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 | * |
9608703e | 79 | * ->i_rwsem |
730633f0 JK |
80 | * ->invalidate_lock (acquired by fs in truncate path) |
81 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) | |
1da177e4 | 82 | * |
c1e8d7c6 | 83 | * ->mmap_lock |
c8c06efa | 84 | * ->i_mmap_rwsem |
b8072f09 | 85 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
b93b0163 | 86 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
1da177e4 | 87 | * |
c1e8d7c6 | 88 | * ->mmap_lock |
730633f0 JK |
89 | * ->invalidate_lock (filemap_fault) |
90 | * ->lock_page (filemap_fault, access_process_vm) | |
1da177e4 | 91 | * |
9608703e | 92 | * ->i_rwsem (generic_perform_write) |
c1e8d7c6 | 93 | * ->mmap_lock (fault_in_pages_readable->do_page_fault) |
1da177e4 | 94 | * |
f758eeab | 95 | * bdi->wb.list_lock |
a66979ab | 96 | * sb_lock (fs/fs-writeback.c) |
b93b0163 | 97 | * ->i_pages lock (__sync_single_inode) |
1da177e4 | 98 | * |
c8c06efa | 99 | * ->i_mmap_rwsem |
1da177e4 LT |
100 | * ->anon_vma.lock (vma_adjust) |
101 | * | |
102 | * ->anon_vma.lock | |
b8072f09 | 103 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 104 | * |
b8072f09 | 105 | * ->page_table_lock or pte_lock |
5d337b91 | 106 | * ->swap_lock (try_to_unmap_one) |
1da177e4 | 107 | * ->private_lock (try_to_unmap_one) |
b93b0163 | 108 | * ->i_pages lock (try_to_unmap_one) |
15b44736 HD |
109 | * ->lruvec->lru_lock (follow_page->mark_page_accessed) |
110 | * ->lruvec->lru_lock (check_pte_range->isolate_lru_page) | |
1da177e4 | 111 | * ->private_lock (page_remove_rmap->set_page_dirty) |
b93b0163 | 112 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
f758eeab | 113 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 114 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 115 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 116 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 117 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
118 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
119 | * | |
c8c06efa | 120 | * ->i_mmap_rwsem |
9a3c531d | 121 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
122 | */ |
123 | ||
5c024e6a | 124 | static void page_cache_delete(struct address_space *mapping, |
91b0abe3 JW |
125 | struct page *page, void *shadow) |
126 | { | |
5c024e6a MW |
127 | XA_STATE(xas, &mapping->i_pages, page->index); |
128 | unsigned int nr = 1; | |
c70b647d | 129 | |
5c024e6a | 130 | mapping_set_update(&xas, mapping); |
c70b647d | 131 | |
5c024e6a MW |
132 | /* hugetlb pages are represented by a single entry in the xarray */ |
133 | if (!PageHuge(page)) { | |
134 | xas_set_order(&xas, page->index, compound_order(page)); | |
d8c6546b | 135 | nr = compound_nr(page); |
5c024e6a | 136 | } |
91b0abe3 | 137 | |
83929372 KS |
138 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
139 | VM_BUG_ON_PAGE(PageTail(page), page); | |
140 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 141 | |
5c024e6a MW |
142 | xas_store(&xas, shadow); |
143 | xas_init_marks(&xas); | |
d3798ae8 | 144 | |
2300638b JK |
145 | page->mapping = NULL; |
146 | /* Leave page->index set: truncation lookup relies upon it */ | |
d3798ae8 | 147 | mapping->nrpages -= nr; |
91b0abe3 JW |
148 | } |
149 | ||
5ecc4d85 JK |
150 | static void unaccount_page_cache_page(struct address_space *mapping, |
151 | struct page *page) | |
1da177e4 | 152 | { |
5ecc4d85 | 153 | int nr; |
1da177e4 | 154 | |
c515e1fd DM |
155 | /* |
156 | * if we're uptodate, flush out into the cleancache, otherwise | |
157 | * invalidate any existing cleancache entries. We can't leave | |
158 | * stale data around in the cleancache once our page is gone | |
159 | */ | |
160 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
161 | cleancache_put_page(page); | |
162 | else | |
3167760f | 163 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 164 | |
83929372 | 165 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
166 | VM_BUG_ON_PAGE(page_mapped(page), page); |
167 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
168 | int mapcount; | |
169 | ||
170 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
171 | current->comm, page_to_pfn(page)); | |
172 | dump_page(page, "still mapped when deleted"); | |
173 | dump_stack(); | |
174 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
175 | ||
176 | mapcount = page_mapcount(page); | |
177 | if (mapping_exiting(mapping) && | |
178 | page_count(page) >= mapcount + 2) { | |
179 | /* | |
180 | * All vmas have already been torn down, so it's | |
181 | * a good bet that actually the page is unmapped, | |
182 | * and we'd prefer not to leak it: if we're wrong, | |
183 | * some other bad page check should catch it later. | |
184 | */ | |
185 | page_mapcount_reset(page); | |
6d061f9f | 186 | page_ref_sub(page, mapcount); |
06b241f3 HD |
187 | } |
188 | } | |
189 | ||
4165b9b4 | 190 | /* hugetlb pages do not participate in page cache accounting. */ |
5ecc4d85 JK |
191 | if (PageHuge(page)) |
192 | return; | |
09612fa6 | 193 | |
6c357848 | 194 | nr = thp_nr_pages(page); |
5ecc4d85 | 195 | |
0d1c2072 | 196 | __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr); |
5ecc4d85 | 197 | if (PageSwapBacked(page)) { |
0d1c2072 | 198 | __mod_lruvec_page_state(page, NR_SHMEM, -nr); |
5ecc4d85 | 199 | if (PageTransHuge(page)) |
57b2847d | 200 | __mod_lruvec_page_state(page, NR_SHMEM_THPS, -nr); |
99cb0dbd | 201 | } else if (PageTransHuge(page)) { |
bf9ecead | 202 | __mod_lruvec_page_state(page, NR_FILE_THPS, -nr); |
09d91cda | 203 | filemap_nr_thps_dec(mapping); |
800d8c63 | 204 | } |
5ecc4d85 JK |
205 | |
206 | /* | |
207 | * At this point page must be either written or cleaned by | |
208 | * truncate. Dirty page here signals a bug and loss of | |
209 | * unwritten data. | |
210 | * | |
211 | * This fixes dirty accounting after removing the page entirely | |
212 | * but leaves PageDirty set: it has no effect for truncated | |
213 | * page and anyway will be cleared before returning page into | |
214 | * buddy allocator. | |
215 | */ | |
216 | if (WARN_ON_ONCE(PageDirty(page))) | |
217 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); | |
218 | } | |
219 | ||
220 | /* | |
221 | * Delete a page from the page cache and free it. Caller has to make | |
222 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 223 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 JK |
224 | */ |
225 | void __delete_from_page_cache(struct page *page, void *shadow) | |
226 | { | |
227 | struct address_space *mapping = page->mapping; | |
228 | ||
229 | trace_mm_filemap_delete_from_page_cache(page); | |
230 | ||
231 | unaccount_page_cache_page(mapping, page); | |
5c024e6a | 232 | page_cache_delete(mapping, page, shadow); |
1da177e4 LT |
233 | } |
234 | ||
59c66c5f JK |
235 | static void page_cache_free_page(struct address_space *mapping, |
236 | struct page *page) | |
237 | { | |
238 | void (*freepage)(struct page *); | |
239 | ||
240 | freepage = mapping->a_ops->freepage; | |
241 | if (freepage) | |
242 | freepage(page); | |
243 | ||
244 | if (PageTransHuge(page) && !PageHuge(page)) { | |
887b22c6 | 245 | page_ref_sub(page, thp_nr_pages(page)); |
59c66c5f JK |
246 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); |
247 | } else { | |
248 | put_page(page); | |
249 | } | |
250 | } | |
251 | ||
702cfbf9 MK |
252 | /** |
253 | * delete_from_page_cache - delete page from page cache | |
254 | * @page: the page which the kernel is trying to remove from page cache | |
255 | * | |
256 | * This must be called only on pages that have been verified to be in the page | |
257 | * cache and locked. It will never put the page into the free list, the caller | |
258 | * has a reference on the page. | |
259 | */ | |
260 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 261 | { |
83929372 | 262 | struct address_space *mapping = page_mapping(page); |
c4843a75 | 263 | unsigned long flags; |
1da177e4 | 264 | |
cd7619d6 | 265 | BUG_ON(!PageLocked(page)); |
b93b0163 | 266 | xa_lock_irqsave(&mapping->i_pages, flags); |
62cccb8c | 267 | __delete_from_page_cache(page, NULL); |
b93b0163 | 268 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
6072d13c | 269 | |
59c66c5f | 270 | page_cache_free_page(mapping, page); |
97cecb5a MK |
271 | } |
272 | EXPORT_SYMBOL(delete_from_page_cache); | |
273 | ||
aa65c29c | 274 | /* |
ef8e5717 | 275 | * page_cache_delete_batch - delete several pages from page cache |
aa65c29c JK |
276 | * @mapping: the mapping to which pages belong |
277 | * @pvec: pagevec with pages to delete | |
278 | * | |
b93b0163 | 279 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
4101196b MWO |
280 | * from the mapping. The function expects @pvec to be sorted by page index |
281 | * and is optimised for it to be dense. | |
b93b0163 | 282 | * It tolerates holes in @pvec (mapping entries at those indices are not |
aa65c29c | 283 | * modified). The function expects only THP head pages to be present in the |
4101196b | 284 | * @pvec. |
aa65c29c | 285 | * |
b93b0163 | 286 | * The function expects the i_pages lock to be held. |
aa65c29c | 287 | */ |
ef8e5717 | 288 | static void page_cache_delete_batch(struct address_space *mapping, |
aa65c29c JK |
289 | struct pagevec *pvec) |
290 | { | |
ef8e5717 | 291 | XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index); |
aa65c29c | 292 | int total_pages = 0; |
4101196b | 293 | int i = 0; |
aa65c29c | 294 | struct page *page; |
aa65c29c | 295 | |
ef8e5717 MW |
296 | mapping_set_update(&xas, mapping); |
297 | xas_for_each(&xas, page, ULONG_MAX) { | |
4101196b | 298 | if (i >= pagevec_count(pvec)) |
aa65c29c | 299 | break; |
4101196b MWO |
300 | |
301 | /* A swap/dax/shadow entry got inserted? Skip it. */ | |
3159f943 | 302 | if (xa_is_value(page)) |
aa65c29c | 303 | continue; |
4101196b MWO |
304 | /* |
305 | * A page got inserted in our range? Skip it. We have our | |
306 | * pages locked so they are protected from being removed. | |
307 | * If we see a page whose index is higher than ours, it | |
308 | * means our page has been removed, which shouldn't be | |
309 | * possible because we're holding the PageLock. | |
310 | */ | |
311 | if (page != pvec->pages[i]) { | |
312 | VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index, | |
313 | page); | |
314 | continue; | |
315 | } | |
316 | ||
317 | WARN_ON_ONCE(!PageLocked(page)); | |
318 | ||
319 | if (page->index == xas.xa_index) | |
aa65c29c | 320 | page->mapping = NULL; |
4101196b MWO |
321 | /* Leave page->index set: truncation lookup relies on it */ |
322 | ||
323 | /* | |
324 | * Move to the next page in the vector if this is a regular | |
325 | * page or the index is of the last sub-page of this compound | |
326 | * page. | |
327 | */ | |
328 | if (page->index + compound_nr(page) - 1 == xas.xa_index) | |
aa65c29c | 329 | i++; |
ef8e5717 | 330 | xas_store(&xas, NULL); |
aa65c29c JK |
331 | total_pages++; |
332 | } | |
333 | mapping->nrpages -= total_pages; | |
334 | } | |
335 | ||
336 | void delete_from_page_cache_batch(struct address_space *mapping, | |
337 | struct pagevec *pvec) | |
338 | { | |
339 | int i; | |
340 | unsigned long flags; | |
341 | ||
342 | if (!pagevec_count(pvec)) | |
343 | return; | |
344 | ||
b93b0163 | 345 | xa_lock_irqsave(&mapping->i_pages, flags); |
aa65c29c JK |
346 | for (i = 0; i < pagevec_count(pvec); i++) { |
347 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); | |
348 | ||
349 | unaccount_page_cache_page(mapping, pvec->pages[i]); | |
350 | } | |
ef8e5717 | 351 | page_cache_delete_batch(mapping, pvec); |
b93b0163 | 352 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
aa65c29c JK |
353 | |
354 | for (i = 0; i < pagevec_count(pvec); i++) | |
355 | page_cache_free_page(mapping, pvec->pages[i]); | |
356 | } | |
357 | ||
d72d9e2a | 358 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
359 | { |
360 | int ret = 0; | |
361 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
362 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
363 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 364 | ret = -ENOSPC; |
7fcbbaf1 JA |
365 | if (test_bit(AS_EIO, &mapping->flags) && |
366 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
367 | ret = -EIO; |
368 | return ret; | |
369 | } | |
d72d9e2a | 370 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 371 | |
76341cab JL |
372 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
373 | { | |
374 | /* Check for outstanding write errors */ | |
375 | if (test_bit(AS_EIO, &mapping->flags)) | |
376 | return -EIO; | |
377 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
378 | return -ENOSPC; | |
379 | return 0; | |
380 | } | |
381 | ||
1da177e4 | 382 | /** |
485bb99b | 383 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
384 | * @mapping: address space structure to write |
385 | * @start: offset in bytes where the range starts | |
469eb4d0 | 386 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 387 | * @sync_mode: enable synchronous operation |
1da177e4 | 388 | * |
485bb99b RD |
389 | * Start writeback against all of a mapping's dirty pages that lie |
390 | * within the byte offsets <start, end> inclusive. | |
391 | * | |
1da177e4 | 392 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 393 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
394 | * these two operations is that if a dirty page/buffer is encountered, it must |
395 | * be waited upon, and not just skipped over. | |
a862f68a MR |
396 | * |
397 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 398 | */ |
ebcf28e1 AM |
399 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
400 | loff_t end, int sync_mode) | |
1da177e4 LT |
401 | { |
402 | int ret; | |
403 | struct writeback_control wbc = { | |
404 | .sync_mode = sync_mode, | |
05fe478d | 405 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
406 | .range_start = start, |
407 | .range_end = end, | |
1da177e4 LT |
408 | }; |
409 | ||
f56753ac | 410 | if (!mapping_can_writeback(mapping) || |
c3aab9a0 | 411 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
1da177e4 LT |
412 | return 0; |
413 | ||
b16b1deb | 414 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 415 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 416 | wbc_detach_inode(&wbc); |
1da177e4 LT |
417 | return ret; |
418 | } | |
419 | ||
420 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
421 | int sync_mode) | |
422 | { | |
111ebb6e | 423 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
424 | } |
425 | ||
426 | int filemap_fdatawrite(struct address_space *mapping) | |
427 | { | |
428 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
429 | } | |
430 | EXPORT_SYMBOL(filemap_fdatawrite); | |
431 | ||
f4c0a0fd | 432 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 433 | loff_t end) |
1da177e4 LT |
434 | { |
435 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
436 | } | |
f4c0a0fd | 437 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 438 | |
485bb99b RD |
439 | /** |
440 | * filemap_flush - mostly a non-blocking flush | |
441 | * @mapping: target address_space | |
442 | * | |
1da177e4 LT |
443 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
444 | * purposes - I/O may not be started against all dirty pages. | |
a862f68a MR |
445 | * |
446 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 LT |
447 | */ |
448 | int filemap_flush(struct address_space *mapping) | |
449 | { | |
450 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
451 | } | |
452 | EXPORT_SYMBOL(filemap_flush); | |
453 | ||
7fc9e472 GR |
454 | /** |
455 | * filemap_range_has_page - check if a page exists in range. | |
456 | * @mapping: address space within which to check | |
457 | * @start_byte: offset in bytes where the range starts | |
458 | * @end_byte: offset in bytes where the range ends (inclusive) | |
459 | * | |
460 | * Find at least one page in the range supplied, usually used to check if | |
461 | * direct writing in this range will trigger a writeback. | |
a862f68a MR |
462 | * |
463 | * Return: %true if at least one page exists in the specified range, | |
464 | * %false otherwise. | |
7fc9e472 GR |
465 | */ |
466 | bool filemap_range_has_page(struct address_space *mapping, | |
467 | loff_t start_byte, loff_t end_byte) | |
468 | { | |
f7b68046 | 469 | struct page *page; |
8fa8e538 MW |
470 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
471 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
7fc9e472 GR |
472 | |
473 | if (end_byte < start_byte) | |
474 | return false; | |
475 | ||
8fa8e538 MW |
476 | rcu_read_lock(); |
477 | for (;;) { | |
478 | page = xas_find(&xas, max); | |
479 | if (xas_retry(&xas, page)) | |
480 | continue; | |
481 | /* Shadow entries don't count */ | |
482 | if (xa_is_value(page)) | |
483 | continue; | |
484 | /* | |
485 | * We don't need to try to pin this page; we're about to | |
486 | * release the RCU lock anyway. It is enough to know that | |
487 | * there was a page here recently. | |
488 | */ | |
489 | break; | |
490 | } | |
491 | rcu_read_unlock(); | |
7fc9e472 | 492 | |
8fa8e538 | 493 | return page != NULL; |
7fc9e472 GR |
494 | } |
495 | EXPORT_SYMBOL(filemap_range_has_page); | |
496 | ||
5e8fcc1a | 497 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 498 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 499 | { |
09cbfeaf KS |
500 | pgoff_t index = start_byte >> PAGE_SHIFT; |
501 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
502 | struct pagevec pvec; |
503 | int nr_pages; | |
1da177e4 | 504 | |
94004ed7 | 505 | if (end_byte < start_byte) |
5e8fcc1a | 506 | return; |
1da177e4 | 507 | |
86679820 | 508 | pagevec_init(&pvec); |
312e9d2f | 509 | while (index <= end) { |
1da177e4 LT |
510 | unsigned i; |
511 | ||
312e9d2f | 512 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 513 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
514 | if (!nr_pages) |
515 | break; | |
516 | ||
1da177e4 LT |
517 | for (i = 0; i < nr_pages; i++) { |
518 | struct page *page = pvec.pages[i]; | |
519 | ||
1da177e4 | 520 | wait_on_page_writeback(page); |
5e8fcc1a | 521 | ClearPageError(page); |
1da177e4 LT |
522 | } |
523 | pagevec_release(&pvec); | |
524 | cond_resched(); | |
525 | } | |
aa750fd7 JN |
526 | } |
527 | ||
528 | /** | |
529 | * filemap_fdatawait_range - wait for writeback to complete | |
530 | * @mapping: address space structure to wait for | |
531 | * @start_byte: offset in bytes where the range starts | |
532 | * @end_byte: offset in bytes where the range ends (inclusive) | |
533 | * | |
534 | * Walk the list of under-writeback pages of the given address space | |
535 | * in the given range and wait for all of them. Check error status of | |
536 | * the address space and return it. | |
537 | * | |
538 | * Since the error status of the address space is cleared by this function, | |
539 | * callers are responsible for checking the return value and handling and/or | |
540 | * reporting the error. | |
a862f68a MR |
541 | * |
542 | * Return: error status of the address space. | |
aa750fd7 JN |
543 | */ |
544 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
545 | loff_t end_byte) | |
546 | { | |
5e8fcc1a JL |
547 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
548 | return filemap_check_errors(mapping); | |
1da177e4 | 549 | } |
d3bccb6f JK |
550 | EXPORT_SYMBOL(filemap_fdatawait_range); |
551 | ||
aa0bfcd9 RZ |
552 | /** |
553 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete | |
554 | * @mapping: address space structure to wait for | |
555 | * @start_byte: offset in bytes where the range starts | |
556 | * @end_byte: offset in bytes where the range ends (inclusive) | |
557 | * | |
558 | * Walk the list of under-writeback pages of the given address space in the | |
559 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), | |
560 | * this function does not clear error status of the address space. | |
561 | * | |
562 | * Use this function if callers don't handle errors themselves. Expected | |
563 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
564 | * fsfreeze(8) | |
565 | */ | |
566 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, | |
567 | loff_t start_byte, loff_t end_byte) | |
568 | { | |
569 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
570 | return filemap_check_and_keep_errors(mapping); | |
571 | } | |
572 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); | |
573 | ||
a823e458 JL |
574 | /** |
575 | * file_fdatawait_range - wait for writeback to complete | |
576 | * @file: file pointing to address space structure to wait for | |
577 | * @start_byte: offset in bytes where the range starts | |
578 | * @end_byte: offset in bytes where the range ends (inclusive) | |
579 | * | |
580 | * Walk the list of under-writeback pages of the address space that file | |
581 | * refers to, in the given range and wait for all of them. Check error | |
582 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
583 | * | |
584 | * Since the error status of the file is advanced by this function, | |
585 | * callers are responsible for checking the return value and handling and/or | |
586 | * reporting the error. | |
a862f68a MR |
587 | * |
588 | * Return: error status of the address space vs. the file->f_wb_err cursor. | |
a823e458 JL |
589 | */ |
590 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
591 | { | |
592 | struct address_space *mapping = file->f_mapping; | |
593 | ||
594 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
595 | return file_check_and_advance_wb_err(file); | |
596 | } | |
597 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 598 | |
aa750fd7 JN |
599 | /** |
600 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
601 | * @mapping: address space structure to wait for | |
602 | * | |
603 | * Walk the list of under-writeback pages of the given address space | |
604 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
605 | * does not clear error status of the address space. | |
606 | * | |
607 | * Use this function if callers don't handle errors themselves. Expected | |
608 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
609 | * fsfreeze(8) | |
a862f68a MR |
610 | * |
611 | * Return: error status of the address space. | |
aa750fd7 | 612 | */ |
76341cab | 613 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 614 | { |
ffb959bb | 615 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 616 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 617 | } |
76341cab | 618 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 619 | |
875d91b1 | 620 | /* Returns true if writeback might be needed or already in progress. */ |
9326c9b2 | 621 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 622 | { |
875d91b1 | 623 | return mapping->nrpages; |
1da177e4 | 624 | } |
1da177e4 | 625 | |
63135aa3 JA |
626 | /** |
627 | * filemap_range_needs_writeback - check if range potentially needs writeback | |
628 | * @mapping: address space within which to check | |
629 | * @start_byte: offset in bytes where the range starts | |
630 | * @end_byte: offset in bytes where the range ends (inclusive) | |
631 | * | |
632 | * Find at least one page in the range supplied, usually used to check if | |
633 | * direct writing in this range will trigger a writeback. Used by O_DIRECT | |
634 | * read/write with IOCB_NOWAIT, to see if the caller needs to do | |
635 | * filemap_write_and_wait_range() before proceeding. | |
636 | * | |
637 | * Return: %true if the caller should do filemap_write_and_wait_range() before | |
638 | * doing O_DIRECT to a page in this range, %false otherwise. | |
639 | */ | |
640 | bool filemap_range_needs_writeback(struct address_space *mapping, | |
641 | loff_t start_byte, loff_t end_byte) | |
642 | { | |
643 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); | |
644 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
645 | struct page *page; | |
646 | ||
647 | if (!mapping_needs_writeback(mapping)) | |
648 | return false; | |
649 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && | |
650 | !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK)) | |
651 | return false; | |
652 | if (end_byte < start_byte) | |
653 | return false; | |
654 | ||
655 | rcu_read_lock(); | |
656 | xas_for_each(&xas, page, max) { | |
657 | if (xas_retry(&xas, page)) | |
658 | continue; | |
659 | if (xa_is_value(page)) | |
660 | continue; | |
661 | if (PageDirty(page) || PageLocked(page) || PageWriteback(page)) | |
662 | break; | |
663 | } | |
664 | rcu_read_unlock(); | |
665 | return page != NULL; | |
666 | } | |
667 | EXPORT_SYMBOL_GPL(filemap_range_needs_writeback); | |
668 | ||
485bb99b RD |
669 | /** |
670 | * filemap_write_and_wait_range - write out & wait on a file range | |
671 | * @mapping: the address_space for the pages | |
672 | * @lstart: offset in bytes where the range starts | |
673 | * @lend: offset in bytes where the range ends (inclusive) | |
674 | * | |
469eb4d0 AM |
675 | * Write out and wait upon file offsets lstart->lend, inclusive. |
676 | * | |
0e056eb5 | 677 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 | 678 | * that this function can be used to write to the very end-of-file (end = -1). |
a862f68a MR |
679 | * |
680 | * Return: error status of the address space. | |
469eb4d0 | 681 | */ |
1da177e4 LT |
682 | int filemap_write_and_wait_range(struct address_space *mapping, |
683 | loff_t lstart, loff_t lend) | |
684 | { | |
28fd1298 | 685 | int err = 0; |
1da177e4 | 686 | |
9326c9b2 | 687 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
688 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
689 | WB_SYNC_ALL); | |
ddf8f376 IW |
690 | /* |
691 | * Even if the above returned error, the pages may be | |
692 | * written partially (e.g. -ENOSPC), so we wait for it. | |
693 | * But the -EIO is special case, it may indicate the worst | |
694 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
695 | */ | |
28fd1298 | 696 | if (err != -EIO) { |
94004ed7 CH |
697 | int err2 = filemap_fdatawait_range(mapping, |
698 | lstart, lend); | |
28fd1298 OH |
699 | if (!err) |
700 | err = err2; | |
cbeaf951 JL |
701 | } else { |
702 | /* Clear any previously stored errors */ | |
703 | filemap_check_errors(mapping); | |
28fd1298 | 704 | } |
865ffef3 DM |
705 | } else { |
706 | err = filemap_check_errors(mapping); | |
1da177e4 | 707 | } |
28fd1298 | 708 | return err; |
1da177e4 | 709 | } |
f6995585 | 710 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 711 | |
5660e13d JL |
712 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
713 | { | |
3acdfd28 | 714 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
715 | |
716 | trace_filemap_set_wb_err(mapping, eseq); | |
717 | } | |
718 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
719 | ||
720 | /** | |
721 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
722 | * and advance wb_err to current one | |
723 | * @file: struct file on which the error is being reported | |
724 | * | |
725 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
726 | * want to report any writeback errors that occurred since the last fsync (or | |
727 | * since the file was opened if there haven't been any). | |
728 | * | |
729 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
730 | * then just quickly return 0. The file is all caught up. | |
731 | * | |
732 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
733 | * it and try to swap it into place. If it works, or another task beat us | |
734 | * to it with the new value, then update the f_wb_err and return the error | |
735 | * portion. The error at this point must be reported via proper channels | |
736 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
737 | * | |
738 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
739 | * value is protected by the f_lock since we must ensure that it reflects | |
740 | * the latest value swapped in for this file descriptor. | |
a862f68a MR |
741 | * |
742 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
743 | */ |
744 | int file_check_and_advance_wb_err(struct file *file) | |
745 | { | |
746 | int err = 0; | |
747 | errseq_t old = READ_ONCE(file->f_wb_err); | |
748 | struct address_space *mapping = file->f_mapping; | |
749 | ||
750 | /* Locklessly handle the common case where nothing has changed */ | |
751 | if (errseq_check(&mapping->wb_err, old)) { | |
752 | /* Something changed, must use slow path */ | |
753 | spin_lock(&file->f_lock); | |
754 | old = file->f_wb_err; | |
755 | err = errseq_check_and_advance(&mapping->wb_err, | |
756 | &file->f_wb_err); | |
757 | trace_file_check_and_advance_wb_err(file, old); | |
758 | spin_unlock(&file->f_lock); | |
759 | } | |
f4e222c5 JL |
760 | |
761 | /* | |
762 | * We're mostly using this function as a drop in replacement for | |
763 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
764 | * that the legacy code would have had on these flags. | |
765 | */ | |
766 | clear_bit(AS_EIO, &mapping->flags); | |
767 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
768 | return err; |
769 | } | |
770 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
771 | ||
772 | /** | |
773 | * file_write_and_wait_range - write out & wait on a file range | |
774 | * @file: file pointing to address_space with pages | |
775 | * @lstart: offset in bytes where the range starts | |
776 | * @lend: offset in bytes where the range ends (inclusive) | |
777 | * | |
778 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
779 | * | |
780 | * Note that @lend is inclusive (describes the last byte to be written) so | |
781 | * that this function can be used to write to the very end-of-file (end = -1). | |
782 | * | |
783 | * After writing out and waiting on the data, we check and advance the | |
784 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
a862f68a MR |
785 | * |
786 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
787 | */ |
788 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
789 | { | |
790 | int err = 0, err2; | |
791 | struct address_space *mapping = file->f_mapping; | |
792 | ||
9326c9b2 | 793 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
794 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
795 | WB_SYNC_ALL); | |
796 | /* See comment of filemap_write_and_wait() */ | |
797 | if (err != -EIO) | |
798 | __filemap_fdatawait_range(mapping, lstart, lend); | |
799 | } | |
800 | err2 = file_check_and_advance_wb_err(file); | |
801 | if (!err) | |
802 | err = err2; | |
803 | return err; | |
804 | } | |
805 | EXPORT_SYMBOL(file_write_and_wait_range); | |
806 | ||
ef6a3c63 MS |
807 | /** |
808 | * replace_page_cache_page - replace a pagecache page with a new one | |
809 | * @old: page to be replaced | |
810 | * @new: page to replace with | |
ef6a3c63 MS |
811 | * |
812 | * This function replaces a page in the pagecache with a new one. On | |
813 | * success it acquires the pagecache reference for the new page and | |
814 | * drops it for the old page. Both the old and new pages must be | |
815 | * locked. This function does not add the new page to the LRU, the | |
816 | * caller must do that. | |
817 | * | |
74d60958 | 818 | * The remove + add is atomic. This function cannot fail. |
ef6a3c63 | 819 | */ |
1f7ef657 | 820 | void replace_page_cache_page(struct page *old, struct page *new) |
ef6a3c63 | 821 | { |
74d60958 MW |
822 | struct address_space *mapping = old->mapping; |
823 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
824 | pgoff_t offset = old->index; | |
825 | XA_STATE(xas, &mapping->i_pages, offset); | |
826 | unsigned long flags; | |
ef6a3c63 | 827 | |
309381fe SL |
828 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
829 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
830 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 831 | |
74d60958 MW |
832 | get_page(new); |
833 | new->mapping = mapping; | |
834 | new->index = offset; | |
ef6a3c63 | 835 | |
0d1c2072 JW |
836 | mem_cgroup_migrate(old, new); |
837 | ||
74d60958 MW |
838 | xas_lock_irqsave(&xas, flags); |
839 | xas_store(&xas, new); | |
4165b9b4 | 840 | |
74d60958 MW |
841 | old->mapping = NULL; |
842 | /* hugetlb pages do not participate in page cache accounting. */ | |
843 | if (!PageHuge(old)) | |
0d1c2072 | 844 | __dec_lruvec_page_state(old, NR_FILE_PAGES); |
74d60958 | 845 | if (!PageHuge(new)) |
0d1c2072 | 846 | __inc_lruvec_page_state(new, NR_FILE_PAGES); |
74d60958 | 847 | if (PageSwapBacked(old)) |
0d1c2072 | 848 | __dec_lruvec_page_state(old, NR_SHMEM); |
74d60958 | 849 | if (PageSwapBacked(new)) |
0d1c2072 | 850 | __inc_lruvec_page_state(new, NR_SHMEM); |
74d60958 | 851 | xas_unlock_irqrestore(&xas, flags); |
74d60958 MW |
852 | if (freepage) |
853 | freepage(old); | |
854 | put_page(old); | |
ef6a3c63 MS |
855 | } |
856 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
857 | ||
16c0cc0c | 858 | noinline int __add_to_page_cache_locked(struct page *page, |
76cd6173 | 859 | struct address_space *mapping, |
c4cf498d | 860 | pgoff_t offset, gfp_t gfp, |
76cd6173 | 861 | void **shadowp) |
1da177e4 | 862 | { |
74d60958 | 863 | XA_STATE(xas, &mapping->i_pages, offset); |
00501b53 | 864 | int huge = PageHuge(page); |
e286781d | 865 | int error; |
da74240e | 866 | bool charged = false; |
e286781d | 867 | |
309381fe SL |
868 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
869 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
74d60958 | 870 | mapping_set_update(&xas, mapping); |
e286781d | 871 | |
09cbfeaf | 872 | get_page(page); |
66a0c8ee KS |
873 | page->mapping = mapping; |
874 | page->index = offset; | |
875 | ||
3fea5a49 | 876 | if (!huge) { |
04f94e3f | 877 | error = mem_cgroup_charge(page, NULL, gfp); |
3fea5a49 JW |
878 | if (error) |
879 | goto error; | |
da74240e | 880 | charged = true; |
3fea5a49 JW |
881 | } |
882 | ||
198b62f8 MWO |
883 | gfp &= GFP_RECLAIM_MASK; |
884 | ||
74d60958 | 885 | do { |
198b62f8 MWO |
886 | unsigned int order = xa_get_order(xas.xa, xas.xa_index); |
887 | void *entry, *old = NULL; | |
888 | ||
889 | if (order > thp_order(page)) | |
890 | xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index), | |
891 | order, gfp); | |
74d60958 | 892 | xas_lock_irq(&xas); |
198b62f8 MWO |
893 | xas_for_each_conflict(&xas, entry) { |
894 | old = entry; | |
895 | if (!xa_is_value(entry)) { | |
896 | xas_set_err(&xas, -EEXIST); | |
897 | goto unlock; | |
898 | } | |
899 | } | |
900 | ||
901 | if (old) { | |
902 | if (shadowp) | |
903 | *shadowp = old; | |
904 | /* entry may have been split before we acquired lock */ | |
905 | order = xa_get_order(xas.xa, xas.xa_index); | |
906 | if (order > thp_order(page)) { | |
907 | xas_split(&xas, old, order); | |
908 | xas_reset(&xas); | |
909 | } | |
910 | } | |
911 | ||
74d60958 MW |
912 | xas_store(&xas, page); |
913 | if (xas_error(&xas)) | |
914 | goto unlock; | |
915 | ||
74d60958 MW |
916 | mapping->nrpages++; |
917 | ||
918 | /* hugetlb pages do not participate in page cache accounting */ | |
919 | if (!huge) | |
0d1c2072 | 920 | __inc_lruvec_page_state(page, NR_FILE_PAGES); |
74d60958 MW |
921 | unlock: |
922 | xas_unlock_irq(&xas); | |
198b62f8 | 923 | } while (xas_nomem(&xas, gfp)); |
74d60958 | 924 | |
3fea5a49 JW |
925 | if (xas_error(&xas)) { |
926 | error = xas_error(&xas); | |
da74240e WL |
927 | if (charged) |
928 | mem_cgroup_uncharge(page); | |
74d60958 | 929 | goto error; |
3fea5a49 | 930 | } |
4165b9b4 | 931 | |
66a0c8ee KS |
932 | trace_mm_filemap_add_to_page_cache(page); |
933 | return 0; | |
74d60958 | 934 | error: |
66a0c8ee KS |
935 | page->mapping = NULL; |
936 | /* Leave page->index set: truncation relies upon it */ | |
09cbfeaf | 937 | put_page(page); |
3fea5a49 | 938 | return error; |
1da177e4 | 939 | } |
cfcbfb13 | 940 | ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO); |
a528910e JW |
941 | |
942 | /** | |
943 | * add_to_page_cache_locked - add a locked page to the pagecache | |
944 | * @page: page to add | |
945 | * @mapping: the page's address_space | |
946 | * @offset: page index | |
947 | * @gfp_mask: page allocation mode | |
948 | * | |
949 | * This function is used to add a page to the pagecache. It must be locked. | |
950 | * This function does not add the page to the LRU. The caller must do that. | |
a862f68a MR |
951 | * |
952 | * Return: %0 on success, negative error code otherwise. | |
a528910e JW |
953 | */ |
954 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
955 | pgoff_t offset, gfp_t gfp_mask) | |
956 | { | |
957 | return __add_to_page_cache_locked(page, mapping, offset, | |
958 | gfp_mask, NULL); | |
959 | } | |
e286781d | 960 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
961 | |
962 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 963 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 964 | { |
a528910e | 965 | void *shadow = NULL; |
4f98a2fe RR |
966 | int ret; |
967 | ||
48c935ad | 968 | __SetPageLocked(page); |
a528910e JW |
969 | ret = __add_to_page_cache_locked(page, mapping, offset, |
970 | gfp_mask, &shadow); | |
971 | if (unlikely(ret)) | |
48c935ad | 972 | __ClearPageLocked(page); |
a528910e JW |
973 | else { |
974 | /* | |
975 | * The page might have been evicted from cache only | |
976 | * recently, in which case it should be activated like | |
977 | * any other repeatedly accessed page. | |
f0281a00 RR |
978 | * The exception is pages getting rewritten; evicting other |
979 | * data from the working set, only to cache data that will | |
980 | * get overwritten with something else, is a waste of memory. | |
a528910e | 981 | */ |
1899ad18 JW |
982 | WARN_ON_ONCE(PageActive(page)); |
983 | if (!(gfp_mask & __GFP_WRITE) && shadow) | |
984 | workingset_refault(page, shadow); | |
a528910e JW |
985 | lru_cache_add(page); |
986 | } | |
1da177e4 LT |
987 | return ret; |
988 | } | |
18bc0bbd | 989 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 990 | |
44110fe3 | 991 | #ifdef CONFIG_NUMA |
2ae88149 | 992 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 993 | { |
c0ff7453 MX |
994 | int n; |
995 | struct page *page; | |
996 | ||
44110fe3 | 997 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
998 | unsigned int cpuset_mems_cookie; |
999 | do { | |
d26914d1 | 1000 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 1001 | n = cpuset_mem_spread_node(); |
96db800f | 1002 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 1003 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 1004 | |
c0ff7453 | 1005 | return page; |
44110fe3 | 1006 | } |
2ae88149 | 1007 | return alloc_pages(gfp, 0); |
44110fe3 | 1008 | } |
2ae88149 | 1009 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
1010 | #endif |
1011 | ||
1da177e4 LT |
1012 | /* |
1013 | * In order to wait for pages to become available there must be | |
1014 | * waitqueues associated with pages. By using a hash table of | |
1015 | * waitqueues where the bucket discipline is to maintain all | |
1016 | * waiters on the same queue and wake all when any of the pages | |
1017 | * become available, and for the woken contexts to check to be | |
1018 | * sure the appropriate page became available, this saves space | |
1019 | * at a cost of "thundering herd" phenomena during rare hash | |
1020 | * collisions. | |
1021 | */ | |
62906027 NP |
1022 | #define PAGE_WAIT_TABLE_BITS 8 |
1023 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
1024 | static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; | |
1025 | ||
1026 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
1da177e4 | 1027 | { |
62906027 | 1028 | return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 1029 | } |
1da177e4 | 1030 | |
62906027 | 1031 | void __init pagecache_init(void) |
1da177e4 | 1032 | { |
62906027 | 1033 | int i; |
1da177e4 | 1034 | |
62906027 NP |
1035 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
1036 | init_waitqueue_head(&page_wait_table[i]); | |
1037 | ||
1038 | page_writeback_init(); | |
1da177e4 | 1039 | } |
1da177e4 | 1040 | |
5ef64cc8 LT |
1041 | /* |
1042 | * The page wait code treats the "wait->flags" somewhat unusually, because | |
5868ec26 | 1043 | * we have multiple different kinds of waits, not just the usual "exclusive" |
5ef64cc8 LT |
1044 | * one. |
1045 | * | |
1046 | * We have: | |
1047 | * | |
1048 | * (a) no special bits set: | |
1049 | * | |
1050 | * We're just waiting for the bit to be released, and when a waker | |
1051 | * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up, | |
1052 | * and remove it from the wait queue. | |
1053 | * | |
1054 | * Simple and straightforward. | |
1055 | * | |
1056 | * (b) WQ_FLAG_EXCLUSIVE: | |
1057 | * | |
1058 | * The waiter is waiting to get the lock, and only one waiter should | |
1059 | * be woken up to avoid any thundering herd behavior. We'll set the | |
1060 | * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue. | |
1061 | * | |
1062 | * This is the traditional exclusive wait. | |
1063 | * | |
5868ec26 | 1064 | * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: |
5ef64cc8 LT |
1065 | * |
1066 | * The waiter is waiting to get the bit, and additionally wants the | |
1067 | * lock to be transferred to it for fair lock behavior. If the lock | |
1068 | * cannot be taken, we stop walking the wait queue without waking | |
1069 | * the waiter. | |
1070 | * | |
1071 | * This is the "fair lock handoff" case, and in addition to setting | |
1072 | * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see | |
1073 | * that it now has the lock. | |
1074 | */ | |
ac6424b9 | 1075 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 1076 | { |
5ef64cc8 | 1077 | unsigned int flags; |
62906027 NP |
1078 | struct wait_page_key *key = arg; |
1079 | struct wait_page_queue *wait_page | |
1080 | = container_of(wait, struct wait_page_queue, wait); | |
1081 | ||
cdc8fcb4 | 1082 | if (!wake_page_match(wait_page, key)) |
62906027 | 1083 | return 0; |
3510ca20 | 1084 | |
9a1ea439 | 1085 | /* |
5ef64cc8 LT |
1086 | * If it's a lock handoff wait, we get the bit for it, and |
1087 | * stop walking (and do not wake it up) if we can't. | |
9a1ea439 | 1088 | */ |
5ef64cc8 LT |
1089 | flags = wait->flags; |
1090 | if (flags & WQ_FLAG_EXCLUSIVE) { | |
1091 | if (test_bit(key->bit_nr, &key->page->flags)) | |
2a9127fc | 1092 | return -1; |
5ef64cc8 LT |
1093 | if (flags & WQ_FLAG_CUSTOM) { |
1094 | if (test_and_set_bit(key->bit_nr, &key->page->flags)) | |
1095 | return -1; | |
1096 | flags |= WQ_FLAG_DONE; | |
1097 | } | |
2a9127fc | 1098 | } |
f62e00cc | 1099 | |
5ef64cc8 LT |
1100 | /* |
1101 | * We are holding the wait-queue lock, but the waiter that | |
1102 | * is waiting for this will be checking the flags without | |
1103 | * any locking. | |
1104 | * | |
1105 | * So update the flags atomically, and wake up the waiter | |
1106 | * afterwards to avoid any races. This store-release pairs | |
1107 | * with the load-acquire in wait_on_page_bit_common(). | |
1108 | */ | |
1109 | smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN); | |
2a9127fc LT |
1110 | wake_up_state(wait->private, mode); |
1111 | ||
1112 | /* | |
1113 | * Ok, we have successfully done what we're waiting for, | |
1114 | * and we can unconditionally remove the wait entry. | |
1115 | * | |
5ef64cc8 LT |
1116 | * Note that this pairs with the "finish_wait()" in the |
1117 | * waiter, and has to be the absolute last thing we do. | |
1118 | * After this list_del_init(&wait->entry) the wait entry | |
2a9127fc LT |
1119 | * might be de-allocated and the process might even have |
1120 | * exited. | |
2a9127fc | 1121 | */ |
c6fe44d9 | 1122 | list_del_init_careful(&wait->entry); |
5ef64cc8 | 1123 | return (flags & WQ_FLAG_EXCLUSIVE) != 0; |
f62e00cc KM |
1124 | } |
1125 | ||
74d81bfa | 1126 | static void wake_up_page_bit(struct page *page, int bit_nr) |
cbbce822 | 1127 | { |
62906027 NP |
1128 | wait_queue_head_t *q = page_waitqueue(page); |
1129 | struct wait_page_key key; | |
1130 | unsigned long flags; | |
11a19c7b | 1131 | wait_queue_entry_t bookmark; |
cbbce822 | 1132 | |
62906027 NP |
1133 | key.page = page; |
1134 | key.bit_nr = bit_nr; | |
1135 | key.page_match = 0; | |
1136 | ||
11a19c7b TC |
1137 | bookmark.flags = 0; |
1138 | bookmark.private = NULL; | |
1139 | bookmark.func = NULL; | |
1140 | INIT_LIST_HEAD(&bookmark.entry); | |
1141 | ||
62906027 | 1142 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1143 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1144 | ||
1145 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1146 | /* | |
1147 | * Take a breather from holding the lock, | |
1148 | * allow pages that finish wake up asynchronously | |
1149 | * to acquire the lock and remove themselves | |
1150 | * from wait queue | |
1151 | */ | |
1152 | spin_unlock_irqrestore(&q->lock, flags); | |
1153 | cpu_relax(); | |
1154 | spin_lock_irqsave(&q->lock, flags); | |
1155 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1156 | } | |
1157 | ||
62906027 NP |
1158 | /* |
1159 | * It is possible for other pages to have collided on the waitqueue | |
1160 | * hash, so in that case check for a page match. That prevents a long- | |
1161 | * term waiter | |
1162 | * | |
1163 | * It is still possible to miss a case here, when we woke page waiters | |
1164 | * and removed them from the waitqueue, but there are still other | |
1165 | * page waiters. | |
1166 | */ | |
1167 | if (!waitqueue_active(q) || !key.page_match) { | |
1168 | ClearPageWaiters(page); | |
1169 | /* | |
1170 | * It's possible to miss clearing Waiters here, when we woke | |
1171 | * our page waiters, but the hashed waitqueue has waiters for | |
1172 | * other pages on it. | |
1173 | * | |
1174 | * That's okay, it's a rare case. The next waker will clear it. | |
1175 | */ | |
1176 | } | |
1177 | spin_unlock_irqrestore(&q->lock, flags); | |
1178 | } | |
74d81bfa NP |
1179 | |
1180 | static void wake_up_page(struct page *page, int bit) | |
1181 | { | |
1182 | if (!PageWaiters(page)) | |
1183 | return; | |
1184 | wake_up_page_bit(page, bit); | |
1185 | } | |
62906027 | 1186 | |
9a1ea439 HD |
1187 | /* |
1188 | * A choice of three behaviors for wait_on_page_bit_common(): | |
1189 | */ | |
1190 | enum behavior { | |
1191 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like | |
1192 | * __lock_page() waiting on then setting PG_locked. | |
1193 | */ | |
1194 | SHARED, /* Hold ref to page and check the bit when woken, like | |
1195 | * wait_on_page_writeback() waiting on PG_writeback. | |
1196 | */ | |
1197 | DROP, /* Drop ref to page before wait, no check when woken, | |
1198 | * like put_and_wait_on_page_locked() on PG_locked. | |
1199 | */ | |
1200 | }; | |
1201 | ||
2a9127fc | 1202 | /* |
5ef64cc8 LT |
1203 | * Attempt to check (or get) the page bit, and mark us done |
1204 | * if successful. | |
2a9127fc LT |
1205 | */ |
1206 | static inline bool trylock_page_bit_common(struct page *page, int bit_nr, | |
1207 | struct wait_queue_entry *wait) | |
1208 | { | |
1209 | if (wait->flags & WQ_FLAG_EXCLUSIVE) { | |
1210 | if (test_and_set_bit(bit_nr, &page->flags)) | |
1211 | return false; | |
1212 | } else if (test_bit(bit_nr, &page->flags)) | |
1213 | return false; | |
1214 | ||
5ef64cc8 | 1215 | wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE; |
2a9127fc LT |
1216 | return true; |
1217 | } | |
1218 | ||
5ef64cc8 LT |
1219 | /* How many times do we accept lock stealing from under a waiter? */ |
1220 | int sysctl_page_lock_unfairness = 5; | |
1221 | ||
62906027 | 1222 | static inline int wait_on_page_bit_common(wait_queue_head_t *q, |
9a1ea439 | 1223 | struct page *page, int bit_nr, int state, enum behavior behavior) |
62906027 | 1224 | { |
5ef64cc8 | 1225 | int unfairness = sysctl_page_lock_unfairness; |
62906027 | 1226 | struct wait_page_queue wait_page; |
ac6424b9 | 1227 | wait_queue_entry_t *wait = &wait_page.wait; |
b1d29ba8 | 1228 | bool thrashing = false; |
9a1ea439 | 1229 | bool delayacct = false; |
eb414681 | 1230 | unsigned long pflags; |
62906027 | 1231 | |
eb414681 | 1232 | if (bit_nr == PG_locked && |
b1d29ba8 | 1233 | !PageUptodate(page) && PageWorkingset(page)) { |
9a1ea439 | 1234 | if (!PageSwapBacked(page)) { |
eb414681 | 1235 | delayacct_thrashing_start(); |
9a1ea439 HD |
1236 | delayacct = true; |
1237 | } | |
eb414681 | 1238 | psi_memstall_enter(&pflags); |
b1d29ba8 JW |
1239 | thrashing = true; |
1240 | } | |
1241 | ||
62906027 NP |
1242 | init_wait(wait); |
1243 | wait->func = wake_page_function; | |
1244 | wait_page.page = page; | |
1245 | wait_page.bit_nr = bit_nr; | |
1246 | ||
5ef64cc8 LT |
1247 | repeat: |
1248 | wait->flags = 0; | |
1249 | if (behavior == EXCLUSIVE) { | |
1250 | wait->flags = WQ_FLAG_EXCLUSIVE; | |
1251 | if (--unfairness < 0) | |
1252 | wait->flags |= WQ_FLAG_CUSTOM; | |
1253 | } | |
1254 | ||
2a9127fc LT |
1255 | /* |
1256 | * Do one last check whether we can get the | |
1257 | * page bit synchronously. | |
1258 | * | |
1259 | * Do the SetPageWaiters() marking before that | |
1260 | * to let any waker we _just_ missed know they | |
1261 | * need to wake us up (otherwise they'll never | |
1262 | * even go to the slow case that looks at the | |
1263 | * page queue), and add ourselves to the wait | |
1264 | * queue if we need to sleep. | |
1265 | * | |
1266 | * This part needs to be done under the queue | |
1267 | * lock to avoid races. | |
1268 | */ | |
1269 | spin_lock_irq(&q->lock); | |
1270 | SetPageWaiters(page); | |
1271 | if (!trylock_page_bit_common(page, bit_nr, wait)) | |
1272 | __add_wait_queue_entry_tail(q, wait); | |
1273 | spin_unlock_irq(&q->lock); | |
62906027 | 1274 | |
2a9127fc LT |
1275 | /* |
1276 | * From now on, all the logic will be based on | |
5ef64cc8 LT |
1277 | * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to |
1278 | * see whether the page bit testing has already | |
1279 | * been done by the wake function. | |
2a9127fc LT |
1280 | * |
1281 | * We can drop our reference to the page. | |
1282 | */ | |
1283 | if (behavior == DROP) | |
1284 | put_page(page); | |
62906027 | 1285 | |
5ef64cc8 LT |
1286 | /* |
1287 | * Note that until the "finish_wait()", or until | |
1288 | * we see the WQ_FLAG_WOKEN flag, we need to | |
1289 | * be very careful with the 'wait->flags', because | |
1290 | * we may race with a waker that sets them. | |
1291 | */ | |
2a9127fc | 1292 | for (;;) { |
5ef64cc8 LT |
1293 | unsigned int flags; |
1294 | ||
62906027 NP |
1295 | set_current_state(state); |
1296 | ||
5ef64cc8 LT |
1297 | /* Loop until we've been woken or interrupted */ |
1298 | flags = smp_load_acquire(&wait->flags); | |
1299 | if (!(flags & WQ_FLAG_WOKEN)) { | |
1300 | if (signal_pending_state(state, current)) | |
1301 | break; | |
1302 | ||
1303 | io_schedule(); | |
1304 | continue; | |
1305 | } | |
1306 | ||
1307 | /* If we were non-exclusive, we're done */ | |
1308 | if (behavior != EXCLUSIVE) | |
a8b169af | 1309 | break; |
9a1ea439 | 1310 | |
5ef64cc8 LT |
1311 | /* If the waker got the lock for us, we're done */ |
1312 | if (flags & WQ_FLAG_DONE) | |
9a1ea439 | 1313 | break; |
2a9127fc | 1314 | |
5ef64cc8 LT |
1315 | /* |
1316 | * Otherwise, if we're getting the lock, we need to | |
1317 | * try to get it ourselves. | |
1318 | * | |
1319 | * And if that fails, we'll have to retry this all. | |
1320 | */ | |
1321 | if (unlikely(test_and_set_bit(bit_nr, &page->flags))) | |
1322 | goto repeat; | |
1323 | ||
1324 | wait->flags |= WQ_FLAG_DONE; | |
1325 | break; | |
62906027 NP |
1326 | } |
1327 | ||
5ef64cc8 LT |
1328 | /* |
1329 | * If a signal happened, this 'finish_wait()' may remove the last | |
1330 | * waiter from the wait-queues, but the PageWaiters bit will remain | |
1331 | * set. That's ok. The next wakeup will take care of it, and trying | |
1332 | * to do it here would be difficult and prone to races. | |
1333 | */ | |
62906027 NP |
1334 | finish_wait(q, wait); |
1335 | ||
eb414681 | 1336 | if (thrashing) { |
9a1ea439 | 1337 | if (delayacct) |
eb414681 JW |
1338 | delayacct_thrashing_end(); |
1339 | psi_memstall_leave(&pflags); | |
1340 | } | |
b1d29ba8 | 1341 | |
62906027 | 1342 | /* |
5ef64cc8 LT |
1343 | * NOTE! The wait->flags weren't stable until we've done the |
1344 | * 'finish_wait()', and we could have exited the loop above due | |
1345 | * to a signal, and had a wakeup event happen after the signal | |
1346 | * test but before the 'finish_wait()'. | |
1347 | * | |
1348 | * So only after the finish_wait() can we reliably determine | |
1349 | * if we got woken up or not, so we can now figure out the final | |
1350 | * return value based on that state without races. | |
1351 | * | |
1352 | * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive | |
1353 | * waiter, but an exclusive one requires WQ_FLAG_DONE. | |
62906027 | 1354 | */ |
5ef64cc8 LT |
1355 | if (behavior == EXCLUSIVE) |
1356 | return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR; | |
62906027 | 1357 | |
2a9127fc | 1358 | return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR; |
62906027 NP |
1359 | } |
1360 | ||
1361 | void wait_on_page_bit(struct page *page, int bit_nr) | |
1362 | { | |
1363 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 | 1364 | wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
62906027 NP |
1365 | } |
1366 | EXPORT_SYMBOL(wait_on_page_bit); | |
1367 | ||
1368 | int wait_on_page_bit_killable(struct page *page, int bit_nr) | |
1369 | { | |
1370 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 | 1371 | return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED); |
cbbce822 | 1372 | } |
4343d008 | 1373 | EXPORT_SYMBOL(wait_on_page_bit_killable); |
cbbce822 | 1374 | |
9a1ea439 HD |
1375 | /** |
1376 | * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked | |
1377 | * @page: The page to wait for. | |
48054625 | 1378 | * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc). |
9a1ea439 HD |
1379 | * |
1380 | * The caller should hold a reference on @page. They expect the page to | |
1381 | * become unlocked relatively soon, but do not wish to hold up migration | |
1382 | * (for example) by holding the reference while waiting for the page to | |
1383 | * come unlocked. After this function returns, the caller should not | |
1384 | * dereference @page. | |
48054625 MWO |
1385 | * |
1386 | * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal. | |
9a1ea439 | 1387 | */ |
48054625 | 1388 | int put_and_wait_on_page_locked(struct page *page, int state) |
9a1ea439 HD |
1389 | { |
1390 | wait_queue_head_t *q; | |
1391 | ||
1392 | page = compound_head(page); | |
1393 | q = page_waitqueue(page); | |
48054625 | 1394 | return wait_on_page_bit_common(q, page, PG_locked, state, DROP); |
9a1ea439 HD |
1395 | } |
1396 | ||
385e1ca5 DH |
1397 | /** |
1398 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
1399 | * @page: Page defining the wait queue of interest |
1400 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
1401 | * |
1402 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
1403 | */ | |
ac6424b9 | 1404 | void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter) |
385e1ca5 DH |
1405 | { |
1406 | wait_queue_head_t *q = page_waitqueue(page); | |
1407 | unsigned long flags; | |
1408 | ||
1409 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1410 | __add_wait_queue_entry_tail(q, waiter); |
62906027 | 1411 | SetPageWaiters(page); |
385e1ca5 DH |
1412 | spin_unlock_irqrestore(&q->lock, flags); |
1413 | } | |
1414 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
1415 | ||
b91e1302 LT |
1416 | #ifndef clear_bit_unlock_is_negative_byte |
1417 | ||
1418 | /* | |
1419 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1420 | * | |
1421 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1422 | * test the sign bit at the same time. But if the architecture does | |
1423 | * not support that special operation, we just do this all by hand | |
1424 | * instead. | |
1425 | * | |
1426 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
ffceeb62 | 1427 | * being cleared, but a memory barrier should be unnecessary since it is |
b91e1302 LT |
1428 | * in the same byte as PG_locked. |
1429 | */ | |
1430 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1431 | { | |
1432 | clear_bit_unlock(nr, mem); | |
1433 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1434 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1435 | } |
1436 | ||
1437 | #endif | |
1438 | ||
1da177e4 | 1439 | /** |
485bb99b | 1440 | * unlock_page - unlock a locked page |
1da177e4 LT |
1441 | * @page: the page |
1442 | * | |
0e9aa675 | 1443 | * Unlocks the page and wakes up sleepers in wait_on_page_locked(). |
1da177e4 | 1444 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup |
da3dae54 | 1445 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
1446 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
1447 | * | |
b91e1302 LT |
1448 | * Note that this depends on PG_waiters being the sign bit in the byte |
1449 | * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to | |
1450 | * clear the PG_locked bit and test PG_waiters at the same time fairly | |
1451 | * portably (architectures that do LL/SC can test any bit, while x86 can | |
1452 | * test the sign bit). | |
1da177e4 | 1453 | */ |
920c7a5d | 1454 | void unlock_page(struct page *page) |
1da177e4 | 1455 | { |
b91e1302 | 1456 | BUILD_BUG_ON(PG_waiters != 7); |
48c935ad | 1457 | page = compound_head(page); |
309381fe | 1458 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
b91e1302 LT |
1459 | if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags)) |
1460 | wake_up_page_bit(page, PG_locked); | |
1da177e4 LT |
1461 | } |
1462 | EXPORT_SYMBOL(unlock_page); | |
1463 | ||
73e10ded DH |
1464 | /** |
1465 | * end_page_private_2 - Clear PG_private_2 and release any waiters | |
1466 | * @page: The page | |
1467 | * | |
1468 | * Clear the PG_private_2 bit on a page and wake up any sleepers waiting for | |
1469 | * this. The page ref held for PG_private_2 being set is released. | |
1470 | * | |
1471 | * This is, for example, used when a netfs page is being written to a local | |
1472 | * disk cache, thereby allowing writes to the cache for the same page to be | |
1473 | * serialised. | |
1474 | */ | |
1475 | void end_page_private_2(struct page *page) | |
1476 | { | |
1477 | page = compound_head(page); | |
1478 | VM_BUG_ON_PAGE(!PagePrivate2(page), page); | |
1479 | clear_bit_unlock(PG_private_2, &page->flags); | |
1480 | wake_up_page_bit(page, PG_private_2); | |
1481 | put_page(page); | |
1482 | } | |
1483 | EXPORT_SYMBOL(end_page_private_2); | |
1484 | ||
1485 | /** | |
1486 | * wait_on_page_private_2 - Wait for PG_private_2 to be cleared on a page | |
1487 | * @page: The page to wait on | |
1488 | * | |
1489 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page. | |
1490 | */ | |
1491 | void wait_on_page_private_2(struct page *page) | |
1492 | { | |
1493 | page = compound_head(page); | |
1494 | while (PagePrivate2(page)) | |
1495 | wait_on_page_bit(page, PG_private_2); | |
1496 | } | |
1497 | EXPORT_SYMBOL(wait_on_page_private_2); | |
1498 | ||
1499 | /** | |
1500 | * wait_on_page_private_2_killable - Wait for PG_private_2 to be cleared on a page | |
1501 | * @page: The page to wait on | |
1502 | * | |
1503 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page or until a | |
1504 | * fatal signal is received by the calling task. | |
1505 | * | |
1506 | * Return: | |
1507 | * - 0 if successful. | |
1508 | * - -EINTR if a fatal signal was encountered. | |
1509 | */ | |
1510 | int wait_on_page_private_2_killable(struct page *page) | |
1511 | { | |
1512 | int ret = 0; | |
1513 | ||
1514 | page = compound_head(page); | |
1515 | while (PagePrivate2(page)) { | |
1516 | ret = wait_on_page_bit_killable(page, PG_private_2); | |
1517 | if (ret < 0) | |
1518 | break; | |
1519 | } | |
1520 | ||
1521 | return ret; | |
1522 | } | |
1523 | EXPORT_SYMBOL(wait_on_page_private_2_killable); | |
1524 | ||
485bb99b RD |
1525 | /** |
1526 | * end_page_writeback - end writeback against a page | |
1527 | * @page: the page | |
1da177e4 LT |
1528 | */ |
1529 | void end_page_writeback(struct page *page) | |
1530 | { | |
888cf2db MG |
1531 | /* |
1532 | * TestClearPageReclaim could be used here but it is an atomic | |
1533 | * operation and overkill in this particular case. Failing to | |
1534 | * shuffle a page marked for immediate reclaim is too mild to | |
1535 | * justify taking an atomic operation penalty at the end of | |
1536 | * ever page writeback. | |
1537 | */ | |
1538 | if (PageReclaim(page)) { | |
1539 | ClearPageReclaim(page); | |
ac6aadb2 | 1540 | rotate_reclaimable_page(page); |
888cf2db | 1541 | } |
ac6aadb2 | 1542 | |
073861ed HD |
1543 | /* |
1544 | * Writeback does not hold a page reference of its own, relying | |
1545 | * on truncation to wait for the clearing of PG_writeback. | |
1546 | * But here we must make sure that the page is not freed and | |
1547 | * reused before the wake_up_page(). | |
1548 | */ | |
1549 | get_page(page); | |
ac6aadb2 MS |
1550 | if (!test_clear_page_writeback(page)) |
1551 | BUG(); | |
1552 | ||
4e857c58 | 1553 | smp_mb__after_atomic(); |
1da177e4 | 1554 | wake_up_page(page, PG_writeback); |
073861ed | 1555 | put_page(page); |
1da177e4 LT |
1556 | } |
1557 | EXPORT_SYMBOL(end_page_writeback); | |
1558 | ||
57d99845 MW |
1559 | /* |
1560 | * After completing I/O on a page, call this routine to update the page | |
1561 | * flags appropriately | |
1562 | */ | |
c11f0c0b | 1563 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1564 | { |
c11f0c0b | 1565 | if (!is_write) { |
57d99845 MW |
1566 | if (!err) { |
1567 | SetPageUptodate(page); | |
1568 | } else { | |
1569 | ClearPageUptodate(page); | |
1570 | SetPageError(page); | |
1571 | } | |
1572 | unlock_page(page); | |
abf54548 | 1573 | } else { |
57d99845 | 1574 | if (err) { |
dd8416c4 MK |
1575 | struct address_space *mapping; |
1576 | ||
57d99845 | 1577 | SetPageError(page); |
dd8416c4 MK |
1578 | mapping = page_mapping(page); |
1579 | if (mapping) | |
1580 | mapping_set_error(mapping, err); | |
57d99845 MW |
1581 | } |
1582 | end_page_writeback(page); | |
1583 | } | |
1584 | } | |
1585 | EXPORT_SYMBOL_GPL(page_endio); | |
1586 | ||
485bb99b RD |
1587 | /** |
1588 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
87066755 | 1589 | * @__page: the page to lock |
1da177e4 | 1590 | */ |
62906027 | 1591 | void __lock_page(struct page *__page) |
1da177e4 | 1592 | { |
62906027 NP |
1593 | struct page *page = compound_head(__page); |
1594 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 HD |
1595 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, |
1596 | EXCLUSIVE); | |
1da177e4 LT |
1597 | } |
1598 | EXPORT_SYMBOL(__lock_page); | |
1599 | ||
62906027 | 1600 | int __lock_page_killable(struct page *__page) |
2687a356 | 1601 | { |
62906027 NP |
1602 | struct page *page = compound_head(__page); |
1603 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 HD |
1604 | return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, |
1605 | EXCLUSIVE); | |
2687a356 | 1606 | } |
18bc0bbd | 1607 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 1608 | |
dd3e6d50 JA |
1609 | int __lock_page_async(struct page *page, struct wait_page_queue *wait) |
1610 | { | |
f32b5dd7 MWO |
1611 | struct wait_queue_head *q = page_waitqueue(page); |
1612 | int ret = 0; | |
1613 | ||
1614 | wait->page = page; | |
1615 | wait->bit_nr = PG_locked; | |
1616 | ||
1617 | spin_lock_irq(&q->lock); | |
1618 | __add_wait_queue_entry_tail(q, &wait->wait); | |
1619 | SetPageWaiters(page); | |
1620 | ret = !trylock_page(page); | |
1621 | /* | |
1622 | * If we were successful now, we know we're still on the | |
1623 | * waitqueue as we're still under the lock. This means it's | |
1624 | * safe to remove and return success, we know the callback | |
1625 | * isn't going to trigger. | |
1626 | */ | |
1627 | if (!ret) | |
1628 | __remove_wait_queue(q, &wait->wait); | |
1629 | else | |
1630 | ret = -EIOCBQUEUED; | |
1631 | spin_unlock_irq(&q->lock); | |
1632 | return ret; | |
dd3e6d50 JA |
1633 | } |
1634 | ||
9a95f3cf PC |
1635 | /* |
1636 | * Return values: | |
c1e8d7c6 | 1637 | * 1 - page is locked; mmap_lock is still held. |
9a95f3cf | 1638 | * 0 - page is not locked. |
3e4e28c5 | 1639 | * mmap_lock has been released (mmap_read_unlock(), unless flags had both |
9a95f3cf | 1640 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
c1e8d7c6 | 1641 | * which case mmap_lock is still held. |
9a95f3cf PC |
1642 | * |
1643 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
c1e8d7c6 | 1644 | * with the page locked and the mmap_lock unperturbed. |
9a95f3cf | 1645 | */ |
d065bd81 ML |
1646 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
1647 | unsigned int flags) | |
1648 | { | |
4064b982 | 1649 | if (fault_flag_allow_retry_first(flags)) { |
37b23e05 | 1650 | /* |
c1e8d7c6 | 1651 | * CAUTION! In this case, mmap_lock is not released |
37b23e05 KM |
1652 | * even though return 0. |
1653 | */ | |
1654 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
1655 | return 0; | |
1656 | ||
d8ed45c5 | 1657 | mmap_read_unlock(mm); |
37b23e05 KM |
1658 | if (flags & FAULT_FLAG_KILLABLE) |
1659 | wait_on_page_locked_killable(page); | |
1660 | else | |
318b275f | 1661 | wait_on_page_locked(page); |
d065bd81 | 1662 | return 0; |
800bca7c HL |
1663 | } |
1664 | if (flags & FAULT_FLAG_KILLABLE) { | |
1665 | int ret; | |
37b23e05 | 1666 | |
800bca7c HL |
1667 | ret = __lock_page_killable(page); |
1668 | if (ret) { | |
1669 | mmap_read_unlock(mm); | |
1670 | return 0; | |
1671 | } | |
1672 | } else { | |
1673 | __lock_page(page); | |
d065bd81 | 1674 | } |
800bca7c HL |
1675 | return 1; |
1676 | ||
d065bd81 ML |
1677 | } |
1678 | ||
e7b563bb | 1679 | /** |
0d3f9296 MW |
1680 | * page_cache_next_miss() - Find the next gap in the page cache. |
1681 | * @mapping: Mapping. | |
1682 | * @index: Index. | |
1683 | * @max_scan: Maximum range to search. | |
e7b563bb | 1684 | * |
0d3f9296 MW |
1685 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1686 | * gap with the lowest index. | |
e7b563bb | 1687 | * |
0d3f9296 MW |
1688 | * This function may be called under the rcu_read_lock. However, this will |
1689 | * not atomically search a snapshot of the cache at a single point in time. | |
1690 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1691 | * created at index 10, page_cache_next_miss covering both indices may | |
1692 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1693 | * |
0d3f9296 MW |
1694 | * Return: The index of the gap if found, otherwise an index outside the |
1695 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1696 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1697 | */ |
0d3f9296 | 1698 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1699 | pgoff_t index, unsigned long max_scan) |
1700 | { | |
0d3f9296 | 1701 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1702 | |
0d3f9296 MW |
1703 | while (max_scan--) { |
1704 | void *entry = xas_next(&xas); | |
1705 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1706 | break; |
0d3f9296 | 1707 | if (xas.xa_index == 0) |
e7b563bb JW |
1708 | break; |
1709 | } | |
1710 | ||
0d3f9296 | 1711 | return xas.xa_index; |
e7b563bb | 1712 | } |
0d3f9296 | 1713 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1714 | |
1715 | /** | |
2346a560 | 1716 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
0d3f9296 MW |
1717 | * @mapping: Mapping. |
1718 | * @index: Index. | |
1719 | * @max_scan: Maximum range to search. | |
e7b563bb | 1720 | * |
0d3f9296 MW |
1721 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1722 | * gap with the highest index. | |
e7b563bb | 1723 | * |
0d3f9296 MW |
1724 | * This function may be called under the rcu_read_lock. However, this will |
1725 | * not atomically search a snapshot of the cache at a single point in time. | |
1726 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1727 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1728 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1729 | * |
0d3f9296 MW |
1730 | * Return: The index of the gap if found, otherwise an index outside the |
1731 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1732 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1733 | */ |
0d3f9296 | 1734 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1735 | pgoff_t index, unsigned long max_scan) |
1736 | { | |
0d3f9296 | 1737 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1738 | |
0d3f9296 MW |
1739 | while (max_scan--) { |
1740 | void *entry = xas_prev(&xas); | |
1741 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1742 | break; |
0d3f9296 | 1743 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1744 | break; |
1745 | } | |
1746 | ||
0d3f9296 | 1747 | return xas.xa_index; |
e7b563bb | 1748 | } |
0d3f9296 | 1749 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1750 | |
44835d20 | 1751 | /* |
bc5a3011 | 1752 | * mapping_get_entry - Get a page cache entry. |
485bb99b | 1753 | * @mapping: the address_space to search |
a6de4b48 | 1754 | * @index: The page cache index. |
0cd6144a | 1755 | * |
4b17f030 | 1756 | * Looks up the page cache slot at @mapping & @index. If there is a |
a6de4b48 | 1757 | * page cache page, the head page is returned with an increased refcount. |
485bb99b | 1758 | * |
139b6a6f JW |
1759 | * If the slot holds a shadow entry of a previously evicted page, or a |
1760 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a | 1761 | * |
a6de4b48 | 1762 | * Return: The head page or shadow entry, %NULL if nothing is found. |
1da177e4 | 1763 | */ |
bc5a3011 MWO |
1764 | static struct page *mapping_get_entry(struct address_space *mapping, |
1765 | pgoff_t index) | |
1da177e4 | 1766 | { |
a6de4b48 | 1767 | XA_STATE(xas, &mapping->i_pages, index); |
4101196b | 1768 | struct page *page; |
1da177e4 | 1769 | |
a60637c8 NP |
1770 | rcu_read_lock(); |
1771 | repeat: | |
4c7472c0 MW |
1772 | xas_reset(&xas); |
1773 | page = xas_load(&xas); | |
1774 | if (xas_retry(&xas, page)) | |
1775 | goto repeat; | |
1776 | /* | |
1777 | * A shadow entry of a recently evicted page, or a swap entry from | |
1778 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1779 | */ | |
1780 | if (!page || xa_is_value(page)) | |
1781 | goto out; | |
83929372 | 1782 | |
4101196b | 1783 | if (!page_cache_get_speculative(page)) |
4c7472c0 | 1784 | goto repeat; |
83929372 | 1785 | |
4c7472c0 | 1786 | /* |
4101196b | 1787 | * Has the page moved or been split? |
4c7472c0 MW |
1788 | * This is part of the lockless pagecache protocol. See |
1789 | * include/linux/pagemap.h for details. | |
1790 | */ | |
1791 | if (unlikely(page != xas_reload(&xas))) { | |
4101196b | 1792 | put_page(page); |
4c7472c0 | 1793 | goto repeat; |
a60637c8 | 1794 | } |
27d20fdd | 1795 | out: |
a60637c8 NP |
1796 | rcu_read_unlock(); |
1797 | ||
1da177e4 LT |
1798 | return page; |
1799 | } | |
1da177e4 | 1800 | |
0cd6144a | 1801 | /** |
2294b32e MWO |
1802 | * pagecache_get_page - Find and get a reference to a page. |
1803 | * @mapping: The address_space to search. | |
1804 | * @index: The page index. | |
1805 | * @fgp_flags: %FGP flags modify how the page is returned. | |
1806 | * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified. | |
1da177e4 | 1807 | * |
2294b32e | 1808 | * Looks up the page cache entry at @mapping & @index. |
0cd6144a | 1809 | * |
2294b32e | 1810 | * @fgp_flags can be zero or more of these flags: |
0e056eb5 | 1811 | * |
2294b32e MWO |
1812 | * * %FGP_ACCESSED - The page will be marked accessed. |
1813 | * * %FGP_LOCK - The page is returned locked. | |
a8cf7f27 MWO |
1814 | * * %FGP_HEAD - If the page is present and a THP, return the head page |
1815 | * rather than the exact page specified by the index. | |
44835d20 MWO |
1816 | * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it |
1817 | * instead of allocating a new page to replace it. | |
2294b32e MWO |
1818 | * * %FGP_CREAT - If no page is present then a new page is allocated using |
1819 | * @gfp_mask and added to the page cache and the VM's LRU list. | |
1820 | * The page is returned locked and with an increased refcount. | |
1821 | * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the | |
1822 | * page is already in cache. If the page was allocated, unlock it before | |
1823 | * returning so the caller can do the same dance. | |
605cad83 YS |
1824 | * * %FGP_WRITE - The page will be written |
1825 | * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask | |
1826 | * * %FGP_NOWAIT - Don't get blocked by page lock | |
1da177e4 | 1827 | * |
2294b32e MWO |
1828 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even |
1829 | * if the %GFP flags specified for %FGP_CREAT are atomic. | |
1da177e4 | 1830 | * |
2457aec6 | 1831 | * If there is a page cache page, it is returned with an increased refcount. |
a862f68a | 1832 | * |
2294b32e | 1833 | * Return: The found page or %NULL otherwise. |
1da177e4 | 1834 | */ |
2294b32e MWO |
1835 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index, |
1836 | int fgp_flags, gfp_t gfp_mask) | |
1da177e4 | 1837 | { |
eb2be189 | 1838 | struct page *page; |
2457aec6 | 1839 | |
1da177e4 | 1840 | repeat: |
bc5a3011 | 1841 | page = mapping_get_entry(mapping, index); |
44835d20 MWO |
1842 | if (xa_is_value(page)) { |
1843 | if (fgp_flags & FGP_ENTRY) | |
1844 | return page; | |
2457aec6 | 1845 | page = NULL; |
44835d20 | 1846 | } |
2457aec6 MG |
1847 | if (!page) |
1848 | goto no_page; | |
1849 | ||
1850 | if (fgp_flags & FGP_LOCK) { | |
1851 | if (fgp_flags & FGP_NOWAIT) { | |
1852 | if (!trylock_page(page)) { | |
09cbfeaf | 1853 | put_page(page); |
2457aec6 MG |
1854 | return NULL; |
1855 | } | |
1856 | } else { | |
1857 | lock_page(page); | |
1858 | } | |
1859 | ||
1860 | /* Has the page been truncated? */ | |
a8cf7f27 | 1861 | if (unlikely(page->mapping != mapping)) { |
2457aec6 | 1862 | unlock_page(page); |
09cbfeaf | 1863 | put_page(page); |
2457aec6 MG |
1864 | goto repeat; |
1865 | } | |
a8cf7f27 | 1866 | VM_BUG_ON_PAGE(!thp_contains(page, index), page); |
2457aec6 MG |
1867 | } |
1868 | ||
c16eb000 | 1869 | if (fgp_flags & FGP_ACCESSED) |
2457aec6 | 1870 | mark_page_accessed(page); |
b9306a79 YS |
1871 | else if (fgp_flags & FGP_WRITE) { |
1872 | /* Clear idle flag for buffer write */ | |
1873 | if (page_is_idle(page)) | |
1874 | clear_page_idle(page); | |
1875 | } | |
a8cf7f27 MWO |
1876 | if (!(fgp_flags & FGP_HEAD)) |
1877 | page = find_subpage(page, index); | |
2457aec6 MG |
1878 | |
1879 | no_page: | |
1880 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1881 | int err; | |
f56753ac | 1882 | if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping)) |
45f87de5 MH |
1883 | gfp_mask |= __GFP_WRITE; |
1884 | if (fgp_flags & FGP_NOFS) | |
1885 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1886 | |
45f87de5 | 1887 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1888 | if (!page) |
1889 | return NULL; | |
2457aec6 | 1890 | |
a75d4c33 | 1891 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
2457aec6 MG |
1892 | fgp_flags |= FGP_LOCK; |
1893 | ||
eb39d618 | 1894 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1895 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1896 | __SetPageReferenced(page); |
2457aec6 | 1897 | |
2294b32e | 1898 | err = add_to_page_cache_lru(page, mapping, index, gfp_mask); |
eb2be189 | 1899 | if (unlikely(err)) { |
09cbfeaf | 1900 | put_page(page); |
eb2be189 NP |
1901 | page = NULL; |
1902 | if (err == -EEXIST) | |
1903 | goto repeat; | |
1da177e4 | 1904 | } |
a75d4c33 JB |
1905 | |
1906 | /* | |
1907 | * add_to_page_cache_lru locks the page, and for mmap we expect | |
1908 | * an unlocked page. | |
1909 | */ | |
1910 | if (page && (fgp_flags & FGP_FOR_MMAP)) | |
1911 | unlock_page(page); | |
1da177e4 | 1912 | } |
2457aec6 | 1913 | |
1da177e4 LT |
1914 | return page; |
1915 | } | |
2457aec6 | 1916 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1917 | |
c7bad633 MWO |
1918 | static inline struct page *find_get_entry(struct xa_state *xas, pgoff_t max, |
1919 | xa_mark_t mark) | |
1920 | { | |
1921 | struct page *page; | |
1922 | ||
1923 | retry: | |
1924 | if (mark == XA_PRESENT) | |
1925 | page = xas_find(xas, max); | |
1926 | else | |
1927 | page = xas_find_marked(xas, max, mark); | |
1928 | ||
1929 | if (xas_retry(xas, page)) | |
1930 | goto retry; | |
1931 | /* | |
1932 | * A shadow entry of a recently evicted page, a swap | |
1933 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1934 | * without attempting to raise page count. | |
1935 | */ | |
1936 | if (!page || xa_is_value(page)) | |
1937 | return page; | |
1938 | ||
1939 | if (!page_cache_get_speculative(page)) | |
1940 | goto reset; | |
1941 | ||
1942 | /* Has the page moved or been split? */ | |
1943 | if (unlikely(page != xas_reload(xas))) { | |
1944 | put_page(page); | |
1945 | goto reset; | |
1946 | } | |
1947 | ||
1948 | return page; | |
1949 | reset: | |
1950 | xas_reset(xas); | |
1951 | goto retry; | |
1952 | } | |
1953 | ||
0cd6144a JW |
1954 | /** |
1955 | * find_get_entries - gang pagecache lookup | |
1956 | * @mapping: The address_space to search | |
1957 | * @start: The starting page cache index | |
ca122fe4 | 1958 | * @end: The final page index (inclusive). |
cf2039af | 1959 | * @pvec: Where the resulting entries are placed. |
0cd6144a JW |
1960 | * @indices: The cache indices corresponding to the entries in @entries |
1961 | * | |
cf2039af MWO |
1962 | * find_get_entries() will search for and return a batch of entries in |
1963 | * the mapping. The entries are placed in @pvec. find_get_entries() | |
1964 | * takes a reference on any actual pages it returns. | |
0cd6144a JW |
1965 | * |
1966 | * The search returns a group of mapping-contiguous page cache entries | |
1967 | * with ascending indexes. There may be holes in the indices due to | |
1968 | * not-present pages. | |
1969 | * | |
139b6a6f JW |
1970 | * Any shadow entries of evicted pages, or swap entries from |
1971 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a | 1972 | * |
71725ed1 HD |
1973 | * If it finds a Transparent Huge Page, head or tail, find_get_entries() |
1974 | * stops at that page: the caller is likely to have a better way to handle | |
1975 | * the compound page as a whole, and then skip its extent, than repeatedly | |
1976 | * calling find_get_entries() to return all its tails. | |
1977 | * | |
a862f68a | 1978 | * Return: the number of pages and shadow entries which were found. |
0cd6144a | 1979 | */ |
ca122fe4 | 1980 | unsigned find_get_entries(struct address_space *mapping, pgoff_t start, |
cf2039af | 1981 | pgoff_t end, struct pagevec *pvec, pgoff_t *indices) |
0cd6144a | 1982 | { |
f280bf09 MW |
1983 | XA_STATE(xas, &mapping->i_pages, start); |
1984 | struct page *page; | |
0cd6144a | 1985 | unsigned int ret = 0; |
cf2039af | 1986 | unsigned nr_entries = PAGEVEC_SIZE; |
0cd6144a JW |
1987 | |
1988 | rcu_read_lock(); | |
ca122fe4 | 1989 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { |
71725ed1 HD |
1990 | /* |
1991 | * Terminate early on finding a THP, to allow the caller to | |
1992 | * handle it all at once; but continue if this is hugetlbfs. | |
1993 | */ | |
c7bad633 MWO |
1994 | if (!xa_is_value(page) && PageTransHuge(page) && |
1995 | !PageHuge(page)) { | |
71725ed1 HD |
1996 | page = find_subpage(page, xas.xa_index); |
1997 | nr_entries = ret + 1; | |
1998 | } | |
c7bad633 | 1999 | |
f280bf09 | 2000 | indices[ret] = xas.xa_index; |
cf2039af | 2001 | pvec->pages[ret] = page; |
0cd6144a JW |
2002 | if (++ret == nr_entries) |
2003 | break; | |
2004 | } | |
2005 | rcu_read_unlock(); | |
cf2039af MWO |
2006 | |
2007 | pvec->nr = ret; | |
0cd6144a JW |
2008 | return ret; |
2009 | } | |
2010 | ||
5c211ba2 MWO |
2011 | /** |
2012 | * find_lock_entries - Find a batch of pagecache entries. | |
2013 | * @mapping: The address_space to search. | |
2014 | * @start: The starting page cache index. | |
2015 | * @end: The final page index (inclusive). | |
2016 | * @pvec: Where the resulting entries are placed. | |
2017 | * @indices: The cache indices of the entries in @pvec. | |
2018 | * | |
2019 | * find_lock_entries() will return a batch of entries from @mapping. | |
2020 | * Swap, shadow and DAX entries are included. Pages are returned | |
2021 | * locked and with an incremented refcount. Pages which are locked by | |
2022 | * somebody else or under writeback are skipped. Only the head page of | |
2023 | * a THP is returned. Pages which are partially outside the range are | |
2024 | * not returned. | |
2025 | * | |
2026 | * The entries have ascending indexes. The indices may not be consecutive | |
2027 | * due to not-present entries, THP pages, pages which could not be locked | |
2028 | * or pages under writeback. | |
2029 | * | |
2030 | * Return: The number of entries which were found. | |
2031 | */ | |
2032 | unsigned find_lock_entries(struct address_space *mapping, pgoff_t start, | |
2033 | pgoff_t end, struct pagevec *pvec, pgoff_t *indices) | |
2034 | { | |
2035 | XA_STATE(xas, &mapping->i_pages, start); | |
2036 | struct page *page; | |
2037 | ||
2038 | rcu_read_lock(); | |
2039 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { | |
2040 | if (!xa_is_value(page)) { | |
2041 | if (page->index < start) | |
2042 | goto put; | |
2043 | VM_BUG_ON_PAGE(page->index != xas.xa_index, page); | |
2044 | if (page->index + thp_nr_pages(page) - 1 > end) | |
2045 | goto put; | |
2046 | if (!trylock_page(page)) | |
2047 | goto put; | |
2048 | if (page->mapping != mapping || PageWriteback(page)) | |
2049 | goto unlock; | |
2050 | VM_BUG_ON_PAGE(!thp_contains(page, xas.xa_index), | |
2051 | page); | |
2052 | } | |
2053 | indices[pvec->nr] = xas.xa_index; | |
2054 | if (!pagevec_add(pvec, page)) | |
2055 | break; | |
2056 | goto next; | |
2057 | unlock: | |
2058 | unlock_page(page); | |
2059 | put: | |
2060 | put_page(page); | |
2061 | next: | |
2d11e738 HD |
2062 | if (!xa_is_value(page) && PageTransHuge(page)) { |
2063 | unsigned int nr_pages = thp_nr_pages(page); | |
2064 | ||
2065 | /* Final THP may cross MAX_LFS_FILESIZE on 32-bit */ | |
2066 | xas_set(&xas, page->index + nr_pages); | |
2067 | if (xas.xa_index < nr_pages) | |
2068 | break; | |
2069 | } | |
5c211ba2 MWO |
2070 | } |
2071 | rcu_read_unlock(); | |
2072 | ||
2073 | return pagevec_count(pvec); | |
2074 | } | |
2075 | ||
1da177e4 | 2076 | /** |
b947cee4 | 2077 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
2078 | * @mapping: The address_space to search |
2079 | * @start: The starting page index | |
b947cee4 | 2080 | * @end: The final page index (inclusive) |
1da177e4 LT |
2081 | * @nr_pages: The maximum number of pages |
2082 | * @pages: Where the resulting pages are placed | |
2083 | * | |
b947cee4 JK |
2084 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
2085 | * pages in the mapping starting at index @start and up to index @end | |
2086 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
2087 | * a reference against the returned pages. | |
1da177e4 LT |
2088 | * |
2089 | * The search returns a group of mapping-contiguous pages with ascending | |
2090 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 2091 | * We also update @start to index the next page for the traversal. |
1da177e4 | 2092 | * |
a862f68a MR |
2093 | * Return: the number of pages which were found. If this number is |
2094 | * smaller than @nr_pages, the end of specified range has been | |
b947cee4 | 2095 | * reached. |
1da177e4 | 2096 | */ |
b947cee4 JK |
2097 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
2098 | pgoff_t end, unsigned int nr_pages, | |
2099 | struct page **pages) | |
1da177e4 | 2100 | { |
fd1b3cee MW |
2101 | XA_STATE(xas, &mapping->i_pages, *start); |
2102 | struct page *page; | |
0fc9d104 KK |
2103 | unsigned ret = 0; |
2104 | ||
2105 | if (unlikely(!nr_pages)) | |
2106 | return 0; | |
a60637c8 NP |
2107 | |
2108 | rcu_read_lock(); | |
c7bad633 | 2109 | while ((page = find_get_entry(&xas, end, XA_PRESENT))) { |
fd1b3cee MW |
2110 | /* Skip over shadow, swap and DAX entries */ |
2111 | if (xa_is_value(page)) | |
8079b1c8 | 2112 | continue; |
a60637c8 | 2113 | |
4101196b | 2114 | pages[ret] = find_subpage(page, xas.xa_index); |
b947cee4 | 2115 | if (++ret == nr_pages) { |
5d3ee42f | 2116 | *start = xas.xa_index + 1; |
b947cee4 JK |
2117 | goto out; |
2118 | } | |
a60637c8 | 2119 | } |
5b280c0c | 2120 | |
b947cee4 JK |
2121 | /* |
2122 | * We come here when there is no page beyond @end. We take care to not | |
2123 | * overflow the index @start as it confuses some of the callers. This | |
fd1b3cee | 2124 | * breaks the iteration when there is a page at index -1 but that is |
b947cee4 JK |
2125 | * already broken anyway. |
2126 | */ | |
2127 | if (end == (pgoff_t)-1) | |
2128 | *start = (pgoff_t)-1; | |
2129 | else | |
2130 | *start = end + 1; | |
2131 | out: | |
a60637c8 | 2132 | rcu_read_unlock(); |
d72dc8a2 | 2133 | |
1da177e4 LT |
2134 | return ret; |
2135 | } | |
2136 | ||
ebf43500 JA |
2137 | /** |
2138 | * find_get_pages_contig - gang contiguous pagecache lookup | |
2139 | * @mapping: The address_space to search | |
2140 | * @index: The starting page index | |
2141 | * @nr_pages: The maximum number of pages | |
2142 | * @pages: Where the resulting pages are placed | |
2143 | * | |
2144 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
2145 | * that the returned number of pages are guaranteed to be contiguous. | |
2146 | * | |
a862f68a | 2147 | * Return: the number of pages which were found. |
ebf43500 JA |
2148 | */ |
2149 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
2150 | unsigned int nr_pages, struct page **pages) | |
2151 | { | |
3ece58a2 MW |
2152 | XA_STATE(xas, &mapping->i_pages, index); |
2153 | struct page *page; | |
0fc9d104 KK |
2154 | unsigned int ret = 0; |
2155 | ||
2156 | if (unlikely(!nr_pages)) | |
2157 | return 0; | |
a60637c8 NP |
2158 | |
2159 | rcu_read_lock(); | |
3ece58a2 | 2160 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { |
3ece58a2 MW |
2161 | if (xas_retry(&xas, page)) |
2162 | continue; | |
2163 | /* | |
2164 | * If the entry has been swapped out, we can stop looking. | |
2165 | * No current caller is looking for DAX entries. | |
2166 | */ | |
2167 | if (xa_is_value(page)) | |
8079b1c8 | 2168 | break; |
ebf43500 | 2169 | |
4101196b | 2170 | if (!page_cache_get_speculative(page)) |
3ece58a2 | 2171 | goto retry; |
83929372 | 2172 | |
4101196b | 2173 | /* Has the page moved or been split? */ |
3ece58a2 MW |
2174 | if (unlikely(page != xas_reload(&xas))) |
2175 | goto put_page; | |
a60637c8 | 2176 | |
4101196b | 2177 | pages[ret] = find_subpage(page, xas.xa_index); |
0fc9d104 KK |
2178 | if (++ret == nr_pages) |
2179 | break; | |
3ece58a2 MW |
2180 | continue; |
2181 | put_page: | |
4101196b | 2182 | put_page(page); |
3ece58a2 MW |
2183 | retry: |
2184 | xas_reset(&xas); | |
ebf43500 | 2185 | } |
a60637c8 NP |
2186 | rcu_read_unlock(); |
2187 | return ret; | |
ebf43500 | 2188 | } |
ef71c15c | 2189 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 2190 | |
485bb99b | 2191 | /** |
c49f50d1 | 2192 | * find_get_pages_range_tag - Find and return head pages matching @tag. |
485bb99b RD |
2193 | * @mapping: the address_space to search |
2194 | * @index: the starting page index | |
72b045ae | 2195 | * @end: The final page index (inclusive) |
485bb99b RD |
2196 | * @tag: the tag index |
2197 | * @nr_pages: the maximum number of pages | |
2198 | * @pages: where the resulting pages are placed | |
2199 | * | |
c49f50d1 MWO |
2200 | * Like find_get_pages(), except we only return head pages which are tagged |
2201 | * with @tag. @index is updated to the index immediately after the last | |
2202 | * page we return, ready for the next iteration. | |
a862f68a MR |
2203 | * |
2204 | * Return: the number of pages which were found. | |
1da177e4 | 2205 | */ |
72b045ae | 2206 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
a6906972 | 2207 | pgoff_t end, xa_mark_t tag, unsigned int nr_pages, |
72b045ae | 2208 | struct page **pages) |
1da177e4 | 2209 | { |
a6906972 MW |
2210 | XA_STATE(xas, &mapping->i_pages, *index); |
2211 | struct page *page; | |
0fc9d104 KK |
2212 | unsigned ret = 0; |
2213 | ||
2214 | if (unlikely(!nr_pages)) | |
2215 | return 0; | |
a60637c8 NP |
2216 | |
2217 | rcu_read_lock(); | |
c7bad633 | 2218 | while ((page = find_get_entry(&xas, end, tag))) { |
a6906972 MW |
2219 | /* |
2220 | * Shadow entries should never be tagged, but this iteration | |
2221 | * is lockless so there is a window for page reclaim to evict | |
2222 | * a page we saw tagged. Skip over it. | |
2223 | */ | |
2224 | if (xa_is_value(page)) | |
139b6a6f | 2225 | continue; |
a60637c8 | 2226 | |
c49f50d1 | 2227 | pages[ret] = page; |
72b045ae | 2228 | if (++ret == nr_pages) { |
c49f50d1 | 2229 | *index = page->index + thp_nr_pages(page); |
72b045ae JK |
2230 | goto out; |
2231 | } | |
a60637c8 | 2232 | } |
5b280c0c | 2233 | |
72b045ae | 2234 | /* |
a6906972 | 2235 | * We come here when we got to @end. We take care to not overflow the |
72b045ae | 2236 | * index @index as it confuses some of the callers. This breaks the |
a6906972 MW |
2237 | * iteration when there is a page at index -1 but that is already |
2238 | * broken anyway. | |
72b045ae JK |
2239 | */ |
2240 | if (end == (pgoff_t)-1) | |
2241 | *index = (pgoff_t)-1; | |
2242 | else | |
2243 | *index = end + 1; | |
2244 | out: | |
a60637c8 | 2245 | rcu_read_unlock(); |
1da177e4 | 2246 | |
1da177e4 LT |
2247 | return ret; |
2248 | } | |
72b045ae | 2249 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 2250 | |
76d42bd9 WF |
2251 | /* |
2252 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
2253 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
2254 | * | |
2255 | * ---R__________________________________________B__________ | |
2256 | * ^ reading here ^ bad block(assume 4k) | |
2257 | * | |
2258 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
2259 | * => failing the whole request => read(R) => read(R+1) => | |
2260 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
2261 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
2262 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
2263 | * | |
2264 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
2265 | */ | |
0f8e2db4 | 2266 | static void shrink_readahead_size_eio(struct file_ra_state *ra) |
76d42bd9 | 2267 | { |
76d42bd9 | 2268 | ra->ra_pages /= 4; |
76d42bd9 WF |
2269 | } |
2270 | ||
cbd59c48 MWO |
2271 | /* |
2272 | * filemap_get_read_batch - Get a batch of pages for read | |
2273 | * | |
2274 | * Get a batch of pages which represent a contiguous range of bytes | |
2275 | * in the file. No tail pages will be returned. If @index is in the | |
2276 | * middle of a THP, the entire THP will be returned. The last page in | |
2277 | * the batch may have Readahead set or be not Uptodate so that the | |
2278 | * caller can take the appropriate action. | |
2279 | */ | |
2280 | static void filemap_get_read_batch(struct address_space *mapping, | |
2281 | pgoff_t index, pgoff_t max, struct pagevec *pvec) | |
2282 | { | |
2283 | XA_STATE(xas, &mapping->i_pages, index); | |
2284 | struct page *head; | |
2285 | ||
2286 | rcu_read_lock(); | |
2287 | for (head = xas_load(&xas); head; head = xas_next(&xas)) { | |
2288 | if (xas_retry(&xas, head)) | |
2289 | continue; | |
2290 | if (xas.xa_index > max || xa_is_value(head)) | |
2291 | break; | |
2292 | if (!page_cache_get_speculative(head)) | |
2293 | goto retry; | |
2294 | ||
2295 | /* Has the page moved or been split? */ | |
2296 | if (unlikely(head != xas_reload(&xas))) | |
2297 | goto put_page; | |
2298 | ||
2299 | if (!pagevec_add(pvec, head)) | |
2300 | break; | |
2301 | if (!PageUptodate(head)) | |
2302 | break; | |
2303 | if (PageReadahead(head)) | |
2304 | break; | |
2305 | xas.xa_index = head->index + thp_nr_pages(head) - 1; | |
2306 | xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK; | |
2307 | continue; | |
2308 | put_page: | |
2309 | put_page(head); | |
2310 | retry: | |
2311 | xas_reset(&xas); | |
2312 | } | |
2313 | rcu_read_unlock(); | |
2314 | } | |
2315 | ||
68430303 MWO |
2316 | static int filemap_read_page(struct file *file, struct address_space *mapping, |
2317 | struct page *page) | |
723ef24b | 2318 | { |
723ef24b KO |
2319 | int error; |
2320 | ||
723ef24b | 2321 | /* |
68430303 MWO |
2322 | * A previous I/O error may have been due to temporary failures, |
2323 | * eg. multipath errors. PG_error will be set again if readpage | |
2324 | * fails. | |
723ef24b KO |
2325 | */ |
2326 | ClearPageError(page); | |
2327 | /* Start the actual read. The read will unlock the page. */ | |
68430303 MWO |
2328 | error = mapping->a_ops->readpage(file, page); |
2329 | if (error) | |
2330 | return error; | |
723ef24b | 2331 | |
aa1ec2f6 | 2332 | error = wait_on_page_locked_killable(page); |
68430303 MWO |
2333 | if (error) |
2334 | return error; | |
aa1ec2f6 MWO |
2335 | if (PageUptodate(page)) |
2336 | return 0; | |
aa1ec2f6 MWO |
2337 | shrink_readahead_size_eio(&file->f_ra); |
2338 | return -EIO; | |
723ef24b KO |
2339 | } |
2340 | ||
fce70da3 MWO |
2341 | static bool filemap_range_uptodate(struct address_space *mapping, |
2342 | loff_t pos, struct iov_iter *iter, struct page *page) | |
2343 | { | |
2344 | int count; | |
2345 | ||
2346 | if (PageUptodate(page)) | |
2347 | return true; | |
2348 | /* pipes can't handle partially uptodate pages */ | |
2349 | if (iov_iter_is_pipe(iter)) | |
2350 | return false; | |
2351 | if (!mapping->a_ops->is_partially_uptodate) | |
2352 | return false; | |
2353 | if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page))) | |
2354 | return false; | |
2355 | ||
2356 | count = iter->count; | |
2357 | if (page_offset(page) > pos) { | |
2358 | count -= page_offset(page) - pos; | |
2359 | pos = 0; | |
2360 | } else { | |
2361 | pos -= page_offset(page); | |
2362 | } | |
2363 | ||
2364 | return mapping->a_ops->is_partially_uptodate(page, pos, count); | |
2365 | } | |
2366 | ||
4612aeef MWO |
2367 | static int filemap_update_page(struct kiocb *iocb, |
2368 | struct address_space *mapping, struct iov_iter *iter, | |
fce70da3 | 2369 | struct page *page) |
723ef24b | 2370 | { |
723ef24b KO |
2371 | int error; |
2372 | ||
730633f0 JK |
2373 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2374 | if (!filemap_invalidate_trylock_shared(mapping)) | |
2375 | return -EAGAIN; | |
2376 | } else { | |
2377 | filemap_invalidate_lock_shared(mapping); | |
2378 | } | |
2379 | ||
87d1d7b6 | 2380 | if (!trylock_page(page)) { |
730633f0 | 2381 | error = -EAGAIN; |
87d1d7b6 | 2382 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) |
730633f0 | 2383 | goto unlock_mapping; |
87d1d7b6 | 2384 | if (!(iocb->ki_flags & IOCB_WAITQ)) { |
730633f0 | 2385 | filemap_invalidate_unlock_shared(mapping); |
bd8a1f36 | 2386 | put_and_wait_on_page_locked(page, TASK_KILLABLE); |
4612aeef | 2387 | return AOP_TRUNCATED_PAGE; |
bd8a1f36 | 2388 | } |
87d1d7b6 MWO |
2389 | error = __lock_page_async(page, iocb->ki_waitq); |
2390 | if (error) | |
730633f0 | 2391 | goto unlock_mapping; |
723ef24b | 2392 | } |
723ef24b | 2393 | |
730633f0 | 2394 | error = AOP_TRUNCATED_PAGE; |
bd8a1f36 | 2395 | if (!page->mapping) |
730633f0 | 2396 | goto unlock; |
723ef24b | 2397 | |
fce70da3 MWO |
2398 | error = 0; |
2399 | if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, page)) | |
2400 | goto unlock; | |
2401 | ||
2402 | error = -EAGAIN; | |
2403 | if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ)) | |
2404 | goto unlock; | |
2405 | ||
68430303 | 2406 | error = filemap_read_page(iocb->ki_filp, mapping, page); |
730633f0 | 2407 | goto unlock_mapping; |
fce70da3 MWO |
2408 | unlock: |
2409 | unlock_page(page); | |
730633f0 JK |
2410 | unlock_mapping: |
2411 | filemap_invalidate_unlock_shared(mapping); | |
2412 | if (error == AOP_TRUNCATED_PAGE) | |
2413 | put_page(page); | |
fce70da3 | 2414 | return error; |
723ef24b KO |
2415 | } |
2416 | ||
f253e185 MWO |
2417 | static int filemap_create_page(struct file *file, |
2418 | struct address_space *mapping, pgoff_t index, | |
2419 | struct pagevec *pvec) | |
723ef24b | 2420 | { |
723ef24b KO |
2421 | struct page *page; |
2422 | int error; | |
2423 | ||
723ef24b KO |
2424 | page = page_cache_alloc(mapping); |
2425 | if (!page) | |
f253e185 | 2426 | return -ENOMEM; |
723ef24b | 2427 | |
730633f0 JK |
2428 | /* |
2429 | * Protect against truncate / hole punch. Grabbing invalidate_lock here | |
2430 | * assures we cannot instantiate and bring uptodate new pagecache pages | |
2431 | * after evicting page cache during truncate and before actually | |
2432 | * freeing blocks. Note that we could release invalidate_lock after | |
2433 | * inserting the page into page cache as the locked page would then be | |
2434 | * enough to synchronize with hole punching. But there are code paths | |
2435 | * such as filemap_update_page() filling in partially uptodate pages or | |
2436 | * ->readpages() that need to hold invalidate_lock while mapping blocks | |
2437 | * for IO so let's hold the lock here as well to keep locking rules | |
2438 | * simple. | |
2439 | */ | |
2440 | filemap_invalidate_lock_shared(mapping); | |
723ef24b | 2441 | error = add_to_page_cache_lru(page, mapping, index, |
f253e185 MWO |
2442 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
2443 | if (error == -EEXIST) | |
2444 | error = AOP_TRUNCATED_PAGE; | |
2445 | if (error) | |
2446 | goto error; | |
2447 | ||
2448 | error = filemap_read_page(file, mapping, page); | |
2449 | if (error) | |
2450 | goto error; | |
2451 | ||
730633f0 | 2452 | filemap_invalidate_unlock_shared(mapping); |
f253e185 MWO |
2453 | pagevec_add(pvec, page); |
2454 | return 0; | |
2455 | error: | |
730633f0 | 2456 | filemap_invalidate_unlock_shared(mapping); |
68430303 | 2457 | put_page(page); |
f253e185 | 2458 | return error; |
723ef24b KO |
2459 | } |
2460 | ||
5963fe03 MWO |
2461 | static int filemap_readahead(struct kiocb *iocb, struct file *file, |
2462 | struct address_space *mapping, struct page *page, | |
2463 | pgoff_t last_index) | |
2464 | { | |
2465 | if (iocb->ki_flags & IOCB_NOIO) | |
2466 | return -EAGAIN; | |
2467 | page_cache_async_readahead(mapping, &file->f_ra, file, page, | |
2468 | page->index, last_index - page->index); | |
2469 | return 0; | |
2470 | } | |
2471 | ||
3a6bae48 | 2472 | static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter, |
ff993ba1 | 2473 | struct pagevec *pvec) |
06c04442 KO |
2474 | { |
2475 | struct file *filp = iocb->ki_filp; | |
2476 | struct address_space *mapping = filp->f_mapping; | |
2477 | struct file_ra_state *ra = &filp->f_ra; | |
2478 | pgoff_t index = iocb->ki_pos >> PAGE_SHIFT; | |
cbd59c48 | 2479 | pgoff_t last_index; |
2642fca6 | 2480 | struct page *page; |
cbd59c48 | 2481 | int err = 0; |
06c04442 | 2482 | |
cbd59c48 | 2483 | last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE); |
2642fca6 | 2484 | retry: |
06c04442 KO |
2485 | if (fatal_signal_pending(current)) |
2486 | return -EINTR; | |
2487 | ||
cbd59c48 | 2488 | filemap_get_read_batch(mapping, index, last_index, pvec); |
2642fca6 MWO |
2489 | if (!pagevec_count(pvec)) { |
2490 | if (iocb->ki_flags & IOCB_NOIO) | |
2491 | return -EAGAIN; | |
2492 | page_cache_sync_readahead(mapping, ra, filp, index, | |
2493 | last_index - index); | |
2494 | filemap_get_read_batch(mapping, index, last_index, pvec); | |
2495 | } | |
f253e185 MWO |
2496 | if (!pagevec_count(pvec)) { |
2497 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ)) | |
2498 | return -EAGAIN; | |
2499 | err = filemap_create_page(filp, mapping, | |
2500 | iocb->ki_pos >> PAGE_SHIFT, pvec); | |
2501 | if (err == AOP_TRUNCATED_PAGE) | |
2642fca6 | 2502 | goto retry; |
f253e185 MWO |
2503 | return err; |
2504 | } | |
06c04442 | 2505 | |
2642fca6 MWO |
2506 | page = pvec->pages[pagevec_count(pvec) - 1]; |
2507 | if (PageReadahead(page)) { | |
2508 | err = filemap_readahead(iocb, filp, mapping, page, last_index); | |
2509 | if (err) | |
2510 | goto err; | |
2511 | } | |
2512 | if (!PageUptodate(page)) { | |
2513 | if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1) | |
2514 | iocb->ki_flags |= IOCB_NOWAIT; | |
2515 | err = filemap_update_page(iocb, mapping, iter, page); | |
2516 | if (err) | |
2517 | goto err; | |
06c04442 KO |
2518 | } |
2519 | ||
2642fca6 | 2520 | return 0; |
cbd59c48 | 2521 | err: |
2642fca6 MWO |
2522 | if (err < 0) |
2523 | put_page(page); | |
2524 | if (likely(--pvec->nr)) | |
ff993ba1 | 2525 | return 0; |
4612aeef | 2526 | if (err == AOP_TRUNCATED_PAGE) |
2642fca6 MWO |
2527 | goto retry; |
2528 | return err; | |
06c04442 KO |
2529 | } |
2530 | ||
485bb99b | 2531 | /** |
87fa0f3e CH |
2532 | * filemap_read - Read data from the page cache. |
2533 | * @iocb: The iocb to read. | |
2534 | * @iter: Destination for the data. | |
2535 | * @already_read: Number of bytes already read by the caller. | |
485bb99b | 2536 | * |
87fa0f3e CH |
2537 | * Copies data from the page cache. If the data is not currently present, |
2538 | * uses the readahead and readpage address_space operations to fetch it. | |
1da177e4 | 2539 | * |
87fa0f3e CH |
2540 | * Return: Total number of bytes copied, including those already read by |
2541 | * the caller. If an error happens before any bytes are copied, returns | |
2542 | * a negative error number. | |
1da177e4 | 2543 | */ |
87fa0f3e CH |
2544 | ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter, |
2545 | ssize_t already_read) | |
1da177e4 | 2546 | { |
47c27bc4 | 2547 | struct file *filp = iocb->ki_filp; |
06c04442 | 2548 | struct file_ra_state *ra = &filp->f_ra; |
36e78914 | 2549 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2550 | struct inode *inode = mapping->host; |
ff993ba1 MWO |
2551 | struct pagevec pvec; |
2552 | int i, error = 0; | |
06c04442 KO |
2553 | bool writably_mapped; |
2554 | loff_t isize, end_offset; | |
1da177e4 | 2555 | |
723ef24b | 2556 | if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2557 | return 0; |
3644e2d2 KO |
2558 | if (unlikely(!iov_iter_count(iter))) |
2559 | return 0; | |
2560 | ||
c2a9737f | 2561 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
cbd59c48 | 2562 | pagevec_init(&pvec); |
c2a9737f | 2563 | |
06c04442 | 2564 | do { |
1da177e4 | 2565 | cond_resched(); |
5abf186a | 2566 | |
723ef24b | 2567 | /* |
06c04442 KO |
2568 | * If we've already successfully copied some data, then we |
2569 | * can no longer safely return -EIOCBQUEUED. Hence mark | |
2570 | * an async read NOWAIT at that point. | |
723ef24b | 2571 | */ |
87fa0f3e | 2572 | if ((iocb->ki_flags & IOCB_WAITQ) && already_read) |
723ef24b KO |
2573 | iocb->ki_flags |= IOCB_NOWAIT; |
2574 | ||
ff993ba1 MWO |
2575 | error = filemap_get_pages(iocb, iter, &pvec); |
2576 | if (error < 0) | |
06c04442 | 2577 | break; |
1da177e4 | 2578 | |
06c04442 KO |
2579 | /* |
2580 | * i_size must be checked after we know the pages are Uptodate. | |
2581 | * | |
2582 | * Checking i_size after the check allows us to calculate | |
2583 | * the correct value for "nr", which means the zero-filled | |
2584 | * part of the page is not copied back to userspace (unless | |
2585 | * another truncate extends the file - this is desired though). | |
2586 | */ | |
2587 | isize = i_size_read(inode); | |
2588 | if (unlikely(iocb->ki_pos >= isize)) | |
2589 | goto put_pages; | |
06c04442 KO |
2590 | end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count); |
2591 | ||
06c04442 KO |
2592 | /* |
2593 | * Once we start copying data, we don't want to be touching any | |
2594 | * cachelines that might be contended: | |
2595 | */ | |
2596 | writably_mapped = mapping_writably_mapped(mapping); | |
2597 | ||
2598 | /* | |
2599 | * When a sequential read accesses a page several times, only | |
2600 | * mark it as accessed the first time. | |
2601 | */ | |
2602 | if (iocb->ki_pos >> PAGE_SHIFT != | |
2603 | ra->prev_pos >> PAGE_SHIFT) | |
ff993ba1 | 2604 | mark_page_accessed(pvec.pages[0]); |
06c04442 | 2605 | |
ff993ba1 | 2606 | for (i = 0; i < pagevec_count(&pvec); i++) { |
cbd59c48 MWO |
2607 | struct page *page = pvec.pages[i]; |
2608 | size_t page_size = thp_size(page); | |
2609 | size_t offset = iocb->ki_pos & (page_size - 1); | |
2610 | size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos, | |
2611 | page_size - offset); | |
2612 | size_t copied; | |
06c04442 | 2613 | |
cbd59c48 MWO |
2614 | if (end_offset < page_offset(page)) |
2615 | break; | |
2616 | if (i > 0) | |
2617 | mark_page_accessed(page); | |
06c04442 KO |
2618 | /* |
2619 | * If users can be writing to this page using arbitrary | |
2620 | * virtual addresses, take care about potential aliasing | |
2621 | * before reading the page on the kernel side. | |
2622 | */ | |
cbd59c48 MWO |
2623 | if (writably_mapped) { |
2624 | int j; | |
2625 | ||
2626 | for (j = 0; j < thp_nr_pages(page); j++) | |
2627 | flush_dcache_page(page + j); | |
2628 | } | |
06c04442 | 2629 | |
cbd59c48 | 2630 | copied = copy_page_to_iter(page, offset, bytes, iter); |
06c04442 | 2631 | |
87fa0f3e | 2632 | already_read += copied; |
06c04442 KO |
2633 | iocb->ki_pos += copied; |
2634 | ra->prev_pos = iocb->ki_pos; | |
2635 | ||
2636 | if (copied < bytes) { | |
2637 | error = -EFAULT; | |
2638 | break; | |
2639 | } | |
1da177e4 | 2640 | } |
06c04442 | 2641 | put_pages: |
ff993ba1 MWO |
2642 | for (i = 0; i < pagevec_count(&pvec); i++) |
2643 | put_page(pvec.pages[i]); | |
cbd59c48 | 2644 | pagevec_reinit(&pvec); |
06c04442 | 2645 | } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error); |
1da177e4 | 2646 | |
0c6aa263 | 2647 | file_accessed(filp); |
06c04442 | 2648 | |
87fa0f3e | 2649 | return already_read ? already_read : error; |
1da177e4 | 2650 | } |
87fa0f3e | 2651 | EXPORT_SYMBOL_GPL(filemap_read); |
1da177e4 | 2652 | |
485bb99b | 2653 | /** |
6abd2322 | 2654 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2655 | * @iocb: kernel I/O control block |
6abd2322 | 2656 | * @iter: destination for the data read |
485bb99b | 2657 | * |
6abd2322 | 2658 | * This is the "read_iter()" routine for all filesystems |
1da177e4 | 2659 | * that can use the page cache directly. |
41da51bc AG |
2660 | * |
2661 | * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall | |
2662 | * be returned when no data can be read without waiting for I/O requests | |
2663 | * to complete; it doesn't prevent readahead. | |
2664 | * | |
2665 | * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O | |
2666 | * requests shall be made for the read or for readahead. When no data | |
2667 | * can be read, -EAGAIN shall be returned. When readahead would be | |
2668 | * triggered, a partial, possibly empty read shall be returned. | |
2669 | * | |
a862f68a MR |
2670 | * Return: |
2671 | * * number of bytes copied, even for partial reads | |
41da51bc | 2672 | * * negative error code (or 0 if IOCB_NOIO) if nothing was read |
1da177e4 LT |
2673 | */ |
2674 | ssize_t | |
ed978a81 | 2675 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2676 | { |
e7080a43 | 2677 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2678 | ssize_t retval = 0; |
e7080a43 NS |
2679 | |
2680 | if (!count) | |
826ea860 | 2681 | return 0; /* skip atime */ |
1da177e4 | 2682 | |
2ba48ce5 | 2683 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2684 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2685 | struct address_space *mapping = file->f_mapping; |
2686 | struct inode *inode = mapping->host; | |
543ade1f | 2687 | loff_t size; |
1da177e4 | 2688 | |
1da177e4 | 2689 | size = i_size_read(inode); |
6be96d3a | 2690 | if (iocb->ki_flags & IOCB_NOWAIT) { |
7a60d6d7 JA |
2691 | if (filemap_range_needs_writeback(mapping, iocb->ki_pos, |
2692 | iocb->ki_pos + count - 1)) | |
6be96d3a GR |
2693 | return -EAGAIN; |
2694 | } else { | |
2695 | retval = filemap_write_and_wait_range(mapping, | |
2696 | iocb->ki_pos, | |
2697 | iocb->ki_pos + count - 1); | |
2698 | if (retval < 0) | |
826ea860 | 2699 | return retval; |
6be96d3a | 2700 | } |
d8d3d94b | 2701 | |
0d5b0cf2 CH |
2702 | file_accessed(file); |
2703 | ||
5ecda137 | 2704 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2705 | if (retval >= 0) { |
c64fb5c7 | 2706 | iocb->ki_pos += retval; |
5ecda137 | 2707 | count -= retval; |
9fe55eea | 2708 | } |
ab2125df PB |
2709 | if (retval != -EIOCBQUEUED) |
2710 | iov_iter_revert(iter, count - iov_iter_count(iter)); | |
66f998f6 | 2711 | |
9fe55eea SW |
2712 | /* |
2713 | * Btrfs can have a short DIO read if we encounter | |
2714 | * compressed extents, so if there was an error, or if | |
2715 | * we've already read everything we wanted to, or if | |
2716 | * there was a short read because we hit EOF, go ahead | |
2717 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2718 | * the rest of the read. Buffered reads will not work for |
2719 | * DAX files, so don't bother trying. | |
9fe55eea | 2720 | */ |
5ecda137 | 2721 | if (retval < 0 || !count || iocb->ki_pos >= size || |
0d5b0cf2 | 2722 | IS_DAX(inode)) |
826ea860 | 2723 | return retval; |
1da177e4 LT |
2724 | } |
2725 | ||
826ea860 | 2726 | return filemap_read(iocb, iter, retval); |
1da177e4 | 2727 | } |
ed978a81 | 2728 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2729 | |
54fa39ac MWO |
2730 | static inline loff_t page_seek_hole_data(struct xa_state *xas, |
2731 | struct address_space *mapping, struct page *page, | |
2732 | loff_t start, loff_t end, bool seek_data) | |
41139aa4 | 2733 | { |
54fa39ac MWO |
2734 | const struct address_space_operations *ops = mapping->a_ops; |
2735 | size_t offset, bsz = i_blocksize(mapping->host); | |
2736 | ||
41139aa4 | 2737 | if (xa_is_value(page) || PageUptodate(page)) |
54fa39ac MWO |
2738 | return seek_data ? start : end; |
2739 | if (!ops->is_partially_uptodate) | |
2740 | return seek_data ? end : start; | |
2741 | ||
2742 | xas_pause(xas); | |
2743 | rcu_read_unlock(); | |
2744 | lock_page(page); | |
2745 | if (unlikely(page->mapping != mapping)) | |
2746 | goto unlock; | |
2747 | ||
2748 | offset = offset_in_thp(page, start) & ~(bsz - 1); | |
2749 | ||
2750 | do { | |
2751 | if (ops->is_partially_uptodate(page, offset, bsz) == seek_data) | |
2752 | break; | |
2753 | start = (start + bsz) & ~(bsz - 1); | |
2754 | offset += bsz; | |
2755 | } while (offset < thp_size(page)); | |
2756 | unlock: | |
2757 | unlock_page(page); | |
2758 | rcu_read_lock(); | |
2759 | return start; | |
41139aa4 MWO |
2760 | } |
2761 | ||
2762 | static inline | |
2763 | unsigned int seek_page_size(struct xa_state *xas, struct page *page) | |
2764 | { | |
2765 | if (xa_is_value(page)) | |
2766 | return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index); | |
2767 | return thp_size(page); | |
2768 | } | |
2769 | ||
2770 | /** | |
2771 | * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache. | |
2772 | * @mapping: Address space to search. | |
2773 | * @start: First byte to consider. | |
2774 | * @end: Limit of search (exclusive). | |
2775 | * @whence: Either SEEK_HOLE or SEEK_DATA. | |
2776 | * | |
2777 | * If the page cache knows which blocks contain holes and which blocks | |
2778 | * contain data, your filesystem can use this function to implement | |
2779 | * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are | |
2780 | * entirely memory-based such as tmpfs, and filesystems which support | |
2781 | * unwritten extents. | |
2782 | * | |
f0953a1b | 2783 | * Return: The requested offset on success, or -ENXIO if @whence specifies |
41139aa4 MWO |
2784 | * SEEK_DATA and there is no data after @start. There is an implicit hole |
2785 | * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start | |
2786 | * and @end contain data. | |
2787 | */ | |
2788 | loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start, | |
2789 | loff_t end, int whence) | |
2790 | { | |
2791 | XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT); | |
ed98b015 | 2792 | pgoff_t max = (end - 1) >> PAGE_SHIFT; |
41139aa4 MWO |
2793 | bool seek_data = (whence == SEEK_DATA); |
2794 | struct page *page; | |
2795 | ||
2796 | if (end <= start) | |
2797 | return -ENXIO; | |
2798 | ||
2799 | rcu_read_lock(); | |
2800 | while ((page = find_get_entry(&xas, max, XA_PRESENT))) { | |
ed98b015 HD |
2801 | loff_t pos = (u64)xas.xa_index << PAGE_SHIFT; |
2802 | unsigned int seek_size; | |
41139aa4 MWO |
2803 | |
2804 | if (start < pos) { | |
2805 | if (!seek_data) | |
2806 | goto unlock; | |
2807 | start = pos; | |
2808 | } | |
2809 | ||
ed98b015 HD |
2810 | seek_size = seek_page_size(&xas, page); |
2811 | pos = round_up(pos + 1, seek_size); | |
54fa39ac MWO |
2812 | start = page_seek_hole_data(&xas, mapping, page, start, pos, |
2813 | seek_data); | |
2814 | if (start < pos) | |
41139aa4 | 2815 | goto unlock; |
ed98b015 HD |
2816 | if (start >= end) |
2817 | break; | |
2818 | if (seek_size > PAGE_SIZE) | |
2819 | xas_set(&xas, pos >> PAGE_SHIFT); | |
41139aa4 MWO |
2820 | if (!xa_is_value(page)) |
2821 | put_page(page); | |
2822 | } | |
41139aa4 | 2823 | if (seek_data) |
ed98b015 | 2824 | start = -ENXIO; |
41139aa4 MWO |
2825 | unlock: |
2826 | rcu_read_unlock(); | |
ed98b015 | 2827 | if (page && !xa_is_value(page)) |
41139aa4 | 2828 | put_page(page); |
41139aa4 MWO |
2829 | if (start > end) |
2830 | return end; | |
2831 | return start; | |
2832 | } | |
2833 | ||
1da177e4 | 2834 | #ifdef CONFIG_MMU |
1da177e4 | 2835 | #define MMAP_LOTSAMISS (100) |
6b4c9f44 | 2836 | /* |
c1e8d7c6 | 2837 | * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock |
6b4c9f44 JB |
2838 | * @vmf - the vm_fault for this fault. |
2839 | * @page - the page to lock. | |
2840 | * @fpin - the pointer to the file we may pin (or is already pinned). | |
2841 | * | |
c1e8d7c6 | 2842 | * This works similar to lock_page_or_retry in that it can drop the mmap_lock. |
6b4c9f44 | 2843 | * It differs in that it actually returns the page locked if it returns 1 and 0 |
c1e8d7c6 | 2844 | * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin |
6b4c9f44 JB |
2845 | * will point to the pinned file and needs to be fput()'ed at a later point. |
2846 | */ | |
2847 | static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page, | |
2848 | struct file **fpin) | |
2849 | { | |
2850 | if (trylock_page(page)) | |
2851 | return 1; | |
2852 | ||
8b0f9fa2 LT |
2853 | /* |
2854 | * NOTE! This will make us return with VM_FAULT_RETRY, but with | |
c1e8d7c6 | 2855 | * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT |
8b0f9fa2 LT |
2856 | * is supposed to work. We have way too many special cases.. |
2857 | */ | |
6b4c9f44 JB |
2858 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
2859 | return 0; | |
2860 | ||
2861 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); | |
2862 | if (vmf->flags & FAULT_FLAG_KILLABLE) { | |
2863 | if (__lock_page_killable(page)) { | |
2864 | /* | |
c1e8d7c6 | 2865 | * We didn't have the right flags to drop the mmap_lock, |
6b4c9f44 JB |
2866 | * but all fault_handlers only check for fatal signals |
2867 | * if we return VM_FAULT_RETRY, so we need to drop the | |
c1e8d7c6 | 2868 | * mmap_lock here and return 0 if we don't have a fpin. |
6b4c9f44 JB |
2869 | */ |
2870 | if (*fpin == NULL) | |
d8ed45c5 | 2871 | mmap_read_unlock(vmf->vma->vm_mm); |
6b4c9f44 JB |
2872 | return 0; |
2873 | } | |
2874 | } else | |
2875 | __lock_page(page); | |
2876 | return 1; | |
2877 | } | |
2878 | ||
1da177e4 | 2879 | |
ef00e08e | 2880 | /* |
6b4c9f44 JB |
2881 | * Synchronous readahead happens when we don't even find a page in the page |
2882 | * cache at all. We don't want to perform IO under the mmap sem, so if we have | |
2883 | * to drop the mmap sem we return the file that was pinned in order for us to do | |
2884 | * that. If we didn't pin a file then we return NULL. The file that is | |
2885 | * returned needs to be fput()'ed when we're done with it. | |
ef00e08e | 2886 | */ |
6b4c9f44 | 2887 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
ef00e08e | 2888 | { |
2a1180f1 JB |
2889 | struct file *file = vmf->vma->vm_file; |
2890 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2891 | struct address_space *mapping = file->f_mapping; |
fcd9ae4f | 2892 | DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff); |
6b4c9f44 | 2893 | struct file *fpin = NULL; |
e630bfac | 2894 | unsigned int mmap_miss; |
ef00e08e LT |
2895 | |
2896 | /* If we don't want any read-ahead, don't bother */ | |
2a1180f1 | 2897 | if (vmf->vma->vm_flags & VM_RAND_READ) |
6b4c9f44 | 2898 | return fpin; |
275b12bf | 2899 | if (!ra->ra_pages) |
6b4c9f44 | 2900 | return fpin; |
ef00e08e | 2901 | |
2a1180f1 | 2902 | if (vmf->vma->vm_flags & VM_SEQ_READ) { |
6b4c9f44 | 2903 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
fcd9ae4f | 2904 | page_cache_sync_ra(&ractl, ra->ra_pages); |
6b4c9f44 | 2905 | return fpin; |
ef00e08e LT |
2906 | } |
2907 | ||
207d04ba | 2908 | /* Avoid banging the cache line if not needed */ |
e630bfac KS |
2909 | mmap_miss = READ_ONCE(ra->mmap_miss); |
2910 | if (mmap_miss < MMAP_LOTSAMISS * 10) | |
2911 | WRITE_ONCE(ra->mmap_miss, ++mmap_miss); | |
ef00e08e LT |
2912 | |
2913 | /* | |
2914 | * Do we miss much more than hit in this file? If so, | |
2915 | * stop bothering with read-ahead. It will only hurt. | |
2916 | */ | |
e630bfac | 2917 | if (mmap_miss > MMAP_LOTSAMISS) |
6b4c9f44 | 2918 | return fpin; |
ef00e08e | 2919 | |
d30a1100 WF |
2920 | /* |
2921 | * mmap read-around | |
2922 | */ | |
6b4c9f44 | 2923 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
db660d46 | 2924 | ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2); |
600e19af RG |
2925 | ra->size = ra->ra_pages; |
2926 | ra->async_size = ra->ra_pages / 4; | |
db660d46 DH |
2927 | ractl._index = ra->start; |
2928 | do_page_cache_ra(&ractl, ra->size, ra->async_size); | |
6b4c9f44 | 2929 | return fpin; |
ef00e08e LT |
2930 | } |
2931 | ||
2932 | /* | |
2933 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
6b4c9f44 | 2934 | * so we want to possibly extend the readahead further. We return the file that |
c1e8d7c6 | 2935 | * was pinned if we have to drop the mmap_lock in order to do IO. |
ef00e08e | 2936 | */ |
6b4c9f44 JB |
2937 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
2938 | struct page *page) | |
ef00e08e | 2939 | { |
2a1180f1 JB |
2940 | struct file *file = vmf->vma->vm_file; |
2941 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2942 | struct address_space *mapping = file->f_mapping; |
6b4c9f44 | 2943 | struct file *fpin = NULL; |
e630bfac | 2944 | unsigned int mmap_miss; |
2a1180f1 | 2945 | pgoff_t offset = vmf->pgoff; |
ef00e08e LT |
2946 | |
2947 | /* If we don't want any read-ahead, don't bother */ | |
5c72feee | 2948 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
6b4c9f44 | 2949 | return fpin; |
e630bfac KS |
2950 | mmap_miss = READ_ONCE(ra->mmap_miss); |
2951 | if (mmap_miss) | |
2952 | WRITE_ONCE(ra->mmap_miss, --mmap_miss); | |
6b4c9f44 JB |
2953 | if (PageReadahead(page)) { |
2954 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); | |
2fad6f5d WF |
2955 | page_cache_async_readahead(mapping, ra, file, |
2956 | page, offset, ra->ra_pages); | |
6b4c9f44 JB |
2957 | } |
2958 | return fpin; | |
ef00e08e LT |
2959 | } |
2960 | ||
485bb99b | 2961 | /** |
54cb8821 | 2962 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 2963 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 2964 | * |
54cb8821 | 2965 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2966 | * mapped memory region to read in file data during a page fault. |
2967 | * | |
2968 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2969 | * it in the page cache, and handles the special cases reasonably without | |
2970 | * having a lot of duplicated code. | |
9a95f3cf | 2971 | * |
c1e8d7c6 | 2972 | * vma->vm_mm->mmap_lock must be held on entry. |
9a95f3cf | 2973 | * |
c1e8d7c6 | 2974 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock |
a4985833 | 2975 | * may be dropped before doing I/O or by lock_page_maybe_drop_mmap(). |
9a95f3cf | 2976 | * |
c1e8d7c6 | 2977 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock |
9a95f3cf PC |
2978 | * has not been released. |
2979 | * | |
2980 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
a862f68a MR |
2981 | * |
2982 | * Return: bitwise-OR of %VM_FAULT_ codes. | |
1da177e4 | 2983 | */ |
2bcd6454 | 2984 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
2985 | { |
2986 | int error; | |
11bac800 | 2987 | struct file *file = vmf->vma->vm_file; |
6b4c9f44 | 2988 | struct file *fpin = NULL; |
1da177e4 | 2989 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 2990 | struct inode *inode = mapping->host; |
ef00e08e | 2991 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 2992 | pgoff_t max_off; |
1da177e4 | 2993 | struct page *page; |
2bcd6454 | 2994 | vm_fault_t ret = 0; |
730633f0 | 2995 | bool mapping_locked = false; |
1da177e4 | 2996 | |
9ab2594f MW |
2997 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2998 | if (unlikely(offset >= max_off)) | |
5307cc1a | 2999 | return VM_FAULT_SIGBUS; |
1da177e4 | 3000 | |
1da177e4 | 3001 | /* |
49426420 | 3002 | * Do we have something in the page cache already? |
1da177e4 | 3003 | */ |
ef00e08e | 3004 | page = find_get_page(mapping, offset); |
730633f0 | 3005 | if (likely(page)) { |
1da177e4 | 3006 | /* |
730633f0 JK |
3007 | * We found the page, so try async readahead before waiting for |
3008 | * the lock. | |
1da177e4 | 3009 | */ |
730633f0 JK |
3010 | if (!(vmf->flags & FAULT_FLAG_TRIED)) |
3011 | fpin = do_async_mmap_readahead(vmf, page); | |
3012 | if (unlikely(!PageUptodate(page))) { | |
3013 | filemap_invalidate_lock_shared(mapping); | |
3014 | mapping_locked = true; | |
3015 | } | |
3016 | } else { | |
ef00e08e | 3017 | /* No page in the page cache at all */ |
ef00e08e | 3018 | count_vm_event(PGMAJFAULT); |
2262185c | 3019 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e | 3020 | ret = VM_FAULT_MAJOR; |
6b4c9f44 | 3021 | fpin = do_sync_mmap_readahead(vmf); |
ef00e08e | 3022 | retry_find: |
730633f0 JK |
3023 | /* |
3024 | * See comment in filemap_create_page() why we need | |
3025 | * invalidate_lock | |
3026 | */ | |
3027 | if (!mapping_locked) { | |
3028 | filemap_invalidate_lock_shared(mapping); | |
3029 | mapping_locked = true; | |
3030 | } | |
a75d4c33 JB |
3031 | page = pagecache_get_page(mapping, offset, |
3032 | FGP_CREAT|FGP_FOR_MMAP, | |
3033 | vmf->gfp_mask); | |
6b4c9f44 JB |
3034 | if (!page) { |
3035 | if (fpin) | |
3036 | goto out_retry; | |
730633f0 | 3037 | filemap_invalidate_unlock_shared(mapping); |
e520e932 | 3038 | return VM_FAULT_OOM; |
6b4c9f44 | 3039 | } |
1da177e4 LT |
3040 | } |
3041 | ||
6b4c9f44 JB |
3042 | if (!lock_page_maybe_drop_mmap(vmf, page, &fpin)) |
3043 | goto out_retry; | |
b522c94d ML |
3044 | |
3045 | /* Did it get truncated? */ | |
585e5a7b | 3046 | if (unlikely(compound_head(page)->mapping != mapping)) { |
b522c94d ML |
3047 | unlock_page(page); |
3048 | put_page(page); | |
3049 | goto retry_find; | |
3050 | } | |
520e5ba4 | 3051 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
b522c94d | 3052 | |
1da177e4 | 3053 | /* |
d00806b1 NP |
3054 | * We have a locked page in the page cache, now we need to check |
3055 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 3056 | */ |
730633f0 JK |
3057 | if (unlikely(!PageUptodate(page))) { |
3058 | /* | |
3059 | * The page was in cache and uptodate and now it is not. | |
3060 | * Strange but possible since we didn't hold the page lock all | |
3061 | * the time. Let's drop everything get the invalidate lock and | |
3062 | * try again. | |
3063 | */ | |
3064 | if (!mapping_locked) { | |
3065 | unlock_page(page); | |
3066 | put_page(page); | |
3067 | goto retry_find; | |
3068 | } | |
1da177e4 | 3069 | goto page_not_uptodate; |
730633f0 | 3070 | } |
1da177e4 | 3071 | |
6b4c9f44 | 3072 | /* |
c1e8d7c6 | 3073 | * We've made it this far and we had to drop our mmap_lock, now is the |
6b4c9f44 JB |
3074 | * time to return to the upper layer and have it re-find the vma and |
3075 | * redo the fault. | |
3076 | */ | |
3077 | if (fpin) { | |
3078 | unlock_page(page); | |
3079 | goto out_retry; | |
3080 | } | |
730633f0 JK |
3081 | if (mapping_locked) |
3082 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 | 3083 | |
ef00e08e LT |
3084 | /* |
3085 | * Found the page and have a reference on it. | |
3086 | * We must recheck i_size under page lock. | |
3087 | */ | |
9ab2594f MW |
3088 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3089 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 3090 | unlock_page(page); |
09cbfeaf | 3091 | put_page(page); |
5307cc1a | 3092 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
3093 | } |
3094 | ||
d0217ac0 | 3095 | vmf->page = page; |
83c54070 | 3096 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 3097 | |
1da177e4 | 3098 | page_not_uptodate: |
1da177e4 LT |
3099 | /* |
3100 | * Umm, take care of errors if the page isn't up-to-date. | |
3101 | * Try to re-read it _once_. We do this synchronously, | |
3102 | * because there really aren't any performance issues here | |
3103 | * and we need to check for errors. | |
3104 | */ | |
6b4c9f44 | 3105 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
d31fa86a | 3106 | error = filemap_read_page(file, mapping, page); |
6b4c9f44 JB |
3107 | if (fpin) |
3108 | goto out_retry; | |
09cbfeaf | 3109 | put_page(page); |
d00806b1 NP |
3110 | |
3111 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 3112 | goto retry_find; |
730633f0 | 3113 | filemap_invalidate_unlock_shared(mapping); |
1da177e4 | 3114 | |
d0217ac0 | 3115 | return VM_FAULT_SIGBUS; |
6b4c9f44 JB |
3116 | |
3117 | out_retry: | |
3118 | /* | |
c1e8d7c6 | 3119 | * We dropped the mmap_lock, we need to return to the fault handler to |
6b4c9f44 JB |
3120 | * re-find the vma and come back and find our hopefully still populated |
3121 | * page. | |
3122 | */ | |
3123 | if (page) | |
3124 | put_page(page); | |
730633f0 JK |
3125 | if (mapping_locked) |
3126 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 JB |
3127 | if (fpin) |
3128 | fput(fpin); | |
3129 | return ret | VM_FAULT_RETRY; | |
54cb8821 NP |
3130 | } |
3131 | EXPORT_SYMBOL(filemap_fault); | |
3132 | ||
f9ce0be7 | 3133 | static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page) |
f1820361 | 3134 | { |
f9ce0be7 KS |
3135 | struct mm_struct *mm = vmf->vma->vm_mm; |
3136 | ||
3137 | /* Huge page is mapped? No need to proceed. */ | |
3138 | if (pmd_trans_huge(*vmf->pmd)) { | |
3139 | unlock_page(page); | |
3140 | put_page(page); | |
3141 | return true; | |
3142 | } | |
3143 | ||
3144 | if (pmd_none(*vmf->pmd) && PageTransHuge(page)) { | |
3145 | vm_fault_t ret = do_set_pmd(vmf, page); | |
3146 | if (!ret) { | |
3147 | /* The page is mapped successfully, reference consumed. */ | |
3148 | unlock_page(page); | |
3149 | return true; | |
3150 | } | |
3151 | } | |
3152 | ||
3153 | if (pmd_none(*vmf->pmd)) { | |
3154 | vmf->ptl = pmd_lock(mm, vmf->pmd); | |
3155 | if (likely(pmd_none(*vmf->pmd))) { | |
3156 | mm_inc_nr_ptes(mm); | |
3157 | pmd_populate(mm, vmf->pmd, vmf->prealloc_pte); | |
3158 | vmf->prealloc_pte = NULL; | |
3159 | } | |
3160 | spin_unlock(vmf->ptl); | |
3161 | } | |
3162 | ||
3163 | /* See comment in handle_pte_fault() */ | |
3164 | if (pmd_devmap_trans_unstable(vmf->pmd)) { | |
3165 | unlock_page(page); | |
3166 | put_page(page); | |
3167 | return true; | |
3168 | } | |
3169 | ||
3170 | return false; | |
3171 | } | |
3172 | ||
3173 | static struct page *next_uptodate_page(struct page *page, | |
3174 | struct address_space *mapping, | |
3175 | struct xa_state *xas, pgoff_t end_pgoff) | |
3176 | { | |
3177 | unsigned long max_idx; | |
3178 | ||
3179 | do { | |
3180 | if (!page) | |
3181 | return NULL; | |
3182 | if (xas_retry(xas, page)) | |
3183 | continue; | |
3184 | if (xa_is_value(page)) | |
3185 | continue; | |
3186 | if (PageLocked(page)) | |
3187 | continue; | |
3188 | if (!page_cache_get_speculative(page)) | |
3189 | continue; | |
3190 | /* Has the page moved or been split? */ | |
3191 | if (unlikely(page != xas_reload(xas))) | |
3192 | goto skip; | |
3193 | if (!PageUptodate(page) || PageReadahead(page)) | |
3194 | goto skip; | |
3195 | if (PageHWPoison(page)) | |
3196 | goto skip; | |
3197 | if (!trylock_page(page)) | |
3198 | goto skip; | |
3199 | if (page->mapping != mapping) | |
3200 | goto unlock; | |
3201 | if (!PageUptodate(page)) | |
3202 | goto unlock; | |
3203 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); | |
3204 | if (xas->xa_index >= max_idx) | |
3205 | goto unlock; | |
3206 | return page; | |
3207 | unlock: | |
3208 | unlock_page(page); | |
3209 | skip: | |
3210 | put_page(page); | |
3211 | } while ((page = xas_next_entry(xas, end_pgoff)) != NULL); | |
3212 | ||
3213 | return NULL; | |
3214 | } | |
3215 | ||
3216 | static inline struct page *first_map_page(struct address_space *mapping, | |
3217 | struct xa_state *xas, | |
3218 | pgoff_t end_pgoff) | |
3219 | { | |
3220 | return next_uptodate_page(xas_find(xas, end_pgoff), | |
3221 | mapping, xas, end_pgoff); | |
3222 | } | |
3223 | ||
3224 | static inline struct page *next_map_page(struct address_space *mapping, | |
3225 | struct xa_state *xas, | |
3226 | pgoff_t end_pgoff) | |
3227 | { | |
3228 | return next_uptodate_page(xas_next_entry(xas, end_pgoff), | |
3229 | mapping, xas, end_pgoff); | |
3230 | } | |
3231 | ||
3232 | vm_fault_t filemap_map_pages(struct vm_fault *vmf, | |
3233 | pgoff_t start_pgoff, pgoff_t end_pgoff) | |
3234 | { | |
3235 | struct vm_area_struct *vma = vmf->vma; | |
3236 | struct file *file = vma->vm_file; | |
f1820361 | 3237 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 3238 | pgoff_t last_pgoff = start_pgoff; |
9d3af4b4 | 3239 | unsigned long addr; |
070e807c | 3240 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
27a83a60 | 3241 | struct page *head, *page; |
e630bfac | 3242 | unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss); |
f9ce0be7 | 3243 | vm_fault_t ret = 0; |
f1820361 KS |
3244 | |
3245 | rcu_read_lock(); | |
f9ce0be7 KS |
3246 | head = first_map_page(mapping, &xas, end_pgoff); |
3247 | if (!head) | |
3248 | goto out; | |
f1820361 | 3249 | |
f9ce0be7 KS |
3250 | if (filemap_map_pmd(vmf, head)) { |
3251 | ret = VM_FAULT_NOPAGE; | |
3252 | goto out; | |
3253 | } | |
f1820361 | 3254 | |
9d3af4b4 WD |
3255 | addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
3256 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); | |
f9ce0be7 | 3257 | do { |
27a83a60 | 3258 | page = find_subpage(head, xas.xa_index); |
f9ce0be7 | 3259 | if (PageHWPoison(page)) |
f1820361 KS |
3260 | goto unlock; |
3261 | ||
e630bfac KS |
3262 | if (mmap_miss > 0) |
3263 | mmap_miss--; | |
7267ec00 | 3264 | |
9d3af4b4 | 3265 | addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
f9ce0be7 | 3266 | vmf->pte += xas.xa_index - last_pgoff; |
070e807c | 3267 | last_pgoff = xas.xa_index; |
f9ce0be7 KS |
3268 | |
3269 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3270 | goto unlock; |
f9ce0be7 | 3271 | |
46bdb427 | 3272 | /* We're about to handle the fault */ |
9d3af4b4 | 3273 | if (vmf->address == addr) |
46bdb427 | 3274 | ret = VM_FAULT_NOPAGE; |
46bdb427 | 3275 | |
9d3af4b4 | 3276 | do_set_pte(vmf, page, addr); |
f9ce0be7 | 3277 | /* no need to invalidate: a not-present page won't be cached */ |
9d3af4b4 | 3278 | update_mmu_cache(vma, addr, vmf->pte); |
27a83a60 | 3279 | unlock_page(head); |
f9ce0be7 | 3280 | continue; |
f1820361 | 3281 | unlock: |
27a83a60 | 3282 | unlock_page(head); |
27a83a60 | 3283 | put_page(head); |
f9ce0be7 KS |
3284 | } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL); |
3285 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3286 | out: | |
f1820361 | 3287 | rcu_read_unlock(); |
e630bfac | 3288 | WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss); |
f9ce0be7 | 3289 | return ret; |
f1820361 KS |
3290 | } |
3291 | EXPORT_SYMBOL(filemap_map_pages); | |
3292 | ||
2bcd6454 | 3293 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 | 3294 | { |
5df1a672 | 3295 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
4fcf1c62 | 3296 | struct page *page = vmf->page; |
2bcd6454 | 3297 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 3298 | |
5df1a672 | 3299 | sb_start_pagefault(mapping->host->i_sb); |
11bac800 | 3300 | file_update_time(vmf->vma->vm_file); |
4fcf1c62 | 3301 | lock_page(page); |
5df1a672 | 3302 | if (page->mapping != mapping) { |
4fcf1c62 JK |
3303 | unlock_page(page); |
3304 | ret = VM_FAULT_NOPAGE; | |
3305 | goto out; | |
3306 | } | |
14da9200 JK |
3307 | /* |
3308 | * We mark the page dirty already here so that when freeze is in | |
3309 | * progress, we are guaranteed that writeback during freezing will | |
3310 | * see the dirty page and writeprotect it again. | |
3311 | */ | |
3312 | set_page_dirty(page); | |
1d1d1a76 | 3313 | wait_for_stable_page(page); |
4fcf1c62 | 3314 | out: |
5df1a672 | 3315 | sb_end_pagefault(mapping->host->i_sb); |
4fcf1c62 JK |
3316 | return ret; |
3317 | } | |
4fcf1c62 | 3318 | |
f0f37e2f | 3319 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 3320 | .fault = filemap_fault, |
f1820361 | 3321 | .map_pages = filemap_map_pages, |
4fcf1c62 | 3322 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
3323 | }; |
3324 | ||
3325 | /* This is used for a general mmap of a disk file */ | |
3326 | ||
68d68ff6 | 3327 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3328 | { |
3329 | struct address_space *mapping = file->f_mapping; | |
3330 | ||
3331 | if (!mapping->a_ops->readpage) | |
3332 | return -ENOEXEC; | |
3333 | file_accessed(file); | |
3334 | vma->vm_ops = &generic_file_vm_ops; | |
3335 | return 0; | |
3336 | } | |
1da177e4 LT |
3337 | |
3338 | /* | |
3339 | * This is for filesystems which do not implement ->writepage. | |
3340 | */ | |
3341 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
3342 | { | |
3343 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
3344 | return -EINVAL; | |
3345 | return generic_file_mmap(file, vma); | |
3346 | } | |
3347 | #else | |
4b96a37d | 3348 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
45397228 | 3349 | { |
4b96a37d | 3350 | return VM_FAULT_SIGBUS; |
45397228 | 3351 | } |
68d68ff6 | 3352 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3353 | { |
3354 | return -ENOSYS; | |
3355 | } | |
68d68ff6 | 3356 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3357 | { |
3358 | return -ENOSYS; | |
3359 | } | |
3360 | #endif /* CONFIG_MMU */ | |
3361 | ||
45397228 | 3362 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
3363 | EXPORT_SYMBOL(generic_file_mmap); |
3364 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
3365 | ||
67f9fd91 SL |
3366 | static struct page *wait_on_page_read(struct page *page) |
3367 | { | |
3368 | if (!IS_ERR(page)) { | |
3369 | wait_on_page_locked(page); | |
3370 | if (!PageUptodate(page)) { | |
09cbfeaf | 3371 | put_page(page); |
67f9fd91 SL |
3372 | page = ERR_PTR(-EIO); |
3373 | } | |
3374 | } | |
3375 | return page; | |
3376 | } | |
3377 | ||
32b63529 | 3378 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 3379 | pgoff_t index, |
5e5358e7 | 3380 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3381 | void *data, |
3382 | gfp_t gfp) | |
1da177e4 | 3383 | { |
eb2be189 | 3384 | struct page *page; |
1da177e4 LT |
3385 | int err; |
3386 | repeat: | |
3387 | page = find_get_page(mapping, index); | |
3388 | if (!page) { | |
453f85d4 | 3389 | page = __page_cache_alloc(gfp); |
eb2be189 NP |
3390 | if (!page) |
3391 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 3392 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 3393 | if (unlikely(err)) { |
09cbfeaf | 3394 | put_page(page); |
eb2be189 NP |
3395 | if (err == -EEXIST) |
3396 | goto repeat; | |
22ecdb4f | 3397 | /* Presumably ENOMEM for xarray node */ |
1da177e4 LT |
3398 | return ERR_PTR(err); |
3399 | } | |
32b63529 MG |
3400 | |
3401 | filler: | |
6c45b454 CH |
3402 | if (filler) |
3403 | err = filler(data, page); | |
3404 | else | |
3405 | err = mapping->a_ops->readpage(data, page); | |
3406 | ||
1da177e4 | 3407 | if (err < 0) { |
09cbfeaf | 3408 | put_page(page); |
32b63529 | 3409 | return ERR_PTR(err); |
1da177e4 | 3410 | } |
1da177e4 | 3411 | |
32b63529 MG |
3412 | page = wait_on_page_read(page); |
3413 | if (IS_ERR(page)) | |
3414 | return page; | |
3415 | goto out; | |
3416 | } | |
1da177e4 LT |
3417 | if (PageUptodate(page)) |
3418 | goto out; | |
3419 | ||
ebded027 | 3420 | /* |
0e9aa675 | 3421 | * Page is not up to date and may be locked due to one of the following |
ebded027 MG |
3422 | * case a: Page is being filled and the page lock is held |
3423 | * case b: Read/write error clearing the page uptodate status | |
3424 | * case c: Truncation in progress (page locked) | |
3425 | * case d: Reclaim in progress | |
3426 | * | |
3427 | * Case a, the page will be up to date when the page is unlocked. | |
3428 | * There is no need to serialise on the page lock here as the page | |
3429 | * is pinned so the lock gives no additional protection. Even if the | |
ce89fddf | 3430 | * page is truncated, the data is still valid if PageUptodate as |
ebded027 MG |
3431 | * it's a race vs truncate race. |
3432 | * Case b, the page will not be up to date | |
3433 | * Case c, the page may be truncated but in itself, the data may still | |
3434 | * be valid after IO completes as it's a read vs truncate race. The | |
3435 | * operation must restart if the page is not uptodate on unlock but | |
3436 | * otherwise serialising on page lock to stabilise the mapping gives | |
3437 | * no additional guarantees to the caller as the page lock is | |
3438 | * released before return. | |
3439 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
3440 | * will be a race with remove_mapping that determines if the mapping | |
3441 | * is valid on unlock but otherwise the data is valid and there is | |
3442 | * no need to serialise with page lock. | |
3443 | * | |
3444 | * As the page lock gives no additional guarantee, we optimistically | |
3445 | * wait on the page to be unlocked and check if it's up to date and | |
3446 | * use the page if it is. Otherwise, the page lock is required to | |
3447 | * distinguish between the different cases. The motivation is that we | |
3448 | * avoid spurious serialisations and wakeups when multiple processes | |
3449 | * wait on the same page for IO to complete. | |
3450 | */ | |
3451 | wait_on_page_locked(page); | |
3452 | if (PageUptodate(page)) | |
3453 | goto out; | |
3454 | ||
3455 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 3456 | lock_page(page); |
ebded027 MG |
3457 | |
3458 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
3459 | if (!page->mapping) { |
3460 | unlock_page(page); | |
09cbfeaf | 3461 | put_page(page); |
32b63529 | 3462 | goto repeat; |
1da177e4 | 3463 | } |
ebded027 MG |
3464 | |
3465 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
3466 | if (PageUptodate(page)) { |
3467 | unlock_page(page); | |
3468 | goto out; | |
3469 | } | |
faffdfa0 XT |
3470 | |
3471 | /* | |
3472 | * A previous I/O error may have been due to temporary | |
3473 | * failures. | |
3474 | * Clear page error before actual read, PG_error will be | |
3475 | * set again if read page fails. | |
3476 | */ | |
3477 | ClearPageError(page); | |
32b63529 MG |
3478 | goto filler; |
3479 | ||
c855ff37 | 3480 | out: |
6fe6900e NP |
3481 | mark_page_accessed(page); |
3482 | return page; | |
3483 | } | |
0531b2aa LT |
3484 | |
3485 | /** | |
67f9fd91 | 3486 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
3487 | * @mapping: the page's address_space |
3488 | * @index: the page index | |
3489 | * @filler: function to perform the read | |
5e5358e7 | 3490 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 3491 | * |
0531b2aa | 3492 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 3493 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
3494 | * |
3495 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3496 | * |
730633f0 JK |
3497 | * The function expects mapping->invalidate_lock to be already held. |
3498 | * | |
a862f68a | 3499 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa | 3500 | */ |
67f9fd91 | 3501 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 3502 | pgoff_t index, |
5e5358e7 | 3503 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3504 | void *data) |
3505 | { | |
d322a8e5 CH |
3506 | return do_read_cache_page(mapping, index, filler, data, |
3507 | mapping_gfp_mask(mapping)); | |
0531b2aa | 3508 | } |
67f9fd91 | 3509 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
3510 | |
3511 | /** | |
3512 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
3513 | * @mapping: the page's address_space | |
3514 | * @index: the page index | |
3515 | * @gfp: the page allocator flags to use if allocating | |
3516 | * | |
3517 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 3518 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
3519 | * |
3520 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3521 | * |
730633f0 JK |
3522 | * The function expects mapping->invalidate_lock to be already held. |
3523 | * | |
a862f68a | 3524 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa LT |
3525 | */ |
3526 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
3527 | pgoff_t index, | |
3528 | gfp_t gfp) | |
3529 | { | |
6c45b454 | 3530 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
0531b2aa LT |
3531 | } |
3532 | EXPORT_SYMBOL(read_cache_page_gfp); | |
3533 | ||
afddba49 NP |
3534 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
3535 | loff_t pos, unsigned len, unsigned flags, | |
3536 | struct page **pagep, void **fsdata) | |
3537 | { | |
3538 | const struct address_space_operations *aops = mapping->a_ops; | |
3539 | ||
4e02ed4b | 3540 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 3541 | pagep, fsdata); |
afddba49 NP |
3542 | } |
3543 | EXPORT_SYMBOL(pagecache_write_begin); | |
3544 | ||
3545 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
3546 | loff_t pos, unsigned len, unsigned copied, | |
3547 | struct page *page, void *fsdata) | |
3548 | { | |
3549 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 3550 | |
4e02ed4b | 3551 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
3552 | } |
3553 | EXPORT_SYMBOL(pagecache_write_end); | |
3554 | ||
a92853b6 KK |
3555 | /* |
3556 | * Warn about a page cache invalidation failure during a direct I/O write. | |
3557 | */ | |
3558 | void dio_warn_stale_pagecache(struct file *filp) | |
3559 | { | |
3560 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); | |
3561 | char pathname[128]; | |
a92853b6 KK |
3562 | char *path; |
3563 | ||
5df1a672 | 3564 | errseq_set(&filp->f_mapping->wb_err, -EIO); |
a92853b6 KK |
3565 | if (__ratelimit(&_rs)) { |
3566 | path = file_path(filp, pathname, sizeof(pathname)); | |
3567 | if (IS_ERR(path)) | |
3568 | path = "(unknown)"; | |
3569 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); | |
3570 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, | |
3571 | current->comm); | |
3572 | } | |
3573 | } | |
3574 | ||
1da177e4 | 3575 | ssize_t |
1af5bb49 | 3576 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3577 | { |
3578 | struct file *file = iocb->ki_filp; | |
3579 | struct address_space *mapping = file->f_mapping; | |
3580 | struct inode *inode = mapping->host; | |
1af5bb49 | 3581 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3582 | ssize_t written; |
a969e903 CH |
3583 | size_t write_len; |
3584 | pgoff_t end; | |
1da177e4 | 3585 | |
0c949334 | 3586 | write_len = iov_iter_count(from); |
09cbfeaf | 3587 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3588 | |
6be96d3a GR |
3589 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3590 | /* If there are pages to writeback, return */ | |
5df1a672 | 3591 | if (filemap_range_has_page(file->f_mapping, pos, |
35f12f0f | 3592 | pos + write_len - 1)) |
6be96d3a GR |
3593 | return -EAGAIN; |
3594 | } else { | |
3595 | written = filemap_write_and_wait_range(mapping, pos, | |
3596 | pos + write_len - 1); | |
3597 | if (written) | |
3598 | goto out; | |
3599 | } | |
a969e903 CH |
3600 | |
3601 | /* | |
3602 | * After a write we want buffered reads to be sure to go to disk to get | |
3603 | * the new data. We invalidate clean cached page from the region we're | |
3604 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3605 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3606 | */ |
55635ba7 | 3607 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3608 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3609 | /* |
3610 | * If a page can not be invalidated, return 0 to fall back | |
3611 | * to buffered write. | |
3612 | */ | |
3613 | if (written) { | |
3614 | if (written == -EBUSY) | |
3615 | return 0; | |
3616 | goto out; | |
a969e903 CH |
3617 | } |
3618 | ||
639a93a5 | 3619 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3620 | |
3621 | /* | |
3622 | * Finally, try again to invalidate clean pages which might have been | |
3623 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3624 | * if the source of the write was an mmap'ed region of the file | |
3625 | * we're writing. Either one is a pretty crazy thing to do, | |
3626 | * so we don't support it 100%. If this invalidation | |
3627 | * fails, tough, the write still worked... | |
332391a9 LC |
3628 | * |
3629 | * Most of the time we do not need this since dio_complete() will do | |
3630 | * the invalidation for us. However there are some file systems that | |
3631 | * do not end up with dio_complete() being called, so let's not break | |
80c1fe90 KK |
3632 | * them by removing it completely. |
3633 | * | |
9266a140 KK |
3634 | * Noticeable example is a blkdev_direct_IO(). |
3635 | * | |
80c1fe90 | 3636 | * Skip invalidation for async writes or if mapping has no pages. |
a969e903 | 3637 | */ |
9266a140 KK |
3638 | if (written > 0 && mapping->nrpages && |
3639 | invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) | |
3640 | dio_warn_stale_pagecache(file); | |
a969e903 | 3641 | |
1da177e4 | 3642 | if (written > 0) { |
0116651c | 3643 | pos += written; |
639a93a5 | 3644 | write_len -= written; |
0116651c NK |
3645 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3646 | i_size_write(inode, pos); | |
1da177e4 LT |
3647 | mark_inode_dirty(inode); |
3648 | } | |
5cb6c6c7 | 3649 | iocb->ki_pos = pos; |
1da177e4 | 3650 | } |
ab2125df PB |
3651 | if (written != -EIOCBQUEUED) |
3652 | iov_iter_revert(from, write_len - iov_iter_count(from)); | |
a969e903 | 3653 | out: |
1da177e4 LT |
3654 | return written; |
3655 | } | |
3656 | EXPORT_SYMBOL(generic_file_direct_write); | |
3657 | ||
eb2be189 NP |
3658 | /* |
3659 | * Find or create a page at the given pagecache position. Return the locked | |
3660 | * page. This function is specifically for buffered writes. | |
3661 | */ | |
54566b2c NP |
3662 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
3663 | pgoff_t index, unsigned flags) | |
eb2be189 | 3664 | { |
eb2be189 | 3665 | struct page *page; |
bbddabe2 | 3666 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 3667 | |
54566b2c | 3668 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
3669 | fgp_flags |= FGP_NOFS; |
3670 | ||
3671 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 3672 | mapping_gfp_mask(mapping)); |
c585a267 | 3673 | if (page) |
2457aec6 | 3674 | wait_for_stable_page(page); |
eb2be189 | 3675 | |
eb2be189 NP |
3676 | return page; |
3677 | } | |
54566b2c | 3678 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 3679 | |
3b93f911 | 3680 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3681 | struct iov_iter *i, loff_t pos) |
3682 | { | |
3683 | struct address_space *mapping = file->f_mapping; | |
3684 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3685 | long status = 0; | |
3686 | ssize_t written = 0; | |
674b892e NP |
3687 | unsigned int flags = 0; |
3688 | ||
afddba49 NP |
3689 | do { |
3690 | struct page *page; | |
afddba49 NP |
3691 | unsigned long offset; /* Offset into pagecache page */ |
3692 | unsigned long bytes; /* Bytes to write to page */ | |
3693 | size_t copied; /* Bytes copied from user */ | |
3694 | void *fsdata; | |
3695 | ||
09cbfeaf KS |
3696 | offset = (pos & (PAGE_SIZE - 1)); |
3697 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3698 | iov_iter_count(i)); |
3699 | ||
3700 | again: | |
00a3d660 LT |
3701 | /* |
3702 | * Bring in the user page that we will copy from _first_. | |
3703 | * Otherwise there's a nasty deadlock on copying from the | |
3704 | * same page as we're writing to, without it being marked | |
3705 | * up-to-date. | |
00a3d660 LT |
3706 | */ |
3707 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
3708 | status = -EFAULT; | |
3709 | break; | |
3710 | } | |
3711 | ||
296291cd JK |
3712 | if (fatal_signal_pending(current)) { |
3713 | status = -EINTR; | |
3714 | break; | |
3715 | } | |
3716 | ||
674b892e | 3717 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3718 | &page, &fsdata); |
2457aec6 | 3719 | if (unlikely(status < 0)) |
afddba49 NP |
3720 | break; |
3721 | ||
931e80e4 | 3722 | if (mapping_writably_mapped(mapping)) |
3723 | flush_dcache_page(page); | |
00a3d660 | 3724 | |
f0b65f39 | 3725 | copied = copy_page_from_iter_atomic(page, offset, bytes, i); |
afddba49 NP |
3726 | flush_dcache_page(page); |
3727 | ||
3728 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3729 | page, fsdata); | |
f0b65f39 AV |
3730 | if (unlikely(status != copied)) { |
3731 | iov_iter_revert(i, copied - max(status, 0L)); | |
3732 | if (unlikely(status < 0)) | |
3733 | break; | |
3734 | } | |
afddba49 NP |
3735 | cond_resched(); |
3736 | ||
bc1bb416 | 3737 | if (unlikely(status == 0)) { |
afddba49 | 3738 | /* |
bc1bb416 AV |
3739 | * A short copy made ->write_end() reject the |
3740 | * thing entirely. Might be memory poisoning | |
3741 | * halfway through, might be a race with munmap, | |
3742 | * might be severe memory pressure. | |
afddba49 | 3743 | */ |
bc1bb416 AV |
3744 | if (copied) |
3745 | bytes = copied; | |
afddba49 NP |
3746 | goto again; |
3747 | } | |
f0b65f39 AV |
3748 | pos += status; |
3749 | written += status; | |
afddba49 NP |
3750 | |
3751 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3752 | } while (iov_iter_count(i)); |
3753 | ||
3754 | return written ? written : status; | |
3755 | } | |
3b93f911 | 3756 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3757 | |
e4dd9de3 | 3758 | /** |
8174202b | 3759 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3760 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3761 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3762 | * |
3763 | * This function does all the work needed for actually writing data to a | |
3764 | * file. It does all basic checks, removes SUID from the file, updates | |
3765 | * modification times and calls proper subroutines depending on whether we | |
3766 | * do direct IO or a standard buffered write. | |
3767 | * | |
9608703e | 3768 | * It expects i_rwsem to be grabbed unless we work on a block device or similar |
e4dd9de3 JK |
3769 | * object which does not need locking at all. |
3770 | * | |
3771 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3772 | * A caller has to handle it. This is mainly due to the fact that we want to | |
9608703e | 3773 | * avoid syncing under i_rwsem. |
a862f68a MR |
3774 | * |
3775 | * Return: | |
3776 | * * number of bytes written, even for truncated writes | |
3777 | * * negative error code if no data has been written at all | |
e4dd9de3 | 3778 | */ |
8174202b | 3779 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3780 | { |
3781 | struct file *file = iocb->ki_filp; | |
68d68ff6 | 3782 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3783 | struct inode *inode = mapping->host; |
3b93f911 | 3784 | ssize_t written = 0; |
1da177e4 | 3785 | ssize_t err; |
3b93f911 | 3786 | ssize_t status; |
1da177e4 | 3787 | |
1da177e4 | 3788 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3789 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3790 | err = file_remove_privs(file); |
1da177e4 LT |
3791 | if (err) |
3792 | goto out; | |
3793 | ||
c3b2da31 JB |
3794 | err = file_update_time(file); |
3795 | if (err) | |
3796 | goto out; | |
1da177e4 | 3797 | |
2ba48ce5 | 3798 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3799 | loff_t pos, endbyte; |
fb5527e6 | 3800 | |
1af5bb49 | 3801 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3802 | /* |
fbbbad4b MW |
3803 | * If the write stopped short of completing, fall back to |
3804 | * buffered writes. Some filesystems do this for writes to | |
3805 | * holes, for example. For DAX files, a buffered write will | |
3806 | * not succeed (even if it did, DAX does not handle dirty | |
3807 | * page-cache pages correctly). | |
1da177e4 | 3808 | */ |
0b8def9d | 3809 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3810 | goto out; |
3811 | ||
0b8def9d | 3812 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3813 | /* |
3b93f911 | 3814 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3815 | * then we want to return the number of bytes which were |
3816 | * direct-written, or the error code if that was zero. Note | |
3817 | * that this differs from normal direct-io semantics, which | |
3818 | * will return -EFOO even if some bytes were written. | |
3819 | */ | |
60bb4529 | 3820 | if (unlikely(status < 0)) { |
3b93f911 | 3821 | err = status; |
fb5527e6 JM |
3822 | goto out; |
3823 | } | |
fb5527e6 JM |
3824 | /* |
3825 | * We need to ensure that the page cache pages are written to | |
3826 | * disk and invalidated to preserve the expected O_DIRECT | |
3827 | * semantics. | |
3828 | */ | |
3b93f911 | 3829 | endbyte = pos + status - 1; |
0b8def9d | 3830 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3831 | if (err == 0) { |
0b8def9d | 3832 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3833 | written += status; |
fb5527e6 | 3834 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3835 | pos >> PAGE_SHIFT, |
3836 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3837 | } else { |
3838 | /* | |
3839 | * We don't know how much we wrote, so just return | |
3840 | * the number of bytes which were direct-written | |
3841 | */ | |
3842 | } | |
3843 | } else { | |
0b8def9d AV |
3844 | written = generic_perform_write(file, from, iocb->ki_pos); |
3845 | if (likely(written > 0)) | |
3846 | iocb->ki_pos += written; | |
fb5527e6 | 3847 | } |
1da177e4 LT |
3848 | out: |
3849 | current->backing_dev_info = NULL; | |
3850 | return written ? written : err; | |
3851 | } | |
8174202b | 3852 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3853 | |
e4dd9de3 | 3854 | /** |
8174202b | 3855 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3856 | * @iocb: IO state structure |
8174202b | 3857 | * @from: iov_iter with data to write |
e4dd9de3 | 3858 | * |
8174202b | 3859 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 | 3860 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
9608703e | 3861 | * and acquires i_rwsem as needed. |
a862f68a MR |
3862 | * Return: |
3863 | * * negative error code if no data has been written at all of | |
3864 | * vfs_fsync_range() failed for a synchronous write | |
3865 | * * number of bytes written, even for truncated writes | |
e4dd9de3 | 3866 | */ |
8174202b | 3867 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3868 | { |
3869 | struct file *file = iocb->ki_filp; | |
148f948b | 3870 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3871 | ssize_t ret; |
1da177e4 | 3872 | |
5955102c | 3873 | inode_lock(inode); |
3309dd04 AV |
3874 | ret = generic_write_checks(iocb, from); |
3875 | if (ret > 0) | |
5f380c7f | 3876 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3877 | inode_unlock(inode); |
1da177e4 | 3878 | |
e2592217 CH |
3879 | if (ret > 0) |
3880 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3881 | return ret; |
3882 | } | |
8174202b | 3883 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3884 | |
cf9a2ae8 DH |
3885 | /** |
3886 | * try_to_release_page() - release old fs-specific metadata on a page | |
3887 | * | |
3888 | * @page: the page which the kernel is trying to free | |
3889 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3890 | * | |
3891 | * The address_space is to try to release any data against the page | |
a862f68a | 3892 | * (presumably at page->private). |
cf9a2ae8 | 3893 | * |
266cf658 DH |
3894 | * This may also be called if PG_fscache is set on a page, indicating that the |
3895 | * page is known to the local caching routines. | |
3896 | * | |
cf9a2ae8 | 3897 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3898 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3899 | * |
a862f68a | 3900 | * Return: %1 if the release was successful, otherwise return zero. |
cf9a2ae8 DH |
3901 | */ |
3902 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3903 | { | |
3904 | struct address_space * const mapping = page->mapping; | |
3905 | ||
3906 | BUG_ON(!PageLocked(page)); | |
3907 | if (PageWriteback(page)) | |
3908 | return 0; | |
3909 | ||
3910 | if (mapping && mapping->a_ops->releasepage) | |
3911 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3912 | return try_to_free_buffers(page); | |
3913 | } | |
3914 | ||
3915 | EXPORT_SYMBOL(try_to_release_page); |