]>
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> |
00501b53 | 34 | #include <linux/hugetlb.h> |
8a9f3ccd | 35 | #include <linux/memcontrol.h> |
c515e1fd | 36 | #include <linux/cleancache.h> |
c7df8ad2 | 37 | #include <linux/shmem_fs.h> |
f1820361 | 38 | #include <linux/rmap.h> |
b1d29ba8 | 39 | #include <linux/delayacct.h> |
eb414681 | 40 | #include <linux/psi.h> |
0f8053a5 NP |
41 | #include "internal.h" |
42 | ||
fe0bfaaf RJ |
43 | #define CREATE_TRACE_POINTS |
44 | #include <trace/events/filemap.h> | |
45 | ||
1da177e4 | 46 | /* |
1da177e4 LT |
47 | * FIXME: remove all knowledge of the buffer layer from the core VM |
48 | */ | |
148f948b | 49 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 50 | |
1da177e4 LT |
51 | #include <asm/mman.h> |
52 | ||
53 | /* | |
54 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
55 | * though. | |
56 | * | |
57 | * Shared mappings now work. 15.8.1995 Bruno. | |
58 | * | |
59 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
60 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
61 | * | |
62 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
63 | */ | |
64 | ||
65 | /* | |
66 | * Lock ordering: | |
67 | * | |
c8c06efa | 68 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 69 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 | 70 | * ->swap_lock (exclusive_swap_page, others) |
b93b0163 | 71 | * ->i_pages lock |
1da177e4 | 72 | * |
1b1dcc1b | 73 | * ->i_mutex |
c8c06efa | 74 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
1da177e4 LT |
75 | * |
76 | * ->mmap_sem | |
c8c06efa | 77 | * ->i_mmap_rwsem |
b8072f09 | 78 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
b93b0163 | 79 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
1da177e4 LT |
80 | * |
81 | * ->mmap_sem | |
82 | * ->lock_page (access_process_vm) | |
83 | * | |
ccad2365 | 84 | * ->i_mutex (generic_perform_write) |
82591e6e | 85 | * ->mmap_sem (fault_in_pages_readable->do_page_fault) |
1da177e4 | 86 | * |
f758eeab | 87 | * bdi->wb.list_lock |
a66979ab | 88 | * sb_lock (fs/fs-writeback.c) |
b93b0163 | 89 | * ->i_pages lock (__sync_single_inode) |
1da177e4 | 90 | * |
c8c06efa | 91 | * ->i_mmap_rwsem |
1da177e4 LT |
92 | * ->anon_vma.lock (vma_adjust) |
93 | * | |
94 | * ->anon_vma.lock | |
b8072f09 | 95 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 96 | * |
b8072f09 | 97 | * ->page_table_lock or pte_lock |
5d337b91 | 98 | * ->swap_lock (try_to_unmap_one) |
1da177e4 | 99 | * ->private_lock (try_to_unmap_one) |
b93b0163 | 100 | * ->i_pages lock (try_to_unmap_one) |
a52633d8 MG |
101 | * ->zone_lru_lock(zone) (follow_page->mark_page_accessed) |
102 | * ->zone_lru_lock(zone) (check_pte_range->isolate_lru_page) | |
1da177e4 | 103 | * ->private_lock (page_remove_rmap->set_page_dirty) |
b93b0163 | 104 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
f758eeab | 105 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 106 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 107 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 108 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 109 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
110 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
111 | * | |
c8c06efa | 112 | * ->i_mmap_rwsem |
9a3c531d | 113 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
114 | */ |
115 | ||
5c024e6a | 116 | static void page_cache_delete(struct address_space *mapping, |
91b0abe3 JW |
117 | struct page *page, void *shadow) |
118 | { | |
5c024e6a MW |
119 | XA_STATE(xas, &mapping->i_pages, page->index); |
120 | unsigned int nr = 1; | |
c70b647d | 121 | |
5c024e6a | 122 | mapping_set_update(&xas, mapping); |
c70b647d | 123 | |
5c024e6a MW |
124 | /* hugetlb pages are represented by a single entry in the xarray */ |
125 | if (!PageHuge(page)) { | |
126 | xas_set_order(&xas, page->index, compound_order(page)); | |
127 | nr = 1U << compound_order(page); | |
128 | } | |
91b0abe3 | 129 | |
83929372 KS |
130 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
131 | VM_BUG_ON_PAGE(PageTail(page), page); | |
132 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 133 | |
5c024e6a MW |
134 | xas_store(&xas, shadow); |
135 | xas_init_marks(&xas); | |
d3798ae8 | 136 | |
2300638b JK |
137 | page->mapping = NULL; |
138 | /* Leave page->index set: truncation lookup relies upon it */ | |
139 | ||
d3798ae8 JW |
140 | if (shadow) { |
141 | mapping->nrexceptional += nr; | |
142 | /* | |
143 | * Make sure the nrexceptional update is committed before | |
144 | * the nrpages update so that final truncate racing | |
145 | * with reclaim does not see both counters 0 at the | |
146 | * same time and miss a shadow entry. | |
147 | */ | |
148 | smp_wmb(); | |
149 | } | |
150 | mapping->nrpages -= nr; | |
91b0abe3 JW |
151 | } |
152 | ||
5ecc4d85 JK |
153 | static void unaccount_page_cache_page(struct address_space *mapping, |
154 | struct page *page) | |
1da177e4 | 155 | { |
5ecc4d85 | 156 | int nr; |
1da177e4 | 157 | |
c515e1fd DM |
158 | /* |
159 | * if we're uptodate, flush out into the cleancache, otherwise | |
160 | * invalidate any existing cleancache entries. We can't leave | |
161 | * stale data around in the cleancache once our page is gone | |
162 | */ | |
163 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
164 | cleancache_put_page(page); | |
165 | else | |
3167760f | 166 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 167 | |
83929372 | 168 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
169 | VM_BUG_ON_PAGE(page_mapped(page), page); |
170 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
171 | int mapcount; | |
172 | ||
173 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
174 | current->comm, page_to_pfn(page)); | |
175 | dump_page(page, "still mapped when deleted"); | |
176 | dump_stack(); | |
177 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
178 | ||
179 | mapcount = page_mapcount(page); | |
180 | if (mapping_exiting(mapping) && | |
181 | page_count(page) >= mapcount + 2) { | |
182 | /* | |
183 | * All vmas have already been torn down, so it's | |
184 | * a good bet that actually the page is unmapped, | |
185 | * and we'd prefer not to leak it: if we're wrong, | |
186 | * some other bad page check should catch it later. | |
187 | */ | |
188 | page_mapcount_reset(page); | |
6d061f9f | 189 | page_ref_sub(page, mapcount); |
06b241f3 HD |
190 | } |
191 | } | |
192 | ||
4165b9b4 | 193 | /* hugetlb pages do not participate in page cache accounting. */ |
5ecc4d85 JK |
194 | if (PageHuge(page)) |
195 | return; | |
09612fa6 | 196 | |
5ecc4d85 JK |
197 | nr = hpage_nr_pages(page); |
198 | ||
199 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr); | |
200 | if (PageSwapBacked(page)) { | |
201 | __mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr); | |
202 | if (PageTransHuge(page)) | |
203 | __dec_node_page_state(page, NR_SHMEM_THPS); | |
204 | } else { | |
205 | VM_BUG_ON_PAGE(PageTransHuge(page), page); | |
800d8c63 | 206 | } |
5ecc4d85 JK |
207 | |
208 | /* | |
209 | * At this point page must be either written or cleaned by | |
210 | * truncate. Dirty page here signals a bug and loss of | |
211 | * unwritten data. | |
212 | * | |
213 | * This fixes dirty accounting after removing the page entirely | |
214 | * but leaves PageDirty set: it has no effect for truncated | |
215 | * page and anyway will be cleared before returning page into | |
216 | * buddy allocator. | |
217 | */ | |
218 | if (WARN_ON_ONCE(PageDirty(page))) | |
219 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); | |
220 | } | |
221 | ||
222 | /* | |
223 | * Delete a page from the page cache and free it. Caller has to make | |
224 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 225 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 JK |
226 | */ |
227 | void __delete_from_page_cache(struct page *page, void *shadow) | |
228 | { | |
229 | struct address_space *mapping = page->mapping; | |
230 | ||
231 | trace_mm_filemap_delete_from_page_cache(page); | |
232 | ||
233 | unaccount_page_cache_page(mapping, page); | |
5c024e6a | 234 | page_cache_delete(mapping, page, shadow); |
1da177e4 LT |
235 | } |
236 | ||
59c66c5f JK |
237 | static void page_cache_free_page(struct address_space *mapping, |
238 | struct page *page) | |
239 | { | |
240 | void (*freepage)(struct page *); | |
241 | ||
242 | freepage = mapping->a_ops->freepage; | |
243 | if (freepage) | |
244 | freepage(page); | |
245 | ||
246 | if (PageTransHuge(page) && !PageHuge(page)) { | |
247 | page_ref_sub(page, HPAGE_PMD_NR); | |
248 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); | |
249 | } else { | |
250 | put_page(page); | |
251 | } | |
252 | } | |
253 | ||
702cfbf9 MK |
254 | /** |
255 | * delete_from_page_cache - delete page from page cache | |
256 | * @page: the page which the kernel is trying to remove from page cache | |
257 | * | |
258 | * This must be called only on pages that have been verified to be in the page | |
259 | * cache and locked. It will never put the page into the free list, the caller | |
260 | * has a reference on the page. | |
261 | */ | |
262 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 263 | { |
83929372 | 264 | struct address_space *mapping = page_mapping(page); |
c4843a75 | 265 | unsigned long flags; |
1da177e4 | 266 | |
cd7619d6 | 267 | BUG_ON(!PageLocked(page)); |
b93b0163 | 268 | xa_lock_irqsave(&mapping->i_pages, flags); |
62cccb8c | 269 | __delete_from_page_cache(page, NULL); |
b93b0163 | 270 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
6072d13c | 271 | |
59c66c5f | 272 | page_cache_free_page(mapping, page); |
97cecb5a MK |
273 | } |
274 | EXPORT_SYMBOL(delete_from_page_cache); | |
275 | ||
aa65c29c | 276 | /* |
ef8e5717 | 277 | * page_cache_delete_batch - delete several pages from page cache |
aa65c29c JK |
278 | * @mapping: the mapping to which pages belong |
279 | * @pvec: pagevec with pages to delete | |
280 | * | |
b93b0163 MW |
281 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
282 | * from the mapping. The function expects @pvec to be sorted by page index. | |
283 | * It tolerates holes in @pvec (mapping entries at those indices are not | |
aa65c29c | 284 | * modified). The function expects only THP head pages to be present in the |
b93b0163 MW |
285 | * @pvec and takes care to delete all corresponding tail pages from the |
286 | * mapping as well. | |
aa65c29c | 287 | * |
b93b0163 | 288 | * The function expects the i_pages lock to be held. |
aa65c29c | 289 | */ |
ef8e5717 | 290 | static void page_cache_delete_batch(struct address_space *mapping, |
aa65c29c JK |
291 | struct pagevec *pvec) |
292 | { | |
ef8e5717 | 293 | XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index); |
aa65c29c JK |
294 | int total_pages = 0; |
295 | int i = 0, tail_pages = 0; | |
296 | struct page *page; | |
aa65c29c | 297 | |
ef8e5717 MW |
298 | mapping_set_update(&xas, mapping); |
299 | xas_for_each(&xas, page, ULONG_MAX) { | |
aa65c29c JK |
300 | if (i >= pagevec_count(pvec) && !tail_pages) |
301 | break; | |
3159f943 | 302 | if (xa_is_value(page)) |
aa65c29c JK |
303 | continue; |
304 | if (!tail_pages) { | |
305 | /* | |
306 | * Some page got inserted in our range? Skip it. We | |
307 | * have our pages locked so they are protected from | |
308 | * being removed. | |
309 | */ | |
ef8e5717 MW |
310 | if (page != pvec->pages[i]) { |
311 | VM_BUG_ON_PAGE(page->index > | |
312 | pvec->pages[i]->index, page); | |
aa65c29c | 313 | continue; |
ef8e5717 | 314 | } |
aa65c29c JK |
315 | WARN_ON_ONCE(!PageLocked(page)); |
316 | if (PageTransHuge(page) && !PageHuge(page)) | |
317 | tail_pages = HPAGE_PMD_NR - 1; | |
318 | page->mapping = NULL; | |
319 | /* | |
320 | * Leave page->index set: truncation lookup relies | |
321 | * upon it | |
322 | */ | |
323 | i++; | |
324 | } else { | |
ef8e5717 MW |
325 | VM_BUG_ON_PAGE(page->index + HPAGE_PMD_NR - tail_pages |
326 | != pvec->pages[i]->index, page); | |
aa65c29c JK |
327 | tail_pages--; |
328 | } | |
ef8e5717 | 329 | xas_store(&xas, NULL); |
aa65c29c JK |
330 | total_pages++; |
331 | } | |
332 | mapping->nrpages -= total_pages; | |
333 | } | |
334 | ||
335 | void delete_from_page_cache_batch(struct address_space *mapping, | |
336 | struct pagevec *pvec) | |
337 | { | |
338 | int i; | |
339 | unsigned long flags; | |
340 | ||
341 | if (!pagevec_count(pvec)) | |
342 | return; | |
343 | ||
b93b0163 | 344 | xa_lock_irqsave(&mapping->i_pages, flags); |
aa65c29c JK |
345 | for (i = 0; i < pagevec_count(pvec); i++) { |
346 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); | |
347 | ||
348 | unaccount_page_cache_page(mapping, pvec->pages[i]); | |
349 | } | |
ef8e5717 | 350 | page_cache_delete_batch(mapping, pvec); |
b93b0163 | 351 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
aa65c29c JK |
352 | |
353 | for (i = 0; i < pagevec_count(pvec); i++) | |
354 | page_cache_free_page(mapping, pvec->pages[i]); | |
355 | } | |
356 | ||
d72d9e2a | 357 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
358 | { |
359 | int ret = 0; | |
360 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
361 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
362 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 363 | ret = -ENOSPC; |
7fcbbaf1 JA |
364 | if (test_bit(AS_EIO, &mapping->flags) && |
365 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
366 | ret = -EIO; |
367 | return ret; | |
368 | } | |
d72d9e2a | 369 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 370 | |
76341cab JL |
371 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
372 | { | |
373 | /* Check for outstanding write errors */ | |
374 | if (test_bit(AS_EIO, &mapping->flags)) | |
375 | return -EIO; | |
376 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
377 | return -ENOSPC; | |
378 | return 0; | |
379 | } | |
380 | ||
1da177e4 | 381 | /** |
485bb99b | 382 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
383 | * @mapping: address space structure to write |
384 | * @start: offset in bytes where the range starts | |
469eb4d0 | 385 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 386 | * @sync_mode: enable synchronous operation |
1da177e4 | 387 | * |
485bb99b RD |
388 | * Start writeback against all of a mapping's dirty pages that lie |
389 | * within the byte offsets <start, end> inclusive. | |
390 | * | |
1da177e4 | 391 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 392 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
393 | * these two operations is that if a dirty page/buffer is encountered, it must |
394 | * be waited upon, and not just skipped over. | |
a862f68a MR |
395 | * |
396 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 397 | */ |
ebcf28e1 AM |
398 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
399 | loff_t end, int sync_mode) | |
1da177e4 LT |
400 | { |
401 | int ret; | |
402 | struct writeback_control wbc = { | |
403 | .sync_mode = sync_mode, | |
05fe478d | 404 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
405 | .range_start = start, |
406 | .range_end = end, | |
1da177e4 LT |
407 | }; |
408 | ||
409 | if (!mapping_cap_writeback_dirty(mapping)) | |
410 | return 0; | |
411 | ||
b16b1deb | 412 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 413 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 414 | wbc_detach_inode(&wbc); |
1da177e4 LT |
415 | return ret; |
416 | } | |
417 | ||
418 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
419 | int sync_mode) | |
420 | { | |
111ebb6e | 421 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
422 | } |
423 | ||
424 | int filemap_fdatawrite(struct address_space *mapping) | |
425 | { | |
426 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
427 | } | |
428 | EXPORT_SYMBOL(filemap_fdatawrite); | |
429 | ||
f4c0a0fd | 430 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 431 | loff_t end) |
1da177e4 LT |
432 | { |
433 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
434 | } | |
f4c0a0fd | 435 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 436 | |
485bb99b RD |
437 | /** |
438 | * filemap_flush - mostly a non-blocking flush | |
439 | * @mapping: target address_space | |
440 | * | |
1da177e4 LT |
441 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
442 | * purposes - I/O may not be started against all dirty pages. | |
a862f68a MR |
443 | * |
444 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 LT |
445 | */ |
446 | int filemap_flush(struct address_space *mapping) | |
447 | { | |
448 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
449 | } | |
450 | EXPORT_SYMBOL(filemap_flush); | |
451 | ||
7fc9e472 GR |
452 | /** |
453 | * filemap_range_has_page - check if a page exists in range. | |
454 | * @mapping: address space within which to check | |
455 | * @start_byte: offset in bytes where the range starts | |
456 | * @end_byte: offset in bytes where the range ends (inclusive) | |
457 | * | |
458 | * Find at least one page in the range supplied, usually used to check if | |
459 | * direct writing in this range will trigger a writeback. | |
a862f68a MR |
460 | * |
461 | * Return: %true if at least one page exists in the specified range, | |
462 | * %false otherwise. | |
7fc9e472 GR |
463 | */ |
464 | bool filemap_range_has_page(struct address_space *mapping, | |
465 | loff_t start_byte, loff_t end_byte) | |
466 | { | |
f7b68046 | 467 | struct page *page; |
8fa8e538 MW |
468 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
469 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
7fc9e472 GR |
470 | |
471 | if (end_byte < start_byte) | |
472 | return false; | |
473 | ||
8fa8e538 MW |
474 | rcu_read_lock(); |
475 | for (;;) { | |
476 | page = xas_find(&xas, max); | |
477 | if (xas_retry(&xas, page)) | |
478 | continue; | |
479 | /* Shadow entries don't count */ | |
480 | if (xa_is_value(page)) | |
481 | continue; | |
482 | /* | |
483 | * We don't need to try to pin this page; we're about to | |
484 | * release the RCU lock anyway. It is enough to know that | |
485 | * there was a page here recently. | |
486 | */ | |
487 | break; | |
488 | } | |
489 | rcu_read_unlock(); | |
7fc9e472 | 490 | |
8fa8e538 | 491 | return page != NULL; |
7fc9e472 GR |
492 | } |
493 | EXPORT_SYMBOL(filemap_range_has_page); | |
494 | ||
5e8fcc1a | 495 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 496 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 497 | { |
09cbfeaf KS |
498 | pgoff_t index = start_byte >> PAGE_SHIFT; |
499 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
500 | struct pagevec pvec; |
501 | int nr_pages; | |
1da177e4 | 502 | |
94004ed7 | 503 | if (end_byte < start_byte) |
5e8fcc1a | 504 | return; |
1da177e4 | 505 | |
86679820 | 506 | pagevec_init(&pvec); |
312e9d2f | 507 | while (index <= end) { |
1da177e4 LT |
508 | unsigned i; |
509 | ||
312e9d2f | 510 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 511 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
512 | if (!nr_pages) |
513 | break; | |
514 | ||
1da177e4 LT |
515 | for (i = 0; i < nr_pages; i++) { |
516 | struct page *page = pvec.pages[i]; | |
517 | ||
1da177e4 | 518 | wait_on_page_writeback(page); |
5e8fcc1a | 519 | ClearPageError(page); |
1da177e4 LT |
520 | } |
521 | pagevec_release(&pvec); | |
522 | cond_resched(); | |
523 | } | |
aa750fd7 JN |
524 | } |
525 | ||
526 | /** | |
527 | * filemap_fdatawait_range - wait for writeback to complete | |
528 | * @mapping: address space structure to wait for | |
529 | * @start_byte: offset in bytes where the range starts | |
530 | * @end_byte: offset in bytes where the range ends (inclusive) | |
531 | * | |
532 | * Walk the list of under-writeback pages of the given address space | |
533 | * in the given range and wait for all of them. Check error status of | |
534 | * the address space and return it. | |
535 | * | |
536 | * Since the error status of the address space is cleared by this function, | |
537 | * callers are responsible for checking the return value and handling and/or | |
538 | * reporting the error. | |
a862f68a MR |
539 | * |
540 | * Return: error status of the address space. | |
aa750fd7 JN |
541 | */ |
542 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
543 | loff_t end_byte) | |
544 | { | |
5e8fcc1a JL |
545 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
546 | return filemap_check_errors(mapping); | |
1da177e4 | 547 | } |
d3bccb6f JK |
548 | EXPORT_SYMBOL(filemap_fdatawait_range); |
549 | ||
a823e458 JL |
550 | /** |
551 | * file_fdatawait_range - wait for writeback to complete | |
552 | * @file: file pointing to address space structure to wait for | |
553 | * @start_byte: offset in bytes where the range starts | |
554 | * @end_byte: offset in bytes where the range ends (inclusive) | |
555 | * | |
556 | * Walk the list of under-writeback pages of the address space that file | |
557 | * refers to, in the given range and wait for all of them. Check error | |
558 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
559 | * | |
560 | * Since the error status of the file is advanced by this function, | |
561 | * callers are responsible for checking the return value and handling and/or | |
562 | * reporting the error. | |
a862f68a MR |
563 | * |
564 | * Return: error status of the address space vs. the file->f_wb_err cursor. | |
a823e458 JL |
565 | */ |
566 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
567 | { | |
568 | struct address_space *mapping = file->f_mapping; | |
569 | ||
570 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
571 | return file_check_and_advance_wb_err(file); | |
572 | } | |
573 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 574 | |
aa750fd7 JN |
575 | /** |
576 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
577 | * @mapping: address space structure to wait for | |
578 | * | |
579 | * Walk the list of under-writeback pages of the given address space | |
580 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
581 | * does not clear error status of the address space. | |
582 | * | |
583 | * Use this function if callers don't handle errors themselves. Expected | |
584 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
585 | * fsfreeze(8) | |
a862f68a MR |
586 | * |
587 | * Return: error status of the address space. | |
aa750fd7 | 588 | */ |
76341cab | 589 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 590 | { |
ffb959bb | 591 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 592 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 593 | } |
76341cab | 594 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 595 | |
9326c9b2 | 596 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 597 | { |
9326c9b2 JL |
598 | return (!dax_mapping(mapping) && mapping->nrpages) || |
599 | (dax_mapping(mapping) && mapping->nrexceptional); | |
1da177e4 | 600 | } |
1da177e4 LT |
601 | |
602 | int filemap_write_and_wait(struct address_space *mapping) | |
603 | { | |
28fd1298 | 604 | int err = 0; |
1da177e4 | 605 | |
9326c9b2 | 606 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
607 | err = filemap_fdatawrite(mapping); |
608 | /* | |
609 | * Even if the above returned error, the pages may be | |
610 | * written partially (e.g. -ENOSPC), so we wait for it. | |
611 | * But the -EIO is special case, it may indicate the worst | |
612 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
613 | */ | |
614 | if (err != -EIO) { | |
615 | int err2 = filemap_fdatawait(mapping); | |
616 | if (!err) | |
617 | err = err2; | |
cbeaf951 JL |
618 | } else { |
619 | /* Clear any previously stored errors */ | |
620 | filemap_check_errors(mapping); | |
28fd1298 | 621 | } |
865ffef3 DM |
622 | } else { |
623 | err = filemap_check_errors(mapping); | |
1da177e4 | 624 | } |
28fd1298 | 625 | return err; |
1da177e4 | 626 | } |
28fd1298 | 627 | EXPORT_SYMBOL(filemap_write_and_wait); |
1da177e4 | 628 | |
485bb99b RD |
629 | /** |
630 | * filemap_write_and_wait_range - write out & wait on a file range | |
631 | * @mapping: the address_space for the pages | |
632 | * @lstart: offset in bytes where the range starts | |
633 | * @lend: offset in bytes where the range ends (inclusive) | |
634 | * | |
469eb4d0 AM |
635 | * Write out and wait upon file offsets lstart->lend, inclusive. |
636 | * | |
0e056eb5 | 637 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 | 638 | * that this function can be used to write to the very end-of-file (end = -1). |
a862f68a MR |
639 | * |
640 | * Return: error status of the address space. | |
469eb4d0 | 641 | */ |
1da177e4 LT |
642 | int filemap_write_and_wait_range(struct address_space *mapping, |
643 | loff_t lstart, loff_t lend) | |
644 | { | |
28fd1298 | 645 | int err = 0; |
1da177e4 | 646 | |
9326c9b2 | 647 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
648 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
649 | WB_SYNC_ALL); | |
650 | /* See comment of filemap_write_and_wait() */ | |
651 | if (err != -EIO) { | |
94004ed7 CH |
652 | int err2 = filemap_fdatawait_range(mapping, |
653 | lstart, lend); | |
28fd1298 OH |
654 | if (!err) |
655 | err = err2; | |
cbeaf951 JL |
656 | } else { |
657 | /* Clear any previously stored errors */ | |
658 | filemap_check_errors(mapping); | |
28fd1298 | 659 | } |
865ffef3 DM |
660 | } else { |
661 | err = filemap_check_errors(mapping); | |
1da177e4 | 662 | } |
28fd1298 | 663 | return err; |
1da177e4 | 664 | } |
f6995585 | 665 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 666 | |
5660e13d JL |
667 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
668 | { | |
3acdfd28 | 669 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
670 | |
671 | trace_filemap_set_wb_err(mapping, eseq); | |
672 | } | |
673 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
674 | ||
675 | /** | |
676 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
677 | * and advance wb_err to current one | |
678 | * @file: struct file on which the error is being reported | |
679 | * | |
680 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
681 | * want to report any writeback errors that occurred since the last fsync (or | |
682 | * since the file was opened if there haven't been any). | |
683 | * | |
684 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
685 | * then just quickly return 0. The file is all caught up. | |
686 | * | |
687 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
688 | * it and try to swap it into place. If it works, or another task beat us | |
689 | * to it with the new value, then update the f_wb_err and return the error | |
690 | * portion. The error at this point must be reported via proper channels | |
691 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
692 | * | |
693 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
694 | * value is protected by the f_lock since we must ensure that it reflects | |
695 | * the latest value swapped in for this file descriptor. | |
a862f68a MR |
696 | * |
697 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
698 | */ |
699 | int file_check_and_advance_wb_err(struct file *file) | |
700 | { | |
701 | int err = 0; | |
702 | errseq_t old = READ_ONCE(file->f_wb_err); | |
703 | struct address_space *mapping = file->f_mapping; | |
704 | ||
705 | /* Locklessly handle the common case where nothing has changed */ | |
706 | if (errseq_check(&mapping->wb_err, old)) { | |
707 | /* Something changed, must use slow path */ | |
708 | spin_lock(&file->f_lock); | |
709 | old = file->f_wb_err; | |
710 | err = errseq_check_and_advance(&mapping->wb_err, | |
711 | &file->f_wb_err); | |
712 | trace_file_check_and_advance_wb_err(file, old); | |
713 | spin_unlock(&file->f_lock); | |
714 | } | |
f4e222c5 JL |
715 | |
716 | /* | |
717 | * We're mostly using this function as a drop in replacement for | |
718 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
719 | * that the legacy code would have had on these flags. | |
720 | */ | |
721 | clear_bit(AS_EIO, &mapping->flags); | |
722 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
723 | return err; |
724 | } | |
725 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
726 | ||
727 | /** | |
728 | * file_write_and_wait_range - write out & wait on a file range | |
729 | * @file: file pointing to address_space with pages | |
730 | * @lstart: offset in bytes where the range starts | |
731 | * @lend: offset in bytes where the range ends (inclusive) | |
732 | * | |
733 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
734 | * | |
735 | * Note that @lend is inclusive (describes the last byte to be written) so | |
736 | * that this function can be used to write to the very end-of-file (end = -1). | |
737 | * | |
738 | * After writing out and waiting on the data, we check and advance the | |
739 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
a862f68a MR |
740 | * |
741 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
742 | */ |
743 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
744 | { | |
745 | int err = 0, err2; | |
746 | struct address_space *mapping = file->f_mapping; | |
747 | ||
9326c9b2 | 748 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
749 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
750 | WB_SYNC_ALL); | |
751 | /* See comment of filemap_write_and_wait() */ | |
752 | if (err != -EIO) | |
753 | __filemap_fdatawait_range(mapping, lstart, lend); | |
754 | } | |
755 | err2 = file_check_and_advance_wb_err(file); | |
756 | if (!err) | |
757 | err = err2; | |
758 | return err; | |
759 | } | |
760 | EXPORT_SYMBOL(file_write_and_wait_range); | |
761 | ||
ef6a3c63 MS |
762 | /** |
763 | * replace_page_cache_page - replace a pagecache page with a new one | |
764 | * @old: page to be replaced | |
765 | * @new: page to replace with | |
766 | * @gfp_mask: allocation mode | |
767 | * | |
768 | * This function replaces a page in the pagecache with a new one. On | |
769 | * success it acquires the pagecache reference for the new page and | |
770 | * drops it for the old page. Both the old and new pages must be | |
771 | * locked. This function does not add the new page to the LRU, the | |
772 | * caller must do that. | |
773 | * | |
74d60958 | 774 | * The remove + add is atomic. This function cannot fail. |
a862f68a MR |
775 | * |
776 | * Return: %0 | |
ef6a3c63 MS |
777 | */ |
778 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
779 | { | |
74d60958 MW |
780 | struct address_space *mapping = old->mapping; |
781 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
782 | pgoff_t offset = old->index; | |
783 | XA_STATE(xas, &mapping->i_pages, offset); | |
784 | unsigned long flags; | |
ef6a3c63 | 785 | |
309381fe SL |
786 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
787 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
788 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 789 | |
74d60958 MW |
790 | get_page(new); |
791 | new->mapping = mapping; | |
792 | new->index = offset; | |
ef6a3c63 | 793 | |
74d60958 MW |
794 | xas_lock_irqsave(&xas, flags); |
795 | xas_store(&xas, new); | |
4165b9b4 | 796 | |
74d60958 MW |
797 | old->mapping = NULL; |
798 | /* hugetlb pages do not participate in page cache accounting. */ | |
799 | if (!PageHuge(old)) | |
800 | __dec_node_page_state(new, NR_FILE_PAGES); | |
801 | if (!PageHuge(new)) | |
802 | __inc_node_page_state(new, NR_FILE_PAGES); | |
803 | if (PageSwapBacked(old)) | |
804 | __dec_node_page_state(new, NR_SHMEM); | |
805 | if (PageSwapBacked(new)) | |
806 | __inc_node_page_state(new, NR_SHMEM); | |
807 | xas_unlock_irqrestore(&xas, flags); | |
808 | mem_cgroup_migrate(old, new); | |
809 | if (freepage) | |
810 | freepage(old); | |
811 | put_page(old); | |
ef6a3c63 | 812 | |
74d60958 | 813 | return 0; |
ef6a3c63 MS |
814 | } |
815 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
816 | ||
a528910e JW |
817 | static int __add_to_page_cache_locked(struct page *page, |
818 | struct address_space *mapping, | |
819 | pgoff_t offset, gfp_t gfp_mask, | |
820 | void **shadowp) | |
1da177e4 | 821 | { |
74d60958 | 822 | XA_STATE(xas, &mapping->i_pages, offset); |
00501b53 JW |
823 | int huge = PageHuge(page); |
824 | struct mem_cgroup *memcg; | |
e286781d | 825 | int error; |
74d60958 | 826 | void *old; |
e286781d | 827 | |
309381fe SL |
828 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
829 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
74d60958 | 830 | mapping_set_update(&xas, mapping); |
e286781d | 831 | |
00501b53 JW |
832 | if (!huge) { |
833 | error = mem_cgroup_try_charge(page, current->mm, | |
f627c2f5 | 834 | gfp_mask, &memcg, false); |
00501b53 JW |
835 | if (error) |
836 | return error; | |
837 | } | |
1da177e4 | 838 | |
09cbfeaf | 839 | get_page(page); |
66a0c8ee KS |
840 | page->mapping = mapping; |
841 | page->index = offset; | |
842 | ||
74d60958 MW |
843 | do { |
844 | xas_lock_irq(&xas); | |
845 | old = xas_load(&xas); | |
846 | if (old && !xa_is_value(old)) | |
847 | xas_set_err(&xas, -EEXIST); | |
848 | xas_store(&xas, page); | |
849 | if (xas_error(&xas)) | |
850 | goto unlock; | |
851 | ||
852 | if (xa_is_value(old)) { | |
853 | mapping->nrexceptional--; | |
854 | if (shadowp) | |
855 | *shadowp = old; | |
856 | } | |
857 | mapping->nrpages++; | |
858 | ||
859 | /* hugetlb pages do not participate in page cache accounting */ | |
860 | if (!huge) | |
861 | __inc_node_page_state(page, NR_FILE_PAGES); | |
862 | unlock: | |
863 | xas_unlock_irq(&xas); | |
864 | } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK)); | |
865 | ||
866 | if (xas_error(&xas)) | |
867 | goto error; | |
4165b9b4 | 868 | |
00501b53 | 869 | if (!huge) |
f627c2f5 | 870 | mem_cgroup_commit_charge(page, memcg, false, false); |
66a0c8ee KS |
871 | trace_mm_filemap_add_to_page_cache(page); |
872 | return 0; | |
74d60958 | 873 | error: |
66a0c8ee KS |
874 | page->mapping = NULL; |
875 | /* Leave page->index set: truncation relies upon it */ | |
00501b53 | 876 | if (!huge) |
f627c2f5 | 877 | mem_cgroup_cancel_charge(page, memcg, false); |
09cbfeaf | 878 | put_page(page); |
74d60958 | 879 | return xas_error(&xas); |
1da177e4 | 880 | } |
a528910e JW |
881 | |
882 | /** | |
883 | * add_to_page_cache_locked - add a locked page to the pagecache | |
884 | * @page: page to add | |
885 | * @mapping: the page's address_space | |
886 | * @offset: page index | |
887 | * @gfp_mask: page allocation mode | |
888 | * | |
889 | * This function is used to add a page to the pagecache. It must be locked. | |
890 | * This function does not add the page to the LRU. The caller must do that. | |
a862f68a MR |
891 | * |
892 | * Return: %0 on success, negative error code otherwise. | |
a528910e JW |
893 | */ |
894 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
895 | pgoff_t offset, gfp_t gfp_mask) | |
896 | { | |
897 | return __add_to_page_cache_locked(page, mapping, offset, | |
898 | gfp_mask, NULL); | |
899 | } | |
e286781d | 900 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
901 | |
902 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 903 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 904 | { |
a528910e | 905 | void *shadow = NULL; |
4f98a2fe RR |
906 | int ret; |
907 | ||
48c935ad | 908 | __SetPageLocked(page); |
a528910e JW |
909 | ret = __add_to_page_cache_locked(page, mapping, offset, |
910 | gfp_mask, &shadow); | |
911 | if (unlikely(ret)) | |
48c935ad | 912 | __ClearPageLocked(page); |
a528910e JW |
913 | else { |
914 | /* | |
915 | * The page might have been evicted from cache only | |
916 | * recently, in which case it should be activated like | |
917 | * any other repeatedly accessed page. | |
f0281a00 RR |
918 | * The exception is pages getting rewritten; evicting other |
919 | * data from the working set, only to cache data that will | |
920 | * get overwritten with something else, is a waste of memory. | |
a528910e | 921 | */ |
1899ad18 JW |
922 | WARN_ON_ONCE(PageActive(page)); |
923 | if (!(gfp_mask & __GFP_WRITE) && shadow) | |
924 | workingset_refault(page, shadow); | |
a528910e JW |
925 | lru_cache_add(page); |
926 | } | |
1da177e4 LT |
927 | return ret; |
928 | } | |
18bc0bbd | 929 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 930 | |
44110fe3 | 931 | #ifdef CONFIG_NUMA |
2ae88149 | 932 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 933 | { |
c0ff7453 MX |
934 | int n; |
935 | struct page *page; | |
936 | ||
44110fe3 | 937 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
938 | unsigned int cpuset_mems_cookie; |
939 | do { | |
d26914d1 | 940 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 941 | n = cpuset_mem_spread_node(); |
96db800f | 942 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 943 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 944 | |
c0ff7453 | 945 | return page; |
44110fe3 | 946 | } |
2ae88149 | 947 | return alloc_pages(gfp, 0); |
44110fe3 | 948 | } |
2ae88149 | 949 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
950 | #endif |
951 | ||
1da177e4 LT |
952 | /* |
953 | * In order to wait for pages to become available there must be | |
954 | * waitqueues associated with pages. By using a hash table of | |
955 | * waitqueues where the bucket discipline is to maintain all | |
956 | * waiters on the same queue and wake all when any of the pages | |
957 | * become available, and for the woken contexts to check to be | |
958 | * sure the appropriate page became available, this saves space | |
959 | * at a cost of "thundering herd" phenomena during rare hash | |
960 | * collisions. | |
961 | */ | |
62906027 NP |
962 | #define PAGE_WAIT_TABLE_BITS 8 |
963 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
964 | static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; | |
965 | ||
966 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
1da177e4 | 967 | { |
62906027 | 968 | return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 969 | } |
1da177e4 | 970 | |
62906027 | 971 | void __init pagecache_init(void) |
1da177e4 | 972 | { |
62906027 | 973 | int i; |
1da177e4 | 974 | |
62906027 NP |
975 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
976 | init_waitqueue_head(&page_wait_table[i]); | |
977 | ||
978 | page_writeback_init(); | |
1da177e4 | 979 | } |
1da177e4 | 980 | |
3510ca20 | 981 | /* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */ |
62906027 NP |
982 | struct wait_page_key { |
983 | struct page *page; | |
984 | int bit_nr; | |
985 | int page_match; | |
986 | }; | |
987 | ||
988 | struct wait_page_queue { | |
989 | struct page *page; | |
990 | int bit_nr; | |
ac6424b9 | 991 | wait_queue_entry_t wait; |
62906027 NP |
992 | }; |
993 | ||
ac6424b9 | 994 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 995 | { |
62906027 NP |
996 | struct wait_page_key *key = arg; |
997 | struct wait_page_queue *wait_page | |
998 | = container_of(wait, struct wait_page_queue, wait); | |
999 | ||
1000 | if (wait_page->page != key->page) | |
1001 | return 0; | |
1002 | key->page_match = 1; | |
f62e00cc | 1003 | |
62906027 NP |
1004 | if (wait_page->bit_nr != key->bit_nr) |
1005 | return 0; | |
3510ca20 | 1006 | |
9a1ea439 HD |
1007 | /* |
1008 | * Stop walking if it's locked. | |
1009 | * Is this safe if put_and_wait_on_page_locked() is in use? | |
1010 | * Yes: the waker must hold a reference to this page, and if PG_locked | |
1011 | * has now already been set by another task, that task must also hold | |
1012 | * a reference to the *same usage* of this page; so there is no need | |
1013 | * to walk on to wake even the put_and_wait_on_page_locked() callers. | |
1014 | */ | |
62906027 | 1015 | if (test_bit(key->bit_nr, &key->page->flags)) |
3510ca20 | 1016 | return -1; |
f62e00cc | 1017 | |
62906027 | 1018 | return autoremove_wake_function(wait, mode, sync, key); |
f62e00cc KM |
1019 | } |
1020 | ||
74d81bfa | 1021 | static void wake_up_page_bit(struct page *page, int bit_nr) |
cbbce822 | 1022 | { |
62906027 NP |
1023 | wait_queue_head_t *q = page_waitqueue(page); |
1024 | struct wait_page_key key; | |
1025 | unsigned long flags; | |
11a19c7b | 1026 | wait_queue_entry_t bookmark; |
cbbce822 | 1027 | |
62906027 NP |
1028 | key.page = page; |
1029 | key.bit_nr = bit_nr; | |
1030 | key.page_match = 0; | |
1031 | ||
11a19c7b TC |
1032 | bookmark.flags = 0; |
1033 | bookmark.private = NULL; | |
1034 | bookmark.func = NULL; | |
1035 | INIT_LIST_HEAD(&bookmark.entry); | |
1036 | ||
62906027 | 1037 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1038 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1039 | ||
1040 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1041 | /* | |
1042 | * Take a breather from holding the lock, | |
1043 | * allow pages that finish wake up asynchronously | |
1044 | * to acquire the lock and remove themselves | |
1045 | * from wait queue | |
1046 | */ | |
1047 | spin_unlock_irqrestore(&q->lock, flags); | |
1048 | cpu_relax(); | |
1049 | spin_lock_irqsave(&q->lock, flags); | |
1050 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1051 | } | |
1052 | ||
62906027 NP |
1053 | /* |
1054 | * It is possible for other pages to have collided on the waitqueue | |
1055 | * hash, so in that case check for a page match. That prevents a long- | |
1056 | * term waiter | |
1057 | * | |
1058 | * It is still possible to miss a case here, when we woke page waiters | |
1059 | * and removed them from the waitqueue, but there are still other | |
1060 | * page waiters. | |
1061 | */ | |
1062 | if (!waitqueue_active(q) || !key.page_match) { | |
1063 | ClearPageWaiters(page); | |
1064 | /* | |
1065 | * It's possible to miss clearing Waiters here, when we woke | |
1066 | * our page waiters, but the hashed waitqueue has waiters for | |
1067 | * other pages on it. | |
1068 | * | |
1069 | * That's okay, it's a rare case. The next waker will clear it. | |
1070 | */ | |
1071 | } | |
1072 | spin_unlock_irqrestore(&q->lock, flags); | |
1073 | } | |
74d81bfa NP |
1074 | |
1075 | static void wake_up_page(struct page *page, int bit) | |
1076 | { | |
1077 | if (!PageWaiters(page)) | |
1078 | return; | |
1079 | wake_up_page_bit(page, bit); | |
1080 | } | |
62906027 | 1081 | |
9a1ea439 HD |
1082 | /* |
1083 | * A choice of three behaviors for wait_on_page_bit_common(): | |
1084 | */ | |
1085 | enum behavior { | |
1086 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like | |
1087 | * __lock_page() waiting on then setting PG_locked. | |
1088 | */ | |
1089 | SHARED, /* Hold ref to page and check the bit when woken, like | |
1090 | * wait_on_page_writeback() waiting on PG_writeback. | |
1091 | */ | |
1092 | DROP, /* Drop ref to page before wait, no check when woken, | |
1093 | * like put_and_wait_on_page_locked() on PG_locked. | |
1094 | */ | |
1095 | }; | |
1096 | ||
62906027 | 1097 | static inline int wait_on_page_bit_common(wait_queue_head_t *q, |
9a1ea439 | 1098 | struct page *page, int bit_nr, int state, enum behavior behavior) |
62906027 NP |
1099 | { |
1100 | struct wait_page_queue wait_page; | |
ac6424b9 | 1101 | wait_queue_entry_t *wait = &wait_page.wait; |
9a1ea439 | 1102 | bool bit_is_set; |
b1d29ba8 | 1103 | bool thrashing = false; |
9a1ea439 | 1104 | bool delayacct = false; |
eb414681 | 1105 | unsigned long pflags; |
62906027 NP |
1106 | int ret = 0; |
1107 | ||
eb414681 | 1108 | if (bit_nr == PG_locked && |
b1d29ba8 | 1109 | !PageUptodate(page) && PageWorkingset(page)) { |
9a1ea439 | 1110 | if (!PageSwapBacked(page)) { |
eb414681 | 1111 | delayacct_thrashing_start(); |
9a1ea439 HD |
1112 | delayacct = true; |
1113 | } | |
eb414681 | 1114 | psi_memstall_enter(&pflags); |
b1d29ba8 JW |
1115 | thrashing = true; |
1116 | } | |
1117 | ||
62906027 | 1118 | init_wait(wait); |
9a1ea439 | 1119 | wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0; |
62906027 NP |
1120 | wait->func = wake_page_function; |
1121 | wait_page.page = page; | |
1122 | wait_page.bit_nr = bit_nr; | |
1123 | ||
1124 | for (;;) { | |
1125 | spin_lock_irq(&q->lock); | |
1126 | ||
2055da97 | 1127 | if (likely(list_empty(&wait->entry))) { |
3510ca20 | 1128 | __add_wait_queue_entry_tail(q, wait); |
62906027 NP |
1129 | SetPageWaiters(page); |
1130 | } | |
1131 | ||
1132 | set_current_state(state); | |
1133 | ||
1134 | spin_unlock_irq(&q->lock); | |
1135 | ||
9a1ea439 HD |
1136 | bit_is_set = test_bit(bit_nr, &page->flags); |
1137 | if (behavior == DROP) | |
1138 | put_page(page); | |
1139 | ||
1140 | if (likely(bit_is_set)) | |
62906027 | 1141 | io_schedule(); |
62906027 | 1142 | |
9a1ea439 | 1143 | if (behavior == EXCLUSIVE) { |
62906027 NP |
1144 | if (!test_and_set_bit_lock(bit_nr, &page->flags)) |
1145 | break; | |
9a1ea439 | 1146 | } else if (behavior == SHARED) { |
62906027 NP |
1147 | if (!test_bit(bit_nr, &page->flags)) |
1148 | break; | |
1149 | } | |
a8b169af | 1150 | |
fa45f116 | 1151 | if (signal_pending_state(state, current)) { |
a8b169af LT |
1152 | ret = -EINTR; |
1153 | break; | |
1154 | } | |
9a1ea439 HD |
1155 | |
1156 | if (behavior == DROP) { | |
1157 | /* | |
1158 | * We can no longer safely access page->flags: | |
1159 | * even if CONFIG_MEMORY_HOTREMOVE is not enabled, | |
1160 | * there is a risk of waiting forever on a page reused | |
1161 | * for something that keeps it locked indefinitely. | |
1162 | * But best check for -EINTR above before breaking. | |
1163 | */ | |
1164 | break; | |
1165 | } | |
62906027 NP |
1166 | } |
1167 | ||
1168 | finish_wait(q, wait); | |
1169 | ||
eb414681 | 1170 | if (thrashing) { |
9a1ea439 | 1171 | if (delayacct) |
eb414681 JW |
1172 | delayacct_thrashing_end(); |
1173 | psi_memstall_leave(&pflags); | |
1174 | } | |
b1d29ba8 | 1175 | |
62906027 NP |
1176 | /* |
1177 | * A signal could leave PageWaiters set. Clearing it here if | |
1178 | * !waitqueue_active would be possible (by open-coding finish_wait), | |
1179 | * but still fail to catch it in the case of wait hash collision. We | |
1180 | * already can fail to clear wait hash collision cases, so don't | |
1181 | * bother with signals either. | |
1182 | */ | |
1183 | ||
1184 | return ret; | |
1185 | } | |
1186 | ||
1187 | void wait_on_page_bit(struct page *page, int bit_nr) | |
1188 | { | |
1189 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 | 1190 | wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
62906027 NP |
1191 | } |
1192 | EXPORT_SYMBOL(wait_on_page_bit); | |
1193 | ||
1194 | int wait_on_page_bit_killable(struct page *page, int bit_nr) | |
1195 | { | |
1196 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 | 1197 | return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED); |
cbbce822 | 1198 | } |
4343d008 | 1199 | EXPORT_SYMBOL(wait_on_page_bit_killable); |
cbbce822 | 1200 | |
9a1ea439 HD |
1201 | /** |
1202 | * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked | |
1203 | * @page: The page to wait for. | |
1204 | * | |
1205 | * The caller should hold a reference on @page. They expect the page to | |
1206 | * become unlocked relatively soon, but do not wish to hold up migration | |
1207 | * (for example) by holding the reference while waiting for the page to | |
1208 | * come unlocked. After this function returns, the caller should not | |
1209 | * dereference @page. | |
1210 | */ | |
1211 | void put_and_wait_on_page_locked(struct page *page) | |
1212 | { | |
1213 | wait_queue_head_t *q; | |
1214 | ||
1215 | page = compound_head(page); | |
1216 | q = page_waitqueue(page); | |
1217 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP); | |
1218 | } | |
1219 | ||
385e1ca5 DH |
1220 | /** |
1221 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
1222 | * @page: Page defining the wait queue of interest |
1223 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
1224 | * |
1225 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
1226 | */ | |
ac6424b9 | 1227 | void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter) |
385e1ca5 DH |
1228 | { |
1229 | wait_queue_head_t *q = page_waitqueue(page); | |
1230 | unsigned long flags; | |
1231 | ||
1232 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1233 | __add_wait_queue_entry_tail(q, waiter); |
62906027 | 1234 | SetPageWaiters(page); |
385e1ca5 DH |
1235 | spin_unlock_irqrestore(&q->lock, flags); |
1236 | } | |
1237 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
1238 | ||
b91e1302 LT |
1239 | #ifndef clear_bit_unlock_is_negative_byte |
1240 | ||
1241 | /* | |
1242 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1243 | * | |
1244 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1245 | * test the sign bit at the same time. But if the architecture does | |
1246 | * not support that special operation, we just do this all by hand | |
1247 | * instead. | |
1248 | * | |
1249 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
1250 | * being cleared, but a memory barrier should be unneccssary since it is | |
1251 | * in the same byte as PG_locked. | |
1252 | */ | |
1253 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1254 | { | |
1255 | clear_bit_unlock(nr, mem); | |
1256 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1257 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1258 | } |
1259 | ||
1260 | #endif | |
1261 | ||
1da177e4 | 1262 | /** |
485bb99b | 1263 | * unlock_page - unlock a locked page |
1da177e4 LT |
1264 | * @page: the page |
1265 | * | |
1266 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
1267 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 1268 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
1269 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
1270 | * | |
b91e1302 LT |
1271 | * Note that this depends on PG_waiters being the sign bit in the byte |
1272 | * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to | |
1273 | * clear the PG_locked bit and test PG_waiters at the same time fairly | |
1274 | * portably (architectures that do LL/SC can test any bit, while x86 can | |
1275 | * test the sign bit). | |
1da177e4 | 1276 | */ |
920c7a5d | 1277 | void unlock_page(struct page *page) |
1da177e4 | 1278 | { |
b91e1302 | 1279 | BUILD_BUG_ON(PG_waiters != 7); |
48c935ad | 1280 | page = compound_head(page); |
309381fe | 1281 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
b91e1302 LT |
1282 | if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags)) |
1283 | wake_up_page_bit(page, PG_locked); | |
1da177e4 LT |
1284 | } |
1285 | EXPORT_SYMBOL(unlock_page); | |
1286 | ||
485bb99b RD |
1287 | /** |
1288 | * end_page_writeback - end writeback against a page | |
1289 | * @page: the page | |
1da177e4 LT |
1290 | */ |
1291 | void end_page_writeback(struct page *page) | |
1292 | { | |
888cf2db MG |
1293 | /* |
1294 | * TestClearPageReclaim could be used here but it is an atomic | |
1295 | * operation and overkill in this particular case. Failing to | |
1296 | * shuffle a page marked for immediate reclaim is too mild to | |
1297 | * justify taking an atomic operation penalty at the end of | |
1298 | * ever page writeback. | |
1299 | */ | |
1300 | if (PageReclaim(page)) { | |
1301 | ClearPageReclaim(page); | |
ac6aadb2 | 1302 | rotate_reclaimable_page(page); |
888cf2db | 1303 | } |
ac6aadb2 MS |
1304 | |
1305 | if (!test_clear_page_writeback(page)) | |
1306 | BUG(); | |
1307 | ||
4e857c58 | 1308 | smp_mb__after_atomic(); |
1da177e4 LT |
1309 | wake_up_page(page, PG_writeback); |
1310 | } | |
1311 | EXPORT_SYMBOL(end_page_writeback); | |
1312 | ||
57d99845 MW |
1313 | /* |
1314 | * After completing I/O on a page, call this routine to update the page | |
1315 | * flags appropriately | |
1316 | */ | |
c11f0c0b | 1317 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1318 | { |
c11f0c0b | 1319 | if (!is_write) { |
57d99845 MW |
1320 | if (!err) { |
1321 | SetPageUptodate(page); | |
1322 | } else { | |
1323 | ClearPageUptodate(page); | |
1324 | SetPageError(page); | |
1325 | } | |
1326 | unlock_page(page); | |
abf54548 | 1327 | } else { |
57d99845 | 1328 | if (err) { |
dd8416c4 MK |
1329 | struct address_space *mapping; |
1330 | ||
57d99845 | 1331 | SetPageError(page); |
dd8416c4 MK |
1332 | mapping = page_mapping(page); |
1333 | if (mapping) | |
1334 | mapping_set_error(mapping, err); | |
57d99845 MW |
1335 | } |
1336 | end_page_writeback(page); | |
1337 | } | |
1338 | } | |
1339 | EXPORT_SYMBOL_GPL(page_endio); | |
1340 | ||
485bb99b RD |
1341 | /** |
1342 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
87066755 | 1343 | * @__page: the page to lock |
1da177e4 | 1344 | */ |
62906027 | 1345 | void __lock_page(struct page *__page) |
1da177e4 | 1346 | { |
62906027 NP |
1347 | struct page *page = compound_head(__page); |
1348 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 HD |
1349 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, |
1350 | EXCLUSIVE); | |
1da177e4 LT |
1351 | } |
1352 | EXPORT_SYMBOL(__lock_page); | |
1353 | ||
62906027 | 1354 | int __lock_page_killable(struct page *__page) |
2687a356 | 1355 | { |
62906027 NP |
1356 | struct page *page = compound_head(__page); |
1357 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 HD |
1358 | return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, |
1359 | EXCLUSIVE); | |
2687a356 | 1360 | } |
18bc0bbd | 1361 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 1362 | |
9a95f3cf PC |
1363 | /* |
1364 | * Return values: | |
1365 | * 1 - page is locked; mmap_sem is still held. | |
1366 | * 0 - page is not locked. | |
1367 | * mmap_sem has been released (up_read()), unless flags had both | |
1368 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in | |
1369 | * which case mmap_sem is still held. | |
1370 | * | |
1371 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
1372 | * with the page locked and the mmap_sem unperturbed. | |
1373 | */ | |
d065bd81 ML |
1374 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
1375 | unsigned int flags) | |
1376 | { | |
37b23e05 KM |
1377 | if (flags & FAULT_FLAG_ALLOW_RETRY) { |
1378 | /* | |
1379 | * CAUTION! In this case, mmap_sem is not released | |
1380 | * even though return 0. | |
1381 | */ | |
1382 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
1383 | return 0; | |
1384 | ||
1385 | up_read(&mm->mmap_sem); | |
1386 | if (flags & FAULT_FLAG_KILLABLE) | |
1387 | wait_on_page_locked_killable(page); | |
1388 | else | |
318b275f | 1389 | wait_on_page_locked(page); |
d065bd81 | 1390 | return 0; |
37b23e05 KM |
1391 | } else { |
1392 | if (flags & FAULT_FLAG_KILLABLE) { | |
1393 | int ret; | |
1394 | ||
1395 | ret = __lock_page_killable(page); | |
1396 | if (ret) { | |
1397 | up_read(&mm->mmap_sem); | |
1398 | return 0; | |
1399 | } | |
1400 | } else | |
1401 | __lock_page(page); | |
1402 | return 1; | |
d065bd81 ML |
1403 | } |
1404 | } | |
1405 | ||
e7b563bb | 1406 | /** |
0d3f9296 MW |
1407 | * page_cache_next_miss() - Find the next gap in the page cache. |
1408 | * @mapping: Mapping. | |
1409 | * @index: Index. | |
1410 | * @max_scan: Maximum range to search. | |
e7b563bb | 1411 | * |
0d3f9296 MW |
1412 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1413 | * gap with the lowest index. | |
e7b563bb | 1414 | * |
0d3f9296 MW |
1415 | * This function may be called under the rcu_read_lock. However, this will |
1416 | * not atomically search a snapshot of the cache at a single point in time. | |
1417 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1418 | * created at index 10, page_cache_next_miss covering both indices may | |
1419 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1420 | * |
0d3f9296 MW |
1421 | * Return: The index of the gap if found, otherwise an index outside the |
1422 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1423 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1424 | */ |
0d3f9296 | 1425 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1426 | pgoff_t index, unsigned long max_scan) |
1427 | { | |
0d3f9296 | 1428 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1429 | |
0d3f9296 MW |
1430 | while (max_scan--) { |
1431 | void *entry = xas_next(&xas); | |
1432 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1433 | break; |
0d3f9296 | 1434 | if (xas.xa_index == 0) |
e7b563bb JW |
1435 | break; |
1436 | } | |
1437 | ||
0d3f9296 | 1438 | return xas.xa_index; |
e7b563bb | 1439 | } |
0d3f9296 | 1440 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1441 | |
1442 | /** | |
0d3f9296 MW |
1443 | * page_cache_prev_miss() - Find the next gap in the page cache. |
1444 | * @mapping: Mapping. | |
1445 | * @index: Index. | |
1446 | * @max_scan: Maximum range to search. | |
e7b563bb | 1447 | * |
0d3f9296 MW |
1448 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1449 | * gap with the highest index. | |
e7b563bb | 1450 | * |
0d3f9296 MW |
1451 | * This function may be called under the rcu_read_lock. However, this will |
1452 | * not atomically search a snapshot of the cache at a single point in time. | |
1453 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1454 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1455 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1456 | * |
0d3f9296 MW |
1457 | * Return: The index of the gap if found, otherwise an index outside the |
1458 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1459 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1460 | */ |
0d3f9296 | 1461 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1462 | pgoff_t index, unsigned long max_scan) |
1463 | { | |
0d3f9296 | 1464 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1465 | |
0d3f9296 MW |
1466 | while (max_scan--) { |
1467 | void *entry = xas_prev(&xas); | |
1468 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1469 | break; |
0d3f9296 | 1470 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1471 | break; |
1472 | } | |
1473 | ||
0d3f9296 | 1474 | return xas.xa_index; |
e7b563bb | 1475 | } |
0d3f9296 | 1476 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1477 | |
485bb99b | 1478 | /** |
0cd6144a | 1479 | * find_get_entry - find and get a page cache entry |
485bb99b | 1480 | * @mapping: the address_space to search |
0cd6144a JW |
1481 | * @offset: the page cache index |
1482 | * | |
1483 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1484 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1485 | * |
139b6a6f JW |
1486 | * If the slot holds a shadow entry of a previously evicted page, or a |
1487 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a | 1488 | * |
a862f68a | 1489 | * Return: the found page or shadow entry, %NULL if nothing is found. |
1da177e4 | 1490 | */ |
0cd6144a | 1491 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1492 | { |
4c7472c0 | 1493 | XA_STATE(xas, &mapping->i_pages, offset); |
83929372 | 1494 | struct page *head, *page; |
1da177e4 | 1495 | |
a60637c8 NP |
1496 | rcu_read_lock(); |
1497 | repeat: | |
4c7472c0 MW |
1498 | xas_reset(&xas); |
1499 | page = xas_load(&xas); | |
1500 | if (xas_retry(&xas, page)) | |
1501 | goto repeat; | |
1502 | /* | |
1503 | * A shadow entry of a recently evicted page, or a swap entry from | |
1504 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1505 | */ | |
1506 | if (!page || xa_is_value(page)) | |
1507 | goto out; | |
83929372 | 1508 | |
4c7472c0 MW |
1509 | head = compound_head(page); |
1510 | if (!page_cache_get_speculative(head)) | |
1511 | goto repeat; | |
83929372 | 1512 | |
4c7472c0 MW |
1513 | /* The page was split under us? */ |
1514 | if (compound_head(page) != head) { | |
1515 | put_page(head); | |
1516 | goto repeat; | |
1517 | } | |
a60637c8 | 1518 | |
4c7472c0 MW |
1519 | /* |
1520 | * Has the page moved? | |
1521 | * This is part of the lockless pagecache protocol. See | |
1522 | * include/linux/pagemap.h for details. | |
1523 | */ | |
1524 | if (unlikely(page != xas_reload(&xas))) { | |
1525 | put_page(head); | |
1526 | goto repeat; | |
a60637c8 | 1527 | } |
27d20fdd | 1528 | out: |
a60637c8 NP |
1529 | rcu_read_unlock(); |
1530 | ||
1da177e4 LT |
1531 | return page; |
1532 | } | |
0cd6144a | 1533 | EXPORT_SYMBOL(find_get_entry); |
1da177e4 | 1534 | |
0cd6144a JW |
1535 | /** |
1536 | * find_lock_entry - locate, pin and lock a page cache entry | |
1537 | * @mapping: the address_space to search | |
1538 | * @offset: the page cache index | |
1539 | * | |
1540 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1541 | * page cache page, it is returned locked and with an increased | |
1542 | * refcount. | |
1543 | * | |
139b6a6f JW |
1544 | * If the slot holds a shadow entry of a previously evicted page, or a |
1545 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a | 1546 | * |
0cd6144a | 1547 | * find_lock_entry() may sleep. |
a862f68a MR |
1548 | * |
1549 | * Return: the found page or shadow entry, %NULL if nothing is found. | |
0cd6144a JW |
1550 | */ |
1551 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1552 | { |
1553 | struct page *page; | |
1554 | ||
1da177e4 | 1555 | repeat: |
0cd6144a | 1556 | page = find_get_entry(mapping, offset); |
4c7472c0 | 1557 | if (page && !xa_is_value(page)) { |
a60637c8 NP |
1558 | lock_page(page); |
1559 | /* Has the page been truncated? */ | |
83929372 | 1560 | if (unlikely(page_mapping(page) != mapping)) { |
a60637c8 | 1561 | unlock_page(page); |
09cbfeaf | 1562 | put_page(page); |
a60637c8 | 1563 | goto repeat; |
1da177e4 | 1564 | } |
83929372 | 1565 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
1da177e4 | 1566 | } |
1da177e4 LT |
1567 | return page; |
1568 | } | |
0cd6144a JW |
1569 | EXPORT_SYMBOL(find_lock_entry); |
1570 | ||
1571 | /** | |
2457aec6 | 1572 | * pagecache_get_page - find and get a page reference |
0cd6144a JW |
1573 | * @mapping: the address_space to search |
1574 | * @offset: the page index | |
2457aec6 | 1575 | * @fgp_flags: PCG flags |
45f87de5 | 1576 | * @gfp_mask: gfp mask to use for the page cache data page allocation |
0cd6144a | 1577 | * |
2457aec6 | 1578 | * Looks up the page cache slot at @mapping & @offset. |
1da177e4 | 1579 | * |
75325189 | 1580 | * PCG flags modify how the page is returned. |
0cd6144a | 1581 | * |
0e056eb5 | 1582 | * @fgp_flags can be: |
1583 | * | |
1584 | * - FGP_ACCESSED: the page will be marked accessed | |
1585 | * - FGP_LOCK: Page is return locked | |
1586 | * - FGP_CREAT: If page is not present then a new page is allocated using | |
1587 | * @gfp_mask and added to the page cache and the VM's LRU | |
1588 | * list. The page is returned locked and with an increased | |
a862f68a | 1589 | * refcount. |
1da177e4 | 1590 | * |
2457aec6 MG |
1591 | * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even |
1592 | * if the GFP flags specified for FGP_CREAT are atomic. | |
1da177e4 | 1593 | * |
2457aec6 | 1594 | * If there is a page cache page, it is returned with an increased refcount. |
a862f68a MR |
1595 | * |
1596 | * Return: the found page or %NULL otherwise. | |
1da177e4 | 1597 | */ |
2457aec6 | 1598 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset, |
45f87de5 | 1599 | int fgp_flags, gfp_t gfp_mask) |
1da177e4 | 1600 | { |
eb2be189 | 1601 | struct page *page; |
2457aec6 | 1602 | |
1da177e4 | 1603 | repeat: |
2457aec6 | 1604 | page = find_get_entry(mapping, offset); |
3159f943 | 1605 | if (xa_is_value(page)) |
2457aec6 MG |
1606 | page = NULL; |
1607 | if (!page) | |
1608 | goto no_page; | |
1609 | ||
1610 | if (fgp_flags & FGP_LOCK) { | |
1611 | if (fgp_flags & FGP_NOWAIT) { | |
1612 | if (!trylock_page(page)) { | |
09cbfeaf | 1613 | put_page(page); |
2457aec6 MG |
1614 | return NULL; |
1615 | } | |
1616 | } else { | |
1617 | lock_page(page); | |
1618 | } | |
1619 | ||
1620 | /* Has the page been truncated? */ | |
1621 | if (unlikely(page->mapping != mapping)) { | |
1622 | unlock_page(page); | |
09cbfeaf | 1623 | put_page(page); |
2457aec6 MG |
1624 | goto repeat; |
1625 | } | |
1626 | VM_BUG_ON_PAGE(page->index != offset, page); | |
1627 | } | |
1628 | ||
c16eb000 | 1629 | if (fgp_flags & FGP_ACCESSED) |
2457aec6 MG |
1630 | mark_page_accessed(page); |
1631 | ||
1632 | no_page: | |
1633 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1634 | int err; | |
1635 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1636 | gfp_mask |= __GFP_WRITE; |
1637 | if (fgp_flags & FGP_NOFS) | |
1638 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1639 | |
45f87de5 | 1640 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1641 | if (!page) |
1642 | return NULL; | |
2457aec6 MG |
1643 | |
1644 | if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK))) | |
1645 | fgp_flags |= FGP_LOCK; | |
1646 | ||
eb39d618 | 1647 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1648 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1649 | __SetPageReferenced(page); |
2457aec6 | 1650 | |
abc1be13 | 1651 | err = add_to_page_cache_lru(page, mapping, offset, gfp_mask); |
eb2be189 | 1652 | if (unlikely(err)) { |
09cbfeaf | 1653 | put_page(page); |
eb2be189 NP |
1654 | page = NULL; |
1655 | if (err == -EEXIST) | |
1656 | goto repeat; | |
1da177e4 | 1657 | } |
1da177e4 | 1658 | } |
2457aec6 | 1659 | |
1da177e4 LT |
1660 | return page; |
1661 | } | |
2457aec6 | 1662 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1663 | |
0cd6144a JW |
1664 | /** |
1665 | * find_get_entries - gang pagecache lookup | |
1666 | * @mapping: The address_space to search | |
1667 | * @start: The starting page cache index | |
1668 | * @nr_entries: The maximum number of entries | |
1669 | * @entries: Where the resulting entries are placed | |
1670 | * @indices: The cache indices corresponding to the entries in @entries | |
1671 | * | |
1672 | * find_get_entries() will search for and return a group of up to | |
1673 | * @nr_entries entries in the mapping. The entries are placed at | |
1674 | * @entries. find_get_entries() takes a reference against any actual | |
1675 | * pages it returns. | |
1676 | * | |
1677 | * The search returns a group of mapping-contiguous page cache entries | |
1678 | * with ascending indexes. There may be holes in the indices due to | |
1679 | * not-present pages. | |
1680 | * | |
139b6a6f JW |
1681 | * Any shadow entries of evicted pages, or swap entries from |
1682 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a | 1683 | * |
a862f68a | 1684 | * Return: the number of pages and shadow entries which were found. |
0cd6144a JW |
1685 | */ |
1686 | unsigned find_get_entries(struct address_space *mapping, | |
1687 | pgoff_t start, unsigned int nr_entries, | |
1688 | struct page **entries, pgoff_t *indices) | |
1689 | { | |
f280bf09 MW |
1690 | XA_STATE(xas, &mapping->i_pages, start); |
1691 | struct page *page; | |
0cd6144a | 1692 | unsigned int ret = 0; |
0cd6144a JW |
1693 | |
1694 | if (!nr_entries) | |
1695 | return 0; | |
1696 | ||
1697 | rcu_read_lock(); | |
f280bf09 MW |
1698 | xas_for_each(&xas, page, ULONG_MAX) { |
1699 | struct page *head; | |
1700 | if (xas_retry(&xas, page)) | |
0cd6144a | 1701 | continue; |
f280bf09 MW |
1702 | /* |
1703 | * A shadow entry of a recently evicted page, a swap | |
1704 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1705 | * without attempting to raise page count. | |
1706 | */ | |
1707 | if (xa_is_value(page)) | |
0cd6144a | 1708 | goto export; |
83929372 KS |
1709 | |
1710 | head = compound_head(page); | |
1711 | if (!page_cache_get_speculative(head)) | |
f280bf09 | 1712 | goto retry; |
83929372 KS |
1713 | |
1714 | /* The page was split under us? */ | |
f280bf09 MW |
1715 | if (compound_head(page) != head) |
1716 | goto put_page; | |
0cd6144a JW |
1717 | |
1718 | /* Has the page moved? */ | |
f280bf09 MW |
1719 | if (unlikely(page != xas_reload(&xas))) |
1720 | goto put_page; | |
1721 | ||
0cd6144a | 1722 | export: |
f280bf09 | 1723 | indices[ret] = xas.xa_index; |
0cd6144a JW |
1724 | entries[ret] = page; |
1725 | if (++ret == nr_entries) | |
1726 | break; | |
f280bf09 MW |
1727 | continue; |
1728 | put_page: | |
1729 | put_page(head); | |
1730 | retry: | |
1731 | xas_reset(&xas); | |
0cd6144a JW |
1732 | } |
1733 | rcu_read_unlock(); | |
1734 | return ret; | |
1735 | } | |
1736 | ||
1da177e4 | 1737 | /** |
b947cee4 | 1738 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
1739 | * @mapping: The address_space to search |
1740 | * @start: The starting page index | |
b947cee4 | 1741 | * @end: The final page index (inclusive) |
1da177e4 LT |
1742 | * @nr_pages: The maximum number of pages |
1743 | * @pages: Where the resulting pages are placed | |
1744 | * | |
b947cee4 JK |
1745 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
1746 | * pages in the mapping starting at index @start and up to index @end | |
1747 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
1748 | * a reference against the returned pages. | |
1da177e4 LT |
1749 | * |
1750 | * The search returns a group of mapping-contiguous pages with ascending | |
1751 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 1752 | * We also update @start to index the next page for the traversal. |
1da177e4 | 1753 | * |
a862f68a MR |
1754 | * Return: the number of pages which were found. If this number is |
1755 | * smaller than @nr_pages, the end of specified range has been | |
b947cee4 | 1756 | * reached. |
1da177e4 | 1757 | */ |
b947cee4 JK |
1758 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
1759 | pgoff_t end, unsigned int nr_pages, | |
1760 | struct page **pages) | |
1da177e4 | 1761 | { |
fd1b3cee MW |
1762 | XA_STATE(xas, &mapping->i_pages, *start); |
1763 | struct page *page; | |
0fc9d104 KK |
1764 | unsigned ret = 0; |
1765 | ||
1766 | if (unlikely(!nr_pages)) | |
1767 | return 0; | |
a60637c8 NP |
1768 | |
1769 | rcu_read_lock(); | |
fd1b3cee MW |
1770 | xas_for_each(&xas, page, end) { |
1771 | struct page *head; | |
1772 | if (xas_retry(&xas, page)) | |
a60637c8 | 1773 | continue; |
fd1b3cee MW |
1774 | /* Skip over shadow, swap and DAX entries */ |
1775 | if (xa_is_value(page)) | |
8079b1c8 | 1776 | continue; |
a60637c8 | 1777 | |
83929372 KS |
1778 | head = compound_head(page); |
1779 | if (!page_cache_get_speculative(head)) | |
fd1b3cee | 1780 | goto retry; |
83929372 KS |
1781 | |
1782 | /* The page was split under us? */ | |
fd1b3cee MW |
1783 | if (compound_head(page) != head) |
1784 | goto put_page; | |
a60637c8 NP |
1785 | |
1786 | /* Has the page moved? */ | |
fd1b3cee MW |
1787 | if (unlikely(page != xas_reload(&xas))) |
1788 | goto put_page; | |
1da177e4 | 1789 | |
a60637c8 | 1790 | pages[ret] = page; |
b947cee4 | 1791 | if (++ret == nr_pages) { |
fd1b3cee | 1792 | *start = page->index + 1; |
b947cee4 JK |
1793 | goto out; |
1794 | } | |
fd1b3cee MW |
1795 | continue; |
1796 | put_page: | |
1797 | put_page(head); | |
1798 | retry: | |
1799 | xas_reset(&xas); | |
a60637c8 | 1800 | } |
5b280c0c | 1801 | |
b947cee4 JK |
1802 | /* |
1803 | * We come here when there is no page beyond @end. We take care to not | |
1804 | * overflow the index @start as it confuses some of the callers. This | |
fd1b3cee | 1805 | * breaks the iteration when there is a page at index -1 but that is |
b947cee4 JK |
1806 | * already broken anyway. |
1807 | */ | |
1808 | if (end == (pgoff_t)-1) | |
1809 | *start = (pgoff_t)-1; | |
1810 | else | |
1811 | *start = end + 1; | |
1812 | out: | |
a60637c8 | 1813 | rcu_read_unlock(); |
d72dc8a2 | 1814 | |
1da177e4 LT |
1815 | return ret; |
1816 | } | |
1817 | ||
ebf43500 JA |
1818 | /** |
1819 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1820 | * @mapping: The address_space to search | |
1821 | * @index: The starting page index | |
1822 | * @nr_pages: The maximum number of pages | |
1823 | * @pages: Where the resulting pages are placed | |
1824 | * | |
1825 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
1826 | * that the returned number of pages are guaranteed to be contiguous. | |
1827 | * | |
a862f68a | 1828 | * Return: the number of pages which were found. |
ebf43500 JA |
1829 | */ |
1830 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
1831 | unsigned int nr_pages, struct page **pages) | |
1832 | { | |
3ece58a2 MW |
1833 | XA_STATE(xas, &mapping->i_pages, index); |
1834 | struct page *page; | |
0fc9d104 KK |
1835 | unsigned int ret = 0; |
1836 | ||
1837 | if (unlikely(!nr_pages)) | |
1838 | return 0; | |
a60637c8 NP |
1839 | |
1840 | rcu_read_lock(); | |
3ece58a2 MW |
1841 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { |
1842 | struct page *head; | |
1843 | if (xas_retry(&xas, page)) | |
1844 | continue; | |
1845 | /* | |
1846 | * If the entry has been swapped out, we can stop looking. | |
1847 | * No current caller is looking for DAX entries. | |
1848 | */ | |
1849 | if (xa_is_value(page)) | |
8079b1c8 | 1850 | break; |
ebf43500 | 1851 | |
83929372 KS |
1852 | head = compound_head(page); |
1853 | if (!page_cache_get_speculative(head)) | |
3ece58a2 | 1854 | goto retry; |
83929372 KS |
1855 | |
1856 | /* The page was split under us? */ | |
3ece58a2 MW |
1857 | if (compound_head(page) != head) |
1858 | goto put_page; | |
a60637c8 NP |
1859 | |
1860 | /* Has the page moved? */ | |
3ece58a2 MW |
1861 | if (unlikely(page != xas_reload(&xas))) |
1862 | goto put_page; | |
a60637c8 NP |
1863 | |
1864 | pages[ret] = page; | |
0fc9d104 KK |
1865 | if (++ret == nr_pages) |
1866 | break; | |
3ece58a2 MW |
1867 | continue; |
1868 | put_page: | |
1869 | put_page(head); | |
1870 | retry: | |
1871 | xas_reset(&xas); | |
ebf43500 | 1872 | } |
a60637c8 NP |
1873 | rcu_read_unlock(); |
1874 | return ret; | |
ebf43500 | 1875 | } |
ef71c15c | 1876 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 1877 | |
485bb99b | 1878 | /** |
72b045ae | 1879 | * find_get_pages_range_tag - find and return pages in given range matching @tag |
485bb99b RD |
1880 | * @mapping: the address_space to search |
1881 | * @index: the starting page index | |
72b045ae | 1882 | * @end: The final page index (inclusive) |
485bb99b RD |
1883 | * @tag: the tag index |
1884 | * @nr_pages: the maximum number of pages | |
1885 | * @pages: where the resulting pages are placed | |
1886 | * | |
1da177e4 | 1887 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 1888 | * @tag. We update @index to index the next page for the traversal. |
a862f68a MR |
1889 | * |
1890 | * Return: the number of pages which were found. | |
1da177e4 | 1891 | */ |
72b045ae | 1892 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
a6906972 | 1893 | pgoff_t end, xa_mark_t tag, unsigned int nr_pages, |
72b045ae | 1894 | struct page **pages) |
1da177e4 | 1895 | { |
a6906972 MW |
1896 | XA_STATE(xas, &mapping->i_pages, *index); |
1897 | struct page *page; | |
0fc9d104 KK |
1898 | unsigned ret = 0; |
1899 | ||
1900 | if (unlikely(!nr_pages)) | |
1901 | return 0; | |
a60637c8 NP |
1902 | |
1903 | rcu_read_lock(); | |
a6906972 MW |
1904 | xas_for_each_marked(&xas, page, end, tag) { |
1905 | struct page *head; | |
1906 | if (xas_retry(&xas, page)) | |
a60637c8 | 1907 | continue; |
a6906972 MW |
1908 | /* |
1909 | * Shadow entries should never be tagged, but this iteration | |
1910 | * is lockless so there is a window for page reclaim to evict | |
1911 | * a page we saw tagged. Skip over it. | |
1912 | */ | |
1913 | if (xa_is_value(page)) | |
139b6a6f | 1914 | continue; |
a60637c8 | 1915 | |
83929372 KS |
1916 | head = compound_head(page); |
1917 | if (!page_cache_get_speculative(head)) | |
a6906972 | 1918 | goto retry; |
a60637c8 | 1919 | |
83929372 | 1920 | /* The page was split under us? */ |
a6906972 MW |
1921 | if (compound_head(page) != head) |
1922 | goto put_page; | |
83929372 | 1923 | |
a60637c8 | 1924 | /* Has the page moved? */ |
a6906972 MW |
1925 | if (unlikely(page != xas_reload(&xas))) |
1926 | goto put_page; | |
a60637c8 NP |
1927 | |
1928 | pages[ret] = page; | |
72b045ae | 1929 | if (++ret == nr_pages) { |
a6906972 | 1930 | *index = page->index + 1; |
72b045ae JK |
1931 | goto out; |
1932 | } | |
a6906972 MW |
1933 | continue; |
1934 | put_page: | |
1935 | put_page(head); | |
1936 | retry: | |
1937 | xas_reset(&xas); | |
a60637c8 | 1938 | } |
5b280c0c | 1939 | |
72b045ae | 1940 | /* |
a6906972 | 1941 | * We come here when we got to @end. We take care to not overflow the |
72b045ae | 1942 | * index @index as it confuses some of the callers. This breaks the |
a6906972 MW |
1943 | * iteration when there is a page at index -1 but that is already |
1944 | * broken anyway. | |
72b045ae JK |
1945 | */ |
1946 | if (end == (pgoff_t)-1) | |
1947 | *index = (pgoff_t)-1; | |
1948 | else | |
1949 | *index = end + 1; | |
1950 | out: | |
a60637c8 | 1951 | rcu_read_unlock(); |
1da177e4 | 1952 | |
1da177e4 LT |
1953 | return ret; |
1954 | } | |
72b045ae | 1955 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 1956 | |
7e7f7749 RZ |
1957 | /** |
1958 | * find_get_entries_tag - find and return entries that match @tag | |
1959 | * @mapping: the address_space to search | |
1960 | * @start: the starting page cache index | |
1961 | * @tag: the tag index | |
1962 | * @nr_entries: the maximum number of entries | |
1963 | * @entries: where the resulting entries are placed | |
1964 | * @indices: the cache indices corresponding to the entries in @entries | |
1965 | * | |
1966 | * Like find_get_entries, except we only return entries which are tagged with | |
1967 | * @tag. | |
a862f68a MR |
1968 | * |
1969 | * Return: the number of entries which were found. | |
7e7f7749 RZ |
1970 | */ |
1971 | unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start, | |
c1901cd3 | 1972 | xa_mark_t tag, unsigned int nr_entries, |
7e7f7749 RZ |
1973 | struct page **entries, pgoff_t *indices) |
1974 | { | |
c1901cd3 MW |
1975 | XA_STATE(xas, &mapping->i_pages, start); |
1976 | struct page *page; | |
7e7f7749 | 1977 | unsigned int ret = 0; |
7e7f7749 RZ |
1978 | |
1979 | if (!nr_entries) | |
1980 | return 0; | |
1981 | ||
1982 | rcu_read_lock(); | |
c1901cd3 MW |
1983 | xas_for_each_marked(&xas, page, ULONG_MAX, tag) { |
1984 | struct page *head; | |
1985 | if (xas_retry(&xas, page)) | |
7e7f7749 | 1986 | continue; |
c1901cd3 MW |
1987 | /* |
1988 | * A shadow entry of a recently evicted page, a swap | |
1989 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1990 | * without attempting to raise page count. | |
1991 | */ | |
1992 | if (xa_is_value(page)) | |
7e7f7749 | 1993 | goto export; |
83929372 KS |
1994 | |
1995 | head = compound_head(page); | |
1996 | if (!page_cache_get_speculative(head)) | |
c1901cd3 | 1997 | goto retry; |
7e7f7749 | 1998 | |
83929372 | 1999 | /* The page was split under us? */ |
c1901cd3 MW |
2000 | if (compound_head(page) != head) |
2001 | goto put_page; | |
83929372 | 2002 | |
7e7f7749 | 2003 | /* Has the page moved? */ |
c1901cd3 MW |
2004 | if (unlikely(page != xas_reload(&xas))) |
2005 | goto put_page; | |
2006 | ||
7e7f7749 | 2007 | export: |
c1901cd3 | 2008 | indices[ret] = xas.xa_index; |
7e7f7749 RZ |
2009 | entries[ret] = page; |
2010 | if (++ret == nr_entries) | |
2011 | break; | |
c1901cd3 MW |
2012 | continue; |
2013 | put_page: | |
2014 | put_page(head); | |
2015 | retry: | |
2016 | xas_reset(&xas); | |
7e7f7749 RZ |
2017 | } |
2018 | rcu_read_unlock(); | |
2019 | return ret; | |
2020 | } | |
2021 | EXPORT_SYMBOL(find_get_entries_tag); | |
2022 | ||
76d42bd9 WF |
2023 | /* |
2024 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
2025 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
2026 | * | |
2027 | * ---R__________________________________________B__________ | |
2028 | * ^ reading here ^ bad block(assume 4k) | |
2029 | * | |
2030 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
2031 | * => failing the whole request => read(R) => read(R+1) => | |
2032 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
2033 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
2034 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
2035 | * | |
2036 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
2037 | */ | |
2038 | static void shrink_readahead_size_eio(struct file *filp, | |
2039 | struct file_ra_state *ra) | |
2040 | { | |
76d42bd9 | 2041 | ra->ra_pages /= 4; |
76d42bd9 WF |
2042 | } |
2043 | ||
485bb99b | 2044 | /** |
47c27bc4 CH |
2045 | * generic_file_buffered_read - generic file read routine |
2046 | * @iocb: the iocb to read | |
6e58e79d AV |
2047 | * @iter: data destination |
2048 | * @written: already copied | |
485bb99b | 2049 | * |
1da177e4 | 2050 | * This is a generic file read routine, and uses the |
485bb99b | 2051 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
2052 | * |
2053 | * This is really ugly. But the goto's actually try to clarify some | |
2054 | * of the logic when it comes to error handling etc. | |
a862f68a MR |
2055 | * |
2056 | * Return: | |
2057 | * * total number of bytes copied, including those the were already @written | |
2058 | * * negative error code if nothing was copied | |
1da177e4 | 2059 | */ |
47c27bc4 | 2060 | static ssize_t generic_file_buffered_read(struct kiocb *iocb, |
6e58e79d | 2061 | struct iov_iter *iter, ssize_t written) |
1da177e4 | 2062 | { |
47c27bc4 | 2063 | struct file *filp = iocb->ki_filp; |
36e78914 | 2064 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2065 | struct inode *inode = mapping->host; |
36e78914 | 2066 | struct file_ra_state *ra = &filp->f_ra; |
47c27bc4 | 2067 | loff_t *ppos = &iocb->ki_pos; |
57f6b96c FW |
2068 | pgoff_t index; |
2069 | pgoff_t last_index; | |
2070 | pgoff_t prev_index; | |
2071 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 2072 | unsigned int prev_offset; |
6e58e79d | 2073 | int error = 0; |
1da177e4 | 2074 | |
c2a9737f | 2075 | if (unlikely(*ppos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2076 | return 0; |
c2a9737f WF |
2077 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
2078 | ||
09cbfeaf KS |
2079 | index = *ppos >> PAGE_SHIFT; |
2080 | prev_index = ra->prev_pos >> PAGE_SHIFT; | |
2081 | prev_offset = ra->prev_pos & (PAGE_SIZE-1); | |
2082 | last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; | |
2083 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 | 2084 | |
1da177e4 LT |
2085 | for (;;) { |
2086 | struct page *page; | |
57f6b96c | 2087 | pgoff_t end_index; |
a32ea1e1 | 2088 | loff_t isize; |
1da177e4 LT |
2089 | unsigned long nr, ret; |
2090 | ||
1da177e4 | 2091 | cond_resched(); |
1da177e4 | 2092 | find_page: |
5abf186a MH |
2093 | if (fatal_signal_pending(current)) { |
2094 | error = -EINTR; | |
2095 | goto out; | |
2096 | } | |
2097 | ||
1da177e4 | 2098 | page = find_get_page(mapping, index); |
3ea89ee8 | 2099 | if (!page) { |
3239d834 MT |
2100 | if (iocb->ki_flags & IOCB_NOWAIT) |
2101 | goto would_block; | |
cf914a7d | 2102 | page_cache_sync_readahead(mapping, |
7ff81078 | 2103 | ra, filp, |
3ea89ee8 FW |
2104 | index, last_index - index); |
2105 | page = find_get_page(mapping, index); | |
2106 | if (unlikely(page == NULL)) | |
2107 | goto no_cached_page; | |
2108 | } | |
2109 | if (PageReadahead(page)) { | |
cf914a7d | 2110 | page_cache_async_readahead(mapping, |
7ff81078 | 2111 | ra, filp, page, |
3ea89ee8 | 2112 | index, last_index - index); |
1da177e4 | 2113 | } |
8ab22b9a | 2114 | if (!PageUptodate(page)) { |
3239d834 MT |
2115 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2116 | put_page(page); | |
2117 | goto would_block; | |
2118 | } | |
2119 | ||
ebded027 MG |
2120 | /* |
2121 | * See comment in do_read_cache_page on why | |
2122 | * wait_on_page_locked is used to avoid unnecessarily | |
2123 | * serialisations and why it's safe. | |
2124 | */ | |
c4b209a4 BVA |
2125 | error = wait_on_page_locked_killable(page); |
2126 | if (unlikely(error)) | |
2127 | goto readpage_error; | |
ebded027 MG |
2128 | if (PageUptodate(page)) |
2129 | goto page_ok; | |
2130 | ||
09cbfeaf | 2131 | if (inode->i_blkbits == PAGE_SHIFT || |
8ab22b9a HH |
2132 | !mapping->a_ops->is_partially_uptodate) |
2133 | goto page_not_up_to_date; | |
6d6d36bc | 2134 | /* pipes can't handle partially uptodate pages */ |
00e23707 | 2135 | if (unlikely(iov_iter_is_pipe(iter))) |
6d6d36bc | 2136 | goto page_not_up_to_date; |
529ae9aa | 2137 | if (!trylock_page(page)) |
8ab22b9a | 2138 | goto page_not_up_to_date; |
8d056cb9 DH |
2139 | /* Did it get truncated before we got the lock? */ |
2140 | if (!page->mapping) | |
2141 | goto page_not_up_to_date_locked; | |
8ab22b9a | 2142 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 2143 | offset, iter->count)) |
8ab22b9a HH |
2144 | goto page_not_up_to_date_locked; |
2145 | unlock_page(page); | |
2146 | } | |
1da177e4 | 2147 | page_ok: |
a32ea1e1 N |
2148 | /* |
2149 | * i_size must be checked after we know the page is Uptodate. | |
2150 | * | |
2151 | * Checking i_size after the check allows us to calculate | |
2152 | * the correct value for "nr", which means the zero-filled | |
2153 | * part of the page is not copied back to userspace (unless | |
2154 | * another truncate extends the file - this is desired though). | |
2155 | */ | |
2156 | ||
2157 | isize = i_size_read(inode); | |
09cbfeaf | 2158 | end_index = (isize - 1) >> PAGE_SHIFT; |
a32ea1e1 | 2159 | if (unlikely(!isize || index > end_index)) { |
09cbfeaf | 2160 | put_page(page); |
a32ea1e1 N |
2161 | goto out; |
2162 | } | |
2163 | ||
2164 | /* nr is the maximum number of bytes to copy from this page */ | |
09cbfeaf | 2165 | nr = PAGE_SIZE; |
a32ea1e1 | 2166 | if (index == end_index) { |
09cbfeaf | 2167 | nr = ((isize - 1) & ~PAGE_MASK) + 1; |
a32ea1e1 | 2168 | if (nr <= offset) { |
09cbfeaf | 2169 | put_page(page); |
a32ea1e1 N |
2170 | goto out; |
2171 | } | |
2172 | } | |
2173 | nr = nr - offset; | |
1da177e4 LT |
2174 | |
2175 | /* If users can be writing to this page using arbitrary | |
2176 | * virtual addresses, take care about potential aliasing | |
2177 | * before reading the page on the kernel side. | |
2178 | */ | |
2179 | if (mapping_writably_mapped(mapping)) | |
2180 | flush_dcache_page(page); | |
2181 | ||
2182 | /* | |
ec0f1637 JK |
2183 | * When a sequential read accesses a page several times, |
2184 | * only mark it as accessed the first time. | |
1da177e4 | 2185 | */ |
ec0f1637 | 2186 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
2187 | mark_page_accessed(page); |
2188 | prev_index = index; | |
2189 | ||
2190 | /* | |
2191 | * Ok, we have the page, and it's up-to-date, so | |
2192 | * now we can copy it to user space... | |
1da177e4 | 2193 | */ |
6e58e79d AV |
2194 | |
2195 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 | 2196 | offset += ret; |
09cbfeaf KS |
2197 | index += offset >> PAGE_SHIFT; |
2198 | offset &= ~PAGE_MASK; | |
6ce745ed | 2199 | prev_offset = offset; |
1da177e4 | 2200 | |
09cbfeaf | 2201 | put_page(page); |
6e58e79d AV |
2202 | written += ret; |
2203 | if (!iov_iter_count(iter)) | |
2204 | goto out; | |
2205 | if (ret < nr) { | |
2206 | error = -EFAULT; | |
2207 | goto out; | |
2208 | } | |
2209 | continue; | |
1da177e4 LT |
2210 | |
2211 | page_not_up_to_date: | |
2212 | /* Get exclusive access to the page ... */ | |
85462323 ON |
2213 | error = lock_page_killable(page); |
2214 | if (unlikely(error)) | |
2215 | goto readpage_error; | |
1da177e4 | 2216 | |
8ab22b9a | 2217 | page_not_up_to_date_locked: |
da6052f7 | 2218 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
2219 | if (!page->mapping) { |
2220 | unlock_page(page); | |
09cbfeaf | 2221 | put_page(page); |
1da177e4 LT |
2222 | continue; |
2223 | } | |
2224 | ||
2225 | /* Did somebody else fill it already? */ | |
2226 | if (PageUptodate(page)) { | |
2227 | unlock_page(page); | |
2228 | goto page_ok; | |
2229 | } | |
2230 | ||
2231 | readpage: | |
91803b49 JM |
2232 | /* |
2233 | * A previous I/O error may have been due to temporary | |
2234 | * failures, eg. multipath errors. | |
2235 | * PG_error will be set again if readpage fails. | |
2236 | */ | |
2237 | ClearPageError(page); | |
1da177e4 LT |
2238 | /* Start the actual read. The read will unlock the page. */ |
2239 | error = mapping->a_ops->readpage(filp, page); | |
2240 | ||
994fc28c ZB |
2241 | if (unlikely(error)) { |
2242 | if (error == AOP_TRUNCATED_PAGE) { | |
09cbfeaf | 2243 | put_page(page); |
6e58e79d | 2244 | error = 0; |
994fc28c ZB |
2245 | goto find_page; |
2246 | } | |
1da177e4 | 2247 | goto readpage_error; |
994fc28c | 2248 | } |
1da177e4 LT |
2249 | |
2250 | if (!PageUptodate(page)) { | |
85462323 ON |
2251 | error = lock_page_killable(page); |
2252 | if (unlikely(error)) | |
2253 | goto readpage_error; | |
1da177e4 LT |
2254 | if (!PageUptodate(page)) { |
2255 | if (page->mapping == NULL) { | |
2256 | /* | |
2ecdc82e | 2257 | * invalidate_mapping_pages got it |
1da177e4 LT |
2258 | */ |
2259 | unlock_page(page); | |
09cbfeaf | 2260 | put_page(page); |
1da177e4 LT |
2261 | goto find_page; |
2262 | } | |
2263 | unlock_page(page); | |
7ff81078 | 2264 | shrink_readahead_size_eio(filp, ra); |
85462323 ON |
2265 | error = -EIO; |
2266 | goto readpage_error; | |
1da177e4 LT |
2267 | } |
2268 | unlock_page(page); | |
2269 | } | |
2270 | ||
1da177e4 LT |
2271 | goto page_ok; |
2272 | ||
2273 | readpage_error: | |
2274 | /* UHHUH! A synchronous read error occurred. Report it */ | |
09cbfeaf | 2275 | put_page(page); |
1da177e4 LT |
2276 | goto out; |
2277 | ||
2278 | no_cached_page: | |
2279 | /* | |
2280 | * Ok, it wasn't cached, so we need to create a new | |
2281 | * page.. | |
2282 | */ | |
453f85d4 | 2283 | page = page_cache_alloc(mapping); |
eb2be189 | 2284 | if (!page) { |
6e58e79d | 2285 | error = -ENOMEM; |
eb2be189 | 2286 | goto out; |
1da177e4 | 2287 | } |
6afdb859 | 2288 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 2289 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 2290 | if (error) { |
09cbfeaf | 2291 | put_page(page); |
6e58e79d AV |
2292 | if (error == -EEXIST) { |
2293 | error = 0; | |
1da177e4 | 2294 | goto find_page; |
6e58e79d | 2295 | } |
1da177e4 LT |
2296 | goto out; |
2297 | } | |
1da177e4 LT |
2298 | goto readpage; |
2299 | } | |
2300 | ||
3239d834 MT |
2301 | would_block: |
2302 | error = -EAGAIN; | |
1da177e4 | 2303 | out: |
7ff81078 | 2304 | ra->prev_pos = prev_index; |
09cbfeaf | 2305 | ra->prev_pos <<= PAGE_SHIFT; |
7ff81078 | 2306 | ra->prev_pos |= prev_offset; |
1da177e4 | 2307 | |
09cbfeaf | 2308 | *ppos = ((loff_t)index << PAGE_SHIFT) + offset; |
0c6aa263 | 2309 | file_accessed(filp); |
6e58e79d | 2310 | return written ? written : error; |
1da177e4 LT |
2311 | } |
2312 | ||
485bb99b | 2313 | /** |
6abd2322 | 2314 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2315 | * @iocb: kernel I/O control block |
6abd2322 | 2316 | * @iter: destination for the data read |
485bb99b | 2317 | * |
6abd2322 | 2318 | * This is the "read_iter()" routine for all filesystems |
1da177e4 | 2319 | * that can use the page cache directly. |
a862f68a MR |
2320 | * Return: |
2321 | * * number of bytes copied, even for partial reads | |
2322 | * * negative error code if nothing was read | |
1da177e4 LT |
2323 | */ |
2324 | ssize_t | |
ed978a81 | 2325 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2326 | { |
e7080a43 | 2327 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2328 | ssize_t retval = 0; |
e7080a43 NS |
2329 | |
2330 | if (!count) | |
2331 | goto out; /* skip atime */ | |
1da177e4 | 2332 | |
2ba48ce5 | 2333 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2334 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2335 | struct address_space *mapping = file->f_mapping; |
2336 | struct inode *inode = mapping->host; | |
543ade1f | 2337 | loff_t size; |
1da177e4 | 2338 | |
1da177e4 | 2339 | size = i_size_read(inode); |
6be96d3a GR |
2340 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2341 | if (filemap_range_has_page(mapping, iocb->ki_pos, | |
2342 | iocb->ki_pos + count - 1)) | |
2343 | return -EAGAIN; | |
2344 | } else { | |
2345 | retval = filemap_write_and_wait_range(mapping, | |
2346 | iocb->ki_pos, | |
2347 | iocb->ki_pos + count - 1); | |
2348 | if (retval < 0) | |
2349 | goto out; | |
2350 | } | |
d8d3d94b | 2351 | |
0d5b0cf2 CH |
2352 | file_accessed(file); |
2353 | ||
5ecda137 | 2354 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2355 | if (retval >= 0) { |
c64fb5c7 | 2356 | iocb->ki_pos += retval; |
5ecda137 | 2357 | count -= retval; |
9fe55eea | 2358 | } |
5b47d59a | 2359 | iov_iter_revert(iter, count - iov_iter_count(iter)); |
66f998f6 | 2360 | |
9fe55eea SW |
2361 | /* |
2362 | * Btrfs can have a short DIO read if we encounter | |
2363 | * compressed extents, so if there was an error, or if | |
2364 | * we've already read everything we wanted to, or if | |
2365 | * there was a short read because we hit EOF, go ahead | |
2366 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2367 | * the rest of the read. Buffered reads will not work for |
2368 | * DAX files, so don't bother trying. | |
9fe55eea | 2369 | */ |
5ecda137 | 2370 | if (retval < 0 || !count || iocb->ki_pos >= size || |
0d5b0cf2 | 2371 | IS_DAX(inode)) |
9fe55eea | 2372 | goto out; |
1da177e4 LT |
2373 | } |
2374 | ||
47c27bc4 | 2375 | retval = generic_file_buffered_read(iocb, iter, retval); |
1da177e4 LT |
2376 | out: |
2377 | return retval; | |
2378 | } | |
ed978a81 | 2379 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2380 | |
1da177e4 | 2381 | #ifdef CONFIG_MMU |
485bb99b RD |
2382 | /** |
2383 | * page_cache_read - adds requested page to the page cache if not already there | |
2384 | * @file: file to read | |
2385 | * @offset: page index | |
62eb320a | 2386 | * @gfp_mask: memory allocation flags |
485bb99b | 2387 | * |
1da177e4 LT |
2388 | * This adds the requested page to the page cache if it isn't already there, |
2389 | * and schedules an I/O to read in its contents from disk. | |
a862f68a MR |
2390 | * |
2391 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 2392 | */ |
c20cd45e | 2393 | static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask) |
1da177e4 LT |
2394 | { |
2395 | struct address_space *mapping = file->f_mapping; | |
99dadfdd | 2396 | struct page *page; |
994fc28c | 2397 | int ret; |
1da177e4 | 2398 | |
994fc28c | 2399 | do { |
453f85d4 | 2400 | page = __page_cache_alloc(gfp_mask); |
994fc28c ZB |
2401 | if (!page) |
2402 | return -ENOMEM; | |
2403 | ||
abc1be13 | 2404 | ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask); |
994fc28c ZB |
2405 | if (ret == 0) |
2406 | ret = mapping->a_ops->readpage(file, page); | |
2407 | else if (ret == -EEXIST) | |
2408 | ret = 0; /* losing race to add is OK */ | |
1da177e4 | 2409 | |
09cbfeaf | 2410 | put_page(page); |
1da177e4 | 2411 | |
994fc28c | 2412 | } while (ret == AOP_TRUNCATED_PAGE); |
99dadfdd | 2413 | |
994fc28c | 2414 | return ret; |
1da177e4 LT |
2415 | } |
2416 | ||
2417 | #define MMAP_LOTSAMISS (100) | |
2418 | ||
ef00e08e LT |
2419 | /* |
2420 | * Synchronous readahead happens when we don't even find | |
2421 | * a page in the page cache at all. | |
2422 | */ | |
2423 | static void do_sync_mmap_readahead(struct vm_area_struct *vma, | |
2424 | struct file_ra_state *ra, | |
2425 | struct file *file, | |
2426 | pgoff_t offset) | |
2427 | { | |
ef00e08e LT |
2428 | struct address_space *mapping = file->f_mapping; |
2429 | ||
2430 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 2431 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e | 2432 | return; |
275b12bf WF |
2433 | if (!ra->ra_pages) |
2434 | return; | |
ef00e08e | 2435 | |
64363aad | 2436 | if (vma->vm_flags & VM_SEQ_READ) { |
7ffc59b4 WF |
2437 | page_cache_sync_readahead(mapping, ra, file, offset, |
2438 | ra->ra_pages); | |
ef00e08e LT |
2439 | return; |
2440 | } | |
2441 | ||
207d04ba AK |
2442 | /* Avoid banging the cache line if not needed */ |
2443 | if (ra->mmap_miss < MMAP_LOTSAMISS * 10) | |
ef00e08e LT |
2444 | ra->mmap_miss++; |
2445 | ||
2446 | /* | |
2447 | * Do we miss much more than hit in this file? If so, | |
2448 | * stop bothering with read-ahead. It will only hurt. | |
2449 | */ | |
2450 | if (ra->mmap_miss > MMAP_LOTSAMISS) | |
2451 | return; | |
2452 | ||
d30a1100 WF |
2453 | /* |
2454 | * mmap read-around | |
2455 | */ | |
600e19af RG |
2456 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
2457 | ra->size = ra->ra_pages; | |
2458 | ra->async_size = ra->ra_pages / 4; | |
275b12bf | 2459 | ra_submit(ra, mapping, file); |
ef00e08e LT |
2460 | } |
2461 | ||
2462 | /* | |
2463 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
2464 | * so we want to possibly extend the readahead further.. | |
2465 | */ | |
2466 | static void do_async_mmap_readahead(struct vm_area_struct *vma, | |
2467 | struct file_ra_state *ra, | |
2468 | struct file *file, | |
2469 | struct page *page, | |
2470 | pgoff_t offset) | |
2471 | { | |
2472 | struct address_space *mapping = file->f_mapping; | |
2473 | ||
2474 | /* If we don't want any read-ahead, don't bother */ | |
64363aad | 2475 | if (vma->vm_flags & VM_RAND_READ) |
ef00e08e LT |
2476 | return; |
2477 | if (ra->mmap_miss > 0) | |
2478 | ra->mmap_miss--; | |
2479 | if (PageReadahead(page)) | |
2fad6f5d WF |
2480 | page_cache_async_readahead(mapping, ra, file, |
2481 | page, offset, ra->ra_pages); | |
ef00e08e LT |
2482 | } |
2483 | ||
485bb99b | 2484 | /** |
54cb8821 | 2485 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 2486 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 2487 | * |
54cb8821 | 2488 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2489 | * mapped memory region to read in file data during a page fault. |
2490 | * | |
2491 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2492 | * it in the page cache, and handles the special cases reasonably without | |
2493 | * having a lot of duplicated code. | |
9a95f3cf PC |
2494 | * |
2495 | * vma->vm_mm->mmap_sem must be held on entry. | |
2496 | * | |
2497 | * If our return value has VM_FAULT_RETRY set, it's because | |
2498 | * lock_page_or_retry() returned 0. | |
2499 | * The mmap_sem has usually been released in this case. | |
2500 | * See __lock_page_or_retry() for the exception. | |
2501 | * | |
2502 | * If our return value does not have VM_FAULT_RETRY set, the mmap_sem | |
2503 | * has not been released. | |
2504 | * | |
2505 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
a862f68a MR |
2506 | * |
2507 | * Return: bitwise-OR of %VM_FAULT_ codes. | |
1da177e4 | 2508 | */ |
2bcd6454 | 2509 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
2510 | { |
2511 | int error; | |
11bac800 | 2512 | struct file *file = vmf->vma->vm_file; |
1da177e4 LT |
2513 | struct address_space *mapping = file->f_mapping; |
2514 | struct file_ra_state *ra = &file->f_ra; | |
2515 | struct inode *inode = mapping->host; | |
ef00e08e | 2516 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 2517 | pgoff_t max_off; |
1da177e4 | 2518 | struct page *page; |
2bcd6454 | 2519 | vm_fault_t ret = 0; |
1da177e4 | 2520 | |
9ab2594f MW |
2521 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2522 | if (unlikely(offset >= max_off)) | |
5307cc1a | 2523 | return VM_FAULT_SIGBUS; |
1da177e4 | 2524 | |
1da177e4 | 2525 | /* |
49426420 | 2526 | * Do we have something in the page cache already? |
1da177e4 | 2527 | */ |
ef00e08e | 2528 | page = find_get_page(mapping, offset); |
45cac65b | 2529 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2530 | /* |
ef00e08e LT |
2531 | * We found the page, so try async readahead before |
2532 | * waiting for the lock. | |
1da177e4 | 2533 | */ |
11bac800 | 2534 | do_async_mmap_readahead(vmf->vma, ra, file, page, offset); |
45cac65b | 2535 | } else if (!page) { |
ef00e08e | 2536 | /* No page in the page cache at all */ |
11bac800 | 2537 | do_sync_mmap_readahead(vmf->vma, ra, file, offset); |
ef00e08e | 2538 | count_vm_event(PGMAJFAULT); |
2262185c | 2539 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e LT |
2540 | ret = VM_FAULT_MAJOR; |
2541 | retry_find: | |
b522c94d | 2542 | page = find_get_page(mapping, offset); |
1da177e4 LT |
2543 | if (!page) |
2544 | goto no_cached_page; | |
2545 | } | |
2546 | ||
11bac800 | 2547 | if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) { |
09cbfeaf | 2548 | put_page(page); |
d065bd81 | 2549 | return ret | VM_FAULT_RETRY; |
d88c0922 | 2550 | } |
b522c94d ML |
2551 | |
2552 | /* Did it get truncated? */ | |
2553 | if (unlikely(page->mapping != mapping)) { | |
2554 | unlock_page(page); | |
2555 | put_page(page); | |
2556 | goto retry_find; | |
2557 | } | |
309381fe | 2558 | VM_BUG_ON_PAGE(page->index != offset, page); |
b522c94d | 2559 | |
1da177e4 | 2560 | /* |
d00806b1 NP |
2561 | * We have a locked page in the page cache, now we need to check |
2562 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2563 | */ |
d00806b1 | 2564 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2565 | goto page_not_uptodate; |
2566 | ||
ef00e08e LT |
2567 | /* |
2568 | * Found the page and have a reference on it. | |
2569 | * We must recheck i_size under page lock. | |
2570 | */ | |
9ab2594f MW |
2571 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2572 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 2573 | unlock_page(page); |
09cbfeaf | 2574 | put_page(page); |
5307cc1a | 2575 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2576 | } |
2577 | ||
d0217ac0 | 2578 | vmf->page = page; |
83c54070 | 2579 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2580 | |
1da177e4 LT |
2581 | no_cached_page: |
2582 | /* | |
2583 | * We're only likely to ever get here if MADV_RANDOM is in | |
2584 | * effect. | |
2585 | */ | |
c20cd45e | 2586 | error = page_cache_read(file, offset, vmf->gfp_mask); |
1da177e4 LT |
2587 | |
2588 | /* | |
2589 | * The page we want has now been added to the page cache. | |
2590 | * In the unlikely event that someone removed it in the | |
2591 | * meantime, we'll just come back here and read it again. | |
2592 | */ | |
2593 | if (error >= 0) | |
2594 | goto retry_find; | |
2595 | ||
2596 | /* | |
2597 | * An error return from page_cache_read can result if the | |
2598 | * system is low on memory, or a problem occurs while trying | |
2599 | * to schedule I/O. | |
2600 | */ | |
3c051324 | 2601 | return vmf_error(error); |
1da177e4 LT |
2602 | |
2603 | page_not_uptodate: | |
1da177e4 LT |
2604 | /* |
2605 | * Umm, take care of errors if the page isn't up-to-date. | |
2606 | * Try to re-read it _once_. We do this synchronously, | |
2607 | * because there really aren't any performance issues here | |
2608 | * and we need to check for errors. | |
2609 | */ | |
1da177e4 | 2610 | ClearPageError(page); |
994fc28c | 2611 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2612 | if (!error) { |
2613 | wait_on_page_locked(page); | |
2614 | if (!PageUptodate(page)) | |
2615 | error = -EIO; | |
2616 | } | |
09cbfeaf | 2617 | put_page(page); |
d00806b1 NP |
2618 | |
2619 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2620 | goto retry_find; |
1da177e4 | 2621 | |
d00806b1 | 2622 | /* Things didn't work out. Return zero to tell the mm layer so. */ |
76d42bd9 | 2623 | shrink_readahead_size_eio(file, ra); |
d0217ac0 | 2624 | return VM_FAULT_SIGBUS; |
54cb8821 NP |
2625 | } |
2626 | EXPORT_SYMBOL(filemap_fault); | |
2627 | ||
82b0f8c3 | 2628 | void filemap_map_pages(struct vm_fault *vmf, |
bae473a4 | 2629 | pgoff_t start_pgoff, pgoff_t end_pgoff) |
f1820361 | 2630 | { |
82b0f8c3 | 2631 | struct file *file = vmf->vma->vm_file; |
f1820361 | 2632 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 2633 | pgoff_t last_pgoff = start_pgoff; |
9ab2594f | 2634 | unsigned long max_idx; |
070e807c | 2635 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
83929372 | 2636 | struct page *head, *page; |
f1820361 KS |
2637 | |
2638 | rcu_read_lock(); | |
070e807c MW |
2639 | xas_for_each(&xas, page, end_pgoff) { |
2640 | if (xas_retry(&xas, page)) | |
2641 | continue; | |
2642 | if (xa_is_value(page)) | |
2cf938aa | 2643 | goto next; |
f1820361 | 2644 | |
83929372 | 2645 | head = compound_head(page); |
e0975b2a MH |
2646 | |
2647 | /* | |
2648 | * Check for a locked page first, as a speculative | |
2649 | * reference may adversely influence page migration. | |
2650 | */ | |
2651 | if (PageLocked(head)) | |
2652 | goto next; | |
83929372 | 2653 | if (!page_cache_get_speculative(head)) |
070e807c | 2654 | goto next; |
f1820361 | 2655 | |
83929372 | 2656 | /* The page was split under us? */ |
070e807c MW |
2657 | if (compound_head(page) != head) |
2658 | goto skip; | |
83929372 | 2659 | |
f1820361 | 2660 | /* Has the page moved? */ |
070e807c MW |
2661 | if (unlikely(page != xas_reload(&xas))) |
2662 | goto skip; | |
f1820361 KS |
2663 | |
2664 | if (!PageUptodate(page) || | |
2665 | PageReadahead(page) || | |
2666 | PageHWPoison(page)) | |
2667 | goto skip; | |
2668 | if (!trylock_page(page)) | |
2669 | goto skip; | |
2670 | ||
2671 | if (page->mapping != mapping || !PageUptodate(page)) | |
2672 | goto unlock; | |
2673 | ||
9ab2594f MW |
2674 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); |
2675 | if (page->index >= max_idx) | |
f1820361 KS |
2676 | goto unlock; |
2677 | ||
f1820361 KS |
2678 | if (file->f_ra.mmap_miss > 0) |
2679 | file->f_ra.mmap_miss--; | |
7267ec00 | 2680 | |
070e807c | 2681 | vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
82b0f8c3 | 2682 | if (vmf->pte) |
070e807c MW |
2683 | vmf->pte += xas.xa_index - last_pgoff; |
2684 | last_pgoff = xas.xa_index; | |
82b0f8c3 | 2685 | if (alloc_set_pte(vmf, NULL, page)) |
7267ec00 | 2686 | goto unlock; |
f1820361 KS |
2687 | unlock_page(page); |
2688 | goto next; | |
2689 | unlock: | |
2690 | unlock_page(page); | |
2691 | skip: | |
09cbfeaf | 2692 | put_page(page); |
f1820361 | 2693 | next: |
7267ec00 | 2694 | /* Huge page is mapped? No need to proceed. */ |
82b0f8c3 | 2695 | if (pmd_trans_huge(*vmf->pmd)) |
7267ec00 | 2696 | break; |
f1820361 KS |
2697 | } |
2698 | rcu_read_unlock(); | |
2699 | } | |
2700 | EXPORT_SYMBOL(filemap_map_pages); | |
2701 | ||
2bcd6454 | 2702 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 JK |
2703 | { |
2704 | struct page *page = vmf->page; | |
11bac800 | 2705 | struct inode *inode = file_inode(vmf->vma->vm_file); |
2bcd6454 | 2706 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 2707 | |
14da9200 | 2708 | sb_start_pagefault(inode->i_sb); |
11bac800 | 2709 | file_update_time(vmf->vma->vm_file); |
4fcf1c62 JK |
2710 | lock_page(page); |
2711 | if (page->mapping != inode->i_mapping) { | |
2712 | unlock_page(page); | |
2713 | ret = VM_FAULT_NOPAGE; | |
2714 | goto out; | |
2715 | } | |
14da9200 JK |
2716 | /* |
2717 | * We mark the page dirty already here so that when freeze is in | |
2718 | * progress, we are guaranteed that writeback during freezing will | |
2719 | * see the dirty page and writeprotect it again. | |
2720 | */ | |
2721 | set_page_dirty(page); | |
1d1d1a76 | 2722 | wait_for_stable_page(page); |
4fcf1c62 | 2723 | out: |
14da9200 | 2724 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2725 | return ret; |
2726 | } | |
4fcf1c62 | 2727 | |
f0f37e2f | 2728 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2729 | .fault = filemap_fault, |
f1820361 | 2730 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2731 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2732 | }; |
2733 | ||
2734 | /* This is used for a general mmap of a disk file */ | |
2735 | ||
2736 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2737 | { | |
2738 | struct address_space *mapping = file->f_mapping; | |
2739 | ||
2740 | if (!mapping->a_ops->readpage) | |
2741 | return -ENOEXEC; | |
2742 | file_accessed(file); | |
2743 | vma->vm_ops = &generic_file_vm_ops; | |
2744 | return 0; | |
2745 | } | |
1da177e4 LT |
2746 | |
2747 | /* | |
2748 | * This is for filesystems which do not implement ->writepage. | |
2749 | */ | |
2750 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2751 | { | |
2752 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2753 | return -EINVAL; | |
2754 | return generic_file_mmap(file, vma); | |
2755 | } | |
2756 | #else | |
4b96a37d | 2757 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
45397228 | 2758 | { |
4b96a37d | 2759 | return VM_FAULT_SIGBUS; |
45397228 | 2760 | } |
1da177e4 LT |
2761 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) |
2762 | { | |
2763 | return -ENOSYS; | |
2764 | } | |
2765 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2766 | { | |
2767 | return -ENOSYS; | |
2768 | } | |
2769 | #endif /* CONFIG_MMU */ | |
2770 | ||
45397228 | 2771 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
2772 | EXPORT_SYMBOL(generic_file_mmap); |
2773 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2774 | ||
67f9fd91 SL |
2775 | static struct page *wait_on_page_read(struct page *page) |
2776 | { | |
2777 | if (!IS_ERR(page)) { | |
2778 | wait_on_page_locked(page); | |
2779 | if (!PageUptodate(page)) { | |
09cbfeaf | 2780 | put_page(page); |
67f9fd91 SL |
2781 | page = ERR_PTR(-EIO); |
2782 | } | |
2783 | } | |
2784 | return page; | |
2785 | } | |
2786 | ||
32b63529 | 2787 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2788 | pgoff_t index, |
5e5358e7 | 2789 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2790 | void *data, |
2791 | gfp_t gfp) | |
1da177e4 | 2792 | { |
eb2be189 | 2793 | struct page *page; |
1da177e4 LT |
2794 | int err; |
2795 | repeat: | |
2796 | page = find_get_page(mapping, index); | |
2797 | if (!page) { | |
453f85d4 | 2798 | page = __page_cache_alloc(gfp); |
eb2be189 NP |
2799 | if (!page) |
2800 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2801 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 2802 | if (unlikely(err)) { |
09cbfeaf | 2803 | put_page(page); |
eb2be189 NP |
2804 | if (err == -EEXIST) |
2805 | goto repeat; | |
22ecdb4f | 2806 | /* Presumably ENOMEM for xarray node */ |
1da177e4 LT |
2807 | return ERR_PTR(err); |
2808 | } | |
32b63529 MG |
2809 | |
2810 | filler: | |
1da177e4 LT |
2811 | err = filler(data, page); |
2812 | if (err < 0) { | |
09cbfeaf | 2813 | put_page(page); |
32b63529 | 2814 | return ERR_PTR(err); |
1da177e4 | 2815 | } |
1da177e4 | 2816 | |
32b63529 MG |
2817 | page = wait_on_page_read(page); |
2818 | if (IS_ERR(page)) | |
2819 | return page; | |
2820 | goto out; | |
2821 | } | |
1da177e4 LT |
2822 | if (PageUptodate(page)) |
2823 | goto out; | |
2824 | ||
ebded027 MG |
2825 | /* |
2826 | * Page is not up to date and may be locked due one of the following | |
2827 | * case a: Page is being filled and the page lock is held | |
2828 | * case b: Read/write error clearing the page uptodate status | |
2829 | * case c: Truncation in progress (page locked) | |
2830 | * case d: Reclaim in progress | |
2831 | * | |
2832 | * Case a, the page will be up to date when the page is unlocked. | |
2833 | * There is no need to serialise on the page lock here as the page | |
2834 | * is pinned so the lock gives no additional protection. Even if the | |
2835 | * the page is truncated, the data is still valid if PageUptodate as | |
2836 | * it's a race vs truncate race. | |
2837 | * Case b, the page will not be up to date | |
2838 | * Case c, the page may be truncated but in itself, the data may still | |
2839 | * be valid after IO completes as it's a read vs truncate race. The | |
2840 | * operation must restart if the page is not uptodate on unlock but | |
2841 | * otherwise serialising on page lock to stabilise the mapping gives | |
2842 | * no additional guarantees to the caller as the page lock is | |
2843 | * released before return. | |
2844 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
2845 | * will be a race with remove_mapping that determines if the mapping | |
2846 | * is valid on unlock but otherwise the data is valid and there is | |
2847 | * no need to serialise with page lock. | |
2848 | * | |
2849 | * As the page lock gives no additional guarantee, we optimistically | |
2850 | * wait on the page to be unlocked and check if it's up to date and | |
2851 | * use the page if it is. Otherwise, the page lock is required to | |
2852 | * distinguish between the different cases. The motivation is that we | |
2853 | * avoid spurious serialisations and wakeups when multiple processes | |
2854 | * wait on the same page for IO to complete. | |
2855 | */ | |
2856 | wait_on_page_locked(page); | |
2857 | if (PageUptodate(page)) | |
2858 | goto out; | |
2859 | ||
2860 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 2861 | lock_page(page); |
ebded027 MG |
2862 | |
2863 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
2864 | if (!page->mapping) { |
2865 | unlock_page(page); | |
09cbfeaf | 2866 | put_page(page); |
32b63529 | 2867 | goto repeat; |
1da177e4 | 2868 | } |
ebded027 MG |
2869 | |
2870 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
2871 | if (PageUptodate(page)) { |
2872 | unlock_page(page); | |
2873 | goto out; | |
2874 | } | |
32b63529 MG |
2875 | goto filler; |
2876 | ||
c855ff37 | 2877 | out: |
6fe6900e NP |
2878 | mark_page_accessed(page); |
2879 | return page; | |
2880 | } | |
0531b2aa LT |
2881 | |
2882 | /** | |
67f9fd91 | 2883 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
2884 | * @mapping: the page's address_space |
2885 | * @index: the page index | |
2886 | * @filler: function to perform the read | |
5e5358e7 | 2887 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 2888 | * |
0531b2aa | 2889 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 2890 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
2891 | * |
2892 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a MR |
2893 | * |
2894 | * Return: up to date page on success, ERR_PTR() on failure. | |
0531b2aa | 2895 | */ |
67f9fd91 | 2896 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 2897 | pgoff_t index, |
5e5358e7 | 2898 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2899 | void *data) |
2900 | { | |
2901 | return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); | |
2902 | } | |
67f9fd91 | 2903 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
2904 | |
2905 | /** | |
2906 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
2907 | * @mapping: the page's address_space | |
2908 | * @index: the page index | |
2909 | * @gfp: the page allocator flags to use if allocating | |
2910 | * | |
2911 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 2912 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
2913 | * |
2914 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a MR |
2915 | * |
2916 | * Return: up to date page on success, ERR_PTR() on failure. | |
0531b2aa LT |
2917 | */ |
2918 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
2919 | pgoff_t index, | |
2920 | gfp_t gfp) | |
2921 | { | |
2922 | filler_t *filler = (filler_t *)mapping->a_ops->readpage; | |
2923 | ||
67f9fd91 | 2924 | return do_read_cache_page(mapping, index, filler, NULL, gfp); |
0531b2aa LT |
2925 | } |
2926 | EXPORT_SYMBOL(read_cache_page_gfp); | |
2927 | ||
9fd91a90 DW |
2928 | /* |
2929 | * Don't operate on ranges the page cache doesn't support, and don't exceed the | |
2930 | * LFS limits. If pos is under the limit it becomes a short access. If it | |
2931 | * exceeds the limit we return -EFBIG. | |
2932 | */ | |
2933 | static int generic_access_check_limits(struct file *file, loff_t pos, | |
2934 | loff_t *count) | |
2935 | { | |
2936 | struct inode *inode = file->f_mapping->host; | |
2937 | loff_t max_size = inode->i_sb->s_maxbytes; | |
2938 | ||
2939 | if (!(file->f_flags & O_LARGEFILE)) | |
2940 | max_size = MAX_NON_LFS; | |
2941 | ||
2942 | if (unlikely(pos >= max_size)) | |
2943 | return -EFBIG; | |
2944 | *count = min(*count, max_size - pos); | |
2945 | return 0; | |
2946 | } | |
2947 | ||
2948 | static int generic_write_check_limits(struct file *file, loff_t pos, | |
2949 | loff_t *count) | |
2950 | { | |
2951 | loff_t limit = rlimit(RLIMIT_FSIZE); | |
2952 | ||
2953 | if (limit != RLIM_INFINITY) { | |
2954 | if (pos >= limit) { | |
2955 | send_sig(SIGXFSZ, current, 0); | |
2956 | return -EFBIG; | |
2957 | } | |
2958 | *count = min(*count, limit - pos); | |
2959 | } | |
2960 | ||
2961 | return generic_access_check_limits(file, pos, count); | |
2962 | } | |
2963 | ||
1da177e4 LT |
2964 | /* |
2965 | * Performs necessary checks before doing a write | |
2966 | * | |
485bb99b | 2967 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
2968 | * Returns appropriate error code that caller should return or |
2969 | * zero in case that write should be allowed. | |
2970 | */ | |
3309dd04 | 2971 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 2972 | { |
3309dd04 | 2973 | struct file *file = iocb->ki_filp; |
1da177e4 | 2974 | struct inode *inode = file->f_mapping->host; |
9fd91a90 DW |
2975 | loff_t count; |
2976 | int ret; | |
1da177e4 | 2977 | |
3309dd04 AV |
2978 | if (!iov_iter_count(from)) |
2979 | return 0; | |
1da177e4 | 2980 | |
0fa6b005 | 2981 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 2982 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 2983 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 2984 | |
6be96d3a GR |
2985 | if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT)) |
2986 | return -EINVAL; | |
2987 | ||
9fd91a90 DW |
2988 | count = iov_iter_count(from); |
2989 | ret = generic_write_check_limits(file, iocb->ki_pos, &count); | |
2990 | if (ret) | |
2991 | return ret; | |
1da177e4 | 2992 | |
9fd91a90 | 2993 | iov_iter_truncate(from, count); |
3309dd04 | 2994 | return iov_iter_count(from); |
1da177e4 LT |
2995 | } |
2996 | EXPORT_SYMBOL(generic_write_checks); | |
2997 | ||
1383a7ed DW |
2998 | /* |
2999 | * Performs necessary checks before doing a clone. | |
3000 | * | |
3001 | * Can adjust amount of bytes to clone. | |
3002 | * Returns appropriate error code that caller should return or | |
3003 | * zero in case the clone should be allowed. | |
3004 | */ | |
3005 | int generic_remap_checks(struct file *file_in, loff_t pos_in, | |
3006 | struct file *file_out, loff_t pos_out, | |
42ec3d4c | 3007 | loff_t *req_count, unsigned int remap_flags) |
1383a7ed DW |
3008 | { |
3009 | struct inode *inode_in = file_in->f_mapping->host; | |
3010 | struct inode *inode_out = file_out->f_mapping->host; | |
3011 | uint64_t count = *req_count; | |
3012 | uint64_t bcount; | |
3013 | loff_t size_in, size_out; | |
3014 | loff_t bs = inode_out->i_sb->s_blocksize; | |
9fd91a90 | 3015 | int ret; |
1383a7ed DW |
3016 | |
3017 | /* The start of both ranges must be aligned to an fs block. */ | |
3018 | if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs)) | |
3019 | return -EINVAL; | |
3020 | ||
3021 | /* Ensure offsets don't wrap. */ | |
3022 | if (pos_in + count < pos_in || pos_out + count < pos_out) | |
3023 | return -EINVAL; | |
3024 | ||
3025 | size_in = i_size_read(inode_in); | |
3026 | size_out = i_size_read(inode_out); | |
3027 | ||
3028 | /* Dedupe requires both ranges to be within EOF. */ | |
3d28193e | 3029 | if ((remap_flags & REMAP_FILE_DEDUP) && |
1383a7ed DW |
3030 | (pos_in >= size_in || pos_in + count > size_in || |
3031 | pos_out >= size_out || pos_out + count > size_out)) | |
3032 | return -EINVAL; | |
3033 | ||
3034 | /* Ensure the infile range is within the infile. */ | |
3035 | if (pos_in >= size_in) | |
3036 | return -EINVAL; | |
3037 | count = min(count, size_in - (uint64_t)pos_in); | |
3038 | ||
9fd91a90 DW |
3039 | ret = generic_access_check_limits(file_in, pos_in, &count); |
3040 | if (ret) | |
3041 | return ret; | |
3042 | ||
3043 | ret = generic_write_check_limits(file_out, pos_out, &count); | |
3044 | if (ret) | |
3045 | return ret; | |
1da177e4 LT |
3046 | |
3047 | /* | |
1383a7ed DW |
3048 | * If the user wanted us to link to the infile's EOF, round up to the |
3049 | * next block boundary for this check. | |
3050 | * | |
3051 | * Otherwise, make sure the count is also block-aligned, having | |
3052 | * already confirmed the starting offsets' block alignment. | |
1da177e4 | 3053 | */ |
1383a7ed DW |
3054 | if (pos_in + count == size_in) { |
3055 | bcount = ALIGN(size_in, bs) - pos_in; | |
3056 | } else { | |
3057 | if (!IS_ALIGNED(count, bs)) | |
eca3654e | 3058 | count = ALIGN_DOWN(count, bs); |
1383a7ed | 3059 | bcount = count; |
1da177e4 LT |
3060 | } |
3061 | ||
1383a7ed DW |
3062 | /* Don't allow overlapped cloning within the same file. */ |
3063 | if (inode_in == inode_out && | |
3064 | pos_out + bcount > pos_in && | |
3065 | pos_out < pos_in + bcount) | |
3066 | return -EINVAL; | |
3067 | ||
1da177e4 | 3068 | /* |
eca3654e DW |
3069 | * We shortened the request but the caller can't deal with that, so |
3070 | * bounce the request back to userspace. | |
1da177e4 | 3071 | */ |
eca3654e | 3072 | if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN)) |
1383a7ed | 3073 | return -EINVAL; |
1da177e4 | 3074 | |
eca3654e | 3075 | *req_count = count; |
1383a7ed | 3076 | return 0; |
1da177e4 | 3077 | } |
1da177e4 | 3078 | |
afddba49 NP |
3079 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
3080 | loff_t pos, unsigned len, unsigned flags, | |
3081 | struct page **pagep, void **fsdata) | |
3082 | { | |
3083 | const struct address_space_operations *aops = mapping->a_ops; | |
3084 | ||
4e02ed4b | 3085 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 3086 | pagep, fsdata); |
afddba49 NP |
3087 | } |
3088 | EXPORT_SYMBOL(pagecache_write_begin); | |
3089 | ||
3090 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
3091 | loff_t pos, unsigned len, unsigned copied, | |
3092 | struct page *page, void *fsdata) | |
3093 | { | |
3094 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 3095 | |
4e02ed4b | 3096 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
3097 | } |
3098 | EXPORT_SYMBOL(pagecache_write_end); | |
3099 | ||
1da177e4 | 3100 | ssize_t |
1af5bb49 | 3101 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3102 | { |
3103 | struct file *file = iocb->ki_filp; | |
3104 | struct address_space *mapping = file->f_mapping; | |
3105 | struct inode *inode = mapping->host; | |
1af5bb49 | 3106 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3107 | ssize_t written; |
a969e903 CH |
3108 | size_t write_len; |
3109 | pgoff_t end; | |
1da177e4 | 3110 | |
0c949334 | 3111 | write_len = iov_iter_count(from); |
09cbfeaf | 3112 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3113 | |
6be96d3a GR |
3114 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3115 | /* If there are pages to writeback, return */ | |
3116 | if (filemap_range_has_page(inode->i_mapping, pos, | |
35f12f0f | 3117 | pos + write_len - 1)) |
6be96d3a GR |
3118 | return -EAGAIN; |
3119 | } else { | |
3120 | written = filemap_write_and_wait_range(mapping, pos, | |
3121 | pos + write_len - 1); | |
3122 | if (written) | |
3123 | goto out; | |
3124 | } | |
a969e903 CH |
3125 | |
3126 | /* | |
3127 | * After a write we want buffered reads to be sure to go to disk to get | |
3128 | * the new data. We invalidate clean cached page from the region we're | |
3129 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3130 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3131 | */ |
55635ba7 | 3132 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3133 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3134 | /* |
3135 | * If a page can not be invalidated, return 0 to fall back | |
3136 | * to buffered write. | |
3137 | */ | |
3138 | if (written) { | |
3139 | if (written == -EBUSY) | |
3140 | return 0; | |
3141 | goto out; | |
a969e903 CH |
3142 | } |
3143 | ||
639a93a5 | 3144 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3145 | |
3146 | /* | |
3147 | * Finally, try again to invalidate clean pages which might have been | |
3148 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3149 | * if the source of the write was an mmap'ed region of the file | |
3150 | * we're writing. Either one is a pretty crazy thing to do, | |
3151 | * so we don't support it 100%. If this invalidation | |
3152 | * fails, tough, the write still worked... | |
332391a9 LC |
3153 | * |
3154 | * Most of the time we do not need this since dio_complete() will do | |
3155 | * the invalidation for us. However there are some file systems that | |
3156 | * do not end up with dio_complete() being called, so let's not break | |
3157 | * them by removing it completely | |
a969e903 | 3158 | */ |
332391a9 LC |
3159 | if (mapping->nrpages) |
3160 | invalidate_inode_pages2_range(mapping, | |
3161 | pos >> PAGE_SHIFT, end); | |
a969e903 | 3162 | |
1da177e4 | 3163 | if (written > 0) { |
0116651c | 3164 | pos += written; |
639a93a5 | 3165 | write_len -= written; |
0116651c NK |
3166 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3167 | i_size_write(inode, pos); | |
1da177e4 LT |
3168 | mark_inode_dirty(inode); |
3169 | } | |
5cb6c6c7 | 3170 | iocb->ki_pos = pos; |
1da177e4 | 3171 | } |
639a93a5 | 3172 | iov_iter_revert(from, write_len - iov_iter_count(from)); |
a969e903 | 3173 | out: |
1da177e4 LT |
3174 | return written; |
3175 | } | |
3176 | EXPORT_SYMBOL(generic_file_direct_write); | |
3177 | ||
eb2be189 NP |
3178 | /* |
3179 | * Find or create a page at the given pagecache position. Return the locked | |
3180 | * page. This function is specifically for buffered writes. | |
3181 | */ | |
54566b2c NP |
3182 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
3183 | pgoff_t index, unsigned flags) | |
eb2be189 | 3184 | { |
eb2be189 | 3185 | struct page *page; |
bbddabe2 | 3186 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 3187 | |
54566b2c | 3188 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
3189 | fgp_flags |= FGP_NOFS; |
3190 | ||
3191 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 3192 | mapping_gfp_mask(mapping)); |
c585a267 | 3193 | if (page) |
2457aec6 | 3194 | wait_for_stable_page(page); |
eb2be189 | 3195 | |
eb2be189 NP |
3196 | return page; |
3197 | } | |
54566b2c | 3198 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 3199 | |
3b93f911 | 3200 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3201 | struct iov_iter *i, loff_t pos) |
3202 | { | |
3203 | struct address_space *mapping = file->f_mapping; | |
3204 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3205 | long status = 0; | |
3206 | ssize_t written = 0; | |
674b892e NP |
3207 | unsigned int flags = 0; |
3208 | ||
afddba49 NP |
3209 | do { |
3210 | struct page *page; | |
afddba49 NP |
3211 | unsigned long offset; /* Offset into pagecache page */ |
3212 | unsigned long bytes; /* Bytes to write to page */ | |
3213 | size_t copied; /* Bytes copied from user */ | |
3214 | void *fsdata; | |
3215 | ||
09cbfeaf KS |
3216 | offset = (pos & (PAGE_SIZE - 1)); |
3217 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3218 | iov_iter_count(i)); |
3219 | ||
3220 | again: | |
00a3d660 LT |
3221 | /* |
3222 | * Bring in the user page that we will copy from _first_. | |
3223 | * Otherwise there's a nasty deadlock on copying from the | |
3224 | * same page as we're writing to, without it being marked | |
3225 | * up-to-date. | |
3226 | * | |
3227 | * Not only is this an optimisation, but it is also required | |
3228 | * to check that the address is actually valid, when atomic | |
3229 | * usercopies are used, below. | |
3230 | */ | |
3231 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
3232 | status = -EFAULT; | |
3233 | break; | |
3234 | } | |
3235 | ||
296291cd JK |
3236 | if (fatal_signal_pending(current)) { |
3237 | status = -EINTR; | |
3238 | break; | |
3239 | } | |
3240 | ||
674b892e | 3241 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3242 | &page, &fsdata); |
2457aec6 | 3243 | if (unlikely(status < 0)) |
afddba49 NP |
3244 | break; |
3245 | ||
931e80e4 | 3246 | if (mapping_writably_mapped(mapping)) |
3247 | flush_dcache_page(page); | |
00a3d660 | 3248 | |
afddba49 | 3249 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
3250 | flush_dcache_page(page); |
3251 | ||
3252 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3253 | page, fsdata); | |
3254 | if (unlikely(status < 0)) | |
3255 | break; | |
3256 | copied = status; | |
3257 | ||
3258 | cond_resched(); | |
3259 | ||
124d3b70 | 3260 | iov_iter_advance(i, copied); |
afddba49 NP |
3261 | if (unlikely(copied == 0)) { |
3262 | /* | |
3263 | * If we were unable to copy any data at all, we must | |
3264 | * fall back to a single segment length write. | |
3265 | * | |
3266 | * If we didn't fallback here, we could livelock | |
3267 | * because not all segments in the iov can be copied at | |
3268 | * once without a pagefault. | |
3269 | */ | |
09cbfeaf | 3270 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
afddba49 NP |
3271 | iov_iter_single_seg_count(i)); |
3272 | goto again; | |
3273 | } | |
afddba49 NP |
3274 | pos += copied; |
3275 | written += copied; | |
3276 | ||
3277 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3278 | } while (iov_iter_count(i)); |
3279 | ||
3280 | return written ? written : status; | |
3281 | } | |
3b93f911 | 3282 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3283 | |
e4dd9de3 | 3284 | /** |
8174202b | 3285 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3286 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3287 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3288 | * |
3289 | * This function does all the work needed for actually writing data to a | |
3290 | * file. It does all basic checks, removes SUID from the file, updates | |
3291 | * modification times and calls proper subroutines depending on whether we | |
3292 | * do direct IO or a standard buffered write. | |
3293 | * | |
3294 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
3295 | * object which does not need locking at all. | |
3296 | * | |
3297 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3298 | * A caller has to handle it. This is mainly due to the fact that we want to | |
3299 | * avoid syncing under i_mutex. | |
a862f68a MR |
3300 | * |
3301 | * Return: | |
3302 | * * number of bytes written, even for truncated writes | |
3303 | * * negative error code if no data has been written at all | |
e4dd9de3 | 3304 | */ |
8174202b | 3305 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3306 | { |
3307 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 3308 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 3309 | struct inode *inode = mapping->host; |
3b93f911 | 3310 | ssize_t written = 0; |
1da177e4 | 3311 | ssize_t err; |
3b93f911 | 3312 | ssize_t status; |
1da177e4 | 3313 | |
1da177e4 | 3314 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3315 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3316 | err = file_remove_privs(file); |
1da177e4 LT |
3317 | if (err) |
3318 | goto out; | |
3319 | ||
c3b2da31 JB |
3320 | err = file_update_time(file); |
3321 | if (err) | |
3322 | goto out; | |
1da177e4 | 3323 | |
2ba48ce5 | 3324 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3325 | loff_t pos, endbyte; |
fb5527e6 | 3326 | |
1af5bb49 | 3327 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3328 | /* |
fbbbad4b MW |
3329 | * If the write stopped short of completing, fall back to |
3330 | * buffered writes. Some filesystems do this for writes to | |
3331 | * holes, for example. For DAX files, a buffered write will | |
3332 | * not succeed (even if it did, DAX does not handle dirty | |
3333 | * page-cache pages correctly). | |
1da177e4 | 3334 | */ |
0b8def9d | 3335 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3336 | goto out; |
3337 | ||
0b8def9d | 3338 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3339 | /* |
3b93f911 | 3340 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3341 | * then we want to return the number of bytes which were |
3342 | * direct-written, or the error code if that was zero. Note | |
3343 | * that this differs from normal direct-io semantics, which | |
3344 | * will return -EFOO even if some bytes were written. | |
3345 | */ | |
60bb4529 | 3346 | if (unlikely(status < 0)) { |
3b93f911 | 3347 | err = status; |
fb5527e6 JM |
3348 | goto out; |
3349 | } | |
fb5527e6 JM |
3350 | /* |
3351 | * We need to ensure that the page cache pages are written to | |
3352 | * disk and invalidated to preserve the expected O_DIRECT | |
3353 | * semantics. | |
3354 | */ | |
3b93f911 | 3355 | endbyte = pos + status - 1; |
0b8def9d | 3356 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3357 | if (err == 0) { |
0b8def9d | 3358 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3359 | written += status; |
fb5527e6 | 3360 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3361 | pos >> PAGE_SHIFT, |
3362 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3363 | } else { |
3364 | /* | |
3365 | * We don't know how much we wrote, so just return | |
3366 | * the number of bytes which were direct-written | |
3367 | */ | |
3368 | } | |
3369 | } else { | |
0b8def9d AV |
3370 | written = generic_perform_write(file, from, iocb->ki_pos); |
3371 | if (likely(written > 0)) | |
3372 | iocb->ki_pos += written; | |
fb5527e6 | 3373 | } |
1da177e4 LT |
3374 | out: |
3375 | current->backing_dev_info = NULL; | |
3376 | return written ? written : err; | |
3377 | } | |
8174202b | 3378 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3379 | |
e4dd9de3 | 3380 | /** |
8174202b | 3381 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3382 | * @iocb: IO state structure |
8174202b | 3383 | * @from: iov_iter with data to write |
e4dd9de3 | 3384 | * |
8174202b | 3385 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
3386 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
3387 | * and acquires i_mutex as needed. | |
a862f68a MR |
3388 | * Return: |
3389 | * * negative error code if no data has been written at all of | |
3390 | * vfs_fsync_range() failed for a synchronous write | |
3391 | * * number of bytes written, even for truncated writes | |
e4dd9de3 | 3392 | */ |
8174202b | 3393 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3394 | { |
3395 | struct file *file = iocb->ki_filp; | |
148f948b | 3396 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3397 | ssize_t ret; |
1da177e4 | 3398 | |
5955102c | 3399 | inode_lock(inode); |
3309dd04 AV |
3400 | ret = generic_write_checks(iocb, from); |
3401 | if (ret > 0) | |
5f380c7f | 3402 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3403 | inode_unlock(inode); |
1da177e4 | 3404 | |
e2592217 CH |
3405 | if (ret > 0) |
3406 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3407 | return ret; |
3408 | } | |
8174202b | 3409 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3410 | |
cf9a2ae8 DH |
3411 | /** |
3412 | * try_to_release_page() - release old fs-specific metadata on a page | |
3413 | * | |
3414 | * @page: the page which the kernel is trying to free | |
3415 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3416 | * | |
3417 | * The address_space is to try to release any data against the page | |
a862f68a | 3418 | * (presumably at page->private). |
cf9a2ae8 | 3419 | * |
266cf658 DH |
3420 | * This may also be called if PG_fscache is set on a page, indicating that the |
3421 | * page is known to the local caching routines. | |
3422 | * | |
cf9a2ae8 | 3423 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3424 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3425 | * |
a862f68a | 3426 | * Return: %1 if the release was successful, otherwise return zero. |
cf9a2ae8 DH |
3427 | */ |
3428 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3429 | { | |
3430 | struct address_space * const mapping = page->mapping; | |
3431 | ||
3432 | BUG_ON(!PageLocked(page)); | |
3433 | if (PageWriteback(page)) | |
3434 | return 0; | |
3435 | ||
3436 | if (mapping && mapping->a_ops->releasepage) | |
3437 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3438 | return try_to_free_buffers(page); | |
3439 | } | |
3440 | ||
3441 | EXPORT_SYMBOL(try_to_release_page); |