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); | |
216 | atomic_sub(mapcount, &page->_count); | |
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 | 599 | |
f9fe48be | 600 | mapping->nrexceptional--; |
4f622938 JK |
601 | if (!dax_mapping(mapping)) { |
602 | if (shadowp) | |
603 | *shadowp = p; | |
604 | if (node) | |
605 | workingset_node_shadows_dec(node); | |
606 | } else { | |
607 | /* DAX can replace empty locked entry with a hole */ | |
608 | WARN_ON_ONCE(p != | |
609 | (void *)(RADIX_TREE_EXCEPTIONAL_ENTRY | | |
610 | RADIX_DAX_ENTRY_LOCK)); | |
611 | /* DAX accounts exceptional entries as normal pages */ | |
612 | if (node) | |
613 | workingset_node_pages_dec(node); | |
614 | } | |
0cd6144a | 615 | } |
449dd698 JW |
616 | radix_tree_replace_slot(slot, page); |
617 | mapping->nrpages++; | |
618 | if (node) { | |
619 | workingset_node_pages_inc(node); | |
620 | /* | |
621 | * Don't track node that contains actual pages. | |
622 | * | |
623 | * Avoid acquiring the list_lru lock if already | |
624 | * untracked. The list_empty() test is safe as | |
625 | * node->private_list is protected by | |
626 | * mapping->tree_lock. | |
627 | */ | |
628 | if (!list_empty(&node->private_list)) | |
629 | list_lru_del(&workingset_shadow_nodes, | |
630 | &node->private_list); | |
631 | } | |
632 | return 0; | |
0cd6144a JW |
633 | } |
634 | ||
a528910e JW |
635 | static int __add_to_page_cache_locked(struct page *page, |
636 | struct address_space *mapping, | |
637 | pgoff_t offset, gfp_t gfp_mask, | |
638 | void **shadowp) | |
1da177e4 | 639 | { |
00501b53 JW |
640 | int huge = PageHuge(page); |
641 | struct mem_cgroup *memcg; | |
e286781d NP |
642 | int error; |
643 | ||
309381fe SL |
644 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
645 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d | 646 | |
00501b53 JW |
647 | if (!huge) { |
648 | error = mem_cgroup_try_charge(page, current->mm, | |
f627c2f5 | 649 | gfp_mask, &memcg, false); |
00501b53 JW |
650 | if (error) |
651 | return error; | |
652 | } | |
1da177e4 | 653 | |
5e4c0d97 | 654 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 655 | if (error) { |
00501b53 | 656 | if (!huge) |
f627c2f5 | 657 | mem_cgroup_cancel_charge(page, memcg, false); |
66a0c8ee KS |
658 | return error; |
659 | } | |
660 | ||
09cbfeaf | 661 | get_page(page); |
66a0c8ee KS |
662 | page->mapping = mapping; |
663 | page->index = offset; | |
664 | ||
665 | spin_lock_irq(&mapping->tree_lock); | |
a528910e | 666 | error = page_cache_tree_insert(mapping, page, shadowp); |
66a0c8ee KS |
667 | radix_tree_preload_end(); |
668 | if (unlikely(error)) | |
669 | goto err_insert; | |
4165b9b4 MH |
670 | |
671 | /* hugetlb pages do not participate in page cache accounting. */ | |
672 | if (!huge) | |
673 | __inc_zone_page_state(page, NR_FILE_PAGES); | |
66a0c8ee | 674 | spin_unlock_irq(&mapping->tree_lock); |
00501b53 | 675 | if (!huge) |
f627c2f5 | 676 | mem_cgroup_commit_charge(page, memcg, false, false); |
66a0c8ee KS |
677 | trace_mm_filemap_add_to_page_cache(page); |
678 | return 0; | |
679 | err_insert: | |
680 | page->mapping = NULL; | |
681 | /* Leave page->index set: truncation relies upon it */ | |
682 | spin_unlock_irq(&mapping->tree_lock); | |
00501b53 | 683 | if (!huge) |
f627c2f5 | 684 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 685 | put_page(page); |
1da177e4 LT |
686 | return error; |
687 | } | |
a528910e JW |
688 | |
689 | /** | |
690 | * add_to_page_cache_locked - add a locked page to the pagecache | |
691 | * @page: page to add | |
692 | * @mapping: the page's address_space | |
693 | * @offset: page index | |
694 | * @gfp_mask: page allocation mode | |
695 | * | |
696 | * This function is used to add a page to the pagecache. It must be locked. | |
697 | * This function does not add the page to the LRU. The caller must do that. | |
698 | */ | |
699 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
700 | pgoff_t offset, gfp_t gfp_mask) | |
701 | { | |
702 | return __add_to_page_cache_locked(page, mapping, offset, | |
703 | gfp_mask, NULL); | |
704 | } | |
e286781d | 705 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
706 | |
707 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 708 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 709 | { |
a528910e | 710 | void *shadow = NULL; |
4f98a2fe RR |
711 | int ret; |
712 | ||
48c935ad | 713 | __SetPageLocked(page); |
a528910e JW |
714 | ret = __add_to_page_cache_locked(page, mapping, offset, |
715 | gfp_mask, &shadow); | |
716 | if (unlikely(ret)) | |
48c935ad | 717 | __ClearPageLocked(page); |
a528910e JW |
718 | else { |
719 | /* | |
720 | * The page might have been evicted from cache only | |
721 | * recently, in which case it should be activated like | |
722 | * any other repeatedly accessed page. | |
723 | */ | |
724 | if (shadow && workingset_refault(shadow)) { | |
725 | SetPageActive(page); | |
726 | workingset_activation(page); | |
727 | } else | |
728 | ClearPageActive(page); | |
729 | lru_cache_add(page); | |
730 | } | |
1da177e4 LT |
731 | return ret; |
732 | } | |
18bc0bbd | 733 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 734 | |
44110fe3 | 735 | #ifdef CONFIG_NUMA |
2ae88149 | 736 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 737 | { |
c0ff7453 MX |
738 | int n; |
739 | struct page *page; | |
740 | ||
44110fe3 | 741 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
742 | unsigned int cpuset_mems_cookie; |
743 | do { | |
d26914d1 | 744 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 745 | n = cpuset_mem_spread_node(); |
96db800f | 746 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 747 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 748 | |
c0ff7453 | 749 | return page; |
44110fe3 | 750 | } |
2ae88149 | 751 | return alloc_pages(gfp, 0); |
44110fe3 | 752 | } |
2ae88149 | 753 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
754 | #endif |
755 | ||
1da177e4 LT |
756 | /* |
757 | * In order to wait for pages to become available there must be | |
758 | * waitqueues associated with pages. By using a hash table of | |
759 | * waitqueues where the bucket discipline is to maintain all | |
760 | * waiters on the same queue and wake all when any of the pages | |
761 | * become available, and for the woken contexts to check to be | |
762 | * sure the appropriate page became available, this saves space | |
763 | * at a cost of "thundering herd" phenomena during rare hash | |
764 | * collisions. | |
765 | */ | |
a4796e37 | 766 | wait_queue_head_t *page_waitqueue(struct page *page) |
1da177e4 LT |
767 | { |
768 | const struct zone *zone = page_zone(page); | |
769 | ||
770 | return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; | |
771 | } | |
a4796e37 | 772 | EXPORT_SYMBOL(page_waitqueue); |
1da177e4 | 773 | |
920c7a5d | 774 | void wait_on_page_bit(struct page *page, int bit_nr) |
1da177e4 LT |
775 | { |
776 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
777 | ||
778 | if (test_bit(bit_nr, &page->flags)) | |
74316201 | 779 | __wait_on_bit(page_waitqueue(page), &wait, bit_wait_io, |
1da177e4 LT |
780 | TASK_UNINTERRUPTIBLE); |
781 | } | |
782 | EXPORT_SYMBOL(wait_on_page_bit); | |
783 | ||
f62e00cc KM |
784 | int wait_on_page_bit_killable(struct page *page, int bit_nr) |
785 | { | |
786 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
787 | ||
788 | if (!test_bit(bit_nr, &page->flags)) | |
789 | return 0; | |
790 | ||
791 | return __wait_on_bit(page_waitqueue(page), &wait, | |
74316201 | 792 | bit_wait_io, TASK_KILLABLE); |
f62e00cc KM |
793 | } |
794 | ||
cbbce822 N |
795 | int wait_on_page_bit_killable_timeout(struct page *page, |
796 | int bit_nr, unsigned long timeout) | |
797 | { | |
798 | DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); | |
799 | ||
800 | wait.key.timeout = jiffies + timeout; | |
801 | if (!test_bit(bit_nr, &page->flags)) | |
802 | return 0; | |
803 | return __wait_on_bit(page_waitqueue(page), &wait, | |
804 | bit_wait_io_timeout, TASK_KILLABLE); | |
805 | } | |
806 | EXPORT_SYMBOL_GPL(wait_on_page_bit_killable_timeout); | |
807 | ||
385e1ca5 DH |
808 | /** |
809 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
810 | * @page: Page defining the wait queue of interest |
811 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
812 | * |
813 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
814 | */ | |
815 | void add_page_wait_queue(struct page *page, wait_queue_t *waiter) | |
816 | { | |
817 | wait_queue_head_t *q = page_waitqueue(page); | |
818 | unsigned long flags; | |
819 | ||
820 | spin_lock_irqsave(&q->lock, flags); | |
821 | __add_wait_queue(q, waiter); | |
822 | spin_unlock_irqrestore(&q->lock, flags); | |
823 | } | |
824 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
825 | ||
1da177e4 | 826 | /** |
485bb99b | 827 | * unlock_page - unlock a locked page |
1da177e4 LT |
828 | * @page: the page |
829 | * | |
830 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
831 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 832 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
833 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
834 | * | |
8413ac9d NP |
835 | * The mb is necessary to enforce ordering between the clear_bit and the read |
836 | * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). | |
1da177e4 | 837 | */ |
920c7a5d | 838 | void unlock_page(struct page *page) |
1da177e4 | 839 | { |
48c935ad | 840 | page = compound_head(page); |
309381fe | 841 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
8413ac9d | 842 | clear_bit_unlock(PG_locked, &page->flags); |
4e857c58 | 843 | smp_mb__after_atomic(); |
1da177e4 LT |
844 | wake_up_page(page, PG_locked); |
845 | } | |
846 | EXPORT_SYMBOL(unlock_page); | |
847 | ||
485bb99b RD |
848 | /** |
849 | * end_page_writeback - end writeback against a page | |
850 | * @page: the page | |
1da177e4 LT |
851 | */ |
852 | void end_page_writeback(struct page *page) | |
853 | { | |
888cf2db MG |
854 | /* |
855 | * TestClearPageReclaim could be used here but it is an atomic | |
856 | * operation and overkill in this particular case. Failing to | |
857 | * shuffle a page marked for immediate reclaim is too mild to | |
858 | * justify taking an atomic operation penalty at the end of | |
859 | * ever page writeback. | |
860 | */ | |
861 | if (PageReclaim(page)) { | |
862 | ClearPageReclaim(page); | |
ac6aadb2 | 863 | rotate_reclaimable_page(page); |
888cf2db | 864 | } |
ac6aadb2 MS |
865 | |
866 | if (!test_clear_page_writeback(page)) | |
867 | BUG(); | |
868 | ||
4e857c58 | 869 | smp_mb__after_atomic(); |
1da177e4 LT |
870 | wake_up_page(page, PG_writeback); |
871 | } | |
872 | EXPORT_SYMBOL(end_page_writeback); | |
873 | ||
57d99845 MW |
874 | /* |
875 | * After completing I/O on a page, call this routine to update the page | |
876 | * flags appropriately | |
877 | */ | |
878 | void page_endio(struct page *page, int rw, int err) | |
879 | { | |
880 | if (rw == READ) { | |
881 | if (!err) { | |
882 | SetPageUptodate(page); | |
883 | } else { | |
884 | ClearPageUptodate(page); | |
885 | SetPageError(page); | |
886 | } | |
887 | unlock_page(page); | |
888 | } else { /* rw == WRITE */ | |
889 | if (err) { | |
890 | SetPageError(page); | |
891 | if (page->mapping) | |
892 | mapping_set_error(page->mapping, err); | |
893 | } | |
894 | end_page_writeback(page); | |
895 | } | |
896 | } | |
897 | EXPORT_SYMBOL_GPL(page_endio); | |
898 | ||
485bb99b RD |
899 | /** |
900 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
901 | * @page: the page to lock | |
1da177e4 | 902 | */ |
920c7a5d | 903 | void __lock_page(struct page *page) |
1da177e4 | 904 | { |
48c935ad KS |
905 | struct page *page_head = compound_head(page); |
906 | DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked); | |
1da177e4 | 907 | |
48c935ad | 908 | __wait_on_bit_lock(page_waitqueue(page_head), &wait, bit_wait_io, |
1da177e4 LT |
909 | TASK_UNINTERRUPTIBLE); |
910 | } | |
911 | EXPORT_SYMBOL(__lock_page); | |
912 | ||
b5606c2d | 913 | int __lock_page_killable(struct page *page) |
2687a356 | 914 | { |
48c935ad KS |
915 | struct page *page_head = compound_head(page); |
916 | DEFINE_WAIT_BIT(wait, &page_head->flags, PG_locked); | |
2687a356 | 917 | |
48c935ad | 918 | return __wait_on_bit_lock(page_waitqueue(page_head), &wait, |
74316201 | 919 | bit_wait_io, TASK_KILLABLE); |
2687a356 | 920 | } |
18bc0bbd | 921 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 922 | |
9a95f3cf PC |
923 | /* |
924 | * Return values: | |
925 | * 1 - page is locked; mmap_sem is still held. | |
926 | * 0 - page is not locked. | |
927 | * mmap_sem has been released (up_read()), unless flags had both | |
928 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
929 | * which case mmap_sem is still held. | |
930 | * | |
931 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
932 | * with the page locked and the mmap_sem unperturbed. | |
933 | */ | |
d065bd81 ML |
934 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
935 | unsigned int flags) | |
936 | { | |
37b23e05 KM |
937 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
938 | /* | |
939 | * CAUTION! In this case, mmap_sem is not released | |
940 | * even though return 0. | |
941 | */ | |
942 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
943 | return 0; | |
944 | ||
945 | up_read(&mm->mmap_sem); | |
946 | if (flags & FAULT_FLAG_KILLABLE) | |
947 | wait_on_page_locked_killable(page); | |
948 | else | |
318b275f | 949 | wait_on_page_locked(page); |
d065bd81 | 950 | return 0; |
37b23e05 KM |
951 | } else { |
952 | if (flags & FAULT_FLAG_KILLABLE) { | |
953 | int ret; | |
954 | ||
955 | ret = __lock_page_killable(page); | |
956 | if (ret) { | |
957 | up_read(&mm->mmap_sem); | |
958 | return 0; | |
959 | } | |
960 | } else | |
961 | __lock_page(page); | |
962 | return 1; | |
d065bd81 ML |
963 | } |
964 | } | |
965 | ||
e7b563bb JW |
966 | /** |
967 | * page_cache_next_hole - find the next hole (not-present entry) | |
968 | * @mapping: mapping | |
969 | * @index: index | |
970 | * @max_scan: maximum range to search | |
971 | * | |
972 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
973 | * lowest indexed hole. | |
974 | * | |
975 | * Returns: the index of the hole if found, otherwise returns an index | |
976 | * outside of the set specified (in which case 'return - index >= | |
977 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
978 | * be returned. | |
979 | * | |
980 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
981 | * like radix_tree_gang_lookup, this will not atomically search a | |
982 | * snapshot of the tree at a single point in time. For example, if a | |
983 | * hole is created at index 5, then subsequently a hole is created at | |
984 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
985 | * if called under rcu_read_lock. | |
986 | */ | |
987 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
988 | pgoff_t index, unsigned long max_scan) | |
989 | { | |
990 | unsigned long i; | |
991 | ||
992 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
993 | struct page *page; |
994 | ||
995 | page = radix_tree_lookup(&mapping->page_tree, index); | |
996 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
997 | break; |
998 | index++; | |
999 | if (index == 0) | |
1000 | break; | |
1001 | } | |
1002 | ||
1003 | return index; | |
1004 | } | |
1005 | EXPORT_SYMBOL(page_cache_next_hole); | |
1006 | ||
1007 | /** | |
1008 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
1009 | * @mapping: mapping | |
1010 | * @index: index | |
1011 | * @max_scan: maximum range to search | |
1012 | * | |
1013 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
1014 | * the first hole. | |
1015 | * | |
1016 | * Returns: the index of the hole if found, otherwise returns an index | |
1017 | * outside of the set specified (in which case 'index - return >= | |
1018 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
1019 | * will be returned. | |
1020 | * | |
1021 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
1022 | * like radix_tree_gang_lookup, this will not atomically search a | |
1023 | * snapshot of the tree at a single point in time. For example, if a | |
1024 | * hole is created at index 10, then subsequently a hole is created at | |
1025 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
1026 | * called under rcu_read_lock. | |
1027 | */ | |
1028 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
1029 | pgoff_t index, unsigned long max_scan) | |
1030 | { | |
1031 | unsigned long i; | |
1032 | ||
1033 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
1034 | struct page *page; |
1035 | ||
1036 | page = radix_tree_lookup(&mapping->page_tree, index); | |
1037 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
1038 | break; |
1039 | index--; | |
1040 | if (index == ULONG_MAX) | |
1041 | break; | |
1042 | } | |
1043 | ||
1044 | return index; | |
1045 | } | |
1046 | EXPORT_SYMBOL(page_cache_prev_hole); | |
1047 | ||
485bb99b | 1048 | /** |
0cd6144a | 1049 | * find_get_entry - find and get a page cache entry |
485bb99b | 1050 | * @mapping: the address_space to search |
0cd6144a JW |
1051 | * @offset: the page cache index |
1052 | * | |
1053 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1054 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1055 | * |
139b6a6f JW |
1056 | * If the slot holds a shadow entry of a previously evicted page, or a |
1057 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1058 | * |
1059 | * Otherwise, %NULL is returned. | |
1da177e4 | 1060 | */ |
0cd6144a | 1061 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1062 | { |
a60637c8 | 1063 | void **pagep; |
1da177e4 LT |
1064 | struct page *page; |
1065 | ||
a60637c8 NP |
1066 | rcu_read_lock(); |
1067 | repeat: | |
1068 | page = NULL; | |
1069 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
1070 | if (pagep) { | |
1071 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
1072 | if (unlikely(!page)) |
1073 | goto out; | |
a2c16d6c | 1074 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1075 | if (radix_tree_deref_retry(page)) |
1076 | goto repeat; | |
1077 | /* | |
139b6a6f JW |
1078 | * A shadow entry of a recently evicted page, |
1079 | * or a swap entry from shmem/tmpfs. Return | |
1080 | * it without attempting to raise page count. | |
8079b1c8 HD |
1081 | */ |
1082 | goto out; | |
a2c16d6c | 1083 | } |
a60637c8 NP |
1084 | if (!page_cache_get_speculative(page)) |
1085 | goto repeat; | |
1086 | ||
1087 | /* | |
1088 | * Has the page moved? | |
1089 | * This is part of the lockless pagecache protocol. See | |
1090 | * include/linux/pagemap.h for details. | |
1091 | */ | |
1092 | if (unlikely(page != *pagep)) { | |
09cbfeaf | 1093 | put_page(page); |
a60637c8 NP |
1094 | goto repeat; |
1095 | } | |
1096 | } | |
27d20fdd | 1097 | out: |
a60637c8 NP |
1098 | rcu_read_unlock(); |
1099 | ||
1da177e4 LT |
1100 | return page; |
1101 | } | |
0cd6144a | 1102 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1103 | |
0cd6144a JW |
1104 | /** |
1105 | * find_lock_entry - locate, pin and lock a page cache entry | |
1106 | * @mapping: the address_space to search | |
1107 | * @offset: the page cache index | |
1108 | * | |
1109 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1110 | * page cache page, it is returned locked and with an increased | |
1111 | * refcount. | |
1112 | * | |
139b6a6f JW |
1113 | * If the slot holds a shadow entry of a previously evicted page, or a |
1114 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1115 | * |
1116 | * Otherwise, %NULL is returned. | |
1117 | * | |
1118 | * find_lock_entry() may sleep. | |
1119 | */ | |
1120 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1121 | { |
1122 | struct page *page; | |
1123 | ||
1da177e4 | 1124 | repeat: |
0cd6144a | 1125 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1126 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1127 | lock_page(page); |
1128 | /* Has the page been truncated? */ | |
1129 | if (unlikely(page->mapping != mapping)) { | |
1130 | unlock_page(page); | |
09cbfeaf | 1131 | put_page(page); |
a60637c8 | 1132 | goto repeat; |
1da177e4 | 1133 | } |
309381fe | 1134 | VM_BUG_ON_PAGE(page->index != offset, page); |
1da177e4 | 1135 | } |
1da177e4 LT |
1136 | return page; |
1137 | } | |
0cd6144a JW |
1138 | EXPORT_SYMBOL(find_lock_entry); |
1139 | ||
1140 | /** | |
2457aec6 | 1141 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1142 | * @mapping: the address_space to search |
1143 | * @offset: the page index | |
2457aec6 | 1144 | * @fgp_flags: PCG flags |
45f87de5 | 1145 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
0cd6144a | 1146 | * |
2457aec6 | 1147 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1148 | * |
75325189 | 1149 | * PCG flags modify how the page is returned. |
0cd6144a | 1150 | * |
2457aec6 MG |
1151 | * FGP_ACCESSED: the page will be marked accessed |
1152 | * FGP_LOCK: Page is return locked | |
1153 | * FGP_CREAT: If page is not present then a new page is allocated using | |
45f87de5 MH |
1154 | * @gfp_mask and added to the page cache and the VM's LRU |
1155 | * list. The page is returned locked and with an increased | |
1156 | * refcount. Otherwise, %NULL is returned. | |
1da177e4 | 1157 | * |
2457aec6 MG |
1158 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1159 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1160 | * |
2457aec6 | 1161 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1162 | */ |
2457aec6 | 1163 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
45f87de5 | 1164 | int fgp_flags, gfp_t gfp_mask) |
1da177e4 | 1165 | { |
eb2be189 | 1166 | struct page *page; |
2457aec6 | 1167 | |
1da177e4 | 1168 | repeat: |
2457aec6 MG |
1169 | page = find_get_entry(mapping, offset); |
1170 | if (radix_tree_exceptional_entry(page)) | |
1171 | page = NULL; | |
1172 | if (!page) | |
1173 | goto no_page; | |
1174 | ||
1175 | if (fgp_flags & FGP_LOCK) { | |
1176 | if (fgp_flags & FGP_NOWAIT) { | |
1177 | if (!trylock_page(page)) { | |
09cbfeaf | 1178 | put_page(page); |
2457aec6 MG |
1179 | return NULL; |
1180 | } | |
1181 | } else { | |
1182 | lock_page(page); | |
1183 | } | |
1184 | ||
1185 | /* Has the page been truncated? */ | |
1186 | if (unlikely(page->mapping != mapping)) { | |
1187 | unlock_page(page); | |
09cbfeaf | 1188 | put_page(page); |
2457aec6 MG |
1189 | goto repeat; |
1190 | } | |
1191 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1192 | } | |
1193 | ||
1194 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1195 | mark_page_accessed(page); | |
1196 | ||
1197 | no_page: | |
1198 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1199 | int err; | |
1200 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1201 | gfp_mask |= __GFP_WRITE; |
1202 | if (fgp_flags & FGP_NOFS) | |
1203 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1204 | |
45f87de5 | 1205 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1206 | if (!page) |
1207 | return NULL; | |
2457aec6 MG |
1208 | |
1209 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1210 | fgp_flags |= FGP_LOCK; | |
1211 | ||
eb39d618 | 1212 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1213 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1214 | __SetPageReferenced(page); |
2457aec6 | 1215 | |
45f87de5 MH |
1216 | err = add_to_page_cache_lru(page, mapping, offset, |
1217 | gfp_mask & GFP_RECLAIM_MASK); | |
eb2be189 | 1218 | if (unlikely(err)) { |
09cbfeaf | 1219 | put_page(page); |
eb2be189 NP |
1220 | page = NULL; |
1221 | if (err == -EEXIST) | |
1222 | goto repeat; | |
1da177e4 | 1223 | } |
1da177e4 | 1224 | } |
2457aec6 | 1225 | |
1da177e4 LT |
1226 | return page; |
1227 | } | |
2457aec6 | 1228 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1229 | |
0cd6144a JW |
1230 | /** |
1231 | * find_get_entries - gang pagecache lookup | |
1232 | * @mapping: The address_space to search | |
1233 | * @start: The starting page cache index | |
1234 | * @nr_entries: The maximum number of entries | |
1235 | * @entries: Where the resulting entries are placed | |
1236 | * @indices: The cache indices corresponding to the entries in @entries | |
1237 | * | |
1238 | * find_get_entries() will search for and return a group of up to | |
1239 | * @nr_entries entries in the mapping. The entries are placed at | |
1240 | * @entries. find_get_entries() takes a reference against any actual | |
1241 | * pages it returns. | |
1242 | * | |
1243 | * The search returns a group of mapping-contiguous page cache entries | |
1244 | * with ascending indexes. There may be holes in the indices due to | |
1245 | * not-present pages. | |
1246 | * | |
139b6a6f JW |
1247 | * Any shadow entries of evicted pages, or swap entries from |
1248 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1249 | * |
1250 | * find_get_entries() returns the number of pages and shadow entries | |
1251 | * which were found. | |
1252 | */ | |
1253 | unsigned find_get_entries(struct address_space *mapping, | |
1254 | pgoff_t start, unsigned int nr_entries, | |
1255 | struct page **entries, pgoff_t *indices) | |
1256 | { | |
1257 | void **slot; | |
1258 | unsigned int ret = 0; | |
1259 | struct radix_tree_iter iter; | |
1260 | ||
1261 | if (!nr_entries) | |
1262 | return 0; | |
1263 | ||
1264 | rcu_read_lock(); | |
0cd6144a JW |
1265 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
1266 | struct page *page; | |
1267 | repeat: | |
1268 | page = radix_tree_deref_slot(slot); | |
1269 | if (unlikely(!page)) | |
1270 | continue; | |
1271 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
1272 | if (radix_tree_deref_retry(page)) { |
1273 | slot = radix_tree_iter_retry(&iter); | |
1274 | continue; | |
1275 | } | |
0cd6144a | 1276 | /* |
f9fe48be RZ |
1277 | * A shadow entry of a recently evicted page, a swap |
1278 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1279 | * without attempting to raise page count. | |
0cd6144a JW |
1280 | */ |
1281 | goto export; | |
1282 | } | |
1283 | if (!page_cache_get_speculative(page)) | |
1284 | goto repeat; | |
1285 | ||
1286 | /* Has the page moved? */ | |
1287 | if (unlikely(page != *slot)) { | |
09cbfeaf | 1288 | put_page(page); |
0cd6144a JW |
1289 | goto repeat; |
1290 | } | |
1291 | export: | |
1292 | indices[ret] = iter.index; | |
1293 | entries[ret] = page; | |
1294 | if (++ret == nr_entries) | |
1295 | break; | |
1296 | } | |
1297 | rcu_read_unlock(); | |
1298 | return ret; | |
1299 | } | |
1300 | ||
1da177e4 LT |
1301 | /** |
1302 | * find_get_pages - gang pagecache lookup | |
1303 | * @mapping: The address_space to search | |
1304 | * @start: The starting page index | |
1305 | * @nr_pages: The maximum number of pages | |
1306 | * @pages: Where the resulting pages are placed | |
1307 | * | |
1308 | * find_get_pages() will search for and return a group of up to | |
1309 | * @nr_pages pages in the mapping. The pages are placed at @pages. | |
1310 | * find_get_pages() takes a reference against the returned pages. | |
1311 | * | |
1312 | * The search returns a group of mapping-contiguous pages with ascending | |
1313 | * indexes. There may be holes in the indices due to not-present pages. | |
1314 | * | |
1315 | * find_get_pages() returns the number of pages which were found. | |
1316 | */ | |
1317 | unsigned find_get_pages(struct address_space *mapping, pgoff_t start, | |
1318 | unsigned int nr_pages, struct page **pages) | |
1319 | { | |
0fc9d104 KK |
1320 | struct radix_tree_iter iter; |
1321 | void **slot; | |
1322 | unsigned ret = 0; | |
1323 | ||
1324 | if (unlikely(!nr_pages)) | |
1325 | return 0; | |
a60637c8 NP |
1326 | |
1327 | rcu_read_lock(); | |
0fc9d104 | 1328 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
a60637c8 NP |
1329 | struct page *page; |
1330 | repeat: | |
0fc9d104 | 1331 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1332 | if (unlikely(!page)) |
1333 | continue; | |
9d8aa4ea | 1334 | |
a2c16d6c | 1335 | if (radix_tree_exception(page)) { |
8079b1c8 | 1336 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1337 | slot = radix_tree_iter_retry(&iter); |
1338 | continue; | |
8079b1c8 | 1339 | } |
a2c16d6c | 1340 | /* |
139b6a6f JW |
1341 | * A shadow entry of a recently evicted page, |
1342 | * or a swap entry from shmem/tmpfs. Skip | |
1343 | * over it. | |
a2c16d6c | 1344 | */ |
8079b1c8 | 1345 | continue; |
27d20fdd | 1346 | } |
a60637c8 NP |
1347 | |
1348 | if (!page_cache_get_speculative(page)) | |
1349 | goto repeat; | |
1350 | ||
1351 | /* Has the page moved? */ | |
0fc9d104 | 1352 | if (unlikely(page != *slot)) { |
09cbfeaf | 1353 | put_page(page); |
a60637c8 NP |
1354 | goto repeat; |
1355 | } | |
1da177e4 | 1356 | |
a60637c8 | 1357 | pages[ret] = page; |
0fc9d104 KK |
1358 | if (++ret == nr_pages) |
1359 | break; | |
a60637c8 | 1360 | } |
5b280c0c | 1361 | |
a60637c8 | 1362 | rcu_read_unlock(); |
1da177e4 LT |
1363 | return ret; |
1364 | } | |
1365 | ||
ebf43500 JA |
1366 | /** |
1367 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1368 | * @mapping: The address_space to search | |
1369 | * @index: The starting page index | |
1370 | * @nr_pages: The maximum number of pages | |
1371 | * @pages: Where the resulting pages are placed | |
1372 | * | |
1373 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1374 | * that the returned number of pages are guaranteed to be contiguous. | |
1375 | * | |
1376 | * find_get_pages_contig() returns the number of pages which were found. | |
1377 | */ | |
1378 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1379 | unsigned int nr_pages, struct page **pages) | |
1380 | { | |
0fc9d104 KK |
1381 | struct radix_tree_iter iter; |
1382 | void **slot; | |
1383 | unsigned int ret = 0; | |
1384 | ||
1385 | if (unlikely(!nr_pages)) | |
1386 | return 0; | |
a60637c8 NP |
1387 | |
1388 | rcu_read_lock(); | |
0fc9d104 | 1389 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
a60637c8 NP |
1390 | struct page *page; |
1391 | repeat: | |
0fc9d104 KK |
1392 | page = radix_tree_deref_slot(slot); |
1393 | /* The hole, there no reason to continue */ | |
a60637c8 | 1394 | if (unlikely(!page)) |
0fc9d104 | 1395 | break; |
9d8aa4ea | 1396 | |
a2c16d6c | 1397 | if (radix_tree_exception(page)) { |
8079b1c8 | 1398 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1399 | slot = radix_tree_iter_retry(&iter); |
1400 | continue; | |
8079b1c8 | 1401 | } |
a2c16d6c | 1402 | /* |
139b6a6f JW |
1403 | * A shadow entry of a recently evicted page, |
1404 | * or a swap entry from shmem/tmpfs. Stop | |
1405 | * looking for contiguous pages. | |
a2c16d6c | 1406 | */ |
8079b1c8 | 1407 | break; |
a2c16d6c | 1408 | } |
ebf43500 | 1409 | |
a60637c8 NP |
1410 | if (!page_cache_get_speculative(page)) |
1411 | goto repeat; | |
1412 | ||
1413 | /* Has the page moved? */ | |
0fc9d104 | 1414 | if (unlikely(page != *slot)) { |
09cbfeaf | 1415 | put_page(page); |
a60637c8 NP |
1416 | goto repeat; |
1417 | } | |
1418 | ||
9cbb4cb2 NP |
1419 | /* |
1420 | * must check mapping and index after taking the ref. | |
1421 | * otherwise we can get both false positives and false | |
1422 | * negatives, which is just confusing to the caller. | |
1423 | */ | |
0fc9d104 | 1424 | if (page->mapping == NULL || page->index != iter.index) { |
09cbfeaf | 1425 | put_page(page); |
9cbb4cb2 NP |
1426 | break; |
1427 | } | |
1428 | ||
a60637c8 | 1429 | pages[ret] = page; |
0fc9d104 KK |
1430 | if (++ret == nr_pages) |
1431 | break; | |
ebf43500 | 1432 | } |
a60637c8 NP |
1433 | rcu_read_unlock(); |
1434 | return ret; | |
ebf43500 | 1435 | } |
ef71c15c | 1436 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1437 | |
485bb99b RD |
1438 | /** |
1439 | * find_get_pages_tag - find and return pages that match @tag | |
1440 | * @mapping: the address_space to search | |
1441 | * @index: the starting page index | |
1442 | * @tag: the tag index | |
1443 | * @nr_pages: the maximum number of pages | |
1444 | * @pages: where the resulting pages are placed | |
1445 | * | |
1da177e4 | 1446 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1447 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 LT |
1448 | */ |
1449 | unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, | |
1450 | int tag, unsigned int nr_pages, struct page **pages) | |
1451 | { | |
0fc9d104 KK |
1452 | struct radix_tree_iter iter; |
1453 | void **slot; | |
1454 | unsigned ret = 0; | |
1455 | ||
1456 | if (unlikely(!nr_pages)) | |
1457 | return 0; | |
a60637c8 NP |
1458 | |
1459 | rcu_read_lock(); | |
0fc9d104 KK |
1460 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1461 | &iter, *index, tag) { | |
a60637c8 NP |
1462 | struct page *page; |
1463 | repeat: | |
0fc9d104 | 1464 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1465 | if (unlikely(!page)) |
1466 | continue; | |
9d8aa4ea | 1467 | |
a2c16d6c | 1468 | if (radix_tree_exception(page)) { |
8079b1c8 | 1469 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1470 | slot = radix_tree_iter_retry(&iter); |
1471 | continue; | |
8079b1c8 | 1472 | } |
a2c16d6c | 1473 | /* |
139b6a6f JW |
1474 | * A shadow entry of a recently evicted page. |
1475 | * | |
1476 | * Those entries should never be tagged, but | |
1477 | * this tree walk is lockless and the tags are | |
1478 | * looked up in bulk, one radix tree node at a | |
1479 | * time, so there is a sizable window for page | |
1480 | * reclaim to evict a page we saw tagged. | |
1481 | * | |
1482 | * Skip over it. | |
a2c16d6c | 1483 | */ |
139b6a6f | 1484 | continue; |
a2c16d6c | 1485 | } |
a60637c8 NP |
1486 | |
1487 | if (!page_cache_get_speculative(page)) | |
1488 | goto repeat; | |
1489 | ||
1490 | /* Has the page moved? */ | |
0fc9d104 | 1491 | if (unlikely(page != *slot)) { |
09cbfeaf | 1492 | put_page(page); |
a60637c8 NP |
1493 | goto repeat; |
1494 | } | |
1495 | ||
1496 | pages[ret] = page; | |
0fc9d104 KK |
1497 | if (++ret == nr_pages) |
1498 | break; | |
a60637c8 | 1499 | } |
5b280c0c | 1500 | |
a60637c8 | 1501 | rcu_read_unlock(); |
1da177e4 | 1502 | |
1da177e4 LT |
1503 | if (ret) |
1504 | *index = pages[ret - 1]->index + 1; | |
a60637c8 | 1505 | |
1da177e4 LT |
1506 | return ret; |
1507 | } | |
ef71c15c | 1508 | EXPORT_SYMBOL(find_get_pages_tag); |
1da177e4 | 1509 | |
7e7f7749 RZ |
1510 | /** |
1511 | * find_get_entries_tag - find and return entries that match @tag | |
1512 | * @mapping: the address_space to search | |
1513 | * @start: the starting page cache index | |
1514 | * @tag: the tag index | |
1515 | * @nr_entries: the maximum number of entries | |
1516 | * @entries: where the resulting entries are placed | |
1517 | * @indices: the cache indices corresponding to the entries in @entries | |
1518 | * | |
1519 | * Like find_get_entries, except we only return entries which are tagged with | |
1520 | * @tag. | |
1521 | */ | |
1522 | unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start, | |
1523 | int tag, unsigned int nr_entries, | |
1524 | struct page **entries, pgoff_t *indices) | |
1525 | { | |
1526 | void **slot; | |
1527 | unsigned int ret = 0; | |
1528 | struct radix_tree_iter iter; | |
1529 | ||
1530 | if (!nr_entries) | |
1531 | return 0; | |
1532 | ||
1533 | rcu_read_lock(); | |
7e7f7749 RZ |
1534 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1535 | &iter, start, tag) { | |
1536 | struct page *page; | |
1537 | repeat: | |
1538 | page = radix_tree_deref_slot(slot); | |
1539 | if (unlikely(!page)) | |
1540 | continue; | |
1541 | if (radix_tree_exception(page)) { | |
1542 | if (radix_tree_deref_retry(page)) { | |
2cf938aa MW |
1543 | slot = radix_tree_iter_retry(&iter); |
1544 | continue; | |
7e7f7749 RZ |
1545 | } |
1546 | ||
1547 | /* | |
1548 | * A shadow entry of a recently evicted page, a swap | |
1549 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1550 | * without attempting to raise page count. | |
1551 | */ | |
1552 | goto export; | |
1553 | } | |
1554 | if (!page_cache_get_speculative(page)) | |
1555 | goto repeat; | |
1556 | ||
1557 | /* Has the page moved? */ | |
1558 | if (unlikely(page != *slot)) { | |
09cbfeaf | 1559 | put_page(page); |
7e7f7749 RZ |
1560 | goto repeat; |
1561 | } | |
1562 | export: | |
1563 | indices[ret] = iter.index; | |
1564 | entries[ret] = page; | |
1565 | if (++ret == nr_entries) | |
1566 | break; | |
1567 | } | |
1568 | rcu_read_unlock(); | |
1569 | return ret; | |
1570 | } | |
1571 | EXPORT_SYMBOL(find_get_entries_tag); | |
1572 | ||
76d42bd9 WF |
1573 | /* |
1574 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1575 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1576 | * | |
1577 | * ---R__________________________________________B__________ | |
1578 | * ^ reading here ^ bad block(assume 4k) | |
1579 | * | |
1580 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1581 | * => failing the whole request => read(R) => read(R+1) => | |
1582 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1583 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1584 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1585 | * | |
1586 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1587 | */ | |
1588 | static void shrink_readahead_size_eio(struct file *filp, | |
1589 | struct file_ra_state *ra) | |
1590 | { | |
76d42bd9 | 1591 | ra->ra_pages /= 4; |
76d42bd9 WF |
1592 | } |
1593 | ||
485bb99b | 1594 | /** |
36e78914 | 1595 | * do_generic_file_read - generic file read routine |
485bb99b RD |
1596 | * @filp: the file to read |
1597 | * @ppos: current file position | |
6e58e79d AV |
1598 | * @iter: data destination |
1599 | * @written: already copied | |
485bb99b | 1600 | * |
1da177e4 | 1601 | * This is a generic file read routine, and uses the |
485bb99b | 1602 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1603 | * |
1604 | * This is really ugly. But the goto's actually try to clarify some | |
1605 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1606 | */ |
6e58e79d AV |
1607 | static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos, |
1608 | struct iov_iter *iter, ssize_t written) | |
1da177e4 | 1609 | { |
36e78914 | 1610 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1611 | struct inode *inode = mapping->host; |
36e78914 | 1612 | struct file_ra_state *ra = &filp->f_ra; |
57f6b96c FW |
1613 | pgoff_t index; |
1614 | pgoff_t last_index; | |
1615 | pgoff_t prev_index; | |
1616 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1617 | unsigned int prev_offset; |
6e58e79d | 1618 | int error = 0; |
1da177e4 | 1619 | |
09cbfeaf KS |
1620 | index = *ppos >> PAGE_SHIFT; |
1621 | prev_index = ra->prev_pos >> PAGE_SHIFT; | |
1622 | prev_offset = ra->prev_pos & (PAGE_SIZE-1); | |
1623 | last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; | |
1624 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 | 1625 | |
1da177e4 LT |
1626 | for (;;) { |
1627 | struct page *page; | |
57f6b96c | 1628 | pgoff_t end_index; |
a32ea1e1 | 1629 | loff_t isize; |
1da177e4 LT |
1630 | unsigned long nr, ret; |
1631 | ||
1da177e4 | 1632 | cond_resched(); |
1da177e4 LT |
1633 | find_page: |
1634 | page = find_get_page(mapping, index); | |
3ea89ee8 | 1635 | if (!page) { |
cf914a7d | 1636 | page_cache_sync_readahead(mapping, |
7ff81078 | 1637 | ra, filp, |
3ea89ee8 FW |
1638 | index, last_index - index); |
1639 | page = find_get_page(mapping, index); | |
1640 | if (unlikely(page == NULL)) | |
1641 | goto no_cached_page; | |
1642 | } | |
1643 | if (PageReadahead(page)) { | |
cf914a7d | 1644 | page_cache_async_readahead(mapping, |
7ff81078 | 1645 | ra, filp, page, |
3ea89ee8 | 1646 | index, last_index - index); |
1da177e4 | 1647 | } |
8ab22b9a | 1648 | if (!PageUptodate(page)) { |
ebded027 MG |
1649 | /* |
1650 | * See comment in do_read_cache_page on why | |
1651 | * wait_on_page_locked is used to avoid unnecessarily | |
1652 | * serialisations and why it's safe. | |
1653 | */ | |
1654 | wait_on_page_locked_killable(page); | |
1655 | if (PageUptodate(page)) | |
1656 | goto page_ok; | |
1657 | ||
09cbfeaf | 1658 | if (inode->i_blkbits == PAGE_SHIFT || |
8ab22b9a HH |
1659 | !mapping->a_ops->is_partially_uptodate) |
1660 | goto page_not_up_to_date; | |
529ae9aa | 1661 | if (!trylock_page(page)) |
8ab22b9a | 1662 | goto page_not_up_to_date; |
8d056cb9 DH |
1663 | /* Did it get truncated before we got the lock? */ |
1664 | if (!page->mapping) | |
1665 | goto page_not_up_to_date_locked; | |
8ab22b9a | 1666 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 1667 | offset, iter->count)) |
8ab22b9a HH |
1668 | goto page_not_up_to_date_locked; |
1669 | unlock_page(page); | |
1670 | } | |
1da177e4 | 1671 | page_ok: |
a32ea1e1 N |
1672 | /* |
1673 | * i_size must be checked after we know the page is Uptodate. | |
1674 | * | |
1675 | * Checking i_size after the check allows us to calculate | |
1676 | * the correct value for "nr", which means the zero-filled | |
1677 | * part of the page is not copied back to userspace (unless | |
1678 | * another truncate extends the file - this is desired though). | |
1679 | */ | |
1680 | ||
1681 | isize = i_size_read(inode); | |
09cbfeaf | 1682 | end_index = (isize - 1) >> PAGE_SHIFT; |
a32ea1e1 | 1683 | if (unlikely(!isize || index > end_index)) { |
09cbfeaf | 1684 | put_page(page); |
a32ea1e1 N |
1685 | goto out; |
1686 | } | |
1687 | ||
1688 | /* nr is the maximum number of bytes to copy from this page */ | |
09cbfeaf | 1689 | nr = PAGE_SIZE; |
a32ea1e1 | 1690 | if (index == end_index) { |
09cbfeaf | 1691 | nr = ((isize - 1) & ~PAGE_MASK) + 1; |
a32ea1e1 | 1692 | if (nr <= offset) { |
09cbfeaf | 1693 | put_page(page); |
a32ea1e1 N |
1694 | goto out; |
1695 | } | |
1696 | } | |
1697 | nr = nr - offset; | |
1da177e4 LT |
1698 | |
1699 | /* If users can be writing to this page using arbitrary | |
1700 | * virtual addresses, take care about potential aliasing | |
1701 | * before reading the page on the kernel side. | |
1702 | */ | |
1703 | if (mapping_writably_mapped(mapping)) | |
1704 | flush_dcache_page(page); | |
1705 | ||
1706 | /* | |
ec0f1637 JK |
1707 | * When a sequential read accesses a page several times, |
1708 | * only mark it as accessed the first time. | |
1da177e4 | 1709 | */ |
ec0f1637 | 1710 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
1711 | mark_page_accessed(page); |
1712 | prev_index = index; | |
1713 | ||
1714 | /* | |
1715 | * Ok, we have the page, and it's up-to-date, so | |
1716 | * now we can copy it to user space... | |
1da177e4 | 1717 | */ |
6e58e79d AV |
1718 | |
1719 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 | 1720 | offset += ret; |
09cbfeaf KS |
1721 | index += offset >> PAGE_SHIFT; |
1722 | offset &= ~PAGE_MASK; | |
6ce745ed | 1723 | prev_offset = offset; |
1da177e4 | 1724 | |
09cbfeaf | 1725 | put_page(page); |
6e58e79d AV |
1726 | written += ret; |
1727 | if (!iov_iter_count(iter)) | |
1728 | goto out; | |
1729 | if (ret < nr) { | |
1730 | error = -EFAULT; | |
1731 | goto out; | |
1732 | } | |
1733 | continue; | |
1da177e4 LT |
1734 | |
1735 | page_not_up_to_date: | |
1736 | /* Get exclusive access to the page ... */ | |
85462323 ON |
1737 | error = lock_page_killable(page); |
1738 | if (unlikely(error)) | |
1739 | goto readpage_error; | |
1da177e4 | 1740 | |
8ab22b9a | 1741 | page_not_up_to_date_locked: |
da6052f7 | 1742 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
1743 | if (!page->mapping) { |
1744 | unlock_page(page); | |
09cbfeaf | 1745 | put_page(page); |
1da177e4 LT |
1746 | continue; |
1747 | } | |
1748 | ||
1749 | /* Did somebody else fill it already? */ | |
1750 | if (PageUptodate(page)) { | |
1751 | unlock_page(page); | |
1752 | goto page_ok; | |
1753 | } | |
1754 | ||
1755 | readpage: | |
91803b49 JM |
1756 | /* |
1757 | * A previous I/O error may have been due to temporary | |
1758 | * failures, eg. multipath errors. | |
1759 | * PG_error will be set again if readpage fails. | |
1760 | */ | |
1761 | ClearPageError(page); | |
1da177e4 LT |
1762 | /* Start the actual read. The read will unlock the page. */ |
1763 | error = mapping->a_ops->readpage(filp, page); | |
1764 | ||
994fc28c ZB |
1765 | if (unlikely(error)) { |
1766 | if (error == AOP_TRUNCATED_PAGE) { | |
09cbfeaf | 1767 | put_page(page); |
6e58e79d | 1768 | error = 0; |
994fc28c ZB |
1769 | goto find_page; |
1770 | } | |
1da177e4 | 1771 | goto readpage_error; |
994fc28c | 1772 | } |
1da177e4 LT |
1773 | |
1774 | if (!PageUptodate(page)) { | |
85462323 ON |
1775 | error = lock_page_killable(page); |
1776 | if (unlikely(error)) | |
1777 | goto readpage_error; | |
1da177e4 LT |
1778 | if (!PageUptodate(page)) { |
1779 | if (page->mapping == NULL) { | |
1780 | /* | |
2ecdc82e | 1781 | * invalidate_mapping_pages got it |
1da177e4 LT |
1782 | */ |
1783 | unlock_page(page); | |
09cbfeaf | 1784 | put_page(page); |
1da177e4 LT |
1785 | goto find_page; |
1786 | } | |
1787 | unlock_page(page); | |
7ff81078 | 1788 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
1789 | error = -EIO; |
1790 | goto readpage_error; | |
1da177e4 LT |
1791 | } |
1792 | unlock_page(page); | |
1793 | } | |
1794 | ||
1da177e4 LT |
1795 | goto page_ok; |
1796 | ||
1797 | readpage_error: | |
1798 | /* UHHUH! A synchronous read error occurred. Report it */ | |
09cbfeaf | 1799 | put_page(page); |
1da177e4 LT |
1800 | goto out; |
1801 | ||
1802 | no_cached_page: | |
1803 | /* | |
1804 | * Ok, it wasn't cached, so we need to create a new | |
1805 | * page.. | |
1806 | */ | |
eb2be189 NP |
1807 | page = page_cache_alloc_cold(mapping); |
1808 | if (!page) { | |
6e58e79d | 1809 | error = -ENOMEM; |
eb2be189 | 1810 | goto out; |
1da177e4 | 1811 | } |
6afdb859 | 1812 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 1813 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 1814 | if (error) { |
09cbfeaf | 1815 | put_page(page); |
6e58e79d AV |
1816 | if (error == -EEXIST) { |
1817 | error = 0; | |
1da177e4 | 1818 | goto find_page; |
6e58e79d | 1819 | } |
1da177e4 LT |
1820 | goto out; |
1821 | } | |
1da177e4 LT |
1822 | goto readpage; |
1823 | } | |
1824 | ||
1825 | out: | |
7ff81078 | 1826 | ra->prev_pos = prev_index; |
09cbfeaf | 1827 | ra->prev_pos <<= PAGE_SHIFT; |
7ff81078 | 1828 | ra->prev_pos |= prev_offset; |
1da177e4 | 1829 | |
09cbfeaf | 1830 | *ppos = ((loff_t)index << PAGE_SHIFT) + offset; |
0c6aa263 | 1831 | file_accessed(filp); |
6e58e79d | 1832 | return written ? written : error; |
1da177e4 LT |
1833 | } |
1834 | ||
485bb99b | 1835 | /** |
6abd2322 | 1836 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 1837 | * @iocb: kernel I/O control block |
6abd2322 | 1838 | * @iter: destination for the data read |
485bb99b | 1839 | * |
6abd2322 | 1840 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
1841 | * that can use the page cache directly. |
1842 | */ | |
1843 | ssize_t | |
ed978a81 | 1844 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 1845 | { |
ed978a81 | 1846 | struct file *file = iocb->ki_filp; |
cb66a7a1 | 1847 | ssize_t retval = 0; |
543ade1f | 1848 | loff_t *ppos = &iocb->ki_pos; |
ed978a81 | 1849 | loff_t pos = *ppos; |
e7080a43 NS |
1850 | size_t count = iov_iter_count(iter); |
1851 | ||
1852 | if (!count) | |
1853 | goto out; /* skip atime */ | |
1da177e4 | 1854 | |
2ba48ce5 | 1855 | if (iocb->ki_flags & IOCB_DIRECT) { |
ed978a81 AV |
1856 | struct address_space *mapping = file->f_mapping; |
1857 | struct inode *inode = mapping->host; | |
543ade1f | 1858 | loff_t size; |
1da177e4 | 1859 | |
1da177e4 | 1860 | size = i_size_read(inode); |
9fe55eea | 1861 | retval = filemap_write_and_wait_range(mapping, pos, |
a6cbcd4a | 1862 | pos + count - 1); |
9fe55eea | 1863 | if (!retval) { |
ed978a81 | 1864 | struct iov_iter data = *iter; |
22c6186e | 1865 | retval = mapping->a_ops->direct_IO(iocb, &data, pos); |
9fe55eea | 1866 | } |
d8d3d94b | 1867 | |
9fe55eea SW |
1868 | if (retval > 0) { |
1869 | *ppos = pos + retval; | |
ed978a81 | 1870 | iov_iter_advance(iter, retval); |
9fe55eea | 1871 | } |
66f998f6 | 1872 | |
9fe55eea SW |
1873 | /* |
1874 | * Btrfs can have a short DIO read if we encounter | |
1875 | * compressed extents, so if there was an error, or if | |
1876 | * we've already read everything we wanted to, or if | |
1877 | * there was a short read because we hit EOF, go ahead | |
1878 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
1879 | * the rest of the read. Buffered reads will not work for |
1880 | * DAX files, so don't bother trying. | |
9fe55eea | 1881 | */ |
fbbbad4b MW |
1882 | if (retval < 0 || !iov_iter_count(iter) || *ppos >= size || |
1883 | IS_DAX(inode)) { | |
ed978a81 | 1884 | file_accessed(file); |
9fe55eea | 1885 | goto out; |
0e0bcae3 | 1886 | } |
1da177e4 LT |
1887 | } |
1888 | ||
ed978a81 | 1889 | retval = do_generic_file_read(file, ppos, iter, retval); |
1da177e4 LT |
1890 | out: |
1891 | return retval; | |
1892 | } | |
ed978a81 | 1893 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 1894 | |
1da177e4 | 1895 | #ifdef CONFIG_MMU |
485bb99b RD |
1896 | /** |
1897 | * page_cache_read - adds requested page to the page cache if not already there | |
1898 | * @file: file to read | |
1899 | * @offset: page index | |
62eb320a | 1900 | * @gfp_mask: memory allocation flags |
485bb99b | 1901 | * |
1da177e4 LT |
1902 | * This adds the requested page to the page cache if it isn't already there, |
1903 | * and schedules an I/O to read in its contents from disk. | |
1904 | */ | |
c20cd45e | 1905 | static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask) |
1da177e4 LT |
1906 | { |
1907 | struct address_space *mapping = file->f_mapping; | |
99dadfdd | 1908 | struct page *page; |
994fc28c | 1909 | int ret; |
1da177e4 | 1910 | |
994fc28c | 1911 | do { |
c20cd45e | 1912 | page = __page_cache_alloc(gfp_mask|__GFP_COLD); |
994fc28c ZB |
1913 | if (!page) |
1914 | return -ENOMEM; | |
1915 | ||
c20cd45e | 1916 | ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL); |
994fc28c ZB |
1917 | if (ret == 0) |
1918 | ret = mapping->a_ops->readpage(file, page); | |
1919 | else if (ret == -EEXIST) | |
1920 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 1921 | |
09cbfeaf | 1922 | put_page(page); |
1da177e4 | 1923 | |
994fc28c | 1924 | } while (ret == AOP_TRUNCATED_PAGE); |
99dadfdd | 1925 | |
994fc28c | 1926 | return ret; |
1da177e4 LT |
1927 | } |
1928 | ||
1929 | #define MMAP_LOTSAMISS (100) | |
1930 | ||
ef00e08e LT |
1931 | /* |
1932 | * Synchronous readahead happens when we don't even find | |
1933 | * a page in the page cache at all. | |
1934 | */ | |
1935 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
1936 | struct file_ra_state *ra, | |
1937 | struct file *file, | |
1938 | pgoff_t offset) | |
1939 | { | |
ef00e08e LT |
1940 | struct address_space *mapping = file->f_mapping; |
1941 | ||
1942 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1943 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 1944 | return; |
275b12bf WF |
1945 | if (!ra->ra_pages) |
1946 | return; | |
ef00e08e | 1947 | |
64363aad | 1948 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
1949 | page_cache_sync_readahead(mapping, ra, file, offset, |
1950 | ra->ra_pages); | |
ef00e08e LT |
1951 | return; |
1952 | } | |
1953 | ||
207d04ba AK |
1954 | /* Avoid banging the cache line if not needed */ |
1955 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
1956 | ra->mmap_miss++; |
1957 | ||
1958 | /* | |
1959 | * Do we miss much more than hit in this file? If so, | |
1960 | * stop bothering with read-ahead. It will only hurt. | |
1961 | */ | |
1962 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
1963 | return; | |
1964 | ||
d30a1100 WF |
1965 | /* |
1966 | * mmap read-around | |
1967 | */ | |
600e19af RG |
1968 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
1969 | ra->size = ra->ra_pages; | |
1970 | ra->async_size = ra->ra_pages / 4; | |
275b12bf | 1971 | ra_submit(ra, mapping, file); |
ef00e08e LT |
1972 | } |
1973 | ||
1974 | /* | |
1975 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
1976 | * so we want to possibly extend the readahead further.. | |
1977 | */ | |
1978 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
1979 | struct file_ra_state *ra, | |
1980 | struct file *file, | |
1981 | struct page *page, | |
1982 | pgoff_t offset) | |
1983 | { | |
1984 | struct address_space *mapping = file->f_mapping; | |
1985 | ||
1986 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 1987 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
1988 | return; |
1989 | if (ra->mmap_miss > 0) | |
1990 | ra->mmap_miss--; | |
1991 | if (PageReadahead(page)) | |
2fad6f5d WF |
1992 | page_cache_async_readahead(mapping, ra, file, |
1993 | page, offset, ra->ra_pages); | |
ef00e08e LT |
1994 | } |
1995 | ||
485bb99b | 1996 | /** |
54cb8821 | 1997 | * filemap_fault - read in file data for page fault handling |
d0217ac0 NP |
1998 | * @vma: vma in which the fault was taken |
1999 | * @vmf: struct vm_fault containing details of the fault | |
485bb99b | 2000 | * |
54cb8821 | 2001 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2002 | * mapped memory region to read in file data during a page fault. |
2003 | * | |
2004 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2005 | * it in the page cache, and handles the special cases reasonably without | |
2006 | * having a lot of duplicated code. | |
9a95f3cf PC |
2007 | * |
2008 | * vma->vm_mm->mmap_sem must be held on entry. | |
2009 | * | |
2010 | * If our return value has VM_FAULT_RETRY set, it's because | |
2011 | * lock_page_or_retry() returned 0. | |
2012 | * The mmap_sem has usually been released in this case. | |
2013 | * See __lock_page_or_retry() for the exception. | |
2014 | * | |
2015 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
2016 | * has not been released. | |
2017 | * | |
2018 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
1da177e4 | 2019 | */ |
d0217ac0 | 2020 | int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
1da177e4 LT |
2021 | { |
2022 | int error; | |
54cb8821 | 2023 | struct file *file = vma->vm_file; |
1da177e4 LT |
2024 | struct address_space *mapping = file->f_mapping; |
2025 | struct file_ra_state *ra = &file->f_ra; | |
2026 | struct inode *inode = mapping->host; | |
ef00e08e | 2027 | pgoff_t offset = vmf->pgoff; |
1da177e4 | 2028 | struct page *page; |
99e3e53f | 2029 | loff_t size; |
83c54070 | 2030 | int ret = 0; |
1da177e4 | 2031 | |
09cbfeaf KS |
2032 | size = round_up(i_size_read(inode), PAGE_SIZE); |
2033 | if (offset >= size >> PAGE_SHIFT) | |
5307cc1a | 2034 | return VM_FAULT_SIGBUS; |
1da177e4 | 2035 | |
1da177e4 | 2036 | /* |
49426420 | 2037 | * Do we have something in the page cache already? |
1da177e4 | 2038 | */ |
ef00e08e | 2039 | page = find_get_page(mapping, offset); |
45cac65b | 2040 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2041 | /* |
ef00e08e LT |
2042 | * We found the page, so try async readahead before |
2043 | * waiting for the lock. | |
1da177e4 | 2044 | */ |
ef00e08e | 2045 | do_async_mmap_readahead(vma, ra, file, page, offset); |
45cac65b | 2046 | } else if (!page) { |
ef00e08e LT |
2047 | /* No page in the page cache at all */ |
2048 | do_sync_mmap_readahead(vma, ra, file, offset); | |
2049 | count_vm_event(PGMAJFAULT); | |
456f998e | 2050 | mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
2051 | ret = VM_FAULT_MAJOR; |
2052 | retry_find: | |
b522c94d | 2053 | page = find_get_page(mapping, offset); |
1da177e4 LT |
2054 | if (!page) |
2055 | goto no_cached_page; | |
2056 | } | |
2057 | ||
d88c0922 | 2058 | if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { |
09cbfeaf | 2059 | put_page(page); |
d065bd81 | 2060 | return ret | VM_FAULT_RETRY; |
d88c0922 | 2061 | } |
b522c94d ML |
2062 | |
2063 | /* Did it get truncated? */ | |
2064 | if (unlikely(page->mapping != mapping)) { | |
2065 | unlock_page(page); | |
2066 | put_page(page); | |
2067 | goto retry_find; | |
2068 | } | |
309381fe | 2069 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 2070 | |
1da177e4 | 2071 | /* |
d00806b1 NP |
2072 | * We have a locked page in the page cache, now we need to check |
2073 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2074 | */ |
d00806b1 | 2075 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2076 | goto page_not_uptodate; |
2077 | ||
ef00e08e LT |
2078 | /* |
2079 | * Found the page and have a reference on it. | |
2080 | * We must recheck i_size under page lock. | |
2081 | */ | |
09cbfeaf KS |
2082 | size = round_up(i_size_read(inode), PAGE_SIZE); |
2083 | if (unlikely(offset >= size >> PAGE_SHIFT)) { | |
d00806b1 | 2084 | unlock_page(page); |
09cbfeaf | 2085 | put_page(page); |
5307cc1a | 2086 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2087 | } |
2088 | ||
d0217ac0 | 2089 | vmf->page = page; |
83c54070 | 2090 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2091 | |
1da177e4 LT |
2092 | no_cached_page: |
2093 | /* | |
2094 | * We're only likely to ever get here if MADV_RANDOM is in | |
2095 | * effect. | |
2096 | */ | |
c20cd45e | 2097 | error = page_cache_read(file, offset, vmf->gfp_mask); |
1da177e4 LT |
2098 | |
2099 | /* | |
2100 | * The page we want has now been added to the page cache. | |
2101 | * In the unlikely event that someone removed it in the | |
2102 | * meantime, we'll just come back here and read it again. | |
2103 | */ | |
2104 | if (error >= 0) | |
2105 | goto retry_find; | |
2106 | ||
2107 | /* | |
2108 | * An error return from page_cache_read can result if the | |
2109 | * system is low on memory, or a problem occurs while trying | |
2110 | * to schedule I/O. | |
2111 | */ | |
2112 | if (error == -ENOMEM) | |
d0217ac0 NP |
2113 | return VM_FAULT_OOM; |
2114 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
2115 | |
2116 | page_not_uptodate: | |
1da177e4 LT |
2117 | /* |
2118 | * Umm, take care of errors if the page isn't up-to-date. | |
2119 | * Try to re-read it _once_. We do this synchronously, | |
2120 | * because there really aren't any performance issues here | |
2121 | * and we need to check for errors. | |
2122 | */ | |
1da177e4 | 2123 | ClearPageError(page); |
994fc28c | 2124 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2125 | if (!error) { |
2126 | wait_on_page_locked(page); | |
2127 | if (!PageUptodate(page)) | |
2128 | error = -EIO; | |
2129 | } | |
09cbfeaf | 2130 | put_page(page); |
d00806b1 NP |
2131 | |
2132 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2133 | goto retry_find; |
1da177e4 | 2134 | |
d00806b1 | 2135 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 2136 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 2137 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
2138 | } |
2139 | EXPORT_SYMBOL(filemap_fault); | |
2140 | ||
f1820361 KS |
2141 | void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf) |
2142 | { | |
2143 | struct radix_tree_iter iter; | |
2144 | void **slot; | |
2145 | struct file *file = vma->vm_file; | |
2146 | struct address_space *mapping = file->f_mapping; | |
2147 | loff_t size; | |
2148 | struct page *page; | |
2149 | unsigned long address = (unsigned long) vmf->virtual_address; | |
2150 | unsigned long addr; | |
2151 | pte_t *pte; | |
2152 | ||
2153 | rcu_read_lock(); | |
2154 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) { | |
2155 | if (iter.index > vmf->max_pgoff) | |
2156 | break; | |
2157 | repeat: | |
2158 | page = radix_tree_deref_slot(slot); | |
2159 | if (unlikely(!page)) | |
2160 | goto next; | |
2161 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2162 | if (radix_tree_deref_retry(page)) { |
2163 | slot = radix_tree_iter_retry(&iter); | |
2164 | continue; | |
2165 | } | |
2166 | goto next; | |
f1820361 KS |
2167 | } |
2168 | ||
2169 | if (!page_cache_get_speculative(page)) | |
2170 | goto repeat; | |
2171 | ||
2172 | /* Has the page moved? */ | |
2173 | if (unlikely(page != *slot)) { | |
09cbfeaf | 2174 | put_page(page); |
f1820361 KS |
2175 | goto repeat; |
2176 | } | |
2177 | ||
2178 | if (!PageUptodate(page) || | |
2179 | PageReadahead(page) || | |
2180 | PageHWPoison(page)) | |
2181 | goto skip; | |
2182 | if (!trylock_page(page)) | |
2183 | goto skip; | |
2184 | ||
2185 | if (page->mapping != mapping || !PageUptodate(page)) | |
2186 | goto unlock; | |
2187 | ||
09cbfeaf KS |
2188 | size = round_up(i_size_read(mapping->host), PAGE_SIZE); |
2189 | if (page->index >= size >> PAGE_SHIFT) | |
f1820361 KS |
2190 | goto unlock; |
2191 | ||
2192 | pte = vmf->pte + page->index - vmf->pgoff; | |
2193 | if (!pte_none(*pte)) | |
2194 | goto unlock; | |
2195 | ||
2196 | if (file->f_ra.mmap_miss > 0) | |
2197 | file->f_ra.mmap_miss--; | |
2198 | addr = address + (page->index - vmf->pgoff) * PAGE_SIZE; | |
2199 | do_set_pte(vma, addr, page, pte, false, false); | |
2200 | unlock_page(page); | |
2201 | goto next; | |
2202 | unlock: | |
2203 | unlock_page(page); | |
2204 | skip: | |
09cbfeaf | 2205 | put_page(page); |
f1820361 KS |
2206 | next: |
2207 | if (iter.index == vmf->max_pgoff) | |
2208 | break; | |
2209 | } | |
2210 | rcu_read_unlock(); | |
2211 | } | |
2212 | EXPORT_SYMBOL(filemap_map_pages); | |
2213 | ||
4fcf1c62 JK |
2214 | int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2215 | { | |
2216 | struct page *page = vmf->page; | |
496ad9aa | 2217 | struct inode *inode = file_inode(vma->vm_file); |
4fcf1c62 JK |
2218 | int ret = VM_FAULT_LOCKED; |
2219 | ||
14da9200 | 2220 | sb_start_pagefault(inode->i_sb); |
4fcf1c62 JK |
2221 | file_update_time(vma->vm_file); |
2222 | lock_page(page); | |
2223 | if (page->mapping != inode->i_mapping) { | |
2224 | unlock_page(page); | |
2225 | ret = VM_FAULT_NOPAGE; | |
2226 | goto out; | |
2227 | } | |
14da9200 JK |
2228 | /* |
2229 | * We mark the page dirty already here so that when freeze is in | |
2230 | * progress, we are guaranteed that writeback during freezing will | |
2231 | * see the dirty page and writeprotect it again. | |
2232 | */ | |
2233 | set_page_dirty(page); | |
1d1d1a76 | 2234 | wait_for_stable_page(page); |
4fcf1c62 | 2235 | out: |
14da9200 | 2236 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2237 | return ret; |
2238 | } | |
2239 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2240 | ||
f0f37e2f | 2241 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2242 | .fault = filemap_fault, |
f1820361 | 2243 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2244 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2245 | }; |
2246 | ||
2247 | /* This is used for a general mmap of a disk file */ | |
2248 | ||
2249 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2250 | { | |
2251 | struct address_space *mapping = file->f_mapping; | |
2252 | ||
2253 | if (!mapping->a_ops->readpage) | |
2254 | return -ENOEXEC; | |
2255 | file_accessed(file); | |
2256 | vma->vm_ops = &generic_file_vm_ops; | |
2257 | return 0; | |
2258 | } | |
1da177e4 LT |
2259 | |
2260 | /* | |
2261 | * This is for filesystems which do not implement ->writepage. | |
2262 | */ | |
2263 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2264 | { | |
2265 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2266 | return -EINVAL; | |
2267 | return generic_file_mmap(file, vma); | |
2268 | } | |
2269 | #else | |
2270 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2271 | { | |
2272 | return -ENOSYS; | |
2273 | } | |
2274 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2275 | { | |
2276 | return -ENOSYS; | |
2277 | } | |
2278 | #endif /* CONFIG_MMU */ | |
2279 | ||
2280 | EXPORT_SYMBOL(generic_file_mmap); | |
2281 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2282 | ||
67f9fd91 SL |
2283 | static struct page *wait_on_page_read(struct page *page) |
2284 | { | |
2285 | if (!IS_ERR(page)) { | |
2286 | wait_on_page_locked(page); | |
2287 | if (!PageUptodate(page)) { | |
09cbfeaf | 2288 | put_page(page); |
67f9fd91 SL |
2289 | page = ERR_PTR(-EIO); |
2290 | } | |
2291 | } | |
2292 | return page; | |
2293 | } | |
2294 | ||
32b63529 | 2295 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2296 | pgoff_t index, |
5e5358e7 | 2297 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2298 | void *data, |
2299 | gfp_t gfp) | |
1da177e4 | 2300 | { |
eb2be189 | 2301 | struct page *page; |
1da177e4 LT |
2302 | int err; |
2303 | repeat: | |
2304 | page = find_get_page(mapping, index); | |
2305 | if (!page) { | |
0531b2aa | 2306 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
2307 | if (!page) |
2308 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2309 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 2310 | if (unlikely(err)) { |
09cbfeaf | 2311 | put_page(page); |
eb2be189 NP |
2312 | if (err == -EEXIST) |
2313 | goto repeat; | |
1da177e4 | 2314 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2315 | return ERR_PTR(err); |
2316 | } | |
32b63529 MG |
2317 | |
2318 | filler: | |
1da177e4 LT |
2319 | err = filler(data, page); |
2320 | if (err < 0) { | |
09cbfeaf | 2321 | put_page(page); |
32b63529 | 2322 | return ERR_PTR(err); |
1da177e4 | 2323 | } |
1da177e4 | 2324 | |
32b63529 MG |
2325 | page = wait_on_page_read(page); |
2326 | if (IS_ERR(page)) | |
2327 | return page; | |
2328 | goto out; | |
2329 | } | |
1da177e4 LT |
2330 | if (PageUptodate(page)) |
2331 | goto out; | |
2332 | ||
ebded027 MG |
2333 | /* |
2334 | * Page is not up to date and may be locked due one of the following | |
2335 | * case a: Page is being filled and the page lock is held | |
2336 | * case b: Read/write error clearing the page uptodate status | |
2337 | * case c: Truncation in progress (page locked) | |
2338 | * case d: Reclaim in progress | |
2339 | * | |
2340 | * Case a, the page will be up to date when the page is unlocked. | |
2341 | * There is no need to serialise on the page lock here as the page | |
2342 | * is pinned so the lock gives no additional protection. Even if the | |
2343 | * the page is truncated, the data is still valid if PageUptodate as | |
2344 | * it's a race vs truncate race. | |
2345 | * Case b, the page will not be up to date | |
2346 | * Case c, the page may be truncated but in itself, the data may still | |
2347 | * be valid after IO completes as it's a read vs truncate race. The | |
2348 | * operation must restart if the page is not uptodate on unlock but | |
2349 | * otherwise serialising on page lock to stabilise the mapping gives | |
2350 | * no additional guarantees to the caller as the page lock is | |
2351 | * released before return. | |
2352 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
2353 | * will be a race with remove_mapping that determines if the mapping | |
2354 | * is valid on unlock but otherwise the data is valid and there is | |
2355 | * no need to serialise with page lock. | |
2356 | * | |
2357 | * As the page lock gives no additional guarantee, we optimistically | |
2358 | * wait on the page to be unlocked and check if it's up to date and | |
2359 | * use the page if it is. Otherwise, the page lock is required to | |
2360 | * distinguish between the different cases. The motivation is that we | |
2361 | * avoid spurious serialisations and wakeups when multiple processes | |
2362 | * wait on the same page for IO to complete. | |
2363 | */ | |
2364 | wait_on_page_locked(page); | |
2365 | if (PageUptodate(page)) | |
2366 | goto out; | |
2367 | ||
2368 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 2369 | lock_page(page); |
ebded027 MG |
2370 | |
2371 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
2372 | if (!page->mapping) { |
2373 | unlock_page(page); | |
09cbfeaf | 2374 | put_page(page); |
32b63529 | 2375 | goto repeat; |
1da177e4 | 2376 | } |
ebded027 MG |
2377 | |
2378 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
2379 | if (PageUptodate(page)) { |
2380 | unlock_page(page); | |
2381 | goto out; | |
2382 | } | |
32b63529 MG |
2383 | goto filler; |
2384 | ||
c855ff37 | 2385 | out: |
6fe6900e NP |
2386 | mark_page_accessed(page); |
2387 | return page; | |
2388 | } | |
0531b2aa LT |
2389 | |
2390 | /** | |
67f9fd91 | 2391 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2392 | * @mapping: the page's address_space |
2393 | * @index: the page index | |
2394 | * @filler: function to perform the read | |
5e5358e7 | 2395 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2396 | * |
0531b2aa | 2397 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2398 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2399 | * |
2400 | * If the page does not get brought uptodate, return -EIO. | |
2401 | */ | |
67f9fd91 | 2402 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2403 | pgoff_t index, |
5e5358e7 | 2404 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2405 | void *data) |
2406 | { | |
2407 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2408 | } | |
67f9fd91 | 2409 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2410 | |
2411 | /** | |
2412 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2413 | * @mapping: the page's address_space | |
2414 | * @index: the page index | |
2415 | * @gfp: the page allocator flags to use if allocating | |
2416 | * | |
2417 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2418 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2419 | * |
2420 | * If the page does not get brought uptodate, return -EIO. | |
2421 | */ | |
2422 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2423 | pgoff_t index, | |
2424 | gfp_t gfp) | |
2425 | { | |
2426 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2427 | ||
67f9fd91 | 2428 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2429 | } |
2430 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2431 | ||
1da177e4 LT |
2432 | /* |
2433 | * Performs necessary checks before doing a write | |
2434 | * | |
485bb99b | 2435 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2436 | * Returns appropriate error code that caller should return or |
2437 | * zero in case that write should be allowed. | |
2438 | */ | |
3309dd04 | 2439 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2440 | { |
3309dd04 | 2441 | struct file *file = iocb->ki_filp; |
1da177e4 | 2442 | struct inode *inode = file->f_mapping->host; |
59e99e5b | 2443 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
3309dd04 | 2444 | loff_t pos; |
1da177e4 | 2445 | |
3309dd04 AV |
2446 | if (!iov_iter_count(from)) |
2447 | return 0; | |
1da177e4 | 2448 | |
0fa6b005 | 2449 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2450 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2451 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2452 | |
3309dd04 | 2453 | pos = iocb->ki_pos; |
1da177e4 | 2454 | |
0fa6b005 | 2455 | if (limit != RLIM_INFINITY) { |
3309dd04 | 2456 | if (iocb->ki_pos >= limit) { |
0fa6b005 AV |
2457 | send_sig(SIGXFSZ, current, 0); |
2458 | return -EFBIG; | |
1da177e4 | 2459 | } |
3309dd04 | 2460 | iov_iter_truncate(from, limit - (unsigned long)pos); |
1da177e4 LT |
2461 | } |
2462 | ||
2463 | /* | |
2464 | * LFS rule | |
2465 | */ | |
3309dd04 | 2466 | if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS && |
1da177e4 | 2467 | !(file->f_flags & O_LARGEFILE))) { |
3309dd04 | 2468 | if (pos >= MAX_NON_LFS) |
1da177e4 | 2469 | return -EFBIG; |
3309dd04 | 2470 | iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos); |
1da177e4 LT |
2471 | } |
2472 | ||
2473 | /* | |
2474 | * Are we about to exceed the fs block limit ? | |
2475 | * | |
2476 | * If we have written data it becomes a short write. If we have | |
2477 | * exceeded without writing data we send a signal and return EFBIG. | |
2478 | * Linus frestrict idea will clean these up nicely.. | |
2479 | */ | |
3309dd04 AV |
2480 | if (unlikely(pos >= inode->i_sb->s_maxbytes)) |
2481 | return -EFBIG; | |
1da177e4 | 2482 | |
3309dd04 AV |
2483 | iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos); |
2484 | return iov_iter_count(from); | |
1da177e4 LT |
2485 | } |
2486 | EXPORT_SYMBOL(generic_write_checks); | |
2487 | ||
afddba49 NP |
2488 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2489 | loff_t pos, unsigned len, unsigned flags, | |
2490 | struct page **pagep, void **fsdata) | |
2491 | { | |
2492 | const struct address_space_operations *aops = mapping->a_ops; | |
2493 | ||
4e02ed4b | 2494 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2495 | pagep, fsdata); |
afddba49 NP |
2496 | } |
2497 | EXPORT_SYMBOL(pagecache_write_begin); | |
2498 | ||
2499 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2500 | loff_t pos, unsigned len, unsigned copied, | |
2501 | struct page *page, void *fsdata) | |
2502 | { | |
2503 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2504 | |
4e02ed4b | 2505 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2506 | } |
2507 | EXPORT_SYMBOL(pagecache_write_end); | |
2508 | ||
1da177e4 | 2509 | ssize_t |
0c949334 | 2510 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos) |
1da177e4 LT |
2511 | { |
2512 | struct file *file = iocb->ki_filp; | |
2513 | struct address_space *mapping = file->f_mapping; | |
2514 | struct inode *inode = mapping->host; | |
2515 | ssize_t written; | |
a969e903 CH |
2516 | size_t write_len; |
2517 | pgoff_t end; | |
26978b8b | 2518 | struct iov_iter data; |
1da177e4 | 2519 | |
0c949334 | 2520 | write_len = iov_iter_count(from); |
09cbfeaf | 2521 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 2522 | |
48b47c56 | 2523 | written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); |
a969e903 CH |
2524 | if (written) |
2525 | goto out; | |
2526 | ||
2527 | /* | |
2528 | * After a write we want buffered reads to be sure to go to disk to get | |
2529 | * the new data. We invalidate clean cached page from the region we're | |
2530 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2531 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 CH |
2532 | */ |
2533 | if (mapping->nrpages) { | |
2534 | written = invalidate_inode_pages2_range(mapping, | |
09cbfeaf | 2535 | pos >> PAGE_SHIFT, end); |
6ccfa806 HH |
2536 | /* |
2537 | * If a page can not be invalidated, return 0 to fall back | |
2538 | * to buffered write. | |
2539 | */ | |
2540 | if (written) { | |
2541 | if (written == -EBUSY) | |
2542 | return 0; | |
a969e903 | 2543 | goto out; |
6ccfa806 | 2544 | } |
a969e903 CH |
2545 | } |
2546 | ||
26978b8b | 2547 | data = *from; |
22c6186e | 2548 | written = mapping->a_ops->direct_IO(iocb, &data, pos); |
a969e903 CH |
2549 | |
2550 | /* | |
2551 | * Finally, try again to invalidate clean pages which might have been | |
2552 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2553 | * if the source of the write was an mmap'ed region of the file | |
2554 | * we're writing. Either one is a pretty crazy thing to do, | |
2555 | * so we don't support it 100%. If this invalidation | |
2556 | * fails, tough, the write still worked... | |
2557 | */ | |
2558 | if (mapping->nrpages) { | |
2559 | invalidate_inode_pages2_range(mapping, | |
09cbfeaf | 2560 | pos >> PAGE_SHIFT, end); |
a969e903 CH |
2561 | } |
2562 | ||
1da177e4 | 2563 | if (written > 0) { |
0116651c | 2564 | pos += written; |
f8579f86 | 2565 | iov_iter_advance(from, written); |
0116651c NK |
2566 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
2567 | i_size_write(inode, pos); | |
1da177e4 LT |
2568 | mark_inode_dirty(inode); |
2569 | } | |
5cb6c6c7 | 2570 | iocb->ki_pos = pos; |
1da177e4 | 2571 | } |
a969e903 | 2572 | out: |
1da177e4 LT |
2573 | return written; |
2574 | } | |
2575 | EXPORT_SYMBOL(generic_file_direct_write); | |
2576 | ||
eb2be189 NP |
2577 | /* |
2578 | * Find or create a page at the given pagecache position. Return the locked | |
2579 | * page. This function is specifically for buffered writes. | |
2580 | */ | |
54566b2c NP |
2581 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2582 | pgoff_t index, unsigned flags) | |
eb2be189 | 2583 | { |
eb2be189 | 2584 | struct page *page; |
2457aec6 | 2585 | int fgp_flags = FGP_LOCK|FGP_ACCESSED|FGP_WRITE|FGP_CREAT; |
0faa70cb | 2586 | |
54566b2c | 2587 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
2588 | fgp_flags |= FGP_NOFS; |
2589 | ||
2590 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 2591 | mapping_gfp_mask(mapping)); |
c585a267 | 2592 | if (page) |
2457aec6 | 2593 | wait_for_stable_page(page); |
eb2be189 | 2594 | |
eb2be189 NP |
2595 | return page; |
2596 | } | |
54566b2c | 2597 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2598 | |
3b93f911 | 2599 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
2600 | struct iov_iter *i, loff_t pos) |
2601 | { | |
2602 | struct address_space *mapping = file->f_mapping; | |
2603 | const struct address_space_operations *a_ops = mapping->a_ops; | |
2604 | long status = 0; | |
2605 | ssize_t written = 0; | |
674b892e NP |
2606 | unsigned int flags = 0; |
2607 | ||
2608 | /* | |
2609 | * Copies from kernel address space cannot fail (NFSD is a big user). | |
2610 | */ | |
777eda2c | 2611 | if (!iter_is_iovec(i)) |
674b892e | 2612 | flags |= AOP_FLAG_UNINTERRUPTIBLE; |
afddba49 NP |
2613 | |
2614 | do { | |
2615 | struct page *page; | |
afddba49 NP |
2616 | unsigned long offset; /* Offset into pagecache page */ |
2617 | unsigned long bytes; /* Bytes to write to page */ | |
2618 | size_t copied; /* Bytes copied from user */ | |
2619 | void *fsdata; | |
2620 | ||
09cbfeaf KS |
2621 | offset = (pos & (PAGE_SIZE - 1)); |
2622 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
2623 | iov_iter_count(i)); |
2624 | ||
2625 | again: | |
00a3d660 LT |
2626 | /* |
2627 | * Bring in the user page that we will copy from _first_. | |
2628 | * Otherwise there's a nasty deadlock on copying from the | |
2629 | * same page as we're writing to, without it being marked | |
2630 | * up-to-date. | |
2631 | * | |
2632 | * Not only is this an optimisation, but it is also required | |
2633 | * to check that the address is actually valid, when atomic | |
2634 | * usercopies are used, below. | |
2635 | */ | |
2636 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
2637 | status = -EFAULT; | |
2638 | break; | |
2639 | } | |
2640 | ||
296291cd JK |
2641 | if (fatal_signal_pending(current)) { |
2642 | status = -EINTR; | |
2643 | break; | |
2644 | } | |
2645 | ||
674b892e | 2646 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 2647 | &page, &fsdata); |
2457aec6 | 2648 | if (unlikely(status < 0)) |
afddba49 NP |
2649 | break; |
2650 | ||
931e80e4 | 2651 | if (mapping_writably_mapped(mapping)) |
2652 | flush_dcache_page(page); | |
00a3d660 | 2653 | |
afddba49 | 2654 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
2655 | flush_dcache_page(page); |
2656 | ||
2657 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
2658 | page, fsdata); | |
2659 | if (unlikely(status < 0)) | |
2660 | break; | |
2661 | copied = status; | |
2662 | ||
2663 | cond_resched(); | |
2664 | ||
124d3b70 | 2665 | iov_iter_advance(i, copied); |
afddba49 NP |
2666 | if (unlikely(copied == 0)) { |
2667 | /* | |
2668 | * If we were unable to copy any data at all, we must | |
2669 | * fall back to a single segment length write. | |
2670 | * | |
2671 | * If we didn't fallback here, we could livelock | |
2672 | * because not all segments in the iov can be copied at | |
2673 | * once without a pagefault. | |
2674 | */ | |
09cbfeaf | 2675 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
afddba49 NP |
2676 | iov_iter_single_seg_count(i)); |
2677 | goto again; | |
2678 | } | |
afddba49 NP |
2679 | pos += copied; |
2680 | written += copied; | |
2681 | ||
2682 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
2683 | } while (iov_iter_count(i)); |
2684 | ||
2685 | return written ? written : status; | |
2686 | } | |
3b93f911 | 2687 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 2688 | |
e4dd9de3 | 2689 | /** |
8174202b | 2690 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 2691 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 2692 | * @from: iov_iter with data to write |
e4dd9de3 JK |
2693 | * |
2694 | * This function does all the work needed for actually writing data to a | |
2695 | * file. It does all basic checks, removes SUID from the file, updates | |
2696 | * modification times and calls proper subroutines depending on whether we | |
2697 | * do direct IO or a standard buffered write. | |
2698 | * | |
2699 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
2700 | * object which does not need locking at all. | |
2701 | * | |
2702 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
2703 | * A caller has to handle it. This is mainly due to the fact that we want to | |
2704 | * avoid syncing under i_mutex. | |
2705 | */ | |
8174202b | 2706 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2707 | { |
2708 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 2709 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 2710 | struct inode *inode = mapping->host; |
3b93f911 | 2711 | ssize_t written = 0; |
1da177e4 | 2712 | ssize_t err; |
3b93f911 | 2713 | ssize_t status; |
1da177e4 | 2714 | |
1da177e4 | 2715 | /* We can write back this queue in page reclaim */ |
de1414a6 | 2716 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 2717 | err = file_remove_privs(file); |
1da177e4 LT |
2718 | if (err) |
2719 | goto out; | |
2720 | ||
c3b2da31 JB |
2721 | err = file_update_time(file); |
2722 | if (err) | |
2723 | goto out; | |
1da177e4 | 2724 | |
2ba48ce5 | 2725 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 2726 | loff_t pos, endbyte; |
fb5527e6 | 2727 | |
0b8def9d | 2728 | written = generic_file_direct_write(iocb, from, iocb->ki_pos); |
1da177e4 | 2729 | /* |
fbbbad4b MW |
2730 | * If the write stopped short of completing, fall back to |
2731 | * buffered writes. Some filesystems do this for writes to | |
2732 | * holes, for example. For DAX files, a buffered write will | |
2733 | * not succeed (even if it did, DAX does not handle dirty | |
2734 | * page-cache pages correctly). | |
1da177e4 | 2735 | */ |
0b8def9d | 2736 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
2737 | goto out; |
2738 | ||
0b8def9d | 2739 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 2740 | /* |
3b93f911 | 2741 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
2742 | * then we want to return the number of bytes which were |
2743 | * direct-written, or the error code if that was zero. Note | |
2744 | * that this differs from normal direct-io semantics, which | |
2745 | * will return -EFOO even if some bytes were written. | |
2746 | */ | |
60bb4529 | 2747 | if (unlikely(status < 0)) { |
3b93f911 | 2748 | err = status; |
fb5527e6 JM |
2749 | goto out; |
2750 | } | |
fb5527e6 JM |
2751 | /* |
2752 | * We need to ensure that the page cache pages are written to | |
2753 | * disk and invalidated to preserve the expected O_DIRECT | |
2754 | * semantics. | |
2755 | */ | |
3b93f911 | 2756 | endbyte = pos + status - 1; |
0b8def9d | 2757 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 2758 | if (err == 0) { |
0b8def9d | 2759 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 2760 | written += status; |
fb5527e6 | 2761 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
2762 | pos >> PAGE_SHIFT, |
2763 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
2764 | } else { |
2765 | /* | |
2766 | * We don't know how much we wrote, so just return | |
2767 | * the number of bytes which were direct-written | |
2768 | */ | |
2769 | } | |
2770 | } else { | |
0b8def9d AV |
2771 | written = generic_perform_write(file, from, iocb->ki_pos); |
2772 | if (likely(written > 0)) | |
2773 | iocb->ki_pos += written; | |
fb5527e6 | 2774 | } |
1da177e4 LT |
2775 | out: |
2776 | current->backing_dev_info = NULL; | |
2777 | return written ? written : err; | |
2778 | } | |
8174202b | 2779 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 2780 | |
e4dd9de3 | 2781 | /** |
8174202b | 2782 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 2783 | * @iocb: IO state structure |
8174202b | 2784 | * @from: iov_iter with data to write |
e4dd9de3 | 2785 | * |
8174202b | 2786 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
2787 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
2788 | * and acquires i_mutex as needed. | |
2789 | */ | |
8174202b | 2790 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2791 | { |
2792 | struct file *file = iocb->ki_filp; | |
148f948b | 2793 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 2794 | ssize_t ret; |
1da177e4 | 2795 | |
5955102c | 2796 | inode_lock(inode); |
3309dd04 AV |
2797 | ret = generic_write_checks(iocb, from); |
2798 | if (ret > 0) | |
5f380c7f | 2799 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 2800 | inode_unlock(inode); |
1da177e4 | 2801 | |
02afc27f | 2802 | if (ret > 0) { |
1da177e4 LT |
2803 | ssize_t err; |
2804 | ||
d311d79d AV |
2805 | err = generic_write_sync(file, iocb->ki_pos - ret, ret); |
2806 | if (err < 0) | |
1da177e4 LT |
2807 | ret = err; |
2808 | } | |
2809 | return ret; | |
2810 | } | |
8174202b | 2811 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 2812 | |
cf9a2ae8 DH |
2813 | /** |
2814 | * try_to_release_page() - release old fs-specific metadata on a page | |
2815 | * | |
2816 | * @page: the page which the kernel is trying to free | |
2817 | * @gfp_mask: memory allocation flags (and I/O mode) | |
2818 | * | |
2819 | * The address_space is to try to release any data against the page | |
2820 | * (presumably at page->private). If the release was successful, return `1'. | |
2821 | * Otherwise return zero. | |
2822 | * | |
266cf658 DH |
2823 | * This may also be called if PG_fscache is set on a page, indicating that the |
2824 | * page is known to the local caching routines. | |
2825 | * | |
cf9a2ae8 | 2826 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 2827 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 2828 | * |
cf9a2ae8 DH |
2829 | */ |
2830 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
2831 | { | |
2832 | struct address_space * const mapping = page->mapping; | |
2833 | ||
2834 | BUG_ON(!PageLocked(page)); | |
2835 | if (PageWriteback(page)) | |
2836 | return 0; | |
2837 | ||
2838 | if (mapping && mapping->a_ops->releasepage) | |
2839 | return mapping->a_ops->releasepage(page, gfp_mask); | |
2840 | return try_to_free_buffers(page); | |
2841 | } | |
2842 | ||
2843 | EXPORT_SYMBOL(try_to_release_page); |