Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/filemap.c | |
3 | * | |
4 | * Copyright (C) 1994-1999 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * This file handles the generic file mmap semantics used by | |
9 | * most "normal" filesystems (but you don't /have/ to use this: | |
10 | * the NFS filesystem used to do this differently, for example) | |
11 | */ | |
b95f1b31 | 12 | #include <linux/export.h> |
1da177e4 | 13 | #include <linux/compiler.h> |
f9fe48be | 14 | #include <linux/dax.h> |
1da177e4 | 15 | #include <linux/fs.h> |
c22ce143 | 16 | #include <linux/uaccess.h> |
c59ede7b | 17 | #include <linux/capability.h> |
1da177e4 | 18 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 19 | #include <linux/gfp.h> |
1da177e4 LT |
20 | #include <linux/mm.h> |
21 | #include <linux/swap.h> | |
22 | #include <linux/mman.h> | |
23 | #include <linux/pagemap.h> | |
24 | #include <linux/file.h> | |
25 | #include <linux/uio.h> | |
26 | #include <linux/hash.h> | |
27 | #include <linux/writeback.h> | |
53253383 | 28 | #include <linux/backing-dev.h> |
1da177e4 LT |
29 | #include <linux/pagevec.h> |
30 | #include <linux/blkdev.h> | |
31 | #include <linux/security.h> | |
44110fe3 | 32 | #include <linux/cpuset.h> |
2f718ffc | 33 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
00501b53 | 34 | #include <linux/hugetlb.h> |
8a9f3ccd | 35 | #include <linux/memcontrol.h> |
c515e1fd | 36 | #include <linux/cleancache.h> |
f1820361 | 37 | #include <linux/rmap.h> |
0f8053a5 NP |
38 | #include "internal.h" |
39 | ||
fe0bfaaf RJ |
40 | #define CREATE_TRACE_POINTS |
41 | #include <trace/events/filemap.h> | |
42 | ||
1da177e4 | 43 | /* |
1da177e4 LT |
44 | * FIXME: remove all knowledge of the buffer layer from the core VM |
45 | */ | |
148f948b | 46 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 47 | |
1da177e4 LT |
48 | #include <asm/mman.h> |
49 | ||
50 | /* | |
51 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
52 | * though. | |
53 | * | |
54 | * Shared mappings now work. 15.8.1995 Bruno. | |
55 | * | |
56 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
57 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
58 | * | |
59 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
60 | */ | |
61 | ||
62 | /* | |
63 | * Lock ordering: | |
64 | * | |
c8c06efa | 65 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 66 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 HD |
67 | * ->swap_lock (exclusive_swap_page, others) |
68 | * ->mapping->tree_lock | |
1da177e4 | 69 | * |
1b1dcc1b | 70 | * ->i_mutex |
c8c06efa | 71 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
1da177e4 LT |
72 | * |
73 | * ->mmap_sem | |
c8c06efa | 74 | * ->i_mmap_rwsem |
b8072f09 | 75 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
76 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
77 | * | |
78 | * ->mmap_sem | |
79 | * ->lock_page (access_process_vm) | |
80 | * | |
ccad2365 | 81 | * ->i_mutex (generic_perform_write) |
82591e6e | 82 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
1da177e4 | 83 | * |
f758eeab | 84 | * bdi->wb.list_lock |
a66979ab | 85 | * sb_lock (fs/fs-writeback.c) |
1da177e4 LT |
86 | * ->mapping->tree_lock (__sync_single_inode) |
87 | * | |
c8c06efa | 88 | * ->i_mmap_rwsem |
1da177e4 LT |
89 | * ->anon_vma.lock (vma_adjust) |
90 | * | |
91 | * ->anon_vma.lock | |
b8072f09 | 92 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 93 | * |
b8072f09 | 94 | * ->page_table_lock or pte_lock |
5d337b91 | 95 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
96 | * ->private_lock (try_to_unmap_one) |
97 | * ->tree_lock (try_to_unmap_one) | |
98 | * ->zone.lru_lock (follow_page->mark_page_accessed) | |
053837fc | 99 | * ->zone.lru_lock (check_pte_range->isolate_lru_page) |
1da177e4 LT |
100 | * ->private_lock (page_remove_rmap->set_page_dirty) |
101 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
f758eeab | 102 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 103 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
c4843a75 | 104 | * ->memcg->move_lock (page_remove_rmap->mem_cgroup_begin_page_stat) |
f758eeab | 105 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 106 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
107 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
108 | * | |
c8c06efa | 109 | * ->i_mmap_rwsem |
9a3c531d | 110 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
111 | */ |
112 | ||
91b0abe3 JW |
113 | static void page_cache_tree_delete(struct address_space *mapping, |
114 | struct page *page, void *shadow) | |
115 | { | |
449dd698 JW |
116 | struct radix_tree_node *node; |
117 | unsigned long index; | |
118 | unsigned int offset; | |
119 | unsigned int tag; | |
120 | void **slot; | |
91b0abe3 | 121 | |
449dd698 JW |
122 | VM_BUG_ON(!PageLocked(page)); |
123 | ||
124 | __radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot); | |
125 | ||
126 | if (shadow) { | |
f9fe48be | 127 | mapping->nrexceptional++; |
91b0abe3 | 128 | /* |
f9fe48be | 129 | * Make sure the nrexceptional update is committed before |
91b0abe3 JW |
130 | * the nrpages update so that final truncate racing |
131 | * with reclaim does not see both counters 0 at the | |
132 | * same time and miss a shadow entry. | |
133 | */ | |
134 | smp_wmb(); | |
449dd698 | 135 | } |
91b0abe3 | 136 | mapping->nrpages--; |
449dd698 JW |
137 | |
138 | if (!node) { | |
139 | /* Clear direct pointer tags in root node */ | |
140 | mapping->page_tree.gfp_mask &= __GFP_BITS_MASK; | |
141 | radix_tree_replace_slot(slot, shadow); | |
142 | return; | |
143 | } | |
144 | ||
145 | /* Clear tree tags for the removed page */ | |
146 | index = page->index; | |
147 | offset = index & RADIX_TREE_MAP_MASK; | |
148 | for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) { | |
149 | if (test_bit(offset, node->tags[tag])) | |
150 | radix_tree_tag_clear(&mapping->page_tree, index, tag); | |
151 | } | |
152 | ||
153 | /* Delete page, swap shadow entry */ | |
154 | radix_tree_replace_slot(slot, shadow); | |
155 | workingset_node_pages_dec(node); | |
156 | if (shadow) | |
157 | workingset_node_shadows_inc(node); | |
158 | else | |
159 | if (__radix_tree_delete_node(&mapping->page_tree, node)) | |
160 | return; | |
161 | ||
162 | /* | |
163 | * Track node that only contains shadow entries. | |
164 | * | |
165 | * Avoid acquiring the list_lru lock if already tracked. The | |
166 | * list_empty() test is safe as node->private_list is | |
167 | * protected by mapping->tree_lock. | |
168 | */ | |
169 | if (!workingset_node_pages(node) && | |
170 | list_empty(&node->private_list)) { | |
171 | node->private_data = mapping; | |
172 | list_lru_add(&workingset_shadow_nodes, &node->private_list); | |
173 | } | |
91b0abe3 JW |
174 | } |
175 | ||
1da177e4 | 176 | /* |
e64a782f | 177 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 178 | * sure the page is locked and that nobody else uses it - or that usage |
c4843a75 GT |
179 | * is safe. The caller must hold the mapping's tree_lock and |
180 | * mem_cgroup_begin_page_stat(). | |
1da177e4 | 181 | */ |
c4843a75 GT |
182 | void __delete_from_page_cache(struct page *page, void *shadow, |
183 | struct mem_cgroup *memcg) | |
1da177e4 LT |
184 | { |
185 | struct address_space *mapping = page->mapping; | |
186 | ||
fe0bfaaf | 187 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
188 | /* |
189 | * if we're uptodate, flush out into the cleancache, otherwise | |
190 | * invalidate any existing cleancache entries. We can't leave | |
191 | * stale data around in the cleancache once our page is gone | |
192 | */ | |
193 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
194 | cleancache_put_page(page); | |
195 | else | |
3167760f | 196 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 197 | |
91b0abe3 JW |
198 | page_cache_tree_delete(mapping, page, shadow); |
199 | ||
1da177e4 | 200 | page->mapping = NULL; |
b85e0eff | 201 | /* Leave page->index set: truncation lookup relies upon it */ |
91b0abe3 | 202 | |
4165b9b4 MH |
203 | /* hugetlb pages do not participate in page cache accounting. */ |
204 | if (!PageHuge(page)) | |
205 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
4b02108a KM |
206 | if (PageSwapBacked(page)) |
207 | __dec_zone_page_state(page, NR_SHMEM); | |
e1534ae9 | 208 | VM_BUG_ON_PAGE(page_mapped(page), page); |
3a692790 LT |
209 | |
210 | /* | |
b9ea2515 KK |
211 | * At this point page must be either written or cleaned by truncate. |
212 | * Dirty page here signals a bug and loss of unwritten data. | |
3a692790 | 213 | * |
b9ea2515 KK |
214 | * This fixes dirty accounting after removing the page entirely but |
215 | * leaves PageDirty set: it has no effect for truncated page and | |
216 | * anyway will be cleared before returning page into buddy allocator. | |
3a692790 | 217 | */ |
b9ea2515 | 218 | if (WARN_ON_ONCE(PageDirty(page))) |
682aa8e1 TH |
219 | account_page_cleaned(page, mapping, memcg, |
220 | inode_to_wb(mapping->host)); | |
1da177e4 LT |
221 | } |
222 | ||
702cfbf9 MK |
223 | /** |
224 | * delete_from_page_cache - delete page from page cache | |
225 | * @page: the page which the kernel is trying to remove from page cache | |
226 | * | |
227 | * This must be called only on pages that have been verified to be in the page | |
228 | * cache and locked. It will never put the page into the free list, the caller | |
229 | * has a reference on the page. | |
230 | */ | |
231 | void delete_from_page_cache(struct page *page) | |
1da177e4 LT |
232 | { |
233 | struct address_space *mapping = page->mapping; | |
c4843a75 GT |
234 | struct mem_cgroup *memcg; |
235 | unsigned long flags; | |
236 | ||
6072d13c | 237 | void (*freepage)(struct page *); |
1da177e4 | 238 | |
cd7619d6 | 239 | BUG_ON(!PageLocked(page)); |
1da177e4 | 240 | |
6072d13c | 241 | freepage = mapping->a_ops->freepage; |
c4843a75 GT |
242 | |
243 | memcg = mem_cgroup_begin_page_stat(page); | |
244 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
245 | __delete_from_page_cache(page, NULL, memcg); | |
246 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
247 | mem_cgroup_end_page_stat(memcg); | |
6072d13c LT |
248 | |
249 | if (freepage) | |
250 | freepage(page); | |
97cecb5a MK |
251 | page_cache_release(page); |
252 | } | |
253 | EXPORT_SYMBOL(delete_from_page_cache); | |
254 | ||
865ffef3 DM |
255 | static int filemap_check_errors(struct address_space *mapping) |
256 | { | |
257 | int ret = 0; | |
258 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
259 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
260 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 261 | ret = -ENOSPC; |
7fcbbaf1 JA |
262 | if (test_bit(AS_EIO, &mapping->flags) && |
263 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
264 | ret = -EIO; |
265 | return ret; | |
266 | } | |
267 | ||
1da177e4 | 268 | /** |
485bb99b | 269 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
270 | * @mapping: address space structure to write |
271 | * @start: offset in bytes where the range starts | |
469eb4d0 | 272 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 273 | * @sync_mode: enable synchronous operation |
1da177e4 | 274 | * |
485bb99b RD |
275 | * Start writeback against all of a mapping's dirty pages that lie |
276 | * within the byte offsets <start, end> inclusive. | |
277 | * | |
1da177e4 | 278 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 279 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
280 | * these two operations is that if a dirty page/buffer is encountered, it must |
281 | * be waited upon, and not just skipped over. | |
282 | */ | |
ebcf28e1 AM |
283 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
284 | loff_t end, int sync_mode) | |
1da177e4 LT |
285 | { |
286 | int ret; | |
287 | struct writeback_control wbc = { | |
288 | .sync_mode = sync_mode, | |
05fe478d | 289 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
290 | .range_start = start, |
291 | .range_end = end, | |
1da177e4 LT |
292 | }; |
293 | ||
294 | if (!mapping_cap_writeback_dirty(mapping)) | |
295 | return 0; | |
296 | ||
b16b1deb | 297 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 298 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 299 | wbc_detach_inode(&wbc); |
1da177e4 LT |
300 | return ret; |
301 | } | |
302 | ||
303 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
304 | int sync_mode) | |
305 | { | |
111ebb6e | 306 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
307 | } |
308 | ||
309 | int filemap_fdatawrite(struct address_space *mapping) | |
310 | { | |
311 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
312 | } | |
313 | EXPORT_SYMBOL(filemap_fdatawrite); | |
314 | ||
f4c0a0fd | 315 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 316 | loff_t end) |
1da177e4 LT |
317 | { |
318 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
319 | } | |
f4c0a0fd | 320 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 321 | |
485bb99b RD |
322 | /** |
323 | * filemap_flush - mostly a non-blocking flush | |
324 | * @mapping: target address_space | |
325 | * | |
1da177e4 LT |
326 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
327 | * purposes - I/O may not be started against all dirty pages. | |
328 | */ | |
329 | int filemap_flush(struct address_space *mapping) | |
330 | { | |
331 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
332 | } | |
333 | EXPORT_SYMBOL(filemap_flush); | |
334 | ||
aa750fd7 JN |
335 | static int __filemap_fdatawait_range(struct address_space *mapping, |
336 | loff_t start_byte, loff_t end_byte) | |
1da177e4 | 337 | { |
94004ed7 CH |
338 | pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; |
339 | pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; | |
1da177e4 LT |
340 | struct pagevec pvec; |
341 | int nr_pages; | |
aa750fd7 | 342 | int ret = 0; |
1da177e4 | 343 | |
94004ed7 | 344 | if (end_byte < start_byte) |
865ffef3 | 345 | goto out; |
1da177e4 LT |
346 | |
347 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
348 | while ((index <= end) && |
349 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
350 | PAGECACHE_TAG_WRITEBACK, | |
351 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
352 | unsigned i; | |
353 | ||
354 | for (i = 0; i < nr_pages; i++) { | |
355 | struct page *page = pvec.pages[i]; | |
356 | ||
357 | /* until radix tree lookup accepts end_index */ | |
358 | if (page->index > end) | |
359 | continue; | |
360 | ||
361 | wait_on_page_writeback(page); | |
212260aa | 362 | if (TestClearPageError(page)) |
1da177e4 LT |
363 | ret = -EIO; |
364 | } | |
365 | pagevec_release(&pvec); | |
366 | cond_resched(); | |
367 | } | |
865ffef3 | 368 | out: |
aa750fd7 JN |
369 | return ret; |
370 | } | |
371 | ||
372 | /** | |
373 | * filemap_fdatawait_range - wait for writeback to complete | |
374 | * @mapping: address space structure to wait for | |
375 | * @start_byte: offset in bytes where the range starts | |
376 | * @end_byte: offset in bytes where the range ends (inclusive) | |
377 | * | |
378 | * Walk the list of under-writeback pages of the given address space | |
379 | * in the given range and wait for all of them. Check error status of | |
380 | * the address space and return it. | |
381 | * | |
382 | * Since the error status of the address space is cleared by this function, | |
383 | * callers are responsible for checking the return value and handling and/or | |
384 | * reporting the error. | |
385 | */ | |
386 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
387 | loff_t end_byte) | |
388 | { | |
389 | int ret, ret2; | |
390 | ||
391 | ret = __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
865ffef3 DM |
392 | ret2 = filemap_check_errors(mapping); |
393 | if (!ret) | |
394 | ret = ret2; | |
1da177e4 LT |
395 | |
396 | return ret; | |
397 | } | |
d3bccb6f JK |
398 | EXPORT_SYMBOL(filemap_fdatawait_range); |
399 | ||
aa750fd7 JN |
400 | /** |
401 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
402 | * @mapping: address space structure to wait for | |
403 | * | |
404 | * Walk the list of under-writeback pages of the given address space | |
405 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
406 | * does not clear error status of the address space. | |
407 | * | |
408 | * Use this function if callers don't handle errors themselves. Expected | |
409 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
410 | * fsfreeze(8) | |
411 | */ | |
412 | void filemap_fdatawait_keep_errors(struct address_space *mapping) | |
413 | { | |
414 | loff_t i_size = i_size_read(mapping->host); | |
415 | ||
416 | if (i_size == 0) | |
417 | return; | |
418 | ||
419 | __filemap_fdatawait_range(mapping, 0, i_size - 1); | |
420 | } | |
421 | ||
1da177e4 | 422 | /** |
485bb99b | 423 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 424 | * @mapping: address space structure to wait for |
485bb99b RD |
425 | * |
426 | * Walk the list of under-writeback pages of the given address space | |
aa750fd7 JN |
427 | * and wait for all of them. Check error status of the address space |
428 | * and return it. | |
429 | * | |
430 | * Since the error status of the address space is cleared by this function, | |
431 | * callers are responsible for checking the return value and handling and/or | |
432 | * reporting the error. | |
1da177e4 LT |
433 | */ |
434 | int filemap_fdatawait(struct address_space *mapping) | |
435 | { | |
436 | loff_t i_size = i_size_read(mapping->host); | |
437 | ||
438 | if (i_size == 0) | |
439 | return 0; | |
440 | ||
94004ed7 | 441 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
442 | } |
443 | EXPORT_SYMBOL(filemap_fdatawait); | |
444 | ||
445 | int filemap_write_and_wait(struct address_space *mapping) | |
446 | { | |
28fd1298 | 447 | int err = 0; |
1da177e4 LT |
448 | |
449 | if (mapping->nrpages) { | |
28fd1298 OH |
450 | err = filemap_fdatawrite(mapping); |
451 | /* | |
452 | * Even if the above returned error, the pages may be | |
453 | * written partially (e.g. -ENOSPC), so we wait for it. | |
454 | * But the -EIO is special case, it may indicate the worst | |
455 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
456 | */ | |
457 | if (err != -EIO) { | |
458 | int err2 = filemap_fdatawait(mapping); | |
459 | if (!err) | |
460 | err = err2; | |
461 | } | |
865ffef3 DM |
462 | } else { |
463 | err = filemap_check_errors(mapping); | |
1da177e4 | 464 | } |
28fd1298 | 465 | return err; |
1da177e4 | 466 | } |
28fd1298 | 467 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 468 | |
485bb99b RD |
469 | /** |
470 | * filemap_write_and_wait_range - write out & wait on a file range | |
471 | * @mapping: the address_space for the pages | |
472 | * @lstart: offset in bytes where the range starts | |
473 | * @lend: offset in bytes where the range ends (inclusive) | |
474 | * | |
469eb4d0 AM |
475 | * Write out and wait upon file offsets lstart->lend, inclusive. |
476 | * | |
477 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
478 | * that this function can be used to write to the very end-of-file (end = -1). | |
479 | */ | |
1da177e4 LT |
480 | int filemap_write_and_wait_range(struct address_space *mapping, |
481 | loff_t lstart, loff_t lend) | |
482 | { | |
28fd1298 | 483 | int err = 0; |
1da177e4 | 484 | |
9973c98e RZ |
485 | if (dax_mapping(mapping) && mapping->nrexceptional) { |
486 | err = dax_writeback_mapping_range(mapping, lstart, lend); | |
487 | if (err) | |
488 | return err; | |
489 | } | |
490 | ||
1da177e4 | 491 | if (mapping->nrpages) { |
28fd1298 OH |
492 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
493 | WB_SYNC_ALL); | |
494 | /* See comment of filemap_write_and_wait() */ | |
495 | if (err != -EIO) { | |
94004ed7 CH |
496 | int err2 = filemap_fdatawait_range(mapping, |
497 | lstart, lend); | |
28fd1298 OH |
498 | if (!err) |
499 | err = err2; | |
500 | } | |
865ffef3 DM |
501 | } else { |
502 | err = filemap_check_errors(mapping); | |
1da177e4 | 503 | } |
28fd1298 | 504 | return err; |
1da177e4 | 505 | } |
f6995585 | 506 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 507 | |
ef6a3c63 MS |
508 | /** |
509 | * replace_page_cache_page - replace a pagecache page with a new one | |
510 | * @old: page to be replaced | |
511 | * @new: page to replace with | |
512 | * @gfp_mask: allocation mode | |
513 | * | |
514 | * This function replaces a page in the pagecache with a new one. On | |
515 | * success it acquires the pagecache reference for the new page and | |
516 | * drops it for the old page. Both the old and new pages must be | |
517 | * locked. This function does not add the new page to the LRU, the | |
518 | * caller must do that. | |
519 | * | |
520 | * The remove + add is atomic. The only way this function can fail is | |
521 | * memory allocation failure. | |
522 | */ | |
523 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
524 | { | |
525 | int error; | |
ef6a3c63 | 526 | |
309381fe SL |
527 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
528 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
529 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 530 | |
ef6a3c63 MS |
531 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
532 | if (!error) { | |
533 | struct address_space *mapping = old->mapping; | |
534 | void (*freepage)(struct page *); | |
c4843a75 GT |
535 | struct mem_cgroup *memcg; |
536 | unsigned long flags; | |
ef6a3c63 MS |
537 | |
538 | pgoff_t offset = old->index; | |
539 | freepage = mapping->a_ops->freepage; | |
540 | ||
541 | page_cache_get(new); | |
542 | new->mapping = mapping; | |
543 | new->index = offset; | |
544 | ||
c4843a75 GT |
545 | memcg = mem_cgroup_begin_page_stat(old); |
546 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
547 | __delete_from_page_cache(old, NULL, memcg); | |
ef6a3c63 MS |
548 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
549 | BUG_ON(error); | |
550 | mapping->nrpages++; | |
4165b9b4 MH |
551 | |
552 | /* | |
553 | * hugetlb pages do not participate in page cache accounting. | |
554 | */ | |
555 | if (!PageHuge(new)) | |
556 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
ef6a3c63 MS |
557 | if (PageSwapBacked(new)) |
558 | __inc_zone_page_state(new, NR_SHMEM); | |
c4843a75 GT |
559 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
560 | mem_cgroup_end_page_stat(memcg); | |
45637bab | 561 | mem_cgroup_replace_page(old, new); |
ef6a3c63 MS |
562 | radix_tree_preload_end(); |
563 | if (freepage) | |
564 | freepage(old); | |
565 | page_cache_release(old); | |
ef6a3c63 MS |
566 | } |
567 | ||
568 | return error; | |
569 | } | |
570 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
571 | ||
0cd6144a | 572 | static int page_cache_tree_insert(struct address_space *mapping, |
a528910e | 573 | struct page *page, void **shadowp) |
0cd6144a | 574 | { |
449dd698 | 575 | struct radix_tree_node *node; |
0cd6144a JW |
576 | void **slot; |
577 | int error; | |
578 | ||
449dd698 JW |
579 | error = __radix_tree_create(&mapping->page_tree, page->index, |
580 | &node, &slot); | |
581 | if (error) | |
582 | return error; | |
583 | if (*slot) { | |
0cd6144a JW |
584 | void *p; |
585 | ||
586 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
587 | if (!radix_tree_exceptional_entry(p)) | |
588 | return -EEXIST; | |
f9fe48be RZ |
589 | |
590 | if (WARN_ON(dax_mapping(mapping))) | |
591 | return -EINVAL; | |
592 | ||
a528910e JW |
593 | if (shadowp) |
594 | *shadowp = p; | |
f9fe48be | 595 | mapping->nrexceptional--; |
449dd698 JW |
596 | if (node) |
597 | workingset_node_shadows_dec(node); | |
0cd6144a | 598 | } |
449dd698 JW |
599 | radix_tree_replace_slot(slot, page); |
600 | mapping->nrpages++; | |
601 | if (node) { | |
602 | workingset_node_pages_inc(node); | |
603 | /* | |
604 | * Don't track node that contains actual pages. | |
605 | * | |
606 | * Avoid acquiring the list_lru lock if already | |
607 | * untracked. The list_empty() test is safe as | |
608 | * node->private_list is protected by | |
609 | * mapping->tree_lock. | |
610 | */ | |
611 | if (!list_empty(&node->private_list)) | |
612 | list_lru_del(&workingset_shadow_nodes, | |
613 | &node->private_list); | |
614 | } | |
615 | return 0; | |
0cd6144a JW |
616 | } |
617 | ||
a528910e JW |
618 | static int __add_to_page_cache_locked(struct page *page, |
619 | struct address_space *mapping, | |
620 | pgoff_t offset, gfp_t gfp_mask, | |
621 | void **shadowp) | |
1da177e4 | 622 | { |
00501b53 JW |
623 | int huge = PageHuge(page); |
624 | struct mem_cgroup *memcg; | |
e286781d NP |
625 | int error; |
626 | ||
309381fe SL |
627 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
628 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d | 629 | |
00501b53 JW |
630 | if (!huge) { |
631 | error = mem_cgroup_try_charge(page, current->mm, | |
f627c2f5 | 632 | gfp_mask, &memcg, false); |
00501b53 JW |
633 | if (error) |
634 | return error; | |
635 | } | |
1da177e4 | 636 | |
5e4c0d97 | 637 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 638 | if (error) { |
00501b53 | 639 | if (!huge) |
f627c2f5 | 640 | mem_cgroup_cancel_charge(page, memcg, false); |
66a0c8ee KS |
641 | return error; |
642 | } | |
643 | ||
644 | page_cache_get(page); | |
645 | page->mapping = mapping; | |
646 | page->index = offset; | |
647 | ||
648 | spin_lock_irq(&mapping->tree_lock); | |
a528910e | 649 | error = page_cache_tree_insert(mapping, page, shadowp); |
66a0c8ee KS |
650 | radix_tree_preload_end(); |
651 | if (unlikely(error)) | |
652 | goto err_insert; | |
4165b9b4 MH |
653 | |
654 | /* hugetlb pages do not participate in page cache accounting. */ | |
655 | if (!huge) | |
656 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
66a0c8ee | 657 | spin_unlock_irq(&mapping->tree_lock); |
00501b53 | 658 | if (!huge) |
f627c2f5 | 659 | mem_cgroup_commit_charge(page, memcg, false, false); |
66a0c8ee KS |
660 | trace_mm_filemap_add_to_page_cache(page); |
661 | return 0; | |
662 | err_insert: | |
663 | page->mapping = NULL; | |
664 | /* Leave page->index set: truncation relies upon it */ | |
665 | spin_unlock_irq(&mapping->tree_lock); | |
00501b53 | 666 | if (!huge) |
f627c2f5 | 667 | mem_cgroup_cancel_charge(page, memcg, false); |
66a0c8ee | 668 | page_cache_release(page); |
1da177e4 LT |
669 | return error; |
670 | } | |
a528910e JW |
671 | |
672 | /** | |
673 | * add_to_page_cache_locked - add a locked page to the pagecache | |
674 | * @page: page to add | |
675 | * @mapping: the page's address_space | |
676 | * @offset: page index | |
677 | * @gfp_mask: page allocation mode | |
678 | * | |
679 | * This function is used to add a page to the pagecache. It must be locked. | |
680 | * This function does not add the page to the LRU. The caller must do that. | |
681 | */ | |
682 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
683 | pgoff_t offset, gfp_t gfp_mask) | |
684 | { | |
685 | return __add_to_page_cache_locked(page, mapping, offset, | |
686 | gfp_mask, NULL); | |
687 | } | |
e286781d | 688 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
689 | |
690 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 691 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 692 | { |
a528910e | 693 | void *shadow = NULL; |
4f98a2fe RR |
694 | int ret; |
695 | ||
48c935ad | 696 | __SetPageLocked(page); |
a528910e JW |
697 | ret = __add_to_page_cache_locked(page, mapping, offset, |
698 | gfp_mask, &shadow); | |
699 | if (unlikely(ret)) | |
48c935ad | 700 | __ClearPageLocked(page); |
a528910e JW |
701 | else { |
702 | /* | |
703 | * The page might have been evicted from cache only | |
704 | * recently, in which case it should be activated like | |
705 | * any other repeatedly accessed page. | |
706 | */ | |
707 | if (shadow && workingset_refault(shadow)) { | |
708 | SetPageActive(page); | |
709 | workingset_activation(page); | |
710 | } else | |
711 | ClearPageActive(page); | |
712 | lru_cache_add(page); | |
713 | } | |
1da177e4 LT |
714 | return ret; |
715 | } | |
18bc0bbd | 716 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 717 | |
44110fe3 | 718 | #ifdef CONFIG_NUMA |
2ae88149 | 719 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 720 | { |
c0ff7453 MX |
721 | int n; |
722 | struct page *page; | |
723 | ||
44110fe3 | 724 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
725 | unsigned int cpuset_mems_cookie; |
726 | do { | |
d26914d1 | 727 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 728 | n = cpuset_mem_spread_node(); |
96db800f | 729 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 730 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 731 | |
c0ff7453 | 732 | return page; |
44110fe3 | 733 | } |
2ae88149 | 734 | return alloc_pages(gfp, 0); |
44110fe3 | 735 | } |
2ae88149 | 736 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
737 | #endif |
738 | ||
1da177e4 LT |
739 | /* |
740 | * In order to wait for pages to become available there must be | |
741 | * waitqueues associated with pages. By using a hash table of | |
742 | * waitqueues where the bucket discipline is to maintain all | |
743 | * waiters on the same queue and wake all when any of the pages | |
744 | * become available, and for the woken contexts to check to be | |
745 | * sure the appropriate page became available, this saves space | |
746 | * at a cost of "thundering herd" phenomena during rare hash | |
747 | * collisions. | |
748 | */ | |
a4796e37 | 749 | wait_queue_head_t *page_waitqueue(struct page *page) |
1da177e4 LT |
750 | { |
751 | const struct zone *zone = page_zone(page); | |
752 | ||
753 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
754 | } | |
a4796e37 | 755 | EXPORT_SYMBOL(page_waitqueue); |
1da177e4 | 756 | |
920c7a5d | 757 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
758 | { |
759 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
760 | ||
761 | if (test_bit(bit_nr, &page->flags)) | |
74316201 | 762 | __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
763 | TASK_UNINTERRUPTIBLE); |
764 | } | |
765 | EXPORT_SYMBOL(wait_on_page_bit); | |
766 | ||
f62e00cc KM |
767 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
768 | { | |
769 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
770 | ||
771 | if (!test_bit(bit_nr, &page->flags)) | |
772 | return 0; | |
773 | ||
774 | return __wait_on_bit(page_waitqueue(page), &wait, | |
74316201 | 775 | bit_wait_io, TASK_KILLABLE); |
f62e00cc KM |
776 | } |
777 | ||
cbbce822 N |
778 | int wait_on_page_bit_killable_timeout(struct page *page, |
779 | int bit_nr, unsigned long timeout) | |
780 | { | |
781 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
782 | ||
783 | wait.key.timeout = jiffies + timeout; | |
784 | if (!test_bit(bit_nr, &page->flags)) | |
785 | return 0; | |
786 | return __wait_on_bit(page_waitqueue(page), &wait, | |
787 | bit_wait_io_timeout, TASK_KILLABLE); | |
788 | } | |
789 | EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout); | |
790 | ||
385e1ca5 DH |
791 | /** |
792 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
793 | * @page: Page defining the wait queue of interest |
794 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
795 | * |
796 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
797 | */ | |
798 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
799 | { | |
800 | wait_queue_head_t *q = page_waitqueue(page); | |
801 | unsigned long flags; | |
802 | ||
803 | spin_lock_irqsave(&q->lock, flags); | |
804 | __add_wait_queue(q, waiter); | |
805 | spin_unlock_irqrestore(&q->lock, flags); | |
806 | } | |
807 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
808 | ||
1da177e4 | 809 | /** |
485bb99b | 810 | * unlock_page - unlock a locked page |
1da177e4 LT |
811 | * @page: the page |
812 | * | |
813 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
814 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 815 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
816 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
817 | * | |
8413ac9d NP |
818 | * The mb is necessary to enforce ordering between the clear_bit and the read |
819 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 820 | */ |
920c7a5d | 821 | void unlock_page(struct page *page) |
1da177e4 | 822 | { |
48c935ad | 823 | page = compound_head(page); |
309381fe | 824 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
8413ac9d | 825 | clear_bit_unlock(PG_locked, &page->flags); |
4e857c58 | 826 | smp_mb__after_atomic(); |
1da177e4 LT |
827 | wake_up_page(page, PG_locked); |
828 | } | |
829 | EXPORT_SYMBOL(unlock_page); | |
830 | ||
485bb99b RD |
831 | /** |
832 | * end_page_writeback - end writeback against a page | |
833 | * @page: the page | |
1da177e4 LT |
834 | */ |
835 | void end_page_writeback(struct page *page) | |
836 | { | |
888cf2db MG |
837 | /* |
838 | * TestClearPageReclaim could be used here but it is an atomic | |
839 | * operation and overkill in this particular case. Failing to | |
840 | * shuffle a page marked for immediate reclaim is too mild to | |
841 | * justify taking an atomic operation penalty at the end of | |
842 | * ever page writeback. | |
843 | */ | |
844 | if (PageReclaim(page)) { | |
845 | ClearPageReclaim(page); | |
ac6aadb2 | 846 | rotate_reclaimable_page(page); |
888cf2db | 847 | } |
ac6aadb2 MS |
848 | |
849 | if (!test_clear_page_writeback(page)) | |
850 | BUG(); | |
851 | ||
4e857c58 | 852 | smp_mb__after_atomic(); |
1da177e4 LT |
853 | wake_up_page(page, PG_writeback); |
854 | } | |
855 | EXPORT_SYMBOL(end_page_writeback); | |
856 | ||
57d99845 MW |
857 | /* |
858 | * After completing I/O on a page, call this routine to update the page | |
859 | * flags appropriately | |
860 | */ | |
861 | void page_endio(struct page *page, int rw, int err) | |
862 | { | |
863 | if (rw == READ) { | |
864 | if (!err) { | |
865 | SetPageUptodate(page); | |
866 | } else { | |
867 | ClearPageUptodate(page); | |
868 | SetPageError(page); | |
869 | } | |
870 | unlock_page(page); | |
871 | } else { /* rw == WRITE */ | |
872 | if (err) { | |
873 | SetPageError(page); | |
874 | if (page->mapping) | |
875 | mapping_set_error(page->mapping, err); | |
876 | } | |
877 | end_page_writeback(page); | |
878 | } | |
879 | } | |
880 | EXPORT_SYMBOL_GPL(page_endio); | |
881 | ||
485bb99b RD |
882 | /** |
883 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
884 | * @page: the page to lock | |
1da177e4 | 885 | */ |
920c7a5d | 886 | void __lock_page(struct page *page) |
1da177e4 | 887 | { |
48c935ad KS |
888 | struct page *page_head = compound_head(page); |
889 | DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked); | |
1da177e4 | 890 | |
48c935ad | 891 | __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io, |
1da177e4 LT |
892 | TASK_UNINTERRUPTIBLE); |
893 | } | |
894 | EXPORT_SYMBOL(__lock_page); | |
895 | ||
b5606c2d | 896 | int __lock_page_killable(struct page *page) |
2687a356 | 897 | { |
48c935ad KS |
898 | struct page *page_head = compound_head(page); |
899 | DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked); | |
2687a356 | 900 | |
48c935ad | 901 | return __wait_on_bit_lock(page_waitqueue(page_head), &wait, |
74316201 | 902 | bit_wait_io, TASK_KILLABLE); |
2687a356 | 903 | } |
18bc0bbd | 904 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 905 | |
9a95f3cf PC |
906 | /* |
907 | * Return values: | |
908 | * 1 - page is locked; mmap_sem is still held. | |
909 | * 0 - page is not locked. | |
910 | * mmap_sem has been released (up_read()), unless flags had both | |
911 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
912 | * which case mmap_sem is still held. | |
913 | * | |
914 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
915 | * with the page locked and the mmap_sem unperturbed. | |
916 | */ | |
d065bd81 ML |
917 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
918 | unsigned int flags) | |
919 | { | |
37b23e05 KM |
920 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
921 | /* | |
922 | * CAUTION! In this case, mmap_sem is not released | |
923 | * even though return 0. | |
924 | */ | |
925 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
926 | return 0; | |
927 | ||
928 | up_read(&mm->mmap_sem); | |
929 | if (flags & FAULT_FLAG_KILLABLE) | |
930 | wait_on_page_locked_killable(page); | |
931 | else | |
318b275f | 932 | wait_on_page_locked(page); |
d065bd81 | 933 | return 0; |
37b23e05 KM |
934 | } else { |
935 | if (flags & FAULT_FLAG_KILLABLE) { | |
936 | int ret; | |
937 | ||
938 | ret = __lock_page_killable(page); | |
939 | if (ret) { | |
940 | up_read(&mm->mmap_sem); | |
941 | return 0; | |
942 | } | |
943 | } else | |
944 | __lock_page(page); | |
945 | return 1; | |
d065bd81 ML |
946 | } |
947 | } | |
948 | ||
e7b563bb JW |
949 | /** |
950 | * page_cache_next_hole - find the next hole (not-present entry) | |
951 | * @mapping: mapping | |
952 | * @index: index | |
953 | * @max_scan: maximum range to search | |
954 | * | |
955 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
956 | * lowest indexed hole. | |
957 | * | |
958 | * Returns: the index of the hole if found, otherwise returns an index | |
959 | * outside of the set specified (in which case 'return - index >= | |
960 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
961 | * be returned. | |
962 | * | |
963 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
964 | * like radix_tree_gang_lookup, this will not atomically search a | |
965 | * snapshot of the tree at a single point in time. For example, if a | |
966 | * hole is created at index 5, then subsequently a hole is created at | |
967 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
968 | * if called under rcu_read_lock. | |
969 | */ | |
970 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
971 | pgoff_t index, unsigned long max_scan) | |
972 | { | |
973 | unsigned long i; | |
974 | ||
975 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
976 | struct page *page; |
977 | ||
978 | page = radix_tree_lookup(&mapping->page_tree, index); | |
979 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
980 | break; |
981 | index++; | |
982 | if (index == 0) | |
983 | break; | |
984 | } | |
985 | ||
986 | return index; | |
987 | } | |
988 | EXPORT_SYMBOL(page_cache_next_hole); | |
989 | ||
990 | /** | |
991 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
992 | * @mapping: mapping | |
993 | * @index: index | |
994 | * @max_scan: maximum range to search | |
995 | * | |
996 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
997 | * the first hole. | |
998 | * | |
999 | * Returns: the index of the hole if found, otherwise returns an index | |
1000 | * outside of the set specified (in which case 'index - return >= | |
1001 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
1002 | * will be returned. | |
1003 | * | |
1004 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
1005 | * like radix_tree_gang_lookup, this will not atomically search a | |
1006 | * snapshot of the tree at a single point in time. For example, if a | |
1007 | * hole is created at index 10, then subsequently a hole is created at | |
1008 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
1009 | * called under rcu_read_lock. | |
1010 | */ | |
1011 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
1012 | pgoff_t index, unsigned long max_scan) | |
1013 | { | |
1014 | unsigned long i; | |
1015 | ||
1016 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
1017 | struct page *page; |
1018 | ||
1019 | page = radix_tree_lookup(&mapping->page_tree, index); | |
1020 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
1021 | break; |
1022 | index--; | |
1023 | if (index == ULONG_MAX) | |
1024 | break; | |
1025 | } | |
1026 | ||
1027 | return index; | |
1028 | } | |
1029 | EXPORT_SYMBOL(page_cache_prev_hole); | |
1030 | ||
485bb99b | 1031 | /** |
0cd6144a | 1032 | * find_get_entry - find and get a page cache entry |
485bb99b | 1033 | * @mapping: the address_space to search |
0cd6144a JW |
1034 | * @offset: the page cache index |
1035 | * | |
1036 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1037 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1038 | * |
139b6a6f JW |
1039 | * If the slot holds a shadow entry of a previously evicted page, or a |
1040 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1041 | * |
1042 | * Otherwise, %NULL is returned. | |
1da177e4 | 1043 | */ |
0cd6144a | 1044 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1045 | { |
a60637c8 | 1046 | void **pagep; |
1da177e4 LT |
1047 | struct page *page; |
1048 | ||
a60637c8 NP |
1049 | rcu_read_lock(); |
1050 | repeat: | |
1051 | page = NULL; | |
1052 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
1053 | if (pagep) { | |
1054 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
1055 | if (unlikely(!page)) |
1056 | goto out; | |
a2c16d6c | 1057 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1058 | if (radix_tree_deref_retry(page)) |
1059 | goto repeat; | |
1060 | /* | |
139b6a6f JW |
1061 | * A shadow entry of a recently evicted page, |
1062 | * or a swap entry from shmem/tmpfs. Return | |
1063 | * it without attempting to raise page count. | |
8079b1c8 HD |
1064 | */ |
1065 | goto out; | |
a2c16d6c | 1066 | } |
a60637c8 NP |
1067 | if (!page_cache_get_speculative(page)) |
1068 | goto repeat; | |
1069 | ||
1070 | /* | |
1071 | * Has the page moved? | |
1072 | * This is part of the lockless pagecache protocol. See | |
1073 | * include/linux/pagemap.h for details. | |
1074 | */ | |
1075 | if (unlikely(page != *pagep)) { | |
1076 | page_cache_release(page); | |
1077 | goto repeat; | |
1078 | } | |
1079 | } | |
27d20fdd | 1080 | out: |
a60637c8 NP |
1081 | rcu_read_unlock(); |
1082 | ||
1da177e4 LT |
1083 | return page; |
1084 | } | |
0cd6144a | 1085 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1086 | |
0cd6144a JW |
1087 | /** |
1088 | * find_lock_entry - locate, pin and lock a page cache entry | |
1089 | * @mapping: the address_space to search | |
1090 | * @offset: the page cache index | |
1091 | * | |
1092 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1093 | * page cache page, it is returned locked and with an increased | |
1094 | * refcount. | |
1095 | * | |
139b6a6f JW |
1096 | * If the slot holds a shadow entry of a previously evicted page, or a |
1097 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1098 | * |
1099 | * Otherwise, %NULL is returned. | |
1100 | * | |
1101 | * find_lock_entry() may sleep. | |
1102 | */ | |
1103 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1104 | { |
1105 | struct page *page; | |
1106 | ||
1da177e4 | 1107 | repeat: |
0cd6144a | 1108 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1109 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1110 | lock_page(page); |
1111 | /* Has the page been truncated? */ | |
1112 | if (unlikely(page->mapping != mapping)) { | |
1113 | unlock_page(page); | |
1114 | page_cache_release(page); | |
1115 | goto repeat; | |
1da177e4 | 1116 | } |
309381fe | 1117 | VM_BUG_ON_PAGE(page->index != offset, page); |
1da177e4 | 1118 | } |
1da177e4 LT |
1119 | return page; |
1120 | } | |
0cd6144a JW |
1121 | EXPORT_SYMBOL(find_lock_entry); |
1122 | ||
1123 | /** | |
2457aec6 | 1124 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1125 | * @mapping: the address_space to search |
1126 | * @offset: the page index | |
2457aec6 | 1127 | * @fgp_flags: PCG flags |
45f87de5 | 1128 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
0cd6144a | 1129 | * |
2457aec6 | 1130 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1131 | * |
75325189 | 1132 | * PCG flags modify how the page is returned. |
0cd6144a | 1133 | * |
2457aec6 MG |
1134 | * FGP_ACCESSED: the page will be marked accessed |
1135 | * FGP_LOCK: Page is return locked | |
1136 | * FGP_CREAT: If page is not present then a new page is allocated using | |
45f87de5 MH |
1137 | * @gfp_mask and added to the page cache and the VM's LRU |
1138 | * list. The page is returned locked and with an increased | |
1139 | * refcount. Otherwise, %NULL is returned. | |
1da177e4 | 1140 | * |
2457aec6 MG |
1141 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1142 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1143 | * |
2457aec6 | 1144 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1145 | */ |
2457aec6 | 1146 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
45f87de5 | 1147 | int fgp_flags, gfp_t gfp_mask) |
1da177e4 | 1148 | { |
eb2be189 | 1149 | struct page *page; |
2457aec6 | 1150 | |
1da177e4 | 1151 | repeat: |
2457aec6 MG |
1152 | page = find_get_entry(mapping, offset); |
1153 | if (radix_tree_exceptional_entry(page)) | |
1154 | page = NULL; | |
1155 | if (!page) | |
1156 | goto no_page; | |
1157 | ||
1158 | if (fgp_flags & FGP_LOCK) { | |
1159 | if (fgp_flags & FGP_NOWAIT) { | |
1160 | if (!trylock_page(page)) { | |
1161 | page_cache_release(page); | |
1162 | return NULL; | |
1163 | } | |
1164 | } else { | |
1165 | lock_page(page); | |
1166 | } | |
1167 | ||
1168 | /* Has the page been truncated? */ | |
1169 | if (unlikely(page->mapping != mapping)) { | |
1170 | unlock_page(page); | |
1171 | page_cache_release(page); | |
1172 | goto repeat; | |
1173 | } | |
1174 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1175 | } | |
1176 | ||
1177 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1178 | mark_page_accessed(page); | |
1179 | ||
1180 | no_page: | |
1181 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1182 | int err; | |
1183 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1184 | gfp_mask |= __GFP_WRITE; |
1185 | if (fgp_flags & FGP_NOFS) | |
1186 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1187 | |
45f87de5 | 1188 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1189 | if (!page) |
1190 | return NULL; | |
2457aec6 MG |
1191 | |
1192 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1193 | fgp_flags |= FGP_LOCK; | |
1194 | ||
eb39d618 | 1195 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1196 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1197 | __SetPageReferenced(page); |
2457aec6 | 1198 | |
45f87de5 MH |
1199 | err = add_to_page_cache_lru(page, mapping, offset, |
1200 | gfp_mask & GFP_RECLAIM_MASK); | |
eb2be189 NP |
1201 | if (unlikely(err)) { |
1202 | page_cache_release(page); | |
1203 | page = NULL; | |
1204 | if (err == -EEXIST) | |
1205 | goto repeat; | |
1da177e4 | 1206 | } |
1da177e4 | 1207 | } |
2457aec6 | 1208 | |
1da177e4 LT |
1209 | return page; |
1210 | } | |
2457aec6 | 1211 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1212 | |
0cd6144a JW |
1213 | /** |
1214 | * find_get_entries - gang pagecache lookup | |
1215 | * @mapping: The address_space to search | |
1216 | * @start: The starting page cache index | |
1217 | * @nr_entries: The maximum number of entries | |
1218 | * @entries: Where the resulting entries are placed | |
1219 | * @indices: The cache indices corresponding to the entries in @entries | |
1220 | * | |
1221 | * find_get_entries() will search for and return a group of up to | |
1222 | * @nr_entries entries in the mapping. The entries are placed at | |
1223 | * @entries. find_get_entries() takes a reference against any actual | |
1224 | * pages it returns. | |
1225 | * | |
1226 | * The search returns a group of mapping-contiguous page cache entries | |
1227 | * with ascending indexes. There may be holes in the indices due to | |
1228 | * not-present pages. | |
1229 | * | |
139b6a6f JW |
1230 | * Any shadow entries of evicted pages, or swap entries from |
1231 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1232 | * |
1233 | * find_get_entries() returns the number of pages and shadow entries | |
1234 | * which were found. | |
1235 | */ | |
1236 | unsigned find_get_entries(struct address_space *mapping, | |
1237 | pgoff_t start, unsigned int nr_entries, | |
1238 | struct page **entries, pgoff_t *indices) | |
1239 | { | |
1240 | void **slot; | |
1241 | unsigned int ret = 0; | |
1242 | struct radix_tree_iter iter; | |
1243 | ||
1244 | if (!nr_entries) | |
1245 | return 0; | |
1246 | ||
1247 | rcu_read_lock(); | |
1248 | restart: | |
1249 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { | |
1250 | struct page *page; | |
1251 | repeat: | |
1252 | page = radix_tree_deref_slot(slot); | |
1253 | if (unlikely(!page)) | |
1254 | continue; | |
1255 | if (radix_tree_exception(page)) { | |
1256 | if (radix_tree_deref_retry(page)) | |
1257 | goto restart; | |
1258 | /* | |
f9fe48be RZ |
1259 | * A shadow entry of a recently evicted page, a swap |
1260 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1261 | * without attempting to raise page count. | |
0cd6144a JW |
1262 | */ |
1263 | goto export; | |
1264 | } | |
1265 | if (!page_cache_get_speculative(page)) | |
1266 | goto repeat; | |
1267 | ||
1268 | /* Has the page moved? */ | |
1269 | if (unlikely(page != *slot)) { | |
1270 | page_cache_release(page); | |
1271 | goto repeat; | |
1272 | } | |
1273 | export: | |
1274 | indices[ret] = iter.index; | |
1275 | entries[ret] = page; | |
1276 | if (++ret == nr_entries) | |
1277 | break; | |
1278 | } | |
1279 | rcu_read_unlock(); | |
1280 | return ret; | |
1281 | } | |
1282 | ||
1da177e4 LT |
1283 | /** |
1284 | * find_get_pages - gang pagecache lookup | |
1285 | * @mapping: The address_space to search | |
1286 | * @start: The starting page index | |
1287 | * @nr_pages: The maximum number of pages | |
1288 | * @pages: Where the resulting pages are placed | |
1289 | * | |
1290 | * find_get_pages() will search for and return a group of up to | |
1291 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
1292 | * find_get_pages() takes a reference against the returned pages. | |
1293 | * | |
1294 | * The search returns a group of mapping-contiguous pages with ascending | |
1295 | * indexes. There may be holes in the indices due to not-present pages. | |
1296 | * | |
1297 | * find_get_pages() returns the number of pages which were found. | |
1298 | */ | |
1299 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
1300 | unsigned int nr_pages, struct page **pages) | |
1301 | { | |
0fc9d104 KK |
1302 | struct radix_tree_iter iter; |
1303 | void **slot; | |
1304 | unsigned ret = 0; | |
1305 | ||
1306 | if (unlikely(!nr_pages)) | |
1307 | return 0; | |
a60637c8 NP |
1308 | |
1309 | rcu_read_lock(); | |
1310 | restart: | |
0fc9d104 | 1311 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
a60637c8 NP |
1312 | struct page *page; |
1313 | repeat: | |
0fc9d104 | 1314 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1315 | if (unlikely(!page)) |
1316 | continue; | |
9d8aa4ea | 1317 | |
a2c16d6c | 1318 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1319 | if (radix_tree_deref_retry(page)) { |
1320 | /* | |
1321 | * Transient condition which can only trigger | |
1322 | * when entry at index 0 moves out of or back | |
1323 | * to root: none yet gotten, safe to restart. | |
1324 | */ | |
0fc9d104 | 1325 | WARN_ON(iter.index); |
8079b1c8 HD |
1326 | goto restart; |
1327 | } | |
a2c16d6c | 1328 | /* |
139b6a6f JW |
1329 | * A shadow entry of a recently evicted page, |
1330 | * or a swap entry from shmem/tmpfs. Skip | |
1331 | * over it. | |
a2c16d6c | 1332 | */ |
8079b1c8 | 1333 | continue; |
27d20fdd | 1334 | } |
a60637c8 NP |
1335 | |
1336 | if (!page_cache_get_speculative(page)) | |
1337 | goto repeat; | |
1338 | ||
1339 | /* Has the page moved? */ | |
0fc9d104 | 1340 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1341 | page_cache_release(page); |
1342 | goto repeat; | |
1343 | } | |
1da177e4 | 1344 | |
a60637c8 | 1345 | pages[ret] = page; |
0fc9d104 KK |
1346 | if (++ret == nr_pages) |
1347 | break; | |
a60637c8 | 1348 | } |
5b280c0c | 1349 | |
a60637c8 | 1350 | rcu_read_unlock(); |
1da177e4 LT |
1351 | return ret; |
1352 | } | |
1353 | ||
ebf43500 JA |
1354 | /** |
1355 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1356 | * @mapping: The address_space to search | |
1357 | * @index: The starting page index | |
1358 | * @nr_pages: The maximum number of pages | |
1359 | * @pages: Where the resulting pages are placed | |
1360 | * | |
1361 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1362 | * that the returned number of pages are guaranteed to be contiguous. | |
1363 | * | |
1364 | * find_get_pages_contig() returns the number of pages which were found. | |
1365 | */ | |
1366 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1367 | unsigned int nr_pages, struct page **pages) | |
1368 | { | |
0fc9d104 KK |
1369 | struct radix_tree_iter iter; |
1370 | void **slot; | |
1371 | unsigned int ret = 0; | |
1372 | ||
1373 | if (unlikely(!nr_pages)) | |
1374 | return 0; | |
a60637c8 NP |
1375 | |
1376 | rcu_read_lock(); | |
1377 | restart: | |
0fc9d104 | 1378 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
a60637c8 NP |
1379 | struct page *page; |
1380 | repeat: | |
0fc9d104 KK |
1381 | page = radix_tree_deref_slot(slot); |
1382 | /* The hole, there no reason to continue */ | |
a60637c8 | 1383 | if (unlikely(!page)) |
0fc9d104 | 1384 | break; |
9d8aa4ea | 1385 | |
a2c16d6c | 1386 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1387 | if (radix_tree_deref_retry(page)) { |
1388 | /* | |
1389 | * Transient condition which can only trigger | |
1390 | * when entry at index 0 moves out of or back | |
1391 | * to root: none yet gotten, safe to restart. | |
1392 | */ | |
1393 | goto restart; | |
1394 | } | |
a2c16d6c | 1395 | /* |
139b6a6f JW |
1396 | * A shadow entry of a recently evicted page, |
1397 | * or a swap entry from shmem/tmpfs. Stop | |
1398 | * looking for contiguous pages. | |
a2c16d6c | 1399 | */ |
8079b1c8 | 1400 | break; |
a2c16d6c | 1401 | } |
ebf43500 | 1402 | |
a60637c8 NP |
1403 | if (!page_cache_get_speculative(page)) |
1404 | goto repeat; | |
1405 | ||
1406 | /* Has the page moved? */ | |
0fc9d104 | 1407 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1408 | page_cache_release(page); |
1409 | goto repeat; | |
1410 | } | |
1411 | ||
9cbb4cb2 NP |
1412 | /* |
1413 | * must check mapping and index after taking the ref. | |
1414 | * otherwise we can get both false positives and false | |
1415 | * negatives, which is just confusing to the caller. | |
1416 | */ | |
0fc9d104 | 1417 | if (page->mapping == NULL || page->index != iter.index) { |
9cbb4cb2 NP |
1418 | page_cache_release(page); |
1419 | break; | |
1420 | } | |
1421 | ||
a60637c8 | 1422 | pages[ret] = page; |
0fc9d104 KK |
1423 | if (++ret == nr_pages) |
1424 | break; | |
ebf43500 | 1425 | } |
a60637c8 NP |
1426 | rcu_read_unlock(); |
1427 | return ret; | |
ebf43500 | 1428 | } |
ef71c15c | 1429 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1430 | |
485bb99b RD |
1431 | /** |
1432 | * find_get_pages_tag - find and return pages that match @tag | |
1433 | * @mapping: the address_space to search | |
1434 | * @index: the starting page index | |
1435 | * @tag: the tag index | |
1436 | * @nr_pages: the maximum number of pages | |
1437 | * @pages: where the resulting pages are placed | |
1438 | * | |
1da177e4 | 1439 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1440 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
1441 | */ |
1442 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
1443 | int tag, unsigned int nr_pages, struct page **pages) | |
1444 | { | |
0fc9d104 KK |
1445 | struct radix_tree_iter iter; |
1446 | void **slot; | |
1447 | unsigned ret = 0; | |
1448 | ||
1449 | if (unlikely(!nr_pages)) | |
1450 | return 0; | |
a60637c8 NP |
1451 | |
1452 | rcu_read_lock(); | |
1453 | restart: | |
0fc9d104 KK |
1454 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1455 | &iter, *index, tag) { | |
a60637c8 NP |
1456 | struct page *page; |
1457 | repeat: | |
0fc9d104 | 1458 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1459 | if (unlikely(!page)) |
1460 | continue; | |
9d8aa4ea | 1461 | |
a2c16d6c | 1462 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1463 | if (radix_tree_deref_retry(page)) { |
1464 | /* | |
1465 | * Transient condition which can only trigger | |
1466 | * when entry at index 0 moves out of or back | |
1467 | * to root: none yet gotten, safe to restart. | |
1468 | */ | |
1469 | goto restart; | |
1470 | } | |
a2c16d6c | 1471 | /* |
139b6a6f JW |
1472 | * A shadow entry of a recently evicted page. |
1473 | * | |
1474 | * Those entries should never be tagged, but | |
1475 | * this tree walk is lockless and the tags are | |
1476 | * looked up in bulk, one radix tree node at a | |
1477 | * time, so there is a sizable window for page | |
1478 | * reclaim to evict a page we saw tagged. | |
1479 | * | |
1480 | * Skip over it. | |
a2c16d6c | 1481 | */ |
139b6a6f | 1482 | continue; |
a2c16d6c | 1483 | } |
a60637c8 NP |
1484 | |
1485 | if (!page_cache_get_speculative(page)) | |
1486 | goto repeat; | |
1487 | ||
1488 | /* Has the page moved? */ | |
0fc9d104 | 1489 | if (unlikely(page != *slot)) { |
a60637c8 NP |
1490 | page_cache_release(page); |
1491 | goto repeat; | |
1492 | } | |
1493 | ||
1494 | pages[ret] = page; | |
0fc9d104 KK |
1495 | if (++ret == nr_pages) |
1496 | break; | |
a60637c8 | 1497 | } |
5b280c0c | 1498 | |
a60637c8 | 1499 | rcu_read_unlock(); |
1da177e4 | 1500 | |
1da177e4 LT |
1501 | if (ret) |
1502 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1503 | |
1da177e4 LT |
1504 | return ret; |
1505 | } | |
ef71c15c | 1506 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1507 | |
7e7f7749 RZ |
1508 | /** |
1509 | * find_get_entries_tag - find and return entries that match @tag | |
1510 | * @mapping: the address_space to search | |
1511 | * @start: the starting page cache index | |
1512 | * @tag: the tag index | |
1513 | * @nr_entries: the maximum number of entries | |
1514 | * @entries: where the resulting entries are placed | |
1515 | * @indices: the cache indices corresponding to the entries in @entries | |
1516 | * | |
1517 | * Like find_get_entries, except we only return entries which are tagged with | |
1518 | * @tag. | |
1519 | */ | |
1520 | unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start, | |
1521 | int tag, unsigned int nr_entries, | |
1522 | struct page **entries, pgoff_t *indices) | |
1523 | { | |
1524 | void **slot; | |
1525 | unsigned int ret = 0; | |
1526 | struct radix_tree_iter iter; | |
1527 | ||
1528 | if (!nr_entries) | |
1529 | return 0; | |
1530 | ||
1531 | rcu_read_lock(); | |
1532 | restart: | |
1533 | radix_tree_for_each_tagged(slot, &mapping->page_tree, | |
1534 | &iter, start, tag) { | |
1535 | struct page *page; | |
1536 | repeat: | |
1537 | page = radix_tree_deref_slot(slot); | |
1538 | if (unlikely(!page)) | |
1539 | continue; | |
1540 | if (radix_tree_exception(page)) { | |
1541 | if (radix_tree_deref_retry(page)) { | |
1542 | /* | |
1543 | * Transient condition which can only trigger | |
1544 | * when entry at index 0 moves out of or back | |
1545 | * to root: none yet gotten, safe to restart. | |
1546 | */ | |
1547 | goto restart; | |
1548 | } | |
1549 | ||
1550 | /* | |
1551 | * A shadow entry of a recently evicted page, a swap | |
1552 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1553 | * without attempting to raise page count. | |
1554 | */ | |
1555 | goto export; | |
1556 | } | |
1557 | if (!page_cache_get_speculative(page)) | |
1558 | goto repeat; | |
1559 | ||
1560 | /* Has the page moved? */ | |
1561 | if (unlikely(page != *slot)) { | |
1562 | page_cache_release(page); | |
1563 | goto repeat; | |
1564 | } | |
1565 | export: | |
1566 | indices[ret] = iter.index; | |
1567 | entries[ret] = page; | |
1568 | if (++ret == nr_entries) | |
1569 | break; | |
1570 | } | |
1571 | rcu_read_unlock(); | |
1572 | return ret; | |
1573 | } | |
1574 | EXPORT_SYMBOL(find_get_entries_tag); | |
1575 | ||
76d42bd9 WF |
1576 | /* |
1577 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1578 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1579 | * | |
1580 | * ---R__________________________________________B__________ | |
1581 | * ^ reading here ^ bad block(assume 4k) | |
1582 | * | |
1583 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1584 | * => failing the whole request => read(R) => read(R+1) => | |
1585 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1586 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1587 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1588 | * | |
1589 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1590 | */ | |
1591 | static void shrink_readahead_size_eio(struct file *filp, | |
1592 | struct file_ra_state *ra) | |
1593 | { | |
76d42bd9 | 1594 | ra->ra_pages /= 4; |
76d42bd9 WF |
1595 | } |
1596 | ||
485bb99b | 1597 | /** |
36e78914 | 1598 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1599 | * @filp: the file to read |
1600 | * @ppos: current file position | |
6e58e79d AV |
1601 | * @iter: data destination |
1602 | * @written: already copied | |
485bb99b | 1603 | * |
1da177e4 | 1604 | * This is a generic file read routine, and uses the |
485bb99b | 1605 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1606 | * |
1607 | * This is really ugly. But the goto's actually try to clarify some | |
1608 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1609 | */ |
6e58e79d AV |
1610 | static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos, |
1611 | struct iov_iter *iter, ssize_t written) | |
1da177e4 | 1612 | { |
36e78914 | 1613 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1614 | struct inode *inode = mapping->host; |
36e78914 | 1615 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1616 | pgoff_t index; |
1617 | pgoff_t last_index; | |
1618 | pgoff_t prev_index; | |
1619 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1620 | unsigned int prev_offset; |
6e58e79d | 1621 | int error = 0; |
1da177e4 | 1622 | |
1da177e4 | 1623 | index = *ppos >> PAGE_CACHE_SHIFT; |
7ff81078 FW |
1624 | prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; |
1625 | prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); | |
6e58e79d | 1626 | last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
1da177e4 LT |
1627 | offset = *ppos & ~PAGE_CACHE_MASK; |
1628 | ||
1da177e4 LT |
1629 | for (;;) { |
1630 | struct page *page; | |
57f6b96c | 1631 | pgoff_t end_index; |
a32ea1e1 | 1632 | loff_t isize; |
1da177e4 LT |
1633 | unsigned long nr, ret; |
1634 | ||
1da177e4 | 1635 | cond_resched(); |
1da177e4 LT |
1636 | find_page: |
1637 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1638 | if (!page) { |
cf914a7d | 1639 | page_cache_sync_readahead(mapping, |
7ff81078 | 1640 | ra, filp, |
3ea89ee8 FW |
1641 | index, last_index - index); |
1642 | page = find_get_page(mapping, index); | |
1643 | if (unlikely(page == NULL)) | |
1644 | goto no_cached_page; | |
1645 | } | |
1646 | if (PageReadahead(page)) { | |
cf914a7d | 1647 | page_cache_async_readahead(mapping, |
7ff81078 | 1648 | ra, filp, page, |
3ea89ee8 | 1649 | index, last_index - index); |
1da177e4 | 1650 | } |
8ab22b9a HH |
1651 | if (!PageUptodate(page)) { |
1652 | if (inode->i_blkbits == PAGE_CACHE_SHIFT || | |
1653 | !mapping->a_ops->is_partially_uptodate) | |
1654 | goto page_not_up_to_date; | |
529ae9aa | 1655 | if (!trylock_page(page)) |
8ab22b9a | 1656 | goto page_not_up_to_date; |
8d056cb9 DH |
1657 | /* Did it get truncated before we got the lock? */ |
1658 | if (!page->mapping) | |
1659 | goto page_not_up_to_date_locked; | |
8ab22b9a | 1660 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 1661 | offset, iter->count)) |
8ab22b9a HH |
1662 | goto page_not_up_to_date_locked; |
1663 | unlock_page(page); | |
1664 | } | |
1da177e4 | 1665 | page_ok: |
a32ea1e1 N |
1666 | /* |
1667 | * i_size must be checked after we know the page is Uptodate. | |
1668 | * | |
1669 | * Checking i_size after the check allows us to calculate | |
1670 | * the correct value for "nr", which means the zero-filled | |
1671 | * part of the page is not copied back to userspace (unless | |
1672 | * another truncate extends the file - this is desired though). | |
1673 | */ | |
1674 | ||
1675 | isize = i_size_read(inode); | |
1676 | end_index = (isize - 1) >> PAGE_CACHE_SHIFT; | |
1677 | if (unlikely(!isize || index > end_index)) { | |
1678 | page_cache_release(page); | |
1679 | goto out; | |
1680 | } | |
1681 | ||
1682 | /* nr is the maximum number of bytes to copy from this page */ | |
1683 | nr = PAGE_CACHE_SIZE; | |
1684 | if (index == end_index) { | |
1685 | nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; | |
1686 | if (nr <= offset) { | |
1687 | page_cache_release(page); | |
1688 | goto out; | |
1689 | } | |
1690 | } | |
1691 | nr = nr - offset; | |
1da177e4 LT |
1692 | |
1693 | /* If users can be writing to this page using arbitrary | |
1694 | * virtual addresses, take care about potential aliasing | |
1695 | * before reading the page on the kernel side. | |
1696 | */ | |
1697 | if (mapping_writably_mapped(mapping)) | |
1698 | flush_dcache_page(page); | |
1699 | ||
1700 | /* | |
ec0f1637 JK |
1701 | * When a sequential read accesses a page several times, |
1702 | * only mark it as accessed the first time. | |
1da177e4 | 1703 | */ |
ec0f1637 | 1704 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1705 | mark_page_accessed(page); |
1706 | prev_index = index; | |
1707 | ||
1708 | /* | |
1709 | * Ok, we have the page, and it's up-to-date, so | |
1710 | * now we can copy it to user space... | |
1da177e4 | 1711 | */ |
6e58e79d AV |
1712 | |
1713 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 LT |
1714 | offset += ret; |
1715 | index += offset >> PAGE_CACHE_SHIFT; | |
1716 | offset &= ~PAGE_CACHE_MASK; | |
6ce745ed | 1717 | prev_offset = offset; |
1da177e4 LT |
1718 | |
1719 | page_cache_release(page); | |
6e58e79d AV |
1720 | written += ret; |
1721 | if (!iov_iter_count(iter)) | |
1722 | goto out; | |
1723 | if (ret < nr) { | |
1724 | error = -EFAULT; | |
1725 | goto out; | |
1726 | } | |
1727 | continue; | |
1da177e4 LT |
1728 | |
1729 | page_not_up_to_date: | |
1730 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1731 | error = lock_page_killable(page); |
1732 | if (unlikely(error)) | |
1733 | goto readpage_error; | |
1da177e4 | 1734 | |
8ab22b9a | 1735 | page_not_up_to_date_locked: |
da6052f7 | 1736 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1737 | if (!page->mapping) { |
1738 | unlock_page(page); | |
1739 | page_cache_release(page); | |
1740 | continue; | |
1741 | } | |
1742 | ||
1743 | /* Did somebody else fill it already? */ | |
1744 | if (PageUptodate(page)) { | |
1745 | unlock_page(page); | |
1746 | goto page_ok; | |
1747 | } | |
1748 | ||
1749 | readpage: | |
91803b49 JM |
1750 | /* |
1751 | * A previous I/O error may have been due to temporary | |
1752 | * failures, eg. multipath errors. | |
1753 | * PG_error will be set again if readpage fails. | |
1754 | */ | |
1755 | ClearPageError(page); | |
1da177e4 LT |
1756 | /* Start the actual read. The read will unlock the page. */ |
1757 | error = mapping->a_ops->readpage(filp, page); | |
1758 | ||
994fc28c ZB |
1759 | if (unlikely(error)) { |
1760 | if (error == AOP_TRUNCATED_PAGE) { | |
1761 | page_cache_release(page); | |
6e58e79d | 1762 | error = 0; |
994fc28c ZB |
1763 | goto find_page; |
1764 | } | |
1da177e4 | 1765 | goto readpage_error; |
994fc28c | 1766 | } |
1da177e4 LT |
1767 | |
1768 | if (!PageUptodate(page)) { | |
85462323 ON |
1769 | error = lock_page_killable(page); |
1770 | if (unlikely(error)) | |
1771 | goto readpage_error; | |
1da177e4 LT |
1772 | if (!PageUptodate(page)) { |
1773 | if (page->mapping == NULL) { | |
1774 | /* | |
2ecdc82e | 1775 | * invalidate_mapping_pages got it |
1da177e4 LT |
1776 | */ |
1777 | unlock_page(page); | |
1778 | page_cache_release(page); | |
1779 | goto find_page; | |
1780 | } | |
1781 | unlock_page(page); | |
7ff81078 | 1782 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1783 | error = -EIO; |
1784 | goto readpage_error; | |
1da177e4 LT |
1785 | } |
1786 | unlock_page(page); | |
1787 | } | |
1788 | ||
1da177e4 LT |
1789 | goto page_ok; |
1790 | ||
1791 | readpage_error: | |
1792 | /* UHHUH! A synchronous read error occurred. Report it */ | |
1da177e4 LT |
1793 | page_cache_release(page); |
1794 | goto out; | |
1795 | ||
1796 | no_cached_page: | |
1797 | /* | |
1798 | * Ok, it wasn't cached, so we need to create a new | |
1799 | * page.. | |
1800 | */ | |
eb2be189 NP |
1801 | page = page_cache_alloc_cold(mapping); |
1802 | if (!page) { | |
6e58e79d | 1803 | error = -ENOMEM; |
eb2be189 | 1804 | goto out; |
1da177e4 | 1805 | } |
6afdb859 | 1806 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 1807 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 1808 | if (error) { |
eb2be189 | 1809 | page_cache_release(page); |
6e58e79d AV |
1810 | if (error == -EEXIST) { |
1811 | error = 0; | |
1da177e4 | 1812 | goto find_page; |
6e58e79d | 1813 | } |
1da177e4 LT |
1814 | goto out; |
1815 | } | |
1da177e4 LT |
1816 | goto readpage; |
1817 | } | |
1818 | ||
1819 | out: | |
7ff81078 FW |
1820 | ra->prev_pos = prev_index; |
1821 | ra->prev_pos <<= PAGE_CACHE_SHIFT; | |
1822 | ra->prev_pos |= prev_offset; | |
1da177e4 | 1823 | |
f4e6b498 | 1824 | *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; |
0c6aa263 | 1825 | file_accessed(filp); |
6e58e79d | 1826 | return written ? written : error; |
1da177e4 LT |
1827 | } |
1828 | ||
485bb99b | 1829 | /** |
6abd2322 | 1830 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 1831 | * @iocb: kernel I/O control block |
6abd2322 | 1832 | * @iter: destination for the data read |
485bb99b | 1833 | * |
6abd2322 | 1834 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
1835 | * that can use the page cache directly. |
1836 | */ | |
1837 | ssize_t | |
ed978a81 | 1838 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 1839 | { |
ed978a81 | 1840 | struct file *file = iocb->ki_filp; |
cb66a7a1 | 1841 | ssize_t retval = 0; |
543ade1f | 1842 | loff_t *ppos = &iocb->ki_pos; |
ed978a81 | 1843 | loff_t pos = *ppos; |
1da177e4 | 1844 | |
2ba48ce5 | 1845 | if (iocb->ki_flags & IOCB_DIRECT) { |
ed978a81 AV |
1846 | struct address_space *mapping = file->f_mapping; |
1847 | struct inode *inode = mapping->host; | |
1848 | size_t count = iov_iter_count(iter); | |
543ade1f | 1849 | loff_t size; |
1da177e4 | 1850 | |
1da177e4 LT |
1851 | if (!count) |
1852 | goto out; /* skip atime */ | |
1853 | size = i_size_read(inode); | |
9fe55eea | 1854 | retval = filemap_write_and_wait_range(mapping, pos, |
a6cbcd4a | 1855 | pos + count - 1); |
9fe55eea | 1856 | if (!retval) { |
ed978a81 | 1857 | struct iov_iter data = *iter; |
22c6186e | 1858 | retval = mapping->a_ops->direct_IO(iocb, &data, pos); |
9fe55eea | 1859 | } |
d8d3d94b | 1860 | |
9fe55eea SW |
1861 | if (retval > 0) { |
1862 | *ppos = pos + retval; | |
ed978a81 | 1863 | iov_iter_advance(iter, retval); |
9fe55eea | 1864 | } |
66f998f6 | 1865 | |
9fe55eea SW |
1866 | /* |
1867 | * Btrfs can have a short DIO read if we encounter | |
1868 | * compressed extents, so if there was an error, or if | |
1869 | * we've already read everything we wanted to, or if | |
1870 | * there was a short read because we hit EOF, go ahead | |
1871 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
1872 | * the rest of the read. Buffered reads will not work for |
1873 | * DAX files, so don't bother trying. | |
9fe55eea | 1874 | */ |
fbbbad4b MW |
1875 | if (retval < 0 || !iov_iter_count(iter) || *ppos >= size || |
1876 | IS_DAX(inode)) { | |
ed978a81 | 1877 | file_accessed(file); |
9fe55eea | 1878 | goto out; |
0e0bcae3 | 1879 | } |
1da177e4 LT |
1880 | } |
1881 | ||
ed978a81 | 1882 | retval = do_generic_file_read(file, ppos, iter, retval); |
1da177e4 LT |
1883 | out: |
1884 | return retval; | |
1885 | } | |
ed978a81 | 1886 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 1887 | |
1da177e4 | 1888 | #ifdef CONFIG_MMU |
485bb99b RD |
1889 | /** |
1890 | * page_cache_read - adds requested page to the page cache if not already there | |
1891 | * @file: file to read | |
1892 | * @offset: page index | |
62eb320a | 1893 | * @gfp_mask: memory allocation flags |
485bb99b | 1894 | * |
1da177e4 LT |
1895 | * This adds the requested page to the page cache if it isn't already there, |
1896 | * and schedules an I/O to read in its contents from disk. | |
1897 | */ | |
c20cd45e | 1898 | static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask) |
1da177e4 LT |
1899 | { |
1900 | struct address_space *mapping = file->f_mapping; | |
99dadfdd | 1901 | struct page *page; |
994fc28c | 1902 | int ret; |
1da177e4 | 1903 | |
994fc28c | 1904 | do { |
c20cd45e | 1905 | page = __page_cache_alloc(gfp_mask|__GFP_COLD); |
994fc28c ZB |
1906 | if (!page) |
1907 | return -ENOMEM; | |
1908 | ||
c20cd45e | 1909 | ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL); |
994fc28c ZB |
1910 | if (ret == 0) |
1911 | ret = mapping->a_ops->readpage(file, page); | |
1912 | else if (ret == -EEXIST) | |
1913 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1914 | |
1da177e4 | 1915 | page_cache_release(page); |
1da177e4 | 1916 | |
994fc28c | 1917 | } while (ret == AOP_TRUNCATED_PAGE); |
99dadfdd | 1918 | |
994fc28c | 1919 | return ret; |
1da177e4 LT |
1920 | } |
1921 | ||
1922 | #define MMAP_LOTSAMISS (100) | |
1923 | ||
ef00e08e LT |
1924 | /* |
1925 | * Synchronous readahead happens when we don't even find | |
1926 | * a page in the page cache at all. | |
1927 | */ | |
1928 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1929 | struct file_ra_state *ra, | |
1930 | struct file *file, | |
1931 | pgoff_t offset) | |
1932 | { | |
ef00e08e LT |
1933 | struct address_space *mapping = file->f_mapping; |
1934 | ||
1935 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1936 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 1937 | return; |
275b12bf WF |
1938 | if (!ra->ra_pages) |
1939 | return; | |
ef00e08e | 1940 | |
64363aad | 1941 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
1942 | page_cache_sync_readahead(mapping, ra, file, offset, |
1943 | ra->ra_pages); | |
ef00e08e LT |
1944 | return; |
1945 | } | |
1946 | ||
207d04ba AK |
1947 | /* Avoid banging the cache line if not needed */ |
1948 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1949 | ra->mmap_miss++; |
1950 | ||
1951 | /* | |
1952 | * Do we miss much more than hit in this file? If so, | |
1953 | * stop bothering with read-ahead. It will only hurt. | |
1954 | */ | |
1955 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1956 | return; | |
1957 | ||
d30a1100 WF |
1958 | /* |
1959 | * mmap read-around | |
1960 | */ | |
600e19af RG |
1961 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
1962 | ra->size = ra->ra_pages; | |
1963 | ra->async_size = ra->ra_pages / 4; | |
275b12bf | 1964 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1965 | } |
1966 | ||
1967 | /* | |
1968 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1969 | * so we want to possibly extend the readahead further.. | |
1970 | */ | |
1971 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1972 | struct file_ra_state *ra, | |
1973 | struct file *file, | |
1974 | struct page *page, | |
1975 | pgoff_t offset) | |
1976 | { | |
1977 | struct address_space *mapping = file->f_mapping; | |
1978 | ||
1979 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1980 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
1981 | return; |
1982 | if (ra->mmap_miss > 0) | |
1983 | ra->mmap_miss--; | |
1984 | if (PageReadahead(page)) | |
2fad6f5d WF |
1985 | page_cache_async_readahead(mapping, ra, file, |
1986 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1987 | } |
1988 | ||
485bb99b | 1989 | /** |
54cb8821 | 1990 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1991 | * @vma: vma in which the fault was taken |
1992 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1993 | * |
54cb8821 | 1994 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1995 | * mapped memory region to read in file data during a page fault. |
1996 | * | |
1997 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
1998 | * it in the page cache, and handles the special cases reasonably without | |
1999 | * having a lot of duplicated code. | |
9a95f3cf PC |
2000 | * |
2001 | * vma->vm_mm->mmap_sem must be held on entry. | |
2002 | * | |
2003 | * If our return value has VM_FAULT_RETRY set, it's because | |
2004 | * lock_page_or_retry() returned 0. | |
2005 | * The mmap_sem has usually been released in this case. | |
2006 | * See __lock_page_or_retry() for the exception. | |
2007 | * | |
2008 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
2009 | * has not been released. | |
2010 | * | |
2011 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
1da177e4 | 2012 | */ |
d0217ac0 | 2013 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
2014 | { |
2015 | int error; | |
54cb8821 | 2016 | struct file *file = vma->vm_file; |
1da177e4 LT |
2017 | struct address_space *mapping = file->f_mapping; |
2018 | struct file_ra_state *ra = &file->f_ra; | |
2019 | struct inode *inode = mapping->host; | |
ef00e08e | 2020 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 2021 | struct page *page; |
99e3e53f | 2022 | loff_t size; |
83c54070 | 2023 | int ret = 0; |
1da177e4 | 2024 | |
99e3e53f KS |
2025 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
2026 | if (offset >= size >> PAGE_CACHE_SHIFT) | |
5307cc1a | 2027 | return VM_FAULT_SIGBUS; |
1da177e4 | 2028 | |
1da177e4 | 2029 | /* |
49426420 | 2030 | * Do we have something in the page cache already? |
1da177e4 | 2031 | */ |
ef00e08e | 2032 | page = find_get_page(mapping, offset); |
45cac65b | 2033 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2034 | /* |
ef00e08e LT |
2035 | * We found the page, so try async readahead before |
2036 | * waiting for the lock. | |
1da177e4 | 2037 | */ |
ef00e08e | 2038 | do_async_mmap_readahead(vma, ra, file, page, offset); |
45cac65b | 2039 | } else if (!page) { |
ef00e08e LT |
2040 | /* No page in the page cache at all */ |
2041 | do_sync_mmap_readahead(vma, ra, file, offset); | |
2042 | count_vm_event(PGMAJFAULT); | |
456f998e | 2043 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
2044 | ret = VM_FAULT_MAJOR; |
2045 | retry_find: | |
b522c94d | 2046 | page = find_get_page(mapping, offset); |
1da177e4 LT |
2047 | if (!page) |
2048 | goto no_cached_page; | |
2049 | } | |
2050 | ||
d88c0922 ML |
2051 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
2052 | page_cache_release(page); | |
d065bd81 | 2053 | return ret | VM_FAULT_RETRY; |
d88c0922 | 2054 | } |
b522c94d ML |
2055 | |
2056 | /* Did it get truncated? */ | |
2057 | if (unlikely(page->mapping != mapping)) { | |
2058 | unlock_page(page); | |
2059 | put_page(page); | |
2060 | goto retry_find; | |
2061 | } | |
309381fe | 2062 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 2063 | |
1da177e4 | 2064 | /* |
d00806b1 NP |
2065 | * We have a locked page in the page cache, now we need to check |
2066 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2067 | */ |
d00806b1 | 2068 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2069 | goto page_not_uptodate; |
2070 | ||
ef00e08e LT |
2071 | /* |
2072 | * Found the page and have a reference on it. | |
2073 | * We must recheck i_size under page lock. | |
2074 | */ | |
99e3e53f KS |
2075 | size = round_up(i_size_read(inode), PAGE_CACHE_SIZE); |
2076 | if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) { | |
d00806b1 | 2077 | unlock_page(page); |
745ad48e | 2078 | page_cache_release(page); |
5307cc1a | 2079 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2080 | } |
2081 | ||
d0217ac0 | 2082 | vmf->page = page; |
83c54070 | 2083 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2084 | |
1da177e4 LT |
2085 | no_cached_page: |
2086 | /* | |
2087 | * We're only likely to ever get here if MADV_RANDOM is in | |
2088 | * effect. | |
2089 | */ | |
c20cd45e | 2090 | error = page_cache_read(file, offset, vmf->gfp_mask); |
1da177e4 LT |
2091 | |
2092 | /* | |
2093 | * The page we want has now been added to the page cache. | |
2094 | * In the unlikely event that someone removed it in the | |
2095 | * meantime, we'll just come back here and read it again. | |
2096 | */ | |
2097 | if (error >= 0) | |
2098 | goto retry_find; | |
2099 | ||
2100 | /* | |
2101 | * An error return from page_cache_read can result if the | |
2102 | * system is low on memory, or a problem occurs while trying | |
2103 | * to schedule I/O. | |
2104 | */ | |
2105 | if (error == -ENOMEM) | |
d0217ac0 NP |
2106 | return VM_FAULT_OOM; |
2107 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
2108 | |
2109 | page_not_uptodate: | |
1da177e4 LT |
2110 | /* |
2111 | * Umm, take care of errors if the page isn't up-to-date. | |
2112 | * Try to re-read it _once_. We do this synchronously, | |
2113 | * because there really aren't any performance issues here | |
2114 | * and we need to check for errors. | |
2115 | */ | |
1da177e4 | 2116 | ClearPageError(page); |
994fc28c | 2117 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2118 | if (!error) { |
2119 | wait_on_page_locked(page); | |
2120 | if (!PageUptodate(page)) | |
2121 | error = -EIO; | |
2122 | } | |
d00806b1 NP |
2123 | page_cache_release(page); |
2124 | ||
2125 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2126 | goto retry_find; |
1da177e4 | 2127 | |
d00806b1 | 2128 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 2129 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 2130 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
2131 | } |
2132 | EXPORT_SYMBOL(filemap_fault); | |
2133 | ||
f1820361 KS |
2134 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
2135 | { | |
2136 | struct radix_tree_iter iter; | |
2137 | void **slot; | |
2138 | struct file *file = vma->vm_file; | |
2139 | struct address_space *mapping = file->f_mapping; | |
2140 | loff_t size; | |
2141 | struct page *page; | |
2142 | unsigned long address = (unsigned long) vmf->virtual_address; | |
2143 | unsigned long addr; | |
2144 | pte_t *pte; | |
2145 | ||
2146 | rcu_read_lock(); | |
2147 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) { | |
2148 | if (iter.index > vmf->max_pgoff) | |
2149 | break; | |
2150 | repeat: | |
2151 | page = radix_tree_deref_slot(slot); | |
2152 | if (unlikely(!page)) | |
2153 | goto next; | |
2154 | if (radix_tree_exception(page)) { | |
2155 | if (radix_tree_deref_retry(page)) | |
2156 | break; | |
2157 | else | |
2158 | goto next; | |
2159 | } | |
2160 | ||
2161 | if (!page_cache_get_speculative(page)) | |
2162 | goto repeat; | |
2163 | ||
2164 | /* Has the page moved? */ | |
2165 | if (unlikely(page != *slot)) { | |
2166 | page_cache_release(page); | |
2167 | goto repeat; | |
2168 | } | |
2169 | ||
2170 | if (!PageUptodate(page) || | |
2171 | PageReadahead(page) || | |
2172 | PageHWPoison(page)) | |
2173 | goto skip; | |
2174 | if (!trylock_page(page)) | |
2175 | goto skip; | |
2176 | ||
2177 | if (page->mapping != mapping || !PageUptodate(page)) | |
2178 | goto unlock; | |
2179 | ||
99e3e53f KS |
2180 | size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE); |
2181 | if (page->index >= size >> PAGE_CACHE_SHIFT) | |
f1820361 KS |
2182 | goto unlock; |
2183 | ||
2184 | pte = vmf->pte + page->index - vmf->pgoff; | |
2185 | if (!pte_none(*pte)) | |
2186 | goto unlock; | |
2187 | ||
2188 | if (file->f_ra.mmap_miss > 0) | |
2189 | file->f_ra.mmap_miss--; | |
2190 | addr = address + (page->index - vmf->pgoff) * PAGE_SIZE; | |
2191 | do_set_pte(vma, addr, page, pte, false, false); | |
2192 | unlock_page(page); | |
2193 | goto next; | |
2194 | unlock: | |
2195 | unlock_page(page); | |
2196 | skip: | |
2197 | page_cache_release(page); | |
2198 | next: | |
2199 | if (iter.index == vmf->max_pgoff) | |
2200 | break; | |
2201 | } | |
2202 | rcu_read_unlock(); | |
2203 | } | |
2204 | EXPORT_SYMBOL(filemap_map_pages); | |
2205 | ||
4fcf1c62 JK |
2206 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2207 | { | |
2208 | struct page *page = vmf->page; | |
496ad9aa | 2209 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
2210 | int ret = VM_FAULT_LOCKED; |
2211 | ||
14da9200 | 2212 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
2213 | file_update_time(vma->vm_file); |
2214 | lock_page(page); | |
2215 | if (page->mapping != inode->i_mapping) { | |
2216 | unlock_page(page); | |
2217 | ret = VM_FAULT_NOPAGE; | |
2218 | goto out; | |
2219 | } | |
14da9200 JK |
2220 | /* |
2221 | * We mark the page dirty already here so that when freeze is in | |
2222 | * progress, we are guaranteed that writeback during freezing will | |
2223 | * see the dirty page and writeprotect it again. | |
2224 | */ | |
2225 | set_page_dirty(page); | |
1d1d1a76 | 2226 | wait_for_stable_page(page); |
4fcf1c62 | 2227 | out: |
14da9200 | 2228 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2229 | return ret; |
2230 | } | |
2231 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2232 | ||
f0f37e2f | 2233 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2234 | .fault = filemap_fault, |
f1820361 | 2235 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2236 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2237 | }; |
2238 | ||
2239 | /* This is used for a general mmap of a disk file */ | |
2240 | ||
2241 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2242 | { | |
2243 | struct address_space *mapping = file->f_mapping; | |
2244 | ||
2245 | if (!mapping->a_ops->readpage) | |
2246 | return -ENOEXEC; | |
2247 | file_accessed(file); | |
2248 | vma->vm_ops = &generic_file_vm_ops; | |
2249 | return 0; | |
2250 | } | |
1da177e4 LT |
2251 | |
2252 | /* | |
2253 | * This is for filesystems which do not implement ->writepage. | |
2254 | */ | |
2255 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2256 | { | |
2257 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2258 | return -EINVAL; | |
2259 | return generic_file_mmap(file, vma); | |
2260 | } | |
2261 | #else | |
2262 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2263 | { | |
2264 | return -ENOSYS; | |
2265 | } | |
2266 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2267 | { | |
2268 | return -ENOSYS; | |
2269 | } | |
2270 | #endif /* CONFIG_MMU */ | |
2271 | ||
2272 | EXPORT_SYMBOL(generic_file_mmap); | |
2273 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2274 | ||
67f9fd91 SL |
2275 | static struct page *wait_on_page_read(struct page *page) |
2276 | { | |
2277 | if (!IS_ERR(page)) { | |
2278 | wait_on_page_locked(page); | |
2279 | if (!PageUptodate(page)) { | |
2280 | page_cache_release(page); | |
2281 | page = ERR_PTR(-EIO); | |
2282 | } | |
2283 | } | |
2284 | return page; | |
2285 | } | |
2286 | ||
6fe6900e | 2287 | static struct page *__read_cache_page(struct address_space *mapping, |
57f6b96c | 2288 | pgoff_t index, |
5e5358e7 | 2289 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2290 | void *data, |
2291 | gfp_t gfp) | |
1da177e4 | 2292 | { |
eb2be189 | 2293 | struct page *page; |
1da177e4 LT |
2294 | int err; |
2295 | repeat: | |
2296 | page = find_get_page(mapping, index); | |
2297 | if (!page) { | |
0531b2aa | 2298 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
2299 | if (!page) |
2300 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2301 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 NP |
2302 | if (unlikely(err)) { |
2303 | page_cache_release(page); | |
2304 | if (err == -EEXIST) | |
2305 | goto repeat; | |
1da177e4 | 2306 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2307 | return ERR_PTR(err); |
2308 | } | |
1da177e4 LT |
2309 | err = filler(data, page); |
2310 | if (err < 0) { | |
2311 | page_cache_release(page); | |
2312 | page = ERR_PTR(err); | |
67f9fd91 SL |
2313 | } else { |
2314 | page = wait_on_page_read(page); | |
1da177e4 LT |
2315 | } |
2316 | } | |
1da177e4 LT |
2317 | return page; |
2318 | } | |
2319 | ||
0531b2aa | 2320 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2321 | pgoff_t index, |
5e5358e7 | 2322 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2323 | void *data, |
2324 | gfp_t gfp) | |
2325 | ||
1da177e4 LT |
2326 | { |
2327 | struct page *page; | |
2328 | int err; | |
2329 | ||
2330 | retry: | |
0531b2aa | 2331 | page = __read_cache_page(mapping, index, filler, data, gfp); |
1da177e4 | 2332 | if (IS_ERR(page)) |
c855ff37 | 2333 | return page; |
1da177e4 LT |
2334 | if (PageUptodate(page)) |
2335 | goto out; | |
2336 | ||
2337 | lock_page(page); | |
2338 | if (!page->mapping) { | |
2339 | unlock_page(page); | |
2340 | page_cache_release(page); | |
2341 | goto retry; | |
2342 | } | |
2343 | if (PageUptodate(page)) { | |
2344 | unlock_page(page); | |
2345 | goto out; | |
2346 | } | |
2347 | err = filler(data, page); | |
2348 | if (err < 0) { | |
2349 | page_cache_release(page); | |
c855ff37 | 2350 | return ERR_PTR(err); |
67f9fd91 SL |
2351 | } else { |
2352 | page = wait_on_page_read(page); | |
2353 | if (IS_ERR(page)) | |
2354 | return page; | |
1da177e4 | 2355 | } |
c855ff37 | 2356 | out: |
6fe6900e NP |
2357 | mark_page_accessed(page); |
2358 | return page; | |
2359 | } | |
0531b2aa LT |
2360 | |
2361 | /** | |
67f9fd91 | 2362 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2363 | * @mapping: the page's address_space |
2364 | * @index: the page index | |
2365 | * @filler: function to perform the read | |
5e5358e7 | 2366 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2367 | * |
0531b2aa | 2368 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2369 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2370 | * |
2371 | * If the page does not get brought uptodate, return -EIO. | |
2372 | */ | |
67f9fd91 | 2373 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2374 | pgoff_t index, |
5e5358e7 | 2375 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2376 | void *data) |
2377 | { | |
2378 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2379 | } | |
67f9fd91 | 2380 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2381 | |
2382 | /** | |
2383 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2384 | * @mapping: the page's address_space | |
2385 | * @index: the page index | |
2386 | * @gfp: the page allocator flags to use if allocating | |
2387 | * | |
2388 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2389 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2390 | * |
2391 | * If the page does not get brought uptodate, return -EIO. | |
2392 | */ | |
2393 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2394 | pgoff_t index, | |
2395 | gfp_t gfp) | |
2396 | { | |
2397 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2398 | ||
67f9fd91 | 2399 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2400 | } |
2401 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2402 | ||
1da177e4 LT |
2403 | /* |
2404 | * Performs necessary checks before doing a write | |
2405 | * | |
485bb99b | 2406 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2407 | * Returns appropriate error code that caller should return or |
2408 | * zero in case that write should be allowed. | |
2409 | */ | |
3309dd04 | 2410 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2411 | { |
3309dd04 | 2412 | struct file *file = iocb->ki_filp; |
1da177e4 | 2413 | struct inode *inode = file->f_mapping->host; |
59e99e5b | 2414 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
3309dd04 | 2415 | loff_t pos; |
1da177e4 | 2416 | |
3309dd04 AV |
2417 | if (!iov_iter_count(from)) |
2418 | return 0; | |
1da177e4 | 2419 | |
0fa6b005 | 2420 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2421 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2422 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2423 | |
3309dd04 | 2424 | pos = iocb->ki_pos; |
1da177e4 | 2425 | |
0fa6b005 | 2426 | if (limit != RLIM_INFINITY) { |
3309dd04 | 2427 | if (iocb->ki_pos >= limit) { |
0fa6b005 AV |
2428 | send_sig(SIGXFSZ, current, 0); |
2429 | return -EFBIG; | |
1da177e4 | 2430 | } |
3309dd04 | 2431 | iov_iter_truncate(from, limit - (unsigned long)pos); |
1da177e4 LT |
2432 | } |
2433 | ||
2434 | /* | |
2435 | * LFS rule | |
2436 | */ | |
3309dd04 | 2437 | if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS && |
1da177e4 | 2438 | !(file->f_flags & O_LARGEFILE))) { |
3309dd04 | 2439 | if (pos >= MAX_NON_LFS) |
1da177e4 | 2440 | return -EFBIG; |
3309dd04 | 2441 | iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos); |
1da177e4 LT |
2442 | } |
2443 | ||
2444 | /* | |
2445 | * Are we about to exceed the fs block limit ? | |
2446 | * | |
2447 | * If we have written data it becomes a short write. If we have | |
2448 | * exceeded without writing data we send a signal and return EFBIG. | |
2449 | * Linus frestrict idea will clean these up nicely.. | |
2450 | */ | |
3309dd04 AV |
2451 | if (unlikely(pos >= inode->i_sb->s_maxbytes)) |
2452 | return -EFBIG; | |
1da177e4 | 2453 | |
3309dd04 AV |
2454 | iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos); |
2455 | return iov_iter_count(from); | |
1da177e4 LT |
2456 | } |
2457 | EXPORT_SYMBOL(generic_write_checks); | |
2458 | ||
afddba49 NP |
2459 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2460 | loff_t pos, unsigned len, unsigned flags, | |
2461 | struct page **pagep, void **fsdata) | |
2462 | { | |
2463 | const struct address_space_operations *aops = mapping->a_ops; | |
2464 | ||
4e02ed4b | 2465 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2466 | pagep, fsdata); |
afddba49 NP |
2467 | } |
2468 | EXPORT_SYMBOL(pagecache_write_begin); | |
2469 | ||
2470 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2471 | loff_t pos, unsigned len, unsigned copied, | |
2472 | struct page *page, void *fsdata) | |
2473 | { | |
2474 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2475 | |
4e02ed4b | 2476 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2477 | } |
2478 | EXPORT_SYMBOL(pagecache_write_end); | |
2479 | ||
1da177e4 | 2480 | ssize_t |
0c949334 | 2481 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos) |
1da177e4 LT |
2482 | { |
2483 | struct file *file = iocb->ki_filp; | |
2484 | struct address_space *mapping = file->f_mapping; | |
2485 | struct inode *inode = mapping->host; | |
2486 | ssize_t written; | |
a969e903 CH |
2487 | size_t write_len; |
2488 | pgoff_t end; | |
26978b8b | 2489 | struct iov_iter data; |
1da177e4 | 2490 | |
0c949334 | 2491 | write_len = iov_iter_count(from); |
a969e903 | 2492 | end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; |
a969e903 | 2493 | |
48b47c56 | 2494 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2495 | if (written) |
2496 | goto out; | |
2497 | ||
2498 | /* | |
2499 | * After a write we want buffered reads to be sure to go to disk to get | |
2500 | * the new data. We invalidate clean cached page from the region we're | |
2501 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2502 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2503 | */ |
2504 | if (mapping->nrpages) { | |
2505 | written = invalidate_inode_pages2_range(mapping, | |
2506 | pos >> PAGE_CACHE_SHIFT, end); | |
6ccfa806 HH |
2507 | /* |
2508 | * If a page can not be invalidated, return 0 to fall back | |
2509 | * to buffered write. | |
2510 | */ | |
2511 | if (written) { | |
2512 | if (written == -EBUSY) | |
2513 | return 0; | |
a969e903 | 2514 | goto out; |
6ccfa806 | 2515 | } |
a969e903 CH |
2516 | } |
2517 | ||
26978b8b | 2518 | data = *from; |
22c6186e | 2519 | written = mapping->a_ops->direct_IO(iocb, &data, pos); |
a969e903 CH |
2520 | |
2521 | /* | |
2522 | * Finally, try again to invalidate clean pages which might have been | |
2523 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2524 | * if the source of the write was an mmap'ed region of the file | |
2525 | * we're writing. Either one is a pretty crazy thing to do, | |
2526 | * so we don't support it 100%. If this invalidation | |
2527 | * fails, tough, the write still worked... | |
2528 | */ | |
2529 | if (mapping->nrpages) { | |
2530 | invalidate_inode_pages2_range(mapping, | |
2531 | pos >> PAGE_CACHE_SHIFT, end); | |
2532 | } | |
2533 | ||
1da177e4 | 2534 | if (written > 0) { |
0116651c | 2535 | pos += written; |
f8579f86 | 2536 | iov_iter_advance(from, written); |
0116651c NK |
2537 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
2538 | i_size_write(inode, pos); | |
1da177e4 LT |
2539 | mark_inode_dirty(inode); |
2540 | } | |
5cb6c6c7 | 2541 | iocb->ki_pos = pos; |
1da177e4 | 2542 | } |
a969e903 | 2543 | out: |
1da177e4 LT |
2544 | return written; |
2545 | } | |
2546 | EXPORT_SYMBOL(generic_file_direct_write); | |
2547 | ||
eb2be189 NP |
2548 | /* |
2549 | * Find or create a page at the given pagecache position. Return the locked | |
2550 | * page. This function is specifically for buffered writes. | |
2551 | */ | |
54566b2c NP |
2552 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2553 | pgoff_t index, unsigned flags) | |
eb2be189 | 2554 | { |
eb2be189 | 2555 | struct page *page; |
2457aec6 | 2556 | int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT; |
0faa70cb | 2557 | |
54566b2c | 2558 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
2559 | fgp_flags |= FGP_NOFS; |
2560 | ||
2561 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 2562 | mapping_gfp_mask(mapping)); |
c585a267 | 2563 | if (page) |
2457aec6 | 2564 | wait_for_stable_page(page); |
eb2be189 | 2565 | |
eb2be189 NP |
2566 | return page; |
2567 | } | |
54566b2c | 2568 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2569 | |
3b93f911 | 2570 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
2571 | struct iov_iter *i, loff_t pos) |
2572 | { | |
2573 | struct address_space *mapping = file->f_mapping; | |
2574 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2575 | long status = 0; | |
2576 | ssize_t written = 0; | |
674b892e NP |
2577 | unsigned int flags = 0; |
2578 | ||
2579 | /* | |
2580 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2581 | */ | |
777eda2c | 2582 | if (!iter_is_iovec(i)) |
674b892e | 2583 | flags |= AOP_FLAG_UNINTERRUPTIBLE; |
afddba49 NP |
2584 | |
2585 | do { | |
2586 | struct page *page; | |
afddba49 NP |
2587 | unsigned long offset; /* Offset into pagecache page */ |
2588 | unsigned long bytes; /* Bytes to write to page */ | |
2589 | size_t copied; /* Bytes copied from user */ | |
2590 | void *fsdata; | |
2591 | ||
2592 | offset = (pos & (PAGE_CACHE_SIZE - 1)); | |
afddba49 NP |
2593 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, |
2594 | iov_iter_count(i)); | |
2595 | ||
2596 | again: | |
00a3d660 LT |
2597 | /* |
2598 | * Bring in the user page that we will copy from _first_. | |
2599 | * Otherwise there's a nasty deadlock on copying from the | |
2600 | * same page as we're writing to, without it being marked | |
2601 | * up-to-date. | |
2602 | * | |
2603 | * Not only is this an optimisation, but it is also required | |
2604 | * to check that the address is actually valid, when atomic | |
2605 | * usercopies are used, below. | |
2606 | */ | |
2607 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2608 | status = -EFAULT; | |
2609 | break; | |
2610 | } | |
2611 | ||
296291cd JK |
2612 | if (fatal_signal_pending(current)) { |
2613 | status = -EINTR; | |
2614 | break; | |
2615 | } | |
2616 | ||
674b892e | 2617 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 2618 | &page, &fsdata); |
2457aec6 | 2619 | if (unlikely(status < 0)) |
afddba49 NP |
2620 | break; |
2621 | ||
931e80e4 | 2622 | if (mapping_writably_mapped(mapping)) |
2623 | flush_dcache_page(page); | |
00a3d660 | 2624 | |
afddba49 | 2625 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
2626 | flush_dcache_page(page); |
2627 | ||
2628 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
2629 | page, fsdata); | |
2630 | if (unlikely(status < 0)) | |
2631 | break; | |
2632 | copied = status; | |
2633 | ||
2634 | cond_resched(); | |
2635 | ||
124d3b70 | 2636 | iov_iter_advance(i, copied); |
afddba49 NP |
2637 | if (unlikely(copied == 0)) { |
2638 | /* | |
2639 | * If we were unable to copy any data at all, we must | |
2640 | * fall back to a single segment length write. | |
2641 | * | |
2642 | * If we didn't fallback here, we could livelock | |
2643 | * because not all segments in the iov can be copied at | |
2644 | * once without a pagefault. | |
2645 | */ | |
2646 | bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, | |
2647 | iov_iter_single_seg_count(i)); | |
2648 | goto again; | |
2649 | } | |
afddba49 NP |
2650 | pos += copied; |
2651 | written += copied; | |
2652 | ||
2653 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
2654 | } while (iov_iter_count(i)); |
2655 | ||
2656 | return written ? written : status; | |
2657 | } | |
3b93f911 | 2658 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 2659 | |
e4dd9de3 | 2660 | /** |
8174202b | 2661 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 2662 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 2663 | * @from: iov_iter with data to write |
e4dd9de3 JK |
2664 | * |
2665 | * This function does all the work needed for actually writing data to a | |
2666 | * file. It does all basic checks, removes SUID from the file, updates | |
2667 | * modification times and calls proper subroutines depending on whether we | |
2668 | * do direct IO or a standard buffered write. | |
2669 | * | |
2670 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2671 | * object which does not need locking at all. | |
2672 | * | |
2673 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2674 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2675 | * avoid syncing under i_mutex. | |
2676 | */ | |
8174202b | 2677 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2678 | { |
2679 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2680 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 2681 | struct inode *inode = mapping->host; |
3b93f911 | 2682 | ssize_t written = 0; |
1da177e4 | 2683 | ssize_t err; |
3b93f911 | 2684 | ssize_t status; |
1da177e4 | 2685 | |
1da177e4 | 2686 | /* We can write back this queue in page reclaim */ |
de1414a6 | 2687 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 2688 | err = file_remove_privs(file); |
1da177e4 LT |
2689 | if (err) |
2690 | goto out; | |
2691 | ||
c3b2da31 JB |
2692 | err = file_update_time(file); |
2693 | if (err) | |
2694 | goto out; | |
1da177e4 | 2695 | |
2ba48ce5 | 2696 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 2697 | loff_t pos, endbyte; |
fb5527e6 | 2698 | |
0b8def9d | 2699 | written = generic_file_direct_write(iocb, from, iocb->ki_pos); |
1da177e4 | 2700 | /* |
fbbbad4b MW |
2701 | * If the write stopped short of completing, fall back to |
2702 | * buffered writes. Some filesystems do this for writes to | |
2703 | * holes, for example. For DAX files, a buffered write will | |
2704 | * not succeed (even if it did, DAX does not handle dirty | |
2705 | * page-cache pages correctly). | |
1da177e4 | 2706 | */ |
0b8def9d | 2707 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
2708 | goto out; |
2709 | ||
0b8def9d | 2710 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 2711 | /* |
3b93f911 | 2712 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
2713 | * then we want to return the number of bytes which were |
2714 | * direct-written, or the error code if that was zero. Note | |
2715 | * that this differs from normal direct-io semantics, which | |
2716 | * will return -EFOO even if some bytes were written. | |
2717 | */ | |
60bb4529 | 2718 | if (unlikely(status < 0)) { |
3b93f911 | 2719 | err = status; |
fb5527e6 JM |
2720 | goto out; |
2721 | } | |
fb5527e6 JM |
2722 | /* |
2723 | * We need to ensure that the page cache pages are written to | |
2724 | * disk and invalidated to preserve the expected O_DIRECT | |
2725 | * semantics. | |
2726 | */ | |
3b93f911 | 2727 | endbyte = pos + status - 1; |
0b8def9d | 2728 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 2729 | if (err == 0) { |
0b8def9d | 2730 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 2731 | written += status; |
fb5527e6 JM |
2732 | invalidate_mapping_pages(mapping, |
2733 | pos >> PAGE_CACHE_SHIFT, | |
2734 | endbyte >> PAGE_CACHE_SHIFT); | |
2735 | } else { | |
2736 | /* | |
2737 | * We don't know how much we wrote, so just return | |
2738 | * the number of bytes which were direct-written | |
2739 | */ | |
2740 | } | |
2741 | } else { | |
0b8def9d AV |
2742 | written = generic_perform_write(file, from, iocb->ki_pos); |
2743 | if (likely(written > 0)) | |
2744 | iocb->ki_pos += written; | |
fb5527e6 | 2745 | } |
1da177e4 LT |
2746 | out: |
2747 | current->backing_dev_info = NULL; | |
2748 | return written ? written : err; | |
2749 | } | |
8174202b | 2750 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 2751 | |
e4dd9de3 | 2752 | /** |
8174202b | 2753 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 2754 | * @iocb: IO state structure |
8174202b | 2755 | * @from: iov_iter with data to write |
e4dd9de3 | 2756 | * |
8174202b | 2757 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
2758 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
2759 | * and acquires i_mutex as needed. | |
2760 | */ | |
8174202b | 2761 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2762 | { |
2763 | struct file *file = iocb->ki_filp; | |
148f948b | 2764 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2765 | ssize_t ret; |
1da177e4 | 2766 | |
5955102c | 2767 | inode_lock(inode); |
3309dd04 AV |
2768 | ret = generic_write_checks(iocb, from); |
2769 | if (ret > 0) | |
5f380c7f | 2770 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 2771 | inode_unlock(inode); |
1da177e4 | 2772 | |
02afc27f | 2773 | if (ret > 0) { |
1da177e4 LT |
2774 | ssize_t err; |
2775 | ||
d311d79d AV |
2776 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
2777 | if (err < 0) | |
1da177e4 LT |
2778 | ret = err; |
2779 | } | |
2780 | return ret; | |
2781 | } | |
8174202b | 2782 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 2783 | |
cf9a2ae8 DH |
2784 | /** |
2785 | * try_to_release_page() - release old fs-specific metadata on a page | |
2786 | * | |
2787 | * @page: the page which the kernel is trying to free | |
2788 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2789 | * | |
2790 | * The address_space is to try to release any data against the page | |
2791 | * (presumably at page->private). If the release was successful, return `1'. | |
2792 | * Otherwise return zero. | |
2793 | * | |
266cf658 DH |
2794 | * This may also be called if PG_fscache is set on a page, indicating that the |
2795 | * page is known to the local caching routines. | |
2796 | * | |
cf9a2ae8 | 2797 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 2798 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 2799 | * |
cf9a2ae8 DH |
2800 | */ |
2801 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2802 | { | |
2803 | struct address_space * const mapping = page->mapping; | |
2804 | ||
2805 | BUG_ON(!PageLocked(page)); | |
2806 | if (PageWriteback(page)) | |
2807 | return 0; | |
2808 | ||
2809 | if (mapping && mapping->a_ops->releasepage) | |
2810 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2811 | return try_to_free_buffers(page); | |
2812 | } | |
2813 | ||
2814 | EXPORT_SYMBOL(try_to_release_page); |