ret) |
379 | ret = ret2; |
380 | |
381 | return ret; |
382 | } |
383 | EXPORT_SYMBOL(filemap_fdatawait_range); |
384 | |
385 | /** |
386 | * filemap_fdatawait - wait for all under-writeback pages to complete |
387 | * @mapping: address space structure to wait for |
388 | * |
389 | * Walk the list of under-writeback pages of the given address space |
390 | * and wait for all of them. |
391 | */ |
392 | int filemap_fdatawait(struct address_space *mapping) |
393 | { |
394 | loff_t i_size = i_size_read(mapping->host); |
395 | |
396 | if (i_size == 0) |
397 | return 0; |
398 | |
399 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
400 | } |
401 | EXPORT_SYMBOL(filemap_fdatawait); |
402 | |
403 | int filemap_write_and_wait(struct address_space *mapping) |
404 | { |
405 | int err = 0; |
406 | |
407 | if (mapping->nrpages) { |
408 | err = filemap_fdatawrite(mapping); |
409 | /* |
410 | * Even if the above returned error, the pages may be |
411 | * written partially (e.g. -ENOSPC), so we wait for it. |
412 | * But the -EIO is special case, it may indicate the worst |
413 | * thing (e.g. bug) happened, so we avoid waiting for it. |
414 | */ |
415 | if (err != -EIO) { |
416 | int err2 = filemap_fdatawait(mapping); |
417 | if (!err) |
418 | err = err2; |
419 | } |
420 | } else { |
421 | err = filemap_check_errors(mapping); |
422 | } |
423 | return err; |
424 | } |
425 | EXPORT_SYMBOL(filemap_write_and_wait); |
426 | |
427 | /** |
428 | * filemap_write_and_wait_range - write out & wait on a file range |
429 | * @mapping: the address_space for the pages |
430 | * @lstart: offset in bytes where the range starts |
431 | * @lend: offset in bytes where the range ends (inclusive) |
432 | * |
433 | * Write out and wait upon file offsets lstart->lend, inclusive. |
434 | * |
435 | * Note that `lend' is inclusive (describes the last byte to be written) so |
436 | * that this function can be used to write to the very end-of-file (end = -1). |
437 | */ |
438 | int filemap_write_and_wait_range(struct address_space *mapping, |
439 | loff_t lstart, loff_t lend) |
440 | { |
441 | int err = 0; |
442 | |
443 | if (mapping->nrpages) { |
444 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
445 | WB_SYNC_ALL); |
446 | /* See comment of filemap_write_and_wait() */ |
447 | if (err != -EIO) { |
448 | int err2 = filemap_fdatawait_range(mapping, |
449 | lstart, lend); |
450 | if (!err) |
451 | err = err2; |
452 | } |
453 | } else { |
454 | err = filemap_check_errors(mapping); |
455 | } |
456 | return err; |
457 | } |
458 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
459 | |
460 | /** |
461 | * replace_page_cache_page - replace a pagecache page with a new one |
462 | * @old: page to be replaced |
463 | * @new: page to replace with |
464 | * @gfp_mask: allocation mode |
465 | * |
466 | * This function replaces a page in the pagecache with a new one. On |
467 | * success it acquires the pagecache reference for the new page and |
468 | * drops it for the old page. Both the old and new pages must be |
469 | * locked. This function does not add the new page to the LRU, the |
470 | * caller must do that. |
471 | * |
472 | * The remove + add is atomic. The only way this function can fail is |
473 | * memory allocation failure. |
474 | */ |
475 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) |
476 | { |
477 | int error; |
478 | |
479 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
480 | VM_BUG_ON_PAGE(!PageLocked(new), new); |
481 | VM_BUG_ON_PAGE(new->mapping, new); |
482 | |
483 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
484 | if (!error) { |
485 | struct address_space *mapping = old->mapping; |
486 | void (*freepage)(struct page *); |
487 | |
488 | pgoff_t offset = old->index; |
489 | freepage = mapping->a_ops->freepage; |
490 | |
491 | page_cache_get(new); |
492 | new->mapping = mapping; |
493 | new->index = offset; |
494 | |
495 | spin_lock_irq(&mapping->tree_lock); |
496 | __delete_from_page_cache(old, NULL); |
497 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
498 | BUG_ON(error); |
499 | mapping->nrpages++; |
500 | __inc_zone_page_state(new, NR_FILE_PAGES); |
501 | if (PageSwapBacked(new)) |
502 | __inc_zone_page_state(new, NR_SHMEM); |
503 | spin_unlock_irq(&mapping->tree_lock); |
504 | /* mem_cgroup codes must not be called under tree_lock */ |
505 | mem_cgroup_replace_page_cache(old, new); |
506 | radix_tree_preload_end(); |
507 | if (freepage) |
508 | freepage(old); |
509 | page_cache_release(old); |
510 | } |
511 | |
512 | return error; |
513 | } |
514 | EXPORT_SYMBOL_GPL(replace_page_cache_page); |
515 | |
516 | static int page_cache_tree_insert(struct address_space *mapping, |
517 | struct page *page, void **shadowp) |
518 | { |
519 | struct radix_tree_node *node; |
520 | void **slot; |
521 | int error; |
522 | |
523 | error = __radix_tree_create(&mapping->page_tree, page->index, |
524 | &node, &slot); |
525 | if (error) |
526 | return error; |
527 | if (*slot) { |
528 | void *p; |
529 | |
530 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); |
531 | if (!radix_tree_exceptional_entry(p)) |
532 | return -EEXIST; |
533 | if (shadowp) |
534 | *shadowp = p; |
535 | mapping->nrshadows--; |
536 | if (node) |
537 | workingset_node_shadows_dec(node); |
538 | } |
539 | radix_tree_replace_slot(slot, page); |
540 | mapping->nrpages++; |
541 | if (node) { |
542 | workingset_node_pages_inc(node); |
543 | /* |
544 | * Don't track node that contains actual pages. |
545 | * |
546 | * Avoid acquiring the list_lru lock if already |
547 | * untracked. The list_empty() test is safe as |
548 | * node->private_list is protected by |
549 | * mapping->tree_lock. |
550 | */ |
551 | if (!list_empty(&node->private_list)) |
552 | list_lru_del(&workingset_shadow_nodes, |
553 | &node->private_list); |
554 | } |
555 | return 0; |
556 | } |
557 | |
558 | static int __add_to_page_cache_locked(struct page *page, |
559 | struct address_space *mapping, |
560 | pgoff_t offset, gfp_t gfp_mask, |
561 | void **shadowp) |
562 | { |
563 | int error; |
564 | |
565 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
566 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); |
567 | |
568 | error = mem_cgroup_charge_file(page, current->mm, |
569 | gfp_mask & GFP_RECLAIM_MASK); |
570 | if (error) |
571 | return error; |
572 | |
573 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
574 | if (error) { |
575 | mem_cgroup_uncharge_cache_page(page); |
576 | return error; |
577 | } |
578 | |
579 | page_cache_get(page); |
580 | page->mapping = mapping; |
581 | page->index = offset; |
582 | |
583 | spin_lock_irq(&mapping->tree_lock); |
584 | error = page_cache_tree_insert(mapping, page, shadowp); |
585 | radix_tree_preload_end(); |
586 | if (unlikely(error)) |
587 | goto err_insert; |
588 | __inc_zone_page_state(page, NR_FILE_PAGES); |
589 | spin_unlock_irq(&mapping->tree_lock); |
590 | trace_mm_filemap_add_to_page_cache(page); |
591 | return 0; |
592 | err_insert: |
593 | page->mapping = NULL; |
594 | /* Leave page->index set: truncation relies upon it */ |
595 | spin_unlock_irq(&mapping->tree_lock); |
596 | mem_cgroup_uncharge_cache_page(page); |
597 | page_cache_release(page); |
598 | return error; |
599 | } |
600 | |
601 | /** |
602 | * add_to_page_cache_locked - add a locked page to the pagecache |
603 | * @page: page to add |
604 | * @mapping: the page's address_space |
605 | * @offset: page index |
606 | * @gfp_mask: page allocation mode |
607 | * |
608 | * This function is used to add a page to the pagecache. It must be locked. |
609 | * This function does not add the page to the LRU. The caller must do that. |
610 | */ |
611 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, |
612 | pgoff_t offset, gfp_t gfp_mask) |
613 | { |
614 | return __add_to_page_cache_locked(page, mapping, offset, |
615 | gfp_mask, NULL); |
616 | } |
617 | EXPORT_SYMBOL(add_to_page_cache_locked); |
618 | |
619 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, |
620 | pgoff_t offset, gfp_t gfp_mask) |
621 | { |
622 | void *shadow = NULL; |
623 | int ret; |
624 | |
625 | __set_page_locked(page); |
626 | ret = __add_to_page_cache_locked(page, mapping, offset, |
627 | gfp_mask, &shadow); |
628 | if (unlikely(ret)) |
629 | __clear_page_locked(page); |
630 | else { |
631 | /* |
632 | * The page might have been evicted from cache only |
633 | * recently, in which case it should be activated like |
634 | * any other repeatedly accessed page. |
635 | */ |
636 | if (shadow && workingset_refault(shadow)) { |
637 | SetPageActive(page); |
638 | workingset_activation(page); |
639 | } else |
640 | ClearPageActive(page); |
641 | lru_cache_add(page); |
642 | } |
643 | return ret; |
644 | } |
645 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
646 | |
647 | #ifdef CONFIG_NUMA |
648 | struct page *__page_cache_alloc(gfp_t gfp) |
649 | { |
650 | int n; |
651 | struct page *page; |
652 | |
653 | if (cpuset_do_page_mem_spread()) { |
654 | unsigned int cpuset_mems_cookie; |
655 | do { |
656 | cpuset_mems_cookie = read_mems_allowed_begin(); |
657 | n = cpuset_mem_spread_node(); |
658 | page = alloc_pages_exact_node(n, gfp, 0); |
659 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
660 | |
661 | return page; |
662 | } |
663 | return alloc_pages(gfp, 0); |
664 | } |
665 | EXPORT_SYMBOL(__page_cache_alloc); |
666 | #endif |
667 | |
668 | /* |
669 | * In order to wait for pages to become available there must be |
670 | * waitqueues associated with pages. By using a hash table of |
671 | * waitqueues where the bucket discipline is to maintain all |
672 | * waiters on the same queue and wake all when any of the pages |
673 | * become available, and for the woken contexts to check to be |
674 | * sure the appropriate page became available, this saves space |
675 | * at a cost of "thundering herd" phenomena during rare hash |
676 | * collisions. |
677 | */ |
678 | static wait_queue_head_t *page_waitqueue(struct page *page) |
679 | { |
680 | const struct zone *zone = page_zone(page); |
681 | |
682 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; |
683 | } |
684 | |
685 | static inline void wake_up_page(struct page *page, int bit) |
686 | { |
687 | __wake_up_bit(page_waitqueue(page), &page->flags, bit); |
688 | } |
689 | |
690 | void wait_on_page_bit(struct page *page, int bit_nr) |
691 | { |
692 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); |
693 | |
694 | if (test_bit(bit_nr, &page->flags)) |
695 | __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page, |
696 | TASK_UNINTERRUPTIBLE); |
697 | } |
698 | EXPORT_SYMBOL(wait_on_page_bit); |
699 | |
700 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
701 | { |
702 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); |
703 | |
704 | if (!test_bit(bit_nr, &page->flags)) |
705 | return 0; |
706 | |
707 | return __wait_on_bit(page_waitqueue(page), &wait, |
708 | sleep_on_page_killable, TASK_KILLABLE); |
709 | } |
710 | |
711 | /** |
712 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue |
713 | * @page: Page defining the wait queue of interest |
714 | * @waiter: Waiter to add to the queue |
715 | * |
716 | * Add an arbitrary @waiter to the wait queue for the nominated @page. |
717 | */ |
718 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) |
719 | { |
720 | wait_queue_head_t *q = page_waitqueue(page); |
721 | unsigned long flags; |
722 | |
723 | spin_lock_irqsave(&q->lock, flags); |
724 | __add_wait_queue(q, waiter); |
725 | spin_unlock_irqrestore(&q->lock, flags); |
726 | } |
727 | EXPORT_SYMBOL_GPL(add_page_wait_queue); |
728 | |
729 | /** |
730 | * unlock_page - unlock a locked page |
731 | * @page: the page |
732 | * |
733 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). |
734 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup |
735 | * mechananism between PageLocked pages and PageWriteback pages is shared. |
736 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
737 | * |
738 | * The mb is necessary to enforce ordering between the clear_bit and the read |
739 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). |
740 | */ |
741 | void unlock_page(struct page *page) |
742 | { |
743 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
744 | clear_bit_unlock(PG_locked, &page->flags); |
745 | smp_mb__after_atomic(); |
746 | wake_up_page(page, PG_locked); |
747 | } |
748 | EXPORT_SYMBOL(unlock_page); |
749 | |
750 | /** |
751 | * end_page_writeback - end writeback against a page |
752 | * @page: the page |
753 | */ |
754 | void end_page_writeback(struct page *page) |
755 | { |
756 | /* |
757 | * TestClearPageReclaim could be used here but it is an atomic |
758 | * operation and overkill in this particular case. Failing to |
759 | * shuffle a page marked for immediate reclaim is too mild to |
760 | * justify taking an atomic operation penalty at the end of |
761 | * ever page writeback. |
762 | */ |
763 | if (PageReclaim(page)) { |
764 | ClearPageReclaim(page); |
765 | rotate_reclaimable_page(page); |
766 | } |
767 | |
768 | if (!test_clear_page_writeback(page)) |
769 | BUG(); |
770 | |
771 | smp_mb__after_atomic(); |
772 | wake_up_page(page, PG_writeback); |
773 | } |
774 | EXPORT_SYMBOL(end_page_writeback); |
775 | |
776 | /* |
777 | * After completing I/O on a page, call this routine to update the page |
778 | * flags appropriately |
779 | */ |
780 | void page_endio(struct page *page, int rw, int err) |
781 | { |
782 | if (rw == READ) { |
783 | if (!err) { |
784 | SetPageUptodate(page); |
785 | } else { |
786 | ClearPageUptodate(page); |
787 | SetPageError(page); |
788 | } |
789 | unlock_page(page); |
790 | } else { /* rw == WRITE */ |
791 | if (err) { |
792 | SetPageError(page); |
793 | if (page->mapping) |
794 | mapping_set_error(page->mapping, err); |
795 | } |
796 | end_page_writeback(page); |
797 | } |
798 | } |
799 | EXPORT_SYMBOL_GPL(page_endio); |
800 | |
801 | /** |
802 | * __lock_page - get a lock on the page, assuming we need to sleep to get it |
803 | * @page: the page to lock |
804 | */ |
805 | void __lock_page(struct page *page) |
806 | { |
807 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); |
808 | |
809 | __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page, |
810 | TASK_UNINTERRUPTIBLE); |
811 | } |
812 | EXPORT_SYMBOL(__lock_page); |
813 | |
814 | int __lock_page_killable(struct page *page) |
815 | { |
816 | DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); |
817 | |
818 | return __wait_on_bit_lock(page_waitqueue(page), &wait, |
819 | sleep_on_page_killable, TASK_KILLABLE); |
820 | } |
821 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
822 | |
823 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
824 | unsigned int flags) |
825 | { |
826 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
827 | /* |
828 | * CAUTION! In this case, mmap_sem is not released |
829 | * even though return 0. |
830 | */ |
831 | if (flags & FAULT_FLAG_RETRY_NOWAIT) |
832 | return 0; |
833 | |
834 | up_read(&mm->mmap_sem); |
835 | if (flags & FAULT_FLAG_KILLABLE) |
836 | wait_on_page_locked_killable(page); |
837 | else |
838 | wait_on_page_locked(page); |
839 | return 0; |
840 | } else { |
841 | if (flags & FAULT_FLAG_KILLABLE) { |
842 | int ret; |
843 | |
844 | ret = __lock_page_killable(page); |
845 | if (ret) { |
846 | up_read(&mm->mmap_sem); |
847 | return 0; |
848 | } |
849 | } else |
850 | __lock_page(page); |
851 | return 1; |
852 | } |
853 | } |
854 | |
855 | /** |
856 | * page_cache_next_hole - find the next hole (not-present entry) |
857 | * @mapping: mapping |
858 | * @index: index |
859 | * @max_scan: maximum range to search |
860 | * |
861 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the |
862 | * lowest indexed hole. |
863 | * |
864 | * Returns: the index of the hole if found, otherwise returns an index |
865 | * outside of the set specified (in which case 'return - index >= |
866 | * max_scan' will be true). In rare cases of index wrap-around, 0 will |
867 | * be returned. |
868 | * |
869 | * page_cache_next_hole may be called under rcu_read_lock. However, |
870 | * like radix_tree_gang_lookup, this will not atomically search a |
871 | * snapshot of the tree at a single point in time. For example, if a |
872 | * hole is created at index 5, then subsequently a hole is created at |
873 | * index 10, page_cache_next_hole covering both indexes may return 10 |
874 | * if called under rcu_read_lock. |
875 | */ |
876 | pgoff_t page_cache_next_hole(struct address_space *mapping, |
877 | pgoff_t index, unsigned long max_scan) |
878 | { |
879 | unsigned long i; |
880 | |
881 | for (i = 0; i < max_scan; i++) { |
882 | struct page *page; |
883 | |
884 | page = radix_tree_lookup(&mapping->page_tree, index); |
885 | if (!page || radix_tree_exceptional_entry(page)) |
886 | break; |
887 | index++; |
888 | if (index == 0) |
889 | break; |
890 | } |
891 | |
892 | return index; |
893 | } |
894 | EXPORT_SYMBOL(page_cache_next_hole); |
895 | |
896 | /** |
897 | * page_cache_prev_hole - find the prev hole (not-present entry) |
898 | * @mapping: mapping |
899 | * @index: index |
900 | * @max_scan: maximum range to search |
901 | * |
902 | * Search backwards in the range [max(index-max_scan+1, 0), index] for |
903 | * the first hole. |
904 | * |
905 | * Returns: the index of the hole if found, otherwise returns an index |
906 | * outside of the set specified (in which case 'index - return >= |
907 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX |
908 | * will be returned. |
909 | * |
910 | * page_cache_prev_hole may be called under rcu_read_lock. However, |
911 | * like radix_tree_gang_lookup, this will not atomically search a |
912 | * snapshot of the tree at a single point in time. For example, if a |
913 | * hole is created at index 10, then subsequently a hole is created at |
914 | * index 5, page_cache_prev_hole covering both indexes may return 5 if |
915 | * called under rcu_read_lock. |
916 | */ |
917 | pgoff_t page_cache_prev_hole(struct address_space *mapping, |
918 | pgoff_t index, unsigned long max_scan) |
919 | { |
920 | unsigned long i; |
921 | |
922 | for (i = 0; i < max_scan; i++) { |
923 | struct page *page; |
924 | |
925 | page = radix_tree_lookup(&mapping->page_tree, index); |
926 | if (!page || radix_tree_exceptional_entry(page)) |
927 | break; |
928 | index--; |
929 | if (index == ULONG_MAX) |
930 | break; |
931 | } |
932 | |
933 | return index; |
934 | } |
935 | EXPORT_SYMBOL(page_cache_prev_hole); |
936 | |
937 | /** |
938 | * find_get_entry - find and get a page cache entry |
939 | * @mapping: the address_space to search |
940 | * @offset: the page cache index |
941 | * |
942 | * Looks up the page cache slot at @mapping & @offset. If there is a |
943 | * page cache page, it is returned with an increased refcount. |
944 | * |
945 | * If the slot holds a shadow entry of a previously evicted page, or a |
946 | * swap entry from shmem/tmpfs, it is returned. |
947 | * |
948 | * Otherwise, %NULL is returned. |
949 | */ |
950 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
951 | { |
952 | void **pagep; |
953 | struct page *page; |
954 | |
955 | rcu_read_lock(); |
956 | repeat: |
957 | page = NULL; |
958 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); |
959 | if (pagep) { |
960 | page = radix_tree_deref_slot(pagep); |
961 | if (unlikely(!page)) |
962 | goto out; |
963 | if (radix_tree_exception(page)) { |
964 | if (radix_tree_deref_retry(page)) |
965 | goto repeat; |
966 | /* |
967 | * A shadow entry of a recently evicted page, |
968 | * or a swap entry from shmem/tmpfs. Return |
969 | * it without attempting to raise page count. |
970 | */ |
971 | goto out; |
972 | } |
973 | if (!page_cache_get_speculative(page)) |
974 | goto repeat; |
975 | |
976 | /* |
977 | * Has the page moved? |
978 | * This is part of the lockless pagecache protocol. See |
979 | * include/linux/pagemap.h for details. |
980 | */ |
981 | if (unlikely(page != *pagep)) { |
982 | page_cache_release(page); |
983 | goto repeat; |
984 | } |
985 | } |
986 | out: |
987 | rcu_read_unlock(); |
988 | |
989 | return page; |
990 | } |
991 | EXPORT_SYMBOL(find_get_entry); |
992 | |
993 | /** |
994 | * find_lock_entry - locate, pin and lock a page cache entry |
995 | * @mapping: the address_space to search |
996 | * @offset: the page cache index |
997 | * |
998 | * Looks up the page cache slot at @mapping & @offset. If there is a |
999 | * page cache page, it is returned locked and with an increased |
1000 | * refcount. |
1001 | * |
1002 | * If the slot holds a shadow entry of a previously evicted page, or a |
1003 | * swap entry from shmem/tmpfs, it is returned. |
1004 | * |
1005 | * Otherwise, %NULL is returned. |
1006 | * |
1007 | * find_lock_entry() may sleep. |
1008 | */ |
1009 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) |
1010 | { |
1011 | struct page *page; |
1012 | |
1013 | repeat: |
1014 | page = find_get_entry(mapping, offset); |
1015 | if (page && !radix_tree_exception(page)) { |
1016 | lock_page(page); |
1017 | /* Has the page been truncated? */ |
1018 | if (unlikely(page->mapping != mapping)) { |
1019 | unlock_page(page); |
1020 | page_cache_release(page); |
1021 | goto repeat; |
1022 | } |
1023 | VM_BUG_ON_PAGE(page->index != offset, page); |
1024 | } |
1025 | return page; |
1026 | } |
1027 | EXPORT_SYMBOL(find_lock_entry); |
1028 | |
1029 | /** |
1030 | * pagecache_get_page - find and get a page reference |
1031 | * @mapping: the address_space to search |
1032 | * @offset: the page index |
1033 | * @fgp_flags: PCG flags |
1034 | * @cache_gfp_mask: gfp mask to use for the page cache data page allocation |
1035 | * @radix_gfp_mask: gfp mask to use for radix tree node allocation |
1036 | * |
1037 | * Looks up the page cache slot at @mapping & @offset. |
1038 | * |
1039 | * PCG flags modify how the page is returned. |
1040 | * |
1041 | * FGP_ACCESSED: the page will be marked accessed |
1042 | * FGP_LOCK: Page is return locked |
1043 | * FGP_CREAT: If page is not present then a new page is allocated using |
1044 | * @cache_gfp_mask and added to the page cache and the VM's LRU |
1045 | * list. If radix tree nodes are allocated during page cache |
1046 | * insertion then @radix_gfp_mask is used. The page is returned |
1047 | * locked and with an increased refcount. Otherwise, %NULL is |
1048 | * returned. |
1049 | * |
1050 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1051 | * if the GFP flags specified for FGP_CREAT are atomic. |
1052 | * |
1053 | * If there is a page cache page, it is returned with an increased refcount. |
1054 | */ |
1055 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
1056 | int fgp_flags, gfp_t cache_gfp_mask, gfp_t radix_gfp_mask) |
1057 | { |
1058 | struct page *page; |
1059 | |
1060 | repeat: |
1061 | page = find_get_entry(mapping, offset); |
1062 | if (radix_tree_exceptional_entry(page)) |
1063 | page = NULL; |
1064 | if (!page) |
1065 | goto no_page; |
1066 | |
1067 | if (fgp_flags & FGP_LOCK) { |
1068 | if (fgp_flags & FGP_NOWAIT) { |
1069 | if (!trylock_page(page)) { |
1070 | page_cache_release(page); |
1071 | return NULL; |
1072 | } |
1073 | } else { |
1074 | lock_page(page); |
1075 | } |
1076 | |
1077 | /* Has the page been truncated? */ |
1078 | if (unlikely(page->mapping != mapping)) { |
1079 | unlock_page(page); |
1080 | page_cache_release(page); |
1081 | goto repeat; |
1082 | } |
1083 | VM_BUG_ON_PAGE(page->index != offset, page); |
1084 | } |
1085 | |
1086 | if (page && (fgp_flags & FGP_ACCESSED)) |
1087 | mark_page_accessed(page); |
1088 | |
1089 | no_page: |
1090 | if (!page && (fgp_flags & FGP_CREAT)) { |
1091 | int err; |
1092 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) |
1093 | cache_gfp_mask |= __GFP_WRITE; |
1094 | if (fgp_flags & FGP_NOFS) { |
1095 | cache_gfp_mask &= ~__GFP_FS; |
1096 | radix_gfp_mask &= ~__GFP_FS; |
1097 | } |
1098 | |
1099 | page = __page_cache_alloc(cache_gfp_mask); |
1100 | if (!page) |
1101 | return NULL; |
1102 | |
1103 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) |
1104 | fgp_flags |= FGP_LOCK; |
1105 | |
1106 | /* Init accessed so avoit atomic mark_page_accessed later */ |
1107 | if (fgp_flags & FGP_ACCESSED) |
1108 | init_page_accessed(page); |
1109 | |
1110 | err = add_to_page_cache_lru(page, mapping, offset, radix_gfp_mask); |
1111 | if (unlikely(err)) { |
1112 | page_cache_release(page); |
1113 | page = NULL; |
1114 | if (err == -EEXIST) |
1115 | goto repeat; |
1116 | } |
1117 | } |
1118 | |
1119 | return page; |
1120 | } |
1121 | EXPORT_SYMBOL(pagecache_get_page); |
1122 | |
1123 | /** |
1124 | * find_get_entries - gang pagecache lookup |
1125 | * @mapping: The address_space to search |
1126 | * @start: The starting page cache index |
1127 | * @nr_entries: The maximum number of entries |
1128 | * @entries: Where the resulting entries are placed |
1129 | * @indices: The cache indices corresponding to the entries in @entries |
1130 | * |
1131 | * find_get_entries() will search for and return a group of up to |
1132 | * @nr_entries entries in the mapping. The entries are placed at |
1133 | * @entries. find_get_entries() takes a reference against any actual |
1134 | * pages it returns. |
1135 | * |
1136 | * The search returns a group of mapping-contiguous page cache entries |
1137 | * with ascending indexes. There may be holes in the indices due to |
1138 | * not-present pages. |
1139 | * |
1140 | * Any shadow entries of evicted pages, or swap entries from |
1141 | * shmem/tmpfs, are included in the returned array. |
1142 | * |
1143 | * find_get_entries() returns the number of pages and shadow entries |
1144 | * which were found. |
1145 | */ |
1146 | unsigned find_get_entries(struct address_space *mapping, |
1147 | pgoff_t start, unsigned int nr_entries, |
1148 | struct page **entries, pgoff_t *indices) |
1149 | { |
1150 | void **slot; |
1151 | unsigned int ret = 0; |
1152 | struct radix_tree_iter iter; |
1153 | |
1154 | if (!nr_entries) |
1155 | return 0; |
1156 | |
1157 | rcu_read_lock(); |
1158 | restart: |
1159 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
1160 | struct page *page; |
1161 | repeat: |
1162 | page = radix_tree_deref_slot(slot); |
1163 | if (unlikely(!page)) |
1164 | continue; |
1165 | if (radix_tree_exception(page)) { |
1166 | if (radix_tree_deref_retry(page)) |
1167 | goto restart; |
1168 | /* |
1169 | * A shadow entry of a recently evicted page, |
1170 | * or a swap entry from shmem/tmpfs. Return |
1171 | * it without attempting to raise page count. |
1172 | */ |
1173 | goto export; |
1174 | } |
1175 | if (!page_cache_get_speculative(page)) |
1176 | goto repeat; |
1177 | |
1178 | /* Has the page moved? */ |
1179 | if (unlikely(page != *slot)) { |
1180 | page_cache_release(page); |
1181 | goto repeat; |
1182 | } |
1183 | export: |
1184 | indices[ret] = iter.index; |
1185 | entries[ret] = page; |
1186 | if (++ret == nr_entries) |
1187 | break; |
1188 | } |
1189 | rcu_read_unlock(); |
1190 | return ret; |
1191 | } |
1192 | |
1193 | /** |
1194 | * find_get_pages - gang pagecache lookup |
1195 | * @mapping: The address_space to search |
1196 | * @start: The starting page index |
1197 | * @nr_pages: The maximum number of pages |
1198 | * @pages: Where the resulting pages are placed |
1199 | * |
1200 | * find_get_pages() will search for and return a group of up to |
1201 | * @nr_pages pages in the mapping. The pages are placed at @pages. |
1202 | * find_get_pages() takes a reference against the returned pages. |
1203 | * |
1204 | * The search returns a group of mapping-contiguous pages with ascending |
1205 | * indexes. There may be holes in the indices due to not-present pages. |
1206 | * |
1207 | * find_get_pages() returns the number of pages which were found. |
1208 | */ |
1209 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, |
1210 | unsigned int nr_pages, struct page **pages) |
1211 | { |
1212 | struct radix_tree_iter iter; |
1213 | void **slot; |
1214 | unsigned ret = 0; |
1215 | |
1216 | if (unlikely(!nr_pages)) |
1217 | return 0; |
1218 | |
1219 | rcu_read_lock(); |
1220 | restart: |
1221 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
1222 | struct page *page; |
1223 | repeat: |
1224 | page = radix_tree_deref_slot(slot); |
1225 | if (unlikely(!page)) |
1226 | continue; |
1227 | |
1228 | if (radix_tree_exception(page)) { |
1229 | if (radix_tree_deref_retry(page)) { |
1230 | /* |
1231 | * Transient condition which can only trigger |
1232 | * when entry at index 0 moves out of or back |
1233 | * to root: none yet gotten, safe to restart. |
1234 | */ |
1235 | WARN_ON(iter.index); |
1236 | goto restart; |
1237 | } |
1238 | /* |
1239 | * A shadow entry of a recently evicted page, |
1240 | * or a swap entry from shmem/tmpfs. Skip |
1241 | * over it. |
1242 | */ |
1243 | continue; |
1244 | } |
1245 | |
1246 | if (!page_cache_get_speculative(page)) |
1247 | goto repeat; |
1248 | |
1249 | /* Has the page moved? */ |
1250 | if (unlikely(page != *slot)) { |
1251 | page_cache_release(page); |
1252 | goto repeat; |
1253 | } |
1254 | |
1255 | pages[ret] = page; |
1256 | if (++ret == nr_pages) |
1257 | break; |
1258 | } |
1259 | |
1260 | rcu_read_unlock(); |
1261 | return ret; |
1262 | } |
1263 | |
1264 | /** |
1265 | * find_get_pages_contig - gang contiguous pagecache lookup |
1266 | * @mapping: The address_space to search |
1267 | * @index: The starting page index |
1268 | * @nr_pages: The maximum number of pages |
1269 | * @pages: Where the resulting pages are placed |
1270 | * |
1271 | * find_get_pages_contig() works exactly like find_get_pages(), except |
1272 | * that the returned number of pages are guaranteed to be contiguous. |
1273 | * |
1274 | * find_get_pages_contig() returns the number of pages which were found. |
1275 | */ |
1276 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, |
1277 | unsigned int nr_pages, struct page **pages) |
1278 | { |
1279 | struct radix_tree_iter iter; |
1280 | void **slot; |
1281 | unsigned int ret = 0; |
1282 | |
1283 | if (unlikely(!nr_pages)) |
1284 | return 0; |
1285 | |
1286 | rcu_read_lock(); |
1287 | restart: |
1288 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
1289 | struct page *page; |
1290 | repeat: |
1291 | page = radix_tree_deref_slot(slot); |
1292 | /* The hole, there no reason to continue */ |
1293 | if (unlikely(!page)) |
1294 | break; |
1295 | |
1296 | if (radix_tree_exception(page)) { |
1297 | if (radix_tree_deref_retry(page)) { |
1298 | /* |
1299 | * Transient condition which can only trigger |
1300 | * when entry at index 0 moves out of or back |
1301 | * to root: none yet gotten, safe to restart. |
1302 | */ |
1303 | goto restart; |
1304 | } |
1305 | /* |
1306 | * A shadow entry of a recently evicted page, |
1307 | * or a swap entry from shmem/tmpfs. Stop |
1308 | * looking for contiguous pages. |
1309 | */ |
1310 | break; |
1311 | } |
1312 | |
1313 | if (!page_cache_get_speculative(page)) |
1314 | goto repeat; |
1315 | |
1316 | /* Has the page moved? */ |
1317 | if (unlikely(page != *slot)) { |
1318 | page_cache_release(page); |
1319 | goto repeat; |
1320 | } |
1321 | |
1322 | /* |
1323 | * must check mapping and index after taking the ref. |
1324 | * otherwise we can get both false positives and false |
1325 | * negatives, which is just confusing to the caller. |
1326 | */ |
1327 | if (page->mapping == NULL || page->index != iter.index) { |
1328 | page_cache_release(page); |
1329 | break; |
1330 | } |
1331 | |
1332 | pages[ret] = page; |
1333 | if (++ret == nr_pages) |
1334 | break; |
1335 | } |
1336 | rcu_read_unlock(); |
1337 | return ret; |
1338 | } |
1339 | EXPORT_SYMBOL(find_get_pages_contig); |
1340 | |
1341 | /** |
1342 | * find_get_pages_tag - find and return pages that match @tag |
1343 | * @mapping: the address_space to search |
1344 | * @index: the starting page index |
1345 | * @tag: the tag index |
1346 | * @nr_pages: the maximum number of pages |
1347 | * @pages: where the resulting pages are placed |
1348 | * |
1349 | * Like find_get_pages, except we only return pages which are tagged with |
1350 | * @tag. We update @index to index the next page for the traversal. |
1351 | */ |
1352 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, |
1353 | int tag, unsigned int nr_pages, struct page **pages) |
1354 | { |
1355 | struct radix_tree_iter iter; |
1356 | void **slot; |
1357 | unsigned ret = 0; |
1358 | |
1359 | if (unlikely(!nr_pages)) |
1360 | return 0; |
1361 | |
1362 | rcu_read_lock(); |
1363 | restart: |
1364 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1365 | &iter, *index, tag) { |
1366 | struct page *page; |
1367 | repeat: |
1368 | page = radix_tree_deref_slot(slot); |
1369 | if (unlikely(!page)) |
1370 | continue; |
1371 | |
1372 | if (radix_tree_exception(page)) { |
1373 | if (radix_tree_deref_retry(page)) { |
1374 | /* |
1375 | * Transient condition which can only trigger |
1376 | * when entry at index 0 moves out of or back |
1377 | * to root: none yet gotten, safe to restart. |
1378 | */ |
1379 | goto restart; |
1380 | } |
1381 | /* |
1382 | * A shadow entry of a recently evicted page. |
1383 | * |
1384 | * Those entries should never be tagged, but |
1385 | * this tree walk is lockless and the tags are |
1386 | * looked up in bulk, one radix tree node at a |
1387 | * time, so there is a sizable window for page |
1388 | * reclaim to evict a page we saw tagged. |
1389 | * |
1390 | * Skip over it. |
1391 | */ |
1392 | continue; |
1393 | } |
1394 | |
1395 | if (!page_cache_get_speculative(page)) |
1396 | goto repeat; |
1397 | |
1398 | /* Has the page moved? */ |
1399 | if (unlikely(page != *slot)) { |
1400 | page_cache_release(page); |
1401 | goto repeat; |
1402 | } |
1403 | |
1404 | pages[ret] = page; |
1405 | if (++ret == nr_pages) |
1406 | break; |
1407 | } |
1408 | |
1409 | rcu_read_unlock(); |
1410 | |
1411 | if (ret) |
1412 | *index = pages[ret - 1]->index + 1; |
1413 | |
1414 | return ret; |
1415 | } |
1416 | EXPORT_SYMBOL(find_get_pages_tag); |
1417 | |
1418 | /* |
1419 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail |
1420 | * a _large_ part of the i/o request. Imagine the worst scenario: |
1421 | * |
1422 | * ---R__________________________________________B__________ |
1423 | * ^ reading here ^ bad block(assume 4k) |
1424 | * |
1425 | * read(R) => miss => readahead(R...B) => media error => frustrating retries |
1426 | * => failing the whole request => read(R) => read(R+1) => |
1427 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => |
1428 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => |
1429 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... |
1430 | * |
1431 | * It is going insane. Fix it by quickly scaling down the readahead size. |
1432 | */ |
1433 | static void shrink_readahead_size_eio(struct file *filp, |
1434 | struct file_ra_state *ra) |
1435 | { |
1436 | ra->ra_pages /= 4; |
1437 | } |
1438 | |
1439 | /** |
1440 | * do_generic_file_read - generic file read routine |
1441 | * @filp: the file to read |
1442 | * @ppos: current file position |
1443 | * @iter: data destination |
1444 | * @written: already copied |
1445 | * |
1446 | * This is a generic file read routine, and uses the |
1447 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1448 | * |
1449 | * This is really ugly. But the goto's actually try to clarify some |
1450 | * of the logic when it comes to error handling etc. |
1451 | */ |
1452 | static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos, |
1453 | struct iov_iter *iter, ssize_t written) |
1454 | { |
1455 | struct address_space *mapping = filp->f_mapping; |
1456 | struct inode *inode = mapping->host; |
1457 | struct file_ra_state *ra = &filp->f_ra; |
1458 | pgoff_t index; |
1459 | pgoff_t last_index; |
1460 | pgoff_t prev_index; |
1461 | unsigned long offset; /* offset into pagecache page */ |
1462 | unsigned int prev_offset; |
1463 | int error = 0; |
1464 | |
1465 | index = *ppos >> PAGE_CACHE_SHIFT; |
1466 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1467 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); |
1468 | last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1469 | offset = *ppos & ~PAGE_CACHE_MASK; |
1470 | |
1471 | for (;;) { |
1472 | struct page *page; |
1473 | pgoff_t end_index; |
1474 | loff_t isize; |
1475 | unsigned long nr, ret; |
1476 | |
1477 | cond_resched(); |
1478 | find_page: |
1479 | page = find_get_page(mapping, index); |
1480 | if (!page) { |
1481 | page_cache_sync_readahead(mapping, |
1482 | ra, filp, |
1483 | index, last_index - index); |
1484 | page = find_get_page(mapping, index); |
1485 | if (unlikely(page == NULL)) |
1486 | goto no_cached_page; |
1487 | } |
1488 | if (PageReadahead(page)) { |
1489 | page_cache_async_readahead(mapping, |
1490 | ra, filp, page, |
1491 | index, last_index - index); |
1492 | } |
1493 | if (!PageUptodate(page)) { |
1494 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || |
1495 | !mapping->a_ops->is_partially_uptodate) |
1496 | goto page_not_up_to_date; |
1497 | if (!trylock_page(page)) |
1498 | goto page_not_up_to_date; |
1499 | /* Did it get truncated before we got the lock? */ |
1500 | if (!page->mapping) |
1501 | goto page_not_up_to_date_locked; |
1502 | if (!mapping->a_ops->is_partially_uptodate(page, |
1503 | offset, iter->count)) |
1504 | goto page_not_up_to_date_locked; |
1505 | unlock_page(page); |
1506 | } |
1507 | page_ok: |
1508 | /* |
1509 | * i_size must be checked after we know the page is Uptodate. |
1510 | * |
1511 | * Checking i_size after the check allows us to calculate |
1512 | * the correct value for "nr", which means the zero-filled |
1513 | * part of the page is not copied back to userspace (unless |
1514 | * another truncate extends the file - this is desired though). |
1515 | */ |
1516 | |
1517 | isize = i_size_read(inode); |
1518 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; |
1519 | if (unlikely(!isize || index > end_index)) { |
1520 | page_cache_release(page); |
1521 | goto out; |
1522 | } |
1523 | |
1524 | /* nr is the maximum number of bytes to copy from this page */ |
1525 | nr = PAGE_CACHE_SIZE; |
1526 | if (index == end_index) { |
1527 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; |
1528 | if (nr <= offset) { |
1529 | page_cache_release(page); |
1530 | goto out; |
1531 | } |
1532 | } |
1533 | nr = nr - offset; |
1534 | |
1535 | /* If users can be writing to this page using arbitrary |
1536 | * virtual addresses, take care about potential aliasing |
1537 | * before reading the page on the kernel side. |
1538 | */ |
1539 | if (mapping_writably_mapped(mapping)) |
1540 | flush_dcache_page(page); |
1541 | |
1542 | /* |
1543 | * When a sequential read accesses a page several times, |
1544 | * only mark it as accessed the first time. |
1545 | */ |
1546 | if (prev_index != index || offset != prev_offset) |
1547 | mark_page_accessed(page); |
1548 | prev_index = index; |
1549 | |
1550 | /* |
1551 | * Ok, we have the page, and it's up-to-date, so |
1552 | * now we can copy it to user space... |
1553 | */ |
1554 | |
1555 | ret = copy_page_to_iter(page, offset, nr, iter); |
1556 | offset += ret; |
1557 | index += offset >> PAGE_CACHE_SHIFT; |
1558 | offset &= ~PAGE_CACHE_MASK; |
1559 | prev_offset = offset; |
1560 | |
1561 | page_cache_release(page); |
1562 | written += ret; |
1563 | if (!iov_iter_count(iter)) |
1564 | goto out; |
1565 | if (ret < nr) { |
1566 | error = -EFAULT; |
1567 | goto out; |
1568 | } |
1569 | continue; |
1570 | |
1571 | page_not_up_to_date: |
1572 | /* Get exclusive access to the page ... */ |
1573 | error = lock_page_killable(page); |
1574 | if (unlikely(error)) |
1575 | goto readpage_error; |
1576 | |
1577 | page_not_up_to_date_locked: |
1578 | /* Did it get truncated before we got the lock? */ |
1579 | if (!page->mapping) { |
1580 | unlock_page(page); |
1581 | page_cache_release(page); |
1582 | continue; |
1583 | } |
1584 | |
1585 | /* Did somebody else fill it already? */ |
1586 | if (PageUptodate(page)) { |
1587 | unlock_page(page); |
1588 | goto page_ok; |
1589 | } |
1590 | |
1591 | readpage: |
1592 | /* |
1593 | * A previous I/O error may have been due to temporary |
1594 | * failures, eg. multipath errors. |
1595 | * PG_error will be set again if readpage fails. |
1596 | */ |
1597 | ClearPageError(page); |
1598 | /* Start the actual read. The read will unlock the page. */ |
1599 | error = mapping->a_ops->readpage(filp, page); |
1600 | |
1601 | if (unlikely(error)) { |
1602 | if (error == AOP_TRUNCATED_PAGE) { |
1603 | page_cache_release(page); |
1604 | error = 0; |
1605 | goto find_page; |
1606 | } |
1607 | goto readpage_error; |
1608 | } |
1609 | |
1610 | if (!PageUptodate(page)) { |
1611 | error = lock_page_killable(page); |
1612 | if (unlikely(error)) |
1613 | goto readpage_error; |
1614 | if (!PageUptodate(page)) { |
1615 | if (page->mapping == NULL) { |
1616 | /* |
1617 | * invalidate_mapping_pages got it |
1618 | */ |
1619 | unlock_page(page); |
1620 | page_cache_release(page); |
1621 | goto find_page; |
1622 | } |
1623 | unlock_page(page); |
1624 | shrink_readahead_size_eio(filp, ra); |
1625 | error = -EIO; |
1626 | goto readpage_error; |
1627 | } |
1628 | unlock_page(page); |
1629 | } |
1630 | |
1631 | goto page_ok; |
1632 | |
1633 | readpage_error: |
1634 | /* UHHUH! A synchronous read error occurred. Report it */ |
1635 | page_cache_release(page); |
1636 | goto out; |
1637 | |
1638 | no_cached_page: |
1639 | /* |
1640 | * Ok, it wasn't cached, so we need to create a new |
1641 | * page.. |
1642 | */ |
1643 | page = page_cache_alloc_cold(mapping); |
1644 | if (!page) { |
1645 | error = -ENOMEM; |
1646 | goto out; |
1647 | } |
1648 | error = add_to_page_cache_lru(page, mapping, |
1649 | index, GFP_KERNEL); |
1650 | if (error) { |
1651 | page_cache_release(page); |
1652 | if (error == -EEXIST) { |
1653 | error = 0; |
1654 | goto find_page; |
1655 | } |
1656 | goto out; |
1657 | } |
1658 | goto readpage; |
1659 | } |
1660 | |
1661 | out: |
1662 | ra->prev_pos = prev_index; |
1663 | ra->prev_pos <<= PAGE_CACHE_SHIFT; |
1664 | ra->prev_pos |= prev_offset; |
1665 | |
1666 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
1667 | file_accessed(filp); |
1668 | return written ? written : error; |
1669 | } |
1670 | |
1671 | /** |
1672 | * generic_file_read_iter - generic filesystem read routine |
1673 | * @iocb: kernel I/O control block |
1674 | * @iter: destination for the data read |
1675 | * |
1676 | * This is the "read_iter()" routine for all filesystems |
1677 | * that can use the page cache directly. |
1678 | */ |
1679 | ssize_t |
1680 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1681 | { |
1682 | struct file *file = iocb->ki_filp; |
1683 | ssize_t retval = 0; |
1684 | loff_t *ppos = &iocb->ki_pos; |
1685 | loff_t pos = *ppos; |
1686 | |
1687 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ |
1688 | if (file->f_flags & O_DIRECT) { |
1689 | struct address_space *mapping = file->f_mapping; |
1690 | struct inode *inode = mapping->host; |
1691 | size_t count = iov_iter_count(iter); |
1692 | loff_t size; |
1693 | |
1694 | if (!count) |
1695 | goto out; /* skip atime */ |
1696 | size = i_size_read(inode); |
1697 | retval = filemap_write_and_wait_range(mapping, pos, |
1698 | pos + count - 1); |
1699 | if (!retval) { |
1700 | struct iov_iter data = *iter; |
1701 | retval = mapping->a_ops->direct_IO(READ, iocb, &data, pos); |
1702 | } |
1703 | |
1704 | if (retval > 0) { |
1705 | *ppos = pos + retval; |
1706 | iov_iter_advance(iter, retval); |
1707 | } |
1708 | |
1709 | /* |
1710 | * Btrfs can have a short DIO read if we encounter |
1711 | * compressed extents, so if there was an error, or if |
1712 | * we've already read everything we wanted to, or if |
1713 | * there was a short read because we hit EOF, go ahead |
1714 | * and return. Otherwise fallthrough to buffered io for |
1715 | * the rest of the read. |
1716 | */ |
1717 | if (retval < 0 || !iov_iter_count(iter) || *ppos >= size) { |
1718 | file_accessed(file); |
1719 | goto out; |
1720 | } |
1721 | } |
1722 | |
1723 | retval = do_generic_file_read(file, ppos, iter, retval); |
1724 | out: |
1725 | return retval; |
1726 | } |
1727 | EXPORT_SYMBOL(generic_file_read_iter); |
1728 | |
1729 | #ifdef CONFIG_MMU |
1730 | /** |
1731 | * page_cache_read - adds requested page to the page cache if not already there |
1732 | * @file: file to read |
1733 | * @offset: page index |
1734 | * |
1735 | * This adds the requested page to the page cache if it isn't already there, |
1736 | * and schedules an I/O to read in its contents from disk. |
1737 | */ |
1738 | static int page_cache_read(struct file *file, pgoff_t offset) |
1739 | { |
1740 | struct address_space *mapping = file->f_mapping; |
1741 | struct page *page; |
1742 | int ret; |
1743 | |
1744 | do { |
1745 | page = page_cache_alloc_cold(mapping); |
1746 | if (!page) |
1747 | return -ENOMEM; |
1748 | |
1749 | ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); |
1750 | if (ret == 0) |
1751 | ret = mapping->a_ops->readpage(file, page); |
1752 | else if (ret == -EEXIST) |
1753 | ret = 0; /* losing race to add is OK */ |
1754 | |
1755 | page_cache_release(page); |
1756 | |
1757 | } while (ret == AOP_TRUNCATED_PAGE); |
1758 | |
1759 | return ret; |
1760 | } |
1761 | |
1762 | #define MMAP_LOTSAMISS (100) |
1763 | |
1764 | /* |
1765 | * Synchronous readahead happens when we don't even find |
1766 | * a page in the page cache at all. |
1767 | */ |
1768 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, |
1769 | struct file_ra_state *ra, |
1770 | struct file *file, |
1771 | pgoff_t offset) |
1772 | { |
1773 | unsigned long ra_pages; |
1774 | struct address_space *mapping = file->f_mapping; |
1775 | |
1776 | /* If we don't want any read-ahead, don't bother */ |
1777 | if (vma->vm_flags & VM_RAND_READ) |
1778 | return; |
1779 | if (!ra->ra_pages) |
1780 | return; |
1781 | |
1782 | if (vma->vm_flags & VM_SEQ_READ) { |
1783 | page_cache_sync_readahead(mapping, ra, file, offset, |
1784 | ra->ra_pages); |
1785 | return; |
1786 | } |
1787 | |
1788 | /* Avoid banging the cache line if not needed */ |
1789 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) |
1790 | ra->mmap_miss++; |
1791 | |
1792 | /* |
1793 | * Do we miss much more than hit in this file? If so, |
1794 | * stop bothering with read-ahead. It will only hurt. |
1795 | */ |
1796 | if (ra->mmap_miss > MMAP_LOTSAMISS) |
1797 | return; |
1798 | |
1799 | /* |
1800 | * mmap read-around |
1801 | */ |
1802 | ra_pages = max_sane_readahead(ra->ra_pages); |
1803 | ra->start = max_t(long, 0, offset - ra_pages / 2); |
1804 | ra->size = ra_pages; |
1805 | ra->async_size = ra_pages / 4; |
1806 | ra_submit(ra, mapping, file); |
1807 | } |
1808 | |
1809 | /* |
1810 | * Asynchronous readahead happens when we find the page and PG_readahead, |
1811 | * so we want to possibly extend the readahead further.. |
1812 | */ |
1813 | static void do_async_mmap_readahead(struct vm_area_struct *vma, |
1814 | struct file_ra_state *ra, |
1815 | struct file *file, |
1816 | struct page *page, |
1817 | pgoff_t offset) |
1818 | { |
1819 | struct address_space *mapping = file->f_mapping; |
1820 | |
1821 | /* If we don't want any read-ahead, don't bother */ |
1822 | if (vma->vm_flags & VM_RAND_READ) |
1823 | return; |
1824 | if (ra->mmap_miss > 0) |
1825 | ra->mmap_miss--; |
1826 | if (PageReadahead(page)) |
1827 | page_cache_async_readahead(mapping, ra, file, |
1828 | page, offset, ra->ra_pages); |
1829 | } |
1830 | |
1831 | /** |
1832 | * filemap_fault - read in file data for page fault handling |
1833 | * @vma: vma in which the fault was taken |
1834 | * @vmf: struct vm_fault containing details of the fault |
1835 | * |
1836 | * filemap_fault() is invoked via the vma operations vector for a |
1837 | * mapped memory region to read in file data during a page fault. |
1838 | * |
1839 | * The goto's are kind of ugly, but this streamlines the normal case of having |
1840 | * it in the page cache, and handles the special cases reasonably without |
1841 | * having a lot of duplicated code. |
1842 | */ |
1843 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1844 | { |
1845 | int error; |
1846 | struct file *file = vma->vm_file; |
1847 | struct address_space *mapping = file->f_mapping; |
1848 | struct file_ra_state *ra = &file->f_ra; |
1849 | struct inode *inode = mapping->host; |
1850 | pgoff_t offset = vmf->pgoff; |
1851 | struct page *page; |
1852 | loff_t size; |
1853 | int ret = 0; |
1854 | |
1855 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
1856 | if (offset >= size >> PAGE_CACHE_SHIFT) |
1857 | return VM_FAULT_SIGBUS; |
1858 | |
1859 | /* |
1860 | * Do we have something in the page cache already? |
1861 | */ |
1862 | page = find_get_page(mapping, offset); |
1863 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1864 | /* |
1865 | * We found the page, so try async readahead before |
1866 | * waiting for the lock. |
1867 | */ |
1868 | do_async_mmap_readahead(vma, ra, file, page, offset); |
1869 | } else if (!page) { |
1870 | /* No page in the page cache at all */ |
1871 | do_sync_mmap_readahead(vma, ra, file, offset); |
1872 | count_vm_event(PGMAJFAULT); |
1873 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
1874 | ret = VM_FAULT_MAJOR; |
1875 | retry_find: |
1876 | page = find_get_page(mapping, offset); |
1877 | if (!page) |
1878 | goto no_cached_page; |
1879 | } |
1880 | |
1881 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
1882 | page_cache_release(page); |
1883 | return ret | VM_FAULT_RETRY; |
1884 | } |
1885 | |
1886 | /* Did it get truncated? */ |
1887 | if (unlikely(page->mapping != mapping)) { |
1888 | unlock_page(page); |
1889 | put_page(page); |
1890 | goto retry_find; |
1891 | } |
1892 | VM_BUG_ON_PAGE(page->index != offset, page); |
1893 | |
1894 | /* |
1895 | * We have a locked page in the page cache, now we need to check |
1896 | * that it's up-to-date. If not, it is going to be due to an error. |
1897 | */ |
1898 | if (unlikely(!PageUptodate(page))) |
1899 | goto page_not_uptodate; |
1900 | |
1901 | /* |
1902 | * Found the page and have a reference on it. |
1903 | * We must recheck i_size under page lock. |
1904 | */ |
1905 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
1906 | if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) { |
1907 | unlock_page(page); |
1908 | page_cache_release(page); |
1909 | return VM_FAULT_SIGBUS; |
1910 | } |
1911 | |
1912 | vmf->page = page; |
1913 | return ret | VM_FAULT_LOCKED; |
1914 | |
1915 | no_cached_page: |
1916 | /* |
1917 | * We're only likely to ever get here if MADV_RANDOM is in |
1918 | * effect. |
1919 | */ |
1920 | error = page_cache_read(file, offset); |
1921 | |
1922 | /* |
1923 | * The page we want has now been added to the page cache. |
1924 | * In the unlikely event that someone removed it in the |
1925 | * meantime, we'll just come back here and read it again. |
1926 | */ |
1927 | if (error >= 0) |
1928 | goto retry_find; |
1929 | |
1930 | /* |
1931 | * An error return from page_cache_read can result if the |
1932 | * system is low on memory, or a problem occurs while trying |
1933 | * to schedule I/O. |
1934 | */ |
1935 | if (error == -ENOMEM) |
1936 | return VM_FAULT_OOM; |
1937 | return VM_FAULT_SIGBUS; |
1938 | |
1939 | page_not_uptodate: |
1940 | /* |
1941 | * Umm, take care of errors if the page isn't up-to-date. |
1942 | * Try to re-read it _once_. We do this synchronously, |
1943 | * because there really aren't any performance issues here |
1944 | * and we need to check for errors. |
1945 | */ |
1946 | ClearPageError(page); |
1947 | error = mapping->a_ops->readpage(file, page); |
1948 | if (!error) { |
1949 | wait_on_page_locked(page); |
1950 | if (!PageUptodate(page)) |
1951 | error = -EIO; |
1952 | } |
1953 | page_cache_release(page); |
1954 | |
1955 | if (!error || error == AOP_TRUNCATED_PAGE) |
1956 | goto retry_find; |
1957 | |
1958 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
1959 | shrink_readahead_size_eio(file, ra); |
1960 | return VM_FAULT_SIGBUS; |
1961 | } |
1962 | EXPORT_SYMBOL(filemap_fault); |
1963 | |
1964 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
1965 | { |
1966 | struct radix_tree_iter iter; |
1967 | void **slot; |
1968 | struct file *file = vma->vm_file; |
1969 | struct address_space *mapping = file->f_mapping; |
1970 | loff_t size; |
1971 | struct page *page; |
1972 | unsigned long address = (unsigned long) vmf->virtual_address; |
1973 | unsigned long addr; |
1974 | pte_t *pte; |
1975 | |
1976 | rcu_read_lock(); |
1977 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) { |
1978 | if (iter.index > vmf->max_pgoff) |
1979 | break; |
1980 | repeat: |
1981 | page = radix_tree_deref_slot(slot); |
1982 | if (unlikely(!page)) |
1983 | goto next; |
1984 | if (radix_tree_exception(page)) { |
1985 | if (radix_tree_deref_retry(page)) |
1986 | break; |
1987 | else |
1988 | goto next; |
1989 | } |
1990 | |
1991 | if (!page_cache_get_speculative(page)) |
1992 | goto repeat; |
1993 | |
1994 | /* Has the page moved? */ |
1995 | if (unlikely(page != *slot)) { |
1996 | page_cache_release(page); |
1997 | goto repeat; |
1998 | } |
1999 | |
2000 | if (!PageUptodate(page) || |
2001 | PageReadahead(page) || |
2002 | PageHWPoison(page)) |
2003 | goto skip; |
2004 | if (!trylock_page(page)) |
2005 | goto skip; |
2006 | |
2007 | if (page->mapping != mapping || !PageUptodate(page)) |
2008 | goto unlock; |
2009 | |
2010 | size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE); |
2011 | if (page->index >= size >> PAGE_CACHE_SHIFT) |
2012 | goto unlock; |
2013 | |
2014 | pte = vmf->pte + page->index - vmf->pgoff; |
2015 | if (!pte_none(*pte)) |
2016 | goto unlock; |
2017 | |
2018 | if (file->f_ra.mmap_miss > 0) |
2019 | file->f_ra.mmap_miss--; |
2020 | addr = address + (page->index - vmf->pgoff) * PAGE_SIZE; |
2021 | do_set_pte(vma, addr, page, pte, false, false); |
2022 | unlock_page(page); |
2023 | goto next; |
2024 | unlock: |
2025 | unlock_page(page); |
2026 | skip: |
2027 | page_cache_release(page); |
2028 | next: |
2029 | if (iter.index == vmf->max_pgoff) |
2030 | break; |
2031 | } |
2032 | rcu_read_unlock(); |
2033 | } |
2034 | EXPORT_SYMBOL(filemap_map_pages); |
2035 | |
2036 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2037 | { |
2038 | struct page *page = vmf->page; |
2039 | struct inode *inode = file_inode(vma->vm_file); |
2040 | int ret = VM_FAULT_LOCKED; |
2041 | |
2042 | sb_start_pagefault(inode->i_sb); |
2043 | file_update_time(vma->vm_file); |
2044 | lock_page(page); |
2045 | if (page->mapping != inode->i_mapping) { |
2046 | unlock_page(page); |
2047 | ret = VM_FAULT_NOPAGE; |
2048 | goto out; |
2049 | } |
2050 | /* |
2051 | * We mark the page dirty already here so that when freeze is in |
2052 | * progress, we are guaranteed that writeback during freezing will |
2053 | * see the dirty page and writeprotect it again. |
2054 | */ |
2055 | set_page_dirty(page); |
2056 | wait_for_stable_page(page); |
2057 | out: |
2058 | sb_end_pagefault(inode->i_sb); |
2059 | return ret; |
2060 | } |
2061 | EXPORT_SYMBOL(filemap_page_mkwrite); |
2062 | |
2063 | const struct vm_operations_struct generic_file_vm_ops = { |
2064 | .fault = filemap_fault, |
2065 | .map_pages = filemap_map_pages, |
2066 | .page_mkwrite = filemap_page_mkwrite, |
2067 | .remap_pages = generic_file_remap_pages, |
2068 | }; |
2069 | |
2070 | /* This is used for a general mmap of a disk file */ |
2071 | |
2072 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) |
2073 | { |
2074 | struct address_space *mapping = file->f_mapping; |
2075 | |
2076 | if (!mapping->a_ops->readpage) |
2077 | return -ENOEXEC; |
2078 | file_accessed(file); |
2079 | vma->vm_ops = &generic_file_vm_ops; |
2080 | return 0; |
2081 | } |
2082 | |
2083 | /* |
2084 | * This is for filesystems which do not implement ->writepage. |
2085 | */ |
2086 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
2087 | { |
2088 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) |
2089 | return -EINVAL; |
2090 | return generic_file_mmap(file, vma); |
2091 | } |
2092 | #else |
2093 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) |
2094 | { |
2095 | return -ENOSYS; |
2096 | } |
2097 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) |
2098 | { |
2099 | return -ENOSYS; |
2100 | } |
2101 | #endif /* CONFIG_MMU */ |
2102 | |
2103 | EXPORT_SYMBOL(generic_file_mmap); |
2104 | EXPORT_SYMBOL(generic_file_readonly_mmap); |
2105 | |
2106 | static struct page *wait_on_page_read(struct page *page) |
2107 | { |
2108 | if (!IS_ERR(page)) { |
2109 | wait_on_page_locked(page); |
2110 | if (!PageUptodate(page)) { |
2111 | page_cache_release(page); |
2112 | page = ERR_PTR(-EIO); |
2113 | } |
2114 | } |
2115 | return page; |
2116 | } |
2117 | |
2118 | static struct page *__read_cache_page(struct address_space *mapping, |
2119 | pgoff_t index, |
2120 | int (*filler)(void *, struct page *), |
2121 | void *data, |
2122 | gfp_t gfp) |
2123 | { |
2124 | struct page *page; |
2125 | int err; |
2126 | repeat: |
2127 | page = find_get_page(mapping, index); |
2128 | if (!page) { |
2129 | page = __page_cache_alloc(gfp | __GFP_COLD); |
2130 | if (!page) |
2131 | return ERR_PTR(-ENOMEM); |
2132 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
2133 | if (unlikely(err)) { |
2134 | page_cache_release(page); |
2135 | if (err == -EEXIST) |
2136 | goto repeat; |
2137 | /* Presumably ENOMEM for radix tree node */ |
2138 | return ERR_PTR(err); |
2139 | } |
2140 | err = filler(data, page); |
2141 | if (err < 0) { |
2142 | page_cache_release(page); |
2143 | page = ERR_PTR(err); |
2144 | } else { |
2145 | page = wait_on_page_read(page); |
2146 | } |
2147 | } |
2148 | return page; |
2149 | } |
2150 | |
2151 | static struct page *do_read_cache_page(struct address_space *mapping, |
2152 | pgoff_t index, |
2153 | int (*filler)(void *, struct page *), |
2154 | void *data, |
2155 | gfp_t gfp) |
2156 | |
2157 | { |
2158 | struct page *page; |
2159 | int err; |
2160 | |
2161 | retry: |
2162 | page = __read_cache_page(mapping, index, filler, data, gfp); |
2163 | if (IS_ERR(page)) |
2164 | return page; |
2165 | if (PageUptodate(page)) |
2166 | goto out; |
2167 | |
2168 | lock_page(page); |
2169 | if (!page->mapping) { |
2170 | unlock_page(page); |
2171 | page_cache_release(page); |
2172 | goto retry; |
2173 | } |
2174 | if (PageUptodate(page)) { |
2175 | unlock_page(page); |
2176 | goto out; |
2177 | } |
2178 | err = filler(data, page); |
2179 | if (err < 0) { |
2180 | page_cache_release(page); |
2181 | return ERR_PTR(err); |
2182 | } else { |
2183 | page = wait_on_page_read(page); |
2184 | if (IS_ERR(page)) |
2185 | return page; |
2186 | } |
2187 | out: |
2188 | mark_page_accessed(page); |
2189 | return page; |
2190 | } |
2191 | |
2192 | /** |
2193 | * read_cache_page - read into page cache, fill it if needed |
2194 | * @mapping: the page's address_space |
2195 | * @index: the page index |
2196 | * @filler: function to perform the read |
2197 | * @data: first arg to filler(data, page) function, often left as NULL |
2198 | * |
2199 | * Read into the page cache. If a page already exists, and PageUptodate() is |
2200 | * not set, try to fill the page and wait for it to become unlocked. |
2201 | * |
2202 | * If the page does not get brought uptodate, return -EIO. |
2203 | */ |
2204 | struct page *read_cache_page(struct address_space *mapping, |
2205 | pgoff_t index, |
2206 | int (*filler)(void *, struct page *), |
2207 | void *data) |
2208 | { |
2209 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); |
2210 | } |
2211 | EXPORT_SYMBOL(read_cache_page); |
2212 | |
2213 | /** |
2214 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. |
2215 | * @mapping: the page's address_space |
2216 | * @index: the page index |
2217 | * @gfp: the page allocator flags to use if allocating |
2218 | * |
2219 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with |
2220 | * any new page allocations done using the specified allocation flags. |
2221 | * |
2222 | * If the page does not get brought uptodate, return -EIO. |
2223 | */ |
2224 | struct page *read_cache_page_gfp(struct address_space *mapping, |
2225 | pgoff_t index, |
2226 | gfp_t gfp) |
2227 | { |
2228 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; |
2229 | |
2230 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
2231 | } |
2232 | EXPORT_SYMBOL(read_cache_page_gfp); |
2233 | |
2234 | /* |
2235 | * Performs necessary checks before doing a write |
2236 | * |
2237 | * Can adjust writing position or amount of bytes to write. |
2238 | * Returns appropriate error code that caller should return or |
2239 | * zero in case that write should be allowed. |
2240 | */ |
2241 | inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) |
2242 | { |
2243 | struct inode *inode = file->f_mapping->host; |
2244 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
2245 | |
2246 | if (unlikely(*pos < 0)) |
2247 | return -EINVAL; |
2248 | |
2249 | if (!isblk) { |
2250 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2251 | if (file->f_flags & O_APPEND) |
2252 | *pos = i_size_read(inode); |
2253 | |
2254 | if (limit != RLIM_INFINITY) { |
2255 | if (*pos >= limit) { |
2256 | send_sig(SIGXFSZ, current, 0); |
2257 | return -EFBIG; |
2258 | } |
2259 | if (*count > limit - (typeof(limit))*pos) { |
2260 | *count = limit - (typeof(limit))*pos; |
2261 | } |
2262 | } |
2263 | } |
2264 | |
2265 | /* |
2266 | * LFS rule |
2267 | */ |
2268 | if (unlikely(*pos + *count > MAX_NON_LFS && |
2269 | !(file->f_flags & O_LARGEFILE))) { |
2270 | if (*pos >= MAX_NON_LFS) { |
2271 | return -EFBIG; |
2272 | } |
2273 | if (*count > MAX_NON_LFS - (unsigned long)*pos) { |
2274 | *count = MAX_NON_LFS - (unsigned long)*pos; |
2275 | } |
2276 | } |
2277 | |
2278 | /* |
2279 | * Are we about to exceed the fs block limit ? |
2280 | * |
2281 | * If we have written data it becomes a short write. If we have |
2282 | * exceeded without writing data we send a signal and return EFBIG. |
2283 | * Linus frestrict idea will clean these up nicely.. |
2284 | */ |
2285 | if (likely(!isblk)) { |
2286 | if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { |
2287 | if (*count || *pos > inode->i_sb->s_maxbytes) { |
2288 | return -EFBIG; |
2289 | } |
2290 | /* zero-length writes at ->s_maxbytes are OK */ |
2291 | } |
2292 | |
2293 | if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) |
2294 | *count = inode->i_sb->s_maxbytes - *pos; |
2295 | } else { |
2296 | #ifdef CONFIG_BLOCK |
2297 | loff_t isize; |
2298 | if (bdev_read_only(I_BDEV(inode))) |
2299 | return -EPERM; |
2300 | isize = i_size_read(inode); |
2301 | if (*pos >= isize) { |
2302 | if (*count || *pos > isize) |
2303 | return -ENOSPC; |
2304 | } |
2305 | |
2306 | if (*pos + *count > isize) |
2307 | *count = isize - *pos; |
2308 | #else |
2309 | return -EPERM; |
2310 | #endif |
2311 | } |
2312 | return 0; |
2313 | } |
2314 | EXPORT_SYMBOL(generic_write_checks); |
2315 | |
2316 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2317 | loff_t pos, unsigned len, unsigned flags, |
2318 | struct page **pagep, void **fsdata) |
2319 | { |
2320 | const struct address_space_operations *aops = mapping->a_ops; |
2321 | |
2322 | return aops->write_begin(file, mapping, pos, len, flags, |
2323 | pagep, fsdata); |
2324 | } |
2325 | EXPORT_SYMBOL(pagecache_write_begin); |
2326 | |
2327 | int pagecache_write_end(struct file *file, struct address_space *mapping, |
2328 | loff_t pos, unsigned len, unsigned copied, |
2329 | struct page *page, void *fsdata) |
2330 | { |
2331 | const struct address_space_operations *aops = mapping->a_ops; |
2332 | |
2333 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
2334 | } |
2335 | EXPORT_SYMBOL(pagecache_write_end); |
2336 | |
2337 | ssize_t |
2338 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos) |
2339 | { |
2340 | struct file *file = iocb->ki_filp; |
2341 | struct address_space *mapping = file->f_mapping; |
2342 | struct inode *inode = mapping->host; |
2343 | ssize_t written; |
2344 | size_t write_len; |
2345 | pgoff_t end; |
2346 | struct iov_iter data; |
2347 | |
2348 | write_len = iov_iter_count(from); |
2349 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; |
2350 | |
2351 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
2352 | if (written) |
2353 | goto out; |
2354 | |
2355 | /* |
2356 | * After a write we want buffered reads to be sure to go to disk to get |
2357 | * the new data. We invalidate clean cached page from the region we're |
2358 | * about to write. We do this *before* the write so that we can return |
2359 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
2360 | */ |
2361 | if (mapping->nrpages) { |
2362 | written = invalidate_inode_pages2_range(mapping, |
2363 | pos >> PAGE_CACHE_SHIFT, end); |
2364 | /* |
2365 | * If a page can not be invalidated, return 0 to fall back |
2366 | * to buffered write. |
2367 | */ |
2368 | if (written) { |
2369 | if (written == -EBUSY) |
2370 | return 0; |
2371 | goto out; |
2372 | } |
2373 | } |
2374 | |
2375 | data = *from; |
2376 | written = mapping->a_ops->direct_IO(WRITE, iocb, &data, pos); |
2377 | |
2378 | /* |
2379 | * Finally, try again to invalidate clean pages which might have been |
2380 | * cached by non-direct readahead, or faulted in by get_user_pages() |
2381 | * if the source of the write was an mmap'ed region of the file |
2382 | * we're writing. Either one is a pretty crazy thing to do, |
2383 | * so we don't support it 100%. If this invalidation |
2384 | * fails, tough, the write still worked... |
2385 | */ |
2386 | if (mapping->nrpages) { |
2387 | invalidate_inode_pages2_range(mapping, |
2388 | pos >> PAGE_CACHE_SHIFT, end); |
2389 | } |
2390 | |
2391 | if (written > 0) { |
2392 | pos += written; |
2393 | iov_iter_advance(from, written); |
2394 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
2395 | i_size_write(inode, pos); |
2396 | mark_inode_dirty(inode); |
2397 | } |
2398 | iocb->ki_pos = pos; |
2399 | } |
2400 | out: |
2401 | return written; |
2402 | } |
2403 | EXPORT_SYMBOL(generic_file_direct_write); |
2404 | |
2405 | /* |
2406 | * Find or create a page at the given pagecache position. Return the locked |
2407 | * page. This function is specifically for buffered writes. |
2408 | */ |
2409 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2410 | pgoff_t index, unsigned flags) |
2411 | { |
2412 | struct page *page; |
2413 | int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT; |
2414 | |
2415 | if (flags & AOP_FLAG_NOFS) |
2416 | fgp_flags |= FGP_NOFS; |
2417 | |
2418 | page = pagecache_get_page(mapping, index, fgp_flags, |
2419 | mapping_gfp_mask(mapping), |
2420 | GFP_KERNEL); |
2421 | if (page) |
2422 | wait_for_stable_page(page); |
2423 | |
2424 | return page; |
2425 | } |
2426 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
2427 | |
2428 | ssize_t generic_perform_write(struct file *file, |
2429 | struct iov_iter *i, loff_t pos) |
2430 | { |
2431 | struct address_space *mapping = file->f_mapping; |
2432 | const struct address_space_operations *a_ops = mapping->a_ops; |
2433 | long status = 0; |
2434 | ssize_t written = 0; |
2435 | unsigned int flags = 0; |
2436 | |
2437 | /* |
2438 | * Copies from kernel address space cannot fail (NFSD is a big user). |
2439 | */ |
2440 | if (segment_eq(get_fs(), KERNEL_DS)) |
2441 | flags |= AOP_FLAG_UNINTERRUPTIBLE; |
2442 | |
2443 | do { |
2444 | struct page *page; |
2445 | unsigned long offset; /* Offset into pagecache page */ |
2446 | unsigned long bytes; /* Bytes to write to page */ |
2447 | size_t copied; /* Bytes copied from user */ |
2448 | void *fsdata; |
2449 | |
2450 | offset = (pos & (PAGE_CACHE_SIZE - 1)); |
2451 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2452 | iov_iter_count(i)); |
2453 | |
2454 | again: |
2455 | /* |
2456 | * Bring in the user page that we will copy from _first_. |
2457 | * Otherwise there's a nasty deadlock on copying from the |
2458 | * same page as we're writing to, without it being marked |
2459 | * up-to-date. |
2460 | * |
2461 | * Not only is this an optimisation, but it is also required |
2462 | * to check that the address is actually valid, when atomic |
2463 | * usercopies are used, below. |
2464 | */ |
2465 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { |
2466 | status = -EFAULT; |
2467 | break; |
2468 | } |
2469 | |
2470 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
2471 | &page, &fsdata); |
2472 | if (unlikely(status < 0)) |
2473 | break; |
2474 | |
2475 | if (mapping_writably_mapped(mapping)) |
2476 | flush_dcache_page(page); |
2477 | |
2478 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
2479 | flush_dcache_page(page); |
2480 | |
2481 | status = a_ops->write_end(file, mapping, pos, bytes, copied, |
2482 | page, fsdata); |
2483 | if (unlikely(status < 0)) |
2484 | break; |
2485 | copied = status; |
2486 | |
2487 | cond_resched(); |
2488 | |
2489 | iov_iter_advance(i, copied); |
2490 | if (unlikely(copied == 0)) { |
2491 | /* |
2492 | * If we were unable to copy any data at all, we must |
2493 | * fall back to a single segment length write. |
2494 | * |
2495 | * If we didn't fallback here, we could livelock |
2496 | * because not all segments in the iov can be copied at |
2497 | * once without a pagefault. |
2498 | */ |
2499 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2500 | iov_iter_single_seg_count(i)); |
2501 | goto again; |
2502 | } |
2503 | pos += copied; |
2504 | written += copied; |
2505 | |
2506 | balance_dirty_pages_ratelimited(mapping); |
2507 | if (fatal_signal_pending(current)) { |
2508 | status = -EINTR; |
2509 | break; |
2510 | } |
2511 | } while (iov_iter_count(i)); |
2512 | |
2513 | return written ? written : status; |
2514 | } |
2515 | EXPORT_SYMBOL(generic_perform_write); |
2516 | |
2517 | /** |
2518 | * __generic_file_write_iter - write data to a file |
2519 | * @iocb: IO state structure (file, offset, etc.) |
2520 | * @from: iov_iter with data to write |
2521 | * |
2522 | * This function does all the work needed for actually writing data to a |
2523 | * file. It does all basic checks, removes SUID from the file, updates |
2524 | * modification times and calls proper subroutines depending on whether we |
2525 | * do direct IO or a standard buffered write. |
2526 | * |
2527 | * It expects i_mutex to be grabbed unless we work on a block device or similar |
2528 | * object which does not need locking at all. |
2529 | * |
2530 | * This function does *not* take care of syncing data in case of O_SYNC write. |
2531 | * A caller has to handle it. This is mainly due to the fact that we want to |
2532 | * avoid syncing under i_mutex. |
2533 | */ |
2534 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
2535 | { |
2536 | struct file *file = iocb->ki_filp; |
2537 | struct address_space * mapping = file->f_mapping; |
2538 | struct inode *inode = mapping->host; |
2539 | loff_t pos = iocb->ki_pos; |
2540 | ssize_t written = 0; |
2541 | ssize_t err; |
2542 | ssize_t status; |
2543 | size_t count = iov_iter_count(from); |
2544 | |
2545 | /* We can write back this queue in page reclaim */ |
2546 | current->backing_dev_info = mapping->backing_dev_info; |
2547 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); |
2548 | if (err) |
2549 | goto out; |
2550 | |
2551 | if (count == 0) |
2552 | goto out; |
2553 | |
2554 | iov_iter_truncate(from, count); |
2555 | |
2556 | err = file_remove_suid(file); |
2557 | if (err) |
2558 | goto out; |
2559 | |
2560 | err = file_update_time(file); |
2561 | if (err) |
2562 | goto out; |
2563 | |
2564 | /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ |
2565 | if (unlikely(file->f_flags & O_DIRECT)) { |
2566 | loff_t endbyte; |
2567 | |
2568 | written = generic_file_direct_write(iocb, from, pos); |
2569 | if (written < 0 || written == count) |
2570 | goto out; |
2571 | |
2572 | /* |
2573 | * direct-io write to a hole: fall through to buffered I/O |
2574 | * for completing the rest of the request. |
2575 | */ |
2576 | pos += written; |
2577 | count -= written; |
2578 | |
2579 | status = generic_perform_write(file, from, pos); |
2580 | /* |
2581 | * If generic_perform_write() returned a synchronous error |
2582 | * then we want to return the number of bytes which were |
2583 | * direct-written, or the error code if that was zero. Note |
2584 | * that this differs from normal direct-io semantics, which |
2585 | * will return -EFOO even if some bytes were written. |
2586 | */ |
2587 | if (unlikely(status < 0) && !written) { |
2588 | err = status; |
2589 | goto out; |
2590 | } |
2591 | iocb->ki_pos = pos + status; |
2592 | /* |
2593 | * We need to ensure that the page cache pages are written to |
2594 | * disk and invalidated to preserve the expected O_DIRECT |
2595 | * semantics. |
2596 | */ |
2597 | endbyte = pos + status - 1; |
2598 | err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); |
2599 | if (err == 0) { |
2600 | written += status; |
2601 | invalidate_mapping_pages(mapping, |
2602 | pos >> PAGE_CACHE_SHIFT, |
2603 | endbyte >> PAGE_CACHE_SHIFT); |
2604 | } else { |
2605 | /* |
2606 | * We don't know how much we wrote, so just return |
2607 | * the number of bytes which were direct-written |
2608 | */ |
2609 | } |
2610 | } else { |
2611 | written = generic_perform_write(file, from, pos); |
2612 | if (likely(written >= 0)) |
2613 | iocb->ki_pos = pos + written; |
2614 | } |
2615 | out: |
2616 | current->backing_dev_info = NULL; |
2617 | return written ? written : err; |
2618 | } |
2619 | EXPORT_SYMBOL(__generic_file_write_iter); |
2620 | |
2621 | /** |
2622 | * generic_file_write_iter - write data to a file |
2623 | * @iocb: IO state structure |
2624 | * @from: iov_iter with data to write |
2625 | * |
2626 | * This is a wrapper around __generic_file_write_iter() to be used by most |
2627 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
2628 | * and acquires i_mutex as needed. |
2629 | */ |
2630 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
2631 | { |
2632 | struct file *file = iocb->ki_filp; |
2633 | struct inode *inode = file->f_mapping->host; |
2634 | ssize_t ret; |
2635 | |
2636 | mutex_lock(&inode->i_mutex); |
2637 | ret = __generic_file_write_iter(iocb, from); |
2638 | mutex_unlock(&inode->i_mutex); |
2639 | |
2640 | if (ret > 0) { |
2641 | ssize_t err; |
2642 | |
2643 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
2644 | if (err < 0) |
2645 | ret = err; |
2646 | } |
2647 | return ret; |
2648 | } |
2649 | EXPORT_SYMBOL(generic_file_write_iter); |
2650 | |
2651 | /** |
2652 | * try_to_release_page() - release old fs-specific metadata on a page |
2653 | * |
2654 | * @page: the page which the kernel is trying to free |
2655 | * @gfp_mask: memory allocation flags (and I/O mode) |
2656 | * |
2657 | * The address_space is to try to release any data against the page |
2658 | * (presumably at page->private). If the release was successful, return `1'. |
2659 | * Otherwise return zero. |
2660 | * |
2661 | * This may also be called if PG_fscache is set on a page, indicating that the |
2662 | * page is known to the local caching routines. |
2663 | * |
2664 | * The @gfp_mask argument specifies whether I/O may be performed to release |
2665 | * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). |
2666 | * |
2667 | */ |
2668 | int try_to_release_page(struct page *page, gfp_t gfp_mask) |
2669 | { |
2670 | struct address_space * const mapping = page->mapping; |
2671 | |
2672 | BUG_ON(!PageLocked(page)); |
2673 | if (PageWriteback(page)) |
2674 | return 0; |
2675 | |
2676 | if (mapping && mapping->a_ops->releasepage) |
2677 | return mapping->a_ops->releasepage(page, gfp_mask); |
2678 | return try_to_free_buffers(page); |
2679 | } |
2680 | |
2681 | EXPORT_SYMBOL(try_to_release_page); |
2682 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9