]>
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> |
3f07c014 | 16 | #include <linux/sched/signal.h> |
c22ce143 | 17 | #include <linux/uaccess.h> |
c59ede7b | 18 | #include <linux/capability.h> |
1da177e4 | 19 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 20 | #include <linux/gfp.h> |
1da177e4 LT |
21 | #include <linux/mm.h> |
22 | #include <linux/swap.h> | |
23 | #include <linux/mman.h> | |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/file.h> | |
26 | #include <linux/uio.h> | |
27 | #include <linux/hash.h> | |
28 | #include <linux/writeback.h> | |
53253383 | 29 | #include <linux/backing-dev.h> |
1da177e4 LT |
30 | #include <linux/pagevec.h> |
31 | #include <linux/blkdev.h> | |
32 | #include <linux/security.h> | |
44110fe3 | 33 | #include <linux/cpuset.h> |
2f718ffc | 34 | #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ |
00501b53 | 35 | #include <linux/hugetlb.h> |
8a9f3ccd | 36 | #include <linux/memcontrol.h> |
c515e1fd | 37 | #include <linux/cleancache.h> |
f1820361 | 38 | #include <linux/rmap.h> |
0f8053a5 NP |
39 | #include "internal.h" |
40 | ||
fe0bfaaf RJ |
41 | #define CREATE_TRACE_POINTS |
42 | #include <trace/events/filemap.h> | |
43 | ||
1da177e4 | 44 | /* |
1da177e4 LT |
45 | * FIXME: remove all knowledge of the buffer layer from the core VM |
46 | */ | |
148f948b | 47 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 48 | |
1da177e4 LT |
49 | #include <asm/mman.h> |
50 | ||
51 | /* | |
52 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
53 | * though. | |
54 | * | |
55 | * Shared mappings now work. 15.8.1995 Bruno. | |
56 | * | |
57 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
58 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
59 | * | |
60 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
61 | */ | |
62 | ||
63 | /* | |
64 | * Lock ordering: | |
65 | * | |
c8c06efa | 66 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 67 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 HD |
68 | * ->swap_lock (exclusive_swap_page, others) |
69 | * ->mapping->tree_lock | |
1da177e4 | 70 | * |
1b1dcc1b | 71 | * ->i_mutex |
c8c06efa | 72 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
1da177e4 LT |
73 | * |
74 | * ->mmap_sem | |
c8c06efa | 75 | * ->i_mmap_rwsem |
b8072f09 | 76 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
1da177e4 LT |
77 | * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) |
78 | * | |
79 | * ->mmap_sem | |
80 | * ->lock_page (access_process_vm) | |
81 | * | |
ccad2365 | 82 | * ->i_mutex (generic_perform_write) |
82591e6e | 83 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
1da177e4 | 84 | * |
f758eeab | 85 | * bdi->wb.list_lock |
a66979ab | 86 | * sb_lock (fs/fs-writeback.c) |
1da177e4 LT |
87 | * ->mapping->tree_lock (__sync_single_inode) |
88 | * | |
c8c06efa | 89 | * ->i_mmap_rwsem |
1da177e4 LT |
90 | * ->anon_vma.lock (vma_adjust) |
91 | * | |
92 | * ->anon_vma.lock | |
b8072f09 | 93 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 94 | * |
b8072f09 | 95 | * ->page_table_lock or pte_lock |
5d337b91 | 96 | * ->swap_lock (try_to_unmap_one) |
1da177e4 LT |
97 | * ->private_lock (try_to_unmap_one) |
98 | * ->tree_lock (try_to_unmap_one) | |
a52633d8 MG |
99 | * ->zone_lru_lock(zone) (follow_page->mark_page_accessed) |
100 | * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page) | |
1da177e4 LT |
101 | * ->private_lock (page_remove_rmap->set_page_dirty) |
102 | * ->tree_lock (page_remove_rmap->set_page_dirty) | |
f758eeab | 103 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 104 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 105 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 106 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 107 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
108 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
109 | * | |
c8c06efa | 110 | * ->i_mmap_rwsem |
9a3c531d | 111 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
112 | */ |
113 | ||
22f2ac51 JW |
114 | static int page_cache_tree_insert(struct address_space *mapping, |
115 | struct page *page, void **shadowp) | |
116 | { | |
117 | struct radix_tree_node *node; | |
118 | void **slot; | |
119 | int error; | |
120 | ||
121 | error = __radix_tree_create(&mapping->page_tree, page->index, 0, | |
122 | &node, &slot); | |
123 | if (error) | |
124 | return error; | |
125 | if (*slot) { | |
126 | void *p; | |
127 | ||
128 | p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock); | |
129 | if (!radix_tree_exceptional_entry(p)) | |
130 | return -EEXIST; | |
131 | ||
132 | mapping->nrexceptional--; | |
d01ad197 RZ |
133 | if (shadowp) |
134 | *shadowp = p; | |
22f2ac51 | 135 | } |
14b46879 JW |
136 | __radix_tree_replace(&mapping->page_tree, node, slot, page, |
137 | workingset_update_node, mapping); | |
22f2ac51 | 138 | mapping->nrpages++; |
22f2ac51 JW |
139 | return 0; |
140 | } | |
141 | ||
91b0abe3 JW |
142 | static void page_cache_tree_delete(struct address_space *mapping, |
143 | struct page *page, void *shadow) | |
144 | { | |
c70b647d KS |
145 | int i, nr; |
146 | ||
147 | /* hugetlb pages are represented by one entry in the radix tree */ | |
148 | nr = PageHuge(page) ? 1 : hpage_nr_pages(page); | |
91b0abe3 | 149 | |
83929372 KS |
150 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
151 | VM_BUG_ON_PAGE(PageTail(page), page); | |
152 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 153 | |
83929372 | 154 | for (i = 0; i < nr; i++) { |
d3798ae8 JW |
155 | struct radix_tree_node *node; |
156 | void **slot; | |
157 | ||
158 | __radix_tree_lookup(&mapping->page_tree, page->index + i, | |
159 | &node, &slot); | |
160 | ||
dbc446b8 | 161 | VM_BUG_ON_PAGE(!node && nr != 1, page); |
449dd698 | 162 | |
14b46879 JW |
163 | radix_tree_clear_tags(&mapping->page_tree, node, slot); |
164 | __radix_tree_replace(&mapping->page_tree, node, slot, shadow, | |
165 | workingset_update_node, mapping); | |
449dd698 | 166 | } |
d3798ae8 JW |
167 | |
168 | if (shadow) { | |
169 | mapping->nrexceptional += nr; | |
170 | /* | |
171 | * Make sure the nrexceptional update is committed before | |
172 | * the nrpages update so that final truncate racing | |
173 | * with reclaim does not see both counters 0 at the | |
174 | * same time and miss a shadow entry. | |
175 | */ | |
176 | smp_wmb(); | |
177 | } | |
178 | mapping->nrpages -= nr; | |
91b0abe3 JW |
179 | } |
180 | ||
1da177e4 | 181 | /* |
e64a782f | 182 | * Delete a page from the page cache and free it. Caller has to make |
1da177e4 | 183 | * sure the page is locked and that nobody else uses it - or that usage |
fdf1cdb9 | 184 | * is safe. The caller must hold the mapping's tree_lock. |
1da177e4 | 185 | */ |
62cccb8c | 186 | void __delete_from_page_cache(struct page *page, void *shadow) |
1da177e4 LT |
187 | { |
188 | struct address_space *mapping = page->mapping; | |
83929372 | 189 | int nr = hpage_nr_pages(page); |
1da177e4 | 190 | |
fe0bfaaf | 191 | trace_mm_filemap_delete_from_page_cache(page); |
c515e1fd DM |
192 | /* |
193 | * if we're uptodate, flush out into the cleancache, otherwise | |
194 | * invalidate any existing cleancache entries. We can't leave | |
195 | * stale data around in the cleancache once our page is gone | |
196 | */ | |
197 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
198 | cleancache_put_page(page); | |
199 | else | |
3167760f | 200 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 201 | |
83929372 | 202 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
203 | VM_BUG_ON_PAGE(page_mapped(page), page); |
204 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
205 | int mapcount; | |
206 | ||
207 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
208 | current->comm, page_to_pfn(page)); | |
209 | dump_page(page, "still mapped when deleted"); | |
210 | dump_stack(); | |
211 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
212 | ||
213 | mapcount = page_mapcount(page); | |
214 | if (mapping_exiting(mapping) && | |
215 | page_count(page) >= mapcount + 2) { | |
216 | /* | |
217 | * All vmas have already been torn down, so it's | |
218 | * a good bet that actually the page is unmapped, | |
219 | * and we'd prefer not to leak it: if we're wrong, | |
220 | * some other bad page check should catch it later. | |
221 | */ | |
222 | page_mapcount_reset(page); | |
6d061f9f | 223 | page_ref_sub(page, mapcount); |
06b241f3 HD |
224 | } |
225 | } | |
226 | ||
91b0abe3 JW |
227 | page_cache_tree_delete(mapping, page, shadow); |
228 | ||
1da177e4 | 229 | page->mapping = NULL; |
b85e0eff | 230 | /* Leave page->index set: truncation lookup relies upon it */ |
91b0abe3 | 231 | |
4165b9b4 | 232 | /* hugetlb pages do not participate in page cache accounting. */ |
09612fa6 NH |
233 | if (PageHuge(page)) |
234 | return; | |
235 | ||
236 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr); | |
800d8c63 | 237 | if (PageSwapBacked(page)) { |
11fb9989 | 238 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr); |
800d8c63 | 239 | if (PageTransHuge(page)) |
11fb9989 | 240 | __dec_node_page_state(page, NR_SHMEM_THPS); |
800d8c63 | 241 | } else { |
09612fa6 | 242 | VM_BUG_ON_PAGE(PageTransHuge(page), page); |
800d8c63 | 243 | } |
3a692790 LT |
244 | |
245 | /* | |
b9ea2515 KK |
246 | * At this point page must be either written or cleaned by truncate. |
247 | * Dirty page here signals a bug and loss of unwritten data. | |
3a692790 | 248 | * |
b9ea2515 KK |
249 | * This fixes dirty accounting after removing the page entirely but |
250 | * leaves PageDirty set: it has no effect for truncated page and | |
251 | * anyway will be cleared before returning page into buddy allocator. | |
3a692790 | 252 | */ |
b9ea2515 | 253 | if (WARN_ON_ONCE(PageDirty(page))) |
62cccb8c | 254 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); |
1da177e4 LT |
255 | } |
256 | ||
702cfbf9 MK |
257 | /** |
258 | * delete_from_page_cache - delete page from page cache | |
259 | * @page: the page which the kernel is trying to remove from page cache | |
260 | * | |
261 | * This must be called only on pages that have been verified to be in the page | |
262 | * cache and locked. It will never put the page into the free list, the caller | |
263 | * has a reference on the page. | |
264 | */ | |
265 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 266 | { |
83929372 | 267 | struct address_space *mapping = page_mapping(page); |
c4843a75 | 268 | unsigned long flags; |
6072d13c | 269 | void (*freepage)(struct page *); |
1da177e4 | 270 | |
cd7619d6 | 271 | BUG_ON(!PageLocked(page)); |
1da177e4 | 272 | |
6072d13c | 273 | freepage = mapping->a_ops->freepage; |
c4843a75 | 274 | |
c4843a75 | 275 | spin_lock_irqsave(&mapping->tree_lock, flags); |
62cccb8c | 276 | __delete_from_page_cache(page, NULL); |
c4843a75 | 277 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6072d13c LT |
278 | |
279 | if (freepage) | |
280 | freepage(page); | |
83929372 KS |
281 | |
282 | if (PageTransHuge(page) && !PageHuge(page)) { | |
283 | page_ref_sub(page, HPAGE_PMD_NR); | |
284 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); | |
285 | } else { | |
286 | put_page(page); | |
287 | } | |
97cecb5a MK |
288 | } |
289 | EXPORT_SYMBOL(delete_from_page_cache); | |
290 | ||
d72d9e2a | 291 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
292 | { |
293 | int ret = 0; | |
294 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
295 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
296 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 297 | ret = -ENOSPC; |
7fcbbaf1 JA |
298 | if (test_bit(AS_EIO, &mapping->flags) && |
299 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
300 | ret = -EIO; |
301 | return ret; | |
302 | } | |
d72d9e2a | 303 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 304 | |
76341cab JL |
305 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
306 | { | |
307 | /* Check for outstanding write errors */ | |
308 | if (test_bit(AS_EIO, &mapping->flags)) | |
309 | return -EIO; | |
310 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
311 | return -ENOSPC; | |
312 | return 0; | |
313 | } | |
314 | ||
1da177e4 | 315 | /** |
485bb99b | 316 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
317 | * @mapping: address space structure to write |
318 | * @start: offset in bytes where the range starts | |
469eb4d0 | 319 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 320 | * @sync_mode: enable synchronous operation |
1da177e4 | 321 | * |
485bb99b RD |
322 | * Start writeback against all of a mapping's dirty pages that lie |
323 | * within the byte offsets <start, end> inclusive. | |
324 | * | |
1da177e4 | 325 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 326 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
327 | * these two operations is that if a dirty page/buffer is encountered, it must |
328 | * be waited upon, and not just skipped over. | |
329 | */ | |
ebcf28e1 AM |
330 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
331 | loff_t end, int sync_mode) | |
1da177e4 LT |
332 | { |
333 | int ret; | |
334 | struct writeback_control wbc = { | |
335 | .sync_mode = sync_mode, | |
05fe478d | 336 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
337 | .range_start = start, |
338 | .range_end = end, | |
1da177e4 LT |
339 | }; |
340 | ||
341 | if (!mapping_cap_writeback_dirty(mapping)) | |
342 | return 0; | |
343 | ||
b16b1deb | 344 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 345 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 346 | wbc_detach_inode(&wbc); |
1da177e4 LT |
347 | return ret; |
348 | } | |
349 | ||
350 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
351 | int sync_mode) | |
352 | { | |
111ebb6e | 353 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
354 | } |
355 | ||
356 | int filemap_fdatawrite(struct address_space *mapping) | |
357 | { | |
358 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
359 | } | |
360 | EXPORT_SYMBOL(filemap_fdatawrite); | |
361 | ||
f4c0a0fd | 362 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 363 | loff_t end) |
1da177e4 LT |
364 | { |
365 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
366 | } | |
f4c0a0fd | 367 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 368 | |
485bb99b RD |
369 | /** |
370 | * filemap_flush - mostly a non-blocking flush | |
371 | * @mapping: target address_space | |
372 | * | |
1da177e4 LT |
373 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
374 | * purposes - I/O may not be started against all dirty pages. | |
375 | */ | |
376 | int filemap_flush(struct address_space *mapping) | |
377 | { | |
378 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
379 | } | |
380 | EXPORT_SYMBOL(filemap_flush); | |
381 | ||
7fc9e472 GR |
382 | /** |
383 | * filemap_range_has_page - check if a page exists in range. | |
384 | * @mapping: address space within which to check | |
385 | * @start_byte: offset in bytes where the range starts | |
386 | * @end_byte: offset in bytes where the range ends (inclusive) | |
387 | * | |
388 | * Find at least one page in the range supplied, usually used to check if | |
389 | * direct writing in this range will trigger a writeback. | |
390 | */ | |
391 | bool filemap_range_has_page(struct address_space *mapping, | |
392 | loff_t start_byte, loff_t end_byte) | |
393 | { | |
394 | pgoff_t index = start_byte >> PAGE_SHIFT; | |
395 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
f7b68046 | 396 | struct page *page; |
7fc9e472 GR |
397 | |
398 | if (end_byte < start_byte) | |
399 | return false; | |
400 | ||
401 | if (mapping->nrpages == 0) | |
402 | return false; | |
403 | ||
f7b68046 | 404 | if (!find_get_pages_range(mapping, &index, end, 1, &page)) |
7fc9e472 | 405 | return false; |
f7b68046 JK |
406 | put_page(page); |
407 | return true; | |
7fc9e472 GR |
408 | } |
409 | EXPORT_SYMBOL(filemap_range_has_page); | |
410 | ||
5e8fcc1a | 411 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 412 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 413 | { |
09cbfeaf KS |
414 | pgoff_t index = start_byte >> PAGE_SHIFT; |
415 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
416 | struct pagevec pvec; |
417 | int nr_pages; | |
1da177e4 | 418 | |
94004ed7 | 419 | if (end_byte < start_byte) |
5e8fcc1a | 420 | return; |
1da177e4 LT |
421 | |
422 | pagevec_init(&pvec, 0); | |
312e9d2f | 423 | while (index <= end) { |
1da177e4 LT |
424 | unsigned i; |
425 | ||
312e9d2f JK |
426 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
427 | end, PAGECACHE_TAG_WRITEBACK, PAGEVEC_SIZE); | |
428 | if (!nr_pages) | |
429 | break; | |
430 | ||
1da177e4 LT |
431 | for (i = 0; i < nr_pages; i++) { |
432 | struct page *page = pvec.pages[i]; | |
433 | ||
1da177e4 | 434 | wait_on_page_writeback(page); |
5e8fcc1a | 435 | ClearPageError(page); |
1da177e4 LT |
436 | } |
437 | pagevec_release(&pvec); | |
438 | cond_resched(); | |
439 | } | |
aa750fd7 JN |
440 | } |
441 | ||
442 | /** | |
443 | * filemap_fdatawait_range - wait for writeback to complete | |
444 | * @mapping: address space structure to wait for | |
445 | * @start_byte: offset in bytes where the range starts | |
446 | * @end_byte: offset in bytes where the range ends (inclusive) | |
447 | * | |
448 | * Walk the list of under-writeback pages of the given address space | |
449 | * in the given range and wait for all of them. Check error status of | |
450 | * the address space and return it. | |
451 | * | |
452 | * Since the error status of the address space is cleared by this function, | |
453 | * callers are responsible for checking the return value and handling and/or | |
454 | * reporting the error. | |
455 | */ | |
456 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
457 | loff_t end_byte) | |
458 | { | |
5e8fcc1a JL |
459 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
460 | return filemap_check_errors(mapping); | |
1da177e4 | 461 | } |
d3bccb6f JK |
462 | EXPORT_SYMBOL(filemap_fdatawait_range); |
463 | ||
a823e458 JL |
464 | /** |
465 | * file_fdatawait_range - wait for writeback to complete | |
466 | * @file: file pointing to address space structure to wait for | |
467 | * @start_byte: offset in bytes where the range starts | |
468 | * @end_byte: offset in bytes where the range ends (inclusive) | |
469 | * | |
470 | * Walk the list of under-writeback pages of the address space that file | |
471 | * refers to, in the given range and wait for all of them. Check error | |
472 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
473 | * | |
474 | * Since the error status of the file is advanced by this function, | |
475 | * callers are responsible for checking the return value and handling and/or | |
476 | * reporting the error. | |
477 | */ | |
478 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
479 | { | |
480 | struct address_space *mapping = file->f_mapping; | |
481 | ||
482 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
483 | return file_check_and_advance_wb_err(file); | |
484 | } | |
485 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 486 | |
aa750fd7 JN |
487 | /** |
488 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
489 | * @mapping: address space structure to wait for | |
490 | * | |
491 | * Walk the list of under-writeback pages of the given address space | |
492 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
493 | * does not clear error status of the address space. | |
494 | * | |
495 | * Use this function if callers don't handle errors themselves. Expected | |
496 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
497 | * fsfreeze(8) | |
498 | */ | |
76341cab | 499 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 500 | { |
ffb959bb | 501 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 502 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 503 | } |
76341cab | 504 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 505 | |
9326c9b2 | 506 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 507 | { |
9326c9b2 JL |
508 | return (!dax_mapping(mapping) && mapping->nrpages) || |
509 | (dax_mapping(mapping) && mapping->nrexceptional); | |
1da177e4 | 510 | } |
1da177e4 LT |
511 | |
512 | int filemap_write_and_wait(struct address_space *mapping) | |
513 | { | |
28fd1298 | 514 | int err = 0; |
1da177e4 | 515 | |
9326c9b2 | 516 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
517 | err = filemap_fdatawrite(mapping); |
518 | /* | |
519 | * Even if the above returned error, the pages may be | |
520 | * written partially (e.g. -ENOSPC), so we wait for it. | |
521 | * But the -EIO is special case, it may indicate the worst | |
522 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
523 | */ | |
524 | if (err != -EIO) { | |
525 | int err2 = filemap_fdatawait(mapping); | |
526 | if (!err) | |
527 | err = err2; | |
cbeaf951 JL |
528 | } else { |
529 | /* Clear any previously stored errors */ | |
530 | filemap_check_errors(mapping); | |
28fd1298 | 531 | } |
865ffef3 DM |
532 | } else { |
533 | err = filemap_check_errors(mapping); | |
1da177e4 | 534 | } |
28fd1298 | 535 | return err; |
1da177e4 | 536 | } |
28fd1298 | 537 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 538 | |
485bb99b RD |
539 | /** |
540 | * filemap_write_and_wait_range - write out & wait on a file range | |
541 | * @mapping: the address_space for the pages | |
542 | * @lstart: offset in bytes where the range starts | |
543 | * @lend: offset in bytes where the range ends (inclusive) | |
544 | * | |
469eb4d0 AM |
545 | * Write out and wait upon file offsets lstart->lend, inclusive. |
546 | * | |
0e056eb5 | 547 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 AM |
548 | * that this function can be used to write to the very end-of-file (end = -1). |
549 | */ | |
1da177e4 LT |
550 | int filemap_write_and_wait_range(struct address_space *mapping, |
551 | loff_t lstart, loff_t lend) | |
552 | { | |
28fd1298 | 553 | int err = 0; |
1da177e4 | 554 | |
9326c9b2 | 555 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
556 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
557 | WB_SYNC_ALL); | |
558 | /* See comment of filemap_write_and_wait() */ | |
559 | if (err != -EIO) { | |
94004ed7 CH |
560 | int err2 = filemap_fdatawait_range(mapping, |
561 | lstart, lend); | |
28fd1298 OH |
562 | if (!err) |
563 | err = err2; | |
cbeaf951 JL |
564 | } else { |
565 | /* Clear any previously stored errors */ | |
566 | filemap_check_errors(mapping); | |
28fd1298 | 567 | } |
865ffef3 DM |
568 | } else { |
569 | err = filemap_check_errors(mapping); | |
1da177e4 | 570 | } |
28fd1298 | 571 | return err; |
1da177e4 | 572 | } |
f6995585 | 573 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 574 | |
5660e13d JL |
575 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
576 | { | |
3acdfd28 | 577 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
578 | |
579 | trace_filemap_set_wb_err(mapping, eseq); | |
580 | } | |
581 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
582 | ||
583 | /** | |
584 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
585 | * and advance wb_err to current one | |
586 | * @file: struct file on which the error is being reported | |
587 | * | |
588 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
589 | * want to report any writeback errors that occurred since the last fsync (or | |
590 | * since the file was opened if there haven't been any). | |
591 | * | |
592 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
593 | * then just quickly return 0. The file is all caught up. | |
594 | * | |
595 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
596 | * it and try to swap it into place. If it works, or another task beat us | |
597 | * to it with the new value, then update the f_wb_err and return the error | |
598 | * portion. The error at this point must be reported via proper channels | |
599 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
600 | * | |
601 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
602 | * value is protected by the f_lock since we must ensure that it reflects | |
603 | * the latest value swapped in for this file descriptor. | |
604 | */ | |
605 | int file_check_and_advance_wb_err(struct file *file) | |
606 | { | |
607 | int err = 0; | |
608 | errseq_t old = READ_ONCE(file->f_wb_err); | |
609 | struct address_space *mapping = file->f_mapping; | |
610 | ||
611 | /* Locklessly handle the common case where nothing has changed */ | |
612 | if (errseq_check(&mapping->wb_err, old)) { | |
613 | /* Something changed, must use slow path */ | |
614 | spin_lock(&file->f_lock); | |
615 | old = file->f_wb_err; | |
616 | err = errseq_check_and_advance(&mapping->wb_err, | |
617 | &file->f_wb_err); | |
618 | trace_file_check_and_advance_wb_err(file, old); | |
619 | spin_unlock(&file->f_lock); | |
620 | } | |
f4e222c5 JL |
621 | |
622 | /* | |
623 | * We're mostly using this function as a drop in replacement for | |
624 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
625 | * that the legacy code would have had on these flags. | |
626 | */ | |
627 | clear_bit(AS_EIO, &mapping->flags); | |
628 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
629 | return err; |
630 | } | |
631 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
632 | ||
633 | /** | |
634 | * file_write_and_wait_range - write out & wait on a file range | |
635 | * @file: file pointing to address_space with pages | |
636 | * @lstart: offset in bytes where the range starts | |
637 | * @lend: offset in bytes where the range ends (inclusive) | |
638 | * | |
639 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
640 | * | |
641 | * Note that @lend is inclusive (describes the last byte to be written) so | |
642 | * that this function can be used to write to the very end-of-file (end = -1). | |
643 | * | |
644 | * After writing out and waiting on the data, we check and advance the | |
645 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
646 | */ | |
647 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
648 | { | |
649 | int err = 0, err2; | |
650 | struct address_space *mapping = file->f_mapping; | |
651 | ||
9326c9b2 | 652 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
653 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
654 | WB_SYNC_ALL); | |
655 | /* See comment of filemap_write_and_wait() */ | |
656 | if (err != -EIO) | |
657 | __filemap_fdatawait_range(mapping, lstart, lend); | |
658 | } | |
659 | err2 = file_check_and_advance_wb_err(file); | |
660 | if (!err) | |
661 | err = err2; | |
662 | return err; | |
663 | } | |
664 | EXPORT_SYMBOL(file_write_and_wait_range); | |
665 | ||
ef6a3c63 MS |
666 | /** |
667 | * replace_page_cache_page - replace a pagecache page with a new one | |
668 | * @old: page to be replaced | |
669 | * @new: page to replace with | |
670 | * @gfp_mask: allocation mode | |
671 | * | |
672 | * This function replaces a page in the pagecache with a new one. On | |
673 | * success it acquires the pagecache reference for the new page and | |
674 | * drops it for the old page. Both the old and new pages must be | |
675 | * locked. This function does not add the new page to the LRU, the | |
676 | * caller must do that. | |
677 | * | |
678 | * The remove + add is atomic. The only way this function can fail is | |
679 | * memory allocation failure. | |
680 | */ | |
681 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
682 | { | |
683 | int error; | |
ef6a3c63 | 684 | |
309381fe SL |
685 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
686 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
687 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 688 | |
ef6a3c63 MS |
689 | error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); |
690 | if (!error) { | |
691 | struct address_space *mapping = old->mapping; | |
692 | void (*freepage)(struct page *); | |
c4843a75 | 693 | unsigned long flags; |
ef6a3c63 MS |
694 | |
695 | pgoff_t offset = old->index; | |
696 | freepage = mapping->a_ops->freepage; | |
697 | ||
09cbfeaf | 698 | get_page(new); |
ef6a3c63 MS |
699 | new->mapping = mapping; |
700 | new->index = offset; | |
701 | ||
c4843a75 | 702 | spin_lock_irqsave(&mapping->tree_lock, flags); |
62cccb8c | 703 | __delete_from_page_cache(old, NULL); |
22f2ac51 | 704 | error = page_cache_tree_insert(mapping, new, NULL); |
ef6a3c63 | 705 | BUG_ON(error); |
4165b9b4 MH |
706 | |
707 | /* | |
708 | * hugetlb pages do not participate in page cache accounting. | |
709 | */ | |
710 | if (!PageHuge(new)) | |
11fb9989 | 711 | __inc_node_page_state(new, NR_FILE_PAGES); |
ef6a3c63 | 712 | if (PageSwapBacked(new)) |
11fb9989 | 713 | __inc_node_page_state(new, NR_SHMEM); |
c4843a75 | 714 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6a93ca8f | 715 | mem_cgroup_migrate(old, new); |
ef6a3c63 MS |
716 | radix_tree_preload_end(); |
717 | if (freepage) | |
718 | freepage(old); | |
09cbfeaf | 719 | put_page(old); |
ef6a3c63 MS |
720 | } |
721 | ||
722 | return error; | |
723 | } | |
724 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
725 | ||
a528910e JW |
726 | static int __add_to_page_cache_locked(struct page *page, |
727 | struct address_space *mapping, | |
728 | pgoff_t offset, gfp_t gfp_mask, | |
729 | void **shadowp) | |
1da177e4 | 730 | { |
00501b53 JW |
731 | int huge = PageHuge(page); |
732 | struct mem_cgroup *memcg; | |
e286781d NP |
733 | int error; |
734 | ||
309381fe SL |
735 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
736 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
e286781d | 737 | |
00501b53 JW |
738 | if (!huge) { |
739 | error = mem_cgroup_try_charge(page, current->mm, | |
f627c2f5 | 740 | gfp_mask, &memcg, false); |
00501b53 JW |
741 | if (error) |
742 | return error; | |
743 | } | |
1da177e4 | 744 | |
5e4c0d97 | 745 | error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM); |
66a0c8ee | 746 | if (error) { |
00501b53 | 747 | if (!huge) |
f627c2f5 | 748 | mem_cgroup_cancel_charge(page, memcg, false); |
66a0c8ee KS |
749 | return error; |
750 | } | |
751 | ||
09cbfeaf | 752 | get_page(page); |
66a0c8ee KS |
753 | page->mapping = mapping; |
754 | page->index = offset; | |
755 | ||
756 | spin_lock_irq(&mapping->tree_lock); | |
a528910e | 757 | error = page_cache_tree_insert(mapping, page, shadowp); |
66a0c8ee KS |
758 | radix_tree_preload_end(); |
759 | if (unlikely(error)) | |
760 | goto err_insert; | |
4165b9b4 MH |
761 | |
762 | /* hugetlb pages do not participate in page cache accounting. */ | |
763 | if (!huge) | |
11fb9989 | 764 | __inc_node_page_state(page, NR_FILE_PAGES); |
66a0c8ee | 765 | spin_unlock_irq(&mapping->tree_lock); |
00501b53 | 766 | if (!huge) |
f627c2f5 | 767 | mem_cgroup_commit_charge(page, memcg, false, false); |
66a0c8ee KS |
768 | trace_mm_filemap_add_to_page_cache(page); |
769 | return 0; | |
770 | err_insert: | |
771 | page->mapping = NULL; | |
772 | /* Leave page->index set: truncation relies upon it */ | |
773 | spin_unlock_irq(&mapping->tree_lock); | |
00501b53 | 774 | if (!huge) |
f627c2f5 | 775 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 776 | put_page(page); |
1da177e4 LT |
777 | return error; |
778 | } | |
a528910e JW |
779 | |
780 | /** | |
781 | * add_to_page_cache_locked - add a locked page to the pagecache | |
782 | * @page: page to add | |
783 | * @mapping: the page's address_space | |
784 | * @offset: page index | |
785 | * @gfp_mask: page allocation mode | |
786 | * | |
787 | * This function is used to add a page to the pagecache. It must be locked. | |
788 | * This function does not add the page to the LRU. The caller must do that. | |
789 | */ | |
790 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
791 | pgoff_t offset, gfp_t gfp_mask) | |
792 | { | |
793 | return __add_to_page_cache_locked(page, mapping, offset, | |
794 | gfp_mask, NULL); | |
795 | } | |
e286781d | 796 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
797 | |
798 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 799 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 800 | { |
a528910e | 801 | void *shadow = NULL; |
4f98a2fe RR |
802 | int ret; |
803 | ||
48c935ad | 804 | __SetPageLocked(page); |
a528910e JW |
805 | ret = __add_to_page_cache_locked(page, mapping, offset, |
806 | gfp_mask, &shadow); | |
807 | if (unlikely(ret)) | |
48c935ad | 808 | __ClearPageLocked(page); |
a528910e JW |
809 | else { |
810 | /* | |
811 | * The page might have been evicted from cache only | |
812 | * recently, in which case it should be activated like | |
813 | * any other repeatedly accessed page. | |
f0281a00 RR |
814 | * The exception is pages getting rewritten; evicting other |
815 | * data from the working set, only to cache data that will | |
816 | * get overwritten with something else, is a waste of memory. | |
a528910e | 817 | */ |
f0281a00 RR |
818 | if (!(gfp_mask & __GFP_WRITE) && |
819 | shadow && workingset_refault(shadow)) { | |
a528910e JW |
820 | SetPageActive(page); |
821 | workingset_activation(page); | |
822 | } else | |
823 | ClearPageActive(page); | |
824 | lru_cache_add(page); | |
825 | } | |
1da177e4 LT |
826 | return ret; |
827 | } | |
18bc0bbd | 828 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 829 | |
44110fe3 | 830 | #ifdef CONFIG_NUMA |
2ae88149 | 831 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 832 | { |
c0ff7453 MX |
833 | int n; |
834 | struct page *page; | |
835 | ||
44110fe3 | 836 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
837 | unsigned int cpuset_mems_cookie; |
838 | do { | |
d26914d1 | 839 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 840 | n = cpuset_mem_spread_node(); |
96db800f | 841 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 842 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 843 | |
c0ff7453 | 844 | return page; |
44110fe3 | 845 | } |
2ae88149 | 846 | return alloc_pages(gfp, 0); |
44110fe3 | 847 | } |
2ae88149 | 848 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
849 | #endif |
850 | ||
1da177e4 LT |
851 | /* |
852 | * In order to wait for pages to become available there must be | |
853 | * waitqueues associated with pages. By using a hash table of | |
854 | * waitqueues where the bucket discipline is to maintain all | |
855 | * waiters on the same queue and wake all when any of the pages | |
856 | * become available, and for the woken contexts to check to be | |
857 | * sure the appropriate page became available, this saves space | |
858 | * at a cost of "thundering herd" phenomena during rare hash | |
859 | * collisions. | |
860 | */ | |
62906027 NP |
861 | #define PAGE_WAIT_TABLE_BITS 8 |
862 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
863 | static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; | |
864 | ||
865 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
1da177e4 | 866 | { |
62906027 | 867 | return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 868 | } |
1da177e4 | 869 | |
62906027 | 870 | void __init pagecache_init(void) |
1da177e4 | 871 | { |
62906027 | 872 | int i; |
1da177e4 | 873 | |
62906027 NP |
874 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
875 | init_waitqueue_head(&page_wait_table[i]); | |
876 | ||
877 | page_writeback_init(); | |
1da177e4 | 878 | } |
1da177e4 | 879 | |
3510ca20 | 880 | /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */ |
62906027 NP |
881 | struct wait_page_key { |
882 | struct page *page; | |
883 | int bit_nr; | |
884 | int page_match; | |
885 | }; | |
886 | ||
887 | struct wait_page_queue { | |
888 | struct page *page; | |
889 | int bit_nr; | |
ac6424b9 | 890 | wait_queue_entry_t wait; |
62906027 NP |
891 | }; |
892 | ||
ac6424b9 | 893 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 894 | { |
62906027 NP |
895 | struct wait_page_key *key = arg; |
896 | struct wait_page_queue *wait_page | |
897 | = container_of(wait, struct wait_page_queue, wait); | |
898 | ||
899 | if (wait_page->page != key->page) | |
900 | return 0; | |
901 | key->page_match = 1; | |
f62e00cc | 902 | |
62906027 NP |
903 | if (wait_page->bit_nr != key->bit_nr) |
904 | return 0; | |
3510ca20 LT |
905 | |
906 | /* Stop walking if it's locked */ | |
62906027 | 907 | if (test_bit(key->bit_nr, &key->page->flags)) |
3510ca20 | 908 | return -1; |
f62e00cc | 909 | |
62906027 | 910 | return autoremove_wake_function(wait, mode, sync, key); |
f62e00cc KM |
911 | } |
912 | ||
74d81bfa | 913 | static void wake_up_page_bit(struct page *page, int bit_nr) |
cbbce822 | 914 | { |
62906027 NP |
915 | wait_queue_head_t *q = page_waitqueue(page); |
916 | struct wait_page_key key; | |
917 | unsigned long flags; | |
11a19c7b | 918 | wait_queue_entry_t bookmark; |
cbbce822 | 919 | |
62906027 NP |
920 | key.page = page; |
921 | key.bit_nr = bit_nr; | |
922 | key.page_match = 0; | |
923 | ||
11a19c7b TC |
924 | bookmark.flags = 0; |
925 | bookmark.private = NULL; | |
926 | bookmark.func = NULL; | |
927 | INIT_LIST_HEAD(&bookmark.entry); | |
928 | ||
62906027 | 929 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
930 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
931 | ||
932 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
933 | /* | |
934 | * Take a breather from holding the lock, | |
935 | * allow pages that finish wake up asynchronously | |
936 | * to acquire the lock and remove themselves | |
937 | * from wait queue | |
938 | */ | |
939 | spin_unlock_irqrestore(&q->lock, flags); | |
940 | cpu_relax(); | |
941 | spin_lock_irqsave(&q->lock, flags); | |
942 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
943 | } | |
944 | ||
62906027 NP |
945 | /* |
946 | * It is possible for other pages to have collided on the waitqueue | |
947 | * hash, so in that case check for a page match. That prevents a long- | |
948 | * term waiter | |
949 | * | |
950 | * It is still possible to miss a case here, when we woke page waiters | |
951 | * and removed them from the waitqueue, but there are still other | |
952 | * page waiters. | |
953 | */ | |
954 | if (!waitqueue_active(q) || !key.page_match) { | |
955 | ClearPageWaiters(page); | |
956 | /* | |
957 | * It's possible to miss clearing Waiters here, when we woke | |
958 | * our page waiters, but the hashed waitqueue has waiters for | |
959 | * other pages on it. | |
960 | * | |
961 | * That's okay, it's a rare case. The next waker will clear it. | |
962 | */ | |
963 | } | |
964 | spin_unlock_irqrestore(&q->lock, flags); | |
965 | } | |
74d81bfa NP |
966 | |
967 | static void wake_up_page(struct page *page, int bit) | |
968 | { | |
969 | if (!PageWaiters(page)) | |
970 | return; | |
971 | wake_up_page_bit(page, bit); | |
972 | } | |
62906027 NP |
973 | |
974 | static inline int wait_on_page_bit_common(wait_queue_head_t *q, | |
975 | struct page *page, int bit_nr, int state, bool lock) | |
976 | { | |
977 | struct wait_page_queue wait_page; | |
ac6424b9 | 978 | wait_queue_entry_t *wait = &wait_page.wait; |
62906027 NP |
979 | int ret = 0; |
980 | ||
981 | init_wait(wait); | |
3510ca20 | 982 | wait->flags = lock ? WQ_FLAG_EXCLUSIVE : 0; |
62906027 NP |
983 | wait->func = wake_page_function; |
984 | wait_page.page = page; | |
985 | wait_page.bit_nr = bit_nr; | |
986 | ||
987 | for (;;) { | |
988 | spin_lock_irq(&q->lock); | |
989 | ||
2055da97 | 990 | if (likely(list_empty(&wait->entry))) { |
3510ca20 | 991 | __add_wait_queue_entry_tail(q, wait); |
62906027 NP |
992 | SetPageWaiters(page); |
993 | } | |
994 | ||
995 | set_current_state(state); | |
996 | ||
997 | spin_unlock_irq(&q->lock); | |
998 | ||
999 | if (likely(test_bit(bit_nr, &page->flags))) { | |
1000 | io_schedule(); | |
62906027 NP |
1001 | } |
1002 | ||
1003 | if (lock) { | |
1004 | if (!test_and_set_bit_lock(bit_nr, &page->flags)) | |
1005 | break; | |
1006 | } else { | |
1007 | if (!test_bit(bit_nr, &page->flags)) | |
1008 | break; | |
1009 | } | |
a8b169af LT |
1010 | |
1011 | if (unlikely(signal_pending_state(state, current))) { | |
1012 | ret = -EINTR; | |
1013 | break; | |
1014 | } | |
62906027 NP |
1015 | } |
1016 | ||
1017 | finish_wait(q, wait); | |
1018 | ||
1019 | /* | |
1020 | * A signal could leave PageWaiters set. Clearing it here if | |
1021 | * !waitqueue_active would be possible (by open-coding finish_wait), | |
1022 | * but still fail to catch it in the case of wait hash collision. We | |
1023 | * already can fail to clear wait hash collision cases, so don't | |
1024 | * bother with signals either. | |
1025 | */ | |
1026 | ||
1027 | return ret; | |
1028 | } | |
1029 | ||
1030 | void wait_on_page_bit(struct page *page, int bit_nr) | |
1031 | { | |
1032 | wait_queue_head_t *q = page_waitqueue(page); | |
1033 | wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false); | |
1034 | } | |
1035 | EXPORT_SYMBOL(wait_on_page_bit); | |
1036 | ||
1037 | int wait_on_page_bit_killable(struct page *page, int bit_nr) | |
1038 | { | |
1039 | wait_queue_head_t *q = page_waitqueue(page); | |
1040 | return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false); | |
cbbce822 | 1041 | } |
cbbce822 | 1042 | |
385e1ca5 DH |
1043 | /** |
1044 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
1045 | * @page: Page defining the wait queue of interest |
1046 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
1047 | * |
1048 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
1049 | */ | |
ac6424b9 | 1050 | void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter) |
385e1ca5 DH |
1051 | { |
1052 | wait_queue_head_t *q = page_waitqueue(page); | |
1053 | unsigned long flags; | |
1054 | ||
1055 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1056 | __add_wait_queue_entry_tail(q, waiter); |
62906027 | 1057 | SetPageWaiters(page); |
385e1ca5 DH |
1058 | spin_unlock_irqrestore(&q->lock, flags); |
1059 | } | |
1060 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
1061 | ||
b91e1302 LT |
1062 | #ifndef clear_bit_unlock_is_negative_byte |
1063 | ||
1064 | /* | |
1065 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1066 | * | |
1067 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1068 | * test the sign bit at the same time. But if the architecture does | |
1069 | * not support that special operation, we just do this all by hand | |
1070 | * instead. | |
1071 | * | |
1072 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
1073 | * being cleared, but a memory barrier should be unneccssary since it is | |
1074 | * in the same byte as PG_locked. | |
1075 | */ | |
1076 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1077 | { | |
1078 | clear_bit_unlock(nr, mem); | |
1079 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1080 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1081 | } |
1082 | ||
1083 | #endif | |
1084 | ||
1da177e4 | 1085 | /** |
485bb99b | 1086 | * unlock_page - unlock a locked page |
1da177e4 LT |
1087 | * @page: the page |
1088 | * | |
1089 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
1090 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 1091 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
1092 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
1093 | * | |
b91e1302 LT |
1094 | * Note that this depends on PG_waiters being the sign bit in the byte |
1095 | * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to | |
1096 | * clear the PG_locked bit and test PG_waiters at the same time fairly | |
1097 | * portably (architectures that do LL/SC can test any bit, while x86 can | |
1098 | * test the sign bit). | |
1da177e4 | 1099 | */ |
920c7a5d | 1100 | void unlock_page(struct page *page) |
1da177e4 | 1101 | { |
b91e1302 | 1102 | BUILD_BUG_ON(PG_waiters != 7); |
48c935ad | 1103 | page = compound_head(page); |
309381fe | 1104 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
b91e1302 LT |
1105 | if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags)) |
1106 | wake_up_page_bit(page, PG_locked); | |
1da177e4 LT |
1107 | } |
1108 | EXPORT_SYMBOL(unlock_page); | |
1109 | ||
485bb99b RD |
1110 | /** |
1111 | * end_page_writeback - end writeback against a page | |
1112 | * @page: the page | |
1da177e4 LT |
1113 | */ |
1114 | void end_page_writeback(struct page *page) | |
1115 | { | |
888cf2db MG |
1116 | /* |
1117 | * TestClearPageReclaim could be used here but it is an atomic | |
1118 | * operation and overkill in this particular case. Failing to | |
1119 | * shuffle a page marked for immediate reclaim is too mild to | |
1120 | * justify taking an atomic operation penalty at the end of | |
1121 | * ever page writeback. | |
1122 | */ | |
1123 | if (PageReclaim(page)) { | |
1124 | ClearPageReclaim(page); | |
ac6aadb2 | 1125 | rotate_reclaimable_page(page); |
888cf2db | 1126 | } |
ac6aadb2 MS |
1127 | |
1128 | if (!test_clear_page_writeback(page)) | |
1129 | BUG(); | |
1130 | ||
4e857c58 | 1131 | smp_mb__after_atomic(); |
1da177e4 LT |
1132 | wake_up_page(page, PG_writeback); |
1133 | } | |
1134 | EXPORT_SYMBOL(end_page_writeback); | |
1135 | ||
57d99845 MW |
1136 | /* |
1137 | * After completing I/O on a page, call this routine to update the page | |
1138 | * flags appropriately | |
1139 | */ | |
c11f0c0b | 1140 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1141 | { |
c11f0c0b | 1142 | if (!is_write) { |
57d99845 MW |
1143 | if (!err) { |
1144 | SetPageUptodate(page); | |
1145 | } else { | |
1146 | ClearPageUptodate(page); | |
1147 | SetPageError(page); | |
1148 | } | |
1149 | unlock_page(page); | |
abf54548 | 1150 | } else { |
57d99845 | 1151 | if (err) { |
dd8416c4 MK |
1152 | struct address_space *mapping; |
1153 | ||
57d99845 | 1154 | SetPageError(page); |
dd8416c4 MK |
1155 | mapping = page_mapping(page); |
1156 | if (mapping) | |
1157 | mapping_set_error(mapping, err); | |
57d99845 MW |
1158 | } |
1159 | end_page_writeback(page); | |
1160 | } | |
1161 | } | |
1162 | EXPORT_SYMBOL_GPL(page_endio); | |
1163 | ||
485bb99b RD |
1164 | /** |
1165 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
87066755 | 1166 | * @__page: the page to lock |
1da177e4 | 1167 | */ |
62906027 | 1168 | void __lock_page(struct page *__page) |
1da177e4 | 1169 | { |
62906027 NP |
1170 | struct page *page = compound_head(__page); |
1171 | wait_queue_head_t *q = page_waitqueue(page); | |
1172 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true); | |
1da177e4 LT |
1173 | } |
1174 | EXPORT_SYMBOL(__lock_page); | |
1175 | ||
62906027 | 1176 | int __lock_page_killable(struct page *__page) |
2687a356 | 1177 | { |
62906027 NP |
1178 | struct page *page = compound_head(__page); |
1179 | wait_queue_head_t *q = page_waitqueue(page); | |
1180 | return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true); | |
2687a356 | 1181 | } |
18bc0bbd | 1182 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 1183 | |
9a95f3cf PC |
1184 | /* |
1185 | * Return values: | |
1186 | * 1 - page is locked; mmap_sem is still held. | |
1187 | * 0 - page is not locked. | |
1188 | * mmap_sem has been released (up_read()), unless flags had both | |
1189 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
1190 | * which case mmap_sem is still held. | |
1191 | * | |
1192 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
1193 | * with the page locked and the mmap_sem unperturbed. | |
1194 | */ | |
d065bd81 ML |
1195 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
1196 | unsigned int flags) | |
1197 | { | |
37b23e05 KM |
1198 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
1199 | /* | |
1200 | * CAUTION! In this case, mmap_sem is not released | |
1201 | * even though return 0. | |
1202 | */ | |
1203 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
1204 | return 0; | |
1205 | ||
1206 | up_read(&mm->mmap_sem); | |
1207 | if (flags & FAULT_FLAG_KILLABLE) | |
1208 | wait_on_page_locked_killable(page); | |
1209 | else | |
318b275f | 1210 | wait_on_page_locked(page); |
d065bd81 | 1211 | return 0; |
37b23e05 KM |
1212 | } else { |
1213 | if (flags & FAULT_FLAG_KILLABLE) { | |
1214 | int ret; | |
1215 | ||
1216 | ret = __lock_page_killable(page); | |
1217 | if (ret) { | |
1218 | up_read(&mm->mmap_sem); | |
1219 | return 0; | |
1220 | } | |
1221 | } else | |
1222 | __lock_page(page); | |
1223 | return 1; | |
d065bd81 ML |
1224 | } |
1225 | } | |
1226 | ||
e7b563bb JW |
1227 | /** |
1228 | * page_cache_next_hole - find the next hole (not-present entry) | |
1229 | * @mapping: mapping | |
1230 | * @index: index | |
1231 | * @max_scan: maximum range to search | |
1232 | * | |
1233 | * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the | |
1234 | * lowest indexed hole. | |
1235 | * | |
1236 | * Returns: the index of the hole if found, otherwise returns an index | |
1237 | * outside of the set specified (in which case 'return - index >= | |
1238 | * max_scan' will be true). In rare cases of index wrap-around, 0 will | |
1239 | * be returned. | |
1240 | * | |
1241 | * page_cache_next_hole may be called under rcu_read_lock. However, | |
1242 | * like radix_tree_gang_lookup, this will not atomically search a | |
1243 | * snapshot of the tree at a single point in time. For example, if a | |
1244 | * hole is created at index 5, then subsequently a hole is created at | |
1245 | * index 10, page_cache_next_hole covering both indexes may return 10 | |
1246 | * if called under rcu_read_lock. | |
1247 | */ | |
1248 | pgoff_t page_cache_next_hole(struct address_space *mapping, | |
1249 | pgoff_t index, unsigned long max_scan) | |
1250 | { | |
1251 | unsigned long i; | |
1252 | ||
1253 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
1254 | struct page *page; |
1255 | ||
1256 | page = radix_tree_lookup(&mapping->page_tree, index); | |
1257 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
1258 | break; |
1259 | index++; | |
1260 | if (index == 0) | |
1261 | break; | |
1262 | } | |
1263 | ||
1264 | return index; | |
1265 | } | |
1266 | EXPORT_SYMBOL(page_cache_next_hole); | |
1267 | ||
1268 | /** | |
1269 | * page_cache_prev_hole - find the prev hole (not-present entry) | |
1270 | * @mapping: mapping | |
1271 | * @index: index | |
1272 | * @max_scan: maximum range to search | |
1273 | * | |
1274 | * Search backwards in the range [max(index-max_scan+1, 0), index] for | |
1275 | * the first hole. | |
1276 | * | |
1277 | * Returns: the index of the hole if found, otherwise returns an index | |
1278 | * outside of the set specified (in which case 'index - return >= | |
1279 | * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX | |
1280 | * will be returned. | |
1281 | * | |
1282 | * page_cache_prev_hole may be called under rcu_read_lock. However, | |
1283 | * like radix_tree_gang_lookup, this will not atomically search a | |
1284 | * snapshot of the tree at a single point in time. For example, if a | |
1285 | * hole is created at index 10, then subsequently a hole is created at | |
1286 | * index 5, page_cache_prev_hole covering both indexes may return 5 if | |
1287 | * called under rcu_read_lock. | |
1288 | */ | |
1289 | pgoff_t page_cache_prev_hole(struct address_space *mapping, | |
1290 | pgoff_t index, unsigned long max_scan) | |
1291 | { | |
1292 | unsigned long i; | |
1293 | ||
1294 | for (i = 0; i < max_scan; i++) { | |
0cd6144a JW |
1295 | struct page *page; |
1296 | ||
1297 | page = radix_tree_lookup(&mapping->page_tree, index); | |
1298 | if (!page || radix_tree_exceptional_entry(page)) | |
e7b563bb JW |
1299 | break; |
1300 | index--; | |
1301 | if (index == ULONG_MAX) | |
1302 | break; | |
1303 | } | |
1304 | ||
1305 | return index; | |
1306 | } | |
1307 | EXPORT_SYMBOL(page_cache_prev_hole); | |
1308 | ||
485bb99b | 1309 | /** |
0cd6144a | 1310 | * find_get_entry - find and get a page cache entry |
485bb99b | 1311 | * @mapping: the address_space to search |
0cd6144a JW |
1312 | * @offset: the page cache index |
1313 | * | |
1314 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1315 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1316 | * |
139b6a6f JW |
1317 | * If the slot holds a shadow entry of a previously evicted page, or a |
1318 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1319 | * |
1320 | * Otherwise, %NULL is returned. | |
1da177e4 | 1321 | */ |
0cd6144a | 1322 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1323 | { |
a60637c8 | 1324 | void **pagep; |
83929372 | 1325 | struct page *head, *page; |
1da177e4 | 1326 | |
a60637c8 NP |
1327 | rcu_read_lock(); |
1328 | repeat: | |
1329 | page = NULL; | |
1330 | pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); | |
1331 | if (pagep) { | |
1332 | page = radix_tree_deref_slot(pagep); | |
27d20fdd NP |
1333 | if (unlikely(!page)) |
1334 | goto out; | |
a2c16d6c | 1335 | if (radix_tree_exception(page)) { |
8079b1c8 HD |
1336 | if (radix_tree_deref_retry(page)) |
1337 | goto repeat; | |
1338 | /* | |
139b6a6f JW |
1339 | * A shadow entry of a recently evicted page, |
1340 | * or a swap entry from shmem/tmpfs. Return | |
1341 | * it without attempting to raise page count. | |
8079b1c8 HD |
1342 | */ |
1343 | goto out; | |
a2c16d6c | 1344 | } |
83929372 KS |
1345 | |
1346 | head = compound_head(page); | |
1347 | if (!page_cache_get_speculative(head)) | |
1348 | goto repeat; | |
1349 | ||
1350 | /* The page was split under us? */ | |
1351 | if (compound_head(page) != head) { | |
1352 | put_page(head); | |
a60637c8 | 1353 | goto repeat; |
83929372 | 1354 | } |
a60637c8 NP |
1355 | |
1356 | /* | |
1357 | * Has the page moved? | |
1358 | * This is part of the lockless pagecache protocol. See | |
1359 | * include/linux/pagemap.h for details. | |
1360 | */ | |
1361 | if (unlikely(page != *pagep)) { | |
83929372 | 1362 | put_page(head); |
a60637c8 NP |
1363 | goto repeat; |
1364 | } | |
1365 | } | |
27d20fdd | 1366 | out: |
a60637c8 NP |
1367 | rcu_read_unlock(); |
1368 | ||
1da177e4 LT |
1369 | return page; |
1370 | } | |
0cd6144a | 1371 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1372 | |
0cd6144a JW |
1373 | /** |
1374 | * find_lock_entry - locate, pin and lock a page cache entry | |
1375 | * @mapping: the address_space to search | |
1376 | * @offset: the page cache index | |
1377 | * | |
1378 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1379 | * page cache page, it is returned locked and with an increased | |
1380 | * refcount. | |
1381 | * | |
139b6a6f JW |
1382 | * If the slot holds a shadow entry of a previously evicted page, or a |
1383 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a JW |
1384 | * |
1385 | * Otherwise, %NULL is returned. | |
1386 | * | |
1387 | * find_lock_entry() may sleep. | |
1388 | */ | |
1389 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1390 | { |
1391 | struct page *page; | |
1392 | ||
1da177e4 | 1393 | repeat: |
0cd6144a | 1394 | page = find_get_entry(mapping, offset); |
a2c16d6c | 1395 | if (page && !radix_tree_exception(page)) { |
a60637c8 NP |
1396 | lock_page(page); |
1397 | /* Has the page been truncated? */ | |
83929372 | 1398 | if (unlikely(page_mapping(page) != mapping)) { |
a60637c8 | 1399 | unlock_page(page); |
09cbfeaf | 1400 | put_page(page); |
a60637c8 | 1401 | goto repeat; |
1da177e4 | 1402 | } |
83929372 | 1403 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
1da177e4 | 1404 | } |
1da177e4 LT |
1405 | return page; |
1406 | } | |
0cd6144a JW |
1407 | EXPORT_SYMBOL(find_lock_entry); |
1408 | ||
1409 | /** | |
2457aec6 | 1410 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1411 | * @mapping: the address_space to search |
1412 | * @offset: the page index | |
2457aec6 | 1413 | * @fgp_flags: PCG flags |
45f87de5 | 1414 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
0cd6144a | 1415 | * |
2457aec6 | 1416 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1417 | * |
75325189 | 1418 | * PCG flags modify how the page is returned. |
0cd6144a | 1419 | * |
0e056eb5 MCC |
1420 | * @fgp_flags can be: |
1421 | * | |
1422 | * - FGP_ACCESSED: the page will be marked accessed | |
1423 | * - FGP_LOCK: Page is return locked | |
1424 | * - FGP_CREAT: If page is not present then a new page is allocated using | |
1425 | * @gfp_mask and added to the page cache and the VM's LRU | |
1426 | * list. The page is returned locked and with an increased | |
1427 | * refcount. Otherwise, NULL is returned. | |
1da177e4 | 1428 | * |
2457aec6 MG |
1429 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1430 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1431 | * |
2457aec6 | 1432 | * If there is a page cache page, it is returned with an increased refcount. |
1da177e4 | 1433 | */ |
2457aec6 | 1434 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
45f87de5 | 1435 | int fgp_flags, gfp_t gfp_mask) |
1da177e4 | 1436 | { |
eb2be189 | 1437 | struct page *page; |
2457aec6 | 1438 | |
1da177e4 | 1439 | repeat: |
2457aec6 MG |
1440 | page = find_get_entry(mapping, offset); |
1441 | if (radix_tree_exceptional_entry(page)) | |
1442 | page = NULL; | |
1443 | if (!page) | |
1444 | goto no_page; | |
1445 | ||
1446 | if (fgp_flags & FGP_LOCK) { | |
1447 | if (fgp_flags & FGP_NOWAIT) { | |
1448 | if (!trylock_page(page)) { | |
09cbfeaf | 1449 | put_page(page); |
2457aec6 MG |
1450 | return NULL; |
1451 | } | |
1452 | } else { | |
1453 | lock_page(page); | |
1454 | } | |
1455 | ||
1456 | /* Has the page been truncated? */ | |
1457 | if (unlikely(page->mapping != mapping)) { | |
1458 | unlock_page(page); | |
09cbfeaf | 1459 | put_page(page); |
2457aec6 MG |
1460 | goto repeat; |
1461 | } | |
1462 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1463 | } | |
1464 | ||
1465 | if (page && (fgp_flags & FGP_ACCESSED)) | |
1466 | mark_page_accessed(page); | |
1467 | ||
1468 | no_page: | |
1469 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1470 | int err; | |
1471 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1472 | gfp_mask |= __GFP_WRITE; |
1473 | if (fgp_flags & FGP_NOFS) | |
1474 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1475 | |
45f87de5 | 1476 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1477 | if (!page) |
1478 | return NULL; | |
2457aec6 MG |
1479 | |
1480 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1481 | fgp_flags |= FGP_LOCK; | |
1482 | ||
eb39d618 | 1483 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1484 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1485 | __SetPageReferenced(page); |
2457aec6 | 1486 | |
45f87de5 MH |
1487 | err = add_to_page_cache_lru(page, mapping, offset, |
1488 | gfp_mask & GFP_RECLAIM_MASK); | |
eb2be189 | 1489 | if (unlikely(err)) { |
09cbfeaf | 1490 | put_page(page); |
eb2be189 NP |
1491 | page = NULL; |
1492 | if (err == -EEXIST) | |
1493 | goto repeat; | |
1da177e4 | 1494 | } |
1da177e4 | 1495 | } |
2457aec6 | 1496 | |
1da177e4 LT |
1497 | return page; |
1498 | } | |
2457aec6 | 1499 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1500 | |
0cd6144a JW |
1501 | /** |
1502 | * find_get_entries - gang pagecache lookup | |
1503 | * @mapping: The address_space to search | |
1504 | * @start: The starting page cache index | |
1505 | * @nr_entries: The maximum number of entries | |
1506 | * @entries: Where the resulting entries are placed | |
1507 | * @indices: The cache indices corresponding to the entries in @entries | |
1508 | * | |
1509 | * find_get_entries() will search for and return a group of up to | |
1510 | * @nr_entries entries in the mapping. The entries are placed at | |
1511 | * @entries. find_get_entries() takes a reference against any actual | |
1512 | * pages it returns. | |
1513 | * | |
1514 | * The search returns a group of mapping-contiguous page cache entries | |
1515 | * with ascending indexes. There may be holes in the indices due to | |
1516 | * not-present pages. | |
1517 | * | |
139b6a6f JW |
1518 | * Any shadow entries of evicted pages, or swap entries from |
1519 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a JW |
1520 | * |
1521 | * find_get_entries() returns the number of pages and shadow entries | |
1522 | * which were found. | |
1523 | */ | |
1524 | unsigned find_get_entries(struct address_space *mapping, | |
1525 | pgoff_t start, unsigned int nr_entries, | |
1526 | struct page **entries, pgoff_t *indices) | |
1527 | { | |
1528 | void **slot; | |
1529 | unsigned int ret = 0; | |
1530 | struct radix_tree_iter iter; | |
1531 | ||
1532 | if (!nr_entries) | |
1533 | return 0; | |
1534 | ||
1535 | rcu_read_lock(); | |
0cd6144a | 1536 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) { |
83929372 | 1537 | struct page *head, *page; |
0cd6144a JW |
1538 | repeat: |
1539 | page = radix_tree_deref_slot(slot); | |
1540 | if (unlikely(!page)) | |
1541 | continue; | |
1542 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
1543 | if (radix_tree_deref_retry(page)) { |
1544 | slot = radix_tree_iter_retry(&iter); | |
1545 | continue; | |
1546 | } | |
0cd6144a | 1547 | /* |
f9fe48be RZ |
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. | |
0cd6144a JW |
1551 | */ |
1552 | goto export; | |
1553 | } | |
83929372 KS |
1554 | |
1555 | head = compound_head(page); | |
1556 | if (!page_cache_get_speculative(head)) | |
1557 | goto repeat; | |
1558 | ||
1559 | /* The page was split under us? */ | |
1560 | if (compound_head(page) != head) { | |
1561 | put_page(head); | |
0cd6144a | 1562 | goto repeat; |
83929372 | 1563 | } |
0cd6144a JW |
1564 | |
1565 | /* Has the page moved? */ | |
1566 | if (unlikely(page != *slot)) { | |
83929372 | 1567 | put_page(head); |
0cd6144a JW |
1568 | goto repeat; |
1569 | } | |
1570 | export: | |
1571 | indices[ret] = iter.index; | |
1572 | entries[ret] = page; | |
1573 | if (++ret == nr_entries) | |
1574 | break; | |
1575 | } | |
1576 | rcu_read_unlock(); | |
1577 | return ret; | |
1578 | } | |
1579 | ||
1da177e4 | 1580 | /** |
b947cee4 | 1581 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
1582 | * @mapping: The address_space to search |
1583 | * @start: The starting page index | |
b947cee4 | 1584 | * @end: The final page index (inclusive) |
1da177e4 LT |
1585 | * @nr_pages: The maximum number of pages |
1586 | * @pages: Where the resulting pages are placed | |
1587 | * | |
b947cee4 JK |
1588 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
1589 | * pages in the mapping starting at index @start and up to index @end | |
1590 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
1591 | * a reference against the returned pages. | |
1da177e4 LT |
1592 | * |
1593 | * The search returns a group of mapping-contiguous pages with ascending | |
1594 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 1595 | * We also update @start to index the next page for the traversal. |
1da177e4 | 1596 | * |
b947cee4 JK |
1597 | * find_get_pages_range() returns the number of pages which were found. If this |
1598 | * number is smaller than @nr_pages, the end of specified range has been | |
1599 | * reached. | |
1da177e4 | 1600 | */ |
b947cee4 JK |
1601 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
1602 | pgoff_t end, unsigned int nr_pages, | |
1603 | struct page **pages) | |
1da177e4 | 1604 | { |
0fc9d104 KK |
1605 | struct radix_tree_iter iter; |
1606 | void **slot; | |
1607 | unsigned ret = 0; | |
1608 | ||
1609 | if (unlikely(!nr_pages)) | |
1610 | return 0; | |
a60637c8 NP |
1611 | |
1612 | rcu_read_lock(); | |
d72dc8a2 | 1613 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, *start) { |
83929372 | 1614 | struct page *head, *page; |
b947cee4 JK |
1615 | |
1616 | if (iter.index > end) | |
1617 | break; | |
a60637c8 | 1618 | repeat: |
0fc9d104 | 1619 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1620 | if (unlikely(!page)) |
1621 | continue; | |
9d8aa4ea | 1622 | |
a2c16d6c | 1623 | if (radix_tree_exception(page)) { |
8079b1c8 | 1624 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1625 | slot = radix_tree_iter_retry(&iter); |
1626 | continue; | |
8079b1c8 | 1627 | } |
a2c16d6c | 1628 | /* |
139b6a6f JW |
1629 | * A shadow entry of a recently evicted page, |
1630 | * or a swap entry from shmem/tmpfs. Skip | |
1631 | * over it. | |
a2c16d6c | 1632 | */ |
8079b1c8 | 1633 | continue; |
27d20fdd | 1634 | } |
a60637c8 | 1635 | |
83929372 KS |
1636 | head = compound_head(page); |
1637 | if (!page_cache_get_speculative(head)) | |
1638 | goto repeat; | |
1639 | ||
1640 | /* The page was split under us? */ | |
1641 | if (compound_head(page) != head) { | |
1642 | put_page(head); | |
a60637c8 | 1643 | goto repeat; |
83929372 | 1644 | } |
a60637c8 NP |
1645 | |
1646 | /* Has the page moved? */ | |
0fc9d104 | 1647 | if (unlikely(page != *slot)) { |
83929372 | 1648 | put_page(head); |
a60637c8 NP |
1649 | goto repeat; |
1650 | } | |
1da177e4 | 1651 | |
a60637c8 | 1652 | pages[ret] = page; |
b947cee4 JK |
1653 | if (++ret == nr_pages) { |
1654 | *start = pages[ret - 1]->index + 1; | |
1655 | goto out; | |
1656 | } | |
a60637c8 | 1657 | } |
5b280c0c | 1658 | |
b947cee4 JK |
1659 | /* |
1660 | * We come here when there is no page beyond @end. We take care to not | |
1661 | * overflow the index @start as it confuses some of the callers. This | |
1662 | * breaks the iteration when there is page at index -1 but that is | |
1663 | * already broken anyway. | |
1664 | */ | |
1665 | if (end == (pgoff_t)-1) | |
1666 | *start = (pgoff_t)-1; | |
1667 | else | |
1668 | *start = end + 1; | |
1669 | out: | |
a60637c8 | 1670 | rcu_read_unlock(); |
d72dc8a2 | 1671 | |
1da177e4 LT |
1672 | return ret; |
1673 | } | |
1674 | ||
ebf43500 JA |
1675 | /** |
1676 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1677 | * @mapping: The address_space to search | |
1678 | * @index: The starting page index | |
1679 | * @nr_pages: The maximum number of pages | |
1680 | * @pages: Where the resulting pages are placed | |
1681 | * | |
1682 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1683 | * that the returned number of pages are guaranteed to be contiguous. | |
1684 | * | |
1685 | * find_get_pages_contig() returns the number of pages which were found. | |
1686 | */ | |
1687 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1688 | unsigned int nr_pages, struct page **pages) | |
1689 | { | |
0fc9d104 KK |
1690 | struct radix_tree_iter iter; |
1691 | void **slot; | |
1692 | unsigned int ret = 0; | |
1693 | ||
1694 | if (unlikely(!nr_pages)) | |
1695 | return 0; | |
a60637c8 NP |
1696 | |
1697 | rcu_read_lock(); | |
0fc9d104 | 1698 | radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) { |
83929372 | 1699 | struct page *head, *page; |
a60637c8 | 1700 | repeat: |
0fc9d104 KK |
1701 | page = radix_tree_deref_slot(slot); |
1702 | /* The hole, there no reason to continue */ | |
a60637c8 | 1703 | if (unlikely(!page)) |
0fc9d104 | 1704 | break; |
9d8aa4ea | 1705 | |
a2c16d6c | 1706 | if (radix_tree_exception(page)) { |
8079b1c8 | 1707 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1708 | slot = radix_tree_iter_retry(&iter); |
1709 | continue; | |
8079b1c8 | 1710 | } |
a2c16d6c | 1711 | /* |
139b6a6f JW |
1712 | * A shadow entry of a recently evicted page, |
1713 | * or a swap entry from shmem/tmpfs. Stop | |
1714 | * looking for contiguous pages. | |
a2c16d6c | 1715 | */ |
8079b1c8 | 1716 | break; |
a2c16d6c | 1717 | } |
ebf43500 | 1718 | |
83929372 KS |
1719 | head = compound_head(page); |
1720 | if (!page_cache_get_speculative(head)) | |
1721 | goto repeat; | |
1722 | ||
1723 | /* The page was split under us? */ | |
1724 | if (compound_head(page) != head) { | |
1725 | put_page(head); | |
a60637c8 | 1726 | goto repeat; |
83929372 | 1727 | } |
a60637c8 NP |
1728 | |
1729 | /* Has the page moved? */ | |
0fc9d104 | 1730 | if (unlikely(page != *slot)) { |
83929372 | 1731 | put_page(head); |
a60637c8 NP |
1732 | goto repeat; |
1733 | } | |
1734 | ||
9cbb4cb2 NP |
1735 | /* |
1736 | * must check mapping and index after taking the ref. | |
1737 | * otherwise we can get both false positives and false | |
1738 | * negatives, which is just confusing to the caller. | |
1739 | */ | |
83929372 | 1740 | if (page->mapping == NULL || page_to_pgoff(page) != iter.index) { |
09cbfeaf | 1741 | put_page(page); |
9cbb4cb2 NP |
1742 | break; |
1743 | } | |
1744 | ||
a60637c8 | 1745 | pages[ret] = page; |
0fc9d104 KK |
1746 | if (++ret == nr_pages) |
1747 | break; | |
ebf43500 | 1748 | } |
a60637c8 NP |
1749 | rcu_read_unlock(); |
1750 | return ret; | |
ebf43500 | 1751 | } |
ef71c15c | 1752 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1753 | |
485bb99b | 1754 | /** |
72b045ae | 1755 | * find_get_pages_range_tag - find and return pages in given range matching @tag |
485bb99b RD |
1756 | * @mapping: the address_space to search |
1757 | * @index: the starting page index | |
72b045ae | 1758 | * @end: The final page index (inclusive) |
485bb99b RD |
1759 | * @tag: the tag index |
1760 | * @nr_pages: the maximum number of pages | |
1761 | * @pages: where the resulting pages are placed | |
1762 | * | |
1da177e4 | 1763 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1764 | * @tag. We update @index to index the next page for the traversal. |
1da177e4 | 1765 | */ |
72b045ae JK |
1766 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
1767 | pgoff_t end, int tag, unsigned int nr_pages, | |
1768 | struct page **pages) | |
1da177e4 | 1769 | { |
0fc9d104 KK |
1770 | struct radix_tree_iter iter; |
1771 | void **slot; | |
1772 | unsigned ret = 0; | |
1773 | ||
1774 | if (unlikely(!nr_pages)) | |
1775 | return 0; | |
a60637c8 NP |
1776 | |
1777 | rcu_read_lock(); | |
0fc9d104 KK |
1778 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1779 | &iter, *index, tag) { | |
83929372 | 1780 | struct page *head, *page; |
72b045ae JK |
1781 | |
1782 | if (iter.index > end) | |
1783 | break; | |
a60637c8 | 1784 | repeat: |
0fc9d104 | 1785 | page = radix_tree_deref_slot(slot); |
a60637c8 NP |
1786 | if (unlikely(!page)) |
1787 | continue; | |
9d8aa4ea | 1788 | |
a2c16d6c | 1789 | if (radix_tree_exception(page)) { |
8079b1c8 | 1790 | if (radix_tree_deref_retry(page)) { |
2cf938aa MW |
1791 | slot = radix_tree_iter_retry(&iter); |
1792 | continue; | |
8079b1c8 | 1793 | } |
a2c16d6c | 1794 | /* |
139b6a6f JW |
1795 | * A shadow entry of a recently evicted page. |
1796 | * | |
1797 | * Those entries should never be tagged, but | |
1798 | * this tree walk is lockless and the tags are | |
1799 | * looked up in bulk, one radix tree node at a | |
1800 | * time, so there is a sizable window for page | |
1801 | * reclaim to evict a page we saw tagged. | |
1802 | * | |
1803 | * Skip over it. | |
a2c16d6c | 1804 | */ |
139b6a6f | 1805 | continue; |
a2c16d6c | 1806 | } |
a60637c8 | 1807 | |
83929372 KS |
1808 | head = compound_head(page); |
1809 | if (!page_cache_get_speculative(head)) | |
a60637c8 NP |
1810 | goto repeat; |
1811 | ||
83929372 KS |
1812 | /* The page was split under us? */ |
1813 | if (compound_head(page) != head) { | |
1814 | put_page(head); | |
1815 | goto repeat; | |
1816 | } | |
1817 | ||
a60637c8 | 1818 | /* Has the page moved? */ |
0fc9d104 | 1819 | if (unlikely(page != *slot)) { |
83929372 | 1820 | put_page(head); |
a60637c8 NP |
1821 | goto repeat; |
1822 | } | |
1823 | ||
1824 | pages[ret] = page; | |
72b045ae JK |
1825 | if (++ret == nr_pages) { |
1826 | *index = pages[ret - 1]->index + 1; | |
1827 | goto out; | |
1828 | } | |
a60637c8 | 1829 | } |
5b280c0c | 1830 | |
72b045ae JK |
1831 | /* |
1832 | * We come here when we got at @end. We take care to not overflow the | |
1833 | * index @index as it confuses some of the callers. This breaks the | |
1834 | * iteration when there is page at index -1 but that is already broken | |
1835 | * anyway. | |
1836 | */ | |
1837 | if (end == (pgoff_t)-1) | |
1838 | *index = (pgoff_t)-1; | |
1839 | else | |
1840 | *index = end + 1; | |
1841 | out: | |
a60637c8 | 1842 | rcu_read_unlock(); |
1da177e4 | 1843 | |
1da177e4 LT |
1844 | return ret; |
1845 | } | |
72b045ae | 1846 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 1847 | |
7e7f7749 RZ |
1848 | /** |
1849 | * find_get_entries_tag - find and return entries that match @tag | |
1850 | * @mapping: the address_space to search | |
1851 | * @start: the starting page cache index | |
1852 | * @tag: the tag index | |
1853 | * @nr_entries: the maximum number of entries | |
1854 | * @entries: where the resulting entries are placed | |
1855 | * @indices: the cache indices corresponding to the entries in @entries | |
1856 | * | |
1857 | * Like find_get_entries, except we only return entries which are tagged with | |
1858 | * @tag. | |
1859 | */ | |
1860 | unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start, | |
1861 | int tag, unsigned int nr_entries, | |
1862 | struct page **entries, pgoff_t *indices) | |
1863 | { | |
1864 | void **slot; | |
1865 | unsigned int ret = 0; | |
1866 | struct radix_tree_iter iter; | |
1867 | ||
1868 | if (!nr_entries) | |
1869 | return 0; | |
1870 | ||
1871 | rcu_read_lock(); | |
7e7f7749 RZ |
1872 | radix_tree_for_each_tagged(slot, &mapping->page_tree, |
1873 | &iter, start, tag) { | |
83929372 | 1874 | struct page *head, *page; |
7e7f7749 RZ |
1875 | repeat: |
1876 | page = radix_tree_deref_slot(slot); | |
1877 | if (unlikely(!page)) | |
1878 | continue; | |
1879 | if (radix_tree_exception(page)) { | |
1880 | if (radix_tree_deref_retry(page)) { | |
2cf938aa MW |
1881 | slot = radix_tree_iter_retry(&iter); |
1882 | continue; | |
7e7f7749 RZ |
1883 | } |
1884 | ||
1885 | /* | |
1886 | * A shadow entry of a recently evicted page, a swap | |
1887 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1888 | * without attempting to raise page count. | |
1889 | */ | |
1890 | goto export; | |
1891 | } | |
83929372 KS |
1892 | |
1893 | head = compound_head(page); | |
1894 | if (!page_cache_get_speculative(head)) | |
7e7f7749 RZ |
1895 | goto repeat; |
1896 | ||
83929372 KS |
1897 | /* The page was split under us? */ |
1898 | if (compound_head(page) != head) { | |
1899 | put_page(head); | |
1900 | goto repeat; | |
1901 | } | |
1902 | ||
7e7f7749 RZ |
1903 | /* Has the page moved? */ |
1904 | if (unlikely(page != *slot)) { | |
83929372 | 1905 | put_page(head); |
7e7f7749 RZ |
1906 | goto repeat; |
1907 | } | |
1908 | export: | |
1909 | indices[ret] = iter.index; | |
1910 | entries[ret] = page; | |
1911 | if (++ret == nr_entries) | |
1912 | break; | |
1913 | } | |
1914 | rcu_read_unlock(); | |
1915 | return ret; | |
1916 | } | |
1917 | EXPORT_SYMBOL(find_get_entries_tag); | |
1918 | ||
76d42bd9 WF |
1919 | /* |
1920 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
1921 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
1922 | * | |
1923 | * ---R__________________________________________B__________ | |
1924 | * ^ reading here ^ bad block(assume 4k) | |
1925 | * | |
1926 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
1927 | * => failing the whole request => read(R) => read(R+1) => | |
1928 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
1929 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
1930 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
1931 | * | |
1932 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
1933 | */ | |
1934 | static void shrink_readahead_size_eio(struct file *filp, | |
1935 | struct file_ra_state *ra) | |
1936 | { | |
76d42bd9 | 1937 | ra->ra_pages /= 4; |
76d42bd9 WF |
1938 | } |
1939 | ||
485bb99b | 1940 | /** |
47c27bc4 CH |
1941 | * generic_file_buffered_read - generic file read routine |
1942 | * @iocb: the iocb to read | |
6e58e79d AV |
1943 | * @iter: data destination |
1944 | * @written: already copied | |
485bb99b | 1945 | * |
1da177e4 | 1946 | * This is a generic file read routine, and uses the |
485bb99b | 1947 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
1948 | * |
1949 | * This is really ugly. But the goto's actually try to clarify some | |
1950 | * of the logic when it comes to error handling etc. | |
1da177e4 | 1951 | */ |
47c27bc4 | 1952 | static ssize_t generic_file_buffered_read(struct kiocb *iocb, |
6e58e79d | 1953 | struct iov_iter *iter, ssize_t written) |
1da177e4 | 1954 | { |
47c27bc4 | 1955 | struct file *filp = iocb->ki_filp; |
36e78914 | 1956 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 1957 | struct inode *inode = mapping->host; |
36e78914 | 1958 | struct file_ra_state *ra = &filp->f_ra; |
47c27bc4 | 1959 | loff_t *ppos = &iocb->ki_pos; |
57f6b96c FW |
1960 | pgoff_t index; |
1961 | pgoff_t last_index; | |
1962 | pgoff_t prev_index; | |
1963 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 1964 | unsigned int prev_offset; |
6e58e79d | 1965 | int error = 0; |
1da177e4 | 1966 | |
c2a9737f | 1967 | if (unlikely(*ppos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 1968 | return 0; |
c2a9737f WF |
1969 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
1970 | ||
09cbfeaf KS |
1971 | index = *ppos >> PAGE_SHIFT; |
1972 | prev_index = ra->prev_pos >> PAGE_SHIFT; | |
1973 | prev_offset = ra->prev_pos & (PAGE_SIZE-1); | |
1974 | last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; | |
1975 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 | 1976 | |
1da177e4 LT |
1977 | for (;;) { |
1978 | struct page *page; | |
57f6b96c | 1979 | pgoff_t end_index; |
a32ea1e1 | 1980 | loff_t isize; |
1da177e4 LT |
1981 | unsigned long nr, ret; |
1982 | ||
1da177e4 | 1983 | cond_resched(); |
1da177e4 | 1984 | find_page: |
5abf186a MH |
1985 | if (fatal_signal_pending(current)) { |
1986 | error = -EINTR; | |
1987 | goto out; | |
1988 | } | |
1989 | ||
1da177e4 | 1990 | page = find_get_page(mapping, index); |
3ea89ee8 | 1991 | if (!page) { |
3239d834 MT |
1992 | if (iocb->ki_flags & IOCB_NOWAIT) |
1993 | goto would_block; | |
cf914a7d | 1994 | page_cache_sync_readahead(mapping, |
7ff81078 | 1995 | ra, filp, |
3ea89ee8 FW |
1996 | index, last_index - index); |
1997 | page = find_get_page(mapping, index); | |
1998 | if (unlikely(page == NULL)) | |
1999 | goto no_cached_page; | |
2000 | } | |
2001 | if (PageReadahead(page)) { | |
cf914a7d | 2002 | page_cache_async_readahead(mapping, |
7ff81078 | 2003 | ra, filp, page, |
3ea89ee8 | 2004 | index, last_index - index); |
1da177e4 | 2005 | } |
8ab22b9a | 2006 | if (!PageUptodate(page)) { |
3239d834 MT |
2007 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2008 | put_page(page); | |
2009 | goto would_block; | |
2010 | } | |
2011 | ||
ebded027 MG |
2012 | /* |
2013 | * See comment in do_read_cache_page on why | |
2014 | * wait_on_page_locked is used to avoid unnecessarily | |
2015 | * serialisations and why it's safe. | |
2016 | */ | |
c4b209a4 BVA |
2017 | error = wait_on_page_locked_killable(page); |
2018 | if (unlikely(error)) | |
2019 | goto readpage_error; | |
ebded027 MG |
2020 | if (PageUptodate(page)) |
2021 | goto page_ok; | |
2022 | ||
09cbfeaf | 2023 | if (inode->i_blkbits == PAGE_SHIFT || |
8ab22b9a HH |
2024 | !mapping->a_ops->is_partially_uptodate) |
2025 | goto page_not_up_to_date; | |
6d6d36bc EG |
2026 | /* pipes can't handle partially uptodate pages */ |
2027 | if (unlikely(iter->type & ITER_PIPE)) | |
2028 | goto page_not_up_to_date; | |
529ae9aa | 2029 | if (!trylock_page(page)) |
8ab22b9a | 2030 | goto page_not_up_to_date; |
8d056cb9 DH |
2031 | /* Did it get truncated before we got the lock? */ |
2032 | if (!page->mapping) | |
2033 | goto page_not_up_to_date_locked; | |
8ab22b9a | 2034 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 2035 | offset, iter->count)) |
8ab22b9a HH |
2036 | goto page_not_up_to_date_locked; |
2037 | unlock_page(page); | |
2038 | } | |
1da177e4 | 2039 | page_ok: |
a32ea1e1 N |
2040 | /* |
2041 | * i_size must be checked after we know the page is Uptodate. | |
2042 | * | |
2043 | * Checking i_size after the check allows us to calculate | |
2044 | * the correct value for "nr", which means the zero-filled | |
2045 | * part of the page is not copied back to userspace (unless | |
2046 | * another truncate extends the file - this is desired though). | |
2047 | */ | |
2048 | ||
2049 | isize = i_size_read(inode); | |
09cbfeaf | 2050 | end_index = (isize - 1) >> PAGE_SHIFT; |
a32ea1e1 | 2051 | if (unlikely(!isize || index > end_index)) { |
09cbfeaf | 2052 | put_page(page); |
a32ea1e1 N |
2053 | goto out; |
2054 | } | |
2055 | ||
2056 | /* nr is the maximum number of bytes to copy from this page */ | |
09cbfeaf | 2057 | nr = PAGE_SIZE; |
a32ea1e1 | 2058 | if (index == end_index) { |
09cbfeaf | 2059 | nr = ((isize - 1) & ~PAGE_MASK) + 1; |
a32ea1e1 | 2060 | if (nr <= offset) { |
09cbfeaf | 2061 | put_page(page); |
a32ea1e1 N |
2062 | goto out; |
2063 | } | |
2064 | } | |
2065 | nr = nr - offset; | |
1da177e4 LT |
2066 | |
2067 | /* If users can be writing to this page using arbitrary | |
2068 | * virtual addresses, take care about potential aliasing | |
2069 | * before reading the page on the kernel side. | |
2070 | */ | |
2071 | if (mapping_writably_mapped(mapping)) | |
2072 | flush_dcache_page(page); | |
2073 | ||
2074 | /* | |
ec0f1637 JK |
2075 | * When a sequential read accesses a page several times, |
2076 | * only mark it as accessed the first time. | |
1da177e4 | 2077 | */ |
ec0f1637 | 2078 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
2079 | mark_page_accessed(page); |
2080 | prev_index = index; | |
2081 | ||
2082 | /* | |
2083 | * Ok, we have the page, and it's up-to-date, so | |
2084 | * now we can copy it to user space... | |
1da177e4 | 2085 | */ |
6e58e79d AV |
2086 | |
2087 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 | 2088 | offset += ret; |
09cbfeaf KS |
2089 | index += offset >> PAGE_SHIFT; |
2090 | offset &= ~PAGE_MASK; | |
6ce745ed | 2091 | prev_offset = offset; |
1da177e4 | 2092 | |
09cbfeaf | 2093 | put_page(page); |
6e58e79d AV |
2094 | written += ret; |
2095 | if (!iov_iter_count(iter)) | |
2096 | goto out; | |
2097 | if (ret < nr) { | |
2098 | error = -EFAULT; | |
2099 | goto out; | |
2100 | } | |
2101 | continue; | |
1da177e4 LT |
2102 | |
2103 | page_not_up_to_date: | |
2104 | /* Get exclusive access to the page ... */ | |
85462323 ON |
2105 | error = lock_page_killable(page); |
2106 | if (unlikely(error)) | |
2107 | goto readpage_error; | |
1da177e4 | 2108 | |
8ab22b9a | 2109 | page_not_up_to_date_locked: |
da6052f7 | 2110 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
2111 | if (!page->mapping) { |
2112 | unlock_page(page); | |
09cbfeaf | 2113 | put_page(page); |
1da177e4 LT |
2114 | continue; |
2115 | } | |
2116 | ||
2117 | /* Did somebody else fill it already? */ | |
2118 | if (PageUptodate(page)) { | |
2119 | unlock_page(page); | |
2120 | goto page_ok; | |
2121 | } | |
2122 | ||
2123 | readpage: | |
91803b49 JM |
2124 | /* |
2125 | * A previous I/O error may have been due to temporary | |
2126 | * failures, eg. multipath errors. | |
2127 | * PG_error will be set again if readpage fails. | |
2128 | */ | |
2129 | ClearPageError(page); | |
1da177e4 LT |
2130 | /* Start the actual read. The read will unlock the page. */ |
2131 | error = mapping->a_ops->readpage(filp, page); | |
2132 | ||
994fc28c ZB |
2133 | if (unlikely(error)) { |
2134 | if (error == AOP_TRUNCATED_PAGE) { | |
09cbfeaf | 2135 | put_page(page); |
6e58e79d | 2136 | error = 0; |
994fc28c ZB |
2137 | goto find_page; |
2138 | } | |
1da177e4 | 2139 | goto readpage_error; |
994fc28c | 2140 | } |
1da177e4 LT |
2141 | |
2142 | if (!PageUptodate(page)) { | |
85462323 ON |
2143 | error = lock_page_killable(page); |
2144 | if (unlikely(error)) | |
2145 | goto readpage_error; | |
1da177e4 LT |
2146 | if (!PageUptodate(page)) { |
2147 | if (page->mapping == NULL) { | |
2148 | /* | |
2ecdc82e | 2149 | * invalidate_mapping_pages got it |
1da177e4 LT |
2150 | */ |
2151 | unlock_page(page); | |
09cbfeaf | 2152 | put_page(page); |
1da177e4 LT |
2153 | goto find_page; |
2154 | } | |
2155 | unlock_page(page); | |
7ff81078 | 2156 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
2157 | error = -EIO; |
2158 | goto readpage_error; | |
1da177e4 LT |
2159 | } |
2160 | unlock_page(page); | |
2161 | } | |
2162 | ||
1da177e4 LT |
2163 | goto page_ok; |
2164 | ||
2165 | readpage_error: | |
2166 | /* UHHUH! A synchronous read error occurred. Report it */ | |
09cbfeaf | 2167 | put_page(page); |
1da177e4 LT |
2168 | goto out; |
2169 | ||
2170 | no_cached_page: | |
2171 | /* | |
2172 | * Ok, it wasn't cached, so we need to create a new | |
2173 | * page.. | |
2174 | */ | |
eb2be189 NP |
2175 | page = page_cache_alloc_cold(mapping); |
2176 | if (!page) { | |
6e58e79d | 2177 | error = -ENOMEM; |
eb2be189 | 2178 | goto out; |
1da177e4 | 2179 | } |
6afdb859 | 2180 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 2181 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 2182 | if (error) { |
09cbfeaf | 2183 | put_page(page); |
6e58e79d AV |
2184 | if (error == -EEXIST) { |
2185 | error = 0; | |
1da177e4 | 2186 | goto find_page; |
6e58e79d | 2187 | } |
1da177e4 LT |
2188 | goto out; |
2189 | } | |
1da177e4 LT |
2190 | goto readpage; |
2191 | } | |
2192 | ||
3239d834 MT |
2193 | would_block: |
2194 | error = -EAGAIN; | |
1da177e4 | 2195 | out: |
7ff81078 | 2196 | ra->prev_pos = prev_index; |
09cbfeaf | 2197 | ra->prev_pos <<= PAGE_SHIFT; |
7ff81078 | 2198 | ra->prev_pos |= prev_offset; |
1da177e4 | 2199 | |
09cbfeaf | 2200 | *ppos = ((loff_t)index << PAGE_SHIFT) + offset; |
0c6aa263 | 2201 | file_accessed(filp); |
6e58e79d | 2202 | return written ? written : error; |
1da177e4 LT |
2203 | } |
2204 | ||
485bb99b | 2205 | /** |
6abd2322 | 2206 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2207 | * @iocb: kernel I/O control block |
6abd2322 | 2208 | * @iter: destination for the data read |
485bb99b | 2209 | * |
6abd2322 | 2210 | * This is the "read_iter()" routine for all filesystems |
1da177e4 LT |
2211 | * that can use the page cache directly. |
2212 | */ | |
2213 | ssize_t | |
ed978a81 | 2214 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2215 | { |
e7080a43 | 2216 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2217 | ssize_t retval = 0; |
e7080a43 NS |
2218 | |
2219 | if (!count) | |
2220 | goto out; /* skip atime */ | |
1da177e4 | 2221 | |
2ba48ce5 | 2222 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2223 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2224 | struct address_space *mapping = file->f_mapping; |
2225 | struct inode *inode = mapping->host; | |
543ade1f | 2226 | loff_t size; |
1da177e4 | 2227 | |
1da177e4 | 2228 | size = i_size_read(inode); |
6be96d3a GR |
2229 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2230 | if (filemap_range_has_page(mapping, iocb->ki_pos, | |
2231 | iocb->ki_pos + count - 1)) | |
2232 | return -EAGAIN; | |
2233 | } else { | |
2234 | retval = filemap_write_and_wait_range(mapping, | |
2235 | iocb->ki_pos, | |
2236 | iocb->ki_pos + count - 1); | |
2237 | if (retval < 0) | |
2238 | goto out; | |
2239 | } | |
d8d3d94b | 2240 | |
0d5b0cf2 CH |
2241 | file_accessed(file); |
2242 | ||
5ecda137 | 2243 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2244 | if (retval >= 0) { |
c64fb5c7 | 2245 | iocb->ki_pos += retval; |
5ecda137 | 2246 | count -= retval; |
9fe55eea | 2247 | } |
5b47d59a | 2248 | iov_iter_revert(iter, count - iov_iter_count(iter)); |
66f998f6 | 2249 | |
9fe55eea SW |
2250 | /* |
2251 | * Btrfs can have a short DIO read if we encounter | |
2252 | * compressed extents, so if there was an error, or if | |
2253 | * we've already read everything we wanted to, or if | |
2254 | * there was a short read because we hit EOF, go ahead | |
2255 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2256 | * the rest of the read. Buffered reads will not work for |
2257 | * DAX files, so don't bother trying. | |
9fe55eea | 2258 | */ |
5ecda137 | 2259 | if (retval < 0 || !count || iocb->ki_pos >= size || |
0d5b0cf2 | 2260 | IS_DAX(inode)) |
9fe55eea | 2261 | goto out; |
1da177e4 LT |
2262 | } |
2263 | ||
47c27bc4 | 2264 | retval = generic_file_buffered_read(iocb, iter, retval); |
1da177e4 LT |
2265 | out: |
2266 | return retval; | |
2267 | } | |
ed978a81 | 2268 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2269 | |
1da177e4 | 2270 | #ifdef CONFIG_MMU |
485bb99b RD |
2271 | /** |
2272 | * page_cache_read - adds requested page to the page cache if not already there | |
2273 | * @file: file to read | |
2274 | * @offset: page index | |
62eb320a | 2275 | * @gfp_mask: memory allocation flags |
485bb99b | 2276 | * |
1da177e4 LT |
2277 | * This adds the requested page to the page cache if it isn't already there, |
2278 | * and schedules an I/O to read in its contents from disk. | |
2279 | */ | |
c20cd45e | 2280 | static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask) |
1da177e4 LT |
2281 | { |
2282 | struct address_space *mapping = file->f_mapping; | |
99dadfdd | 2283 | struct page *page; |
994fc28c | 2284 | int ret; |
1da177e4 | 2285 | |
994fc28c | 2286 | do { |
c20cd45e | 2287 | page = __page_cache_alloc(gfp_mask|__GFP_COLD); |
994fc28c ZB |
2288 | if (!page) |
2289 | return -ENOMEM; | |
2290 | ||
c20cd45e | 2291 | ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask & GFP_KERNEL); |
994fc28c ZB |
2292 | if (ret == 0) |
2293 | ret = mapping->a_ops->readpage(file, page); | |
2294 | else if (ret == -EEXIST) | |
2295 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 2296 | |
09cbfeaf | 2297 | put_page(page); |
1da177e4 | 2298 | |
994fc28c | 2299 | } while (ret == AOP_TRUNCATED_PAGE); |
99dadfdd | 2300 | |
994fc28c | 2301 | return ret; |
1da177e4 LT |
2302 | } |
2303 | ||
2304 | #define MMAP_LOTSAMISS (100) | |
2305 | ||
ef00e08e LT |
2306 | /* |
2307 | * Synchronous readahead happens when we don't even find | |
2308 | * a page in the page cache at all. | |
2309 | */ | |
2310 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
2311 | struct file_ra_state *ra, | |
2312 | struct file *file, | |
2313 | pgoff_t offset) | |
2314 | { | |
ef00e08e LT |
2315 | struct address_space *mapping = file->f_mapping; |
2316 | ||
2317 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 2318 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 2319 | return; |
275b12bf WF |
2320 | if (!ra->ra_pages) |
2321 | return; | |
ef00e08e | 2322 | |
64363aad | 2323 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
2324 | page_cache_sync_readahead(mapping, ra, file, offset, |
2325 | ra->ra_pages); | |
ef00e08e LT |
2326 | return; |
2327 | } | |
2328 | ||
207d04ba AK |
2329 | /* Avoid banging the cache line if not needed */ |
2330 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
2331 | ra->mmap_miss++; |
2332 | ||
2333 | /* | |
2334 | * Do we miss much more than hit in this file? If so, | |
2335 | * stop bothering with read-ahead. It will only hurt. | |
2336 | */ | |
2337 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
2338 | return; | |
2339 | ||
d30a1100 WF |
2340 | /* |
2341 | * mmap read-around | |
2342 | */ | |
600e19af RG |
2343 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
2344 | ra->size = ra->ra_pages; | |
2345 | ra->async_size = ra->ra_pages / 4; | |
275b12bf | 2346 | ra_submit(ra, mapping, file); |
ef00e08e LT |
2347 | } |
2348 | ||
2349 | /* | |
2350 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
2351 | * so we want to possibly extend the readahead further.. | |
2352 | */ | |
2353 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
2354 | struct file_ra_state *ra, | |
2355 | struct file *file, | |
2356 | struct page *page, | |
2357 | pgoff_t offset) | |
2358 | { | |
2359 | struct address_space *mapping = file->f_mapping; | |
2360 | ||
2361 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 2362 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
2363 | return; |
2364 | if (ra->mmap_miss > 0) | |
2365 | ra->mmap_miss--; | |
2366 | if (PageReadahead(page)) | |
2fad6f5d WF |
2367 | page_cache_async_readahead(mapping, ra, file, |
2368 | page, offset, ra->ra_pages); | |
ef00e08e LT |
2369 | } |
2370 | ||
485bb99b | 2371 | /** |
54cb8821 | 2372 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 2373 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 2374 | * |
54cb8821 | 2375 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2376 | * mapped memory region to read in file data during a page fault. |
2377 | * | |
2378 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2379 | * it in the page cache, and handles the special cases reasonably without | |
2380 | * having a lot of duplicated code. | |
9a95f3cf PC |
2381 | * |
2382 | * vma->vm_mm->mmap_sem must be held on entry. | |
2383 | * | |
2384 | * If our return value has VM_FAULT_RETRY set, it's because | |
2385 | * lock_page_or_retry() returned 0. | |
2386 | * The mmap_sem has usually been released in this case. | |
2387 | * See __lock_page_or_retry() for the exception. | |
2388 | * | |
2389 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
2390 | * has not been released. | |
2391 | * | |
2392 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
1da177e4 | 2393 | */ |
11bac800 | 2394 | int filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
2395 | { |
2396 | int error; | |
11bac800 | 2397 | struct file *file = vmf->vma->vm_file; |
1da177e4 LT |
2398 | struct address_space *mapping = file->f_mapping; |
2399 | struct file_ra_state *ra = &file->f_ra; | |
2400 | struct inode *inode = mapping->host; | |
ef00e08e | 2401 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 2402 | pgoff_t max_off; |
1da177e4 | 2403 | struct page *page; |
83c54070 | 2404 | int ret = 0; |
1da177e4 | 2405 | |
9ab2594f MW |
2406 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2407 | if (unlikely(offset >= max_off)) | |
5307cc1a | 2408 | return VM_FAULT_SIGBUS; |
1da177e4 | 2409 | |
1da177e4 | 2410 | /* |
49426420 | 2411 | * Do we have something in the page cache already? |
1da177e4 | 2412 | */ |
ef00e08e | 2413 | page = find_get_page(mapping, offset); |
45cac65b | 2414 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2415 | /* |
ef00e08e LT |
2416 | * We found the page, so try async readahead before |
2417 | * waiting for the lock. | |
1da177e4 | 2418 | */ |
11bac800 | 2419 | do_async_mmap_readahead(vmf->vma, ra, file, page, offset); |
45cac65b | 2420 | } else if (!page) { |
ef00e08e | 2421 | /* No page in the page cache at all */ |
11bac800 | 2422 | do_sync_mmap_readahead(vmf->vma, ra, file, offset); |
ef00e08e | 2423 | count_vm_event(PGMAJFAULT); |
2262185c | 2424 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
2425 | ret = VM_FAULT_MAJOR; |
2426 | retry_find: | |
b522c94d | 2427 | page = find_get_page(mapping, offset); |
1da177e4 LT |
2428 | if (!page) |
2429 | goto no_cached_page; | |
2430 | } | |
2431 | ||
11bac800 | 2432 | if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) { |
09cbfeaf | 2433 | put_page(page); |
d065bd81 | 2434 | return ret | VM_FAULT_RETRY; |
d88c0922 | 2435 | } |
b522c94d ML |
2436 | |
2437 | /* Did it get truncated? */ | |
2438 | if (unlikely(page->mapping != mapping)) { | |
2439 | unlock_page(page); | |
2440 | put_page(page); | |
2441 | goto retry_find; | |
2442 | } | |
309381fe | 2443 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 2444 | |
1da177e4 | 2445 | /* |
d00806b1 NP |
2446 | * We have a locked page in the page cache, now we need to check |
2447 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2448 | */ |
d00806b1 | 2449 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2450 | goto page_not_uptodate; |
2451 | ||
ef00e08e LT |
2452 | /* |
2453 | * Found the page and have a reference on it. | |
2454 | * We must recheck i_size under page lock. | |
2455 | */ | |
9ab2594f MW |
2456 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2457 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 2458 | unlock_page(page); |
09cbfeaf | 2459 | put_page(page); |
5307cc1a | 2460 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2461 | } |
2462 | ||
d0217ac0 | 2463 | vmf->page = page; |
83c54070 | 2464 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2465 | |
1da177e4 LT |
2466 | no_cached_page: |
2467 | /* | |
2468 | * We're only likely to ever get here if MADV_RANDOM is in | |
2469 | * effect. | |
2470 | */ | |
c20cd45e | 2471 | error = page_cache_read(file, offset, vmf->gfp_mask); |
1da177e4 LT |
2472 | |
2473 | /* | |
2474 | * The page we want has now been added to the page cache. | |
2475 | * In the unlikely event that someone removed it in the | |
2476 | * meantime, we'll just come back here and read it again. | |
2477 | */ | |
2478 | if (error >= 0) | |
2479 | goto retry_find; | |
2480 | ||
2481 | /* | |
2482 | * An error return from page_cache_read can result if the | |
2483 | * system is low on memory, or a problem occurs while trying | |
2484 | * to schedule I/O. | |
2485 | */ | |
2486 | if (error == -ENOMEM) | |
d0217ac0 NP |
2487 | return VM_FAULT_OOM; |
2488 | return VM_FAULT_SIGBUS; | |
1da177e4 LT |
2489 | |
2490 | page_not_uptodate: | |
1da177e4 LT |
2491 | /* |
2492 | * Umm, take care of errors if the page isn't up-to-date. | |
2493 | * Try to re-read it _once_. We do this synchronously, | |
2494 | * because there really aren't any performance issues here | |
2495 | * and we need to check for errors. | |
2496 | */ | |
1da177e4 | 2497 | ClearPageError(page); |
994fc28c | 2498 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2499 | if (!error) { |
2500 | wait_on_page_locked(page); | |
2501 | if (!PageUptodate(page)) | |
2502 | error = -EIO; | |
2503 | } | |
09cbfeaf | 2504 | put_page(page); |
d00806b1 NP |
2505 | |
2506 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2507 | goto retry_find; |
1da177e4 | 2508 | |
d00806b1 | 2509 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 2510 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 2511 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
2512 | } |
2513 | EXPORT_SYMBOL(filemap_fault); | |
2514 | ||
82b0f8c3 | 2515 | void filemap_map_pages(struct vm_fault *vmf, |
bae473a4 | 2516 | pgoff_t start_pgoff, pgoff_t end_pgoff) |
f1820361 KS |
2517 | { |
2518 | struct radix_tree_iter iter; | |
2519 | void **slot; | |
82b0f8c3 | 2520 | struct file *file = vmf->vma->vm_file; |
f1820361 | 2521 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 2522 | pgoff_t last_pgoff = start_pgoff; |
9ab2594f | 2523 | unsigned long max_idx; |
83929372 | 2524 | struct page *head, *page; |
f1820361 KS |
2525 | |
2526 | rcu_read_lock(); | |
bae473a4 KS |
2527 | radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, |
2528 | start_pgoff) { | |
2529 | if (iter.index > end_pgoff) | |
f1820361 KS |
2530 | break; |
2531 | repeat: | |
2532 | page = radix_tree_deref_slot(slot); | |
2533 | if (unlikely(!page)) | |
2534 | goto next; | |
2535 | if (radix_tree_exception(page)) { | |
2cf938aa MW |
2536 | if (radix_tree_deref_retry(page)) { |
2537 | slot = radix_tree_iter_retry(&iter); | |
2538 | continue; | |
2539 | } | |
2540 | goto next; | |
f1820361 KS |
2541 | } |
2542 | ||
83929372 KS |
2543 | head = compound_head(page); |
2544 | if (!page_cache_get_speculative(head)) | |
f1820361 KS |
2545 | goto repeat; |
2546 | ||
83929372 KS |
2547 | /* The page was split under us? */ |
2548 | if (compound_head(page) != head) { | |
2549 | put_page(head); | |
2550 | goto repeat; | |
2551 | } | |
2552 | ||
f1820361 KS |
2553 | /* Has the page moved? */ |
2554 | if (unlikely(page != *slot)) { | |
83929372 | 2555 | put_page(head); |
f1820361 KS |
2556 | goto repeat; |
2557 | } | |
2558 | ||
2559 | if (!PageUptodate(page) || | |
2560 | PageReadahead(page) || | |
2561 | PageHWPoison(page)) | |
2562 | goto skip; | |
2563 | if (!trylock_page(page)) | |
2564 | goto skip; | |
2565 | ||
2566 | if (page->mapping != mapping || !PageUptodate(page)) | |
2567 | goto unlock; | |
2568 | ||
9ab2594f MW |
2569 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); |
2570 | if (page->index >= max_idx) | |
f1820361 KS |
2571 | goto unlock; |
2572 | ||
f1820361 KS |
2573 | if (file->f_ra.mmap_miss > 0) |
2574 | file->f_ra.mmap_miss--; | |
7267ec00 | 2575 | |
82b0f8c3 JK |
2576 | vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT; |
2577 | if (vmf->pte) | |
2578 | vmf->pte += iter.index - last_pgoff; | |
7267ec00 | 2579 | last_pgoff = iter.index; |
82b0f8c3 | 2580 | if (alloc_set_pte(vmf, NULL, page)) |
7267ec00 | 2581 | goto unlock; |
f1820361 KS |
2582 | unlock_page(page); |
2583 | goto next; | |
2584 | unlock: | |
2585 | unlock_page(page); | |
2586 | skip: | |
09cbfeaf | 2587 | put_page(page); |
f1820361 | 2588 | next: |
7267ec00 | 2589 | /* Huge page is mapped? No need to proceed. */ |
82b0f8c3 | 2590 | if (pmd_trans_huge(*vmf->pmd)) |
7267ec00 | 2591 | break; |
bae473a4 | 2592 | if (iter.index == end_pgoff) |
f1820361 KS |
2593 | break; |
2594 | } | |
2595 | rcu_read_unlock(); | |
2596 | } | |
2597 | EXPORT_SYMBOL(filemap_map_pages); | |
2598 | ||
11bac800 | 2599 | int filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 JK |
2600 | { |
2601 | struct page *page = vmf->page; | |
11bac800 | 2602 | struct inode *inode = file_inode(vmf->vma->vm_file); |
4fcf1c62 JK |
2603 | int ret = VM_FAULT_LOCKED; |
2604 | ||
14da9200 | 2605 | sb_start_pagefault(inode->i_sb); |
11bac800 | 2606 | file_update_time(vmf->vma->vm_file); |
4fcf1c62 JK |
2607 | lock_page(page); |
2608 | if (page->mapping != inode->i_mapping) { | |
2609 | unlock_page(page); | |
2610 | ret = VM_FAULT_NOPAGE; | |
2611 | goto out; | |
2612 | } | |
14da9200 JK |
2613 | /* |
2614 | * We mark the page dirty already here so that when freeze is in | |
2615 | * progress, we are guaranteed that writeback during freezing will | |
2616 | * see the dirty page and writeprotect it again. | |
2617 | */ | |
2618 | set_page_dirty(page); | |
1d1d1a76 | 2619 | wait_for_stable_page(page); |
4fcf1c62 | 2620 | out: |
14da9200 | 2621 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2622 | return ret; |
2623 | } | |
2624 | EXPORT_SYMBOL(filemap_page_mkwrite); | |
2625 | ||
f0f37e2f | 2626 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2627 | .fault = filemap_fault, |
f1820361 | 2628 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2629 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2630 | }; |
2631 | ||
2632 | /* This is used for a general mmap of a disk file */ | |
2633 | ||
2634 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2635 | { | |
2636 | struct address_space *mapping = file->f_mapping; | |
2637 | ||
2638 | if (!mapping->a_ops->readpage) | |
2639 | return -ENOEXEC; | |
2640 | file_accessed(file); | |
2641 | vma->vm_ops = &generic_file_vm_ops; | |
2642 | return 0; | |
2643 | } | |
1da177e4 LT |
2644 | |
2645 | /* | |
2646 | * This is for filesystems which do not implement ->writepage. | |
2647 | */ | |
2648 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2649 | { | |
2650 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2651 | return -EINVAL; | |
2652 | return generic_file_mmap(file, vma); | |
2653 | } | |
2654 | #else | |
2655 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2656 | { | |
2657 | return -ENOSYS; | |
2658 | } | |
2659 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2660 | { | |
2661 | return -ENOSYS; | |
2662 | } | |
2663 | #endif /* CONFIG_MMU */ | |
2664 | ||
2665 | EXPORT_SYMBOL(generic_file_mmap); | |
2666 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2667 | ||
67f9fd91 SL |
2668 | static struct page *wait_on_page_read(struct page *page) |
2669 | { | |
2670 | if (!IS_ERR(page)) { | |
2671 | wait_on_page_locked(page); | |
2672 | if (!PageUptodate(page)) { | |
09cbfeaf | 2673 | put_page(page); |
67f9fd91 SL |
2674 | page = ERR_PTR(-EIO); |
2675 | } | |
2676 | } | |
2677 | return page; | |
2678 | } | |
2679 | ||
32b63529 | 2680 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2681 | pgoff_t index, |
5e5358e7 | 2682 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2683 | void *data, |
2684 | gfp_t gfp) | |
1da177e4 | 2685 | { |
eb2be189 | 2686 | struct page *page; |
1da177e4 LT |
2687 | int err; |
2688 | repeat: | |
2689 | page = find_get_page(mapping, index); | |
2690 | if (!page) { | |
0531b2aa | 2691 | page = __page_cache_alloc(gfp | __GFP_COLD); |
eb2be189 NP |
2692 | if (!page) |
2693 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2694 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 2695 | if (unlikely(err)) { |
09cbfeaf | 2696 | put_page(page); |
eb2be189 NP |
2697 | if (err == -EEXIST) |
2698 | goto repeat; | |
1da177e4 | 2699 | /* Presumably ENOMEM for radix tree node */ |
1da177e4 LT |
2700 | return ERR_PTR(err); |
2701 | } | |
32b63529 MG |
2702 | |
2703 | filler: | |
1da177e4 LT |
2704 | err = filler(data, page); |
2705 | if (err < 0) { | |
09cbfeaf | 2706 | put_page(page); |
32b63529 | 2707 | return ERR_PTR(err); |
1da177e4 | 2708 | } |
1da177e4 | 2709 | |
32b63529 MG |
2710 | page = wait_on_page_read(page); |
2711 | if (IS_ERR(page)) | |
2712 | return page; | |
2713 | goto out; | |
2714 | } | |
1da177e4 LT |
2715 | if (PageUptodate(page)) |
2716 | goto out; | |
2717 | ||
ebded027 MG |
2718 | /* |
2719 | * Page is not up to date and may be locked due one of the following | |
2720 | * case a: Page is being filled and the page lock is held | |
2721 | * case b: Read/write error clearing the page uptodate status | |
2722 | * case c: Truncation in progress (page locked) | |
2723 | * case d: Reclaim in progress | |
2724 | * | |
2725 | * Case a, the page will be up to date when the page is unlocked. | |
2726 | * There is no need to serialise on the page lock here as the page | |
2727 | * is pinned so the lock gives no additional protection. Even if the | |
2728 | * the page is truncated, the data is still valid if PageUptodate as | |
2729 | * it's a race vs truncate race. | |
2730 | * Case b, the page will not be up to date | |
2731 | * Case c, the page may be truncated but in itself, the data may still | |
2732 | * be valid after IO completes as it's a read vs truncate race. The | |
2733 | * operation must restart if the page is not uptodate on unlock but | |
2734 | * otherwise serialising on page lock to stabilise the mapping gives | |
2735 | * no additional guarantees to the caller as the page lock is | |
2736 | * released before return. | |
2737 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
2738 | * will be a race with remove_mapping that determines if the mapping | |
2739 | * is valid on unlock but otherwise the data is valid and there is | |
2740 | * no need to serialise with page lock. | |
2741 | * | |
2742 | * As the page lock gives no additional guarantee, we optimistically | |
2743 | * wait on the page to be unlocked and check if it's up to date and | |
2744 | * use the page if it is. Otherwise, the page lock is required to | |
2745 | * distinguish between the different cases. The motivation is that we | |
2746 | * avoid spurious serialisations and wakeups when multiple processes | |
2747 | * wait on the same page for IO to complete. | |
2748 | */ | |
2749 | wait_on_page_locked(page); | |
2750 | if (PageUptodate(page)) | |
2751 | goto out; | |
2752 | ||
2753 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 2754 | lock_page(page); |
ebded027 MG |
2755 | |
2756 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
2757 | if (!page->mapping) { |
2758 | unlock_page(page); | |
09cbfeaf | 2759 | put_page(page); |
32b63529 | 2760 | goto repeat; |
1da177e4 | 2761 | } |
ebded027 MG |
2762 | |
2763 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
2764 | if (PageUptodate(page)) { |
2765 | unlock_page(page); | |
2766 | goto out; | |
2767 | } | |
32b63529 MG |
2768 | goto filler; |
2769 | ||
c855ff37 | 2770 | out: |
6fe6900e NP |
2771 | mark_page_accessed(page); |
2772 | return page; | |
2773 | } | |
0531b2aa LT |
2774 | |
2775 | /** | |
67f9fd91 | 2776 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2777 | * @mapping: the page's address_space |
2778 | * @index: the page index | |
2779 | * @filler: function to perform the read | |
5e5358e7 | 2780 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2781 | * |
0531b2aa | 2782 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2783 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2784 | * |
2785 | * If the page does not get brought uptodate, return -EIO. | |
2786 | */ | |
67f9fd91 | 2787 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2788 | pgoff_t index, |
5e5358e7 | 2789 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2790 | void *data) |
2791 | { | |
2792 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2793 | } | |
67f9fd91 | 2794 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2795 | |
2796 | /** | |
2797 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2798 | * @mapping: the page's address_space | |
2799 | * @index: the page index | |
2800 | * @gfp: the page allocator flags to use if allocating | |
2801 | * | |
2802 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2803 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2804 | * |
2805 | * If the page does not get brought uptodate, return -EIO. | |
2806 | */ | |
2807 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2808 | pgoff_t index, | |
2809 | gfp_t gfp) | |
2810 | { | |
2811 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2812 | ||
67f9fd91 | 2813 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2814 | } |
2815 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2816 | ||
1da177e4 LT |
2817 | /* |
2818 | * Performs necessary checks before doing a write | |
2819 | * | |
485bb99b | 2820 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2821 | * Returns appropriate error code that caller should return or |
2822 | * zero in case that write should be allowed. | |
2823 | */ | |
3309dd04 | 2824 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2825 | { |
3309dd04 | 2826 | struct file *file = iocb->ki_filp; |
1da177e4 | 2827 | struct inode *inode = file->f_mapping->host; |
59e99e5b | 2828 | unsigned long limit = rlimit(RLIMIT_FSIZE); |
3309dd04 | 2829 | loff_t pos; |
1da177e4 | 2830 | |
3309dd04 AV |
2831 | if (!iov_iter_count(from)) |
2832 | return 0; | |
1da177e4 | 2833 | |
0fa6b005 | 2834 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2835 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2836 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2837 | |
3309dd04 | 2838 | pos = iocb->ki_pos; |
1da177e4 | 2839 | |
6be96d3a GR |
2840 | if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT)) |
2841 | return -EINVAL; | |
2842 | ||
0fa6b005 | 2843 | if (limit != RLIM_INFINITY) { |
3309dd04 | 2844 | if (iocb->ki_pos >= limit) { |
0fa6b005 AV |
2845 | send_sig(SIGXFSZ, current, 0); |
2846 | return -EFBIG; | |
1da177e4 | 2847 | } |
3309dd04 | 2848 | iov_iter_truncate(from, limit - (unsigned long)pos); |
1da177e4 LT |
2849 | } |
2850 | ||
2851 | /* | |
2852 | * LFS rule | |
2853 | */ | |
3309dd04 | 2854 | if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS && |
1da177e4 | 2855 | !(file->f_flags & O_LARGEFILE))) { |
3309dd04 | 2856 | if (pos >= MAX_NON_LFS) |
1da177e4 | 2857 | return -EFBIG; |
3309dd04 | 2858 | iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos); |
1da177e4 LT |
2859 | } |
2860 | ||
2861 | /* | |
2862 | * Are we about to exceed the fs block limit ? | |
2863 | * | |
2864 | * If we have written data it becomes a short write. If we have | |
2865 | * exceeded without writing data we send a signal and return EFBIG. | |
2866 | * Linus frestrict idea will clean these up nicely.. | |
2867 | */ | |
3309dd04 AV |
2868 | if (unlikely(pos >= inode->i_sb->s_maxbytes)) |
2869 | return -EFBIG; | |
1da177e4 | 2870 | |
3309dd04 AV |
2871 | iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos); |
2872 | return iov_iter_count(from); | |
1da177e4 LT |
2873 | } |
2874 | EXPORT_SYMBOL(generic_write_checks); | |
2875 | ||
afddba49 NP |
2876 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
2877 | loff_t pos, unsigned len, unsigned flags, | |
2878 | struct page **pagep, void **fsdata) | |
2879 | { | |
2880 | const struct address_space_operations *aops = mapping->a_ops; | |
2881 | ||
4e02ed4b | 2882 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 2883 | pagep, fsdata); |
afddba49 NP |
2884 | } |
2885 | EXPORT_SYMBOL(pagecache_write_begin); | |
2886 | ||
2887 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
2888 | loff_t pos, unsigned len, unsigned copied, | |
2889 | struct page *page, void *fsdata) | |
2890 | { | |
2891 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 2892 | |
4e02ed4b | 2893 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
2894 | } |
2895 | EXPORT_SYMBOL(pagecache_write_end); | |
2896 | ||
1da177e4 | 2897 | ssize_t |
1af5bb49 | 2898 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
2899 | { |
2900 | struct file *file = iocb->ki_filp; | |
2901 | struct address_space *mapping = file->f_mapping; | |
2902 | struct inode *inode = mapping->host; | |
1af5bb49 | 2903 | loff_t pos = iocb->ki_pos; |
1da177e4 | 2904 | ssize_t written; |
a969e903 CH |
2905 | size_t write_len; |
2906 | pgoff_t end; | |
1da177e4 | 2907 | |
0c949334 | 2908 | write_len = iov_iter_count(from); |
09cbfeaf | 2909 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 2910 | |
6be96d3a GR |
2911 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2912 | /* If there are pages to writeback, return */ | |
2913 | if (filemap_range_has_page(inode->i_mapping, pos, | |
2914 | pos + iov_iter_count(from))) | |
2915 | return -EAGAIN; | |
2916 | } else { | |
2917 | written = filemap_write_and_wait_range(mapping, pos, | |
2918 | pos + write_len - 1); | |
2919 | if (written) | |
2920 | goto out; | |
2921 | } | |
a969e903 CH |
2922 | |
2923 | /* | |
2924 | * After a write we want buffered reads to be sure to go to disk to get | |
2925 | * the new data. We invalidate clean cached page from the region we're | |
2926 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 2927 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 2928 | */ |
55635ba7 | 2929 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 2930 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
2931 | /* |
2932 | * If a page can not be invalidated, return 0 to fall back | |
2933 | * to buffered write. | |
2934 | */ | |
2935 | if (written) { | |
2936 | if (written == -EBUSY) | |
2937 | return 0; | |
2938 | goto out; | |
a969e903 CH |
2939 | } |
2940 | ||
639a93a5 | 2941 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
2942 | |
2943 | /* | |
2944 | * Finally, try again to invalidate clean pages which might have been | |
2945 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
2946 | * if the source of the write was an mmap'ed region of the file | |
2947 | * we're writing. Either one is a pretty crazy thing to do, | |
2948 | * so we don't support it 100%. If this invalidation | |
2949 | * fails, tough, the write still worked... | |
332391a9 LC |
2950 | * |
2951 | * Most of the time we do not need this since dio_complete() will do | |
2952 | * the invalidation for us. However there are some file systems that | |
2953 | * do not end up with dio_complete() being called, so let's not break | |
2954 | * them by removing it completely | |
a969e903 | 2955 | */ |
332391a9 LC |
2956 | if (mapping->nrpages) |
2957 | invalidate_inode_pages2_range(mapping, | |
2958 | pos >> PAGE_SHIFT, end); | |
a969e903 | 2959 | |
1da177e4 | 2960 | if (written > 0) { |
0116651c | 2961 | pos += written; |
639a93a5 | 2962 | write_len -= written; |
0116651c NK |
2963 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
2964 | i_size_write(inode, pos); | |
1da177e4 LT |
2965 | mark_inode_dirty(inode); |
2966 | } | |
5cb6c6c7 | 2967 | iocb->ki_pos = pos; |
1da177e4 | 2968 | } |
639a93a5 | 2969 | iov_iter_revert(from, write_len - iov_iter_count(from)); |
a969e903 | 2970 | out: |
1da177e4 LT |
2971 | return written; |
2972 | } | |
2973 | EXPORT_SYMBOL(generic_file_direct_write); | |
2974 | ||
eb2be189 NP |
2975 | /* |
2976 | * Find or create a page at the given pagecache position. Return the locked | |
2977 | * page. This function is specifically for buffered writes. | |
2978 | */ | |
54566b2c NP |
2979 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
2980 | pgoff_t index, unsigned flags) | |
eb2be189 | 2981 | { |
eb2be189 | 2982 | struct page *page; |
bbddabe2 | 2983 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 2984 | |
54566b2c | 2985 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
2986 | fgp_flags |= FGP_NOFS; |
2987 | ||
2988 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 2989 | mapping_gfp_mask(mapping)); |
c585a267 | 2990 | if (page) |
2457aec6 | 2991 | wait_for_stable_page(page); |
eb2be189 | 2992 | |
eb2be189 NP |
2993 | return page; |
2994 | } | |
54566b2c | 2995 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 2996 | |
3b93f911 | 2997 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
2998 | struct iov_iter *i, loff_t pos) |
2999 | { | |
3000 | struct address_space *mapping = file->f_mapping; | |
3001 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3002 | long status = 0; | |
3003 | ssize_t written = 0; | |
674b892e NP |
3004 | unsigned int flags = 0; |
3005 | ||
afddba49 NP |
3006 | do { |
3007 | struct page *page; | |
afddba49 NP |
3008 | unsigned long offset; /* Offset into pagecache page */ |
3009 | unsigned long bytes; /* Bytes to write to page */ | |
3010 | size_t copied; /* Bytes copied from user */ | |
3011 | void *fsdata; | |
3012 | ||
09cbfeaf KS |
3013 | offset = (pos & (PAGE_SIZE - 1)); |
3014 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3015 | iov_iter_count(i)); |
3016 | ||
3017 | again: | |
00a3d660 LT |
3018 | /* |
3019 | * Bring in the user page that we will copy from _first_. | |
3020 | * Otherwise there's a nasty deadlock on copying from the | |
3021 | * same page as we're writing to, without it being marked | |
3022 | * up-to-date. | |
3023 | * | |
3024 | * Not only is this an optimisation, but it is also required | |
3025 | * to check that the address is actually valid, when atomic | |
3026 | * usercopies are used, below. | |
3027 | */ | |
3028 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
3029 | status = -EFAULT; | |
3030 | break; | |
3031 | } | |
3032 | ||
296291cd JK |
3033 | if (fatal_signal_pending(current)) { |
3034 | status = -EINTR; | |
3035 | break; | |
3036 | } | |
3037 | ||
674b892e | 3038 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3039 | &page, &fsdata); |
2457aec6 | 3040 | if (unlikely(status < 0)) |
afddba49 NP |
3041 | break; |
3042 | ||
931e80e4 | 3043 | if (mapping_writably_mapped(mapping)) |
3044 | flush_dcache_page(page); | |
00a3d660 | 3045 | |
afddba49 | 3046 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
3047 | flush_dcache_page(page); |
3048 | ||
3049 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3050 | page, fsdata); | |
3051 | if (unlikely(status < 0)) | |
3052 | break; | |
3053 | copied = status; | |
3054 | ||
3055 | cond_resched(); | |
3056 | ||
124d3b70 | 3057 | iov_iter_advance(i, copied); |
afddba49 NP |
3058 | if (unlikely(copied == 0)) { |
3059 | /* | |
3060 | * If we were unable to copy any data at all, we must | |
3061 | * fall back to a single segment length write. | |
3062 | * | |
3063 | * If we didn't fallback here, we could livelock | |
3064 | * because not all segments in the iov can be copied at | |
3065 | * once without a pagefault. | |
3066 | */ | |
09cbfeaf | 3067 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
afddba49 NP |
3068 | iov_iter_single_seg_count(i)); |
3069 | goto again; | |
3070 | } | |
afddba49 NP |
3071 | pos += copied; |
3072 | written += copied; | |
3073 | ||
3074 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3075 | } while (iov_iter_count(i)); |
3076 | ||
3077 | return written ? written : status; | |
3078 | } | |
3b93f911 | 3079 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3080 | |
e4dd9de3 | 3081 | /** |
8174202b | 3082 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3083 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3084 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3085 | * |
3086 | * This function does all the work needed for actually writing data to a | |
3087 | * file. It does all basic checks, removes SUID from the file, updates | |
3088 | * modification times and calls proper subroutines depending on whether we | |
3089 | * do direct IO or a standard buffered write. | |
3090 | * | |
3091 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
3092 | * object which does not need locking at all. | |
3093 | * | |
3094 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3095 | * A caller has to handle it. This is mainly due to the fact that we want to | |
3096 | * avoid syncing under i_mutex. | |
3097 | */ | |
8174202b | 3098 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3099 | { |
3100 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 3101 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 3102 | struct inode *inode = mapping->host; |
3b93f911 | 3103 | ssize_t written = 0; |
1da177e4 | 3104 | ssize_t err; |
3b93f911 | 3105 | ssize_t status; |
1da177e4 | 3106 | |
1da177e4 | 3107 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3108 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3109 | err = file_remove_privs(file); |
1da177e4 LT |
3110 | if (err) |
3111 | goto out; | |
3112 | ||
c3b2da31 JB |
3113 | err = file_update_time(file); |
3114 | if (err) | |
3115 | goto out; | |
1da177e4 | 3116 | |
2ba48ce5 | 3117 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3118 | loff_t pos, endbyte; |
fb5527e6 | 3119 | |
1af5bb49 | 3120 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3121 | /* |
fbbbad4b MW |
3122 | * If the write stopped short of completing, fall back to |
3123 | * buffered writes. Some filesystems do this for writes to | |
3124 | * holes, for example. For DAX files, a buffered write will | |
3125 | * not succeed (even if it did, DAX does not handle dirty | |
3126 | * page-cache pages correctly). | |
1da177e4 | 3127 | */ |
0b8def9d | 3128 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3129 | goto out; |
3130 | ||
0b8def9d | 3131 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3132 | /* |
3b93f911 | 3133 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3134 | * then we want to return the number of bytes which were |
3135 | * direct-written, or the error code if that was zero. Note | |
3136 | * that this differs from normal direct-io semantics, which | |
3137 | * will return -EFOO even if some bytes were written. | |
3138 | */ | |
60bb4529 | 3139 | if (unlikely(status < 0)) { |
3b93f911 | 3140 | err = status; |
fb5527e6 JM |
3141 | goto out; |
3142 | } | |
fb5527e6 JM |
3143 | /* |
3144 | * We need to ensure that the page cache pages are written to | |
3145 | * disk and invalidated to preserve the expected O_DIRECT | |
3146 | * semantics. | |
3147 | */ | |
3b93f911 | 3148 | endbyte = pos + status - 1; |
0b8def9d | 3149 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3150 | if (err == 0) { |
0b8def9d | 3151 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3152 | written += status; |
fb5527e6 | 3153 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3154 | pos >> PAGE_SHIFT, |
3155 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3156 | } else { |
3157 | /* | |
3158 | * We don't know how much we wrote, so just return | |
3159 | * the number of bytes which were direct-written | |
3160 | */ | |
3161 | } | |
3162 | } else { | |
0b8def9d AV |
3163 | written = generic_perform_write(file, from, iocb->ki_pos); |
3164 | if (likely(written > 0)) | |
3165 | iocb->ki_pos += written; | |
fb5527e6 | 3166 | } |
1da177e4 LT |
3167 | out: |
3168 | current->backing_dev_info = NULL; | |
3169 | return written ? written : err; | |
3170 | } | |
8174202b | 3171 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3172 | |
e4dd9de3 | 3173 | /** |
8174202b | 3174 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3175 | * @iocb: IO state structure |
8174202b | 3176 | * @from: iov_iter with data to write |
e4dd9de3 | 3177 | * |
8174202b | 3178 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
3179 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
3180 | * and acquires i_mutex as needed. | |
3181 | */ | |
8174202b | 3182 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3183 | { |
3184 | struct file *file = iocb->ki_filp; | |
148f948b | 3185 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3186 | ssize_t ret; |
1da177e4 | 3187 | |
5955102c | 3188 | inode_lock(inode); |
3309dd04 AV |
3189 | ret = generic_write_checks(iocb, from); |
3190 | if (ret > 0) | |
5f380c7f | 3191 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3192 | inode_unlock(inode); |
1da177e4 | 3193 | |
e2592217 CH |
3194 | if (ret > 0) |
3195 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3196 | return ret; |
3197 | } | |
8174202b | 3198 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3199 | |
cf9a2ae8 DH |
3200 | /** |
3201 | * try_to_release_page() - release old fs-specific metadata on a page | |
3202 | * | |
3203 | * @page: the page which the kernel is trying to free | |
3204 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3205 | * | |
3206 | * The address_space is to try to release any data against the page | |
0e056eb5 | 3207 | * (presumably at page->private). If the release was successful, return '1'. |
cf9a2ae8 DH |
3208 | * Otherwise return zero. |
3209 | * | |
266cf658 DH |
3210 | * This may also be called if PG_fscache is set on a page, indicating that the |
3211 | * page is known to the local caching routines. | |
3212 | * | |
cf9a2ae8 | 3213 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3214 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3215 | * |
cf9a2ae8 DH |
3216 | */ |
3217 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3218 | { | |
3219 | struct address_space * const mapping = page->mapping; | |
3220 | ||
3221 | BUG_ON(!PageLocked(page)); | |
3222 | if (PageWriteback(page)) | |
3223 | return 0; | |
3224 | ||
3225 | if (mapping && mapping->a_ops->releasepage) | |
3226 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3227 | return try_to_free_buffers(page); | |
3228 | } | |
3229 | ||
3230 | EXPORT_SYMBOL(try_to_release_page); |