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