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) |
81f8c3a4 | 104 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
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 |
fdf1cdb9 | 179 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 180 | */ |
62cccb8c | 181 | void __delete_from_page_cache(struct page *page, void *shadow) |
1da177e4 LT |
182 | { |
183 | struct address_space *mapping = page->mapping; | |
184 | ||
fe0bfaaf | 185 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
186 | /* |
187 | * if we're uptodate, flush out into the cleancache, otherwise | |
188 | * invalidate any existing cleancache entries. We can't leave | |
189 | * stale data around in the cleancache once our page is gone | |
190 | */ | |
191 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
192 | cleancache_put_page(page); | |
193 | else | |
3167760f | 194 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 195 | |
06b241f3 HD |
196 | VM_BUG_ON_PAGE(page_mapped(page), page); |
197 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
198 | int mapcount; | |
199 | ||
200 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
201 | current->comm, page_to_pfn(page)); | |
202 | dump_page(page, "still mapped when deleted"); | |
203 | dump_stack(); | |
204 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
205 | ||
206 | mapcount = page_mapcount(page); | |
207 | if (mapping_exiting(mapping) && | |
208 | page_count(page) >= mapcount + 2) { | |
209 | /* | |
210 | * All vmas have already been torn down, so it's | |
211 | * a good bet that actually the page is unmapped, | |
212 | * and we'd prefer not to leak it: if we're wrong, | |
213 | * some other bad page check should catch it later. | |
214 | */ | |
215 | page_mapcount_reset(page); | |
6d061f9f | 216 | page_ref_sub(page, mapcount); |
06b241f3 HD |
217 | } |
218 | } | |
219 | ||
91b0abe3 JW |
220 | page_cache_tree_delete(mapping, page, shadow); |
221 | ||
1da177e4 | 222 | page->mapping = NULL; |
b85e0eff | 223 | /* Leave page->index set: truncation lookup relies upon it */ |
91b0abe3 | 224 | |
4165b9b4 MH |
225 | /* hugetlb pages do not participate in page cache accounting. */ |
226 | if (!PageHuge(page)) | |
227 | __dec_zone_page_state(page, NR_FILE_PAGES); | |
4b02108a KM |
228 | if (PageSwapBacked(page)) |
229 | __dec_zone_page_state(page, NR_SHMEM); | |
3a692790 LT |
230 | |
231 | /* | |
b9ea2515 KK |
232 | * At this point page must be either written or cleaned by truncate. |
233 | * Dirty page here signals a bug and loss of unwritten data. | |
3a692790 | 234 | * |
b9ea2515 KK |
235 | * This fixes dirty accounting after removing the page entirely but |
236 | * leaves PageDirty set: it has no effect for truncated page and | |
237 | * anyway will be cleared before returning page into buddy allocator. | |
3a692790 | 238 | */ |
b9ea2515 | 239 | if (WARN_ON_ONCE(PageDirty(page))) |
62cccb8c | 240 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); |
1da177e4 LT |
241 | } |
242 | ||
702cfbf9 MK |
243 | /** |
244 | * delete_from_page_cache - delete page from page cache | |
245 | * @page: the page which the kernel is trying to remove from page cache | |
246 | * | |
247 | * This must be called only on pages that have been verified to be in the page | |
248 | * cache and locked. It will never put the page into the free list, the caller | |
249 | * has a reference on the page. | |
250 | */ | |
251 | void delete_from_page_cache(struct page *page) | |
1da177e4 LT |
252 | { |
253 | struct address_space *mapping = page->mapping; | |
c4843a75 GT |
254 | unsigned long flags; |
255 | ||
6072d13c | 256 | void (*freepage)(struct page *); |
1da177e4 | 257 | |
cd7619d6 | 258 | BUG_ON(!PageLocked(page)); |
1da177e4 | 259 | |
6072d13c | 260 | freepage = mapping->a_ops->freepage; |
c4843a75 | 261 | |
c4843a75 | 262 | spin_lock_irqsave(&mapping->tree_lock, flags); |
62cccb8c | 263 | __delete_from_page_cache(page, NULL); |
c4843a75 | 264 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6072d13c LT |
265 | |
266 | if (freepage) | |
267 | freepage(page); | |
09cbfeaf | 268 | put_page(page); |
97cecb5a MK |
269 | } |
270 | EXPORT_SYMBOL(delete_from_page_cache); | |
271 | ||
865ffef3 DM |
272 | static int filemap_check_errors(struct address_space *mapping) |
273 | { | |
274 | int ret = 0; | |
275 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
276 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
277 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 278 | ret = -ENOSPC; |
7fcbbaf1 JA |
279 | if (test_bit(AS_EIO, &mapping->flags) && |
280 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
281 | ret = -EIO; |
282 | return ret; | |
283 | } | |
284 | ||
1da177e4 | 285 | /** |
485bb99b | 286 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
287 | * @mapping: address space structure to write |
288 | * @start: offset in bytes where the range starts | |
469eb4d0 | 289 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 290 | * @sync_mode: enable synchronous operation |
1da177e4 | 291 | * |
485bb99b RD |
292 | * Start writeback against all of a mapping's dirty pages that lie |
293 | * within the byte offsets <start, end> inclusive. | |
294 | * | |
1da177e4 | 295 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 296 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
297 | * these two operations is that if a dirty page/buffer is encountered, it must |
298 | * be waited upon, and not just skipped over. | |
299 | */ | |
ebcf28e1 AM |
300 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
301 | loff_t end, int sync_mode) | |
1da177e4 LT |
302 | { |
303 | int ret; | |
304 | struct writeback_control wbc = { | |
305 | .sync_mode = sync_mode, | |
05fe478d | 306 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
307 | .range_start = start, |
308 | .range_end = end, | |
1da177e4 LT |
309 | }; |
310 | ||
311 | if (!mapping_cap_writeback_dirty(mapping)) | |
312 | return 0; | |
313 | ||
b16b1deb | 314 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 315 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 316 | wbc_detach_inode(&wbc); |
1da177e4 LT |
317 | return ret; |
318 | } | |
319 | ||
320 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
321 | int sync_mode) | |
322 | { | |
111ebb6e | 323 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
324 | } |
325 | ||
326 | int filemap_fdatawrite(struct address_space *mapping) | |
327 | { | |
328 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
329 | } | |
330 | EXPORT_SYMBOL(filemap_fdatawrite); | |
331 | ||
f4c0a0fd | 332 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 333 | loff_t end) |
1da177e4 LT |
334 | { |
335 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
336 | } | |
f4c0a0fd | 337 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 338 | |
485bb99b RD |
339 | /** |
340 | * filemap_flush - mostly a non-blocking flush | |
341 | * @mapping: target address_space | |
342 | * | |
1da177e4 LT |
343 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
344 | * purposes - I/O may not be started against all dirty pages. | |
345 | */ | |
346 | int filemap_flush(struct address_space *mapping) | |
347 | { | |
348 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
349 | } | |
350 | EXPORT_SYMBOL(filemap_flush); | |
351 | ||
aa750fd7 JN |
352 | static int __filemap_fdatawait_range(struct address_space *mapping, |
353 | loff_t start_byte, loff_t end_byte) | |
1da177e4 | 354 | { |
09cbfeaf KS |
355 | pgoff_t index = start_byte >> PAGE_SHIFT; |
356 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
357 | struct pagevec pvec; |
358 | int nr_pages; | |
aa750fd7 | 359 | int ret = 0; |
1da177e4 | 360 | |
94004ed7 | 361 | if (end_byte < start_byte) |
865ffef3 | 362 | goto out; |
1da177e4 LT |
363 | |
364 | pagevec_init(&pvec, 0); | |
1da177e4 LT |
365 | while ((index <= end) && |
366 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
367 | PAGECACHE_TAG_WRITEBACK, | |
368 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { | |
369 | unsigned i; | |
370 | ||
371 | for (i = 0; i < nr_pages; i++) { | |
372 | struct page *page = pvec.pages[i]; | |
373 | ||
374 | /* until radix tree lookup accepts end_index */ | |
375 | if (page->index > end) | |
376 | continue; | |
377 | ||
378 | wait_on_page_writeback(page); | |
212260aa | 379 | if (TestClearPageError(page)) |
1da177e4 LT |
380 | ret = -EIO; |
381 | } | |
382 | pagevec_release(&pvec); | |
383 | cond_resched(); | |
384 | } | |
865ffef3 | 385 | out: |
aa750fd7 JN |
386 | return ret; |
387 | } | |
388 | ||
389 | /** | |
390 | * filemap_fdatawait_range - wait for writeback to complete | |
391 | * @mapping: address space structure to wait for | |
392 | * @start_byte: offset in bytes where the range starts | |
393 | * @end_byte: offset in bytes where the range ends (inclusive) | |
394 | * | |
395 | * Walk the list of under-writeback pages of the given address space | |
396 | * in the given range and wait for all of them. Check error status of | |
397 | * the address space and return it. | |
398 | * | |
399 | * Since the error status of the address space is cleared by this function, | |
400 | * callers are responsible for checking the return value and handling and/or | |
401 | * reporting the error. | |
402 | */ | |
403 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
404 | loff_t end_byte) | |
405 | { | |
406 | int ret, ret2; | |
407 | ||
408 | ret = __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
865ffef3 DM |
409 | ret2 = filemap_check_errors(mapping); |
410 | if (!ret) | |
411 | ret = ret2; | |
1da177e4 LT |
412 | |
413 | return ret; | |
414 | } | |
d3bccb6f JK |
415 | EXPORT_SYMBOL(filemap_fdatawait_range); |
416 | ||
aa750fd7 JN |
417 | /** |
418 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
419 | * @mapping: address space structure to wait for | |
420 | * | |
421 | * Walk the list of under-writeback pages of the given address space | |
422 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
423 | * does not clear error status of the address space. | |
424 | * | |
425 | * Use this function if callers don't handle errors themselves. Expected | |
426 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
427 | * fsfreeze(8) | |
428 | */ | |
429 | void filemap_fdatawait_keep_errors(struct address_space *mapping) | |
430 | { | |
431 | loff_t i_size = i_size_read(mapping->host); | |
432 | ||
433 | if (i_size == 0) | |
434 | return; | |
435 | ||
436 | __filemap_fdatawait_range(mapping, 0, i_size - 1); | |
437 | } | |
438 | ||
1da177e4 | 439 | /** |
485bb99b | 440 | * filemap_fdatawait - wait for all under-writeback pages to complete |
1da177e4 | 441 | * @mapping: address space structure to wait for |
485bb99b RD |
442 | * |
443 | * Walk the list of under-writeback pages of the given address space | |
aa750fd7 JN |
444 | * and wait for all of them. Check error status of the address space |
445 | * and return it. | |
446 | * | |
447 | * Since the error status of the address space is cleared by this function, | |
448 | * callers are responsible for checking the return value and handling and/or | |
449 | * reporting the error. | |
1da177e4 LT |
450 | */ |
451 | int filemap_fdatawait(struct address_space *mapping) | |
452 | { | |
453 | loff_t i_size = i_size_read(mapping->host); | |
454 | ||
455 | if (i_size == 0) | |
456 | return 0; | |
457 | ||
94004ed7 | 458 | return filemap_fdatawait_range(mapping, 0, i_size - 1); |
1da177e4 LT |
459 | } |
460 | EXPORT_SYMBOL(filemap_fdatawait); | |
461 | ||
462 | int filemap_write_and_wait(struct address_space *mapping) | |
463 | { | |
28fd1298 | 464 | int err = 0; |
1da177e4 | 465 | |
7f6d5b52 RZ |
466 | if ((!dax_mapping(mapping) && mapping->nrpages) || |
467 | (dax_mapping(mapping) && mapping->nrexceptional)) { | |
28fd1298 OH |
468 | err = filemap_fdatawrite(mapping); |
469 | /* | |
470 | * Even if the above returned error, the pages may be | |
471 | * written partially (e.g. -ENOSPC), so we wait for it. | |
472 | * But the -EIO is special case, it may indicate the worst | |
473 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
474 | */ | |
475 | if (err != -EIO) { | |
476 | int err2 = filemap_fdatawait(mapping); | |
477 | if (!err) | |
478 | err = err2; | |
479 | } | |
865ffef3 DM |
480 | } else { |
481 | err = filemap_check_errors(mapping); | |
1da177e4 | 482 | } |
28fd1298 | 483 | return err; |
1da177e4 | 484 | } |
28fd1298 | 485 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 486 | |
485bb99b RD |
487 | /** |
488 | * filemap_write_and_wait_range - write out & wait on a file range | |
489 | * @mapping: the address_space for the pages | |
490 | * @lstart: offset in bytes where the range starts | |
491 | * @lend: offset in bytes where the range ends (inclusive) | |
492 | * | |
469eb4d0 AM |
493 | * Write out and wait upon file offsets lstart->lend, inclusive. |
494 | * | |
495 | * Note that `lend' is inclusive (describes the last byte to be written) so | |
496 | * that this function can be used to write to the very end-of-file (end = -1). | |
497 | */ | |
1da177e4 LT |
498 | int filemap_write_and_wait_range(struct address_space *mapping, |
499 | loff_t lstart, loff_t lend) | |
500 | { | |
28fd1298 | 501 | int err = 0; |
1da177e4 | 502 | |
7f6d5b52 RZ |
503 | if ((!dax_mapping(mapping) && mapping->nrpages) || |
504 | (dax_mapping(mapping) && mapping->nrexceptional)) { | |
28fd1298 OH |
505 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
506 | WB_SYNC_ALL); | |
507 | /* See comment of filemap_write_and_wait() */ | |
508 | if (err != -EIO) { | |
94004ed7 CH |
509 | int err2 = filemap_fdatawait_range(mapping, |
510 | lstart, lend); | |
28fd1298 OH |
511 | if (!err) |
512 | err = err2; | |
513 | } | |
865ffef3 DM |
514 | } else { |
515 | err = filemap_check_errors(mapping); | |
1da177e4 | 516 | } |
28fd1298 | 517 | return err; |
1da177e4 | 518 | } |
f6995585 | 519 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 520 | |
ef6a3c63 MS |
521 | /** |
522 | * replace_page_cache_page - replace a pagecache page with a new one | |
523 | * @old: page to be replaced | |
524 | * @new: page to replace with | |
525 | * @gfp_mask: allocation mode | |
526 | * | |
527 | * This function replaces a page in the pagecache with a new one. On | |
528 | * success it acquires the pagecache reference for the new page and | |
529 | * drops it for the old page. Both the old and new pages must be | |
530 | * locked. This function does not add the new page to the LRU, the | |
531 | * caller must do that. | |
532 | * | |
533 | * The remove + add is atomic. The only way this function can fail is | |
534 | * memory allocation failure. | |
535 | */ | |
536 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
537 | { | |
538 | int error; | |
ef6a3c63 | 539 | |
309381fe SL |
540 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
541 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
542 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 543 | |
ef6a3c63 MS |
544 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
545 | if (!error) { | |
546 | struct address_space *mapping = old->mapping; | |
547 | void (*freepage)(struct page *); | |
c4843a75 | 548 | unsigned long flags; |
ef6a3c63 MS |
549 | |
550 | pgoff_t offset = old->index; | |
551 | freepage = mapping->a_ops->freepage; | |
552 | ||
09cbfeaf | 553 | get_page(new); |
ef6a3c63 MS |
554 | new->mapping = mapping; |
555 | new->index = offset; | |
556 | ||
c4843a75 | 557 | spin_lock_irqsave(&mapping->tree_lock, flags); |
62cccb8c | 558 | __delete_from_page_cache(old, NULL); |
ef6a3c63 MS |
559 | error = radix_tree_insert(&mapping->page_tree, offset, new); |
560 | BUG_ON(error); | |
561 | mapping->nrpages++; | |
4165b9b4 MH |
562 | |
563 | /* | |
564 | * hugetlb pages do not participate in page cache accounting. | |
565 | */ | |
566 | if (!PageHuge(new)) | |
567 | __inc_zone_page_state(new, NR_FILE_PAGES); | |
ef6a3c63 MS |
568 | if (PageSwapBacked(new)) |
569 | __inc_zone_page_state(new, NR_SHMEM); | |
c4843a75 | 570 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6a93ca8f | 571 | mem_cgroup_migrate(old, new); |
ef6a3c63 MS |
572 | radix_tree_preload_end(); |
573 | if (freepage) | |
574 | freepage(old); | |
09cbfeaf | 575 | put_page(old); |
ef6a3c63 MS |
576 | } |
577 | ||
578 | return error; | |
579 | } | |
580 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
581 | ||
0cd6144a | 582 | static int page_cache_tree_insert(struct address_space *mapping, |
a528910e | 583 | struct page *page, void **shadowp) |
0cd6144a | 584 | { |
449dd698 | 585 | struct radix_tree_node *node; |
0cd6144a JW |
586 | void **slot; |
587 | int error; | |
588 | ||
e6145236 | 589 | error = __radix_tree_create(&mapping->page_tree, page->index, 0, |
449dd698 JW |
590 | &node, &slot); |
591 | if (error) | |
592 | return error; | |
593 | if (*slot) { | |
0cd6144a JW |
594 | void *p; |
595 | ||
596 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
597 | if (!radix_tree_exceptional_entry(p)) | |
598 | return -EEXIST; | |
f9fe48be RZ |
599 | |
600 | if (WARN_ON(dax_mapping(mapping))) | |
601 | return -EINVAL; | |
602 | ||
a528910e JW |
603 | if (shadowp) |
604 | *shadowp = p; | |
f9fe48be | 605 | mapping->nrexceptional--; |
449dd698 JW |
606 | if (node) |
607 | workingset_node_shadows_dec(node); | |
0cd6144a | 608 | } |
449dd698 JW |
609 | radix_tree_replace_slot(slot, page); |
610 | mapping->nrpages++; | |
611 | if (node) { | |
612 | workingset_node_pages_inc(node); | |
613 | /* | |
614 | * Don't track node that contains actual pages. | |
615 | * | |
616 | * Avoid acquiring the list_lru lock if already | |
617 | * untracked. The list_empty() test is safe as | |
618 | * node->private_list is protected by | |
619 | * mapping->tree_lock. | |
620 | */ | |
621 | if (!list_empty(&node->private_list)) | |
622 | list_lru_del(&workingset_shadow_nodes, | |
623 | &node->private_list); | |
624 | } | |
625 | return 0; | |
0cd6144a JW |
626 | } |
627 | ||
a528910e JW |
628 | static int __add_to_page_cache_locked(struct page *page, |
629 | struct address_space *mapping, | |
630 | pgoff_t offset, gfp_t gfp_mask, | |
631 | void **shadowp) | |
1da177e4 | 632 | { |
00501b53 JW |
633 | int huge = PageHuge(page); |
634 | struct mem_cgroup *memcg; | |
e286781d NP |
635 | int error; |
636 | ||
309381fe SL |
637 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
638 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d | 639 | |
00501b53 JW |
640 | if (!huge) { |
641 | error = mem_cgroup_try_charge(page, current->mm, | |
f627c2f5 | 642 | gfp_mask, &memcg, false); |
00501b53 JW |
643 | if (error) |
644 | return error; | |
645 | } | |
1da177e4 | 646 | |
5e4c0d97 | 647 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 648 | if (error) { |
00501b53 | 649 | if (!huge) |
f627c2f5 | 650 | mem_cgroup_cancel_charge(page, memcg, false); |
66a0c8ee KS |
651 | return error; |
652 | } | |
653 | ||
09cbfeaf | 654 | get_page(page); |
66a0c8ee KS |
655 | page->mapping = mapping; |
656 | page->index = offset; | |
657 | ||
658 | spin_lock_irq(&mapping->tree_lock); | |
a528910e | 659 | error = page_cache_tree_insert(mapping, page, shadowp); |
66a0c8ee KS |
660 | radix_tree_preload_end(); |
661 | if (unlikely(error)) | |
662 | goto err_insert; | |
4165b9b4 MH |
663 | |
664 | /* hugetlb pages do not participate in page cache accounting. */ | |
665 | if (!huge) | |
666 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
66a0c8ee | 667 | spin_unlock_irq(&mapping->tree_lock); |
00501b53 | 668 | if (!huge) |
f627c2f5 | 669 | mem_cgroup_commit_charge(page, memcg, false, false); |
66a0c8ee KS |
670 | trace_mm_filemap_add_to_page_cache(page); |
671 | return 0; | |
672 | err_insert: | |
673 | page->mapping = NULL; | |
674 | /* Leave page->index set: truncation relies upon it */ | |
675 | spin_unlock_irq(&mapping->tree_lock); | |
00501b53 | 676 | if (!huge) |
f627c2f5 | 677 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 678 | put_page(page); |
1da177e4 LT |
679 | return error; |
680 | } | |
a528910e JW |
681 | |
682 | /** | |
683 | * add_to_page_cache_locked - add a locked page to the pagecache | |
684 | * @page: page to add | |
685 | * @mapping: the page's address_space | |
686 | * @offset: page index | |
687 | * @gfp_mask: page allocation mode | |
688 | * | |
689 | * This function is used to add a page to the pagecache. It must be locked. | |
690 | * This function does not add the page to the LRU. The caller must do that. | |
691 | */ | |
692 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
693 | pgoff_t offset, gfp_t gfp_mask) | |
694 | { | |
695 | return __add_to_page_cache_locked(page, mapping, offset, | |
696 | gfp_mask, NULL); | |
697 | } | |
e286781d | 698 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
699 | |
700 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 701 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 702 | { |
a528910e | 703 | void *shadow = NULL; |
4f98a2fe RR |
704 | int ret; |
705 | ||
48c935ad | 706 | __SetPageLocked(page); |
a528910e JW |
707 | ret = __add_to_page_cache_locked(page, mapping, offset, |
708 | gfp_mask, &shadow); | |
709 | if (unlikely(ret)) | |
48c935ad | 710 | __ClearPageLocked(page); |
a528910e JW |
711 | else { |
712 | /* | |
713 | * The page might have been evicted from cache only | |
714 | * recently, in which case it should be activated like | |
715 | * any other repeatedly accessed page. | |
f0281a00 RR |
716 | * The exception is pages getting rewritten; evicting other |
717 | * data from the working set, only to cache data that will | |
718 | * get overwritten with something else, is a waste of memory. | |
a528910e | 719 | */ |
f0281a00 RR |
720 | if (!(gfp_mask & __GFP_WRITE) && |
721 | shadow && workingset_refault(shadow)) { | |
a528910e JW |
722 | SetPageActive(page); |
723 | workingset_activation(page); | |
724 | } else | |
725 | ClearPageActive(page); | |
726 | lru_cache_add(page); | |
727 | } | |
1da177e4 LT |
728 | return ret; |
729 | } | |
18bc0bbd | 730 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 731 | |
44110fe3 | 732 | #ifdef CONFIG_NUMA |
2ae88149 | 733 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 734 | { |
c0ff7453 MX |
735 | int n; |
736 | struct page *page; | |
737 | ||
44110fe3 | 738 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
739 | unsigned int cpuset_mems_cookie; |
740 | do { | |
d26914d1 | 741 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 742 | n = cpuset_mem_spread_node(); |
96db800f | 743 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 744 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 745 | |
c0ff7453 | 746 | return page; |
44110fe3 | 747 | } |
2ae88149 | 748 | return alloc_pages(gfp, 0); |
44110fe3 | 749 | } |
2ae88149 | 750 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
751 | #endif |
752 | ||
1da177e4 LT |
753 | /* |
754 | * In order to wait for pages to become available there must be | |
755 | * waitqueues associated with pages. By using a hash table of | |
756 | * waitqueues where the bucket discipline is to maintain all | |
757 | * waiters on the same queue and wake all when any of the pages | |
758 | * become available, and for the woken contexts to check to be | |
759 | * sure the appropriate page became available, this saves space | |
760 | * at a cost of "thundering herd" phenomena during rare hash | |
761 | * collisions. | |
762 | */ | |
a4796e37 | 763 | wait_queue_head_t *page_waitqueue(struct page *page) |
1da177e4 LT |
764 | { |
765 | const struct zone *zone = page_zone(page); | |
766 | ||
767 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
768 | } | |
a4796e37 | 769 | EXPORT_SYMBOL(page_waitqueue); |
1da177e4 | 770 | |
920c7a5d | 771 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
772 | { |
773 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
774 | ||
775 | if (test_bit(bit_nr, &page->flags)) | |
74316201 | 776 | __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
777 | TASK_UNINTERRUPTIBLE); |
778 | } | |
779 | EXPORT_SYMBOL(wait_on_page_bit); | |
780 | ||
f62e00cc KM |
781 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
782 | { | |
783 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
784 | ||
785 | if (!test_bit(bit_nr, &page->flags)) | |
786 | return 0; | |
787 | ||
788 | return __wait_on_bit(page_waitqueue(page), &wait, | |
74316201 | 789 | bit_wait_io, TASK_KILLABLE); |
f62e00cc KM |
790 | } |
791 | ||
cbbce822 N |
792 | int wait_on_page_bit_killable_timeout(struct page *page, |
793 | int bit_nr, unsigned long timeout) | |
794 | { | |
795 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
796 | ||
797 | wait.key.timeout = jiffies + timeout; | |
798 | if (!test_bit(bit_nr, &page->flags)) | |
799 | return 0; | |
800 | return __wait_on_bit(page_waitqueue(page), &wait, | |
801 | bit_wait_io_timeout, TASK_KILLABLE); | |
802 | } | |
803 | EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout); | |
804 | ||
385e1ca5 DH |
805 | /** |
806 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
807 | * @page: Page defining the wait queue of interest |
808 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
809 | * |
810 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
811 | */ | |
812 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
813 | { | |
814 | wait_queue_head_t *q = page_waitqueue(page); | |
815 | unsigned long flags; | |
816 | ||
817 | spin_lock_irqsave(&q->lock, flags); | |
818 | __add_wait_queue(q, waiter); | |
819 | spin_unlock_irqrestore(&q->lock, flags); | |
820 | } | |
821 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
822 | ||
1da177e4 | 823 | /** |
485bb99b | 824 | * unlock_page - unlock a locked page |
1da177e4 LT |
825 | * @page: the page |
826 | * | |
827 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
828 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 829 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
830 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
831 | * | |
8413ac9d NP |
832 | * The mb is necessary to enforce ordering between the clear_bit and the read |
833 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 834 | */ |
920c7a5d | 835 | void unlock_page(struct page *page) |
1da177e4 | 836 | { |
48c935ad | 837 | page = compound_head(page); |
309381fe | 838 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
8413ac9d | 839 | clear_bit_unlock(PG_locked, &page->flags); |
4e857c58 | 840 | smp_mb__after_atomic(); |
1da177e4 LT |
841 | wake_up_page(page, PG_locked); |
842 | } | |
843 | EXPORT_SYMBOL(unlock_page); | |
844 | ||
485bb99b RD |
845 | /** |
846 | * end_page_writeback - end writeback against a page | |
847 | * @page: the page | |
1da177e4 LT |
848 | */ |
849 | void end_page_writeback(struct page *page) | |
850 | { | |
888cf2db MG |
851 | /* |
852 | * TestClearPageReclaim could be used here but it is an atomic | |
853 | * operation and overkill in this particular case. Failing to | |
854 | * shuffle a page marked for immediate reclaim is too mild to | |
855 | * justify taking an atomic operation penalty at the end of | |
856 | * ever page writeback. | |
857 | */ | |
858 | if (PageReclaim(page)) { | |
859 | ClearPageReclaim(page); | |
ac6aadb2 | 860 | rotate_reclaimable_page(page); |
888cf2db | 861 | } |
ac6aadb2 MS |
862 | |
863 | if (!test_clear_page_writeback(page)) | |
864 | BUG(); | |
865 | ||
4e857c58 | 866 | smp_mb__after_atomic(); |
1da177e4 LT |
867 | wake_up_page(page, PG_writeback); |
868 | } | |
869 | EXPORT_SYMBOL(end_page_writeback); | |
870 | ||
57d99845 MW |
871 | /* |
872 | * After completing I/O on a page, call this routine to update the page | |
873 | * flags appropriately | |
874 | */ | |
875 | void page_endio(struct page *page, int rw, int err) | |
876 | { | |
877 | if (rw == READ) { | |
878 | if (!err) { | |
879 | SetPageUptodate(page); | |
880 | } else { | |
881 | ClearPageUptodate(page); | |
882 | SetPageError(page); | |
883 | } | |
884 | unlock_page(page); | |
885 | } else { /* rw == WRITE */ | |
886 | if (err) { | |
887 | SetPageError(page); | |
888 | if (page->mapping) | |
889 | mapping_set_error(page->mapping, err); | |
890 | } | |
891 | end_page_writeback(page); | |
892 | } | |
893 | } | |
894 | EXPORT_SYMBOL_GPL(page_endio); | |
895 | ||
485bb99b RD |
896 | /** |
897 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
898 | * @page: the page to lock | |
1da177e4 | 899 | */ |
920c7a5d | 900 | void __lock_page(struct page *page) |
1da177e4 | 901 | { |
48c935ad KS |
902 | struct page *page_head = compound_head(page); |
903 | DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked); | |
1da177e4 | 904 | |
48c935ad | 905 | __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io, |
1da177e4 LT |
906 | TASK_UNINTERRUPTIBLE); |
907 | } | |
908 | EXPORT_SYMBOL(__lock_page); | |
909 | ||
b5606c2d | 910 | int __lock_page_killable(struct page *page) |
2687a356 | 911 | { |
48c935ad KS |
912 | struct page *page_head = compound_head(page); |
913 | DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked); | |
2687a356 | 914 | |
48c935ad | 915 | return __wait_on_bit_lock(page_waitqueue(page_head), &wait, |
74316201 | 916 | bit_wait_io, TASK_KILLABLE); |
2687a356 | 917 | } |
18bc0bbd | 918 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 919 | |
9a95f3cf PC |
920 | /* |
921 | * Return values: | |
922 | * 1 - page is locked; mmap_sem is still held. | |
923 | * 0 - page is not locked. | |
924 | * mmap_sem has been released (up_read()), unless flags had both | |
925 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
926 | * which case mmap_sem is still held. | |
927 | * | |
928 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
929 | * with the page locked and the mmap_sem unperturbed. | |
930 | */ | |
d065bd81 ML |
931 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
932 | unsigned int flags) | |
933 | { | |
37b23e05 KM |
934 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
935 | /* | |
936 | * CAUTION! In this case, mmap_sem is not released | |
937 | * even though return 0. | |
938 | */ | |
939 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
940 | return 0; | |
941 | ||
942 | up_read(&mm->mmap_sem); | |
943 | if (flags & FAULT_FLAG_KILLABLE) | |
944 | wait_on_page_locked_killable(page); | |
945 | else | |
318b275f | 946 | wait_on_page_locked(page); |
d065bd81 | 947 | return 0; |
37b23e05 KM |
948 | } else { |
949 | if (flags & FAULT_FLAG_KILLABLE) { | |
950 | int ret; | |
951 | ||
952 | ret = __lock_page_killable(page); | |
953 | if (ret) { | |
954 | up_read(&mm->mmap_sem); | |
955 | return 0; | |
956 | } | |
957 | } else | |
958 | __lock_page(page); | |
959 | return 1; | |
d065bd81 ML |
960 | } |
961 | } | |
962 | ||
e7b563bb JW |
963 | /** |
964 | * page_cache_next_hole - find the next hole (not-present entry) | |
965 | * @mapping: mapping | |
966 | * @index: index | |
967 | * @max_scan: maximum range to search | |
968 | * | |
969 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
970 | * lowest indexed hole. | |
971 | * | |
972 | * Returns: the index of the hole if found, otherwise returns an index | |
973 | * outside of the set specified (in which case 'return - index >= | |
974 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
975 | * be returned. | |
976 | * | |
977 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
978 | * like radix_tree_gang_lookup, this will not atomically search a | |
979 | * snapshot of the tree at a single point in time. For example, if a | |
980 | * hole is created at index 5, then subsequently a hole is created at | |
981 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
982 | * if called under rcu_read_lock. | |
983 | */ | |
984 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
985 | pgoff_t index, unsigned long max_scan) | |
986 | { | |
987 | unsigned long i; | |
988 | ||
989 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
990 | struct page *page; |
991 | ||
992 | page = radix_tree_lookup(&mapping->page_tree, index); | |
993 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
994 | break; |
995 | index++; | |
996 | if (index == 0) | |
997 | break; | |
998 | } | |
999 | ||
1000 | return index; | |
1001 | } | |
1002 | EXPORT_SYMBOL(page_cache_next_hole); | |
1003 | ||
1004 | /** | |
1005 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
1006 | * @mapping: mapping | |
1007 | * @index: index | |
1008 | * @max_scan: maximum range to search | |
1009 | * | |
1010 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
1011 | * the first hole. | |
1012 | * | |
1013 | * Returns: the index of the hole if found, otherwise returns an index | |
1014 | * outside of the set specified (in which case 'index - return >= | |
1015 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
1016 | * will be returned. | |
1017 | * | |
1018 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
1019 | * like radix_tree_gang_lookup, this will not atomically search a | |
1020 | * snapshot of the tree at a single point in time. For example, if a | |
1021 | * hole is created at index 10, then subsequently a hole is created at | |
1022 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
1023 | * called under rcu_read_lock. | |
1024 | */ | |
1025 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
1026 | pgoff_t index, unsigned long max_scan) | |
1027 | { | |
1028 | unsigned long i; | |
1029 | ||
1030 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
1031 | struct page *page; |
1032 | ||
1033 | page = radix_tree_lookup(&mapping->page_tree, index); | |
1034 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
1035 | break; |
1036 | index--; | |
1037 | if (index == ULONG_MAX) | |
1038 | break; | |
1039 | } | |
1040 | ||
1041 | return index; | |
1042 | } | |
1043 | EXPORT_SYMBOL(page_cache_prev_hole); | |
1044 | ||
485bb99b | 1045 | /** |
0cd6144a | 1046 | * find_get_entry - find and get a page cache entry |
485bb99b | 1047 | * @mapping: the address_space to search |
0cd6144a JW |
1048 | * @offset: the page cache index |
1049 | * | |
1050 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1051 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1052 | * |
139b6a6f JW |
1053 | * If the slot holds a shadow entry of a previously evicted page, or a |
1054 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1055 | * |
1056 | * Otherwise, %NULL is returned. | |
1da177e4 | 1057 | */ |
0cd6144a | 1058 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1059 | { |
a60637c8 | 1060 | void **pagep; |
1da177e4 LT |
1061 | struct page *page; |
1062 | ||
a60637c8 NP |
1063 | rcu_read_lock(); |
1064 | repeat: | |
1065 | page = NULL; | |
1066 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
1067 | if (pagep) { | |
1068 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
1069 | if (unlikely(!page)) |
1070 | goto out; | |
a2c16d6c | 1071 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1072 | if (radix_tree_deref_retry(page)) |
1073 | goto repeat; | |
1074 | /* | |
139b6a6f JW |
1075 | * A shadow entry of a recently evicted page, |
1076 | * or a swap entry from shmem/tmpfs. Return | |
1077 | * it without attempting to raise page count. | |
8079b1c8 HD |
1078 | */ |
1079 | goto out; | |
a2c16d6c | 1080 | } |
a60637c8 NP |
1081 | if (!page_cache_get_speculative(page)) |
1082 | goto repeat; | |
1083 | ||
1084 | /* | |
1085 | * Has the page moved? | |
1086 | * This is part of the lockless pagecache protocol. See | |
1087 | * include/linux/pagemap.h for details. | |
1088 | */ | |
1089 | if (unlikely(page != *pagep)) { | |
09cbfeaf | 1090 | put_page(page); |
a60637c8 NP |
1091 | goto repeat; |
1092 | } | |
1093 | } | |
27d20fdd | 1094 | out: |
a60637c8 NP |
1095 | rcu_read_unlock(); |
1096 | ||
1da177e4 LT |
1097 | return page; |
1098 | } | |
0cd6144a | 1099 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1100 | |
0cd6144a JW |
1101 | /** |
1102 | * find_lock_entry - locate, pin and lock a page cache entry | |
1103 | * @mapping: the address_space to search | |
1104 | * @offset: the page cache index | |
1105 | * | |
1106 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1107 | * page cache page, it is returned locked and with an increased | |
1108 | * refcount. | |
1109 | * | |
139b6a6f JW |
1110 | * If the slot holds a shadow entry of a previously evicted page, or a |
1111 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1112 | * |
1113 | * Otherwise, %NULL is returned. | |
1114 | * | |
1115 | * find_lock_entry() may sleep. | |
1116 | */ | |
1117 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1118 | { |
1119 | struct page *page; | |
1120 | ||
1da177e4 | 1121 | repeat: |
0cd6144a | 1122 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1123 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1124 | lock_page(page); |
1125 | /* Has the page been truncated? */ | |
1126 | if (unlikely(page->mapping != mapping)) { | |
1127 | unlock_page(page); | |
09cbfeaf | 1128 | put_page(page); |
a60637c8 | 1129 | goto repeat; |
1da177e4 | 1130 | } |
309381fe | 1131 | VM_BUG_ON_PAGE(page->index != offset, page); |
1da177e4 | 1132 | } |
1da177e4 LT |
1133 | return page; |
1134 | } | |
0cd6144a JW |
1135 | EXPORT_SYMBOL(find_lock_entry); |
1136 | ||
1137 | /** | |
2457aec6 | 1138 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1139 | * @mapping: the address_space to search |
1140 | * @offset: the page index | |
2457aec6 | 1141 | * @fgp_flags: PCG flags |
45f87de5 | 1142 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
0cd6144a | 1143 | * |
2457aec6 | 1144 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1145 | * |
75325189 | 1146 | * PCG flags modify how the page is returned. |
0cd6144a | 1147 | * |
2457aec6 MG |
1148 | * FGP_ACCESSED: the page will be marked accessed |
1149 | * FGP_LOCK: Page is return locked | |
1150 | * FGP_CREAT: If page is not present then a new page is allocated using | |
45f87de5 MH |
1151 | * @gfp_mask and added to the page cache and the VM's LRU |
1152 | * list. The page is returned locked and with an increased | |
1153 | * refcount. Otherwise, %NULL is returned. | |
1da177e4 | 1154 | * |
2457aec6 MG |
1155 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1156 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1157 | * |
2457aec6 | 1158 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1159 | */ |
2457aec6 | 1160 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
45f87de5 | 1161 | int fgp_flags, gfp_t gfp_mask) |
1da177e4 | 1162 | { |
eb2be189 | 1163 | struct page *page; |
2457aec6 | 1164 | |
1da177e4 | 1165 | repeat: |
2457aec6 MG |
1166 | page = find_get_entry(mapping, offset); |
1167 | if (radix_tree_exceptional_entry(page)) | |
1168 | page = NULL; | |
1169 | if (!page) | |
1170 | goto no_page; | |
1171 | ||
1172 | if (fgp_flags & FGP_LOCK) { | |
1173 | if (fgp_flags & FGP_NOWAIT) { | |
1174 | if (!trylock_page(page)) { | |
09cbfeaf | 1175 | put_page(page); |
2457aec6 MG |
1176 | return NULL; |
1177 | } | |
1178 | } else { | |
1179 | lock_page(page); | |
1180 | } | |
1181 | ||
1182 | /* Has the page been truncated? */ | |
1183 | if (unlikely(page->mapping != mapping)) { | |
1184 | unlock_page(page); | |
09cbfeaf | 1185 | put_page(page); |
2457aec6 MG |
1186 | goto repeat; |
1187 | } | |
1188 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1189 | } | |
1190 | ||
1191 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1192 | mark_page_accessed(page); | |
1193 | ||
1194 | no_page: | |
1195 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1196 | int err; | |
1197 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1198 | gfp_mask |= __GFP_WRITE; |
1199 | if (fgp_flags & FGP_NOFS) | |
1200 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1201 | |
45f87de5 | 1202 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1203 | if (!page) |
1204 | return NULL; | |
2457aec6 MG |
1205 | |
1206 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1207 | fgp_flags |= FGP_LOCK; | |
1208 | ||
eb39d618 | 1209 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1210 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1211 | __SetPageReferenced(page); |
2457aec6 | 1212 | |
45f87de5 MH |
1213 | err = add_to_page_cache_lru(page, mapping, offset, |
1214 | gfp_mask & GFP_RECLAIM_MASK); | |
eb2be189 | 1215 | if (unlikely(err)) { |
09cbfeaf | 1216 | put_page(page); |
eb2be189 NP |
1217 | page = NULL; |
1218 | if (err == -EEXIST) | |
1219 | goto repeat; | |
1da177e4 | 1220 | } |
1da177e4 | 1221 | } |
2457aec6 | 1222 | |
1da177e4 LT |
1223 | return page; |
1224 | } | |
2457aec6 | 1225 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1226 | |
0cd6144a JW |
1227 | /** |
1228 | * find_get_entries - gang pagecache lookup | |
1229 | * @mapping: The address_space to search | |
1230 | * @start: The starting page cache index | |
1231 | * @nr_entries: The maximum number of entries | |
1232 | * @entries: Where the resulting entries are placed | |
1233 | * @indices: The cache indices corresponding to the entries in @entries | |
1234 | * | |
1235 | * find_get_entries() will search for and return a group of up to | |
1236 | * @nr_entries entries in the mapping. The entries are placed at | |
1237 | * @entries. find_get_entries() takes a reference against any actual | |
1238 | * pages it returns. | |
1239 | * | |
1240 | * The search returns a group of mapping-contiguous page cache entries | |
1241 | * with ascending indexes. There may be holes in the indices due to | |
1242 | * not-present pages. | |
1243 | * | |
139b6a6f JW |
1244 | * Any shadow entries of evicted pages, or swap entries from |
1245 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1246 | * |
1247 | * find_get_entries() returns the number of pages and shadow entries | |
1248 | * which were found. | |
1249 | */ | |
1250 | unsigned find_get_entries(struct address_space *mapping, | |
1251 | pgoff_t start, unsigned int nr_entries, | |
1252 | struct page **entries, pgoff_t *indices) | |
1253 | { | |
1254 | void **slot; | |
1255 | unsigned int ret = 0; | |
1256 | struct radix_tree_iter iter; | |
1257 | ||
1258 | if (!nr_entries) | |
1259 | return 0; | |
1260 | ||
1261 | rcu_read_lock(); | |
0cd6144a JW |
1262 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
1263 | struct page *page; | |
1264 | repeat: | |
1265 | page = radix_tree_deref_slot(slot); | |
1266 | if (unlikely(!page)) | |
1267 | continue; | |
1268 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
1269 | if (radix_tree_deref_retry(page)) { |
1270 | slot = radix_tree_iter_retry(&iter); | |
1271 | continue; | |
1272 | } | |
0cd6144a | 1273 | /* |
f9fe48be RZ |
1274 | * A shadow entry of a recently evicted page, a swap |
1275 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1276 | * without attempting to raise page count. | |
0cd6144a JW |
1277 | */ |
1278 | goto export; | |
1279 | } | |
1280 | if (!page_cache_get_speculative(page)) | |
1281 | goto repeat; | |
1282 | ||
1283 | /* Has the page moved? */ | |
1284 | if (unlikely(page != *slot)) { | |
09cbfeaf | 1285 | put_page(page); |
0cd6144a JW |
1286 | goto repeat; |
1287 | } | |
1288 | export: | |
1289 | indices[ret] = iter.index; | |
1290 | entries[ret] = page; | |
1291 | if (++ret == nr_entries) | |
1292 | break; | |
1293 | } | |
1294 | rcu_read_unlock(); | |
1295 | return ret; | |
1296 | } | |
1297 | ||
1da177e4 LT |
1298 | /** |
1299 | * find_get_pages - gang pagecache lookup | |
1300 | * @mapping: The address_space to search | |
1301 | * @start: The starting page index | |
1302 | * @nr_pages: The maximum number of pages | |
1303 | * @pages: Where the resulting pages are placed | |
1304 | * | |
1305 | * find_get_pages() will search for and return a group of up to | |
1306 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
1307 | * find_get_pages() takes a reference against the returned pages. | |
1308 | * | |
1309 | * The search returns a group of mapping-contiguous pages with ascending | |
1310 | * indexes. There may be holes in the indices due to not-present pages. | |
1311 | * | |
1312 | * find_get_pages() returns the number of pages which were found. | |
1313 | */ | |
1314 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
1315 | unsigned int nr_pages, struct page **pages) | |
1316 | { | |
0fc9d104 KK |
1317 | struct radix_tree_iter iter; |
1318 | void **slot; | |
1319 | unsigned ret = 0; | |
1320 | ||
1321 | if (unlikely(!nr_pages)) | |
1322 | return 0; | |
a60637c8 NP |
1323 | |
1324 | rcu_read_lock(); | |
0fc9d104 | 1325 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
a60637c8 NP |
1326 | struct page *page; |
1327 | repeat: | |
0fc9d104 | 1328 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1329 | if (unlikely(!page)) |
1330 | continue; | |
9d8aa4ea | 1331 | |
a2c16d6c | 1332 | if (radix_tree_exception(page)) { |
8079b1c8 | 1333 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1334 | slot = radix_tree_iter_retry(&iter); |
1335 | continue; | |
8079b1c8 | 1336 | } |
a2c16d6c | 1337 | /* |
139b6a6f JW |
1338 | * A shadow entry of a recently evicted page, |
1339 | * or a swap entry from shmem/tmpfs. Skip | |
1340 | * over it. | |
a2c16d6c | 1341 | */ |
8079b1c8 | 1342 | continue; |
27d20fdd | 1343 | } |
a60637c8 NP |
1344 | |
1345 | if (!page_cache_get_speculative(page)) | |
1346 | goto repeat; | |
1347 | ||
1348 | /* Has the page moved? */ | |
0fc9d104 | 1349 | if (unlikely(page != *slot)) { |
09cbfeaf | 1350 | put_page(page); |
a60637c8 NP |
1351 | goto repeat; |
1352 | } | |
1da177e4 | 1353 | |
a60637c8 | 1354 | pages[ret] = page; |
0fc9d104 KK |
1355 | if (++ret == nr_pages) |
1356 | break; | |
a60637c8 | 1357 | } |
5b280c0c | 1358 | |
a60637c8 | 1359 | rcu_read_unlock(); |
1da177e4 LT |
1360 | return ret; |
1361 | } | |
1362 | ||
ebf43500 JA |
1363 | /** |
1364 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1365 | * @mapping: The address_space to search | |
1366 | * @index: The starting page index | |
1367 | * @nr_pages: The maximum number of pages | |
1368 | * @pages: Where the resulting pages are placed | |
1369 | * | |
1370 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1371 | * that the returned number of pages are guaranteed to be contiguous. | |
1372 | * | |
1373 | * find_get_pages_contig() returns the number of pages which were found. | |
1374 | */ | |
1375 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1376 | unsigned int nr_pages, struct page **pages) | |
1377 | { | |
0fc9d104 KK |
1378 | struct radix_tree_iter iter; |
1379 | void **slot; | |
1380 | unsigned int ret = 0; | |
1381 | ||
1382 | if (unlikely(!nr_pages)) | |
1383 | return 0; | |
a60637c8 NP |
1384 | |
1385 | rcu_read_lock(); | |
0fc9d104 | 1386 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
a60637c8 NP |
1387 | struct page *page; |
1388 | repeat: | |
0fc9d104 KK |
1389 | page = radix_tree_deref_slot(slot); |
1390 | /* The hole, there no reason to continue */ | |
a60637c8 | 1391 | if (unlikely(!page)) |
0fc9d104 | 1392 | break; |
9d8aa4ea | 1393 | |
a2c16d6c | 1394 | if (radix_tree_exception(page)) { |
8079b1c8 | 1395 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1396 | slot = radix_tree_iter_retry(&iter); |
1397 | continue; | |
8079b1c8 | 1398 | } |
a2c16d6c | 1399 | /* |
139b6a6f JW |
1400 | * A shadow entry of a recently evicted page, |
1401 | * or a swap entry from shmem/tmpfs. Stop | |
1402 | * looking for contiguous pages. | |
a2c16d6c | 1403 | */ |
8079b1c8 | 1404 | break; |
a2c16d6c | 1405 | } |
ebf43500 | 1406 | |
a60637c8 NP |
1407 | if (!page_cache_get_speculative(page)) |
1408 | goto repeat; | |
1409 | ||
1410 | /* Has the page moved? */ | |
0fc9d104 | 1411 | if (unlikely(page != *slot)) { |
09cbfeaf | 1412 | put_page(page); |
a60637c8 NP |
1413 | goto repeat; |
1414 | } | |
1415 | ||
9cbb4cb2 NP |
1416 | /* |
1417 | * must check mapping and index after taking the ref. | |
1418 | * otherwise we can get both false positives and false | |
1419 | * negatives, which is just confusing to the caller. | |
1420 | */ | |
0fc9d104 | 1421 | if (page->mapping == NULL || page->index != iter.index) { |
09cbfeaf | 1422 | put_page(page); |
9cbb4cb2 NP |
1423 | break; |
1424 | } | |
1425 | ||
a60637c8 | 1426 | pages[ret] = page; |
0fc9d104 KK |
1427 | if (++ret == nr_pages) |
1428 | break; | |
ebf43500 | 1429 | } |
a60637c8 NP |
1430 | rcu_read_unlock(); |
1431 | return ret; | |
ebf43500 | 1432 | } |
ef71c15c | 1433 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1434 | |
485bb99b RD |
1435 | /** |
1436 | * find_get_pages_tag - find and return pages that match @tag | |
1437 | * @mapping: the address_space to search | |
1438 | * @index: the starting page index | |
1439 | * @tag: the tag index | |
1440 | * @nr_pages: the maximum number of pages | |
1441 | * @pages: where the resulting pages are placed | |
1442 | * | |
1da177e4 | 1443 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1444 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
1445 | */ |
1446 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
1447 | int tag, unsigned int nr_pages, struct page **pages) | |
1448 | { | |
0fc9d104 KK |
1449 | struct radix_tree_iter iter; |
1450 | void **slot; | |
1451 | unsigned ret = 0; | |
1452 | ||
1453 | if (unlikely(!nr_pages)) | |
1454 | return 0; | |
a60637c8 NP |
1455 | |
1456 | rcu_read_lock(); | |
0fc9d104 KK |
1457 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1458 | &iter, *index, tag) { | |
a60637c8 NP |
1459 | struct page *page; |
1460 | repeat: | |
0fc9d104 | 1461 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1462 | if (unlikely(!page)) |
1463 | continue; | |
9d8aa4ea | 1464 | |
a2c16d6c | 1465 | if (radix_tree_exception(page)) { |
8079b1c8 | 1466 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1467 | slot = radix_tree_iter_retry(&iter); |
1468 | continue; | |
8079b1c8 | 1469 | } |
a2c16d6c | 1470 | /* |
139b6a6f JW |
1471 | * A shadow entry of a recently evicted page. |
1472 | * | |
1473 | * Those entries should never be tagged, but | |
1474 | * this tree walk is lockless and the tags are | |
1475 | * looked up in bulk, one radix tree node at a | |
1476 | * time, so there is a sizable window for page | |
1477 | * reclaim to evict a page we saw tagged. | |
1478 | * | |
1479 | * Skip over it. | |
a2c16d6c | 1480 | */ |
139b6a6f | 1481 | continue; |
a2c16d6c | 1482 | } |
a60637c8 NP |
1483 | |
1484 | if (!page_cache_get_speculative(page)) | |
1485 | goto repeat; | |
1486 | ||
1487 | /* Has the page moved? */ | |
0fc9d104 | 1488 | if (unlikely(page != *slot)) { |
09cbfeaf | 1489 | put_page(page); |
a60637c8 NP |
1490 | goto repeat; |
1491 | } | |
1492 | ||
1493 | pages[ret] = page; | |
0fc9d104 KK |
1494 | if (++ret == nr_pages) |
1495 | break; | |
a60637c8 | 1496 | } |
5b280c0c | 1497 | |
a60637c8 | 1498 | rcu_read_unlock(); |
1da177e4 | 1499 | |
1da177e4 LT |
1500 | if (ret) |
1501 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1502 | |
1da177e4 LT |
1503 | return ret; |
1504 | } | |
ef71c15c | 1505 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1506 | |
7e7f7749 RZ |
1507 | /** |
1508 | * find_get_entries_tag - find and return entries that match @tag | |
1509 | * @mapping: the address_space to search | |
1510 | * @start: the starting page cache index | |
1511 | * @tag: the tag index | |
1512 | * @nr_entries: the maximum number of entries | |
1513 | * @entries: where the resulting entries are placed | |
1514 | * @indices: the cache indices corresponding to the entries in @entries | |
1515 | * | |
1516 | * Like find_get_entries, except we only return entries which are tagged with | |
1517 | * @tag. | |
1518 | */ | |
1519 | unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start, | |
1520 | int tag, unsigned int nr_entries, | |
1521 | struct page **entries, pgoff_t *indices) | |
1522 | { | |
1523 | void **slot; | |
1524 | unsigned int ret = 0; | |
1525 | struct radix_tree_iter iter; | |
1526 | ||
1527 | if (!nr_entries) | |
1528 | return 0; | |
1529 | ||
1530 | rcu_read_lock(); | |
7e7f7749 RZ |
1531 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1532 | &iter, start, tag) { | |
1533 | struct page *page; | |
1534 | repeat: | |
1535 | page = radix_tree_deref_slot(slot); | |
1536 | if (unlikely(!page)) | |
1537 | continue; | |
1538 | if (radix_tree_exception(page)) { | |
1539 | if (radix_tree_deref_retry(page)) { | |
2cf938aa MW |
1540 | slot = radix_tree_iter_retry(&iter); |
1541 | continue; | |
7e7f7749 RZ |
1542 | } |
1543 | ||
1544 | /* | |
1545 | * A shadow entry of a recently evicted page, a swap | |
1546 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1547 | * without attempting to raise page count. | |
1548 | */ | |
1549 | goto export; | |
1550 | } | |
1551 | if (!page_cache_get_speculative(page)) | |
1552 | goto repeat; | |
1553 | ||
1554 | /* Has the page moved? */ | |
1555 | if (unlikely(page != *slot)) { | |
09cbfeaf | 1556 | put_page(page); |
7e7f7749 RZ |
1557 | goto repeat; |
1558 | } | |
1559 | export: | |
1560 | indices[ret] = iter.index; | |
1561 | entries[ret] = page; | |
1562 | if (++ret == nr_entries) | |
1563 | break; | |
1564 | } | |
1565 | rcu_read_unlock(); | |
1566 | return ret; | |
1567 | } | |
1568 | EXPORT_SYMBOL(find_get_entries_tag); | |
1569 | ||
76d42bd9 WF |
1570 | /* |
1571 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1572 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1573 | * | |
1574 | * ---R__________________________________________B__________ | |
1575 | * ^ reading here ^ bad block(assume 4k) | |
1576 | * | |
1577 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1578 | * => failing the whole request => read(R) => read(R+1) => | |
1579 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1580 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1581 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1582 | * | |
1583 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1584 | */ | |
1585 | static void shrink_readahead_size_eio(struct file *filp, | |
1586 | struct file_ra_state *ra) | |
1587 | { | |
76d42bd9 | 1588 | ra->ra_pages /= 4; |
76d42bd9 WF |
1589 | } |
1590 | ||
485bb99b | 1591 | /** |
36e78914 | 1592 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1593 | * @filp: the file to read |
1594 | * @ppos: current file position | |
6e58e79d AV |
1595 | * @iter: data destination |
1596 | * @written: already copied | |
485bb99b | 1597 | * |
1da177e4 | 1598 | * This is a generic file read routine, and uses the |
485bb99b | 1599 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1600 | * |
1601 | * This is really ugly. But the goto's actually try to clarify some | |
1602 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1603 | */ |
6e58e79d AV |
1604 | static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos, |
1605 | struct iov_iter *iter, ssize_t written) | |
1da177e4 | 1606 | { |
36e78914 | 1607 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1608 | struct inode *inode = mapping->host; |
36e78914 | 1609 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1610 | pgoff_t index; |
1611 | pgoff_t last_index; | |
1612 | pgoff_t prev_index; | |
1613 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1614 | unsigned int prev_offset; |
6e58e79d | 1615 | int error = 0; |
1da177e4 | 1616 | |
09cbfeaf KS |
1617 | index = *ppos >> PAGE_SHIFT; |
1618 | prev_index = ra->prev_pos >> PAGE_SHIFT; | |
1619 | prev_offset = ra->prev_pos & (PAGE_SIZE-1); | |
1620 | last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; | |
1621 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 | 1622 | |
1da177e4 LT |
1623 | for (;;) { |
1624 | struct page *page; | |
57f6b96c | 1625 | pgoff_t end_index; |
a32ea1e1 | 1626 | loff_t isize; |
1da177e4 LT |
1627 | unsigned long nr, ret; |
1628 | ||
1da177e4 | 1629 | cond_resched(); |
1da177e4 LT |
1630 | find_page: |
1631 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1632 | if (!page) { |
cf914a7d | 1633 | page_cache_sync_readahead(mapping, |
7ff81078 | 1634 | ra, filp, |
3ea89ee8 FW |
1635 | index, last_index - index); |
1636 | page = find_get_page(mapping, index); | |
1637 | if (unlikely(page == NULL)) | |
1638 | goto no_cached_page; | |
1639 | } | |
1640 | if (PageReadahead(page)) { | |
cf914a7d | 1641 | page_cache_async_readahead(mapping, |
7ff81078 | 1642 | ra, filp, page, |
3ea89ee8 | 1643 | index, last_index - index); |
1da177e4 | 1644 | } |
8ab22b9a | 1645 | if (!PageUptodate(page)) { |
ebded027 MG |
1646 | /* |
1647 | * See comment in do_read_cache_page on why | |
1648 | * wait_on_page_locked is used to avoid unnecessarily | |
1649 | * serialisations and why it's safe. | |
1650 | */ | |
1651 | wait_on_page_locked_killable(page); | |
1652 | if (PageUptodate(page)) | |
1653 | goto page_ok; | |
1654 | ||
09cbfeaf | 1655 | if (inode->i_blkbits == PAGE_SHIFT || |
8ab22b9a HH |
1656 | !mapping->a_ops->is_partially_uptodate) |
1657 | goto page_not_up_to_date; | |
529ae9aa | 1658 | if (!trylock_page(page)) |
8ab22b9a | 1659 | goto page_not_up_to_date; |
8d056cb9 DH |
1660 | /* Did it get truncated before we got the lock? */ |
1661 | if (!page->mapping) | |
1662 | goto page_not_up_to_date_locked; | |
8ab22b9a | 1663 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 1664 | offset, iter->count)) |
8ab22b9a HH |
1665 | goto page_not_up_to_date_locked; |
1666 | unlock_page(page); | |
1667 | } | |
1da177e4 | 1668 | page_ok: |
a32ea1e1 N |
1669 | /* |
1670 | * i_size must be checked after we know the page is Uptodate. | |
1671 | * | |
1672 | * Checking i_size after the check allows us to calculate | |
1673 | * the correct value for "nr", which means the zero-filled | |
1674 | * part of the page is not copied back to userspace (unless | |
1675 | * another truncate extends the file - this is desired though). | |
1676 | */ | |
1677 | ||
1678 | isize = i_size_read(inode); | |
09cbfeaf | 1679 | end_index = (isize - 1) >> PAGE_SHIFT; |
a32ea1e1 | 1680 | if (unlikely(!isize || index > end_index)) { |
09cbfeaf | 1681 | put_page(page); |
a32ea1e1 N |
1682 | goto out; |
1683 | } | |
1684 | ||
1685 | /* nr is the maximum number of bytes to copy from this page */ | |
09cbfeaf | 1686 | nr = PAGE_SIZE; |
a32ea1e1 | 1687 | if (index == end_index) { |
09cbfeaf | 1688 | nr = ((isize - 1) & ~PAGE_MASK) + 1; |
a32ea1e1 | 1689 | if (nr <= offset) { |
09cbfeaf | 1690 | put_page(page); |
a32ea1e1 N |
1691 | goto out; |
1692 | } | |
1693 | } | |
1694 | nr = nr - offset; | |
1da177e4 LT |
1695 | |
1696 | /* If users can be writing to this page using arbitrary | |
1697 | * virtual addresses, take care about potential aliasing | |
1698 | * before reading the page on the kernel side. | |
1699 | */ | |
1700 | if (mapping_writably_mapped(mapping)) | |
1701 | flush_dcache_page(page); | |
1702 | ||
1703 | /* | |
ec0f1637 JK |
1704 | * When a sequential read accesses a page several times, |
1705 | * only mark it as accessed the first time. | |
1da177e4 | 1706 | */ |
ec0f1637 | 1707 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1708 | mark_page_accessed(page); |
1709 | prev_index = index; | |
1710 | ||
1711 | /* | |
1712 | * Ok, we have the page, and it's up-to-date, so | |
1713 | * now we can copy it to user space... | |
1da177e4 | 1714 | */ |
6e58e79d AV |
1715 | |
1716 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 | 1717 | offset += ret; |
09cbfeaf KS |
1718 | index += offset >> PAGE_SHIFT; |
1719 | offset &= ~PAGE_MASK; | |
6ce745ed | 1720 | prev_offset = offset; |
1da177e4 | 1721 | |
09cbfeaf | 1722 | put_page(page); |
6e58e79d AV |
1723 | written += ret; |
1724 | if (!iov_iter_count(iter)) | |
1725 | goto out; | |
1726 | if (ret < nr) { | |
1727 | error = -EFAULT; | |
1728 | goto out; | |
1729 | } | |
1730 | continue; | |
1da177e4 LT |
1731 | |
1732 | page_not_up_to_date: | |
1733 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1734 | error = lock_page_killable(page); |
1735 | if (unlikely(error)) | |
1736 | goto readpage_error; | |
1da177e4 | 1737 | |
8ab22b9a | 1738 | page_not_up_to_date_locked: |
da6052f7 | 1739 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1740 | if (!page->mapping) { |
1741 | unlock_page(page); | |
09cbfeaf | 1742 | put_page(page); |
1da177e4 LT |
1743 | continue; |
1744 | } | |
1745 | ||
1746 | /* Did somebody else fill it already? */ | |
1747 | if (PageUptodate(page)) { | |
1748 | unlock_page(page); | |
1749 | goto page_ok; | |
1750 | } | |
1751 | ||
1752 | readpage: | |
91803b49 JM |
1753 | /* |
1754 | * A previous I/O error may have been due to temporary | |
1755 | * failures, eg. multipath errors. | |
1756 | * PG_error will be set again if readpage fails. | |
1757 | */ | |
1758 | ClearPageError(page); | |
1da177e4 LT |
1759 | /* Start the actual read. The read will unlock the page. */ |
1760 | error = mapping->a_ops->readpage(filp, page); | |
1761 | ||
994fc28c ZB |
1762 | if (unlikely(error)) { |
1763 | if (error == AOP_TRUNCATED_PAGE) { | |
09cbfeaf | 1764 | put_page(page); |
6e58e79d | 1765 | error = 0; |
994fc28c ZB |
1766 | goto find_page; |
1767 | } | |
1da177e4 | 1768 | goto readpage_error; |
994fc28c | 1769 | } |
1da177e4 LT |
1770 | |
1771 | if (!PageUptodate(page)) { | |
85462323 ON |
1772 | error = lock_page_killable(page); |
1773 | if (unlikely(error)) | |
1774 | goto readpage_error; | |
1da177e4 LT |
1775 | if (!PageUptodate(page)) { |
1776 | if (page->mapping == NULL) { | |
1777 | /* | |
2ecdc82e | 1778 | * invalidate_mapping_pages got it |
1da177e4 LT |
1779 | */ |
1780 | unlock_page(page); | |
09cbfeaf | 1781 | put_page(page); |
1da177e4 LT |
1782 | goto find_page; |
1783 | } | |
1784 | unlock_page(page); | |
7ff81078 | 1785 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1786 | error = -EIO; |
1787 | goto readpage_error; | |
1da177e4 LT |
1788 | } |
1789 | unlock_page(page); | |
1790 | } | |
1791 | ||
1da177e4 LT |
1792 | goto page_ok; |
1793 | ||
1794 | readpage_error: | |
1795 | /* UHHUH! A synchronous read error occurred. Report it */ | |
09cbfeaf | 1796 | put_page(page); |
1da177e4 LT |
1797 | goto out; |
1798 | ||
1799 | no_cached_page: | |
1800 | /* | |
1801 | * Ok, it wasn't cached, so we need to create a new | |
1802 | * page.. | |
1803 | */ | |
eb2be189 NP |
1804 | page = page_cache_alloc_cold(mapping); |
1805 | if (!page) { | |
6e58e79d | 1806 | error = -ENOMEM; |
eb2be189 | 1807 | goto out; |
1da177e4 | 1808 | } |
6afdb859 | 1809 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 1810 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 1811 | if (error) { |
09cbfeaf | 1812 | put_page(page); |
6e58e79d AV |
1813 | if (error == -EEXIST) { |
1814 | error = 0; | |
1da177e4 | 1815 | goto find_page; |
6e58e79d | 1816 | } |
1da177e4 LT |
1817 | goto out; |
1818 | } | |
1da177e4 LT |
1819 | goto readpage; |
1820 | } | |
1821 | ||
1822 | out: | |
7ff81078 | 1823 | ra->prev_pos = prev_index; |
09cbfeaf | 1824 | ra->prev_pos <<= PAGE_SHIFT; |
7ff81078 | 1825 | ra->prev_pos |= prev_offset; |
1da177e4 | 1826 | |
09cbfeaf | 1827 | *ppos = ((loff_t)index << PAGE_SHIFT) + offset; |
0c6aa263 | 1828 | file_accessed(filp); |
6e58e79d | 1829 | return written ? written : error; |
1da177e4 LT |
1830 | } |
1831 | ||
485bb99b | 1832 | /** |
6abd2322 | 1833 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 1834 | * @iocb: kernel I/O control block |
6abd2322 | 1835 | * @iter: destination for the data read |
485bb99b | 1836 | * |
6abd2322 | 1837 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
1838 | * that can use the page cache directly. |
1839 | */ | |
1840 | ssize_t | |
ed978a81 | 1841 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 1842 | { |
ed978a81 | 1843 | struct file *file = iocb->ki_filp; |
cb66a7a1 | 1844 | ssize_t retval = 0; |
e7080a43 NS |
1845 | size_t count = iov_iter_count(iter); |
1846 | ||
1847 | if (!count) | |
1848 | goto out; /* skip atime */ | |
1da177e4 | 1849 | |
2ba48ce5 | 1850 | if (iocb->ki_flags & IOCB_DIRECT) { |
ed978a81 AV |
1851 | struct address_space *mapping = file->f_mapping; |
1852 | struct inode *inode = mapping->host; | |
543ade1f | 1853 | loff_t size; |
1da177e4 | 1854 | |
1da177e4 | 1855 | size = i_size_read(inode); |
c64fb5c7 CH |
1856 | retval = filemap_write_and_wait_range(mapping, iocb->ki_pos, |
1857 | iocb->ki_pos + count - 1); | |
9fe55eea | 1858 | if (!retval) { |
ed978a81 | 1859 | struct iov_iter data = *iter; |
c8b8e32d | 1860 | retval = mapping->a_ops->direct_IO(iocb, &data); |
9fe55eea | 1861 | } |
d8d3d94b | 1862 | |
9fe55eea | 1863 | if (retval > 0) { |
c64fb5c7 | 1864 | iocb->ki_pos += retval; |
ed978a81 | 1865 | iov_iter_advance(iter, retval); |
9fe55eea | 1866 | } |
66f998f6 | 1867 | |
9fe55eea SW |
1868 | /* |
1869 | * Btrfs can have a short DIO read if we encounter | |
1870 | * compressed extents, so if there was an error, or if | |
1871 | * we've already read everything we wanted to, or if | |
1872 | * there was a short read because we hit EOF, go ahead | |
1873 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
1874 | * the rest of the read. Buffered reads will not work for |
1875 | * DAX files, so don't bother trying. | |
9fe55eea | 1876 | */ |
c64fb5c7 | 1877 | if (retval < 0 || !iov_iter_count(iter) || iocb->ki_pos >= size || |
fbbbad4b | 1878 | IS_DAX(inode)) { |
ed978a81 | 1879 | file_accessed(file); |
9fe55eea | 1880 | goto out; |
0e0bcae3 | 1881 | } |
1da177e4 LT |
1882 | } |
1883 | ||
c64fb5c7 | 1884 | retval = do_generic_file_read(file, &iocb->ki_pos, iter, retval); |
1da177e4 LT |
1885 | out: |
1886 | return retval; | |
1887 | } | |
ed978a81 | 1888 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 1889 | |
1da177e4 | 1890 | #ifdef CONFIG_MMU |
485bb99b RD |
1891 | /** |
1892 | * page_cache_read - adds requested page to the page cache if not already there | |
1893 | * @file: file to read | |
1894 | * @offset: page index | |
62eb320a | 1895 | * @gfp_mask: memory allocation flags |
485bb99b | 1896 | * |
1da177e4 LT |
1897 | * This adds the requested page to the page cache if it isn't already there, |
1898 | * and schedules an I/O to read in its contents from disk. | |
1899 | */ | |
c20cd45e | 1900 | static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask) |
1da177e4 LT |
1901 | { |
1902 | struct address_space *mapping = file->f_mapping; | |
99dadfdd | 1903 | struct page *page; |
994fc28c | 1904 | int ret; |
1da177e4 | 1905 | |
994fc28c | 1906 | do { |
c20cd45e | 1907 | page = __page_cache_alloc(gfp_mask|__GFP_COLD); |
994fc28c ZB |
1908 | if (!page) |
1909 | return -ENOMEM; | |
1910 | ||
c20cd45e | 1911 | ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL); |
994fc28c ZB |
1912 | if (ret == 0) |
1913 | ret = mapping->a_ops->readpage(file, page); | |
1914 | else if (ret == -EEXIST) | |
1915 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1916 | |
09cbfeaf | 1917 | put_page(page); |
1da177e4 | 1918 | |
994fc28c | 1919 | } while (ret == AOP_TRUNCATED_PAGE); |
99dadfdd | 1920 | |
994fc28c | 1921 | return ret; |
1da177e4 LT |
1922 | } |
1923 | ||
1924 | #define MMAP_LOTSAMISS (100) | |
1925 | ||
ef00e08e LT |
1926 | /* |
1927 | * Synchronous readahead happens when we don't even find | |
1928 | * a page in the page cache at all. | |
1929 | */ | |
1930 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1931 | struct file_ra_state *ra, | |
1932 | struct file *file, | |
1933 | pgoff_t offset) | |
1934 | { | |
ef00e08e LT |
1935 | struct address_space *mapping = file->f_mapping; |
1936 | ||
1937 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1938 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 1939 | return; |
275b12bf WF |
1940 | if (!ra->ra_pages) |
1941 | return; | |
ef00e08e | 1942 | |
64363aad | 1943 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
1944 | page_cache_sync_readahead(mapping, ra, file, offset, |
1945 | ra->ra_pages); | |
ef00e08e LT |
1946 | return; |
1947 | } | |
1948 | ||
207d04ba AK |
1949 | /* Avoid banging the cache line if not needed */ |
1950 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1951 | ra->mmap_miss++; |
1952 | ||
1953 | /* | |
1954 | * Do we miss much more than hit in this file? If so, | |
1955 | * stop bothering with read-ahead. It will only hurt. | |
1956 | */ | |
1957 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1958 | return; | |
1959 | ||
d30a1100 WF |
1960 | /* |
1961 | * mmap read-around | |
1962 | */ | |
600e19af RG |
1963 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
1964 | ra->size = ra->ra_pages; | |
1965 | ra->async_size = ra->ra_pages / 4; | |
275b12bf | 1966 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1967 | } |
1968 | ||
1969 | /* | |
1970 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1971 | * so we want to possibly extend the readahead further.. | |
1972 | */ | |
1973 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1974 | struct file_ra_state *ra, | |
1975 | struct file *file, | |
1976 | struct page *page, | |
1977 | pgoff_t offset) | |
1978 | { | |
1979 | struct address_space *mapping = file->f_mapping; | |
1980 | ||
1981 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1982 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
1983 | return; |
1984 | if (ra->mmap_miss > 0) | |
1985 | ra->mmap_miss--; | |
1986 | if (PageReadahead(page)) | |
2fad6f5d WF |
1987 | page_cache_async_readahead(mapping, ra, file, |
1988 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1989 | } |
1990 | ||
485bb99b | 1991 | /** |
54cb8821 | 1992 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1993 | * @vma: vma in which the fault was taken |
1994 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 1995 | * |
54cb8821 | 1996 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
1997 | * mapped memory region to read in file data during a page fault. |
1998 | * | |
1999 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2000 | * it in the page cache, and handles the special cases reasonably without | |
2001 | * having a lot of duplicated code. | |
9a95f3cf PC |
2002 | * |
2003 | * vma->vm_mm->mmap_sem must be held on entry. | |
2004 | * | |
2005 | * If our return value has VM_FAULT_RETRY set, it's because | |
2006 | * lock_page_or_retry() returned 0. | |
2007 | * The mmap_sem has usually been released in this case. | |
2008 | * See __lock_page_or_retry() for the exception. | |
2009 | * | |
2010 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
2011 | * has not been released. | |
2012 | * | |
2013 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
1da177e4 | 2014 | */ |
d0217ac0 | 2015 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
2016 | { |
2017 | int error; | |
54cb8821 | 2018 | struct file *file = vma->vm_file; |
1da177e4 LT |
2019 | struct address_space *mapping = file->f_mapping; |
2020 | struct file_ra_state *ra = &file->f_ra; | |
2021 | struct inode *inode = mapping->host; | |
ef00e08e | 2022 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 2023 | struct page *page; |
99e3e53f | 2024 | loff_t size; |
83c54070 | 2025 | int ret = 0; |
1da177e4 | 2026 | |
09cbfeaf KS |
2027 | size = round_up(i_size_read(inode), PAGE_SIZE); |
2028 | if (offset >= size >> PAGE_SHIFT) | |
5307cc1a | 2029 | return VM_FAULT_SIGBUS; |
1da177e4 | 2030 | |
1da177e4 | 2031 | /* |
49426420 | 2032 | * Do we have something in the page cache already? |
1da177e4 | 2033 | */ |
ef00e08e | 2034 | page = find_get_page(mapping, offset); |
45cac65b | 2035 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2036 | /* |
ef00e08e LT |
2037 | * We found the page, so try async readahead before |
2038 | * waiting for the lock. | |
1da177e4 | 2039 | */ |
ef00e08e | 2040 | do_async_mmap_readahead(vma, ra, file, page, offset); |
45cac65b | 2041 | } else if (!page) { |
ef00e08e LT |
2042 | /* No page in the page cache at all */ |
2043 | do_sync_mmap_readahead(vma, ra, file, offset); | |
2044 | count_vm_event(PGMAJFAULT); | |
456f998e | 2045 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
2046 | ret = VM_FAULT_MAJOR; |
2047 | retry_find: | |
b522c94d | 2048 | page = find_get_page(mapping, offset); |
1da177e4 LT |
2049 | if (!page) |
2050 | goto no_cached_page; | |
2051 | } | |
2052 | ||
d88c0922 | 2053 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
09cbfeaf | 2054 | put_page(page); |
d065bd81 | 2055 | return ret | VM_FAULT_RETRY; |
d88c0922 | 2056 | } |
b522c94d ML |
2057 | |
2058 | /* Did it get truncated? */ | |
2059 | if (unlikely(page->mapping != mapping)) { | |
2060 | unlock_page(page); | |
2061 | put_page(page); | |
2062 | goto retry_find; | |
2063 | } | |
309381fe | 2064 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 2065 | |
1da177e4 | 2066 | /* |
d00806b1 NP |
2067 | * We have a locked page in the page cache, now we need to check |
2068 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2069 | */ |
d00806b1 | 2070 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2071 | goto page_not_uptodate; |
2072 | ||
ef00e08e LT |
2073 | /* |
2074 | * Found the page and have a reference on it. | |
2075 | * We must recheck i_size under page lock. | |
2076 | */ | |
09cbfeaf KS |
2077 | size = round_up(i_size_read(inode), PAGE_SIZE); |
2078 | if (unlikely(offset >= size >> PAGE_SHIFT)) { | |
d00806b1 | 2079 | unlock_page(page); |
09cbfeaf | 2080 | put_page(page); |
5307cc1a | 2081 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2082 | } |
2083 | ||
d0217ac0 | 2084 | vmf->page = page; |
83c54070 | 2085 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2086 | |
1da177e4 LT |
2087 | no_cached_page: |
2088 | /* | |
2089 | * We're only likely to ever get here if MADV_RANDOM is in | |
2090 | * effect. | |
2091 | */ | |
c20cd45e | 2092 | error = page_cache_read(file, offset, vmf->gfp_mask); |
1da177e4 LT |
2093 | |
2094 | /* | |
2095 | * The page we want has now been added to the page cache. | |
2096 | * In the unlikely event that someone removed it in the | |
2097 | * meantime, we'll just come back here and read it again. | |
2098 | */ | |
2099 | if (error >= 0) | |
2100 | goto retry_find; | |
2101 | ||
2102 | /* | |
2103 | * An error return from page_cache_read can result if the | |
2104 | * system is low on memory, or a problem occurs while trying | |
2105 | * to schedule I/O. | |
2106 | */ | |
2107 | if (error == -ENOMEM) | |
d0217ac0 NP |
2108 | return VM_FAULT_OOM; |
2109 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
2110 | |
2111 | page_not_uptodate: | |
1da177e4 LT |
2112 | /* |
2113 | * Umm, take care of errors if the page isn't up-to-date. | |
2114 | * Try to re-read it _once_. We do this synchronously, | |
2115 | * because there really aren't any performance issues here | |
2116 | * and we need to check for errors. | |
2117 | */ | |
1da177e4 | 2118 | ClearPageError(page); |
994fc28c | 2119 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2120 | if (!error) { |
2121 | wait_on_page_locked(page); | |
2122 | if (!PageUptodate(page)) | |
2123 | error = -EIO; | |
2124 | } | |
09cbfeaf | 2125 | put_page(page); |
d00806b1 NP |
2126 | |
2127 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2128 | goto retry_find; |
1da177e4 | 2129 | |
d00806b1 | 2130 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 2131 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 2132 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
2133 | } |
2134 | EXPORT_SYMBOL(filemap_fault); | |
2135 | ||
f1820361 KS |
2136 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
2137 | { | |
2138 | struct radix_tree_iter iter; | |
2139 | void **slot; | |
2140 | struct file *file = vma->vm_file; | |
2141 | struct address_space *mapping = file->f_mapping; | |
2142 | loff_t size; | |
2143 | struct page *page; | |
2144 | unsigned long address = (unsigned long) vmf->virtual_address; | |
2145 | unsigned long addr; | |
2146 | pte_t *pte; | |
2147 | ||
2148 | rcu_read_lock(); | |
2149 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) { | |
2150 | if (iter.index > vmf->max_pgoff) | |
2151 | break; | |
2152 | repeat: | |
2153 | page = radix_tree_deref_slot(slot); | |
2154 | if (unlikely(!page)) | |
2155 | goto next; | |
2156 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2157 | if (radix_tree_deref_retry(page)) { |
2158 | slot = radix_tree_iter_retry(&iter); | |
2159 | continue; | |
2160 | } | |
2161 | goto next; | |
f1820361 KS |
2162 | } |
2163 | ||
2164 | if (!page_cache_get_speculative(page)) | |
2165 | goto repeat; | |
2166 | ||
2167 | /* Has the page moved? */ | |
2168 | if (unlikely(page != *slot)) { | |
09cbfeaf | 2169 | put_page(page); |
f1820361 KS |
2170 | goto repeat; |
2171 | } | |
2172 | ||
2173 | if (!PageUptodate(page) || | |
2174 | PageReadahead(page) || | |
2175 | PageHWPoison(page)) | |
2176 | goto skip; | |
2177 | if (!trylock_page(page)) | |
2178 | goto skip; | |
2179 | ||
2180 | if (page->mapping != mapping || !PageUptodate(page)) | |
2181 | goto unlock; | |
2182 | ||
09cbfeaf KS |
2183 | size = round_up(i_size_read(mapping->host), PAGE_SIZE); |
2184 | if (page->index >= size >> PAGE_SHIFT) | |
f1820361 KS |
2185 | goto unlock; |
2186 | ||
2187 | pte = vmf->pte + page->index - vmf->pgoff; | |
2188 | if (!pte_none(*pte)) | |
2189 | goto unlock; | |
2190 | ||
2191 | if (file->f_ra.mmap_miss > 0) | |
2192 | file->f_ra.mmap_miss--; | |
2193 | addr = address + (page->index - vmf->pgoff) * PAGE_SIZE; | |
2194 | do_set_pte(vma, addr, page, pte, false, false); | |
2195 | unlock_page(page); | |
2196 | goto next; | |
2197 | unlock: | |
2198 | unlock_page(page); | |
2199 | skip: | |
09cbfeaf | 2200 | put_page(page); |
f1820361 KS |
2201 | next: |
2202 | if (iter.index == vmf->max_pgoff) | |
2203 | break; | |
2204 | } | |
2205 | rcu_read_unlock(); | |
2206 | } | |
2207 | EXPORT_SYMBOL(filemap_map_pages); | |
2208 | ||
4fcf1c62 JK |
2209 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2210 | { | |
2211 | struct page *page = vmf->page; | |
496ad9aa | 2212 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
2213 | int ret = VM_FAULT_LOCKED; |
2214 | ||
14da9200 | 2215 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
2216 | file_update_time(vma->vm_file); |
2217 | lock_page(page); | |
2218 | if (page->mapping != inode->i_mapping) { | |
2219 | unlock_page(page); | |
2220 | ret = VM_FAULT_NOPAGE; | |
2221 | goto out; | |
2222 | } | |
14da9200 JK |
2223 | /* |
2224 | * We mark the page dirty already here so that when freeze is in | |
2225 | * progress, we are guaranteed that writeback during freezing will | |
2226 | * see the dirty page and writeprotect it again. | |
2227 | */ | |
2228 | set_page_dirty(page); | |
1d1d1a76 | 2229 | wait_for_stable_page(page); |
4fcf1c62 | 2230 | out: |
14da9200 | 2231 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2232 | return ret; |
2233 | } | |
2234 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2235 | ||
f0f37e2f | 2236 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2237 | .fault = filemap_fault, |
f1820361 | 2238 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2239 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2240 | }; |
2241 | ||
2242 | /* This is used for a general mmap of a disk file */ | |
2243 | ||
2244 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2245 | { | |
2246 | struct address_space *mapping = file->f_mapping; | |
2247 | ||
2248 | if (!mapping->a_ops->readpage) | |
2249 | return -ENOEXEC; | |
2250 | file_accessed(file); | |
2251 | vma->vm_ops = &generic_file_vm_ops; | |
2252 | return 0; | |
2253 | } | |
1da177e4 LT |
2254 | |
2255 | /* | |
2256 | * This is for filesystems which do not implement ->writepage. | |
2257 | */ | |
2258 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2259 | { | |
2260 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2261 | return -EINVAL; | |
2262 | return generic_file_mmap(file, vma); | |
2263 | } | |
2264 | #else | |
2265 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2266 | { | |
2267 | return -ENOSYS; | |
2268 | } | |
2269 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2270 | { | |
2271 | return -ENOSYS; | |
2272 | } | |
2273 | #endif /* CONFIG_MMU */ | |
2274 | ||
2275 | EXPORT_SYMBOL(generic_file_mmap); | |
2276 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2277 | ||
67f9fd91 SL |
2278 | static struct page *wait_on_page_read(struct page *page) |
2279 | { | |
2280 | if (!IS_ERR(page)) { | |
2281 | wait_on_page_locked(page); | |
2282 | if (!PageUptodate(page)) { | |
09cbfeaf | 2283 | put_page(page); |
67f9fd91 SL |
2284 | page = ERR_PTR(-EIO); |
2285 | } | |
2286 | } | |
2287 | return page; | |
2288 | } | |
2289 | ||
32b63529 | 2290 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2291 | pgoff_t index, |
5e5358e7 | 2292 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2293 | void *data, |
2294 | gfp_t gfp) | |
1da177e4 | 2295 | { |
eb2be189 | 2296 | struct page *page; |
1da177e4 LT |
2297 | int err; |
2298 | repeat: | |
2299 | page = find_get_page(mapping, index); | |
2300 | if (!page) { | |
0531b2aa | 2301 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
2302 | if (!page) |
2303 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2304 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 2305 | if (unlikely(err)) { |
09cbfeaf | 2306 | put_page(page); |
eb2be189 NP |
2307 | if (err == -EEXIST) |
2308 | goto repeat; | |
1da177e4 | 2309 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2310 | return ERR_PTR(err); |
2311 | } | |
32b63529 MG |
2312 | |
2313 | filler: | |
1da177e4 LT |
2314 | err = filler(data, page); |
2315 | if (err < 0) { | |
09cbfeaf | 2316 | put_page(page); |
32b63529 | 2317 | return ERR_PTR(err); |
1da177e4 | 2318 | } |
1da177e4 | 2319 | |
32b63529 MG |
2320 | page = wait_on_page_read(page); |
2321 | if (IS_ERR(page)) | |
2322 | return page; | |
2323 | goto out; | |
2324 | } | |
1da177e4 LT |
2325 | if (PageUptodate(page)) |
2326 | goto out; | |
2327 | ||
ebded027 MG |
2328 | /* |
2329 | * Page is not up to date and may be locked due one of the following | |
2330 | * case a: Page is being filled and the page lock is held | |
2331 | * case b: Read/write error clearing the page uptodate status | |
2332 | * case c: Truncation in progress (page locked) | |
2333 | * case d: Reclaim in progress | |
2334 | * | |
2335 | * Case a, the page will be up to date when the page is unlocked. | |
2336 | * There is no need to serialise on the page lock here as the page | |
2337 | * is pinned so the lock gives no additional protection. Even if the | |
2338 | * the page is truncated, the data is still valid if PageUptodate as | |
2339 | * it's a race vs truncate race. | |
2340 | * Case b, the page will not be up to date | |
2341 | * Case c, the page may be truncated but in itself, the data may still | |
2342 | * be valid after IO completes as it's a read vs truncate race. The | |
2343 | * operation must restart if the page is not uptodate on unlock but | |
2344 | * otherwise serialising on page lock to stabilise the mapping gives | |
2345 | * no additional guarantees to the caller as the page lock is | |
2346 | * released before return. | |
2347 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
2348 | * will be a race with remove_mapping that determines if the mapping | |
2349 | * is valid on unlock but otherwise the data is valid and there is | |
2350 | * no need to serialise with page lock. | |
2351 | * | |
2352 | * As the page lock gives no additional guarantee, we optimistically | |
2353 | * wait on the page to be unlocked and check if it's up to date and | |
2354 | * use the page if it is. Otherwise, the page lock is required to | |
2355 | * distinguish between the different cases. The motivation is that we | |
2356 | * avoid spurious serialisations and wakeups when multiple processes | |
2357 | * wait on the same page for IO to complete. | |
2358 | */ | |
2359 | wait_on_page_locked(page); | |
2360 | if (PageUptodate(page)) | |
2361 | goto out; | |
2362 | ||
2363 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 2364 | lock_page(page); |
ebded027 MG |
2365 | |
2366 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
2367 | if (!page->mapping) { |
2368 | unlock_page(page); | |
09cbfeaf | 2369 | put_page(page); |
32b63529 | 2370 | goto repeat; |
1da177e4 | 2371 | } |
ebded027 MG |
2372 | |
2373 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
2374 | if (PageUptodate(page)) { |
2375 | unlock_page(page); | |
2376 | goto out; | |
2377 | } | |
32b63529 MG |
2378 | goto filler; |
2379 | ||
c855ff37 | 2380 | out: |
6fe6900e NP |
2381 | mark_page_accessed(page); |
2382 | return page; | |
2383 | } | |
0531b2aa LT |
2384 | |
2385 | /** | |
67f9fd91 | 2386 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2387 | * @mapping: the page's address_space |
2388 | * @index: the page index | |
2389 | * @filler: function to perform the read | |
5e5358e7 | 2390 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2391 | * |
0531b2aa | 2392 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2393 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2394 | * |
2395 | * If the page does not get brought uptodate, return -EIO. | |
2396 | */ | |
67f9fd91 | 2397 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2398 | pgoff_t index, |
5e5358e7 | 2399 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2400 | void *data) |
2401 | { | |
2402 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2403 | } | |
67f9fd91 | 2404 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2405 | |
2406 | /** | |
2407 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2408 | * @mapping: the page's address_space | |
2409 | * @index: the page index | |
2410 | * @gfp: the page allocator flags to use if allocating | |
2411 | * | |
2412 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2413 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2414 | * |
2415 | * If the page does not get brought uptodate, return -EIO. | |
2416 | */ | |
2417 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2418 | pgoff_t index, | |
2419 | gfp_t gfp) | |
2420 | { | |
2421 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2422 | ||
67f9fd91 | 2423 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2424 | } |
2425 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2426 | ||
1da177e4 LT |
2427 | /* |
2428 | * Performs necessary checks before doing a write | |
2429 | * | |
485bb99b | 2430 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2431 | * Returns appropriate error code that caller should return or |
2432 | * zero in case that write should be allowed. | |
2433 | */ | |
3309dd04 | 2434 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2435 | { |
3309dd04 | 2436 | struct file *file = iocb->ki_filp; |
1da177e4 | 2437 | struct inode *inode = file->f_mapping->host; |
59e99e5b | 2438 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
3309dd04 | 2439 | loff_t pos; |
1da177e4 | 2440 | |
3309dd04 AV |
2441 | if (!iov_iter_count(from)) |
2442 | return 0; | |
1da177e4 | 2443 | |
0fa6b005 | 2444 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2445 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2446 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2447 | |
3309dd04 | 2448 | pos = iocb->ki_pos; |
1da177e4 | 2449 | |
0fa6b005 | 2450 | if (limit != RLIM_INFINITY) { |
3309dd04 | 2451 | if (iocb->ki_pos >= limit) { |
0fa6b005 AV |
2452 | send_sig(SIGXFSZ, current, 0); |
2453 | return -EFBIG; | |
1da177e4 | 2454 | } |
3309dd04 | 2455 | iov_iter_truncate(from, limit - (unsigned long)pos); |
1da177e4 LT |
2456 | } |
2457 | ||
2458 | /* | |
2459 | * LFS rule | |
2460 | */ | |
3309dd04 | 2461 | if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS && |
1da177e4 | 2462 | !(file->f_flags & O_LARGEFILE))) { |
3309dd04 | 2463 | if (pos >= MAX_NON_LFS) |
1da177e4 | 2464 | return -EFBIG; |
3309dd04 | 2465 | iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos); |
1da177e4 LT |
2466 | } |
2467 | ||
2468 | /* | |
2469 | * Are we about to exceed the fs block limit ? | |
2470 | * | |
2471 | * If we have written data it becomes a short write. If we have | |
2472 | * exceeded without writing data we send a signal and return EFBIG. | |
2473 | * Linus frestrict idea will clean these up nicely.. | |
2474 | */ | |
3309dd04 AV |
2475 | if (unlikely(pos >= inode->i_sb->s_maxbytes)) |
2476 | return -EFBIG; | |
1da177e4 | 2477 | |
3309dd04 AV |
2478 | iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos); |
2479 | return iov_iter_count(from); | |
1da177e4 LT |
2480 | } |
2481 | EXPORT_SYMBOL(generic_write_checks); | |
2482 | ||
afddba49 NP |
2483 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2484 | loff_t pos, unsigned len, unsigned flags, | |
2485 | struct page **pagep, void **fsdata) | |
2486 | { | |
2487 | const struct address_space_operations *aops = mapping->a_ops; | |
2488 | ||
4e02ed4b | 2489 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2490 | pagep, fsdata); |
afddba49 NP |
2491 | } |
2492 | EXPORT_SYMBOL(pagecache_write_begin); | |
2493 | ||
2494 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2495 | loff_t pos, unsigned len, unsigned copied, | |
2496 | struct page *page, void *fsdata) | |
2497 | { | |
2498 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2499 | |
4e02ed4b | 2500 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2501 | } |
2502 | EXPORT_SYMBOL(pagecache_write_end); | |
2503 | ||
1da177e4 | 2504 | ssize_t |
1af5bb49 | 2505 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2506 | { |
2507 | struct file *file = iocb->ki_filp; | |
2508 | struct address_space *mapping = file->f_mapping; | |
2509 | struct inode *inode = mapping->host; | |
1af5bb49 | 2510 | loff_t pos = iocb->ki_pos; |
1da177e4 | 2511 | ssize_t written; |
a969e903 CH |
2512 | size_t write_len; |
2513 | pgoff_t end; | |
26978b8b | 2514 | struct iov_iter data; |
1da177e4 | 2515 | |
0c949334 | 2516 | write_len = iov_iter_count(from); |
09cbfeaf | 2517 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 2518 | |
48b47c56 | 2519 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2520 | if (written) |
2521 | goto out; | |
2522 | ||
2523 | /* | |
2524 | * After a write we want buffered reads to be sure to go to disk to get | |
2525 | * the new data. We invalidate clean cached page from the region we're | |
2526 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2527 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2528 | */ |
2529 | if (mapping->nrpages) { | |
2530 | written = invalidate_inode_pages2_range(mapping, | |
09cbfeaf | 2531 | pos >> PAGE_SHIFT, end); |
6ccfa806 HH |
2532 | /* |
2533 | * If a page can not be invalidated, return 0 to fall back | |
2534 | * to buffered write. | |
2535 | */ | |
2536 | if (written) { | |
2537 | if (written == -EBUSY) | |
2538 | return 0; | |
a969e903 | 2539 | goto out; |
6ccfa806 | 2540 | } |
a969e903 CH |
2541 | } |
2542 | ||
26978b8b | 2543 | data = *from; |
c8b8e32d | 2544 | written = mapping->a_ops->direct_IO(iocb, &data); |
a969e903 CH |
2545 | |
2546 | /* | |
2547 | * Finally, try again to invalidate clean pages which might have been | |
2548 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2549 | * if the source of the write was an mmap'ed region of the file | |
2550 | * we're writing. Either one is a pretty crazy thing to do, | |
2551 | * so we don't support it 100%. If this invalidation | |
2552 | * fails, tough, the write still worked... | |
2553 | */ | |
2554 | if (mapping->nrpages) { | |
2555 | invalidate_inode_pages2_range(mapping, | |
09cbfeaf | 2556 | pos >> PAGE_SHIFT, end); |
a969e903 CH |
2557 | } |
2558 | ||
1da177e4 | 2559 | if (written > 0) { |
0116651c | 2560 | pos += written; |
f8579f86 | 2561 | iov_iter_advance(from, written); |
0116651c NK |
2562 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
2563 | i_size_write(inode, pos); | |
1da177e4 LT |
2564 | mark_inode_dirty(inode); |
2565 | } | |
5cb6c6c7 | 2566 | iocb->ki_pos = pos; |
1da177e4 | 2567 | } |
a969e903 | 2568 | out: |
1da177e4 LT |
2569 | return written; |
2570 | } | |
2571 | EXPORT_SYMBOL(generic_file_direct_write); | |
2572 | ||
eb2be189 NP |
2573 | /* |
2574 | * Find or create a page at the given pagecache position. Return the locked | |
2575 | * page. This function is specifically for buffered writes. | |
2576 | */ | |
54566b2c NP |
2577 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2578 | pgoff_t index, unsigned flags) | |
eb2be189 | 2579 | { |
eb2be189 | 2580 | struct page *page; |
bbddabe2 | 2581 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 2582 | |
54566b2c | 2583 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
2584 | fgp_flags |= FGP_NOFS; |
2585 | ||
2586 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 2587 | mapping_gfp_mask(mapping)); |
c585a267 | 2588 | if (page) |
2457aec6 | 2589 | wait_for_stable_page(page); |
eb2be189 | 2590 | |
eb2be189 NP |
2591 | return page; |
2592 | } | |
54566b2c | 2593 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2594 | |
3b93f911 | 2595 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
2596 | struct iov_iter *i, loff_t pos) |
2597 | { | |
2598 | struct address_space *mapping = file->f_mapping; | |
2599 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2600 | long status = 0; | |
2601 | ssize_t written = 0; | |
674b892e NP |
2602 | unsigned int flags = 0; |
2603 | ||
2604 | /* | |
2605 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2606 | */ | |
777eda2c | 2607 | if (!iter_is_iovec(i)) |
674b892e | 2608 | flags |= AOP_FLAG_UNINTERRUPTIBLE; |
afddba49 NP |
2609 | |
2610 | do { | |
2611 | struct page *page; | |
afddba49 NP |
2612 | unsigned long offset; /* Offset into pagecache page */ |
2613 | unsigned long bytes; /* Bytes to write to page */ | |
2614 | size_t copied; /* Bytes copied from user */ | |
2615 | void *fsdata; | |
2616 | ||
09cbfeaf KS |
2617 | offset = (pos & (PAGE_SIZE - 1)); |
2618 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
2619 | iov_iter_count(i)); |
2620 | ||
2621 | again: | |
00a3d660 LT |
2622 | /* |
2623 | * Bring in the user page that we will copy from _first_. | |
2624 | * Otherwise there's a nasty deadlock on copying from the | |
2625 | * same page as we're writing to, without it being marked | |
2626 | * up-to-date. | |
2627 | * | |
2628 | * Not only is this an optimisation, but it is also required | |
2629 | * to check that the address is actually valid, when atomic | |
2630 | * usercopies are used, below. | |
2631 | */ | |
2632 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2633 | status = -EFAULT; | |
2634 | break; | |
2635 | } | |
2636 | ||
296291cd JK |
2637 | if (fatal_signal_pending(current)) { |
2638 | status = -EINTR; | |
2639 | break; | |
2640 | } | |
2641 | ||
674b892e | 2642 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 2643 | &page, &fsdata); |
2457aec6 | 2644 | if (unlikely(status < 0)) |
afddba49 NP |
2645 | break; |
2646 | ||
931e80e4 | 2647 | if (mapping_writably_mapped(mapping)) |
2648 | flush_dcache_page(page); | |
00a3d660 | 2649 | |
afddba49 | 2650 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
2651 | flush_dcache_page(page); |
2652 | ||
2653 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
2654 | page, fsdata); | |
2655 | if (unlikely(status < 0)) | |
2656 | break; | |
2657 | copied = status; | |
2658 | ||
2659 | cond_resched(); | |
2660 | ||
124d3b70 | 2661 | iov_iter_advance(i, copied); |
afddba49 NP |
2662 | if (unlikely(copied == 0)) { |
2663 | /* | |
2664 | * If we were unable to copy any data at all, we must | |
2665 | * fall back to a single segment length write. | |
2666 | * | |
2667 | * If we didn't fallback here, we could livelock | |
2668 | * because not all segments in the iov can be copied at | |
2669 | * once without a pagefault. | |
2670 | */ | |
09cbfeaf | 2671 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
afddba49 NP |
2672 | iov_iter_single_seg_count(i)); |
2673 | goto again; | |
2674 | } | |
afddba49 NP |
2675 | pos += copied; |
2676 | written += copied; | |
2677 | ||
2678 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
2679 | } while (iov_iter_count(i)); |
2680 | ||
2681 | return written ? written : status; | |
2682 | } | |
3b93f911 | 2683 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 2684 | |
e4dd9de3 | 2685 | /** |
8174202b | 2686 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 2687 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 2688 | * @from: iov_iter with data to write |
e4dd9de3 JK |
2689 | * |
2690 | * This function does all the work needed for actually writing data to a | |
2691 | * file. It does all basic checks, removes SUID from the file, updates | |
2692 | * modification times and calls proper subroutines depending on whether we | |
2693 | * do direct IO or a standard buffered write. | |
2694 | * | |
2695 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2696 | * object which does not need locking at all. | |
2697 | * | |
2698 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2699 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2700 | * avoid syncing under i_mutex. | |
2701 | */ | |
8174202b | 2702 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2703 | { |
2704 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2705 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 2706 | struct inode *inode = mapping->host; |
3b93f911 | 2707 | ssize_t written = 0; |
1da177e4 | 2708 | ssize_t err; |
3b93f911 | 2709 | ssize_t status; |
1da177e4 | 2710 | |
1da177e4 | 2711 | /* We can write back this queue in page reclaim */ |
de1414a6 | 2712 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 2713 | err = file_remove_privs(file); |
1da177e4 LT |
2714 | if (err) |
2715 | goto out; | |
2716 | ||
c3b2da31 JB |
2717 | err = file_update_time(file); |
2718 | if (err) | |
2719 | goto out; | |
1da177e4 | 2720 | |
2ba48ce5 | 2721 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 2722 | loff_t pos, endbyte; |
fb5527e6 | 2723 | |
1af5bb49 | 2724 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 2725 | /* |
fbbbad4b MW |
2726 | * If the write stopped short of completing, fall back to |
2727 | * buffered writes. Some filesystems do this for writes to | |
2728 | * holes, for example. For DAX files, a buffered write will | |
2729 | * not succeed (even if it did, DAX does not handle dirty | |
2730 | * page-cache pages correctly). | |
1da177e4 | 2731 | */ |
0b8def9d | 2732 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
2733 | goto out; |
2734 | ||
0b8def9d | 2735 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 2736 | /* |
3b93f911 | 2737 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
2738 | * then we want to return the number of bytes which were |
2739 | * direct-written, or the error code if that was zero. Note | |
2740 | * that this differs from normal direct-io semantics, which | |
2741 | * will return -EFOO even if some bytes were written. | |
2742 | */ | |
60bb4529 | 2743 | if (unlikely(status < 0)) { |
3b93f911 | 2744 | err = status; |
fb5527e6 JM |
2745 | goto out; |
2746 | } | |
fb5527e6 JM |
2747 | /* |
2748 | * We need to ensure that the page cache pages are written to | |
2749 | * disk and invalidated to preserve the expected O_DIRECT | |
2750 | * semantics. | |
2751 | */ | |
3b93f911 | 2752 | endbyte = pos + status - 1; |
0b8def9d | 2753 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 2754 | if (err == 0) { |
0b8def9d | 2755 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 2756 | written += status; |
fb5527e6 | 2757 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
2758 | pos >> PAGE_SHIFT, |
2759 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
2760 | } else { |
2761 | /* | |
2762 | * We don't know how much we wrote, so just return | |
2763 | * the number of bytes which were direct-written | |
2764 | */ | |
2765 | } | |
2766 | } else { | |
0b8def9d AV |
2767 | written = generic_perform_write(file, from, iocb->ki_pos); |
2768 | if (likely(written > 0)) | |
2769 | iocb->ki_pos += written; | |
fb5527e6 | 2770 | } |
1da177e4 LT |
2771 | out: |
2772 | current->backing_dev_info = NULL; | |
2773 | return written ? written : err; | |
2774 | } | |
8174202b | 2775 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 2776 | |
e4dd9de3 | 2777 | /** |
8174202b | 2778 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 2779 | * @iocb: IO state structure |
8174202b | 2780 | * @from: iov_iter with data to write |
e4dd9de3 | 2781 | * |
8174202b | 2782 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
2783 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
2784 | * and acquires i_mutex as needed. | |
2785 | */ | |
8174202b | 2786 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2787 | { |
2788 | struct file *file = iocb->ki_filp; | |
148f948b | 2789 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2790 | ssize_t ret; |
1da177e4 | 2791 | |
5955102c | 2792 | inode_lock(inode); |
3309dd04 AV |
2793 | ret = generic_write_checks(iocb, from); |
2794 | if (ret > 0) | |
5f380c7f | 2795 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 2796 | inode_unlock(inode); |
1da177e4 | 2797 | |
e2592217 CH |
2798 | if (ret > 0) |
2799 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
2800 | return ret; |
2801 | } | |
8174202b | 2802 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 2803 | |
cf9a2ae8 DH |
2804 | /** |
2805 | * try_to_release_page() - release old fs-specific metadata on a page | |
2806 | * | |
2807 | * @page: the page which the kernel is trying to free | |
2808 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2809 | * | |
2810 | * The address_space is to try to release any data against the page | |
2811 | * (presumably at page->private). If the release was successful, return `1'. | |
2812 | * Otherwise return zero. | |
2813 | * | |
266cf658 DH |
2814 | * This may also be called if PG_fscache is set on a page, indicating that the |
2815 | * page is known to the local caching routines. | |
2816 | * | |
cf9a2ae8 | 2817 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 2818 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 2819 | * |
cf9a2ae8 DH |
2820 | */ |
2821 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2822 | { | |
2823 | struct address_space * const mapping = page->mapping; | |
2824 | ||
2825 | BUG_ON(!PageLocked(page)); | |
2826 | if (PageWriteback(page)) | |
2827 | return 0; | |
2828 | ||
2829 | if (mapping && mapping->a_ops->releasepage) | |
2830 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2831 | return try_to_free_buffers(page); | |
2832 | } | |
2833 | ||
2834 | EXPORT_SYMBOL(try_to_release_page); |