Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/filemap.c | |
4 | * | |
5 | * Copyright (C) 1994-1999 Linus Torvalds | |
6 | */ | |
7 | ||
8 | /* | |
9 | * This file handles the generic file mmap semantics used by | |
10 | * most "normal" filesystems (but you don't /have/ to use this: | |
11 | * the NFS filesystem used to do this differently, for example) | |
12 | */ | |
b95f1b31 | 13 | #include <linux/export.h> |
1da177e4 | 14 | #include <linux/compiler.h> |
f9fe48be | 15 | #include <linux/dax.h> |
1da177e4 | 16 | #include <linux/fs.h> |
3f07c014 | 17 | #include <linux/sched/signal.h> |
c22ce143 | 18 | #include <linux/uaccess.h> |
c59ede7b | 19 | #include <linux/capability.h> |
1da177e4 | 20 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 21 | #include <linux/gfp.h> |
1da177e4 LT |
22 | #include <linux/mm.h> |
23 | #include <linux/swap.h> | |
24 | #include <linux/mman.h> | |
25 | #include <linux/pagemap.h> | |
26 | #include <linux/file.h> | |
27 | #include <linux/uio.h> | |
cfcbfb13 | 28 | #include <linux/error-injection.h> |
1da177e4 LT |
29 | #include <linux/hash.h> |
30 | #include <linux/writeback.h> | |
53253383 | 31 | #include <linux/backing-dev.h> |
1da177e4 LT |
32 | #include <linux/pagevec.h> |
33 | #include <linux/blkdev.h> | |
34 | #include <linux/security.h> | |
44110fe3 | 35 | #include <linux/cpuset.h> |
00501b53 | 36 | #include <linux/hugetlb.h> |
8a9f3ccd | 37 | #include <linux/memcontrol.h> |
c515e1fd | 38 | #include <linux/cleancache.h> |
c7df8ad2 | 39 | #include <linux/shmem_fs.h> |
f1820361 | 40 | #include <linux/rmap.h> |
b1d29ba8 | 41 | #include <linux/delayacct.h> |
eb414681 | 42 | #include <linux/psi.h> |
d0e6a582 | 43 | #include <linux/ramfs.h> |
b9306a79 | 44 | #include <linux/page_idle.h> |
f9ce0be7 | 45 | #include <asm/pgalloc.h> |
de591a82 | 46 | #include <asm/tlbflush.h> |
0f8053a5 NP |
47 | #include "internal.h" |
48 | ||
fe0bfaaf RJ |
49 | #define CREATE_TRACE_POINTS |
50 | #include <trace/events/filemap.h> | |
51 | ||
1da177e4 | 52 | /* |
1da177e4 LT |
53 | * FIXME: remove all knowledge of the buffer layer from the core VM |
54 | */ | |
148f948b | 55 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 56 | |
1da177e4 LT |
57 | #include <asm/mman.h> |
58 | ||
59 | /* | |
60 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
61 | * though. | |
62 | * | |
63 | * Shared mappings now work. 15.8.1995 Bruno. | |
64 | * | |
65 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
66 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
67 | * | |
68 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
69 | */ | |
70 | ||
71 | /* | |
72 | * Lock ordering: | |
73 | * | |
c8c06efa | 74 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 75 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 | 76 | * ->swap_lock (exclusive_swap_page, others) |
b93b0163 | 77 | * ->i_pages lock |
1da177e4 | 78 | * |
1b1dcc1b | 79 | * ->i_mutex |
c8c06efa | 80 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
1da177e4 | 81 | * |
c1e8d7c6 | 82 | * ->mmap_lock |
c8c06efa | 83 | * ->i_mmap_rwsem |
b8072f09 | 84 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
b93b0163 | 85 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
1da177e4 | 86 | * |
c1e8d7c6 | 87 | * ->mmap_lock |
1da177e4 LT |
88 | * ->lock_page (access_process_vm) |
89 | * | |
ccad2365 | 90 | * ->i_mutex (generic_perform_write) |
c1e8d7c6 | 91 | * ->mmap_lock (fault_in_pages_readable->do_page_fault) |
1da177e4 | 92 | * |
f758eeab | 93 | * bdi->wb.list_lock |
a66979ab | 94 | * sb_lock (fs/fs-writeback.c) |
b93b0163 | 95 | * ->i_pages lock (__sync_single_inode) |
1da177e4 | 96 | * |
c8c06efa | 97 | * ->i_mmap_rwsem |
1da177e4 LT |
98 | * ->anon_vma.lock (vma_adjust) |
99 | * | |
100 | * ->anon_vma.lock | |
b8072f09 | 101 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 102 | * |
b8072f09 | 103 | * ->page_table_lock or pte_lock |
5d337b91 | 104 | * ->swap_lock (try_to_unmap_one) |
1da177e4 | 105 | * ->private_lock (try_to_unmap_one) |
b93b0163 | 106 | * ->i_pages lock (try_to_unmap_one) |
15b44736 HD |
107 | * ->lruvec->lru_lock (follow_page->mark_page_accessed) |
108 | * ->lruvec->lru_lock (check_pte_range->isolate_lru_page) | |
1da177e4 | 109 | * ->private_lock (page_remove_rmap->set_page_dirty) |
b93b0163 | 110 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
f758eeab | 111 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 112 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 113 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 114 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 115 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
116 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
117 | * | |
c8c06efa | 118 | * ->i_mmap_rwsem |
9a3c531d | 119 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
120 | */ |
121 | ||
5c024e6a | 122 | static void page_cache_delete(struct address_space *mapping, |
91b0abe3 JW |
123 | struct page *page, void *shadow) |
124 | { | |
5c024e6a MW |
125 | XA_STATE(xas, &mapping->i_pages, page->index); |
126 | unsigned int nr = 1; | |
c70b647d | 127 | |
5c024e6a | 128 | mapping_set_update(&xas, mapping); |
c70b647d | 129 | |
5c024e6a MW |
130 | /* hugetlb pages are represented by a single entry in the xarray */ |
131 | if (!PageHuge(page)) { | |
132 | xas_set_order(&xas, page->index, compound_order(page)); | |
d8c6546b | 133 | nr = compound_nr(page); |
5c024e6a | 134 | } |
91b0abe3 | 135 | |
83929372 KS |
136 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
137 | VM_BUG_ON_PAGE(PageTail(page), page); | |
138 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 139 | |
5c024e6a MW |
140 | xas_store(&xas, shadow); |
141 | xas_init_marks(&xas); | |
d3798ae8 | 142 | |
2300638b JK |
143 | page->mapping = NULL; |
144 | /* Leave page->index set: truncation lookup relies upon it */ | |
d3798ae8 | 145 | mapping->nrpages -= nr; |
91b0abe3 JW |
146 | } |
147 | ||
5ecc4d85 JK |
148 | static void unaccount_page_cache_page(struct address_space *mapping, |
149 | struct page *page) | |
1da177e4 | 150 | { |
5ecc4d85 | 151 | int nr; |
1da177e4 | 152 | |
c515e1fd DM |
153 | /* |
154 | * if we're uptodate, flush out into the cleancache, otherwise | |
155 | * invalidate any existing cleancache entries. We can't leave | |
156 | * stale data around in the cleancache once our page is gone | |
157 | */ | |
158 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
159 | cleancache_put_page(page); | |
160 | else | |
3167760f | 161 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 162 | |
83929372 | 163 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
164 | VM_BUG_ON_PAGE(page_mapped(page), page); |
165 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
166 | int mapcount; | |
167 | ||
168 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
169 | current->comm, page_to_pfn(page)); | |
170 | dump_page(page, "still mapped when deleted"); | |
171 | dump_stack(); | |
172 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
173 | ||
174 | mapcount = page_mapcount(page); | |
175 | if (mapping_exiting(mapping) && | |
176 | page_count(page) >= mapcount + 2) { | |
177 | /* | |
178 | * All vmas have already been torn down, so it's | |
179 | * a good bet that actually the page is unmapped, | |
180 | * and we'd prefer not to leak it: if we're wrong, | |
181 | * some other bad page check should catch it later. | |
182 | */ | |
183 | page_mapcount_reset(page); | |
6d061f9f | 184 | page_ref_sub(page, mapcount); |
06b241f3 HD |
185 | } |
186 | } | |
187 | ||
4165b9b4 | 188 | /* hugetlb pages do not participate in page cache accounting. */ |
5ecc4d85 JK |
189 | if (PageHuge(page)) |
190 | return; | |
09612fa6 | 191 | |
6c357848 | 192 | nr = thp_nr_pages(page); |
5ecc4d85 | 193 | |
0d1c2072 | 194 | __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr); |
5ecc4d85 | 195 | if (PageSwapBacked(page)) { |
0d1c2072 | 196 | __mod_lruvec_page_state(page, NR_SHMEM, -nr); |
5ecc4d85 | 197 | if (PageTransHuge(page)) |
57b2847d | 198 | __mod_lruvec_page_state(page, NR_SHMEM_THPS, -nr); |
99cb0dbd | 199 | } else if (PageTransHuge(page)) { |
bf9ecead | 200 | __mod_lruvec_page_state(page, NR_FILE_THPS, -nr); |
09d91cda | 201 | filemap_nr_thps_dec(mapping); |
800d8c63 | 202 | } |
5ecc4d85 JK |
203 | |
204 | /* | |
205 | * At this point page must be either written or cleaned by | |
206 | * truncate. Dirty page here signals a bug and loss of | |
207 | * unwritten data. | |
208 | * | |
209 | * This fixes dirty accounting after removing the page entirely | |
210 | * but leaves PageDirty set: it has no effect for truncated | |
211 | * page and anyway will be cleared before returning page into | |
212 | * buddy allocator. | |
213 | */ | |
214 | if (WARN_ON_ONCE(PageDirty(page))) | |
215 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); | |
216 | } | |
217 | ||
218 | /* | |
219 | * Delete a page from the page cache and free it. Caller has to make | |
220 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 221 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 JK |
222 | */ |
223 | void __delete_from_page_cache(struct page *page, void *shadow) | |
224 | { | |
225 | struct address_space *mapping = page->mapping; | |
226 | ||
227 | trace_mm_filemap_delete_from_page_cache(page); | |
228 | ||
229 | unaccount_page_cache_page(mapping, page); | |
5c024e6a | 230 | page_cache_delete(mapping, page, shadow); |
1da177e4 LT |
231 | } |
232 | ||
59c66c5f JK |
233 | static void page_cache_free_page(struct address_space *mapping, |
234 | struct page *page) | |
235 | { | |
236 | void (*freepage)(struct page *); | |
237 | ||
238 | freepage = mapping->a_ops->freepage; | |
239 | if (freepage) | |
240 | freepage(page); | |
241 | ||
242 | if (PageTransHuge(page) && !PageHuge(page)) { | |
887b22c6 | 243 | page_ref_sub(page, thp_nr_pages(page)); |
59c66c5f JK |
244 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); |
245 | } else { | |
246 | put_page(page); | |
247 | } | |
248 | } | |
249 | ||
702cfbf9 MK |
250 | /** |
251 | * delete_from_page_cache - delete page from page cache | |
252 | * @page: the page which the kernel is trying to remove from page cache | |
253 | * | |
254 | * This must be called only on pages that have been verified to be in the page | |
255 | * cache and locked. It will never put the page into the free list, the caller | |
256 | * has a reference on the page. | |
257 | */ | |
258 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 259 | { |
83929372 | 260 | struct address_space *mapping = page_mapping(page); |
c4843a75 | 261 | unsigned long flags; |
1da177e4 | 262 | |
cd7619d6 | 263 | BUG_ON(!PageLocked(page)); |
b93b0163 | 264 | xa_lock_irqsave(&mapping->i_pages, flags); |
62cccb8c | 265 | __delete_from_page_cache(page, NULL); |
b93b0163 | 266 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
6072d13c | 267 | |
59c66c5f | 268 | page_cache_free_page(mapping, page); |
97cecb5a MK |
269 | } |
270 | EXPORT_SYMBOL(delete_from_page_cache); | |
271 | ||
aa65c29c | 272 | /* |
ef8e5717 | 273 | * page_cache_delete_batch - delete several pages from page cache |
aa65c29c JK |
274 | * @mapping: the mapping to which pages belong |
275 | * @pvec: pagevec with pages to delete | |
276 | * | |
b93b0163 | 277 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
4101196b MWO |
278 | * from the mapping. The function expects @pvec to be sorted by page index |
279 | * and is optimised for it to be dense. | |
b93b0163 | 280 | * It tolerates holes in @pvec (mapping entries at those indices are not |
aa65c29c | 281 | * modified). The function expects only THP head pages to be present in the |
4101196b | 282 | * @pvec. |
aa65c29c | 283 | * |
b93b0163 | 284 | * The function expects the i_pages lock to be held. |
aa65c29c | 285 | */ |
ef8e5717 | 286 | static void page_cache_delete_batch(struct address_space *mapping, |
aa65c29c JK |
287 | struct pagevec *pvec) |
288 | { | |
ef8e5717 | 289 | XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index); |
aa65c29c | 290 | int total_pages = 0; |
4101196b | 291 | int i = 0; |
aa65c29c | 292 | struct page *page; |
aa65c29c | 293 | |
ef8e5717 MW |
294 | mapping_set_update(&xas, mapping); |
295 | xas_for_each(&xas, page, ULONG_MAX) { | |
4101196b | 296 | if (i >= pagevec_count(pvec)) |
aa65c29c | 297 | break; |
4101196b MWO |
298 | |
299 | /* A swap/dax/shadow entry got inserted? Skip it. */ | |
3159f943 | 300 | if (xa_is_value(page)) |
aa65c29c | 301 | continue; |
4101196b MWO |
302 | /* |
303 | * A page got inserted in our range? Skip it. We have our | |
304 | * pages locked so they are protected from being removed. | |
305 | * If we see a page whose index is higher than ours, it | |
306 | * means our page has been removed, which shouldn't be | |
307 | * possible because we're holding the PageLock. | |
308 | */ | |
309 | if (page != pvec->pages[i]) { | |
310 | VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index, | |
311 | page); | |
312 | continue; | |
313 | } | |
314 | ||
315 | WARN_ON_ONCE(!PageLocked(page)); | |
316 | ||
317 | if (page->index == xas.xa_index) | |
aa65c29c | 318 | page->mapping = NULL; |
4101196b MWO |
319 | /* Leave page->index set: truncation lookup relies on it */ |
320 | ||
321 | /* | |
322 | * Move to the next page in the vector if this is a regular | |
323 | * page or the index is of the last sub-page of this compound | |
324 | * page. | |
325 | */ | |
326 | if (page->index + compound_nr(page) - 1 == xas.xa_index) | |
aa65c29c | 327 | i++; |
ef8e5717 | 328 | xas_store(&xas, NULL); |
aa65c29c JK |
329 | total_pages++; |
330 | } | |
331 | mapping->nrpages -= total_pages; | |
332 | } | |
333 | ||
334 | void delete_from_page_cache_batch(struct address_space *mapping, | |
335 | struct pagevec *pvec) | |
336 | { | |
337 | int i; | |
338 | unsigned long flags; | |
339 | ||
340 | if (!pagevec_count(pvec)) | |
341 | return; | |
342 | ||
b93b0163 | 343 | xa_lock_irqsave(&mapping->i_pages, flags); |
aa65c29c JK |
344 | for (i = 0; i < pagevec_count(pvec); i++) { |
345 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); | |
346 | ||
347 | unaccount_page_cache_page(mapping, pvec->pages[i]); | |
348 | } | |
ef8e5717 | 349 | page_cache_delete_batch(mapping, pvec); |
b93b0163 | 350 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
aa65c29c JK |
351 | |
352 | for (i = 0; i < pagevec_count(pvec); i++) | |
353 | page_cache_free_page(mapping, pvec->pages[i]); | |
354 | } | |
355 | ||
d72d9e2a | 356 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
357 | { |
358 | int ret = 0; | |
359 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
360 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
361 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 362 | ret = -ENOSPC; |
7fcbbaf1 JA |
363 | if (test_bit(AS_EIO, &mapping->flags) && |
364 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
365 | ret = -EIO; |
366 | return ret; | |
367 | } | |
d72d9e2a | 368 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 369 | |
76341cab JL |
370 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
371 | { | |
372 | /* Check for outstanding write errors */ | |
373 | if (test_bit(AS_EIO, &mapping->flags)) | |
374 | return -EIO; | |
375 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
376 | return -ENOSPC; | |
377 | return 0; | |
378 | } | |
379 | ||
1da177e4 | 380 | /** |
485bb99b | 381 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
382 | * @mapping: address space structure to write |
383 | * @start: offset in bytes where the range starts | |
469eb4d0 | 384 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 385 | * @sync_mode: enable synchronous operation |
1da177e4 | 386 | * |
485bb99b RD |
387 | * Start writeback against all of a mapping's dirty pages that lie |
388 | * within the byte offsets <start, end> inclusive. | |
389 | * | |
1da177e4 | 390 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 391 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
392 | * these two operations is that if a dirty page/buffer is encountered, it must |
393 | * be waited upon, and not just skipped over. | |
a862f68a MR |
394 | * |
395 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 396 | */ |
ebcf28e1 AM |
397 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
398 | loff_t end, int sync_mode) | |
1da177e4 LT |
399 | { |
400 | int ret; | |
401 | struct writeback_control wbc = { | |
402 | .sync_mode = sync_mode, | |
05fe478d | 403 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
404 | .range_start = start, |
405 | .range_end = end, | |
1da177e4 LT |
406 | }; |
407 | ||
f56753ac | 408 | if (!mapping_can_writeback(mapping) || |
c3aab9a0 | 409 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
1da177e4 LT |
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 | ||
aa0bfcd9 RZ |
550 | /** |
551 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete | |
552 | * @mapping: 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 given address space in the | |
557 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), | |
558 | * this function does not clear error status of the address space. | |
559 | * | |
560 | * Use this function if callers don't handle errors themselves. Expected | |
561 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
562 | * fsfreeze(8) | |
563 | */ | |
564 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, | |
565 | loff_t start_byte, loff_t end_byte) | |
566 | { | |
567 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
568 | return filemap_check_and_keep_errors(mapping); | |
569 | } | |
570 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); | |
571 | ||
a823e458 JL |
572 | /** |
573 | * file_fdatawait_range - wait for writeback to complete | |
574 | * @file: file pointing to address space structure to wait for | |
575 | * @start_byte: offset in bytes where the range starts | |
576 | * @end_byte: offset in bytes where the range ends (inclusive) | |
577 | * | |
578 | * Walk the list of under-writeback pages of the address space that file | |
579 | * refers to, in the given range and wait for all of them. Check error | |
580 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
581 | * | |
582 | * Since the error status of the file is advanced by this function, | |
583 | * callers are responsible for checking the return value and handling and/or | |
584 | * reporting the error. | |
a862f68a MR |
585 | * |
586 | * Return: error status of the address space vs. the file->f_wb_err cursor. | |
a823e458 JL |
587 | */ |
588 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
589 | { | |
590 | struct address_space *mapping = file->f_mapping; | |
591 | ||
592 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
593 | return file_check_and_advance_wb_err(file); | |
594 | } | |
595 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 596 | |
aa750fd7 JN |
597 | /** |
598 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
599 | * @mapping: address space structure to wait for | |
600 | * | |
601 | * Walk the list of under-writeback pages of the given address space | |
602 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
603 | * does not clear error status of the address space. | |
604 | * | |
605 | * Use this function if callers don't handle errors themselves. Expected | |
606 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
607 | * fsfreeze(8) | |
a862f68a MR |
608 | * |
609 | * Return: error status of the address space. | |
aa750fd7 | 610 | */ |
76341cab | 611 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 612 | { |
ffb959bb | 613 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 614 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 615 | } |
76341cab | 616 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 617 | |
875d91b1 | 618 | /* Returns true if writeback might be needed or already in progress. */ |
9326c9b2 | 619 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 620 | { |
875d91b1 | 621 | return mapping->nrpages; |
1da177e4 | 622 | } |
1da177e4 | 623 | |
63135aa3 JA |
624 | /** |
625 | * filemap_range_needs_writeback - check if range potentially needs writeback | |
626 | * @mapping: address space within which to check | |
627 | * @start_byte: offset in bytes where the range starts | |
628 | * @end_byte: offset in bytes where the range ends (inclusive) | |
629 | * | |
630 | * Find at least one page in the range supplied, usually used to check if | |
631 | * direct writing in this range will trigger a writeback. Used by O_DIRECT | |
632 | * read/write with IOCB_NOWAIT, to see if the caller needs to do | |
633 | * filemap_write_and_wait_range() before proceeding. | |
634 | * | |
635 | * Return: %true if the caller should do filemap_write_and_wait_range() before | |
636 | * doing O_DIRECT to a page in this range, %false otherwise. | |
637 | */ | |
638 | bool filemap_range_needs_writeback(struct address_space *mapping, | |
639 | loff_t start_byte, loff_t end_byte) | |
640 | { | |
641 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); | |
642 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
643 | struct page *page; | |
644 | ||
645 | if (!mapping_needs_writeback(mapping)) | |
646 | return false; | |
647 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && | |
648 | !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK)) | |
649 | return false; | |
650 | if (end_byte < start_byte) | |
651 | return false; | |
652 | ||
653 | rcu_read_lock(); | |
654 | xas_for_each(&xas, page, max) { | |
655 | if (xas_retry(&xas, page)) | |
656 | continue; | |
657 | if (xa_is_value(page)) | |
658 | continue; | |
659 | if (PageDirty(page) || PageLocked(page) || PageWriteback(page)) | |
660 | break; | |
661 | } | |
662 | rcu_read_unlock(); | |
663 | return page != NULL; | |
664 | } | |
665 | EXPORT_SYMBOL_GPL(filemap_range_needs_writeback); | |
666 | ||
485bb99b RD |
667 | /** |
668 | * filemap_write_and_wait_range - write out & wait on a file range | |
669 | * @mapping: the address_space for the pages | |
670 | * @lstart: offset in bytes where the range starts | |
671 | * @lend: offset in bytes where the range ends (inclusive) | |
672 | * | |
469eb4d0 AM |
673 | * Write out and wait upon file offsets lstart->lend, inclusive. |
674 | * | |
0e056eb5 | 675 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 | 676 | * that this function can be used to write to the very end-of-file (end = -1). |
a862f68a MR |
677 | * |
678 | * Return: error status of the address space. | |
469eb4d0 | 679 | */ |
1da177e4 LT |
680 | int filemap_write_and_wait_range(struct address_space *mapping, |
681 | loff_t lstart, loff_t lend) | |
682 | { | |
28fd1298 | 683 | int err = 0; |
1da177e4 | 684 | |
9326c9b2 | 685 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
686 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
687 | WB_SYNC_ALL); | |
ddf8f376 IW |
688 | /* |
689 | * Even if the above returned error, the pages may be | |
690 | * written partially (e.g. -ENOSPC), so we wait for it. | |
691 | * But the -EIO is special case, it may indicate the worst | |
692 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
693 | */ | |
28fd1298 | 694 | if (err != -EIO) { |
94004ed7 CH |
695 | int err2 = filemap_fdatawait_range(mapping, |
696 | lstart, lend); | |
28fd1298 OH |
697 | if (!err) |
698 | err = err2; | |
cbeaf951 JL |
699 | } else { |
700 | /* Clear any previously stored errors */ | |
701 | filemap_check_errors(mapping); | |
28fd1298 | 702 | } |
865ffef3 DM |
703 | } else { |
704 | err = filemap_check_errors(mapping); | |
1da177e4 | 705 | } |
28fd1298 | 706 | return err; |
1da177e4 | 707 | } |
f6995585 | 708 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 709 | |
5660e13d JL |
710 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
711 | { | |
3acdfd28 | 712 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
713 | |
714 | trace_filemap_set_wb_err(mapping, eseq); | |
715 | } | |
716 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
717 | ||
718 | /** | |
719 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
720 | * and advance wb_err to current one | |
721 | * @file: struct file on which the error is being reported | |
722 | * | |
723 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
724 | * want to report any writeback errors that occurred since the last fsync (or | |
725 | * since the file was opened if there haven't been any). | |
726 | * | |
727 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
728 | * then just quickly return 0. The file is all caught up. | |
729 | * | |
730 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
731 | * it and try to swap it into place. If it works, or another task beat us | |
732 | * to it with the new value, then update the f_wb_err and return the error | |
733 | * portion. The error at this point must be reported via proper channels | |
734 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
735 | * | |
736 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
737 | * value is protected by the f_lock since we must ensure that it reflects | |
738 | * the latest value swapped in for this file descriptor. | |
a862f68a MR |
739 | * |
740 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
741 | */ |
742 | int file_check_and_advance_wb_err(struct file *file) | |
743 | { | |
744 | int err = 0; | |
745 | errseq_t old = READ_ONCE(file->f_wb_err); | |
746 | struct address_space *mapping = file->f_mapping; | |
747 | ||
748 | /* Locklessly handle the common case where nothing has changed */ | |
749 | if (errseq_check(&mapping->wb_err, old)) { | |
750 | /* Something changed, must use slow path */ | |
751 | spin_lock(&file->f_lock); | |
752 | old = file->f_wb_err; | |
753 | err = errseq_check_and_advance(&mapping->wb_err, | |
754 | &file->f_wb_err); | |
755 | trace_file_check_and_advance_wb_err(file, old); | |
756 | spin_unlock(&file->f_lock); | |
757 | } | |
f4e222c5 JL |
758 | |
759 | /* | |
760 | * We're mostly using this function as a drop in replacement for | |
761 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
762 | * that the legacy code would have had on these flags. | |
763 | */ | |
764 | clear_bit(AS_EIO, &mapping->flags); | |
765 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
766 | return err; |
767 | } | |
768 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
769 | ||
770 | /** | |
771 | * file_write_and_wait_range - write out & wait on a file range | |
772 | * @file: file pointing to address_space with pages | |
773 | * @lstart: offset in bytes where the range starts | |
774 | * @lend: offset in bytes where the range ends (inclusive) | |
775 | * | |
776 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
777 | * | |
778 | * Note that @lend is inclusive (describes the last byte to be written) so | |
779 | * that this function can be used to write to the very end-of-file (end = -1). | |
780 | * | |
781 | * After writing out and waiting on the data, we check and advance the | |
782 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
a862f68a MR |
783 | * |
784 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
785 | */ |
786 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
787 | { | |
788 | int err = 0, err2; | |
789 | struct address_space *mapping = file->f_mapping; | |
790 | ||
9326c9b2 | 791 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
792 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
793 | WB_SYNC_ALL); | |
794 | /* See comment of filemap_write_and_wait() */ | |
795 | if (err != -EIO) | |
796 | __filemap_fdatawait_range(mapping, lstart, lend); | |
797 | } | |
798 | err2 = file_check_and_advance_wb_err(file); | |
799 | if (!err) | |
800 | err = err2; | |
801 | return err; | |
802 | } | |
803 | EXPORT_SYMBOL(file_write_and_wait_range); | |
804 | ||
ef6a3c63 MS |
805 | /** |
806 | * replace_page_cache_page - replace a pagecache page with a new one | |
807 | * @old: page to be replaced | |
808 | * @new: page to replace with | |
ef6a3c63 MS |
809 | * |
810 | * This function replaces a page in the pagecache with a new one. On | |
811 | * success it acquires the pagecache reference for the new page and | |
812 | * drops it for the old page. Both the old and new pages must be | |
813 | * locked. This function does not add the new page to the LRU, the | |
814 | * caller must do that. | |
815 | * | |
74d60958 | 816 | * The remove + add is atomic. This function cannot fail. |
ef6a3c63 | 817 | */ |
1f7ef657 | 818 | void replace_page_cache_page(struct page *old, struct page *new) |
ef6a3c63 | 819 | { |
74d60958 MW |
820 | struct address_space *mapping = old->mapping; |
821 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
822 | pgoff_t offset = old->index; | |
823 | XA_STATE(xas, &mapping->i_pages, offset); | |
824 | unsigned long flags; | |
ef6a3c63 | 825 | |
309381fe SL |
826 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
827 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
828 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 829 | |
74d60958 MW |
830 | get_page(new); |
831 | new->mapping = mapping; | |
832 | new->index = offset; | |
ef6a3c63 | 833 | |
0d1c2072 JW |
834 | mem_cgroup_migrate(old, new); |
835 | ||
74d60958 MW |
836 | xas_lock_irqsave(&xas, flags); |
837 | xas_store(&xas, new); | |
4165b9b4 | 838 | |
74d60958 MW |
839 | old->mapping = NULL; |
840 | /* hugetlb pages do not participate in page cache accounting. */ | |
841 | if (!PageHuge(old)) | |
0d1c2072 | 842 | __dec_lruvec_page_state(old, NR_FILE_PAGES); |
74d60958 | 843 | if (!PageHuge(new)) |
0d1c2072 | 844 | __inc_lruvec_page_state(new, NR_FILE_PAGES); |
74d60958 | 845 | if (PageSwapBacked(old)) |
0d1c2072 | 846 | __dec_lruvec_page_state(old, NR_SHMEM); |
74d60958 | 847 | if (PageSwapBacked(new)) |
0d1c2072 | 848 | __inc_lruvec_page_state(new, NR_SHMEM); |
74d60958 | 849 | xas_unlock_irqrestore(&xas, flags); |
74d60958 MW |
850 | if (freepage) |
851 | freepage(old); | |
852 | put_page(old); | |
ef6a3c63 MS |
853 | } |
854 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
855 | ||
16c0cc0c | 856 | noinline int __add_to_page_cache_locked(struct page *page, |
76cd6173 | 857 | struct address_space *mapping, |
c4cf498d | 858 | pgoff_t offset, gfp_t gfp, |
76cd6173 | 859 | void **shadowp) |
1da177e4 | 860 | { |
74d60958 | 861 | XA_STATE(xas, &mapping->i_pages, offset); |
00501b53 | 862 | int huge = PageHuge(page); |
e286781d | 863 | int error; |
da74240e | 864 | bool charged = false; |
e286781d | 865 | |
309381fe SL |
866 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
867 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
74d60958 | 868 | mapping_set_update(&xas, mapping); |
e286781d | 869 | |
09cbfeaf | 870 | get_page(page); |
66a0c8ee KS |
871 | page->mapping = mapping; |
872 | page->index = offset; | |
873 | ||
3fea5a49 | 874 | if (!huge) { |
04f94e3f | 875 | error = mem_cgroup_charge(page, NULL, gfp); |
3fea5a49 JW |
876 | if (error) |
877 | goto error; | |
da74240e | 878 | charged = true; |
3fea5a49 JW |
879 | } |
880 | ||
198b62f8 MWO |
881 | gfp &= GFP_RECLAIM_MASK; |
882 | ||
74d60958 | 883 | do { |
198b62f8 MWO |
884 | unsigned int order = xa_get_order(xas.xa, xas.xa_index); |
885 | void *entry, *old = NULL; | |
886 | ||
887 | if (order > thp_order(page)) | |
888 | xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index), | |
889 | order, gfp); | |
74d60958 | 890 | xas_lock_irq(&xas); |
198b62f8 MWO |
891 | xas_for_each_conflict(&xas, entry) { |
892 | old = entry; | |
893 | if (!xa_is_value(entry)) { | |
894 | xas_set_err(&xas, -EEXIST); | |
895 | goto unlock; | |
896 | } | |
897 | } | |
898 | ||
899 | if (old) { | |
900 | if (shadowp) | |
901 | *shadowp = old; | |
902 | /* entry may have been split before we acquired lock */ | |
903 | order = xa_get_order(xas.xa, xas.xa_index); | |
904 | if (order > thp_order(page)) { | |
905 | xas_split(&xas, old, order); | |
906 | xas_reset(&xas); | |
907 | } | |
908 | } | |
909 | ||
74d60958 MW |
910 | xas_store(&xas, page); |
911 | if (xas_error(&xas)) | |
912 | goto unlock; | |
913 | ||
74d60958 MW |
914 | mapping->nrpages++; |
915 | ||
916 | /* hugetlb pages do not participate in page cache accounting */ | |
917 | if (!huge) | |
0d1c2072 | 918 | __inc_lruvec_page_state(page, NR_FILE_PAGES); |
74d60958 MW |
919 | unlock: |
920 | xas_unlock_irq(&xas); | |
198b62f8 | 921 | } while (xas_nomem(&xas, gfp)); |
74d60958 | 922 | |
3fea5a49 JW |
923 | if (xas_error(&xas)) { |
924 | error = xas_error(&xas); | |
da74240e WL |
925 | if (charged) |
926 | mem_cgroup_uncharge(page); | |
74d60958 | 927 | goto error; |
3fea5a49 | 928 | } |
4165b9b4 | 929 | |
66a0c8ee KS |
930 | trace_mm_filemap_add_to_page_cache(page); |
931 | return 0; | |
74d60958 | 932 | error: |
66a0c8ee KS |
933 | page->mapping = NULL; |
934 | /* Leave page->index set: truncation relies upon it */ | |
09cbfeaf | 935 | put_page(page); |
3fea5a49 | 936 | return error; |
1da177e4 | 937 | } |
cfcbfb13 | 938 | ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO); |
a528910e JW |
939 | |
940 | /** | |
941 | * add_to_page_cache_locked - add a locked page to the pagecache | |
942 | * @page: page to add | |
943 | * @mapping: the page's address_space | |
944 | * @offset: page index | |
945 | * @gfp_mask: page allocation mode | |
946 | * | |
947 | * This function is used to add a page to the pagecache. It must be locked. | |
948 | * This function does not add the page to the LRU. The caller must do that. | |
a862f68a MR |
949 | * |
950 | * Return: %0 on success, negative error code otherwise. | |
a528910e JW |
951 | */ |
952 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
953 | pgoff_t offset, gfp_t gfp_mask) | |
954 | { | |
955 | return __add_to_page_cache_locked(page, mapping, offset, | |
956 | gfp_mask, NULL); | |
957 | } | |
e286781d | 958 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
959 | |
960 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 961 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 962 | { |
a528910e | 963 | void *shadow = NULL; |
4f98a2fe RR |
964 | int ret; |
965 | ||
48c935ad | 966 | __SetPageLocked(page); |
a528910e JW |
967 | ret = __add_to_page_cache_locked(page, mapping, offset, |
968 | gfp_mask, &shadow); | |
969 | if (unlikely(ret)) | |
48c935ad | 970 | __ClearPageLocked(page); |
a528910e JW |
971 | else { |
972 | /* | |
973 | * The page might have been evicted from cache only | |
974 | * recently, in which case it should be activated like | |
975 | * any other repeatedly accessed page. | |
f0281a00 RR |
976 | * The exception is pages getting rewritten; evicting other |
977 | * data from the working set, only to cache data that will | |
978 | * get overwritten with something else, is a waste of memory. | |
a528910e | 979 | */ |
1899ad18 JW |
980 | WARN_ON_ONCE(PageActive(page)); |
981 | if (!(gfp_mask & __GFP_WRITE) && shadow) | |
982 | workingset_refault(page, shadow); | |
a528910e JW |
983 | lru_cache_add(page); |
984 | } | |
1da177e4 LT |
985 | return ret; |
986 | } | |
18bc0bbd | 987 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 988 | |
44110fe3 | 989 | #ifdef CONFIG_NUMA |
2ae88149 | 990 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 991 | { |
c0ff7453 MX |
992 | int n; |
993 | struct page *page; | |
994 | ||
44110fe3 | 995 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
996 | unsigned int cpuset_mems_cookie; |
997 | do { | |
d26914d1 | 998 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 999 | n = cpuset_mem_spread_node(); |
96db800f | 1000 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 1001 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 1002 | |
c0ff7453 | 1003 | return page; |
44110fe3 | 1004 | } |
2ae88149 | 1005 | return alloc_pages(gfp, 0); |
44110fe3 | 1006 | } |
2ae88149 | 1007 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
1008 | #endif |
1009 | ||
1da177e4 LT |
1010 | /* |
1011 | * In order to wait for pages to become available there must be | |
1012 | * waitqueues associated with pages. By using a hash table of | |
1013 | * waitqueues where the bucket discipline is to maintain all | |
1014 | * waiters on the same queue and wake all when any of the pages | |
1015 | * become available, and for the woken contexts to check to be | |
1016 | * sure the appropriate page became available, this saves space | |
1017 | * at a cost of "thundering herd" phenomena during rare hash | |
1018 | * collisions. | |
1019 | */ | |
62906027 NP |
1020 | #define PAGE_WAIT_TABLE_BITS 8 |
1021 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
1022 | static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; | |
1023 | ||
1024 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
1da177e4 | 1025 | { |
62906027 | 1026 | return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 1027 | } |
1da177e4 | 1028 | |
62906027 | 1029 | void __init pagecache_init(void) |
1da177e4 | 1030 | { |
62906027 | 1031 | int i; |
1da177e4 | 1032 | |
62906027 NP |
1033 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
1034 | init_waitqueue_head(&page_wait_table[i]); | |
1035 | ||
1036 | page_writeback_init(); | |
1da177e4 | 1037 | } |
1da177e4 | 1038 | |
5ef64cc8 LT |
1039 | /* |
1040 | * The page wait code treats the "wait->flags" somewhat unusually, because | |
5868ec26 | 1041 | * we have multiple different kinds of waits, not just the usual "exclusive" |
5ef64cc8 LT |
1042 | * one. |
1043 | * | |
1044 | * We have: | |
1045 | * | |
1046 | * (a) no special bits set: | |
1047 | * | |
1048 | * We're just waiting for the bit to be released, and when a waker | |
1049 | * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up, | |
1050 | * and remove it from the wait queue. | |
1051 | * | |
1052 | * Simple and straightforward. | |
1053 | * | |
1054 | * (b) WQ_FLAG_EXCLUSIVE: | |
1055 | * | |
1056 | * The waiter is waiting to get the lock, and only one waiter should | |
1057 | * be woken up to avoid any thundering herd behavior. We'll set the | |
1058 | * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue. | |
1059 | * | |
1060 | * This is the traditional exclusive wait. | |
1061 | * | |
5868ec26 | 1062 | * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: |
5ef64cc8 LT |
1063 | * |
1064 | * The waiter is waiting to get the bit, and additionally wants the | |
1065 | * lock to be transferred to it for fair lock behavior. If the lock | |
1066 | * cannot be taken, we stop walking the wait queue without waking | |
1067 | * the waiter. | |
1068 | * | |
1069 | * This is the "fair lock handoff" case, and in addition to setting | |
1070 | * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see | |
1071 | * that it now has the lock. | |
1072 | */ | |
ac6424b9 | 1073 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 1074 | { |
5ef64cc8 | 1075 | unsigned int flags; |
62906027 NP |
1076 | struct wait_page_key *key = arg; |
1077 | struct wait_page_queue *wait_page | |
1078 | = container_of(wait, struct wait_page_queue, wait); | |
1079 | ||
cdc8fcb4 | 1080 | if (!wake_page_match(wait_page, key)) |
62906027 | 1081 | return 0; |
3510ca20 | 1082 | |
9a1ea439 | 1083 | /* |
5ef64cc8 LT |
1084 | * If it's a lock handoff wait, we get the bit for it, and |
1085 | * stop walking (and do not wake it up) if we can't. | |
9a1ea439 | 1086 | */ |
5ef64cc8 LT |
1087 | flags = wait->flags; |
1088 | if (flags & WQ_FLAG_EXCLUSIVE) { | |
1089 | if (test_bit(key->bit_nr, &key->page->flags)) | |
2a9127fc | 1090 | return -1; |
5ef64cc8 LT |
1091 | if (flags & WQ_FLAG_CUSTOM) { |
1092 | if (test_and_set_bit(key->bit_nr, &key->page->flags)) | |
1093 | return -1; | |
1094 | flags |= WQ_FLAG_DONE; | |
1095 | } | |
2a9127fc | 1096 | } |
f62e00cc | 1097 | |
5ef64cc8 LT |
1098 | /* |
1099 | * We are holding the wait-queue lock, but the waiter that | |
1100 | * is waiting for this will be checking the flags without | |
1101 | * any locking. | |
1102 | * | |
1103 | * So update the flags atomically, and wake up the waiter | |
1104 | * afterwards to avoid any races. This store-release pairs | |
1105 | * with the load-acquire in wait_on_page_bit_common(). | |
1106 | */ | |
1107 | smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN); | |
2a9127fc LT |
1108 | wake_up_state(wait->private, mode); |
1109 | ||
1110 | /* | |
1111 | * Ok, we have successfully done what we're waiting for, | |
1112 | * and we can unconditionally remove the wait entry. | |
1113 | * | |
5ef64cc8 LT |
1114 | * Note that this pairs with the "finish_wait()" in the |
1115 | * waiter, and has to be the absolute last thing we do. | |
1116 | * After this list_del_init(&wait->entry) the wait entry | |
2a9127fc LT |
1117 | * might be de-allocated and the process might even have |
1118 | * exited. | |
2a9127fc | 1119 | */ |
c6fe44d9 | 1120 | list_del_init_careful(&wait->entry); |
5ef64cc8 | 1121 | return (flags & WQ_FLAG_EXCLUSIVE) != 0; |
f62e00cc KM |
1122 | } |
1123 | ||
74d81bfa | 1124 | static void wake_up_page_bit(struct page *page, int bit_nr) |
cbbce822 | 1125 | { |
62906027 NP |
1126 | wait_queue_head_t *q = page_waitqueue(page); |
1127 | struct wait_page_key key; | |
1128 | unsigned long flags; | |
11a19c7b | 1129 | wait_queue_entry_t bookmark; |
cbbce822 | 1130 | |
62906027 NP |
1131 | key.page = page; |
1132 | key.bit_nr = bit_nr; | |
1133 | key.page_match = 0; | |
1134 | ||
11a19c7b TC |
1135 | bookmark.flags = 0; |
1136 | bookmark.private = NULL; | |
1137 | bookmark.func = NULL; | |
1138 | INIT_LIST_HEAD(&bookmark.entry); | |
1139 | ||
62906027 | 1140 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1141 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1142 | ||
1143 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1144 | /* | |
1145 | * Take a breather from holding the lock, | |
1146 | * allow pages that finish wake up asynchronously | |
1147 | * to acquire the lock and remove themselves | |
1148 | * from wait queue | |
1149 | */ | |
1150 | spin_unlock_irqrestore(&q->lock, flags); | |
1151 | cpu_relax(); | |
1152 | spin_lock_irqsave(&q->lock, flags); | |
1153 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1154 | } | |
1155 | ||
62906027 NP |
1156 | /* |
1157 | * It is possible for other pages to have collided on the waitqueue | |
1158 | * hash, so in that case check for a page match. That prevents a long- | |
1159 | * term waiter | |
1160 | * | |
1161 | * It is still possible to miss a case here, when we woke page waiters | |
1162 | * and removed them from the waitqueue, but there are still other | |
1163 | * page waiters. | |
1164 | */ | |
1165 | if (!waitqueue_active(q) || !key.page_match) { | |
1166 | ClearPageWaiters(page); | |
1167 | /* | |
1168 | * It's possible to miss clearing Waiters here, when we woke | |
1169 | * our page waiters, but the hashed waitqueue has waiters for | |
1170 | * other pages on it. | |
1171 | * | |
1172 | * That's okay, it's a rare case. The next waker will clear it. | |
1173 | */ | |
1174 | } | |
1175 | spin_unlock_irqrestore(&q->lock, flags); | |
1176 | } | |
74d81bfa NP |
1177 | |
1178 | static void wake_up_page(struct page *page, int bit) | |
1179 | { | |
1180 | if (!PageWaiters(page)) | |
1181 | return; | |
1182 | wake_up_page_bit(page, bit); | |
1183 | } | |
62906027 | 1184 | |
9a1ea439 HD |
1185 | /* |
1186 | * A choice of three behaviors for wait_on_page_bit_common(): | |
1187 | */ | |
1188 | enum behavior { | |
1189 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like | |
1190 | * __lock_page() waiting on then setting PG_locked. | |
1191 | */ | |
1192 | SHARED, /* Hold ref to page and check the bit when woken, like | |
1193 | * wait_on_page_writeback() waiting on PG_writeback. | |
1194 | */ | |
1195 | DROP, /* Drop ref to page before wait, no check when woken, | |
1196 | * like put_and_wait_on_page_locked() on PG_locked. | |
1197 | */ | |
1198 | }; | |
1199 | ||
2a9127fc | 1200 | /* |
5ef64cc8 LT |
1201 | * Attempt to check (or get) the page bit, and mark us done |
1202 | * if successful. | |
2a9127fc LT |
1203 | */ |
1204 | static inline bool trylock_page_bit_common(struct page *page, int bit_nr, | |
1205 | struct wait_queue_entry *wait) | |
1206 | { | |
1207 | if (wait->flags & WQ_FLAG_EXCLUSIVE) { | |
1208 | if (test_and_set_bit(bit_nr, &page->flags)) | |
1209 | return false; | |
1210 | } else if (test_bit(bit_nr, &page->flags)) | |
1211 | return false; | |
1212 | ||
5ef64cc8 | 1213 | wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE; |
2a9127fc LT |
1214 | return true; |
1215 | } | |
1216 | ||
5ef64cc8 LT |
1217 | /* How many times do we accept lock stealing from under a waiter? */ |
1218 | int sysctl_page_lock_unfairness = 5; | |
1219 | ||
62906027 | 1220 | static inline int wait_on_page_bit_common(wait_queue_head_t *q, |
9a1ea439 | 1221 | struct page *page, int bit_nr, int state, enum behavior behavior) |
62906027 | 1222 | { |
5ef64cc8 | 1223 | int unfairness = sysctl_page_lock_unfairness; |
62906027 | 1224 | struct wait_page_queue wait_page; |
ac6424b9 | 1225 | wait_queue_entry_t *wait = &wait_page.wait; |
b1d29ba8 | 1226 | bool thrashing = false; |
9a1ea439 | 1227 | bool delayacct = false; |
eb414681 | 1228 | unsigned long pflags; |
62906027 | 1229 | |
eb414681 | 1230 | if (bit_nr == PG_locked && |
b1d29ba8 | 1231 | !PageUptodate(page) && PageWorkingset(page)) { |
9a1ea439 | 1232 | if (!PageSwapBacked(page)) { |
eb414681 | 1233 | delayacct_thrashing_start(); |
9a1ea439 HD |
1234 | delayacct = true; |
1235 | } | |
eb414681 | 1236 | psi_memstall_enter(&pflags); |
b1d29ba8 JW |
1237 | thrashing = true; |
1238 | } | |
1239 | ||
62906027 NP |
1240 | init_wait(wait); |
1241 | wait->func = wake_page_function; | |
1242 | wait_page.page = page; | |
1243 | wait_page.bit_nr = bit_nr; | |
1244 | ||
5ef64cc8 LT |
1245 | repeat: |
1246 | wait->flags = 0; | |
1247 | if (behavior == EXCLUSIVE) { | |
1248 | wait->flags = WQ_FLAG_EXCLUSIVE; | |
1249 | if (--unfairness < 0) | |
1250 | wait->flags |= WQ_FLAG_CUSTOM; | |
1251 | } | |
1252 | ||
2a9127fc LT |
1253 | /* |
1254 | * Do one last check whether we can get the | |
1255 | * page bit synchronously. | |
1256 | * | |
1257 | * Do the SetPageWaiters() marking before that | |
1258 | * to let any waker we _just_ missed know they | |
1259 | * need to wake us up (otherwise they'll never | |
1260 | * even go to the slow case that looks at the | |
1261 | * page queue), and add ourselves to the wait | |
1262 | * queue if we need to sleep. | |
1263 | * | |
1264 | * This part needs to be done under the queue | |
1265 | * lock to avoid races. | |
1266 | */ | |
1267 | spin_lock_irq(&q->lock); | |
1268 | SetPageWaiters(page); | |
1269 | if (!trylock_page_bit_common(page, bit_nr, wait)) | |
1270 | __add_wait_queue_entry_tail(q, wait); | |
1271 | spin_unlock_irq(&q->lock); | |
62906027 | 1272 | |
2a9127fc LT |
1273 | /* |
1274 | * From now on, all the logic will be based on | |
5ef64cc8 LT |
1275 | * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to |
1276 | * see whether the page bit testing has already | |
1277 | * been done by the wake function. | |
2a9127fc LT |
1278 | * |
1279 | * We can drop our reference to the page. | |
1280 | */ | |
1281 | if (behavior == DROP) | |
1282 | put_page(page); | |
62906027 | 1283 | |
5ef64cc8 LT |
1284 | /* |
1285 | * Note that until the "finish_wait()", or until | |
1286 | * we see the WQ_FLAG_WOKEN flag, we need to | |
1287 | * be very careful with the 'wait->flags', because | |
1288 | * we may race with a waker that sets them. | |
1289 | */ | |
2a9127fc | 1290 | for (;;) { |
5ef64cc8 LT |
1291 | unsigned int flags; |
1292 | ||
62906027 NP |
1293 | set_current_state(state); |
1294 | ||
5ef64cc8 LT |
1295 | /* Loop until we've been woken or interrupted */ |
1296 | flags = smp_load_acquire(&wait->flags); | |
1297 | if (!(flags & WQ_FLAG_WOKEN)) { | |
1298 | if (signal_pending_state(state, current)) | |
1299 | break; | |
1300 | ||
1301 | io_schedule(); | |
1302 | continue; | |
1303 | } | |
1304 | ||
1305 | /* If we were non-exclusive, we're done */ | |
1306 | if (behavior != EXCLUSIVE) | |
a8b169af | 1307 | break; |
9a1ea439 | 1308 | |
5ef64cc8 LT |
1309 | /* If the waker got the lock for us, we're done */ |
1310 | if (flags & WQ_FLAG_DONE) | |
9a1ea439 | 1311 | break; |
2a9127fc | 1312 | |
5ef64cc8 LT |
1313 | /* |
1314 | * Otherwise, if we're getting the lock, we need to | |
1315 | * try to get it ourselves. | |
1316 | * | |
1317 | * And if that fails, we'll have to retry this all. | |
1318 | */ | |
1319 | if (unlikely(test_and_set_bit(bit_nr, &page->flags))) | |
1320 | goto repeat; | |
1321 | ||
1322 | wait->flags |= WQ_FLAG_DONE; | |
1323 | break; | |
62906027 NP |
1324 | } |
1325 | ||
5ef64cc8 LT |
1326 | /* |
1327 | * If a signal happened, this 'finish_wait()' may remove the last | |
1328 | * waiter from the wait-queues, but the PageWaiters bit will remain | |
1329 | * set. That's ok. The next wakeup will take care of it, and trying | |
1330 | * to do it here would be difficult and prone to races. | |
1331 | */ | |
62906027 NP |
1332 | finish_wait(q, wait); |
1333 | ||
eb414681 | 1334 | if (thrashing) { |
9a1ea439 | 1335 | if (delayacct) |
eb414681 JW |
1336 | delayacct_thrashing_end(); |
1337 | psi_memstall_leave(&pflags); | |
1338 | } | |
b1d29ba8 | 1339 | |
62906027 | 1340 | /* |
5ef64cc8 LT |
1341 | * NOTE! The wait->flags weren't stable until we've done the |
1342 | * 'finish_wait()', and we could have exited the loop above due | |
1343 | * to a signal, and had a wakeup event happen after the signal | |
1344 | * test but before the 'finish_wait()'. | |
1345 | * | |
1346 | * So only after the finish_wait() can we reliably determine | |
1347 | * if we got woken up or not, so we can now figure out the final | |
1348 | * return value based on that state without races. | |
1349 | * | |
1350 | * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive | |
1351 | * waiter, but an exclusive one requires WQ_FLAG_DONE. | |
62906027 | 1352 | */ |
5ef64cc8 LT |
1353 | if (behavior == EXCLUSIVE) |
1354 | return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR; | |
62906027 | 1355 | |
2a9127fc | 1356 | return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR; |
62906027 NP |
1357 | } |
1358 | ||
1359 | void wait_on_page_bit(struct page *page, int bit_nr) | |
1360 | { | |
1361 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 | 1362 | wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
62906027 NP |
1363 | } |
1364 | EXPORT_SYMBOL(wait_on_page_bit); | |
1365 | ||
1366 | int wait_on_page_bit_killable(struct page *page, int bit_nr) | |
1367 | { | |
1368 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 | 1369 | return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED); |
cbbce822 | 1370 | } |
4343d008 | 1371 | EXPORT_SYMBOL(wait_on_page_bit_killable); |
cbbce822 | 1372 | |
9a1ea439 HD |
1373 | /** |
1374 | * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked | |
1375 | * @page: The page to wait for. | |
48054625 | 1376 | * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc). |
9a1ea439 HD |
1377 | * |
1378 | * The caller should hold a reference on @page. They expect the page to | |
1379 | * become unlocked relatively soon, but do not wish to hold up migration | |
1380 | * (for example) by holding the reference while waiting for the page to | |
1381 | * come unlocked. After this function returns, the caller should not | |
1382 | * dereference @page. | |
48054625 MWO |
1383 | * |
1384 | * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal. | |
9a1ea439 | 1385 | */ |
48054625 | 1386 | int put_and_wait_on_page_locked(struct page *page, int state) |
9a1ea439 HD |
1387 | { |
1388 | wait_queue_head_t *q; | |
1389 | ||
1390 | page = compound_head(page); | |
1391 | q = page_waitqueue(page); | |
48054625 | 1392 | return wait_on_page_bit_common(q, page, PG_locked, state, DROP); |
9a1ea439 HD |
1393 | } |
1394 | ||
385e1ca5 DH |
1395 | /** |
1396 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
1397 | * @page: Page defining the wait queue of interest |
1398 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
1399 | * |
1400 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
1401 | */ | |
ac6424b9 | 1402 | void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter) |
385e1ca5 DH |
1403 | { |
1404 | wait_queue_head_t *q = page_waitqueue(page); | |
1405 | unsigned long flags; | |
1406 | ||
1407 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1408 | __add_wait_queue_entry_tail(q, waiter); |
62906027 | 1409 | SetPageWaiters(page); |
385e1ca5 DH |
1410 | spin_unlock_irqrestore(&q->lock, flags); |
1411 | } | |
1412 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
1413 | ||
b91e1302 LT |
1414 | #ifndef clear_bit_unlock_is_negative_byte |
1415 | ||
1416 | /* | |
1417 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1418 | * | |
1419 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1420 | * test the sign bit at the same time. But if the architecture does | |
1421 | * not support that special operation, we just do this all by hand | |
1422 | * instead. | |
1423 | * | |
1424 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
ffceeb62 | 1425 | * being cleared, but a memory barrier should be unnecessary since it is |
b91e1302 LT |
1426 | * in the same byte as PG_locked. |
1427 | */ | |
1428 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1429 | { | |
1430 | clear_bit_unlock(nr, mem); | |
1431 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1432 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1433 | } |
1434 | ||
1435 | #endif | |
1436 | ||
1da177e4 | 1437 | /** |
485bb99b | 1438 | * unlock_page - unlock a locked page |
1da177e4 LT |
1439 | * @page: the page |
1440 | * | |
0e9aa675 | 1441 | * Unlocks the page and wakes up sleepers in wait_on_page_locked(). |
1da177e4 | 1442 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup |
da3dae54 | 1443 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
1444 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
1445 | * | |
b91e1302 LT |
1446 | * Note that this depends on PG_waiters being the sign bit in the byte |
1447 | * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to | |
1448 | * clear the PG_locked bit and test PG_waiters at the same time fairly | |
1449 | * portably (architectures that do LL/SC can test any bit, while x86 can | |
1450 | * test the sign bit). | |
1da177e4 | 1451 | */ |
920c7a5d | 1452 | void unlock_page(struct page *page) |
1da177e4 | 1453 | { |
b91e1302 | 1454 | BUILD_BUG_ON(PG_waiters != 7); |
48c935ad | 1455 | page = compound_head(page); |
309381fe | 1456 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
b91e1302 LT |
1457 | if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags)) |
1458 | wake_up_page_bit(page, PG_locked); | |
1da177e4 LT |
1459 | } |
1460 | EXPORT_SYMBOL(unlock_page); | |
1461 | ||
73e10ded DH |
1462 | /** |
1463 | * end_page_private_2 - Clear PG_private_2 and release any waiters | |
1464 | * @page: The page | |
1465 | * | |
1466 | * Clear the PG_private_2 bit on a page and wake up any sleepers waiting for | |
1467 | * this. The page ref held for PG_private_2 being set is released. | |
1468 | * | |
1469 | * This is, for example, used when a netfs page is being written to a local | |
1470 | * disk cache, thereby allowing writes to the cache for the same page to be | |
1471 | * serialised. | |
1472 | */ | |
1473 | void end_page_private_2(struct page *page) | |
1474 | { | |
1475 | page = compound_head(page); | |
1476 | VM_BUG_ON_PAGE(!PagePrivate2(page), page); | |
1477 | clear_bit_unlock(PG_private_2, &page->flags); | |
1478 | wake_up_page_bit(page, PG_private_2); | |
1479 | put_page(page); | |
1480 | } | |
1481 | EXPORT_SYMBOL(end_page_private_2); | |
1482 | ||
1483 | /** | |
1484 | * wait_on_page_private_2 - Wait for PG_private_2 to be cleared on a page | |
1485 | * @page: The page to wait on | |
1486 | * | |
1487 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page. | |
1488 | */ | |
1489 | void wait_on_page_private_2(struct page *page) | |
1490 | { | |
1491 | page = compound_head(page); | |
1492 | while (PagePrivate2(page)) | |
1493 | wait_on_page_bit(page, PG_private_2); | |
1494 | } | |
1495 | EXPORT_SYMBOL(wait_on_page_private_2); | |
1496 | ||
1497 | /** | |
1498 | * wait_on_page_private_2_killable - Wait for PG_private_2 to be cleared on a page | |
1499 | * @page: The page to wait on | |
1500 | * | |
1501 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page or until a | |
1502 | * fatal signal is received by the calling task. | |
1503 | * | |
1504 | * Return: | |
1505 | * - 0 if successful. | |
1506 | * - -EINTR if a fatal signal was encountered. | |
1507 | */ | |
1508 | int wait_on_page_private_2_killable(struct page *page) | |
1509 | { | |
1510 | int ret = 0; | |
1511 | ||
1512 | page = compound_head(page); | |
1513 | while (PagePrivate2(page)) { | |
1514 | ret = wait_on_page_bit_killable(page, PG_private_2); | |
1515 | if (ret < 0) | |
1516 | break; | |
1517 | } | |
1518 | ||
1519 | return ret; | |
1520 | } | |
1521 | EXPORT_SYMBOL(wait_on_page_private_2_killable); | |
1522 | ||
485bb99b RD |
1523 | /** |
1524 | * end_page_writeback - end writeback against a page | |
1525 | * @page: the page | |
1da177e4 LT |
1526 | */ |
1527 | void end_page_writeback(struct page *page) | |
1528 | { | |
888cf2db MG |
1529 | /* |
1530 | * TestClearPageReclaim could be used here but it is an atomic | |
1531 | * operation and overkill in this particular case. Failing to | |
1532 | * shuffle a page marked for immediate reclaim is too mild to | |
1533 | * justify taking an atomic operation penalty at the end of | |
1534 | * ever page writeback. | |
1535 | */ | |
1536 | if (PageReclaim(page)) { | |
1537 | ClearPageReclaim(page); | |
ac6aadb2 | 1538 | rotate_reclaimable_page(page); |
888cf2db | 1539 | } |
ac6aadb2 | 1540 | |
073861ed HD |
1541 | /* |
1542 | * Writeback does not hold a page reference of its own, relying | |
1543 | * on truncation to wait for the clearing of PG_writeback. | |
1544 | * But here we must make sure that the page is not freed and | |
1545 | * reused before the wake_up_page(). | |
1546 | */ | |
1547 | get_page(page); | |
ac6aadb2 MS |
1548 | if (!test_clear_page_writeback(page)) |
1549 | BUG(); | |
1550 | ||
4e857c58 | 1551 | smp_mb__after_atomic(); |
1da177e4 | 1552 | wake_up_page(page, PG_writeback); |
073861ed | 1553 | put_page(page); |
1da177e4 LT |
1554 | } |
1555 | EXPORT_SYMBOL(end_page_writeback); | |
1556 | ||
57d99845 MW |
1557 | /* |
1558 | * After completing I/O on a page, call this routine to update the page | |
1559 | * flags appropriately | |
1560 | */ | |
c11f0c0b | 1561 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1562 | { |
c11f0c0b | 1563 | if (!is_write) { |
57d99845 MW |
1564 | if (!err) { |
1565 | SetPageUptodate(page); | |
1566 | } else { | |
1567 | ClearPageUptodate(page); | |
1568 | SetPageError(page); | |
1569 | } | |
1570 | unlock_page(page); | |
abf54548 | 1571 | } else { |
57d99845 | 1572 | if (err) { |
dd8416c4 MK |
1573 | struct address_space *mapping; |
1574 | ||
57d99845 | 1575 | SetPageError(page); |
dd8416c4 MK |
1576 | mapping = page_mapping(page); |
1577 | if (mapping) | |
1578 | mapping_set_error(mapping, err); | |
57d99845 MW |
1579 | } |
1580 | end_page_writeback(page); | |
1581 | } | |
1582 | } | |
1583 | EXPORT_SYMBOL_GPL(page_endio); | |
1584 | ||
485bb99b RD |
1585 | /** |
1586 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
87066755 | 1587 | * @__page: the page to lock |
1da177e4 | 1588 | */ |
62906027 | 1589 | void __lock_page(struct page *__page) |
1da177e4 | 1590 | { |
62906027 NP |
1591 | struct page *page = compound_head(__page); |
1592 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 HD |
1593 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, |
1594 | EXCLUSIVE); | |
1da177e4 LT |
1595 | } |
1596 | EXPORT_SYMBOL(__lock_page); | |
1597 | ||
62906027 | 1598 | int __lock_page_killable(struct page *__page) |
2687a356 | 1599 | { |
62906027 NP |
1600 | struct page *page = compound_head(__page); |
1601 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 HD |
1602 | return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, |
1603 | EXCLUSIVE); | |
2687a356 | 1604 | } |
18bc0bbd | 1605 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 1606 | |
dd3e6d50 JA |
1607 | int __lock_page_async(struct page *page, struct wait_page_queue *wait) |
1608 | { | |
f32b5dd7 MWO |
1609 | struct wait_queue_head *q = page_waitqueue(page); |
1610 | int ret = 0; | |
1611 | ||
1612 | wait->page = page; | |
1613 | wait->bit_nr = PG_locked; | |
1614 | ||
1615 | spin_lock_irq(&q->lock); | |
1616 | __add_wait_queue_entry_tail(q, &wait->wait); | |
1617 | SetPageWaiters(page); | |
1618 | ret = !trylock_page(page); | |
1619 | /* | |
1620 | * If we were successful now, we know we're still on the | |
1621 | * waitqueue as we're still under the lock. This means it's | |
1622 | * safe to remove and return success, we know the callback | |
1623 | * isn't going to trigger. | |
1624 | */ | |
1625 | if (!ret) | |
1626 | __remove_wait_queue(q, &wait->wait); | |
1627 | else | |
1628 | ret = -EIOCBQUEUED; | |
1629 | spin_unlock_irq(&q->lock); | |
1630 | return ret; | |
dd3e6d50 JA |
1631 | } |
1632 | ||
9a95f3cf PC |
1633 | /* |
1634 | * Return values: | |
c1e8d7c6 | 1635 | * 1 - page is locked; mmap_lock is still held. |
9a95f3cf | 1636 | * 0 - page is not locked. |
3e4e28c5 | 1637 | * mmap_lock has been released (mmap_read_unlock(), unless flags had both |
9a95f3cf | 1638 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
c1e8d7c6 | 1639 | * which case mmap_lock is still held. |
9a95f3cf PC |
1640 | * |
1641 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
c1e8d7c6 | 1642 | * with the page locked and the mmap_lock unperturbed. |
9a95f3cf | 1643 | */ |
d065bd81 ML |
1644 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
1645 | unsigned int flags) | |
1646 | { | |
4064b982 | 1647 | if (fault_flag_allow_retry_first(flags)) { |
37b23e05 | 1648 | /* |
c1e8d7c6 | 1649 | * CAUTION! In this case, mmap_lock is not released |
37b23e05 KM |
1650 | * even though return 0. |
1651 | */ | |
1652 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
1653 | return 0; | |
1654 | ||
d8ed45c5 | 1655 | mmap_read_unlock(mm); |
37b23e05 KM |
1656 | if (flags & FAULT_FLAG_KILLABLE) |
1657 | wait_on_page_locked_killable(page); | |
1658 | else | |
318b275f | 1659 | wait_on_page_locked(page); |
d065bd81 | 1660 | return 0; |
800bca7c HL |
1661 | } |
1662 | if (flags & FAULT_FLAG_KILLABLE) { | |
1663 | int ret; | |
37b23e05 | 1664 | |
800bca7c HL |
1665 | ret = __lock_page_killable(page); |
1666 | if (ret) { | |
1667 | mmap_read_unlock(mm); | |
1668 | return 0; | |
1669 | } | |
1670 | } else { | |
1671 | __lock_page(page); | |
d065bd81 | 1672 | } |
800bca7c HL |
1673 | return 1; |
1674 | ||
d065bd81 ML |
1675 | } |
1676 | ||
e7b563bb | 1677 | /** |
0d3f9296 MW |
1678 | * page_cache_next_miss() - Find the next gap in the page cache. |
1679 | * @mapping: Mapping. | |
1680 | * @index: Index. | |
1681 | * @max_scan: Maximum range to search. | |
e7b563bb | 1682 | * |
0d3f9296 MW |
1683 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1684 | * gap with the lowest index. | |
e7b563bb | 1685 | * |
0d3f9296 MW |
1686 | * This function may be called under the rcu_read_lock. However, this will |
1687 | * not atomically search a snapshot of the cache at a single point in time. | |
1688 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1689 | * created at index 10, page_cache_next_miss covering both indices may | |
1690 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1691 | * |
0d3f9296 MW |
1692 | * Return: The index of the gap if found, otherwise an index outside the |
1693 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1694 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1695 | */ |
0d3f9296 | 1696 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1697 | pgoff_t index, unsigned long max_scan) |
1698 | { | |
0d3f9296 | 1699 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1700 | |
0d3f9296 MW |
1701 | while (max_scan--) { |
1702 | void *entry = xas_next(&xas); | |
1703 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1704 | break; |
0d3f9296 | 1705 | if (xas.xa_index == 0) |
e7b563bb JW |
1706 | break; |
1707 | } | |
1708 | ||
0d3f9296 | 1709 | return xas.xa_index; |
e7b563bb | 1710 | } |
0d3f9296 | 1711 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1712 | |
1713 | /** | |
2346a560 | 1714 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
0d3f9296 MW |
1715 | * @mapping: Mapping. |
1716 | * @index: Index. | |
1717 | * @max_scan: Maximum range to search. | |
e7b563bb | 1718 | * |
0d3f9296 MW |
1719 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1720 | * gap with the highest index. | |
e7b563bb | 1721 | * |
0d3f9296 MW |
1722 | * This function may be called under the rcu_read_lock. However, this will |
1723 | * not atomically search a snapshot of the cache at a single point in time. | |
1724 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1725 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1726 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1727 | * |
0d3f9296 MW |
1728 | * Return: The index of the gap if found, otherwise an index outside the |
1729 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1730 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1731 | */ |
0d3f9296 | 1732 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1733 | pgoff_t index, unsigned long max_scan) |
1734 | { | |
0d3f9296 | 1735 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1736 | |
0d3f9296 MW |
1737 | while (max_scan--) { |
1738 | void *entry = xas_prev(&xas); | |
1739 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1740 | break; |
0d3f9296 | 1741 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1742 | break; |
1743 | } | |
1744 | ||
0d3f9296 | 1745 | return xas.xa_index; |
e7b563bb | 1746 | } |
0d3f9296 | 1747 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1748 | |
44835d20 | 1749 | /* |
bc5a3011 | 1750 | * mapping_get_entry - Get a page cache entry. |
485bb99b | 1751 | * @mapping: the address_space to search |
a6de4b48 | 1752 | * @index: The page cache index. |
0cd6144a | 1753 | * |
4b17f030 | 1754 | * Looks up the page cache slot at @mapping & @index. If there is a |
a6de4b48 | 1755 | * page cache page, the head page is returned with an increased refcount. |
485bb99b | 1756 | * |
139b6a6f JW |
1757 | * If the slot holds a shadow entry of a previously evicted page, or a |
1758 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a | 1759 | * |
a6de4b48 | 1760 | * Return: The head page or shadow entry, %NULL if nothing is found. |
1da177e4 | 1761 | */ |
bc5a3011 MWO |
1762 | static struct page *mapping_get_entry(struct address_space *mapping, |
1763 | pgoff_t index) | |
1da177e4 | 1764 | { |
a6de4b48 | 1765 | XA_STATE(xas, &mapping->i_pages, index); |
4101196b | 1766 | struct page *page; |
1da177e4 | 1767 | |
a60637c8 NP |
1768 | rcu_read_lock(); |
1769 | repeat: | |
4c7472c0 MW |
1770 | xas_reset(&xas); |
1771 | page = xas_load(&xas); | |
1772 | if (xas_retry(&xas, page)) | |
1773 | goto repeat; | |
1774 | /* | |
1775 | * A shadow entry of a recently evicted page, or a swap entry from | |
1776 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1777 | */ | |
1778 | if (!page || xa_is_value(page)) | |
1779 | goto out; | |
83929372 | 1780 | |
4101196b | 1781 | if (!page_cache_get_speculative(page)) |
4c7472c0 | 1782 | goto repeat; |
83929372 | 1783 | |
4c7472c0 | 1784 | /* |
4101196b | 1785 | * Has the page moved or been split? |
4c7472c0 MW |
1786 | * This is part of the lockless pagecache protocol. See |
1787 | * include/linux/pagemap.h for details. | |
1788 | */ | |
1789 | if (unlikely(page != xas_reload(&xas))) { | |
4101196b | 1790 | put_page(page); |
4c7472c0 | 1791 | goto repeat; |
a60637c8 | 1792 | } |
27d20fdd | 1793 | out: |
a60637c8 NP |
1794 | rcu_read_unlock(); |
1795 | ||
1da177e4 LT |
1796 | return page; |
1797 | } | |
1da177e4 | 1798 | |
0cd6144a | 1799 | /** |
2294b32e MWO |
1800 | * pagecache_get_page - Find and get a reference to a page. |
1801 | * @mapping: The address_space to search. | |
1802 | * @index: The page index. | |
1803 | * @fgp_flags: %FGP flags modify how the page is returned. | |
1804 | * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified. | |
1da177e4 | 1805 | * |
2294b32e | 1806 | * Looks up the page cache entry at @mapping & @index. |
0cd6144a | 1807 | * |
2294b32e | 1808 | * @fgp_flags can be zero or more of these flags: |
0e056eb5 | 1809 | * |
2294b32e MWO |
1810 | * * %FGP_ACCESSED - The page will be marked accessed. |
1811 | * * %FGP_LOCK - The page is returned locked. | |
a8cf7f27 MWO |
1812 | * * %FGP_HEAD - If the page is present and a THP, return the head page |
1813 | * rather than the exact page specified by the index. | |
44835d20 MWO |
1814 | * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it |
1815 | * instead of allocating a new page to replace it. | |
2294b32e MWO |
1816 | * * %FGP_CREAT - If no page is present then a new page is allocated using |
1817 | * @gfp_mask and added to the page cache and the VM's LRU list. | |
1818 | * The page is returned locked and with an increased refcount. | |
1819 | * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the | |
1820 | * page is already in cache. If the page was allocated, unlock it before | |
1821 | * returning so the caller can do the same dance. | |
605cad83 YS |
1822 | * * %FGP_WRITE - The page will be written |
1823 | * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask | |
1824 | * * %FGP_NOWAIT - Don't get blocked by page lock | |
1da177e4 | 1825 | * |
2294b32e MWO |
1826 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even |
1827 | * if the %GFP flags specified for %FGP_CREAT are atomic. | |
1da177e4 | 1828 | * |
2457aec6 | 1829 | * If there is a page cache page, it is returned with an increased refcount. |
a862f68a | 1830 | * |
2294b32e | 1831 | * Return: The found page or %NULL otherwise. |
1da177e4 | 1832 | */ |
2294b32e MWO |
1833 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index, |
1834 | int fgp_flags, gfp_t gfp_mask) | |
1da177e4 | 1835 | { |
eb2be189 | 1836 | struct page *page; |
2457aec6 | 1837 | |
1da177e4 | 1838 | repeat: |
bc5a3011 | 1839 | page = mapping_get_entry(mapping, index); |
44835d20 MWO |
1840 | if (xa_is_value(page)) { |
1841 | if (fgp_flags & FGP_ENTRY) | |
1842 | return page; | |
2457aec6 | 1843 | page = NULL; |
44835d20 | 1844 | } |
2457aec6 MG |
1845 | if (!page) |
1846 | goto no_page; | |
1847 | ||
1848 | if (fgp_flags & FGP_LOCK) { | |
1849 | if (fgp_flags & FGP_NOWAIT) { | |
1850 | if (!trylock_page(page)) { | |
09cbfeaf | 1851 | put_page(page); |
2457aec6 MG |
1852 | return NULL; |
1853 | } | |
1854 | } else { | |
1855 | lock_page(page); | |
1856 | } | |
1857 | ||
1858 | /* Has the page been truncated? */ | |
a8cf7f27 | 1859 | if (unlikely(page->mapping != mapping)) { |
2457aec6 | 1860 | unlock_page(page); |
09cbfeaf | 1861 | put_page(page); |
2457aec6 MG |
1862 | goto repeat; |
1863 | } | |
a8cf7f27 | 1864 | VM_BUG_ON_PAGE(!thp_contains(page, index), page); |
2457aec6 MG |
1865 | } |
1866 | ||
c16eb000 | 1867 | if (fgp_flags & FGP_ACCESSED) |
2457aec6 | 1868 | mark_page_accessed(page); |
b9306a79 YS |
1869 | else if (fgp_flags & FGP_WRITE) { |
1870 | /* Clear idle flag for buffer write */ | |
1871 | if (page_is_idle(page)) | |
1872 | clear_page_idle(page); | |
1873 | } | |
a8cf7f27 MWO |
1874 | if (!(fgp_flags & FGP_HEAD)) |
1875 | page = find_subpage(page, index); | |
2457aec6 MG |
1876 | |
1877 | no_page: | |
1878 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1879 | int err; | |
f56753ac | 1880 | if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping)) |
45f87de5 MH |
1881 | gfp_mask |= __GFP_WRITE; |
1882 | if (fgp_flags & FGP_NOFS) | |
1883 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1884 | |
45f87de5 | 1885 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1886 | if (!page) |
1887 | return NULL; | |
2457aec6 | 1888 | |
a75d4c33 | 1889 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
2457aec6 MG |
1890 | fgp_flags |= FGP_LOCK; |
1891 | ||
eb39d618 | 1892 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1893 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1894 | __SetPageReferenced(page); |
2457aec6 | 1895 | |
2294b32e | 1896 | err = add_to_page_cache_lru(page, mapping, index, gfp_mask); |
eb2be189 | 1897 | if (unlikely(err)) { |
09cbfeaf | 1898 | put_page(page); |
eb2be189 NP |
1899 | page = NULL; |
1900 | if (err == -EEXIST) | |
1901 | goto repeat; | |
1da177e4 | 1902 | } |
a75d4c33 JB |
1903 | |
1904 | /* | |
1905 | * add_to_page_cache_lru locks the page, and for mmap we expect | |
1906 | * an unlocked page. | |
1907 | */ | |
1908 | if (page && (fgp_flags & FGP_FOR_MMAP)) | |
1909 | unlock_page(page); | |
1da177e4 | 1910 | } |
2457aec6 | 1911 | |
1da177e4 LT |
1912 | return page; |
1913 | } | |
2457aec6 | 1914 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1915 | |
c7bad633 MWO |
1916 | static inline struct page *find_get_entry(struct xa_state *xas, pgoff_t max, |
1917 | xa_mark_t mark) | |
1918 | { | |
1919 | struct page *page; | |
1920 | ||
1921 | retry: | |
1922 | if (mark == XA_PRESENT) | |
1923 | page = xas_find(xas, max); | |
1924 | else | |
1925 | page = xas_find_marked(xas, max, mark); | |
1926 | ||
1927 | if (xas_retry(xas, page)) | |
1928 | goto retry; | |
1929 | /* | |
1930 | * A shadow entry of a recently evicted page, a swap | |
1931 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1932 | * without attempting to raise page count. | |
1933 | */ | |
1934 | if (!page || xa_is_value(page)) | |
1935 | return page; | |
1936 | ||
1937 | if (!page_cache_get_speculative(page)) | |
1938 | goto reset; | |
1939 | ||
1940 | /* Has the page moved or been split? */ | |
1941 | if (unlikely(page != xas_reload(xas))) { | |
1942 | put_page(page); | |
1943 | goto reset; | |
1944 | } | |
1945 | ||
1946 | return page; | |
1947 | reset: | |
1948 | xas_reset(xas); | |
1949 | goto retry; | |
1950 | } | |
1951 | ||
0cd6144a JW |
1952 | /** |
1953 | * find_get_entries - gang pagecache lookup | |
1954 | * @mapping: The address_space to search | |
1955 | * @start: The starting page cache index | |
ca122fe4 | 1956 | * @end: The final page index (inclusive). |
cf2039af | 1957 | * @pvec: Where the resulting entries are placed. |
0cd6144a JW |
1958 | * @indices: The cache indices corresponding to the entries in @entries |
1959 | * | |
cf2039af MWO |
1960 | * find_get_entries() will search for and return a batch of entries in |
1961 | * the mapping. The entries are placed in @pvec. find_get_entries() | |
1962 | * takes a reference on any actual pages it returns. | |
0cd6144a JW |
1963 | * |
1964 | * The search returns a group of mapping-contiguous page cache entries | |
1965 | * with ascending indexes. There may be holes in the indices due to | |
1966 | * not-present pages. | |
1967 | * | |
139b6a6f JW |
1968 | * Any shadow entries of evicted pages, or swap entries from |
1969 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a | 1970 | * |
71725ed1 HD |
1971 | * If it finds a Transparent Huge Page, head or tail, find_get_entries() |
1972 | * stops at that page: the caller is likely to have a better way to handle | |
1973 | * the compound page as a whole, and then skip its extent, than repeatedly | |
1974 | * calling find_get_entries() to return all its tails. | |
1975 | * | |
a862f68a | 1976 | * Return: the number of pages and shadow entries which were found. |
0cd6144a | 1977 | */ |
ca122fe4 | 1978 | unsigned find_get_entries(struct address_space *mapping, pgoff_t start, |
cf2039af | 1979 | pgoff_t end, struct pagevec *pvec, pgoff_t *indices) |
0cd6144a | 1980 | { |
f280bf09 MW |
1981 | XA_STATE(xas, &mapping->i_pages, start); |
1982 | struct page *page; | |
0cd6144a | 1983 | unsigned int ret = 0; |
cf2039af | 1984 | unsigned nr_entries = PAGEVEC_SIZE; |
0cd6144a JW |
1985 | |
1986 | rcu_read_lock(); | |
ca122fe4 | 1987 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { |
71725ed1 HD |
1988 | /* |
1989 | * Terminate early on finding a THP, to allow the caller to | |
1990 | * handle it all at once; but continue if this is hugetlbfs. | |
1991 | */ | |
c7bad633 MWO |
1992 | if (!xa_is_value(page) && PageTransHuge(page) && |
1993 | !PageHuge(page)) { | |
71725ed1 HD |
1994 | page = find_subpage(page, xas.xa_index); |
1995 | nr_entries = ret + 1; | |
1996 | } | |
c7bad633 | 1997 | |
f280bf09 | 1998 | indices[ret] = xas.xa_index; |
cf2039af | 1999 | pvec->pages[ret] = page; |
0cd6144a JW |
2000 | if (++ret == nr_entries) |
2001 | break; | |
2002 | } | |
2003 | rcu_read_unlock(); | |
cf2039af MWO |
2004 | |
2005 | pvec->nr = ret; | |
0cd6144a JW |
2006 | return ret; |
2007 | } | |
2008 | ||
5c211ba2 MWO |
2009 | /** |
2010 | * find_lock_entries - Find a batch of pagecache entries. | |
2011 | * @mapping: The address_space to search. | |
2012 | * @start: The starting page cache index. | |
2013 | * @end: The final page index (inclusive). | |
2014 | * @pvec: Where the resulting entries are placed. | |
2015 | * @indices: The cache indices of the entries in @pvec. | |
2016 | * | |
2017 | * find_lock_entries() will return a batch of entries from @mapping. | |
2018 | * Swap, shadow and DAX entries are included. Pages are returned | |
2019 | * locked and with an incremented refcount. Pages which are locked by | |
2020 | * somebody else or under writeback are skipped. Only the head page of | |
2021 | * a THP is returned. Pages which are partially outside the range are | |
2022 | * not returned. | |
2023 | * | |
2024 | * The entries have ascending indexes. The indices may not be consecutive | |
2025 | * due to not-present entries, THP pages, pages which could not be locked | |
2026 | * or pages under writeback. | |
2027 | * | |
2028 | * Return: The number of entries which were found. | |
2029 | */ | |
2030 | unsigned find_lock_entries(struct address_space *mapping, pgoff_t start, | |
2031 | pgoff_t end, struct pagevec *pvec, pgoff_t *indices) | |
2032 | { | |
2033 | XA_STATE(xas, &mapping->i_pages, start); | |
2034 | struct page *page; | |
2035 | ||
2036 | rcu_read_lock(); | |
2037 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { | |
2038 | if (!xa_is_value(page)) { | |
2039 | if (page->index < start) | |
2040 | goto put; | |
2041 | VM_BUG_ON_PAGE(page->index != xas.xa_index, page); | |
2042 | if (page->index + thp_nr_pages(page) - 1 > end) | |
2043 | goto put; | |
2044 | if (!trylock_page(page)) | |
2045 | goto put; | |
2046 | if (page->mapping != mapping || PageWriteback(page)) | |
2047 | goto unlock; | |
2048 | VM_BUG_ON_PAGE(!thp_contains(page, xas.xa_index), | |
2049 | page); | |
2050 | } | |
2051 | indices[pvec->nr] = xas.xa_index; | |
2052 | if (!pagevec_add(pvec, page)) | |
2053 | break; | |
2054 | goto next; | |
2055 | unlock: | |
2056 | unlock_page(page); | |
2057 | put: | |
2058 | put_page(page); | |
2059 | next: | |
2d11e738 HD |
2060 | if (!xa_is_value(page) && PageTransHuge(page)) { |
2061 | unsigned int nr_pages = thp_nr_pages(page); | |
2062 | ||
2063 | /* Final THP may cross MAX_LFS_FILESIZE on 32-bit */ | |
2064 | xas_set(&xas, page->index + nr_pages); | |
2065 | if (xas.xa_index < nr_pages) | |
2066 | break; | |
2067 | } | |
5c211ba2 MWO |
2068 | } |
2069 | rcu_read_unlock(); | |
2070 | ||
2071 | return pagevec_count(pvec); | |
2072 | } | |
2073 | ||
1da177e4 | 2074 | /** |
b947cee4 | 2075 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
2076 | * @mapping: The address_space to search |
2077 | * @start: The starting page index | |
b947cee4 | 2078 | * @end: The final page index (inclusive) |
1da177e4 LT |
2079 | * @nr_pages: The maximum number of pages |
2080 | * @pages: Where the resulting pages are placed | |
2081 | * | |
b947cee4 JK |
2082 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
2083 | * pages in the mapping starting at index @start and up to index @end | |
2084 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
2085 | * a reference against the returned pages. | |
1da177e4 LT |
2086 | * |
2087 | * The search returns a group of mapping-contiguous pages with ascending | |
2088 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 2089 | * We also update @start to index the next page for the traversal. |
1da177e4 | 2090 | * |
a862f68a MR |
2091 | * Return: the number of pages which were found. If this number is |
2092 | * smaller than @nr_pages, the end of specified range has been | |
b947cee4 | 2093 | * reached. |
1da177e4 | 2094 | */ |
b947cee4 JK |
2095 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
2096 | pgoff_t end, unsigned int nr_pages, | |
2097 | struct page **pages) | |
1da177e4 | 2098 | { |
fd1b3cee MW |
2099 | XA_STATE(xas, &mapping->i_pages, *start); |
2100 | struct page *page; | |
0fc9d104 KK |
2101 | unsigned ret = 0; |
2102 | ||
2103 | if (unlikely(!nr_pages)) | |
2104 | return 0; | |
a60637c8 NP |
2105 | |
2106 | rcu_read_lock(); | |
c7bad633 | 2107 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { |
fd1b3cee MW |
2108 | /* Skip over shadow, swap and DAX entries */ |
2109 | if (xa_is_value(page)) | |
8079b1c8 | 2110 | continue; |
a60637c8 | 2111 | |
4101196b | 2112 | pages[ret] = find_subpage(page, xas.xa_index); |
b947cee4 | 2113 | if (++ret == nr_pages) { |
5d3ee42f | 2114 | *start = xas.xa_index + 1; |
b947cee4 JK |
2115 | goto out; |
2116 | } | |
a60637c8 | 2117 | } |
5b280c0c | 2118 | |
b947cee4 JK |
2119 | /* |
2120 | * We come here when there is no page beyond @end. We take care to not | |
2121 | * overflow the index @start as it confuses some of the callers. This | |
fd1b3cee | 2122 | * breaks the iteration when there is a page at index -1 but that is |
b947cee4 JK |
2123 | * already broken anyway. |
2124 | */ | |
2125 | if (end == (pgoff_t)-1) | |
2126 | *start = (pgoff_t)-1; | |
2127 | else | |
2128 | *start = end + 1; | |
2129 | out: | |
a60637c8 | 2130 | rcu_read_unlock(); |
d72dc8a2 | 2131 | |
1da177e4 LT |
2132 | return ret; |
2133 | } | |
2134 | ||
ebf43500 JA |
2135 | /** |
2136 | * find_get_pages_contig - gang contiguous pagecache lookup | |
2137 | * @mapping: The address_space to search | |
2138 | * @index: The starting page index | |
2139 | * @nr_pages: The maximum number of pages | |
2140 | * @pages: Where the resulting pages are placed | |
2141 | * | |
2142 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
2143 | * that the returned number of pages are guaranteed to be contiguous. | |
2144 | * | |
a862f68a | 2145 | * Return: the number of pages which were found. |
ebf43500 JA |
2146 | */ |
2147 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
2148 | unsigned int nr_pages, struct page **pages) | |
2149 | { | |
3ece58a2 MW |
2150 | XA_STATE(xas, &mapping->i_pages, index); |
2151 | struct page *page; | |
0fc9d104 KK |
2152 | unsigned int ret = 0; |
2153 | ||
2154 | if (unlikely(!nr_pages)) | |
2155 | return 0; | |
a60637c8 NP |
2156 | |
2157 | rcu_read_lock(); | |
3ece58a2 | 2158 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { |
3ece58a2 MW |
2159 | if (xas_retry(&xas, page)) |
2160 | continue; | |
2161 | /* | |
2162 | * If the entry has been swapped out, we can stop looking. | |
2163 | * No current caller is looking for DAX entries. | |
2164 | */ | |
2165 | if (xa_is_value(page)) | |
8079b1c8 | 2166 | break; |
ebf43500 | 2167 | |
4101196b | 2168 | if (!page_cache_get_speculative(page)) |
3ece58a2 | 2169 | goto retry; |
83929372 | 2170 | |
4101196b | 2171 | /* Has the page moved or been split? */ |
3ece58a2 MW |
2172 | if (unlikely(page != xas_reload(&xas))) |
2173 | goto put_page; | |
a60637c8 | 2174 | |
4101196b | 2175 | pages[ret] = find_subpage(page, xas.xa_index); |
0fc9d104 KK |
2176 | if (++ret == nr_pages) |
2177 | break; | |
3ece58a2 MW |
2178 | continue; |
2179 | put_page: | |
4101196b | 2180 | put_page(page); |
3ece58a2 MW |
2181 | retry: |
2182 | xas_reset(&xas); | |
ebf43500 | 2183 | } |
a60637c8 NP |
2184 | rcu_read_unlock(); |
2185 | return ret; | |
ebf43500 | 2186 | } |
ef71c15c | 2187 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 2188 | |
485bb99b | 2189 | /** |
c49f50d1 | 2190 | * find_get_pages_range_tag - Find and return head pages matching @tag. |
485bb99b RD |
2191 | * @mapping: the address_space to search |
2192 | * @index: the starting page index | |
72b045ae | 2193 | * @end: The final page index (inclusive) |
485bb99b RD |
2194 | * @tag: the tag index |
2195 | * @nr_pages: the maximum number of pages | |
2196 | * @pages: where the resulting pages are placed | |
2197 | * | |
c49f50d1 MWO |
2198 | * Like find_get_pages(), except we only return head pages which are tagged |
2199 | * with @tag. @index is updated to the index immediately after the last | |
2200 | * page we return, ready for the next iteration. | |
a862f68a MR |
2201 | * |
2202 | * Return: the number of pages which were found. | |
1da177e4 | 2203 | */ |
72b045ae | 2204 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
a6906972 | 2205 | pgoff_t end, xa_mark_t tag, unsigned int nr_pages, |
72b045ae | 2206 | struct page **pages) |
1da177e4 | 2207 | { |
a6906972 MW |
2208 | XA_STATE(xas, &mapping->i_pages, *index); |
2209 | struct page *page; | |
0fc9d104 KK |
2210 | unsigned ret = 0; |
2211 | ||
2212 | if (unlikely(!nr_pages)) | |
2213 | return 0; | |
a60637c8 NP |
2214 | |
2215 | rcu_read_lock(); | |
c7bad633 | 2216 | while ((page = find_get_entry(&xas, end, tag))) { |
a6906972 MW |
2217 | /* |
2218 | * Shadow entries should never be tagged, but this iteration | |
2219 | * is lockless so there is a window for page reclaim to evict | |
2220 | * a page we saw tagged. Skip over it. | |
2221 | */ | |
2222 | if (xa_is_value(page)) | |
139b6a6f | 2223 | continue; |
a60637c8 | 2224 | |
c49f50d1 | 2225 | pages[ret] = page; |
72b045ae | 2226 | if (++ret == nr_pages) { |
c49f50d1 | 2227 | *index = page->index + thp_nr_pages(page); |
72b045ae JK |
2228 | goto out; |
2229 | } | |
a60637c8 | 2230 | } |
5b280c0c | 2231 | |
72b045ae | 2232 | /* |
a6906972 | 2233 | * We come here when we got to @end. We take care to not overflow the |
72b045ae | 2234 | * index @index as it confuses some of the callers. This breaks the |
a6906972 MW |
2235 | * iteration when there is a page at index -1 but that is already |
2236 | * broken anyway. | |
72b045ae JK |
2237 | */ |
2238 | if (end == (pgoff_t)-1) | |
2239 | *index = (pgoff_t)-1; | |
2240 | else | |
2241 | *index = end + 1; | |
2242 | out: | |
a60637c8 | 2243 | rcu_read_unlock(); |
1da177e4 | 2244 | |
1da177e4 LT |
2245 | return ret; |
2246 | } | |
72b045ae | 2247 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 2248 | |
76d42bd9 WF |
2249 | /* |
2250 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
2251 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
2252 | * | |
2253 | * ---R__________________________________________B__________ | |
2254 | * ^ reading here ^ bad block(assume 4k) | |
2255 | * | |
2256 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
2257 | * => failing the whole request => read(R) => read(R+1) => | |
2258 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
2259 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
2260 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
2261 | * | |
2262 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
2263 | */ | |
0f8e2db4 | 2264 | static void shrink_readahead_size_eio(struct file_ra_state *ra) |
76d42bd9 | 2265 | { |
76d42bd9 | 2266 | ra->ra_pages /= 4; |
76d42bd9 WF |
2267 | } |
2268 | ||
cbd59c48 MWO |
2269 | /* |
2270 | * filemap_get_read_batch - Get a batch of pages for read | |
2271 | * | |
2272 | * Get a batch of pages which represent a contiguous range of bytes | |
2273 | * in the file. No tail pages will be returned. If @index is in the | |
2274 | * middle of a THP, the entire THP will be returned. The last page in | |
2275 | * the batch may have Readahead set or be not Uptodate so that the | |
2276 | * caller can take the appropriate action. | |
2277 | */ | |
2278 | static void filemap_get_read_batch(struct address_space *mapping, | |
2279 | pgoff_t index, pgoff_t max, struct pagevec *pvec) | |
2280 | { | |
2281 | XA_STATE(xas, &mapping->i_pages, index); | |
2282 | struct page *head; | |
2283 | ||
2284 | rcu_read_lock(); | |
2285 | for (head = xas_load(&xas); head; head = xas_next(&xas)) { | |
2286 | if (xas_retry(&xas, head)) | |
2287 | continue; | |
2288 | if (xas.xa_index > max || xa_is_value(head)) | |
2289 | break; | |
2290 | if (!page_cache_get_speculative(head)) | |
2291 | goto retry; | |
2292 | ||
2293 | /* Has the page moved or been split? */ | |
2294 | if (unlikely(head != xas_reload(&xas))) | |
2295 | goto put_page; | |
2296 | ||
2297 | if (!pagevec_add(pvec, head)) | |
2298 | break; | |
2299 | if (!PageUptodate(head)) | |
2300 | break; | |
2301 | if (PageReadahead(head)) | |
2302 | break; | |
2303 | xas.xa_index = head->index + thp_nr_pages(head) - 1; | |
2304 | xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK; | |
2305 | continue; | |
2306 | put_page: | |
2307 | put_page(head); | |
2308 | retry: | |
2309 | xas_reset(&xas); | |
2310 | } | |
2311 | rcu_read_unlock(); | |
2312 | } | |
2313 | ||
68430303 MWO |
2314 | static int filemap_read_page(struct file *file, struct address_space *mapping, |
2315 | struct page *page) | |
723ef24b | 2316 | { |
723ef24b KO |
2317 | int error; |
2318 | ||
723ef24b | 2319 | /* |
68430303 MWO |
2320 | * A previous I/O error may have been due to temporary failures, |
2321 | * eg. multipath errors. PG_error will be set again if readpage | |
2322 | * fails. | |
723ef24b KO |
2323 | */ |
2324 | ClearPageError(page); | |
2325 | /* Start the actual read. The read will unlock the page. */ | |
68430303 MWO |
2326 | error = mapping->a_ops->readpage(file, page); |
2327 | if (error) | |
2328 | return error; | |
723ef24b | 2329 | |
aa1ec2f6 | 2330 | error = wait_on_page_locked_killable(page); |
68430303 MWO |
2331 | if (error) |
2332 | return error; | |
aa1ec2f6 MWO |
2333 | if (PageUptodate(page)) |
2334 | return 0; | |
aa1ec2f6 MWO |
2335 | shrink_readahead_size_eio(&file->f_ra); |
2336 | return -EIO; | |
723ef24b KO |
2337 | } |
2338 | ||
fce70da3 MWO |
2339 | static bool filemap_range_uptodate(struct address_space *mapping, |
2340 | loff_t pos, struct iov_iter *iter, struct page *page) | |
2341 | { | |
2342 | int count; | |
2343 | ||
2344 | if (PageUptodate(page)) | |
2345 | return true; | |
2346 | /* pipes can't handle partially uptodate pages */ | |
2347 | if (iov_iter_is_pipe(iter)) | |
2348 | return false; | |
2349 | if (!mapping->a_ops->is_partially_uptodate) | |
2350 | return false; | |
2351 | if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page))) | |
2352 | return false; | |
2353 | ||
2354 | count = iter->count; | |
2355 | if (page_offset(page) > pos) { | |
2356 | count -= page_offset(page) - pos; | |
2357 | pos = 0; | |
2358 | } else { | |
2359 | pos -= page_offset(page); | |
2360 | } | |
2361 | ||
2362 | return mapping->a_ops->is_partially_uptodate(page, pos, count); | |
2363 | } | |
2364 | ||
4612aeef MWO |
2365 | static int filemap_update_page(struct kiocb *iocb, |
2366 | struct address_space *mapping, struct iov_iter *iter, | |
fce70da3 | 2367 | struct page *page) |
723ef24b | 2368 | { |
723ef24b KO |
2369 | int error; |
2370 | ||
87d1d7b6 MWO |
2371 | if (!trylock_page(page)) { |
2372 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) | |
2373 | return -EAGAIN; | |
2374 | if (!(iocb->ki_flags & IOCB_WAITQ)) { | |
bd8a1f36 | 2375 | put_and_wait_on_page_locked(page, TASK_KILLABLE); |
4612aeef | 2376 | return AOP_TRUNCATED_PAGE; |
bd8a1f36 | 2377 | } |
87d1d7b6 MWO |
2378 | error = __lock_page_async(page, iocb->ki_waitq); |
2379 | if (error) | |
2380 | return error; | |
723ef24b | 2381 | } |
723ef24b | 2382 | |
bd8a1f36 MWO |
2383 | if (!page->mapping) |
2384 | goto truncated; | |
723ef24b | 2385 | |
fce70da3 MWO |
2386 | error = 0; |
2387 | if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, page)) | |
2388 | goto unlock; | |
2389 | ||
2390 | error = -EAGAIN; | |
2391 | if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ)) | |
2392 | goto unlock; | |
2393 | ||
68430303 | 2394 | error = filemap_read_page(iocb->ki_filp, mapping, page); |
68430303 | 2395 | if (error == AOP_TRUNCATED_PAGE) |
4612aeef MWO |
2396 | put_page(page); |
2397 | return error; | |
bd8a1f36 MWO |
2398 | truncated: |
2399 | unlock_page(page); | |
2400 | put_page(page); | |
4612aeef | 2401 | return AOP_TRUNCATED_PAGE; |
fce70da3 MWO |
2402 | unlock: |
2403 | unlock_page(page); | |
2404 | return error; | |
723ef24b KO |
2405 | } |
2406 | ||
f253e185 MWO |
2407 | static int filemap_create_page(struct file *file, |
2408 | struct address_space *mapping, pgoff_t index, | |
2409 | struct pagevec *pvec) | |
723ef24b | 2410 | { |
723ef24b KO |
2411 | struct page *page; |
2412 | int error; | |
2413 | ||
723ef24b KO |
2414 | page = page_cache_alloc(mapping); |
2415 | if (!page) | |
f253e185 | 2416 | return -ENOMEM; |
723ef24b KO |
2417 | |
2418 | error = add_to_page_cache_lru(page, mapping, index, | |
f253e185 MWO |
2419 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
2420 | if (error == -EEXIST) | |
2421 | error = AOP_TRUNCATED_PAGE; | |
2422 | if (error) | |
2423 | goto error; | |
2424 | ||
2425 | error = filemap_read_page(file, mapping, page); | |
2426 | if (error) | |
2427 | goto error; | |
2428 | ||
2429 | pagevec_add(pvec, page); | |
2430 | return 0; | |
2431 | error: | |
68430303 | 2432 | put_page(page); |
f253e185 | 2433 | return error; |
723ef24b KO |
2434 | } |
2435 | ||
5963fe03 MWO |
2436 | static int filemap_readahead(struct kiocb *iocb, struct file *file, |
2437 | struct address_space *mapping, struct page *page, | |
2438 | pgoff_t last_index) | |
2439 | { | |
2440 | if (iocb->ki_flags & IOCB_NOIO) | |
2441 | return -EAGAIN; | |
2442 | page_cache_async_readahead(mapping, &file->f_ra, file, page, | |
2443 | page->index, last_index - page->index); | |
2444 | return 0; | |
2445 | } | |
2446 | ||
3a6bae48 | 2447 | static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter, |
ff993ba1 | 2448 | struct pagevec *pvec) |
06c04442 KO |
2449 | { |
2450 | struct file *filp = iocb->ki_filp; | |
2451 | struct address_space *mapping = filp->f_mapping; | |
2452 | struct file_ra_state *ra = &filp->f_ra; | |
2453 | pgoff_t index = iocb->ki_pos >> PAGE_SHIFT; | |
cbd59c48 | 2454 | pgoff_t last_index; |
2642fca6 | 2455 | struct page *page; |
cbd59c48 | 2456 | int err = 0; |
06c04442 | 2457 | |
cbd59c48 | 2458 | last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE); |
2642fca6 | 2459 | retry: |
06c04442 KO |
2460 | if (fatal_signal_pending(current)) |
2461 | return -EINTR; | |
2462 | ||
cbd59c48 | 2463 | filemap_get_read_batch(mapping, index, last_index, pvec); |
2642fca6 MWO |
2464 | if (!pagevec_count(pvec)) { |
2465 | if (iocb->ki_flags & IOCB_NOIO) | |
2466 | return -EAGAIN; | |
2467 | page_cache_sync_readahead(mapping, ra, filp, index, | |
2468 | last_index - index); | |
2469 | filemap_get_read_batch(mapping, index, last_index, pvec); | |
2470 | } | |
f253e185 MWO |
2471 | if (!pagevec_count(pvec)) { |
2472 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ)) | |
2473 | return -EAGAIN; | |
2474 | err = filemap_create_page(filp, mapping, | |
2475 | iocb->ki_pos >> PAGE_SHIFT, pvec); | |
2476 | if (err == AOP_TRUNCATED_PAGE) | |
2642fca6 | 2477 | goto retry; |
f253e185 MWO |
2478 | return err; |
2479 | } | |
06c04442 | 2480 | |
2642fca6 MWO |
2481 | page = pvec->pages[pagevec_count(pvec) - 1]; |
2482 | if (PageReadahead(page)) { | |
2483 | err = filemap_readahead(iocb, filp, mapping, page, last_index); | |
2484 | if (err) | |
2485 | goto err; | |
2486 | } | |
2487 | if (!PageUptodate(page)) { | |
2488 | if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1) | |
2489 | iocb->ki_flags |= IOCB_NOWAIT; | |
2490 | err = filemap_update_page(iocb, mapping, iter, page); | |
2491 | if (err) | |
2492 | goto err; | |
06c04442 KO |
2493 | } |
2494 | ||
2642fca6 | 2495 | return 0; |
cbd59c48 | 2496 | err: |
2642fca6 MWO |
2497 | if (err < 0) |
2498 | put_page(page); | |
2499 | if (likely(--pvec->nr)) | |
ff993ba1 | 2500 | return 0; |
4612aeef | 2501 | if (err == AOP_TRUNCATED_PAGE) |
2642fca6 MWO |
2502 | goto retry; |
2503 | return err; | |
06c04442 KO |
2504 | } |
2505 | ||
485bb99b | 2506 | /** |
87fa0f3e CH |
2507 | * filemap_read - Read data from the page cache. |
2508 | * @iocb: The iocb to read. | |
2509 | * @iter: Destination for the data. | |
2510 | * @already_read: Number of bytes already read by the caller. | |
485bb99b | 2511 | * |
87fa0f3e CH |
2512 | * Copies data from the page cache. If the data is not currently present, |
2513 | * uses the readahead and readpage address_space operations to fetch it. | |
1da177e4 | 2514 | * |
87fa0f3e CH |
2515 | * Return: Total number of bytes copied, including those already read by |
2516 | * the caller. If an error happens before any bytes are copied, returns | |
2517 | * a negative error number. | |
1da177e4 | 2518 | */ |
87fa0f3e CH |
2519 | ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter, |
2520 | ssize_t already_read) | |
1da177e4 | 2521 | { |
47c27bc4 | 2522 | struct file *filp = iocb->ki_filp; |
06c04442 | 2523 | struct file_ra_state *ra = &filp->f_ra; |
36e78914 | 2524 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2525 | struct inode *inode = mapping->host; |
ff993ba1 MWO |
2526 | struct pagevec pvec; |
2527 | int i, error = 0; | |
06c04442 KO |
2528 | bool writably_mapped; |
2529 | loff_t isize, end_offset; | |
1da177e4 | 2530 | |
723ef24b | 2531 | if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2532 | return 0; |
3644e2d2 KO |
2533 | if (unlikely(!iov_iter_count(iter))) |
2534 | return 0; | |
2535 | ||
c2a9737f | 2536 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
cbd59c48 | 2537 | pagevec_init(&pvec); |
c2a9737f | 2538 | |
06c04442 | 2539 | do { |
1da177e4 | 2540 | cond_resched(); |
5abf186a | 2541 | |
723ef24b | 2542 | /* |
06c04442 KO |
2543 | * If we've already successfully copied some data, then we |
2544 | * can no longer safely return -EIOCBQUEUED. Hence mark | |
2545 | * an async read NOWAIT at that point. | |
723ef24b | 2546 | */ |
87fa0f3e | 2547 | if ((iocb->ki_flags & IOCB_WAITQ) && already_read) |
723ef24b KO |
2548 | iocb->ki_flags |= IOCB_NOWAIT; |
2549 | ||
ff993ba1 MWO |
2550 | error = filemap_get_pages(iocb, iter, &pvec); |
2551 | if (error < 0) | |
06c04442 | 2552 | break; |
1da177e4 | 2553 | |
06c04442 KO |
2554 | /* |
2555 | * i_size must be checked after we know the pages are Uptodate. | |
2556 | * | |
2557 | * Checking i_size after the check allows us to calculate | |
2558 | * the correct value for "nr", which means the zero-filled | |
2559 | * part of the page is not copied back to userspace (unless | |
2560 | * another truncate extends the file - this is desired though). | |
2561 | */ | |
2562 | isize = i_size_read(inode); | |
2563 | if (unlikely(iocb->ki_pos >= isize)) | |
2564 | goto put_pages; | |
06c04442 KO |
2565 | end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count); |
2566 | ||
06c04442 KO |
2567 | /* |
2568 | * Once we start copying data, we don't want to be touching any | |
2569 | * cachelines that might be contended: | |
2570 | */ | |
2571 | writably_mapped = mapping_writably_mapped(mapping); | |
2572 | ||
2573 | /* | |
2574 | * When a sequential read accesses a page several times, only | |
2575 | * mark it as accessed the first time. | |
2576 | */ | |
2577 | if (iocb->ki_pos >> PAGE_SHIFT != | |
2578 | ra->prev_pos >> PAGE_SHIFT) | |
ff993ba1 | 2579 | mark_page_accessed(pvec.pages[0]); |
06c04442 | 2580 | |
ff993ba1 | 2581 | for (i = 0; i < pagevec_count(&pvec); i++) { |
cbd59c48 MWO |
2582 | struct page *page = pvec.pages[i]; |
2583 | size_t page_size = thp_size(page); | |
2584 | size_t offset = iocb->ki_pos & (page_size - 1); | |
2585 | size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos, | |
2586 | page_size - offset); | |
2587 | size_t copied; | |
06c04442 | 2588 | |
cbd59c48 MWO |
2589 | if (end_offset < page_offset(page)) |
2590 | break; | |
2591 | if (i > 0) | |
2592 | mark_page_accessed(page); | |
06c04442 KO |
2593 | /* |
2594 | * If users can be writing to this page using arbitrary | |
2595 | * virtual addresses, take care about potential aliasing | |
2596 | * before reading the page on the kernel side. | |
2597 | */ | |
cbd59c48 MWO |
2598 | if (writably_mapped) { |
2599 | int j; | |
2600 | ||
2601 | for (j = 0; j < thp_nr_pages(page); j++) | |
2602 | flush_dcache_page(page + j); | |
2603 | } | |
06c04442 | 2604 | |
cbd59c48 | 2605 | copied = copy_page_to_iter(page, offset, bytes, iter); |
06c04442 | 2606 | |
87fa0f3e | 2607 | already_read += copied; |
06c04442 KO |
2608 | iocb->ki_pos += copied; |
2609 | ra->prev_pos = iocb->ki_pos; | |
2610 | ||
2611 | if (copied < bytes) { | |
2612 | error = -EFAULT; | |
2613 | break; | |
2614 | } | |
1da177e4 | 2615 | } |
06c04442 | 2616 | put_pages: |
ff993ba1 MWO |
2617 | for (i = 0; i < pagevec_count(&pvec); i++) |
2618 | put_page(pvec.pages[i]); | |
cbd59c48 | 2619 | pagevec_reinit(&pvec); |
06c04442 | 2620 | } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error); |
1da177e4 | 2621 | |
0c6aa263 | 2622 | file_accessed(filp); |
06c04442 | 2623 | |
87fa0f3e | 2624 | return already_read ? already_read : error; |
1da177e4 | 2625 | } |
87fa0f3e | 2626 | EXPORT_SYMBOL_GPL(filemap_read); |
1da177e4 | 2627 | |
485bb99b | 2628 | /** |
6abd2322 | 2629 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2630 | * @iocb: kernel I/O control block |
6abd2322 | 2631 | * @iter: destination for the data read |
485bb99b | 2632 | * |
6abd2322 | 2633 | * This is the "read_iter()" routine for all filesystems |
1da177e4 | 2634 | * that can use the page cache directly. |
41da51bc AG |
2635 | * |
2636 | * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall | |
2637 | * be returned when no data can be read without waiting for I/O requests | |
2638 | * to complete; it doesn't prevent readahead. | |
2639 | * | |
2640 | * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O | |
2641 | * requests shall be made for the read or for readahead. When no data | |
2642 | * can be read, -EAGAIN shall be returned. When readahead would be | |
2643 | * triggered, a partial, possibly empty read shall be returned. | |
2644 | * | |
a862f68a MR |
2645 | * Return: |
2646 | * * number of bytes copied, even for partial reads | |
41da51bc | 2647 | * * negative error code (or 0 if IOCB_NOIO) if nothing was read |
1da177e4 LT |
2648 | */ |
2649 | ssize_t | |
ed978a81 | 2650 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2651 | { |
e7080a43 | 2652 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2653 | ssize_t retval = 0; |
e7080a43 NS |
2654 | |
2655 | if (!count) | |
826ea860 | 2656 | return 0; /* skip atime */ |
1da177e4 | 2657 | |
2ba48ce5 | 2658 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2659 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2660 | struct address_space *mapping = file->f_mapping; |
2661 | struct inode *inode = mapping->host; | |
543ade1f | 2662 | loff_t size; |
1da177e4 | 2663 | |
1da177e4 | 2664 | size = i_size_read(inode); |
6be96d3a | 2665 | if (iocb->ki_flags & IOCB_NOWAIT) { |
7a60d6d7 JA |
2666 | if (filemap_range_needs_writeback(mapping, iocb->ki_pos, |
2667 | iocb->ki_pos + count - 1)) | |
6be96d3a GR |
2668 | return -EAGAIN; |
2669 | } else { | |
2670 | retval = filemap_write_and_wait_range(mapping, | |
2671 | iocb->ki_pos, | |
2672 | iocb->ki_pos + count - 1); | |
2673 | if (retval < 0) | |
826ea860 | 2674 | return retval; |
6be96d3a | 2675 | } |
d8d3d94b | 2676 | |
0d5b0cf2 CH |
2677 | file_accessed(file); |
2678 | ||
5ecda137 | 2679 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2680 | if (retval >= 0) { |
c64fb5c7 | 2681 | iocb->ki_pos += retval; |
5ecda137 | 2682 | count -= retval; |
9fe55eea | 2683 | } |
ab2125df PB |
2684 | if (retval != -EIOCBQUEUED) |
2685 | iov_iter_revert(iter, count - iov_iter_count(iter)); | |
66f998f6 | 2686 | |
9fe55eea SW |
2687 | /* |
2688 | * Btrfs can have a short DIO read if we encounter | |
2689 | * compressed extents, so if there was an error, or if | |
2690 | * we've already read everything we wanted to, or if | |
2691 | * there was a short read because we hit EOF, go ahead | |
2692 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2693 | * the rest of the read. Buffered reads will not work for |
2694 | * DAX files, so don't bother trying. | |
9fe55eea | 2695 | */ |
5ecda137 | 2696 | if (retval < 0 || !count || iocb->ki_pos >= size || |
0d5b0cf2 | 2697 | IS_DAX(inode)) |
826ea860 | 2698 | return retval; |
1da177e4 LT |
2699 | } |
2700 | ||
826ea860 | 2701 | return filemap_read(iocb, iter, retval); |
1da177e4 | 2702 | } |
ed978a81 | 2703 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2704 | |
54fa39ac MWO |
2705 | static inline loff_t page_seek_hole_data(struct xa_state *xas, |
2706 | struct address_space *mapping, struct page *page, | |
2707 | loff_t start, loff_t end, bool seek_data) | |
41139aa4 | 2708 | { |
54fa39ac MWO |
2709 | const struct address_space_operations *ops = mapping->a_ops; |
2710 | size_t offset, bsz = i_blocksize(mapping->host); | |
2711 | ||
41139aa4 | 2712 | if (xa_is_value(page) || PageUptodate(page)) |
54fa39ac MWO |
2713 | return seek_data ? start : end; |
2714 | if (!ops->is_partially_uptodate) | |
2715 | return seek_data ? end : start; | |
2716 | ||
2717 | xas_pause(xas); | |
2718 | rcu_read_unlock(); | |
2719 | lock_page(page); | |
2720 | if (unlikely(page->mapping != mapping)) | |
2721 | goto unlock; | |
2722 | ||
2723 | offset = offset_in_thp(page, start) & ~(bsz - 1); | |
2724 | ||
2725 | do { | |
2726 | if (ops->is_partially_uptodate(page, offset, bsz) == seek_data) | |
2727 | break; | |
2728 | start = (start + bsz) & ~(bsz - 1); | |
2729 | offset += bsz; | |
2730 | } while (offset < thp_size(page)); | |
2731 | unlock: | |
2732 | unlock_page(page); | |
2733 | rcu_read_lock(); | |
2734 | return start; | |
41139aa4 MWO |
2735 | } |
2736 | ||
2737 | static inline | |
2738 | unsigned int seek_page_size(struct xa_state *xas, struct page *page) | |
2739 | { | |
2740 | if (xa_is_value(page)) | |
2741 | return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index); | |
2742 | return thp_size(page); | |
2743 | } | |
2744 | ||
2745 | /** | |
2746 | * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache. | |
2747 | * @mapping: Address space to search. | |
2748 | * @start: First byte to consider. | |
2749 | * @end: Limit of search (exclusive). | |
2750 | * @whence: Either SEEK_HOLE or SEEK_DATA. | |
2751 | * | |
2752 | * If the page cache knows which blocks contain holes and which blocks | |
2753 | * contain data, your filesystem can use this function to implement | |
2754 | * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are | |
2755 | * entirely memory-based such as tmpfs, and filesystems which support | |
2756 | * unwritten extents. | |
2757 | * | |
f0953a1b | 2758 | * Return: The requested offset on success, or -ENXIO if @whence specifies |
41139aa4 MWO |
2759 | * SEEK_DATA and there is no data after @start. There is an implicit hole |
2760 | * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start | |
2761 | * and @end contain data. | |
2762 | */ | |
2763 | loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start, | |
2764 | loff_t end, int whence) | |
2765 | { | |
2766 | XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT); | |
ed98b015 | 2767 | pgoff_t max = (end - 1) >> PAGE_SHIFT; |
41139aa4 MWO |
2768 | bool seek_data = (whence == SEEK_DATA); |
2769 | struct page *page; | |
2770 | ||
2771 | if (end <= start) | |
2772 | return -ENXIO; | |
2773 | ||
2774 | rcu_read_lock(); | |
2775 | while ((page = find_get_entry(&xas, max, XA_PRESENT))) { | |
ed98b015 HD |
2776 | loff_t pos = (u64)xas.xa_index << PAGE_SHIFT; |
2777 | unsigned int seek_size; | |
41139aa4 MWO |
2778 | |
2779 | if (start < pos) { | |
2780 | if (!seek_data) | |
2781 | goto unlock; | |
2782 | start = pos; | |
2783 | } | |
2784 | ||
ed98b015 HD |
2785 | seek_size = seek_page_size(&xas, page); |
2786 | pos = round_up(pos + 1, seek_size); | |
54fa39ac MWO |
2787 | start = page_seek_hole_data(&xas, mapping, page, start, pos, |
2788 | seek_data); | |
2789 | if (start < pos) | |
41139aa4 | 2790 | goto unlock; |
ed98b015 HD |
2791 | if (start >= end) |
2792 | break; | |
2793 | if (seek_size > PAGE_SIZE) | |
2794 | xas_set(&xas, pos >> PAGE_SHIFT); | |
41139aa4 MWO |
2795 | if (!xa_is_value(page)) |
2796 | put_page(page); | |
2797 | } | |
41139aa4 | 2798 | if (seek_data) |
ed98b015 | 2799 | start = -ENXIO; |
41139aa4 MWO |
2800 | unlock: |
2801 | rcu_read_unlock(); | |
ed98b015 | 2802 | if (page && !xa_is_value(page)) |
41139aa4 | 2803 | put_page(page); |
41139aa4 MWO |
2804 | if (start > end) |
2805 | return end; | |
2806 | return start; | |
2807 | } | |
2808 | ||
1da177e4 | 2809 | #ifdef CONFIG_MMU |
1da177e4 | 2810 | #define MMAP_LOTSAMISS (100) |
6b4c9f44 | 2811 | /* |
c1e8d7c6 | 2812 | * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock |
6b4c9f44 JB |
2813 | * @vmf - the vm_fault for this fault. |
2814 | * @page - the page to lock. | |
2815 | * @fpin - the pointer to the file we may pin (or is already pinned). | |
2816 | * | |
c1e8d7c6 | 2817 | * This works similar to lock_page_or_retry in that it can drop the mmap_lock. |
6b4c9f44 | 2818 | * It differs in that it actually returns the page locked if it returns 1 and 0 |
c1e8d7c6 | 2819 | * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin |
6b4c9f44 JB |
2820 | * will point to the pinned file and needs to be fput()'ed at a later point. |
2821 | */ | |
2822 | static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page, | |
2823 | struct file **fpin) | |
2824 | { | |
2825 | if (trylock_page(page)) | |
2826 | return 1; | |
2827 | ||
8b0f9fa2 LT |
2828 | /* |
2829 | * NOTE! This will make us return with VM_FAULT_RETRY, but with | |
c1e8d7c6 | 2830 | * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT |
8b0f9fa2 LT |
2831 | * is supposed to work. We have way too many special cases.. |
2832 | */ | |
6b4c9f44 JB |
2833 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
2834 | return 0; | |
2835 | ||
2836 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); | |
2837 | if (vmf->flags & FAULT_FLAG_KILLABLE) { | |
2838 | if (__lock_page_killable(page)) { | |
2839 | /* | |
c1e8d7c6 | 2840 | * We didn't have the right flags to drop the mmap_lock, |
6b4c9f44 JB |
2841 | * but all fault_handlers only check for fatal signals |
2842 | * if we return VM_FAULT_RETRY, so we need to drop the | |
c1e8d7c6 | 2843 | * mmap_lock here and return 0 if we don't have a fpin. |
6b4c9f44 JB |
2844 | */ |
2845 | if (*fpin == NULL) | |
d8ed45c5 | 2846 | mmap_read_unlock(vmf->vma->vm_mm); |
6b4c9f44 JB |
2847 | return 0; |
2848 | } | |
2849 | } else | |
2850 | __lock_page(page); | |
2851 | return 1; | |
2852 | } | |
2853 | ||
1da177e4 | 2854 | |
ef00e08e | 2855 | /* |
6b4c9f44 JB |
2856 | * Synchronous readahead happens when we don't even find a page in the page |
2857 | * cache at all. We don't want to perform IO under the mmap sem, so if we have | |
2858 | * to drop the mmap sem we return the file that was pinned in order for us to do | |
2859 | * that. If we didn't pin a file then we return NULL. The file that is | |
2860 | * returned needs to be fput()'ed when we're done with it. | |
ef00e08e | 2861 | */ |
6b4c9f44 | 2862 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
ef00e08e | 2863 | { |
2a1180f1 JB |
2864 | struct file *file = vmf->vma->vm_file; |
2865 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2866 | struct address_space *mapping = file->f_mapping; |
fcd9ae4f | 2867 | DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff); |
6b4c9f44 | 2868 | struct file *fpin = NULL; |
e630bfac | 2869 | unsigned int mmap_miss; |
ef00e08e LT |
2870 | |
2871 | /* If we don't want any read-ahead, don't bother */ | |
2a1180f1 | 2872 | if (vmf->vma->vm_flags & VM_RAND_READ) |
6b4c9f44 | 2873 | return fpin; |
275b12bf | 2874 | if (!ra->ra_pages) |
6b4c9f44 | 2875 | return fpin; |
ef00e08e | 2876 | |
2a1180f1 | 2877 | if (vmf->vma->vm_flags & VM_SEQ_READ) { |
6b4c9f44 | 2878 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
fcd9ae4f | 2879 | page_cache_sync_ra(&ractl, ra->ra_pages); |
6b4c9f44 | 2880 | return fpin; |
ef00e08e LT |
2881 | } |
2882 | ||
207d04ba | 2883 | /* Avoid banging the cache line if not needed */ |
e630bfac KS |
2884 | mmap_miss = READ_ONCE(ra->mmap_miss); |
2885 | if (mmap_miss < MMAP_LOTSAMISS * 10) | |
2886 | WRITE_ONCE(ra->mmap_miss, ++mmap_miss); | |
ef00e08e LT |
2887 | |
2888 | /* | |
2889 | * Do we miss much more than hit in this file? If so, | |
2890 | * stop bothering with read-ahead. It will only hurt. | |
2891 | */ | |
e630bfac | 2892 | if (mmap_miss > MMAP_LOTSAMISS) |
6b4c9f44 | 2893 | return fpin; |
ef00e08e | 2894 | |
d30a1100 WF |
2895 | /* |
2896 | * mmap read-around | |
2897 | */ | |
6b4c9f44 | 2898 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
db660d46 | 2899 | ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2); |
600e19af RG |
2900 | ra->size = ra->ra_pages; |
2901 | ra->async_size = ra->ra_pages / 4; | |
db660d46 DH |
2902 | ractl._index = ra->start; |
2903 | do_page_cache_ra(&ractl, ra->size, ra->async_size); | |
6b4c9f44 | 2904 | return fpin; |
ef00e08e LT |
2905 | } |
2906 | ||
2907 | /* | |
2908 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
6b4c9f44 | 2909 | * so we want to possibly extend the readahead further. We return the file that |
c1e8d7c6 | 2910 | * was pinned if we have to drop the mmap_lock in order to do IO. |
ef00e08e | 2911 | */ |
6b4c9f44 JB |
2912 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
2913 | struct page *page) | |
ef00e08e | 2914 | { |
2a1180f1 JB |
2915 | struct file *file = vmf->vma->vm_file; |
2916 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2917 | struct address_space *mapping = file->f_mapping; |
6b4c9f44 | 2918 | struct file *fpin = NULL; |
e630bfac | 2919 | unsigned int mmap_miss; |
2a1180f1 | 2920 | pgoff_t offset = vmf->pgoff; |
ef00e08e LT |
2921 | |
2922 | /* If we don't want any read-ahead, don't bother */ | |
5c72feee | 2923 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
6b4c9f44 | 2924 | return fpin; |
e630bfac KS |
2925 | mmap_miss = READ_ONCE(ra->mmap_miss); |
2926 | if (mmap_miss) | |
2927 | WRITE_ONCE(ra->mmap_miss, --mmap_miss); | |
6b4c9f44 JB |
2928 | if (PageReadahead(page)) { |
2929 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); | |
2fad6f5d WF |
2930 | page_cache_async_readahead(mapping, ra, file, |
2931 | page, offset, ra->ra_pages); | |
6b4c9f44 JB |
2932 | } |
2933 | return fpin; | |
ef00e08e LT |
2934 | } |
2935 | ||
485bb99b | 2936 | /** |
54cb8821 | 2937 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 2938 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 2939 | * |
54cb8821 | 2940 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2941 | * mapped memory region to read in file data during a page fault. |
2942 | * | |
2943 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2944 | * it in the page cache, and handles the special cases reasonably without | |
2945 | * having a lot of duplicated code. | |
9a95f3cf | 2946 | * |
c1e8d7c6 | 2947 | * vma->vm_mm->mmap_lock must be held on entry. |
9a95f3cf | 2948 | * |
c1e8d7c6 | 2949 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock |
a4985833 | 2950 | * may be dropped before doing I/O or by lock_page_maybe_drop_mmap(). |
9a95f3cf | 2951 | * |
c1e8d7c6 | 2952 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock |
9a95f3cf PC |
2953 | * has not been released. |
2954 | * | |
2955 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
a862f68a MR |
2956 | * |
2957 | * Return: bitwise-OR of %VM_FAULT_ codes. | |
1da177e4 | 2958 | */ |
2bcd6454 | 2959 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
2960 | { |
2961 | int error; | |
11bac800 | 2962 | struct file *file = vmf->vma->vm_file; |
6b4c9f44 | 2963 | struct file *fpin = NULL; |
1da177e4 | 2964 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 2965 | struct inode *inode = mapping->host; |
ef00e08e | 2966 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 2967 | pgoff_t max_off; |
1da177e4 | 2968 | struct page *page; |
2bcd6454 | 2969 | vm_fault_t ret = 0; |
1da177e4 | 2970 | |
9ab2594f MW |
2971 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2972 | if (unlikely(offset >= max_off)) | |
5307cc1a | 2973 | return VM_FAULT_SIGBUS; |
1da177e4 | 2974 | |
1da177e4 | 2975 | /* |
49426420 | 2976 | * Do we have something in the page cache already? |
1da177e4 | 2977 | */ |
ef00e08e | 2978 | page = find_get_page(mapping, offset); |
45cac65b | 2979 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2980 | /* |
ef00e08e LT |
2981 | * We found the page, so try async readahead before |
2982 | * waiting for the lock. | |
1da177e4 | 2983 | */ |
6b4c9f44 | 2984 | fpin = do_async_mmap_readahead(vmf, page); |
45cac65b | 2985 | } else if (!page) { |
ef00e08e | 2986 | /* No page in the page cache at all */ |
ef00e08e | 2987 | count_vm_event(PGMAJFAULT); |
2262185c | 2988 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e | 2989 | ret = VM_FAULT_MAJOR; |
6b4c9f44 | 2990 | fpin = do_sync_mmap_readahead(vmf); |
ef00e08e | 2991 | retry_find: |
a75d4c33 JB |
2992 | page = pagecache_get_page(mapping, offset, |
2993 | FGP_CREAT|FGP_FOR_MMAP, | |
2994 | vmf->gfp_mask); | |
6b4c9f44 JB |
2995 | if (!page) { |
2996 | if (fpin) | |
2997 | goto out_retry; | |
e520e932 | 2998 | return VM_FAULT_OOM; |
6b4c9f44 | 2999 | } |
1da177e4 LT |
3000 | } |
3001 | ||
6b4c9f44 JB |
3002 | if (!lock_page_maybe_drop_mmap(vmf, page, &fpin)) |
3003 | goto out_retry; | |
b522c94d ML |
3004 | |
3005 | /* Did it get truncated? */ | |
585e5a7b | 3006 | if (unlikely(compound_head(page)->mapping != mapping)) { |
b522c94d ML |
3007 | unlock_page(page); |
3008 | put_page(page); | |
3009 | goto retry_find; | |
3010 | } | |
520e5ba4 | 3011 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
b522c94d | 3012 | |
1da177e4 | 3013 | /* |
d00806b1 NP |
3014 | * We have a locked page in the page cache, now we need to check |
3015 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 3016 | */ |
d00806b1 | 3017 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
3018 | goto page_not_uptodate; |
3019 | ||
6b4c9f44 | 3020 | /* |
c1e8d7c6 | 3021 | * We've made it this far and we had to drop our mmap_lock, now is the |
6b4c9f44 JB |
3022 | * time to return to the upper layer and have it re-find the vma and |
3023 | * redo the fault. | |
3024 | */ | |
3025 | if (fpin) { | |
3026 | unlock_page(page); | |
3027 | goto out_retry; | |
3028 | } | |
3029 | ||
ef00e08e LT |
3030 | /* |
3031 | * Found the page and have a reference on it. | |
3032 | * We must recheck i_size under page lock. | |
3033 | */ | |
9ab2594f MW |
3034 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3035 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 3036 | unlock_page(page); |
09cbfeaf | 3037 | put_page(page); |
5307cc1a | 3038 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
3039 | } |
3040 | ||
d0217ac0 | 3041 | vmf->page = page; |
83c54070 | 3042 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 3043 | |
1da177e4 | 3044 | page_not_uptodate: |
1da177e4 LT |
3045 | /* |
3046 | * Umm, take care of errors if the page isn't up-to-date. | |
3047 | * Try to re-read it _once_. We do this synchronously, | |
3048 | * because there really aren't any performance issues here | |
3049 | * and we need to check for errors. | |
3050 | */ | |
6b4c9f44 | 3051 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
d31fa86a | 3052 | error = filemap_read_page(file, mapping, page); |
6b4c9f44 JB |
3053 | if (fpin) |
3054 | goto out_retry; | |
09cbfeaf | 3055 | put_page(page); |
d00806b1 NP |
3056 | |
3057 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 3058 | goto retry_find; |
1da177e4 | 3059 | |
d0217ac0 | 3060 | return VM_FAULT_SIGBUS; |
6b4c9f44 JB |
3061 | |
3062 | out_retry: | |
3063 | /* | |
c1e8d7c6 | 3064 | * We dropped the mmap_lock, we need to return to the fault handler to |
6b4c9f44 JB |
3065 | * re-find the vma and come back and find our hopefully still populated |
3066 | * page. | |
3067 | */ | |
3068 | if (page) | |
3069 | put_page(page); | |
3070 | if (fpin) | |
3071 | fput(fpin); | |
3072 | return ret | VM_FAULT_RETRY; | |
54cb8821 NP |
3073 | } |
3074 | EXPORT_SYMBOL(filemap_fault); | |
3075 | ||
f9ce0be7 | 3076 | static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page) |
f1820361 | 3077 | { |
f9ce0be7 KS |
3078 | struct mm_struct *mm = vmf->vma->vm_mm; |
3079 | ||
3080 | /* Huge page is mapped? No need to proceed. */ | |
3081 | if (pmd_trans_huge(*vmf->pmd)) { | |
3082 | unlock_page(page); | |
3083 | put_page(page); | |
3084 | return true; | |
3085 | } | |
3086 | ||
3087 | if (pmd_none(*vmf->pmd) && PageTransHuge(page)) { | |
3088 | vm_fault_t ret = do_set_pmd(vmf, page); | |
3089 | if (!ret) { | |
3090 | /* The page is mapped successfully, reference consumed. */ | |
3091 | unlock_page(page); | |
3092 | return true; | |
3093 | } | |
3094 | } | |
3095 | ||
3096 | if (pmd_none(*vmf->pmd)) { | |
3097 | vmf->ptl = pmd_lock(mm, vmf->pmd); | |
3098 | if (likely(pmd_none(*vmf->pmd))) { | |
3099 | mm_inc_nr_ptes(mm); | |
3100 | pmd_populate(mm, vmf->pmd, vmf->prealloc_pte); | |
3101 | vmf->prealloc_pte = NULL; | |
3102 | } | |
3103 | spin_unlock(vmf->ptl); | |
3104 | } | |
3105 | ||
3106 | /* See comment in handle_pte_fault() */ | |
3107 | if (pmd_devmap_trans_unstable(vmf->pmd)) { | |
3108 | unlock_page(page); | |
3109 | put_page(page); | |
3110 | return true; | |
3111 | } | |
3112 | ||
3113 | return false; | |
3114 | } | |
3115 | ||
3116 | static struct page *next_uptodate_page(struct page *page, | |
3117 | struct address_space *mapping, | |
3118 | struct xa_state *xas, pgoff_t end_pgoff) | |
3119 | { | |
3120 | unsigned long max_idx; | |
3121 | ||
3122 | do { | |
3123 | if (!page) | |
3124 | return NULL; | |
3125 | if (xas_retry(xas, page)) | |
3126 | continue; | |
3127 | if (xa_is_value(page)) | |
3128 | continue; | |
3129 | if (PageLocked(page)) | |
3130 | continue; | |
3131 | if (!page_cache_get_speculative(page)) | |
3132 | continue; | |
3133 | /* Has the page moved or been split? */ | |
3134 | if (unlikely(page != xas_reload(xas))) | |
3135 | goto skip; | |
3136 | if (!PageUptodate(page) || PageReadahead(page)) | |
3137 | goto skip; | |
3138 | if (PageHWPoison(page)) | |
3139 | goto skip; | |
3140 | if (!trylock_page(page)) | |
3141 | goto skip; | |
3142 | if (page->mapping != mapping) | |
3143 | goto unlock; | |
3144 | if (!PageUptodate(page)) | |
3145 | goto unlock; | |
3146 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); | |
3147 | if (xas->xa_index >= max_idx) | |
3148 | goto unlock; | |
3149 | return page; | |
3150 | unlock: | |
3151 | unlock_page(page); | |
3152 | skip: | |
3153 | put_page(page); | |
3154 | } while ((page = xas_next_entry(xas, end_pgoff)) != NULL); | |
3155 | ||
3156 | return NULL; | |
3157 | } | |
3158 | ||
3159 | static inline struct page *first_map_page(struct address_space *mapping, | |
3160 | struct xa_state *xas, | |
3161 | pgoff_t end_pgoff) | |
3162 | { | |
3163 | return next_uptodate_page(xas_find(xas, end_pgoff), | |
3164 | mapping, xas, end_pgoff); | |
3165 | } | |
3166 | ||
3167 | static inline struct page *next_map_page(struct address_space *mapping, | |
3168 | struct xa_state *xas, | |
3169 | pgoff_t end_pgoff) | |
3170 | { | |
3171 | return next_uptodate_page(xas_next_entry(xas, end_pgoff), | |
3172 | mapping, xas, end_pgoff); | |
3173 | } | |
3174 | ||
3175 | vm_fault_t filemap_map_pages(struct vm_fault *vmf, | |
3176 | pgoff_t start_pgoff, pgoff_t end_pgoff) | |
3177 | { | |
3178 | struct vm_area_struct *vma = vmf->vma; | |
3179 | struct file *file = vma->vm_file; | |
f1820361 | 3180 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 3181 | pgoff_t last_pgoff = start_pgoff; |
9d3af4b4 | 3182 | unsigned long addr; |
070e807c | 3183 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
27a83a60 | 3184 | struct page *head, *page; |
e630bfac | 3185 | unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss); |
f9ce0be7 | 3186 | vm_fault_t ret = 0; |
f1820361 KS |
3187 | |
3188 | rcu_read_lock(); | |
f9ce0be7 KS |
3189 | head = first_map_page(mapping, &xas, end_pgoff); |
3190 | if (!head) | |
3191 | goto out; | |
f1820361 | 3192 | |
f9ce0be7 KS |
3193 | if (filemap_map_pmd(vmf, head)) { |
3194 | ret = VM_FAULT_NOPAGE; | |
3195 | goto out; | |
3196 | } | |
f1820361 | 3197 | |
9d3af4b4 WD |
3198 | addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
3199 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); | |
f9ce0be7 | 3200 | do { |
27a83a60 | 3201 | page = find_subpage(head, xas.xa_index); |
f9ce0be7 | 3202 | if (PageHWPoison(page)) |
f1820361 KS |
3203 | goto unlock; |
3204 | ||
e630bfac KS |
3205 | if (mmap_miss > 0) |
3206 | mmap_miss--; | |
7267ec00 | 3207 | |
9d3af4b4 | 3208 | addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
f9ce0be7 | 3209 | vmf->pte += xas.xa_index - last_pgoff; |
070e807c | 3210 | last_pgoff = xas.xa_index; |
f9ce0be7 KS |
3211 | |
3212 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3213 | goto unlock; |
f9ce0be7 | 3214 | |
46bdb427 | 3215 | /* We're about to handle the fault */ |
9d3af4b4 | 3216 | if (vmf->address == addr) |
46bdb427 | 3217 | ret = VM_FAULT_NOPAGE; |
46bdb427 | 3218 | |
9d3af4b4 | 3219 | do_set_pte(vmf, page, addr); |
f9ce0be7 | 3220 | /* no need to invalidate: a not-present page won't be cached */ |
9d3af4b4 | 3221 | update_mmu_cache(vma, addr, vmf->pte); |
27a83a60 | 3222 | unlock_page(head); |
f9ce0be7 | 3223 | continue; |
f1820361 | 3224 | unlock: |
27a83a60 | 3225 | unlock_page(head); |
27a83a60 | 3226 | put_page(head); |
f9ce0be7 KS |
3227 | } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL); |
3228 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3229 | out: | |
f1820361 | 3230 | rcu_read_unlock(); |
e630bfac | 3231 | WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss); |
f9ce0be7 | 3232 | return ret; |
f1820361 KS |
3233 | } |
3234 | EXPORT_SYMBOL(filemap_map_pages); | |
3235 | ||
2bcd6454 | 3236 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 | 3237 | { |
5df1a672 | 3238 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
4fcf1c62 | 3239 | struct page *page = vmf->page; |
2bcd6454 | 3240 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 3241 | |
5df1a672 | 3242 | sb_start_pagefault(mapping->host->i_sb); |
11bac800 | 3243 | file_update_time(vmf->vma->vm_file); |
4fcf1c62 | 3244 | lock_page(page); |
5df1a672 | 3245 | if (page->mapping != mapping) { |
4fcf1c62 JK |
3246 | unlock_page(page); |
3247 | ret = VM_FAULT_NOPAGE; | |
3248 | goto out; | |
3249 | } | |
14da9200 JK |
3250 | /* |
3251 | * We mark the page dirty already here so that when freeze is in | |
3252 | * progress, we are guaranteed that writeback during freezing will | |
3253 | * see the dirty page and writeprotect it again. | |
3254 | */ | |
3255 | set_page_dirty(page); | |
1d1d1a76 | 3256 | wait_for_stable_page(page); |
4fcf1c62 | 3257 | out: |
5df1a672 | 3258 | sb_end_pagefault(mapping->host->i_sb); |
4fcf1c62 JK |
3259 | return ret; |
3260 | } | |
4fcf1c62 | 3261 | |
f0f37e2f | 3262 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 3263 | .fault = filemap_fault, |
f1820361 | 3264 | .map_pages = filemap_map_pages, |
4fcf1c62 | 3265 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
3266 | }; |
3267 | ||
3268 | /* This is used for a general mmap of a disk file */ | |
3269 | ||
68d68ff6 | 3270 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3271 | { |
3272 | struct address_space *mapping = file->f_mapping; | |
3273 | ||
3274 | if (!mapping->a_ops->readpage) | |
3275 | return -ENOEXEC; | |
3276 | file_accessed(file); | |
3277 | vma->vm_ops = &generic_file_vm_ops; | |
3278 | return 0; | |
3279 | } | |
1da177e4 LT |
3280 | |
3281 | /* | |
3282 | * This is for filesystems which do not implement ->writepage. | |
3283 | */ | |
3284 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
3285 | { | |
3286 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
3287 | return -EINVAL; | |
3288 | return generic_file_mmap(file, vma); | |
3289 | } | |
3290 | #else | |
4b96a37d | 3291 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
45397228 | 3292 | { |
4b96a37d | 3293 | return VM_FAULT_SIGBUS; |
45397228 | 3294 | } |
68d68ff6 | 3295 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3296 | { |
3297 | return -ENOSYS; | |
3298 | } | |
68d68ff6 | 3299 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3300 | { |
3301 | return -ENOSYS; | |
3302 | } | |
3303 | #endif /* CONFIG_MMU */ | |
3304 | ||
45397228 | 3305 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
3306 | EXPORT_SYMBOL(generic_file_mmap); |
3307 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
3308 | ||
67f9fd91 SL |
3309 | static struct page *wait_on_page_read(struct page *page) |
3310 | { | |
3311 | if (!IS_ERR(page)) { | |
3312 | wait_on_page_locked(page); | |
3313 | if (!PageUptodate(page)) { | |
09cbfeaf | 3314 | put_page(page); |
67f9fd91 SL |
3315 | page = ERR_PTR(-EIO); |
3316 | } | |
3317 | } | |
3318 | return page; | |
3319 | } | |
3320 | ||
32b63529 | 3321 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 3322 | pgoff_t index, |
5e5358e7 | 3323 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3324 | void *data, |
3325 | gfp_t gfp) | |
1da177e4 | 3326 | { |
eb2be189 | 3327 | struct page *page; |
1da177e4 LT |
3328 | int err; |
3329 | repeat: | |
3330 | page = find_get_page(mapping, index); | |
3331 | if (!page) { | |
453f85d4 | 3332 | page = __page_cache_alloc(gfp); |
eb2be189 NP |
3333 | if (!page) |
3334 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 3335 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 3336 | if (unlikely(err)) { |
09cbfeaf | 3337 | put_page(page); |
eb2be189 NP |
3338 | if (err == -EEXIST) |
3339 | goto repeat; | |
22ecdb4f | 3340 | /* Presumably ENOMEM for xarray node */ |
1da177e4 LT |
3341 | return ERR_PTR(err); |
3342 | } | |
32b63529 MG |
3343 | |
3344 | filler: | |
6c45b454 CH |
3345 | if (filler) |
3346 | err = filler(data, page); | |
3347 | else | |
3348 | err = mapping->a_ops->readpage(data, page); | |
3349 | ||
1da177e4 | 3350 | if (err < 0) { |
09cbfeaf | 3351 | put_page(page); |
32b63529 | 3352 | return ERR_PTR(err); |
1da177e4 | 3353 | } |
1da177e4 | 3354 | |
32b63529 MG |
3355 | page = wait_on_page_read(page); |
3356 | if (IS_ERR(page)) | |
3357 | return page; | |
3358 | goto out; | |
3359 | } | |
1da177e4 LT |
3360 | if (PageUptodate(page)) |
3361 | goto out; | |
3362 | ||
ebded027 | 3363 | /* |
0e9aa675 | 3364 | * Page is not up to date and may be locked due to one of the following |
ebded027 MG |
3365 | * case a: Page is being filled and the page lock is held |
3366 | * case b: Read/write error clearing the page uptodate status | |
3367 | * case c: Truncation in progress (page locked) | |
3368 | * case d: Reclaim in progress | |
3369 | * | |
3370 | * Case a, the page will be up to date when the page is unlocked. | |
3371 | * There is no need to serialise on the page lock here as the page | |
3372 | * is pinned so the lock gives no additional protection. Even if the | |
ce89fddf | 3373 | * page is truncated, the data is still valid if PageUptodate as |
ebded027 MG |
3374 | * it's a race vs truncate race. |
3375 | * Case b, the page will not be up to date | |
3376 | * Case c, the page may be truncated but in itself, the data may still | |
3377 | * be valid after IO completes as it's a read vs truncate race. The | |
3378 | * operation must restart if the page is not uptodate on unlock but | |
3379 | * otherwise serialising on page lock to stabilise the mapping gives | |
3380 | * no additional guarantees to the caller as the page lock is | |
3381 | * released before return. | |
3382 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
3383 | * will be a race with remove_mapping that determines if the mapping | |
3384 | * is valid on unlock but otherwise the data is valid and there is | |
3385 | * no need to serialise with page lock. | |
3386 | * | |
3387 | * As the page lock gives no additional guarantee, we optimistically | |
3388 | * wait on the page to be unlocked and check if it's up to date and | |
3389 | * use the page if it is. Otherwise, the page lock is required to | |
3390 | * distinguish between the different cases. The motivation is that we | |
3391 | * avoid spurious serialisations and wakeups when multiple processes | |
3392 | * wait on the same page for IO to complete. | |
3393 | */ | |
3394 | wait_on_page_locked(page); | |
3395 | if (PageUptodate(page)) | |
3396 | goto out; | |
3397 | ||
3398 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 3399 | lock_page(page); |
ebded027 MG |
3400 | |
3401 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
3402 | if (!page->mapping) { |
3403 | unlock_page(page); | |
09cbfeaf | 3404 | put_page(page); |
32b63529 | 3405 | goto repeat; |
1da177e4 | 3406 | } |
ebded027 MG |
3407 | |
3408 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
3409 | if (PageUptodate(page)) { |
3410 | unlock_page(page); | |
3411 | goto out; | |
3412 | } | |
faffdfa0 XT |
3413 | |
3414 | /* | |
3415 | * A previous I/O error may have been due to temporary | |
3416 | * failures. | |
3417 | * Clear page error before actual read, PG_error will be | |
3418 | * set again if read page fails. | |
3419 | */ | |
3420 | ClearPageError(page); | |
32b63529 MG |
3421 | goto filler; |
3422 | ||
c855ff37 | 3423 | out: |
6fe6900e NP |
3424 | mark_page_accessed(page); |
3425 | return page; | |
3426 | } | |
0531b2aa LT |
3427 | |
3428 | /** | |
67f9fd91 | 3429 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
3430 | * @mapping: the page's address_space |
3431 | * @index: the page index | |
3432 | * @filler: function to perform the read | |
5e5358e7 | 3433 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 3434 | * |
0531b2aa | 3435 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 3436 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
3437 | * |
3438 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a MR |
3439 | * |
3440 | * Return: up to date page on success, ERR_PTR() on failure. | |
0531b2aa | 3441 | */ |
67f9fd91 | 3442 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 3443 | pgoff_t index, |
5e5358e7 | 3444 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3445 | void *data) |
3446 | { | |
d322a8e5 CH |
3447 | return do_read_cache_page(mapping, index, filler, data, |
3448 | mapping_gfp_mask(mapping)); | |
0531b2aa | 3449 | } |
67f9fd91 | 3450 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
3451 | |
3452 | /** | |
3453 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
3454 | * @mapping: the page's address_space | |
3455 | * @index: the page index | |
3456 | * @gfp: the page allocator flags to use if allocating | |
3457 | * | |
3458 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 3459 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
3460 | * |
3461 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a MR |
3462 | * |
3463 | * Return: up to date page on success, ERR_PTR() on failure. | |
0531b2aa LT |
3464 | */ |
3465 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
3466 | pgoff_t index, | |
3467 | gfp_t gfp) | |
3468 | { | |
6c45b454 | 3469 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
0531b2aa LT |
3470 | } |
3471 | EXPORT_SYMBOL(read_cache_page_gfp); | |
3472 | ||
afddba49 NP |
3473 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
3474 | loff_t pos, unsigned len, unsigned flags, | |
3475 | struct page **pagep, void **fsdata) | |
3476 | { | |
3477 | const struct address_space_operations *aops = mapping->a_ops; | |
3478 | ||
4e02ed4b | 3479 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 3480 | pagep, fsdata); |
afddba49 NP |
3481 | } |
3482 | EXPORT_SYMBOL(pagecache_write_begin); | |
3483 | ||
3484 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
3485 | loff_t pos, unsigned len, unsigned copied, | |
3486 | struct page *page, void *fsdata) | |
3487 | { | |
3488 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 3489 | |
4e02ed4b | 3490 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
3491 | } |
3492 | EXPORT_SYMBOL(pagecache_write_end); | |
3493 | ||
a92853b6 KK |
3494 | /* |
3495 | * Warn about a page cache invalidation failure during a direct I/O write. | |
3496 | */ | |
3497 | void dio_warn_stale_pagecache(struct file *filp) | |
3498 | { | |
3499 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); | |
3500 | char pathname[128]; | |
a92853b6 KK |
3501 | char *path; |
3502 | ||
5df1a672 | 3503 | errseq_set(&filp->f_mapping->wb_err, -EIO); |
a92853b6 KK |
3504 | if (__ratelimit(&_rs)) { |
3505 | path = file_path(filp, pathname, sizeof(pathname)); | |
3506 | if (IS_ERR(path)) | |
3507 | path = "(unknown)"; | |
3508 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); | |
3509 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, | |
3510 | current->comm); | |
3511 | } | |
3512 | } | |
3513 | ||
1da177e4 | 3514 | ssize_t |
1af5bb49 | 3515 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3516 | { |
3517 | struct file *file = iocb->ki_filp; | |
3518 | struct address_space *mapping = file->f_mapping; | |
3519 | struct inode *inode = mapping->host; | |
1af5bb49 | 3520 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3521 | ssize_t written; |
a969e903 CH |
3522 | size_t write_len; |
3523 | pgoff_t end; | |
1da177e4 | 3524 | |
0c949334 | 3525 | write_len = iov_iter_count(from); |
09cbfeaf | 3526 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3527 | |
6be96d3a GR |
3528 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3529 | /* If there are pages to writeback, return */ | |
5df1a672 | 3530 | if (filemap_range_has_page(file->f_mapping, pos, |
35f12f0f | 3531 | pos + write_len - 1)) |
6be96d3a GR |
3532 | return -EAGAIN; |
3533 | } else { | |
3534 | written = filemap_write_and_wait_range(mapping, pos, | |
3535 | pos + write_len - 1); | |
3536 | if (written) | |
3537 | goto out; | |
3538 | } | |
a969e903 CH |
3539 | |
3540 | /* | |
3541 | * After a write we want buffered reads to be sure to go to disk to get | |
3542 | * the new data. We invalidate clean cached page from the region we're | |
3543 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3544 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3545 | */ |
55635ba7 | 3546 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3547 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3548 | /* |
3549 | * If a page can not be invalidated, return 0 to fall back | |
3550 | * to buffered write. | |
3551 | */ | |
3552 | if (written) { | |
3553 | if (written == -EBUSY) | |
3554 | return 0; | |
3555 | goto out; | |
a969e903 CH |
3556 | } |
3557 | ||
639a93a5 | 3558 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3559 | |
3560 | /* | |
3561 | * Finally, try again to invalidate clean pages which might have been | |
3562 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3563 | * if the source of the write was an mmap'ed region of the file | |
3564 | * we're writing. Either one is a pretty crazy thing to do, | |
3565 | * so we don't support it 100%. If this invalidation | |
3566 | * fails, tough, the write still worked... | |
332391a9 LC |
3567 | * |
3568 | * Most of the time we do not need this since dio_complete() will do | |
3569 | * the invalidation for us. However there are some file systems that | |
3570 | * do not end up with dio_complete() being called, so let's not break | |
80c1fe90 KK |
3571 | * them by removing it completely. |
3572 | * | |
9266a140 KK |
3573 | * Noticeable example is a blkdev_direct_IO(). |
3574 | * | |
80c1fe90 | 3575 | * Skip invalidation for async writes or if mapping has no pages. |
a969e903 | 3576 | */ |
9266a140 KK |
3577 | if (written > 0 && mapping->nrpages && |
3578 | invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) | |
3579 | dio_warn_stale_pagecache(file); | |
a969e903 | 3580 | |
1da177e4 | 3581 | if (written > 0) { |
0116651c | 3582 | pos += written; |
639a93a5 | 3583 | write_len -= written; |
0116651c NK |
3584 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3585 | i_size_write(inode, pos); | |
1da177e4 LT |
3586 | mark_inode_dirty(inode); |
3587 | } | |
5cb6c6c7 | 3588 | iocb->ki_pos = pos; |
1da177e4 | 3589 | } |
ab2125df PB |
3590 | if (written != -EIOCBQUEUED) |
3591 | iov_iter_revert(from, write_len - iov_iter_count(from)); | |
a969e903 | 3592 | out: |
1da177e4 LT |
3593 | return written; |
3594 | } | |
3595 | EXPORT_SYMBOL(generic_file_direct_write); | |
3596 | ||
eb2be189 NP |
3597 | /* |
3598 | * Find or create a page at the given pagecache position. Return the locked | |
3599 | * page. This function is specifically for buffered writes. | |
3600 | */ | |
54566b2c NP |
3601 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
3602 | pgoff_t index, unsigned flags) | |
eb2be189 | 3603 | { |
eb2be189 | 3604 | struct page *page; |
bbddabe2 | 3605 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 3606 | |
54566b2c | 3607 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
3608 | fgp_flags |= FGP_NOFS; |
3609 | ||
3610 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 3611 | mapping_gfp_mask(mapping)); |
c585a267 | 3612 | if (page) |
2457aec6 | 3613 | wait_for_stable_page(page); |
eb2be189 | 3614 | |
eb2be189 NP |
3615 | return page; |
3616 | } | |
54566b2c | 3617 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 3618 | |
3b93f911 | 3619 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3620 | struct iov_iter *i, loff_t pos) |
3621 | { | |
3622 | struct address_space *mapping = file->f_mapping; | |
3623 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3624 | long status = 0; | |
3625 | ssize_t written = 0; | |
674b892e NP |
3626 | unsigned int flags = 0; |
3627 | ||
afddba49 NP |
3628 | do { |
3629 | struct page *page; | |
afddba49 NP |
3630 | unsigned long offset; /* Offset into pagecache page */ |
3631 | unsigned long bytes; /* Bytes to write to page */ | |
3632 | size_t copied; /* Bytes copied from user */ | |
3633 | void *fsdata; | |
3634 | ||
09cbfeaf KS |
3635 | offset = (pos & (PAGE_SIZE - 1)); |
3636 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3637 | iov_iter_count(i)); |
3638 | ||
3639 | again: | |
00a3d660 LT |
3640 | /* |
3641 | * Bring in the user page that we will copy from _first_. | |
3642 | * Otherwise there's a nasty deadlock on copying from the | |
3643 | * same page as we're writing to, without it being marked | |
3644 | * up-to-date. | |
00a3d660 LT |
3645 | */ |
3646 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
3647 | status = -EFAULT; | |
3648 | break; | |
3649 | } | |
3650 | ||
296291cd JK |
3651 | if (fatal_signal_pending(current)) { |
3652 | status = -EINTR; | |
3653 | break; | |
3654 | } | |
3655 | ||
674b892e | 3656 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3657 | &page, &fsdata); |
2457aec6 | 3658 | if (unlikely(status < 0)) |
afddba49 NP |
3659 | break; |
3660 | ||
931e80e4 | 3661 | if (mapping_writably_mapped(mapping)) |
3662 | flush_dcache_page(page); | |
00a3d660 | 3663 | |
f0b65f39 | 3664 | copied = copy_page_from_iter_atomic(page, offset, bytes, i); |
afddba49 NP |
3665 | flush_dcache_page(page); |
3666 | ||
3667 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3668 | page, fsdata); | |
f0b65f39 AV |
3669 | if (unlikely(status != copied)) { |
3670 | iov_iter_revert(i, copied - max(status, 0L)); | |
3671 | if (unlikely(status < 0)) | |
3672 | break; | |
3673 | } | |
afddba49 NP |
3674 | cond_resched(); |
3675 | ||
bc1bb416 | 3676 | if (unlikely(status == 0)) { |
afddba49 | 3677 | /* |
bc1bb416 AV |
3678 | * A short copy made ->write_end() reject the |
3679 | * thing entirely. Might be memory poisoning | |
3680 | * halfway through, might be a race with munmap, | |
3681 | * might be severe memory pressure. | |
afddba49 | 3682 | */ |
bc1bb416 AV |
3683 | if (copied) |
3684 | bytes = copied; | |
afddba49 NP |
3685 | goto again; |
3686 | } | |
f0b65f39 AV |
3687 | pos += status; |
3688 | written += status; | |
afddba49 NP |
3689 | |
3690 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3691 | } while (iov_iter_count(i)); |
3692 | ||
3693 | return written ? written : status; | |
3694 | } | |
3b93f911 | 3695 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3696 | |
e4dd9de3 | 3697 | /** |
8174202b | 3698 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3699 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3700 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3701 | * |
3702 | * This function does all the work needed for actually writing data to a | |
3703 | * file. It does all basic checks, removes SUID from the file, updates | |
3704 | * modification times and calls proper subroutines depending on whether we | |
3705 | * do direct IO or a standard buffered write. | |
3706 | * | |
3707 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
3708 | * object which does not need locking at all. | |
3709 | * | |
3710 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3711 | * A caller has to handle it. This is mainly due to the fact that we want to | |
3712 | * avoid syncing under i_mutex. | |
a862f68a MR |
3713 | * |
3714 | * Return: | |
3715 | * * number of bytes written, even for truncated writes | |
3716 | * * negative error code if no data has been written at all | |
e4dd9de3 | 3717 | */ |
8174202b | 3718 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3719 | { |
3720 | struct file *file = iocb->ki_filp; | |
68d68ff6 | 3721 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3722 | struct inode *inode = mapping->host; |
3b93f911 | 3723 | ssize_t written = 0; |
1da177e4 | 3724 | ssize_t err; |
3b93f911 | 3725 | ssize_t status; |
1da177e4 | 3726 | |
1da177e4 | 3727 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3728 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3729 | err = file_remove_privs(file); |
1da177e4 LT |
3730 | if (err) |
3731 | goto out; | |
3732 | ||
c3b2da31 JB |
3733 | err = file_update_time(file); |
3734 | if (err) | |
3735 | goto out; | |
1da177e4 | 3736 | |
2ba48ce5 | 3737 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3738 | loff_t pos, endbyte; |
fb5527e6 | 3739 | |
1af5bb49 | 3740 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3741 | /* |
fbbbad4b MW |
3742 | * If the write stopped short of completing, fall back to |
3743 | * buffered writes. Some filesystems do this for writes to | |
3744 | * holes, for example. For DAX files, a buffered write will | |
3745 | * not succeed (even if it did, DAX does not handle dirty | |
3746 | * page-cache pages correctly). | |
1da177e4 | 3747 | */ |
0b8def9d | 3748 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3749 | goto out; |
3750 | ||
0b8def9d | 3751 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3752 | /* |
3b93f911 | 3753 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3754 | * then we want to return the number of bytes which were |
3755 | * direct-written, or the error code if that was zero. Note | |
3756 | * that this differs from normal direct-io semantics, which | |
3757 | * will return -EFOO even if some bytes were written. | |
3758 | */ | |
60bb4529 | 3759 | if (unlikely(status < 0)) { |
3b93f911 | 3760 | err = status; |
fb5527e6 JM |
3761 | goto out; |
3762 | } | |
fb5527e6 JM |
3763 | /* |
3764 | * We need to ensure that the page cache pages are written to | |
3765 | * disk and invalidated to preserve the expected O_DIRECT | |
3766 | * semantics. | |
3767 | */ | |
3b93f911 | 3768 | endbyte = pos + status - 1; |
0b8def9d | 3769 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3770 | if (err == 0) { |
0b8def9d | 3771 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3772 | written += status; |
fb5527e6 | 3773 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3774 | pos >> PAGE_SHIFT, |
3775 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3776 | } else { |
3777 | /* | |
3778 | * We don't know how much we wrote, so just return | |
3779 | * the number of bytes which were direct-written | |
3780 | */ | |
3781 | } | |
3782 | } else { | |
0b8def9d AV |
3783 | written = generic_perform_write(file, from, iocb->ki_pos); |
3784 | if (likely(written > 0)) | |
3785 | iocb->ki_pos += written; | |
fb5527e6 | 3786 | } |
1da177e4 LT |
3787 | out: |
3788 | current->backing_dev_info = NULL; | |
3789 | return written ? written : err; | |
3790 | } | |
8174202b | 3791 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3792 | |
e4dd9de3 | 3793 | /** |
8174202b | 3794 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3795 | * @iocb: IO state structure |
8174202b | 3796 | * @from: iov_iter with data to write |
e4dd9de3 | 3797 | * |
8174202b | 3798 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
3799 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
3800 | * and acquires i_mutex as needed. | |
a862f68a MR |
3801 | * Return: |
3802 | * * negative error code if no data has been written at all of | |
3803 | * vfs_fsync_range() failed for a synchronous write | |
3804 | * * number of bytes written, even for truncated writes | |
e4dd9de3 | 3805 | */ |
8174202b | 3806 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3807 | { |
3808 | struct file *file = iocb->ki_filp; | |
148f948b | 3809 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3810 | ssize_t ret; |
1da177e4 | 3811 | |
5955102c | 3812 | inode_lock(inode); |
3309dd04 AV |
3813 | ret = generic_write_checks(iocb, from); |
3814 | if (ret > 0) | |
5f380c7f | 3815 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3816 | inode_unlock(inode); |
1da177e4 | 3817 | |
e2592217 CH |
3818 | if (ret > 0) |
3819 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3820 | return ret; |
3821 | } | |
8174202b | 3822 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3823 | |
cf9a2ae8 DH |
3824 | /** |
3825 | * try_to_release_page() - release old fs-specific metadata on a page | |
3826 | * | |
3827 | * @page: the page which the kernel is trying to free | |
3828 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3829 | * | |
3830 | * The address_space is to try to release any data against the page | |
a862f68a | 3831 | * (presumably at page->private). |
cf9a2ae8 | 3832 | * |
266cf658 DH |
3833 | * This may also be called if PG_fscache is set on a page, indicating that the |
3834 | * page is known to the local caching routines. | |
3835 | * | |
cf9a2ae8 | 3836 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3837 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3838 | * |
a862f68a | 3839 | * Return: %1 if the release was successful, otherwise return zero. |
cf9a2ae8 DH |
3840 | */ |
3841 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3842 | { | |
3843 | struct address_space * const mapping = page->mapping; | |
3844 | ||
3845 | BUG_ON(!PageLocked(page)); | |
3846 | if (PageWriteback(page)) | |
3847 | return 0; | |
3848 | ||
3849 | if (mapping && mapping->a_ops->releasepage) | |
3850 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3851 | return try_to_free_buffers(page); | |
3852 | } | |
3853 | ||
3854 | EXPORT_SYMBOL(try_to_release_page); |