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> |
0f8053a5 NP |
46 | #include "internal.h" |
47 | ||
fe0bfaaf RJ |
48 | #define CREATE_TRACE_POINTS |
49 | #include <trace/events/filemap.h> | |
50 | ||
1da177e4 | 51 | /* |
1da177e4 LT |
52 | * FIXME: remove all knowledge of the buffer layer from the core VM |
53 | */ | |
148f948b | 54 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 55 | |
1da177e4 LT |
56 | #include <asm/mman.h> |
57 | ||
58 | /* | |
59 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
60 | * though. | |
61 | * | |
62 | * Shared mappings now work. 15.8.1995 Bruno. | |
63 | * | |
64 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
65 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
66 | * | |
67 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
68 | */ | |
69 | ||
70 | /* | |
71 | * Lock ordering: | |
72 | * | |
c8c06efa | 73 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 74 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 | 75 | * ->swap_lock (exclusive_swap_page, others) |
b93b0163 | 76 | * ->i_pages lock |
1da177e4 | 77 | * |
1b1dcc1b | 78 | * ->i_mutex |
c8c06efa | 79 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
1da177e4 | 80 | * |
c1e8d7c6 | 81 | * ->mmap_lock |
c8c06efa | 82 | * ->i_mmap_rwsem |
b8072f09 | 83 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
b93b0163 | 84 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
1da177e4 | 85 | * |
c1e8d7c6 | 86 | * ->mmap_lock |
1da177e4 LT |
87 | * ->lock_page (access_process_vm) |
88 | * | |
ccad2365 | 89 | * ->i_mutex (generic_perform_write) |
c1e8d7c6 | 90 | * ->mmap_lock (fault_in_pages_readable->do_page_fault) |
1da177e4 | 91 | * |
f758eeab | 92 | * bdi->wb.list_lock |
a66979ab | 93 | * sb_lock (fs/fs-writeback.c) |
b93b0163 | 94 | * ->i_pages lock (__sync_single_inode) |
1da177e4 | 95 | * |
c8c06efa | 96 | * ->i_mmap_rwsem |
1da177e4 LT |
97 | * ->anon_vma.lock (vma_adjust) |
98 | * | |
99 | * ->anon_vma.lock | |
b8072f09 | 100 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 101 | * |
b8072f09 | 102 | * ->page_table_lock or pte_lock |
5d337b91 | 103 | * ->swap_lock (try_to_unmap_one) |
1da177e4 | 104 | * ->private_lock (try_to_unmap_one) |
b93b0163 | 105 | * ->i_pages lock (try_to_unmap_one) |
15b44736 HD |
106 | * ->lruvec->lru_lock (follow_page->mark_page_accessed) |
107 | * ->lruvec->lru_lock (check_pte_range->isolate_lru_page) | |
1da177e4 | 108 | * ->private_lock (page_remove_rmap->set_page_dirty) |
b93b0163 | 109 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
f758eeab | 110 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 111 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 112 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 113 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 114 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
115 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
116 | * | |
c8c06efa | 117 | * ->i_mmap_rwsem |
9a3c531d | 118 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
119 | */ |
120 | ||
5c024e6a | 121 | static void page_cache_delete(struct address_space *mapping, |
91b0abe3 JW |
122 | struct page *page, void *shadow) |
123 | { | |
5c024e6a MW |
124 | XA_STATE(xas, &mapping->i_pages, page->index); |
125 | unsigned int nr = 1; | |
c70b647d | 126 | |
5c024e6a | 127 | mapping_set_update(&xas, mapping); |
c70b647d | 128 | |
5c024e6a MW |
129 | /* hugetlb pages are represented by a single entry in the xarray */ |
130 | if (!PageHuge(page)) { | |
131 | xas_set_order(&xas, page->index, compound_order(page)); | |
d8c6546b | 132 | nr = compound_nr(page); |
5c024e6a | 133 | } |
91b0abe3 | 134 | |
83929372 KS |
135 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
136 | VM_BUG_ON_PAGE(PageTail(page), page); | |
137 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 138 | |
5c024e6a MW |
139 | xas_store(&xas, shadow); |
140 | xas_init_marks(&xas); | |
d3798ae8 | 141 | |
2300638b JK |
142 | page->mapping = NULL; |
143 | /* Leave page->index set: truncation lookup relies upon it */ | |
144 | ||
d3798ae8 JW |
145 | if (shadow) { |
146 | mapping->nrexceptional += nr; | |
147 | /* | |
148 | * Make sure the nrexceptional update is committed before | |
149 | * the nrpages update so that final truncate racing | |
150 | * with reclaim does not see both counters 0 at the | |
151 | * same time and miss a shadow entry. | |
152 | */ | |
153 | smp_wmb(); | |
154 | } | |
155 | mapping->nrpages -= nr; | |
91b0abe3 JW |
156 | } |
157 | ||
5ecc4d85 JK |
158 | static void unaccount_page_cache_page(struct address_space *mapping, |
159 | struct page *page) | |
1da177e4 | 160 | { |
5ecc4d85 | 161 | int nr; |
1da177e4 | 162 | |
c515e1fd DM |
163 | /* |
164 | * if we're uptodate, flush out into the cleancache, otherwise | |
165 | * invalidate any existing cleancache entries. We can't leave | |
166 | * stale data around in the cleancache once our page is gone | |
167 | */ | |
168 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
169 | cleancache_put_page(page); | |
170 | else | |
3167760f | 171 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 172 | |
83929372 | 173 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
174 | VM_BUG_ON_PAGE(page_mapped(page), page); |
175 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
176 | int mapcount; | |
177 | ||
178 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
179 | current->comm, page_to_pfn(page)); | |
180 | dump_page(page, "still mapped when deleted"); | |
181 | dump_stack(); | |
182 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
183 | ||
184 | mapcount = page_mapcount(page); | |
185 | if (mapping_exiting(mapping) && | |
186 | page_count(page) >= mapcount + 2) { | |
187 | /* | |
188 | * All vmas have already been torn down, so it's | |
189 | * a good bet that actually the page is unmapped, | |
190 | * and we'd prefer not to leak it: if we're wrong, | |
191 | * some other bad page check should catch it later. | |
192 | */ | |
193 | page_mapcount_reset(page); | |
6d061f9f | 194 | page_ref_sub(page, mapcount); |
06b241f3 HD |
195 | } |
196 | } | |
197 | ||
4165b9b4 | 198 | /* hugetlb pages do not participate in page cache accounting. */ |
5ecc4d85 JK |
199 | if (PageHuge(page)) |
200 | return; | |
09612fa6 | 201 | |
6c357848 | 202 | nr = thp_nr_pages(page); |
5ecc4d85 | 203 | |
0d1c2072 | 204 | __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr); |
5ecc4d85 | 205 | if (PageSwapBacked(page)) { |
0d1c2072 | 206 | __mod_lruvec_page_state(page, NR_SHMEM, -nr); |
5ecc4d85 | 207 | if (PageTransHuge(page)) |
b8eddff8 | 208 | __dec_lruvec_page_state(page, NR_SHMEM_THPS); |
99cb0dbd | 209 | } else if (PageTransHuge(page)) { |
b8eddff8 | 210 | __dec_lruvec_page_state(page, NR_FILE_THPS); |
09d91cda | 211 | filemap_nr_thps_dec(mapping); |
800d8c63 | 212 | } |
5ecc4d85 JK |
213 | |
214 | /* | |
215 | * At this point page must be either written or cleaned by | |
216 | * truncate. Dirty page here signals a bug and loss of | |
217 | * unwritten data. | |
218 | * | |
219 | * This fixes dirty accounting after removing the page entirely | |
220 | * but leaves PageDirty set: it has no effect for truncated | |
221 | * page and anyway will be cleared before returning page into | |
222 | * buddy allocator. | |
223 | */ | |
224 | if (WARN_ON_ONCE(PageDirty(page))) | |
225 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); | |
226 | } | |
227 | ||
228 | /* | |
229 | * Delete a page from the page cache and free it. Caller has to make | |
230 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 231 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 JK |
232 | */ |
233 | void __delete_from_page_cache(struct page *page, void *shadow) | |
234 | { | |
235 | struct address_space *mapping = page->mapping; | |
236 | ||
237 | trace_mm_filemap_delete_from_page_cache(page); | |
238 | ||
239 | unaccount_page_cache_page(mapping, page); | |
5c024e6a | 240 | page_cache_delete(mapping, page, shadow); |
1da177e4 LT |
241 | } |
242 | ||
59c66c5f JK |
243 | static void page_cache_free_page(struct address_space *mapping, |
244 | struct page *page) | |
245 | { | |
246 | void (*freepage)(struct page *); | |
247 | ||
248 | freepage = mapping->a_ops->freepage; | |
249 | if (freepage) | |
250 | freepage(page); | |
251 | ||
252 | if (PageTransHuge(page) && !PageHuge(page)) { | |
887b22c6 | 253 | page_ref_sub(page, thp_nr_pages(page)); |
59c66c5f JK |
254 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); |
255 | } else { | |
256 | put_page(page); | |
257 | } | |
258 | } | |
259 | ||
702cfbf9 MK |
260 | /** |
261 | * delete_from_page_cache - delete page from page cache | |
262 | * @page: the page which the kernel is trying to remove from page cache | |
263 | * | |
264 | * This must be called only on pages that have been verified to be in the page | |
265 | * cache and locked. It will never put the page into the free list, the caller | |
266 | * has a reference on the page. | |
267 | */ | |
268 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 269 | { |
83929372 | 270 | struct address_space *mapping = page_mapping(page); |
c4843a75 | 271 | unsigned long flags; |
1da177e4 | 272 | |
cd7619d6 | 273 | BUG_ON(!PageLocked(page)); |
b93b0163 | 274 | xa_lock_irqsave(&mapping->i_pages, flags); |
62cccb8c | 275 | __delete_from_page_cache(page, NULL); |
b93b0163 | 276 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
6072d13c | 277 | |
59c66c5f | 278 | page_cache_free_page(mapping, page); |
97cecb5a MK |
279 | } |
280 | EXPORT_SYMBOL(delete_from_page_cache); | |
281 | ||
aa65c29c | 282 | /* |
ef8e5717 | 283 | * page_cache_delete_batch - delete several pages from page cache |
aa65c29c JK |
284 | * @mapping: the mapping to which pages belong |
285 | * @pvec: pagevec with pages to delete | |
286 | * | |
b93b0163 | 287 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
4101196b MWO |
288 | * from the mapping. The function expects @pvec to be sorted by page index |
289 | * and is optimised for it to be dense. | |
b93b0163 | 290 | * It tolerates holes in @pvec (mapping entries at those indices are not |
aa65c29c | 291 | * modified). The function expects only THP head pages to be present in the |
4101196b | 292 | * @pvec. |
aa65c29c | 293 | * |
b93b0163 | 294 | * The function expects the i_pages lock to be held. |
aa65c29c | 295 | */ |
ef8e5717 | 296 | static void page_cache_delete_batch(struct address_space *mapping, |
aa65c29c JK |
297 | struct pagevec *pvec) |
298 | { | |
ef8e5717 | 299 | XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index); |
aa65c29c | 300 | int total_pages = 0; |
4101196b | 301 | int i = 0; |
aa65c29c | 302 | struct page *page; |
aa65c29c | 303 | |
ef8e5717 MW |
304 | mapping_set_update(&xas, mapping); |
305 | xas_for_each(&xas, page, ULONG_MAX) { | |
4101196b | 306 | if (i >= pagevec_count(pvec)) |
aa65c29c | 307 | break; |
4101196b MWO |
308 | |
309 | /* A swap/dax/shadow entry got inserted? Skip it. */ | |
3159f943 | 310 | if (xa_is_value(page)) |
aa65c29c | 311 | continue; |
4101196b MWO |
312 | /* |
313 | * A page got inserted in our range? Skip it. We have our | |
314 | * pages locked so they are protected from being removed. | |
315 | * If we see a page whose index is higher than ours, it | |
316 | * means our page has been removed, which shouldn't be | |
317 | * possible because we're holding the PageLock. | |
318 | */ | |
319 | if (page != pvec->pages[i]) { | |
320 | VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index, | |
321 | page); | |
322 | continue; | |
323 | } | |
324 | ||
325 | WARN_ON_ONCE(!PageLocked(page)); | |
326 | ||
327 | if (page->index == xas.xa_index) | |
aa65c29c | 328 | page->mapping = NULL; |
4101196b MWO |
329 | /* Leave page->index set: truncation lookup relies on it */ |
330 | ||
331 | /* | |
332 | * Move to the next page in the vector if this is a regular | |
333 | * page or the index is of the last sub-page of this compound | |
334 | * page. | |
335 | */ | |
336 | if (page->index + compound_nr(page) - 1 == xas.xa_index) | |
aa65c29c | 337 | i++; |
ef8e5717 | 338 | xas_store(&xas, NULL); |
aa65c29c JK |
339 | total_pages++; |
340 | } | |
341 | mapping->nrpages -= total_pages; | |
342 | } | |
343 | ||
344 | void delete_from_page_cache_batch(struct address_space *mapping, | |
345 | struct pagevec *pvec) | |
346 | { | |
347 | int i; | |
348 | unsigned long flags; | |
349 | ||
350 | if (!pagevec_count(pvec)) | |
351 | return; | |
352 | ||
b93b0163 | 353 | xa_lock_irqsave(&mapping->i_pages, flags); |
aa65c29c JK |
354 | for (i = 0; i < pagevec_count(pvec); i++) { |
355 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); | |
356 | ||
357 | unaccount_page_cache_page(mapping, pvec->pages[i]); | |
358 | } | |
ef8e5717 | 359 | page_cache_delete_batch(mapping, pvec); |
b93b0163 | 360 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
aa65c29c JK |
361 | |
362 | for (i = 0; i < pagevec_count(pvec); i++) | |
363 | page_cache_free_page(mapping, pvec->pages[i]); | |
364 | } | |
365 | ||
d72d9e2a | 366 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
367 | { |
368 | int ret = 0; | |
369 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
370 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
371 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 372 | ret = -ENOSPC; |
7fcbbaf1 JA |
373 | if (test_bit(AS_EIO, &mapping->flags) && |
374 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
375 | ret = -EIO; |
376 | return ret; | |
377 | } | |
d72d9e2a | 378 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 379 | |
76341cab JL |
380 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
381 | { | |
382 | /* Check for outstanding write errors */ | |
383 | if (test_bit(AS_EIO, &mapping->flags)) | |
384 | return -EIO; | |
385 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
386 | return -ENOSPC; | |
387 | return 0; | |
388 | } | |
389 | ||
1da177e4 | 390 | /** |
485bb99b | 391 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
392 | * @mapping: address space structure to write |
393 | * @start: offset in bytes where the range starts | |
469eb4d0 | 394 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 395 | * @sync_mode: enable synchronous operation |
1da177e4 | 396 | * |
485bb99b RD |
397 | * Start writeback against all of a mapping's dirty pages that lie |
398 | * within the byte offsets <start, end> inclusive. | |
399 | * | |
1da177e4 | 400 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 401 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
402 | * these two operations is that if a dirty page/buffer is encountered, it must |
403 | * be waited upon, and not just skipped over. | |
a862f68a MR |
404 | * |
405 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 406 | */ |
ebcf28e1 AM |
407 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
408 | loff_t end, int sync_mode) | |
1da177e4 LT |
409 | { |
410 | int ret; | |
411 | struct writeback_control wbc = { | |
412 | .sync_mode = sync_mode, | |
05fe478d | 413 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
414 | .range_start = start, |
415 | .range_end = end, | |
1da177e4 LT |
416 | }; |
417 | ||
f56753ac | 418 | if (!mapping_can_writeback(mapping) || |
c3aab9a0 | 419 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
1da177e4 LT |
420 | return 0; |
421 | ||
b16b1deb | 422 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 423 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 424 | wbc_detach_inode(&wbc); |
1da177e4 LT |
425 | return ret; |
426 | } | |
427 | ||
428 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
429 | int sync_mode) | |
430 | { | |
111ebb6e | 431 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
432 | } |
433 | ||
434 | int filemap_fdatawrite(struct address_space *mapping) | |
435 | { | |
436 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
437 | } | |
438 | EXPORT_SYMBOL(filemap_fdatawrite); | |
439 | ||
f4c0a0fd | 440 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 441 | loff_t end) |
1da177e4 LT |
442 | { |
443 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
444 | } | |
f4c0a0fd | 445 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 446 | |
485bb99b RD |
447 | /** |
448 | * filemap_flush - mostly a non-blocking flush | |
449 | * @mapping: target address_space | |
450 | * | |
1da177e4 LT |
451 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
452 | * purposes - I/O may not be started against all dirty pages. | |
a862f68a MR |
453 | * |
454 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 LT |
455 | */ |
456 | int filemap_flush(struct address_space *mapping) | |
457 | { | |
458 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
459 | } | |
460 | EXPORT_SYMBOL(filemap_flush); | |
461 | ||
7fc9e472 GR |
462 | /** |
463 | * filemap_range_has_page - check if a page exists in range. | |
464 | * @mapping: address space within which to check | |
465 | * @start_byte: offset in bytes where the range starts | |
466 | * @end_byte: offset in bytes where the range ends (inclusive) | |
467 | * | |
468 | * Find at least one page in the range supplied, usually used to check if | |
469 | * direct writing in this range will trigger a writeback. | |
a862f68a MR |
470 | * |
471 | * Return: %true if at least one page exists in the specified range, | |
472 | * %false otherwise. | |
7fc9e472 GR |
473 | */ |
474 | bool filemap_range_has_page(struct address_space *mapping, | |
475 | loff_t start_byte, loff_t end_byte) | |
476 | { | |
f7b68046 | 477 | struct page *page; |
8fa8e538 MW |
478 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
479 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
7fc9e472 GR |
480 | |
481 | if (end_byte < start_byte) | |
482 | return false; | |
483 | ||
8fa8e538 MW |
484 | rcu_read_lock(); |
485 | for (;;) { | |
486 | page = xas_find(&xas, max); | |
487 | if (xas_retry(&xas, page)) | |
488 | continue; | |
489 | /* Shadow entries don't count */ | |
490 | if (xa_is_value(page)) | |
491 | continue; | |
492 | /* | |
493 | * We don't need to try to pin this page; we're about to | |
494 | * release the RCU lock anyway. It is enough to know that | |
495 | * there was a page here recently. | |
496 | */ | |
497 | break; | |
498 | } | |
499 | rcu_read_unlock(); | |
7fc9e472 | 500 | |
8fa8e538 | 501 | return page != NULL; |
7fc9e472 GR |
502 | } |
503 | EXPORT_SYMBOL(filemap_range_has_page); | |
504 | ||
5e8fcc1a | 505 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 506 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 507 | { |
09cbfeaf KS |
508 | pgoff_t index = start_byte >> PAGE_SHIFT; |
509 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
510 | struct pagevec pvec; |
511 | int nr_pages; | |
1da177e4 | 512 | |
94004ed7 | 513 | if (end_byte < start_byte) |
5e8fcc1a | 514 | return; |
1da177e4 | 515 | |
86679820 | 516 | pagevec_init(&pvec); |
312e9d2f | 517 | while (index <= end) { |
1da177e4 LT |
518 | unsigned i; |
519 | ||
312e9d2f | 520 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 521 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
522 | if (!nr_pages) |
523 | break; | |
524 | ||
1da177e4 LT |
525 | for (i = 0; i < nr_pages; i++) { |
526 | struct page *page = pvec.pages[i]; | |
527 | ||
1da177e4 | 528 | wait_on_page_writeback(page); |
5e8fcc1a | 529 | ClearPageError(page); |
1da177e4 LT |
530 | } |
531 | pagevec_release(&pvec); | |
532 | cond_resched(); | |
533 | } | |
aa750fd7 JN |
534 | } |
535 | ||
536 | /** | |
537 | * filemap_fdatawait_range - wait for writeback to complete | |
538 | * @mapping: address space structure to wait for | |
539 | * @start_byte: offset in bytes where the range starts | |
540 | * @end_byte: offset in bytes where the range ends (inclusive) | |
541 | * | |
542 | * Walk the list of under-writeback pages of the given address space | |
543 | * in the given range and wait for all of them. Check error status of | |
544 | * the address space and return it. | |
545 | * | |
546 | * Since the error status of the address space is cleared by this function, | |
547 | * callers are responsible for checking the return value and handling and/or | |
548 | * reporting the error. | |
a862f68a MR |
549 | * |
550 | * Return: error status of the address space. | |
aa750fd7 JN |
551 | */ |
552 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
553 | loff_t end_byte) | |
554 | { | |
5e8fcc1a JL |
555 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
556 | return filemap_check_errors(mapping); | |
1da177e4 | 557 | } |
d3bccb6f JK |
558 | EXPORT_SYMBOL(filemap_fdatawait_range); |
559 | ||
aa0bfcd9 RZ |
560 | /** |
561 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete | |
562 | * @mapping: address space structure to wait for | |
563 | * @start_byte: offset in bytes where the range starts | |
564 | * @end_byte: offset in bytes where the range ends (inclusive) | |
565 | * | |
566 | * Walk the list of under-writeback pages of the given address space in the | |
567 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), | |
568 | * this function does not clear error status of the address space. | |
569 | * | |
570 | * Use this function if callers don't handle errors themselves. Expected | |
571 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
572 | * fsfreeze(8) | |
573 | */ | |
574 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, | |
575 | loff_t start_byte, loff_t end_byte) | |
576 | { | |
577 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
578 | return filemap_check_and_keep_errors(mapping); | |
579 | } | |
580 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); | |
581 | ||
a823e458 JL |
582 | /** |
583 | * file_fdatawait_range - wait for writeback to complete | |
584 | * @file: file pointing to address space structure to wait for | |
585 | * @start_byte: offset in bytes where the range starts | |
586 | * @end_byte: offset in bytes where the range ends (inclusive) | |
587 | * | |
588 | * Walk the list of under-writeback pages of the address space that file | |
589 | * refers to, in the given range and wait for all of them. Check error | |
590 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
591 | * | |
592 | * Since the error status of the file is advanced by this function, | |
593 | * callers are responsible for checking the return value and handling and/or | |
594 | * reporting the error. | |
a862f68a MR |
595 | * |
596 | * Return: error status of the address space vs. the file->f_wb_err cursor. | |
a823e458 JL |
597 | */ |
598 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
599 | { | |
600 | struct address_space *mapping = file->f_mapping; | |
601 | ||
602 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
603 | return file_check_and_advance_wb_err(file); | |
604 | } | |
605 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 606 | |
aa750fd7 JN |
607 | /** |
608 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
609 | * @mapping: address space structure to wait for | |
610 | * | |
611 | * Walk the list of under-writeback pages of the given address space | |
612 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
613 | * does not clear error status of the address space. | |
614 | * | |
615 | * Use this function if callers don't handle errors themselves. Expected | |
616 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
617 | * fsfreeze(8) | |
a862f68a MR |
618 | * |
619 | * Return: error status of the address space. | |
aa750fd7 | 620 | */ |
76341cab | 621 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 622 | { |
ffb959bb | 623 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 624 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 625 | } |
76341cab | 626 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 627 | |
875d91b1 | 628 | /* Returns true if writeback might be needed or already in progress. */ |
9326c9b2 | 629 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 630 | { |
875d91b1 KK |
631 | if (dax_mapping(mapping)) |
632 | return mapping->nrexceptional; | |
633 | ||
634 | return mapping->nrpages; | |
1da177e4 | 635 | } |
1da177e4 | 636 | |
485bb99b RD |
637 | /** |
638 | * filemap_write_and_wait_range - write out & wait on a file range | |
639 | * @mapping: the address_space for the pages | |
640 | * @lstart: offset in bytes where the range starts | |
641 | * @lend: offset in bytes where the range ends (inclusive) | |
642 | * | |
469eb4d0 AM |
643 | * Write out and wait upon file offsets lstart->lend, inclusive. |
644 | * | |
0e056eb5 | 645 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 | 646 | * that this function can be used to write to the very end-of-file (end = -1). |
a862f68a MR |
647 | * |
648 | * Return: error status of the address space. | |
469eb4d0 | 649 | */ |
1da177e4 LT |
650 | int filemap_write_and_wait_range(struct address_space *mapping, |
651 | loff_t lstart, loff_t lend) | |
652 | { | |
28fd1298 | 653 | int err = 0; |
1da177e4 | 654 | |
9326c9b2 | 655 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
656 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
657 | WB_SYNC_ALL); | |
ddf8f376 IW |
658 | /* |
659 | * Even if the above returned error, the pages may be | |
660 | * written partially (e.g. -ENOSPC), so we wait for it. | |
661 | * But the -EIO is special case, it may indicate the worst | |
662 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
663 | */ | |
28fd1298 | 664 | if (err != -EIO) { |
94004ed7 CH |
665 | int err2 = filemap_fdatawait_range(mapping, |
666 | lstart, lend); | |
28fd1298 OH |
667 | if (!err) |
668 | err = err2; | |
cbeaf951 JL |
669 | } else { |
670 | /* Clear any previously stored errors */ | |
671 | filemap_check_errors(mapping); | |
28fd1298 | 672 | } |
865ffef3 DM |
673 | } else { |
674 | err = filemap_check_errors(mapping); | |
1da177e4 | 675 | } |
28fd1298 | 676 | return err; |
1da177e4 | 677 | } |
f6995585 | 678 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 679 | |
5660e13d JL |
680 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
681 | { | |
3acdfd28 | 682 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
683 | |
684 | trace_filemap_set_wb_err(mapping, eseq); | |
685 | } | |
686 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
687 | ||
688 | /** | |
689 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
690 | * and advance wb_err to current one | |
691 | * @file: struct file on which the error is being reported | |
692 | * | |
693 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
694 | * want to report any writeback errors that occurred since the last fsync (or | |
695 | * since the file was opened if there haven't been any). | |
696 | * | |
697 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
698 | * then just quickly return 0. The file is all caught up. | |
699 | * | |
700 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
701 | * it and try to swap it into place. If it works, or another task beat us | |
702 | * to it with the new value, then update the f_wb_err and return the error | |
703 | * portion. The error at this point must be reported via proper channels | |
704 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
705 | * | |
706 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
707 | * value is protected by the f_lock since we must ensure that it reflects | |
708 | * the latest value swapped in for this file descriptor. | |
a862f68a MR |
709 | * |
710 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
711 | */ |
712 | int file_check_and_advance_wb_err(struct file *file) | |
713 | { | |
714 | int err = 0; | |
715 | errseq_t old = READ_ONCE(file->f_wb_err); | |
716 | struct address_space *mapping = file->f_mapping; | |
717 | ||
718 | /* Locklessly handle the common case where nothing has changed */ | |
719 | if (errseq_check(&mapping->wb_err, old)) { | |
720 | /* Something changed, must use slow path */ | |
721 | spin_lock(&file->f_lock); | |
722 | old = file->f_wb_err; | |
723 | err = errseq_check_and_advance(&mapping->wb_err, | |
724 | &file->f_wb_err); | |
725 | trace_file_check_and_advance_wb_err(file, old); | |
726 | spin_unlock(&file->f_lock); | |
727 | } | |
f4e222c5 JL |
728 | |
729 | /* | |
730 | * We're mostly using this function as a drop in replacement for | |
731 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
732 | * that the legacy code would have had on these flags. | |
733 | */ | |
734 | clear_bit(AS_EIO, &mapping->flags); | |
735 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
736 | return err; |
737 | } | |
738 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
739 | ||
740 | /** | |
741 | * file_write_and_wait_range - write out & wait on a file range | |
742 | * @file: file pointing to address_space with pages | |
743 | * @lstart: offset in bytes where the range starts | |
744 | * @lend: offset in bytes where the range ends (inclusive) | |
745 | * | |
746 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
747 | * | |
748 | * Note that @lend is inclusive (describes the last byte to be written) so | |
749 | * that this function can be used to write to the very end-of-file (end = -1). | |
750 | * | |
751 | * After writing out and waiting on the data, we check and advance the | |
752 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
a862f68a MR |
753 | * |
754 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
755 | */ |
756 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
757 | { | |
758 | int err = 0, err2; | |
759 | struct address_space *mapping = file->f_mapping; | |
760 | ||
9326c9b2 | 761 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
762 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
763 | WB_SYNC_ALL); | |
764 | /* See comment of filemap_write_and_wait() */ | |
765 | if (err != -EIO) | |
766 | __filemap_fdatawait_range(mapping, lstart, lend); | |
767 | } | |
768 | err2 = file_check_and_advance_wb_err(file); | |
769 | if (!err) | |
770 | err = err2; | |
771 | return err; | |
772 | } | |
773 | EXPORT_SYMBOL(file_write_and_wait_range); | |
774 | ||
ef6a3c63 MS |
775 | /** |
776 | * replace_page_cache_page - replace a pagecache page with a new one | |
777 | * @old: page to be replaced | |
778 | * @new: page to replace with | |
779 | * @gfp_mask: allocation mode | |
780 | * | |
781 | * This function replaces a page in the pagecache with a new one. On | |
782 | * success it acquires the pagecache reference for the new page and | |
783 | * drops it for the old page. Both the old and new pages must be | |
784 | * locked. This function does not add the new page to the LRU, the | |
785 | * caller must do that. | |
786 | * | |
74d60958 | 787 | * The remove + add is atomic. This function cannot fail. |
a862f68a MR |
788 | * |
789 | * Return: %0 | |
ef6a3c63 MS |
790 | */ |
791 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
792 | { | |
74d60958 MW |
793 | struct address_space *mapping = old->mapping; |
794 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
795 | pgoff_t offset = old->index; | |
796 | XA_STATE(xas, &mapping->i_pages, offset); | |
797 | unsigned long flags; | |
ef6a3c63 | 798 | |
309381fe SL |
799 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
800 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
801 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 802 | |
74d60958 MW |
803 | get_page(new); |
804 | new->mapping = mapping; | |
805 | new->index = offset; | |
ef6a3c63 | 806 | |
0d1c2072 JW |
807 | mem_cgroup_migrate(old, new); |
808 | ||
74d60958 MW |
809 | xas_lock_irqsave(&xas, flags); |
810 | xas_store(&xas, new); | |
4165b9b4 | 811 | |
74d60958 MW |
812 | old->mapping = NULL; |
813 | /* hugetlb pages do not participate in page cache accounting. */ | |
814 | if (!PageHuge(old)) | |
0d1c2072 | 815 | __dec_lruvec_page_state(old, NR_FILE_PAGES); |
74d60958 | 816 | if (!PageHuge(new)) |
0d1c2072 | 817 | __inc_lruvec_page_state(new, NR_FILE_PAGES); |
74d60958 | 818 | if (PageSwapBacked(old)) |
0d1c2072 | 819 | __dec_lruvec_page_state(old, NR_SHMEM); |
74d60958 | 820 | if (PageSwapBacked(new)) |
0d1c2072 | 821 | __inc_lruvec_page_state(new, NR_SHMEM); |
74d60958 | 822 | xas_unlock_irqrestore(&xas, flags); |
74d60958 MW |
823 | if (freepage) |
824 | freepage(old); | |
825 | put_page(old); | |
ef6a3c63 | 826 | |
74d60958 | 827 | return 0; |
ef6a3c63 MS |
828 | } |
829 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
830 | ||
16c0cc0c | 831 | noinline int __add_to_page_cache_locked(struct page *page, |
76cd6173 | 832 | struct address_space *mapping, |
c4cf498d | 833 | pgoff_t offset, gfp_t gfp, |
76cd6173 | 834 | void **shadowp) |
1da177e4 | 835 | { |
74d60958 | 836 | XA_STATE(xas, &mapping->i_pages, offset); |
00501b53 | 837 | int huge = PageHuge(page); |
e286781d NP |
838 | int error; |
839 | ||
309381fe SL |
840 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
841 | VM_BUG_ON_PAGE(PageSwapBacked(page), page); | |
74d60958 | 842 | mapping_set_update(&xas, mapping); |
e286781d | 843 | |
09cbfeaf | 844 | get_page(page); |
66a0c8ee KS |
845 | page->mapping = mapping; |
846 | page->index = offset; | |
847 | ||
3fea5a49 | 848 | if (!huge) { |
198b62f8 | 849 | error = mem_cgroup_charge(page, current->mm, gfp); |
3fea5a49 JW |
850 | if (error) |
851 | goto error; | |
852 | } | |
853 | ||
198b62f8 MWO |
854 | gfp &= GFP_RECLAIM_MASK; |
855 | ||
74d60958 | 856 | do { |
198b62f8 MWO |
857 | unsigned int order = xa_get_order(xas.xa, xas.xa_index); |
858 | void *entry, *old = NULL; | |
859 | ||
860 | if (order > thp_order(page)) | |
861 | xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index), | |
862 | order, gfp); | |
74d60958 | 863 | xas_lock_irq(&xas); |
198b62f8 MWO |
864 | xas_for_each_conflict(&xas, entry) { |
865 | old = entry; | |
866 | if (!xa_is_value(entry)) { | |
867 | xas_set_err(&xas, -EEXIST); | |
868 | goto unlock; | |
869 | } | |
870 | } | |
871 | ||
872 | if (old) { | |
873 | if (shadowp) | |
874 | *shadowp = old; | |
875 | /* entry may have been split before we acquired lock */ | |
876 | order = xa_get_order(xas.xa, xas.xa_index); | |
877 | if (order > thp_order(page)) { | |
878 | xas_split(&xas, old, order); | |
879 | xas_reset(&xas); | |
880 | } | |
881 | } | |
882 | ||
74d60958 MW |
883 | xas_store(&xas, page); |
884 | if (xas_error(&xas)) | |
885 | goto unlock; | |
886 | ||
198b62f8 | 887 | if (old) |
74d60958 | 888 | mapping->nrexceptional--; |
74d60958 MW |
889 | mapping->nrpages++; |
890 | ||
891 | /* hugetlb pages do not participate in page cache accounting */ | |
892 | if (!huge) | |
0d1c2072 | 893 | __inc_lruvec_page_state(page, NR_FILE_PAGES); |
74d60958 MW |
894 | unlock: |
895 | xas_unlock_irq(&xas); | |
198b62f8 | 896 | } while (xas_nomem(&xas, gfp)); |
74d60958 | 897 | |
3fea5a49 JW |
898 | if (xas_error(&xas)) { |
899 | error = xas_error(&xas); | |
74d60958 | 900 | goto error; |
3fea5a49 | 901 | } |
4165b9b4 | 902 | |
66a0c8ee KS |
903 | trace_mm_filemap_add_to_page_cache(page); |
904 | return 0; | |
74d60958 | 905 | error: |
66a0c8ee KS |
906 | page->mapping = NULL; |
907 | /* Leave page->index set: truncation relies upon it */ | |
09cbfeaf | 908 | put_page(page); |
3fea5a49 | 909 | return error; |
1da177e4 | 910 | } |
cfcbfb13 | 911 | ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO); |
a528910e JW |
912 | |
913 | /** | |
914 | * add_to_page_cache_locked - add a locked page to the pagecache | |
915 | * @page: page to add | |
916 | * @mapping: the page's address_space | |
917 | * @offset: page index | |
918 | * @gfp_mask: page allocation mode | |
919 | * | |
920 | * This function is used to add a page to the pagecache. It must be locked. | |
921 | * This function does not add the page to the LRU. The caller must do that. | |
a862f68a MR |
922 | * |
923 | * Return: %0 on success, negative error code otherwise. | |
a528910e JW |
924 | */ |
925 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
926 | pgoff_t offset, gfp_t gfp_mask) | |
927 | { | |
928 | return __add_to_page_cache_locked(page, mapping, offset, | |
929 | gfp_mask, NULL); | |
930 | } | |
e286781d | 931 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
932 | |
933 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 934 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 935 | { |
a528910e | 936 | void *shadow = NULL; |
4f98a2fe RR |
937 | int ret; |
938 | ||
48c935ad | 939 | __SetPageLocked(page); |
a528910e JW |
940 | ret = __add_to_page_cache_locked(page, mapping, offset, |
941 | gfp_mask, &shadow); | |
942 | if (unlikely(ret)) | |
48c935ad | 943 | __ClearPageLocked(page); |
a528910e JW |
944 | else { |
945 | /* | |
946 | * The page might have been evicted from cache only | |
947 | * recently, in which case it should be activated like | |
948 | * any other repeatedly accessed page. | |
f0281a00 RR |
949 | * The exception is pages getting rewritten; evicting other |
950 | * data from the working set, only to cache data that will | |
951 | * get overwritten with something else, is a waste of memory. | |
a528910e | 952 | */ |
1899ad18 JW |
953 | WARN_ON_ONCE(PageActive(page)); |
954 | if (!(gfp_mask & __GFP_WRITE) && shadow) | |
955 | workingset_refault(page, shadow); | |
a528910e JW |
956 | lru_cache_add(page); |
957 | } | |
1da177e4 LT |
958 | return ret; |
959 | } | |
18bc0bbd | 960 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 961 | |
44110fe3 | 962 | #ifdef CONFIG_NUMA |
2ae88149 | 963 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 964 | { |
c0ff7453 MX |
965 | int n; |
966 | struct page *page; | |
967 | ||
44110fe3 | 968 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
969 | unsigned int cpuset_mems_cookie; |
970 | do { | |
d26914d1 | 971 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 972 | n = cpuset_mem_spread_node(); |
96db800f | 973 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 974 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 975 | |
c0ff7453 | 976 | return page; |
44110fe3 | 977 | } |
2ae88149 | 978 | return alloc_pages(gfp, 0); |
44110fe3 | 979 | } |
2ae88149 | 980 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
981 | #endif |
982 | ||
1da177e4 LT |
983 | /* |
984 | * In order to wait for pages to become available there must be | |
985 | * waitqueues associated with pages. By using a hash table of | |
986 | * waitqueues where the bucket discipline is to maintain all | |
987 | * waiters on the same queue and wake all when any of the pages | |
988 | * become available, and for the woken contexts to check to be | |
989 | * sure the appropriate page became available, this saves space | |
990 | * at a cost of "thundering herd" phenomena during rare hash | |
991 | * collisions. | |
992 | */ | |
62906027 NP |
993 | #define PAGE_WAIT_TABLE_BITS 8 |
994 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
995 | static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; | |
996 | ||
997 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
1da177e4 | 998 | { |
62906027 | 999 | return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 1000 | } |
1da177e4 | 1001 | |
62906027 | 1002 | void __init pagecache_init(void) |
1da177e4 | 1003 | { |
62906027 | 1004 | int i; |
1da177e4 | 1005 | |
62906027 NP |
1006 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
1007 | init_waitqueue_head(&page_wait_table[i]); | |
1008 | ||
1009 | page_writeback_init(); | |
1da177e4 | 1010 | } |
1da177e4 | 1011 | |
5ef64cc8 LT |
1012 | /* |
1013 | * The page wait code treats the "wait->flags" somewhat unusually, because | |
5868ec26 | 1014 | * we have multiple different kinds of waits, not just the usual "exclusive" |
5ef64cc8 LT |
1015 | * one. |
1016 | * | |
1017 | * We have: | |
1018 | * | |
1019 | * (a) no special bits set: | |
1020 | * | |
1021 | * We're just waiting for the bit to be released, and when a waker | |
1022 | * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up, | |
1023 | * and remove it from the wait queue. | |
1024 | * | |
1025 | * Simple and straightforward. | |
1026 | * | |
1027 | * (b) WQ_FLAG_EXCLUSIVE: | |
1028 | * | |
1029 | * The waiter is waiting to get the lock, and only one waiter should | |
1030 | * be woken up to avoid any thundering herd behavior. We'll set the | |
1031 | * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue. | |
1032 | * | |
1033 | * This is the traditional exclusive wait. | |
1034 | * | |
5868ec26 | 1035 | * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: |
5ef64cc8 LT |
1036 | * |
1037 | * The waiter is waiting to get the bit, and additionally wants the | |
1038 | * lock to be transferred to it for fair lock behavior. If the lock | |
1039 | * cannot be taken, we stop walking the wait queue without waking | |
1040 | * the waiter. | |
1041 | * | |
1042 | * This is the "fair lock handoff" case, and in addition to setting | |
1043 | * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see | |
1044 | * that it now has the lock. | |
1045 | */ | |
ac6424b9 | 1046 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 1047 | { |
5ef64cc8 | 1048 | unsigned int flags; |
62906027 NP |
1049 | struct wait_page_key *key = arg; |
1050 | struct wait_page_queue *wait_page | |
1051 | = container_of(wait, struct wait_page_queue, wait); | |
1052 | ||
cdc8fcb4 | 1053 | if (!wake_page_match(wait_page, key)) |
62906027 | 1054 | return 0; |
3510ca20 | 1055 | |
9a1ea439 | 1056 | /* |
5ef64cc8 LT |
1057 | * If it's a lock handoff wait, we get the bit for it, and |
1058 | * stop walking (and do not wake it up) if we can't. | |
9a1ea439 | 1059 | */ |
5ef64cc8 LT |
1060 | flags = wait->flags; |
1061 | if (flags & WQ_FLAG_EXCLUSIVE) { | |
1062 | if (test_bit(key->bit_nr, &key->page->flags)) | |
2a9127fc | 1063 | return -1; |
5ef64cc8 LT |
1064 | if (flags & WQ_FLAG_CUSTOM) { |
1065 | if (test_and_set_bit(key->bit_nr, &key->page->flags)) | |
1066 | return -1; | |
1067 | flags |= WQ_FLAG_DONE; | |
1068 | } | |
2a9127fc | 1069 | } |
f62e00cc | 1070 | |
5ef64cc8 LT |
1071 | /* |
1072 | * We are holding the wait-queue lock, but the waiter that | |
1073 | * is waiting for this will be checking the flags without | |
1074 | * any locking. | |
1075 | * | |
1076 | * So update the flags atomically, and wake up the waiter | |
1077 | * afterwards to avoid any races. This store-release pairs | |
1078 | * with the load-acquire in wait_on_page_bit_common(). | |
1079 | */ | |
1080 | smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN); | |
2a9127fc LT |
1081 | wake_up_state(wait->private, mode); |
1082 | ||
1083 | /* | |
1084 | * Ok, we have successfully done what we're waiting for, | |
1085 | * and we can unconditionally remove the wait entry. | |
1086 | * | |
5ef64cc8 LT |
1087 | * Note that this pairs with the "finish_wait()" in the |
1088 | * waiter, and has to be the absolute last thing we do. | |
1089 | * After this list_del_init(&wait->entry) the wait entry | |
2a9127fc LT |
1090 | * might be de-allocated and the process might even have |
1091 | * exited. | |
2a9127fc | 1092 | */ |
c6fe44d9 | 1093 | list_del_init_careful(&wait->entry); |
5ef64cc8 | 1094 | return (flags & WQ_FLAG_EXCLUSIVE) != 0; |
f62e00cc KM |
1095 | } |
1096 | ||
74d81bfa | 1097 | static void wake_up_page_bit(struct page *page, int bit_nr) |
cbbce822 | 1098 | { |
62906027 NP |
1099 | wait_queue_head_t *q = page_waitqueue(page); |
1100 | struct wait_page_key key; | |
1101 | unsigned long flags; | |
11a19c7b | 1102 | wait_queue_entry_t bookmark; |
cbbce822 | 1103 | |
62906027 NP |
1104 | key.page = page; |
1105 | key.bit_nr = bit_nr; | |
1106 | key.page_match = 0; | |
1107 | ||
11a19c7b TC |
1108 | bookmark.flags = 0; |
1109 | bookmark.private = NULL; | |
1110 | bookmark.func = NULL; | |
1111 | INIT_LIST_HEAD(&bookmark.entry); | |
1112 | ||
62906027 | 1113 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1114 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1115 | ||
1116 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1117 | /* | |
1118 | * Take a breather from holding the lock, | |
1119 | * allow pages that finish wake up asynchronously | |
1120 | * to acquire the lock and remove themselves | |
1121 | * from wait queue | |
1122 | */ | |
1123 | spin_unlock_irqrestore(&q->lock, flags); | |
1124 | cpu_relax(); | |
1125 | spin_lock_irqsave(&q->lock, flags); | |
1126 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1127 | } | |
1128 | ||
62906027 NP |
1129 | /* |
1130 | * It is possible for other pages to have collided on the waitqueue | |
1131 | * hash, so in that case check for a page match. That prevents a long- | |
1132 | * term waiter | |
1133 | * | |
1134 | * It is still possible to miss a case here, when we woke page waiters | |
1135 | * and removed them from the waitqueue, but there are still other | |
1136 | * page waiters. | |
1137 | */ | |
1138 | if (!waitqueue_active(q) || !key.page_match) { | |
1139 | ClearPageWaiters(page); | |
1140 | /* | |
1141 | * It's possible to miss clearing Waiters here, when we woke | |
1142 | * our page waiters, but the hashed waitqueue has waiters for | |
1143 | * other pages on it. | |
1144 | * | |
1145 | * That's okay, it's a rare case. The next waker will clear it. | |
1146 | */ | |
1147 | } | |
1148 | spin_unlock_irqrestore(&q->lock, flags); | |
1149 | } | |
74d81bfa NP |
1150 | |
1151 | static void wake_up_page(struct page *page, int bit) | |
1152 | { | |
1153 | if (!PageWaiters(page)) | |
1154 | return; | |
1155 | wake_up_page_bit(page, bit); | |
1156 | } | |
62906027 | 1157 | |
9a1ea439 HD |
1158 | /* |
1159 | * A choice of three behaviors for wait_on_page_bit_common(): | |
1160 | */ | |
1161 | enum behavior { | |
1162 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like | |
1163 | * __lock_page() waiting on then setting PG_locked. | |
1164 | */ | |
1165 | SHARED, /* Hold ref to page and check the bit when woken, like | |
1166 | * wait_on_page_writeback() waiting on PG_writeback. | |
1167 | */ | |
1168 | DROP, /* Drop ref to page before wait, no check when woken, | |
1169 | * like put_and_wait_on_page_locked() on PG_locked. | |
1170 | */ | |
1171 | }; | |
1172 | ||
2a9127fc | 1173 | /* |
5ef64cc8 LT |
1174 | * Attempt to check (or get) the page bit, and mark us done |
1175 | * if successful. | |
2a9127fc LT |
1176 | */ |
1177 | static inline bool trylock_page_bit_common(struct page *page, int bit_nr, | |
1178 | struct wait_queue_entry *wait) | |
1179 | { | |
1180 | if (wait->flags & WQ_FLAG_EXCLUSIVE) { | |
1181 | if (test_and_set_bit(bit_nr, &page->flags)) | |
1182 | return false; | |
1183 | } else if (test_bit(bit_nr, &page->flags)) | |
1184 | return false; | |
1185 | ||
5ef64cc8 | 1186 | wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE; |
2a9127fc LT |
1187 | return true; |
1188 | } | |
1189 | ||
5ef64cc8 LT |
1190 | /* How many times do we accept lock stealing from under a waiter? */ |
1191 | int sysctl_page_lock_unfairness = 5; | |
1192 | ||
62906027 | 1193 | static inline int wait_on_page_bit_common(wait_queue_head_t *q, |
9a1ea439 | 1194 | struct page *page, int bit_nr, int state, enum behavior behavior) |
62906027 | 1195 | { |
5ef64cc8 | 1196 | int unfairness = sysctl_page_lock_unfairness; |
62906027 | 1197 | struct wait_page_queue wait_page; |
ac6424b9 | 1198 | wait_queue_entry_t *wait = &wait_page.wait; |
b1d29ba8 | 1199 | bool thrashing = false; |
9a1ea439 | 1200 | bool delayacct = false; |
eb414681 | 1201 | unsigned long pflags; |
62906027 | 1202 | |
eb414681 | 1203 | if (bit_nr == PG_locked && |
b1d29ba8 | 1204 | !PageUptodate(page) && PageWorkingset(page)) { |
9a1ea439 | 1205 | if (!PageSwapBacked(page)) { |
eb414681 | 1206 | delayacct_thrashing_start(); |
9a1ea439 HD |
1207 | delayacct = true; |
1208 | } | |
eb414681 | 1209 | psi_memstall_enter(&pflags); |
b1d29ba8 JW |
1210 | thrashing = true; |
1211 | } | |
1212 | ||
62906027 NP |
1213 | init_wait(wait); |
1214 | wait->func = wake_page_function; | |
1215 | wait_page.page = page; | |
1216 | wait_page.bit_nr = bit_nr; | |
1217 | ||
5ef64cc8 LT |
1218 | repeat: |
1219 | wait->flags = 0; | |
1220 | if (behavior == EXCLUSIVE) { | |
1221 | wait->flags = WQ_FLAG_EXCLUSIVE; | |
1222 | if (--unfairness < 0) | |
1223 | wait->flags |= WQ_FLAG_CUSTOM; | |
1224 | } | |
1225 | ||
2a9127fc LT |
1226 | /* |
1227 | * Do one last check whether we can get the | |
1228 | * page bit synchronously. | |
1229 | * | |
1230 | * Do the SetPageWaiters() marking before that | |
1231 | * to let any waker we _just_ missed know they | |
1232 | * need to wake us up (otherwise they'll never | |
1233 | * even go to the slow case that looks at the | |
1234 | * page queue), and add ourselves to the wait | |
1235 | * queue if we need to sleep. | |
1236 | * | |
1237 | * This part needs to be done under the queue | |
1238 | * lock to avoid races. | |
1239 | */ | |
1240 | spin_lock_irq(&q->lock); | |
1241 | SetPageWaiters(page); | |
1242 | if (!trylock_page_bit_common(page, bit_nr, wait)) | |
1243 | __add_wait_queue_entry_tail(q, wait); | |
1244 | spin_unlock_irq(&q->lock); | |
62906027 | 1245 | |
2a9127fc LT |
1246 | /* |
1247 | * From now on, all the logic will be based on | |
5ef64cc8 LT |
1248 | * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to |
1249 | * see whether the page bit testing has already | |
1250 | * been done by the wake function. | |
2a9127fc LT |
1251 | * |
1252 | * We can drop our reference to the page. | |
1253 | */ | |
1254 | if (behavior == DROP) | |
1255 | put_page(page); | |
62906027 | 1256 | |
5ef64cc8 LT |
1257 | /* |
1258 | * Note that until the "finish_wait()", or until | |
1259 | * we see the WQ_FLAG_WOKEN flag, we need to | |
1260 | * be very careful with the 'wait->flags', because | |
1261 | * we may race with a waker that sets them. | |
1262 | */ | |
2a9127fc | 1263 | for (;;) { |
5ef64cc8 LT |
1264 | unsigned int flags; |
1265 | ||
62906027 NP |
1266 | set_current_state(state); |
1267 | ||
5ef64cc8 LT |
1268 | /* Loop until we've been woken or interrupted */ |
1269 | flags = smp_load_acquire(&wait->flags); | |
1270 | if (!(flags & WQ_FLAG_WOKEN)) { | |
1271 | if (signal_pending_state(state, current)) | |
1272 | break; | |
1273 | ||
1274 | io_schedule(); | |
1275 | continue; | |
1276 | } | |
1277 | ||
1278 | /* If we were non-exclusive, we're done */ | |
1279 | if (behavior != EXCLUSIVE) | |
a8b169af | 1280 | break; |
9a1ea439 | 1281 | |
5ef64cc8 LT |
1282 | /* If the waker got the lock for us, we're done */ |
1283 | if (flags & WQ_FLAG_DONE) | |
9a1ea439 | 1284 | break; |
2a9127fc | 1285 | |
5ef64cc8 LT |
1286 | /* |
1287 | * Otherwise, if we're getting the lock, we need to | |
1288 | * try to get it ourselves. | |
1289 | * | |
1290 | * And if that fails, we'll have to retry this all. | |
1291 | */ | |
1292 | if (unlikely(test_and_set_bit(bit_nr, &page->flags))) | |
1293 | goto repeat; | |
1294 | ||
1295 | wait->flags |= WQ_FLAG_DONE; | |
1296 | break; | |
62906027 NP |
1297 | } |
1298 | ||
5ef64cc8 LT |
1299 | /* |
1300 | * If a signal happened, this 'finish_wait()' may remove the last | |
1301 | * waiter from the wait-queues, but the PageWaiters bit will remain | |
1302 | * set. That's ok. The next wakeup will take care of it, and trying | |
1303 | * to do it here would be difficult and prone to races. | |
1304 | */ | |
62906027 NP |
1305 | finish_wait(q, wait); |
1306 | ||
eb414681 | 1307 | if (thrashing) { |
9a1ea439 | 1308 | if (delayacct) |
eb414681 JW |
1309 | delayacct_thrashing_end(); |
1310 | psi_memstall_leave(&pflags); | |
1311 | } | |
b1d29ba8 | 1312 | |
62906027 | 1313 | /* |
5ef64cc8 LT |
1314 | * NOTE! The wait->flags weren't stable until we've done the |
1315 | * 'finish_wait()', and we could have exited the loop above due | |
1316 | * to a signal, and had a wakeup event happen after the signal | |
1317 | * test but before the 'finish_wait()'. | |
1318 | * | |
1319 | * So only after the finish_wait() can we reliably determine | |
1320 | * if we got woken up or not, so we can now figure out the final | |
1321 | * return value based on that state without races. | |
1322 | * | |
1323 | * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive | |
1324 | * waiter, but an exclusive one requires WQ_FLAG_DONE. | |
62906027 | 1325 | */ |
5ef64cc8 LT |
1326 | if (behavior == EXCLUSIVE) |
1327 | return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR; | |
62906027 | 1328 | |
2a9127fc | 1329 | return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR; |
62906027 NP |
1330 | } |
1331 | ||
1332 | void wait_on_page_bit(struct page *page, int bit_nr) | |
1333 | { | |
1334 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 | 1335 | wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
62906027 NP |
1336 | } |
1337 | EXPORT_SYMBOL(wait_on_page_bit); | |
1338 | ||
1339 | int wait_on_page_bit_killable(struct page *page, int bit_nr) | |
1340 | { | |
1341 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 | 1342 | return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED); |
cbbce822 | 1343 | } |
4343d008 | 1344 | EXPORT_SYMBOL(wait_on_page_bit_killable); |
cbbce822 | 1345 | |
dd3e6d50 JA |
1346 | static int __wait_on_page_locked_async(struct page *page, |
1347 | struct wait_page_queue *wait, bool set) | |
1348 | { | |
1349 | struct wait_queue_head *q = page_waitqueue(page); | |
1350 | int ret = 0; | |
1351 | ||
1352 | wait->page = page; | |
1353 | wait->bit_nr = PG_locked; | |
1354 | ||
1355 | spin_lock_irq(&q->lock); | |
1356 | __add_wait_queue_entry_tail(q, &wait->wait); | |
1357 | SetPageWaiters(page); | |
1358 | if (set) | |
1359 | ret = !trylock_page(page); | |
1360 | else | |
1361 | ret = PageLocked(page); | |
1362 | /* | |
8958b249 | 1363 | * If we were successful now, we know we're still on the |
dd3e6d50 JA |
1364 | * waitqueue as we're still under the lock. This means it's |
1365 | * safe to remove and return success, we know the callback | |
1366 | * isn't going to trigger. | |
1367 | */ | |
1368 | if (!ret) | |
1369 | __remove_wait_queue(q, &wait->wait); | |
1370 | else | |
1371 | ret = -EIOCBQUEUED; | |
1372 | spin_unlock_irq(&q->lock); | |
1373 | return ret; | |
1374 | } | |
1375 | ||
1a0a7853 JA |
1376 | static int wait_on_page_locked_async(struct page *page, |
1377 | struct wait_page_queue *wait) | |
1378 | { | |
1379 | if (!PageLocked(page)) | |
1380 | return 0; | |
1381 | return __wait_on_page_locked_async(compound_head(page), wait, false); | |
1382 | } | |
1383 | ||
9a1ea439 HD |
1384 | /** |
1385 | * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked | |
1386 | * @page: The page to wait for. | |
1387 | * | |
1388 | * The caller should hold a reference on @page. They expect the page to | |
1389 | * become unlocked relatively soon, but do not wish to hold up migration | |
1390 | * (for example) by holding the reference while waiting for the page to | |
1391 | * come unlocked. After this function returns, the caller should not | |
1392 | * dereference @page. | |
1393 | */ | |
1394 | void put_and_wait_on_page_locked(struct page *page) | |
1395 | { | |
1396 | wait_queue_head_t *q; | |
1397 | ||
1398 | page = compound_head(page); | |
1399 | q = page_waitqueue(page); | |
1400 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP); | |
1401 | } | |
1402 | ||
385e1ca5 DH |
1403 | /** |
1404 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
1405 | * @page: Page defining the wait queue of interest |
1406 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
1407 | * |
1408 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
1409 | */ | |
ac6424b9 | 1410 | void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter) |
385e1ca5 DH |
1411 | { |
1412 | wait_queue_head_t *q = page_waitqueue(page); | |
1413 | unsigned long flags; | |
1414 | ||
1415 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1416 | __add_wait_queue_entry_tail(q, waiter); |
62906027 | 1417 | SetPageWaiters(page); |
385e1ca5 DH |
1418 | spin_unlock_irqrestore(&q->lock, flags); |
1419 | } | |
1420 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
1421 | ||
b91e1302 LT |
1422 | #ifndef clear_bit_unlock_is_negative_byte |
1423 | ||
1424 | /* | |
1425 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1426 | * | |
1427 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1428 | * test the sign bit at the same time. But if the architecture does | |
1429 | * not support that special operation, we just do this all by hand | |
1430 | * instead. | |
1431 | * | |
1432 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
ffceeb62 | 1433 | * being cleared, but a memory barrier should be unnecessary since it is |
b91e1302 LT |
1434 | * in the same byte as PG_locked. |
1435 | */ | |
1436 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1437 | { | |
1438 | clear_bit_unlock(nr, mem); | |
1439 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1440 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1441 | } |
1442 | ||
1443 | #endif | |
1444 | ||
1da177e4 | 1445 | /** |
485bb99b | 1446 | * unlock_page - unlock a locked page |
1da177e4 LT |
1447 | * @page: the page |
1448 | * | |
0e9aa675 | 1449 | * Unlocks the page and wakes up sleepers in wait_on_page_locked(). |
1da177e4 | 1450 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup |
da3dae54 | 1451 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
1452 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
1453 | * | |
b91e1302 LT |
1454 | * Note that this depends on PG_waiters being the sign bit in the byte |
1455 | * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to | |
1456 | * clear the PG_locked bit and test PG_waiters at the same time fairly | |
1457 | * portably (architectures that do LL/SC can test any bit, while x86 can | |
1458 | * test the sign bit). | |
1da177e4 | 1459 | */ |
920c7a5d | 1460 | void unlock_page(struct page *page) |
1da177e4 | 1461 | { |
b91e1302 | 1462 | BUILD_BUG_ON(PG_waiters != 7); |
48c935ad | 1463 | page = compound_head(page); |
309381fe | 1464 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
b91e1302 LT |
1465 | if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags)) |
1466 | wake_up_page_bit(page, PG_locked); | |
1da177e4 LT |
1467 | } |
1468 | EXPORT_SYMBOL(unlock_page); | |
1469 | ||
485bb99b RD |
1470 | /** |
1471 | * end_page_writeback - end writeback against a page | |
1472 | * @page: the page | |
1da177e4 LT |
1473 | */ |
1474 | void end_page_writeback(struct page *page) | |
1475 | { | |
888cf2db MG |
1476 | /* |
1477 | * TestClearPageReclaim could be used here but it is an atomic | |
1478 | * operation and overkill in this particular case. Failing to | |
1479 | * shuffle a page marked for immediate reclaim is too mild to | |
1480 | * justify taking an atomic operation penalty at the end of | |
1481 | * ever page writeback. | |
1482 | */ | |
1483 | if (PageReclaim(page)) { | |
1484 | ClearPageReclaim(page); | |
ac6aadb2 | 1485 | rotate_reclaimable_page(page); |
888cf2db | 1486 | } |
ac6aadb2 | 1487 | |
073861ed HD |
1488 | /* |
1489 | * Writeback does not hold a page reference of its own, relying | |
1490 | * on truncation to wait for the clearing of PG_writeback. | |
1491 | * But here we must make sure that the page is not freed and | |
1492 | * reused before the wake_up_page(). | |
1493 | */ | |
1494 | get_page(page); | |
ac6aadb2 MS |
1495 | if (!test_clear_page_writeback(page)) |
1496 | BUG(); | |
1497 | ||
4e857c58 | 1498 | smp_mb__after_atomic(); |
1da177e4 | 1499 | wake_up_page(page, PG_writeback); |
073861ed | 1500 | put_page(page); |
1da177e4 LT |
1501 | } |
1502 | EXPORT_SYMBOL(end_page_writeback); | |
1503 | ||
57d99845 MW |
1504 | /* |
1505 | * After completing I/O on a page, call this routine to update the page | |
1506 | * flags appropriately | |
1507 | */ | |
c11f0c0b | 1508 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1509 | { |
c11f0c0b | 1510 | if (!is_write) { |
57d99845 MW |
1511 | if (!err) { |
1512 | SetPageUptodate(page); | |
1513 | } else { | |
1514 | ClearPageUptodate(page); | |
1515 | SetPageError(page); | |
1516 | } | |
1517 | unlock_page(page); | |
abf54548 | 1518 | } else { |
57d99845 | 1519 | if (err) { |
dd8416c4 MK |
1520 | struct address_space *mapping; |
1521 | ||
57d99845 | 1522 | SetPageError(page); |
dd8416c4 MK |
1523 | mapping = page_mapping(page); |
1524 | if (mapping) | |
1525 | mapping_set_error(mapping, err); | |
57d99845 MW |
1526 | } |
1527 | end_page_writeback(page); | |
1528 | } | |
1529 | } | |
1530 | EXPORT_SYMBOL_GPL(page_endio); | |
1531 | ||
485bb99b RD |
1532 | /** |
1533 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
87066755 | 1534 | * @__page: the page to lock |
1da177e4 | 1535 | */ |
62906027 | 1536 | void __lock_page(struct page *__page) |
1da177e4 | 1537 | { |
62906027 NP |
1538 | struct page *page = compound_head(__page); |
1539 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 HD |
1540 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, |
1541 | EXCLUSIVE); | |
1da177e4 LT |
1542 | } |
1543 | EXPORT_SYMBOL(__lock_page); | |
1544 | ||
62906027 | 1545 | int __lock_page_killable(struct page *__page) |
2687a356 | 1546 | { |
62906027 NP |
1547 | struct page *page = compound_head(__page); |
1548 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 HD |
1549 | return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, |
1550 | EXCLUSIVE); | |
2687a356 | 1551 | } |
18bc0bbd | 1552 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 1553 | |
dd3e6d50 JA |
1554 | int __lock_page_async(struct page *page, struct wait_page_queue *wait) |
1555 | { | |
1556 | return __wait_on_page_locked_async(page, wait, true); | |
1557 | } | |
1558 | ||
9a95f3cf PC |
1559 | /* |
1560 | * Return values: | |
c1e8d7c6 | 1561 | * 1 - page is locked; mmap_lock is still held. |
9a95f3cf | 1562 | * 0 - page is not locked. |
3e4e28c5 | 1563 | * mmap_lock has been released (mmap_read_unlock(), unless flags had both |
9a95f3cf | 1564 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
c1e8d7c6 | 1565 | * which case mmap_lock is still held. |
9a95f3cf PC |
1566 | * |
1567 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
c1e8d7c6 | 1568 | * with the page locked and the mmap_lock unperturbed. |
9a95f3cf | 1569 | */ |
d065bd81 ML |
1570 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
1571 | unsigned int flags) | |
1572 | { | |
4064b982 | 1573 | if (fault_flag_allow_retry_first(flags)) { |
37b23e05 | 1574 | /* |
c1e8d7c6 | 1575 | * CAUTION! In this case, mmap_lock is not released |
37b23e05 KM |
1576 | * even though return 0. |
1577 | */ | |
1578 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
1579 | return 0; | |
1580 | ||
d8ed45c5 | 1581 | mmap_read_unlock(mm); |
37b23e05 KM |
1582 | if (flags & FAULT_FLAG_KILLABLE) |
1583 | wait_on_page_locked_killable(page); | |
1584 | else | |
318b275f | 1585 | wait_on_page_locked(page); |
d065bd81 | 1586 | return 0; |
800bca7c HL |
1587 | } |
1588 | if (flags & FAULT_FLAG_KILLABLE) { | |
1589 | int ret; | |
37b23e05 | 1590 | |
800bca7c HL |
1591 | ret = __lock_page_killable(page); |
1592 | if (ret) { | |
1593 | mmap_read_unlock(mm); | |
1594 | return 0; | |
1595 | } | |
1596 | } else { | |
1597 | __lock_page(page); | |
d065bd81 | 1598 | } |
800bca7c HL |
1599 | return 1; |
1600 | ||
d065bd81 ML |
1601 | } |
1602 | ||
e7b563bb | 1603 | /** |
0d3f9296 MW |
1604 | * page_cache_next_miss() - Find the next gap in the page cache. |
1605 | * @mapping: Mapping. | |
1606 | * @index: Index. | |
1607 | * @max_scan: Maximum range to search. | |
e7b563bb | 1608 | * |
0d3f9296 MW |
1609 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1610 | * gap with the lowest index. | |
e7b563bb | 1611 | * |
0d3f9296 MW |
1612 | * This function may be called under the rcu_read_lock. However, this will |
1613 | * not atomically search a snapshot of the cache at a single point in time. | |
1614 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1615 | * created at index 10, page_cache_next_miss covering both indices may | |
1616 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1617 | * |
0d3f9296 MW |
1618 | * Return: The index of the gap if found, otherwise an index outside the |
1619 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1620 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1621 | */ |
0d3f9296 | 1622 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1623 | pgoff_t index, unsigned long max_scan) |
1624 | { | |
0d3f9296 | 1625 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1626 | |
0d3f9296 MW |
1627 | while (max_scan--) { |
1628 | void *entry = xas_next(&xas); | |
1629 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1630 | break; |
0d3f9296 | 1631 | if (xas.xa_index == 0) |
e7b563bb JW |
1632 | break; |
1633 | } | |
1634 | ||
0d3f9296 | 1635 | return xas.xa_index; |
e7b563bb | 1636 | } |
0d3f9296 | 1637 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1638 | |
1639 | /** | |
2346a560 | 1640 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
0d3f9296 MW |
1641 | * @mapping: Mapping. |
1642 | * @index: Index. | |
1643 | * @max_scan: Maximum range to search. | |
e7b563bb | 1644 | * |
0d3f9296 MW |
1645 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1646 | * gap with the highest index. | |
e7b563bb | 1647 | * |
0d3f9296 MW |
1648 | * This function may be called under the rcu_read_lock. However, this will |
1649 | * not atomically search a snapshot of the cache at a single point in time. | |
1650 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1651 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1652 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1653 | * |
0d3f9296 MW |
1654 | * Return: The index of the gap if found, otherwise an index outside the |
1655 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1656 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1657 | */ |
0d3f9296 | 1658 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1659 | pgoff_t index, unsigned long max_scan) |
1660 | { | |
0d3f9296 | 1661 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1662 | |
0d3f9296 MW |
1663 | while (max_scan--) { |
1664 | void *entry = xas_prev(&xas); | |
1665 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1666 | break; |
0d3f9296 | 1667 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1668 | break; |
1669 | } | |
1670 | ||
0d3f9296 | 1671 | return xas.xa_index; |
e7b563bb | 1672 | } |
0d3f9296 | 1673 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1674 | |
485bb99b | 1675 | /** |
0cd6144a | 1676 | * find_get_entry - find and get a page cache entry |
485bb99b | 1677 | * @mapping: the address_space to search |
a6de4b48 | 1678 | * @index: The page cache index. |
0cd6144a JW |
1679 | * |
1680 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
a6de4b48 | 1681 | * page cache page, the head page is returned with an increased refcount. |
485bb99b | 1682 | * |
139b6a6f JW |
1683 | * If the slot holds a shadow entry of a previously evicted page, or a |
1684 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a | 1685 | * |
a6de4b48 | 1686 | * Return: The head page or shadow entry, %NULL if nothing is found. |
1da177e4 | 1687 | */ |
a6de4b48 | 1688 | struct page *find_get_entry(struct address_space *mapping, pgoff_t index) |
1da177e4 | 1689 | { |
a6de4b48 | 1690 | XA_STATE(xas, &mapping->i_pages, index); |
4101196b | 1691 | struct page *page; |
1da177e4 | 1692 | |
a60637c8 NP |
1693 | rcu_read_lock(); |
1694 | repeat: | |
4c7472c0 MW |
1695 | xas_reset(&xas); |
1696 | page = xas_load(&xas); | |
1697 | if (xas_retry(&xas, page)) | |
1698 | goto repeat; | |
1699 | /* | |
1700 | * A shadow entry of a recently evicted page, or a swap entry from | |
1701 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1702 | */ | |
1703 | if (!page || xa_is_value(page)) | |
1704 | goto out; | |
83929372 | 1705 | |
4101196b | 1706 | if (!page_cache_get_speculative(page)) |
4c7472c0 | 1707 | goto repeat; |
83929372 | 1708 | |
4c7472c0 | 1709 | /* |
4101196b | 1710 | * Has the page moved or been split? |
4c7472c0 MW |
1711 | * This is part of the lockless pagecache protocol. See |
1712 | * include/linux/pagemap.h for details. | |
1713 | */ | |
1714 | if (unlikely(page != xas_reload(&xas))) { | |
4101196b | 1715 | put_page(page); |
4c7472c0 | 1716 | goto repeat; |
a60637c8 | 1717 | } |
27d20fdd | 1718 | out: |
a60637c8 NP |
1719 | rcu_read_unlock(); |
1720 | ||
1da177e4 LT |
1721 | return page; |
1722 | } | |
1da177e4 | 1723 | |
0cd6144a | 1724 | /** |
63ec1973 MWO |
1725 | * find_lock_entry - Locate and lock a page cache entry. |
1726 | * @mapping: The address_space to search. | |
1727 | * @index: The page cache index. | |
0cd6144a | 1728 | * |
63ec1973 MWO |
1729 | * Looks up the page at @mapping & @index. If there is a page in the |
1730 | * cache, the head page is returned locked and with an increased refcount. | |
0cd6144a | 1731 | * |
139b6a6f JW |
1732 | * If the slot holds a shadow entry of a previously evicted page, or a |
1733 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a | 1734 | * |
63ec1973 MWO |
1735 | * Context: May sleep. |
1736 | * Return: The head page or shadow entry, %NULL if nothing is found. | |
0cd6144a | 1737 | */ |
63ec1973 | 1738 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t index) |
1da177e4 LT |
1739 | { |
1740 | struct page *page; | |
1741 | ||
1da177e4 | 1742 | repeat: |
63ec1973 | 1743 | page = find_get_entry(mapping, index); |
4c7472c0 | 1744 | if (page && !xa_is_value(page)) { |
a60637c8 NP |
1745 | lock_page(page); |
1746 | /* Has the page been truncated? */ | |
63ec1973 | 1747 | if (unlikely(page->mapping != mapping)) { |
a60637c8 | 1748 | unlock_page(page); |
09cbfeaf | 1749 | put_page(page); |
a60637c8 | 1750 | goto repeat; |
1da177e4 | 1751 | } |
63ec1973 | 1752 | VM_BUG_ON_PAGE(!thp_contains(page, index), page); |
1da177e4 | 1753 | } |
1da177e4 LT |
1754 | return page; |
1755 | } | |
0cd6144a JW |
1756 | |
1757 | /** | |
2294b32e MWO |
1758 | * pagecache_get_page - Find and get a reference to a page. |
1759 | * @mapping: The address_space to search. | |
1760 | * @index: The page index. | |
1761 | * @fgp_flags: %FGP flags modify how the page is returned. | |
1762 | * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified. | |
1da177e4 | 1763 | * |
2294b32e | 1764 | * Looks up the page cache entry at @mapping & @index. |
0cd6144a | 1765 | * |
2294b32e | 1766 | * @fgp_flags can be zero or more of these flags: |
0e056eb5 | 1767 | * |
2294b32e MWO |
1768 | * * %FGP_ACCESSED - The page will be marked accessed. |
1769 | * * %FGP_LOCK - The page is returned locked. | |
a8cf7f27 MWO |
1770 | * * %FGP_HEAD - If the page is present and a THP, return the head page |
1771 | * rather than the exact page specified by the index. | |
2294b32e MWO |
1772 | * * %FGP_CREAT - If no page is present then a new page is allocated using |
1773 | * @gfp_mask and added to the page cache and the VM's LRU list. | |
1774 | * The page is returned locked and with an increased refcount. | |
1775 | * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the | |
1776 | * page is already in cache. If the page was allocated, unlock it before | |
1777 | * returning so the caller can do the same dance. | |
605cad83 YS |
1778 | * * %FGP_WRITE - The page will be written |
1779 | * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask | |
1780 | * * %FGP_NOWAIT - Don't get blocked by page lock | |
1da177e4 | 1781 | * |
2294b32e MWO |
1782 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even |
1783 | * if the %GFP flags specified for %FGP_CREAT are atomic. | |
1da177e4 | 1784 | * |
2457aec6 | 1785 | * If there is a page cache page, it is returned with an increased refcount. |
a862f68a | 1786 | * |
2294b32e | 1787 | * Return: The found page or %NULL otherwise. |
1da177e4 | 1788 | */ |
2294b32e MWO |
1789 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index, |
1790 | int fgp_flags, gfp_t gfp_mask) | |
1da177e4 | 1791 | { |
eb2be189 | 1792 | struct page *page; |
2457aec6 | 1793 | |
1da177e4 | 1794 | repeat: |
2294b32e | 1795 | page = find_get_entry(mapping, index); |
3159f943 | 1796 | if (xa_is_value(page)) |
2457aec6 MG |
1797 | page = NULL; |
1798 | if (!page) | |
1799 | goto no_page; | |
1800 | ||
1801 | if (fgp_flags & FGP_LOCK) { | |
1802 | if (fgp_flags & FGP_NOWAIT) { | |
1803 | if (!trylock_page(page)) { | |
09cbfeaf | 1804 | put_page(page); |
2457aec6 MG |
1805 | return NULL; |
1806 | } | |
1807 | } else { | |
1808 | lock_page(page); | |
1809 | } | |
1810 | ||
1811 | /* Has the page been truncated? */ | |
a8cf7f27 | 1812 | if (unlikely(page->mapping != mapping)) { |
2457aec6 | 1813 | unlock_page(page); |
09cbfeaf | 1814 | put_page(page); |
2457aec6 MG |
1815 | goto repeat; |
1816 | } | |
a8cf7f27 | 1817 | VM_BUG_ON_PAGE(!thp_contains(page, index), page); |
2457aec6 MG |
1818 | } |
1819 | ||
c16eb000 | 1820 | if (fgp_flags & FGP_ACCESSED) |
2457aec6 | 1821 | mark_page_accessed(page); |
b9306a79 YS |
1822 | else if (fgp_flags & FGP_WRITE) { |
1823 | /* Clear idle flag for buffer write */ | |
1824 | if (page_is_idle(page)) | |
1825 | clear_page_idle(page); | |
1826 | } | |
a8cf7f27 MWO |
1827 | if (!(fgp_flags & FGP_HEAD)) |
1828 | page = find_subpage(page, index); | |
2457aec6 MG |
1829 | |
1830 | no_page: | |
1831 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1832 | int err; | |
f56753ac | 1833 | if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping)) |
45f87de5 MH |
1834 | gfp_mask |= __GFP_WRITE; |
1835 | if (fgp_flags & FGP_NOFS) | |
1836 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1837 | |
45f87de5 | 1838 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1839 | if (!page) |
1840 | return NULL; | |
2457aec6 | 1841 | |
a75d4c33 | 1842 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
2457aec6 MG |
1843 | fgp_flags |= FGP_LOCK; |
1844 | ||
eb39d618 | 1845 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1846 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1847 | __SetPageReferenced(page); |
2457aec6 | 1848 | |
2294b32e | 1849 | err = add_to_page_cache_lru(page, mapping, index, gfp_mask); |
eb2be189 | 1850 | if (unlikely(err)) { |
09cbfeaf | 1851 | put_page(page); |
eb2be189 NP |
1852 | page = NULL; |
1853 | if (err == -EEXIST) | |
1854 | goto repeat; | |
1da177e4 | 1855 | } |
a75d4c33 JB |
1856 | |
1857 | /* | |
1858 | * add_to_page_cache_lru locks the page, and for mmap we expect | |
1859 | * an unlocked page. | |
1860 | */ | |
1861 | if (page && (fgp_flags & FGP_FOR_MMAP)) | |
1862 | unlock_page(page); | |
1da177e4 | 1863 | } |
2457aec6 | 1864 | |
1da177e4 LT |
1865 | return page; |
1866 | } | |
2457aec6 | 1867 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1868 | |
0cd6144a JW |
1869 | /** |
1870 | * find_get_entries - gang pagecache lookup | |
1871 | * @mapping: The address_space to search | |
1872 | * @start: The starting page cache index | |
1873 | * @nr_entries: The maximum number of entries | |
1874 | * @entries: Where the resulting entries are placed | |
1875 | * @indices: The cache indices corresponding to the entries in @entries | |
1876 | * | |
1877 | * find_get_entries() will search for and return a group of up to | |
1878 | * @nr_entries entries in the mapping. The entries are placed at | |
1879 | * @entries. find_get_entries() takes a reference against any actual | |
1880 | * pages it returns. | |
1881 | * | |
1882 | * The search returns a group of mapping-contiguous page cache entries | |
1883 | * with ascending indexes. There may be holes in the indices due to | |
1884 | * not-present pages. | |
1885 | * | |
139b6a6f JW |
1886 | * Any shadow entries of evicted pages, or swap entries from |
1887 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a | 1888 | * |
71725ed1 HD |
1889 | * If it finds a Transparent Huge Page, head or tail, find_get_entries() |
1890 | * stops at that page: the caller is likely to have a better way to handle | |
1891 | * the compound page as a whole, and then skip its extent, than repeatedly | |
1892 | * calling find_get_entries() to return all its tails. | |
1893 | * | |
a862f68a | 1894 | * Return: the number of pages and shadow entries which were found. |
0cd6144a JW |
1895 | */ |
1896 | unsigned find_get_entries(struct address_space *mapping, | |
1897 | pgoff_t start, unsigned int nr_entries, | |
1898 | struct page **entries, pgoff_t *indices) | |
1899 | { | |
f280bf09 MW |
1900 | XA_STATE(xas, &mapping->i_pages, start); |
1901 | struct page *page; | |
0cd6144a | 1902 | unsigned int ret = 0; |
0cd6144a JW |
1903 | |
1904 | if (!nr_entries) | |
1905 | return 0; | |
1906 | ||
1907 | rcu_read_lock(); | |
f280bf09 | 1908 | xas_for_each(&xas, page, ULONG_MAX) { |
f280bf09 | 1909 | if (xas_retry(&xas, page)) |
0cd6144a | 1910 | continue; |
f280bf09 MW |
1911 | /* |
1912 | * A shadow entry of a recently evicted page, a swap | |
1913 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1914 | * without attempting to raise page count. | |
1915 | */ | |
1916 | if (xa_is_value(page)) | |
0cd6144a | 1917 | goto export; |
83929372 | 1918 | |
4101196b | 1919 | if (!page_cache_get_speculative(page)) |
f280bf09 | 1920 | goto retry; |
83929372 | 1921 | |
4101196b | 1922 | /* Has the page moved or been split? */ |
f280bf09 MW |
1923 | if (unlikely(page != xas_reload(&xas))) |
1924 | goto put_page; | |
1925 | ||
71725ed1 HD |
1926 | /* |
1927 | * Terminate early on finding a THP, to allow the caller to | |
1928 | * handle it all at once; but continue if this is hugetlbfs. | |
1929 | */ | |
1930 | if (PageTransHuge(page) && !PageHuge(page)) { | |
1931 | page = find_subpage(page, xas.xa_index); | |
1932 | nr_entries = ret + 1; | |
1933 | } | |
0cd6144a | 1934 | export: |
f280bf09 | 1935 | indices[ret] = xas.xa_index; |
0cd6144a JW |
1936 | entries[ret] = page; |
1937 | if (++ret == nr_entries) | |
1938 | break; | |
f280bf09 MW |
1939 | continue; |
1940 | put_page: | |
4101196b | 1941 | put_page(page); |
f280bf09 MW |
1942 | retry: |
1943 | xas_reset(&xas); | |
0cd6144a JW |
1944 | } |
1945 | rcu_read_unlock(); | |
1946 | return ret; | |
1947 | } | |
1948 | ||
1da177e4 | 1949 | /** |
b947cee4 | 1950 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
1951 | * @mapping: The address_space to search |
1952 | * @start: The starting page index | |
b947cee4 | 1953 | * @end: The final page index (inclusive) |
1da177e4 LT |
1954 | * @nr_pages: The maximum number of pages |
1955 | * @pages: Where the resulting pages are placed | |
1956 | * | |
b947cee4 JK |
1957 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
1958 | * pages in the mapping starting at index @start and up to index @end | |
1959 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
1960 | * a reference against the returned pages. | |
1da177e4 LT |
1961 | * |
1962 | * The search returns a group of mapping-contiguous pages with ascending | |
1963 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 1964 | * We also update @start to index the next page for the traversal. |
1da177e4 | 1965 | * |
a862f68a MR |
1966 | * Return: the number of pages which were found. If this number is |
1967 | * smaller than @nr_pages, the end of specified range has been | |
b947cee4 | 1968 | * reached. |
1da177e4 | 1969 | */ |
b947cee4 JK |
1970 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
1971 | pgoff_t end, unsigned int nr_pages, | |
1972 | struct page **pages) | |
1da177e4 | 1973 | { |
fd1b3cee MW |
1974 | XA_STATE(xas, &mapping->i_pages, *start); |
1975 | struct page *page; | |
0fc9d104 KK |
1976 | unsigned ret = 0; |
1977 | ||
1978 | if (unlikely(!nr_pages)) | |
1979 | return 0; | |
a60637c8 NP |
1980 | |
1981 | rcu_read_lock(); | |
fd1b3cee | 1982 | xas_for_each(&xas, page, end) { |
fd1b3cee | 1983 | if (xas_retry(&xas, page)) |
a60637c8 | 1984 | continue; |
fd1b3cee MW |
1985 | /* Skip over shadow, swap and DAX entries */ |
1986 | if (xa_is_value(page)) | |
8079b1c8 | 1987 | continue; |
a60637c8 | 1988 | |
4101196b | 1989 | if (!page_cache_get_speculative(page)) |
fd1b3cee | 1990 | goto retry; |
83929372 | 1991 | |
4101196b | 1992 | /* Has the page moved or been split? */ |
fd1b3cee MW |
1993 | if (unlikely(page != xas_reload(&xas))) |
1994 | goto put_page; | |
1da177e4 | 1995 | |
4101196b | 1996 | pages[ret] = find_subpage(page, xas.xa_index); |
b947cee4 | 1997 | if (++ret == nr_pages) { |
5d3ee42f | 1998 | *start = xas.xa_index + 1; |
b947cee4 JK |
1999 | goto out; |
2000 | } | |
fd1b3cee MW |
2001 | continue; |
2002 | put_page: | |
4101196b | 2003 | put_page(page); |
fd1b3cee MW |
2004 | retry: |
2005 | xas_reset(&xas); | |
a60637c8 | 2006 | } |
5b280c0c | 2007 | |
b947cee4 JK |
2008 | /* |
2009 | * We come here when there is no page beyond @end. We take care to not | |
2010 | * overflow the index @start as it confuses some of the callers. This | |
fd1b3cee | 2011 | * breaks the iteration when there is a page at index -1 but that is |
b947cee4 JK |
2012 | * already broken anyway. |
2013 | */ | |
2014 | if (end == (pgoff_t)-1) | |
2015 | *start = (pgoff_t)-1; | |
2016 | else | |
2017 | *start = end + 1; | |
2018 | out: | |
a60637c8 | 2019 | rcu_read_unlock(); |
d72dc8a2 | 2020 | |
1da177e4 LT |
2021 | return ret; |
2022 | } | |
2023 | ||
ebf43500 JA |
2024 | /** |
2025 | * find_get_pages_contig - gang contiguous pagecache lookup | |
2026 | * @mapping: The address_space to search | |
2027 | * @index: The starting page index | |
2028 | * @nr_pages: The maximum number of pages | |
2029 | * @pages: Where the resulting pages are placed | |
2030 | * | |
2031 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
2032 | * that the returned number of pages are guaranteed to be contiguous. | |
2033 | * | |
a862f68a | 2034 | * Return: the number of pages which were found. |
ebf43500 JA |
2035 | */ |
2036 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
2037 | unsigned int nr_pages, struct page **pages) | |
2038 | { | |
3ece58a2 MW |
2039 | XA_STATE(xas, &mapping->i_pages, index); |
2040 | struct page *page; | |
0fc9d104 KK |
2041 | unsigned int ret = 0; |
2042 | ||
2043 | if (unlikely(!nr_pages)) | |
2044 | return 0; | |
a60637c8 NP |
2045 | |
2046 | rcu_read_lock(); | |
3ece58a2 | 2047 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { |
3ece58a2 MW |
2048 | if (xas_retry(&xas, page)) |
2049 | continue; | |
2050 | /* | |
2051 | * If the entry has been swapped out, we can stop looking. | |
2052 | * No current caller is looking for DAX entries. | |
2053 | */ | |
2054 | if (xa_is_value(page)) | |
8079b1c8 | 2055 | break; |
ebf43500 | 2056 | |
4101196b | 2057 | if (!page_cache_get_speculative(page)) |
3ece58a2 | 2058 | goto retry; |
83929372 | 2059 | |
4101196b | 2060 | /* Has the page moved or been split? */ |
3ece58a2 MW |
2061 | if (unlikely(page != xas_reload(&xas))) |
2062 | goto put_page; | |
a60637c8 | 2063 | |
4101196b | 2064 | pages[ret] = find_subpage(page, xas.xa_index); |
0fc9d104 KK |
2065 | if (++ret == nr_pages) |
2066 | break; | |
3ece58a2 MW |
2067 | continue; |
2068 | put_page: | |
4101196b | 2069 | put_page(page); |
3ece58a2 MW |
2070 | retry: |
2071 | xas_reset(&xas); | |
ebf43500 | 2072 | } |
a60637c8 NP |
2073 | rcu_read_unlock(); |
2074 | return ret; | |
ebf43500 | 2075 | } |
ef71c15c | 2076 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 2077 | |
485bb99b | 2078 | /** |
72b045ae | 2079 | * find_get_pages_range_tag - find and return pages in given range matching @tag |
485bb99b RD |
2080 | * @mapping: the address_space to search |
2081 | * @index: the starting page index | |
72b045ae | 2082 | * @end: The final page index (inclusive) |
485bb99b RD |
2083 | * @tag: the tag index |
2084 | * @nr_pages: the maximum number of pages | |
2085 | * @pages: where the resulting pages are placed | |
2086 | * | |
1da177e4 | 2087 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 2088 | * @tag. We update @index to index the next page for the traversal. |
a862f68a MR |
2089 | * |
2090 | * Return: the number of pages which were found. | |
1da177e4 | 2091 | */ |
72b045ae | 2092 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
a6906972 | 2093 | pgoff_t end, xa_mark_t tag, unsigned int nr_pages, |
72b045ae | 2094 | struct page **pages) |
1da177e4 | 2095 | { |
a6906972 MW |
2096 | XA_STATE(xas, &mapping->i_pages, *index); |
2097 | struct page *page; | |
0fc9d104 KK |
2098 | unsigned ret = 0; |
2099 | ||
2100 | if (unlikely(!nr_pages)) | |
2101 | return 0; | |
a60637c8 NP |
2102 | |
2103 | rcu_read_lock(); | |
a6906972 | 2104 | xas_for_each_marked(&xas, page, end, tag) { |
a6906972 | 2105 | if (xas_retry(&xas, page)) |
a60637c8 | 2106 | continue; |
a6906972 MW |
2107 | /* |
2108 | * Shadow entries should never be tagged, but this iteration | |
2109 | * is lockless so there is a window for page reclaim to evict | |
2110 | * a page we saw tagged. Skip over it. | |
2111 | */ | |
2112 | if (xa_is_value(page)) | |
139b6a6f | 2113 | continue; |
a60637c8 | 2114 | |
4101196b | 2115 | if (!page_cache_get_speculative(page)) |
a6906972 | 2116 | goto retry; |
a60637c8 | 2117 | |
4101196b | 2118 | /* Has the page moved or been split? */ |
a6906972 MW |
2119 | if (unlikely(page != xas_reload(&xas))) |
2120 | goto put_page; | |
a60637c8 | 2121 | |
4101196b | 2122 | pages[ret] = find_subpage(page, xas.xa_index); |
72b045ae | 2123 | if (++ret == nr_pages) { |
5d3ee42f | 2124 | *index = xas.xa_index + 1; |
72b045ae JK |
2125 | goto out; |
2126 | } | |
a6906972 MW |
2127 | continue; |
2128 | put_page: | |
4101196b | 2129 | put_page(page); |
a6906972 MW |
2130 | retry: |
2131 | xas_reset(&xas); | |
a60637c8 | 2132 | } |
5b280c0c | 2133 | |
72b045ae | 2134 | /* |
a6906972 | 2135 | * We come here when we got to @end. We take care to not overflow the |
72b045ae | 2136 | * index @index as it confuses some of the callers. This breaks the |
a6906972 MW |
2137 | * iteration when there is a page at index -1 but that is already |
2138 | * broken anyway. | |
72b045ae JK |
2139 | */ |
2140 | if (end == (pgoff_t)-1) | |
2141 | *index = (pgoff_t)-1; | |
2142 | else | |
2143 | *index = end + 1; | |
2144 | out: | |
a60637c8 | 2145 | rcu_read_unlock(); |
1da177e4 | 2146 | |
1da177e4 LT |
2147 | return ret; |
2148 | } | |
72b045ae | 2149 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 2150 | |
76d42bd9 WF |
2151 | /* |
2152 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
2153 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
2154 | * | |
2155 | * ---R__________________________________________B__________ | |
2156 | * ^ reading here ^ bad block(assume 4k) | |
2157 | * | |
2158 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
2159 | * => failing the whole request => read(R) => read(R+1) => | |
2160 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
2161 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
2162 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
2163 | * | |
2164 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
2165 | */ | |
0f8e2db4 | 2166 | static void shrink_readahead_size_eio(struct file_ra_state *ra) |
76d42bd9 | 2167 | { |
76d42bd9 | 2168 | ra->ra_pages /= 4; |
76d42bd9 WF |
2169 | } |
2170 | ||
723ef24b KO |
2171 | static int lock_page_for_iocb(struct kiocb *iocb, struct page *page) |
2172 | { | |
2173 | if (iocb->ki_flags & IOCB_WAITQ) | |
2174 | return lock_page_async(page, iocb->ki_waitq); | |
2175 | else if (iocb->ki_flags & IOCB_NOWAIT) | |
2176 | return trylock_page(page) ? 0 : -EAGAIN; | |
2177 | else | |
2178 | return lock_page_killable(page); | |
2179 | } | |
2180 | ||
723ef24b KO |
2181 | static struct page * |
2182 | generic_file_buffered_read_readpage(struct kiocb *iocb, | |
2183 | struct file *filp, | |
2184 | struct address_space *mapping, | |
2185 | struct page *page) | |
2186 | { | |
2187 | struct file_ra_state *ra = &filp->f_ra; | |
2188 | int error; | |
2189 | ||
2190 | if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT)) { | |
2191 | unlock_page(page); | |
2192 | put_page(page); | |
2193 | return ERR_PTR(-EAGAIN); | |
2194 | } | |
2195 | ||
2196 | /* | |
2197 | * A previous I/O error may have been due to temporary | |
2198 | * failures, eg. multipath errors. | |
2199 | * PG_error will be set again if readpage fails. | |
2200 | */ | |
2201 | ClearPageError(page); | |
2202 | /* Start the actual read. The read will unlock the page. */ | |
2203 | error = mapping->a_ops->readpage(filp, page); | |
2204 | ||
2205 | if (unlikely(error)) { | |
2206 | put_page(page); | |
2207 | return error != AOP_TRUNCATED_PAGE ? ERR_PTR(error) : NULL; | |
2208 | } | |
2209 | ||
2210 | if (!PageUptodate(page)) { | |
2211 | error = lock_page_for_iocb(iocb, page); | |
2212 | if (unlikely(error)) { | |
2213 | put_page(page); | |
2214 | return ERR_PTR(error); | |
2215 | } | |
2216 | if (!PageUptodate(page)) { | |
2217 | if (page->mapping == NULL) { | |
2218 | /* | |
2219 | * invalidate_mapping_pages got it | |
2220 | */ | |
2221 | unlock_page(page); | |
2222 | put_page(page); | |
2223 | return NULL; | |
2224 | } | |
2225 | unlock_page(page); | |
2226 | shrink_readahead_size_eio(ra); | |
2227 | put_page(page); | |
2228 | return ERR_PTR(-EIO); | |
2229 | } | |
2230 | unlock_page(page); | |
2231 | } | |
2232 | ||
2233 | return page; | |
2234 | } | |
2235 | ||
2236 | static struct page * | |
2237 | generic_file_buffered_read_pagenotuptodate(struct kiocb *iocb, | |
2238 | struct file *filp, | |
2239 | struct iov_iter *iter, | |
2240 | struct page *page, | |
2241 | loff_t pos, loff_t count) | |
2242 | { | |
2243 | struct address_space *mapping = filp->f_mapping; | |
2244 | struct inode *inode = mapping->host; | |
2245 | int error; | |
2246 | ||
2247 | /* | |
2248 | * See comment in do_read_cache_page on why | |
2249 | * wait_on_page_locked is used to avoid unnecessarily | |
2250 | * serialisations and why it's safe. | |
2251 | */ | |
2252 | if (iocb->ki_flags & IOCB_WAITQ) { | |
2253 | error = wait_on_page_locked_async(page, | |
2254 | iocb->ki_waitq); | |
2255 | } else { | |
2256 | error = wait_on_page_locked_killable(page); | |
2257 | } | |
2258 | if (unlikely(error)) { | |
2259 | put_page(page); | |
2260 | return ERR_PTR(error); | |
2261 | } | |
2262 | if (PageUptodate(page)) | |
2263 | return page; | |
2264 | ||
2265 | if (inode->i_blkbits == PAGE_SHIFT || | |
2266 | !mapping->a_ops->is_partially_uptodate) | |
2267 | goto page_not_up_to_date; | |
2268 | /* pipes can't handle partially uptodate pages */ | |
2269 | if (unlikely(iov_iter_is_pipe(iter))) | |
2270 | goto page_not_up_to_date; | |
2271 | if (!trylock_page(page)) | |
2272 | goto page_not_up_to_date; | |
2273 | /* Did it get truncated before we got the lock? */ | |
2274 | if (!page->mapping) | |
2275 | goto page_not_up_to_date_locked; | |
2276 | if (!mapping->a_ops->is_partially_uptodate(page, | |
2277 | pos & ~PAGE_MASK, count)) | |
2278 | goto page_not_up_to_date_locked; | |
2279 | unlock_page(page); | |
2280 | return page; | |
2281 | ||
2282 | page_not_up_to_date: | |
2283 | /* Get exclusive access to the page ... */ | |
2284 | error = lock_page_for_iocb(iocb, page); | |
2285 | if (unlikely(error)) { | |
2286 | put_page(page); | |
2287 | return ERR_PTR(error); | |
2288 | } | |
2289 | ||
2290 | page_not_up_to_date_locked: | |
2291 | /* Did it get truncated before we got the lock? */ | |
2292 | if (!page->mapping) { | |
2293 | unlock_page(page); | |
2294 | put_page(page); | |
2295 | return NULL; | |
2296 | } | |
2297 | ||
2298 | /* Did somebody else fill it already? */ | |
2299 | if (PageUptodate(page)) { | |
2300 | unlock_page(page); | |
2301 | return page; | |
2302 | } | |
2303 | ||
2304 | return generic_file_buffered_read_readpage(iocb, filp, mapping, page); | |
2305 | } | |
2306 | ||
2307 | static struct page * | |
2308 | generic_file_buffered_read_no_cached_page(struct kiocb *iocb, | |
2309 | struct iov_iter *iter) | |
2310 | { | |
2311 | struct file *filp = iocb->ki_filp; | |
2312 | struct address_space *mapping = filp->f_mapping; | |
2313 | pgoff_t index = iocb->ki_pos >> PAGE_SHIFT; | |
2314 | struct page *page; | |
2315 | int error; | |
2316 | ||
2317 | if (iocb->ki_flags & IOCB_NOIO) | |
2318 | return ERR_PTR(-EAGAIN); | |
2319 | ||
2320 | /* | |
2321 | * Ok, it wasn't cached, so we need to create a new | |
2322 | * page.. | |
2323 | */ | |
2324 | page = page_cache_alloc(mapping); | |
2325 | if (!page) | |
2326 | return ERR_PTR(-ENOMEM); | |
2327 | ||
2328 | error = add_to_page_cache_lru(page, mapping, index, | |
2329 | mapping_gfp_constraint(mapping, GFP_KERNEL)); | |
2330 | if (error) { | |
2331 | put_page(page); | |
2332 | return error != -EEXIST ? ERR_PTR(error) : NULL; | |
2333 | } | |
2334 | ||
2335 | return generic_file_buffered_read_readpage(iocb, filp, mapping, page); | |
2336 | } | |
2337 | ||
06c04442 KO |
2338 | static int generic_file_buffered_read_get_pages(struct kiocb *iocb, |
2339 | struct iov_iter *iter, | |
2340 | struct page **pages, | |
2341 | unsigned int nr) | |
2342 | { | |
2343 | struct file *filp = iocb->ki_filp; | |
2344 | struct address_space *mapping = filp->f_mapping; | |
2345 | struct file_ra_state *ra = &filp->f_ra; | |
2346 | pgoff_t index = iocb->ki_pos >> PAGE_SHIFT; | |
2347 | pgoff_t last_index = (iocb->ki_pos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; | |
2348 | int i, j, nr_got, err = 0; | |
2349 | ||
2350 | nr = min_t(unsigned long, last_index - index, nr); | |
2351 | find_page: | |
2352 | if (fatal_signal_pending(current)) | |
2353 | return -EINTR; | |
2354 | ||
2355 | nr_got = find_get_pages_contig(mapping, index, nr, pages); | |
2356 | if (nr_got) | |
2357 | goto got_pages; | |
2358 | ||
2359 | if (iocb->ki_flags & IOCB_NOIO) | |
2360 | return -EAGAIN; | |
2361 | ||
2362 | page_cache_sync_readahead(mapping, ra, filp, index, last_index - index); | |
2363 | ||
2364 | nr_got = find_get_pages_contig(mapping, index, nr, pages); | |
2365 | if (nr_got) | |
2366 | goto got_pages; | |
2367 | ||
2368 | pages[0] = generic_file_buffered_read_no_cached_page(iocb, iter); | |
2369 | err = PTR_ERR_OR_ZERO(pages[0]); | |
2370 | if (!IS_ERR_OR_NULL(pages[0])) | |
2371 | nr_got = 1; | |
2372 | got_pages: | |
2373 | for (i = 0; i < nr_got; i++) { | |
2374 | struct page *page = pages[i]; | |
2375 | pgoff_t pg_index = index + i; | |
2376 | loff_t pg_pos = max(iocb->ki_pos, | |
2377 | (loff_t) pg_index << PAGE_SHIFT); | |
2378 | loff_t pg_count = iocb->ki_pos + iter->count - pg_pos; | |
2379 | ||
2380 | if (PageReadahead(page)) { | |
2381 | if (iocb->ki_flags & IOCB_NOIO) { | |
2382 | for (j = i; j < nr_got; j++) | |
2383 | put_page(pages[j]); | |
2384 | nr_got = i; | |
2385 | err = -EAGAIN; | |
2386 | break; | |
2387 | } | |
2388 | page_cache_async_readahead(mapping, ra, filp, page, | |
2389 | pg_index, last_index - pg_index); | |
2390 | } | |
2391 | ||
2392 | if (!PageUptodate(page)) { | |
2393 | if ((iocb->ki_flags & IOCB_NOWAIT) || | |
2394 | ((iocb->ki_flags & IOCB_WAITQ) && i)) { | |
2395 | for (j = i; j < nr_got; j++) | |
2396 | put_page(pages[j]); | |
2397 | nr_got = i; | |
2398 | err = -EAGAIN; | |
2399 | break; | |
2400 | } | |
2401 | ||
2402 | page = generic_file_buffered_read_pagenotuptodate(iocb, | |
2403 | filp, iter, page, pg_pos, pg_count); | |
2404 | if (IS_ERR_OR_NULL(page)) { | |
2405 | for (j = i + 1; j < nr_got; j++) | |
2406 | put_page(pages[j]); | |
2407 | nr_got = i; | |
2408 | err = PTR_ERR_OR_ZERO(page); | |
2409 | break; | |
2410 | } | |
2411 | } | |
2412 | } | |
2413 | ||
2414 | if (likely(nr_got)) | |
2415 | return nr_got; | |
2416 | if (err) | |
2417 | return err; | |
2418 | /* | |
2419 | * No pages and no error means we raced and should retry: | |
2420 | */ | |
2421 | goto find_page; | |
2422 | } | |
2423 | ||
485bb99b | 2424 | /** |
47c27bc4 CH |
2425 | * generic_file_buffered_read - generic file read routine |
2426 | * @iocb: the iocb to read | |
6e58e79d AV |
2427 | * @iter: data destination |
2428 | * @written: already copied | |
485bb99b | 2429 | * |
1da177e4 | 2430 | * This is a generic file read routine, and uses the |
485bb99b | 2431 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
2432 | * |
2433 | * This is really ugly. But the goto's actually try to clarify some | |
2434 | * of the logic when it comes to error handling etc. | |
a862f68a MR |
2435 | * |
2436 | * Return: | |
2437 | * * total number of bytes copied, including those the were already @written | |
2438 | * * negative error code if nothing was copied | |
1da177e4 | 2439 | */ |
d85dc2e1 | 2440 | ssize_t generic_file_buffered_read(struct kiocb *iocb, |
6e58e79d | 2441 | struct iov_iter *iter, ssize_t written) |
1da177e4 | 2442 | { |
47c27bc4 | 2443 | struct file *filp = iocb->ki_filp; |
06c04442 | 2444 | struct file_ra_state *ra = &filp->f_ra; |
36e78914 | 2445 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2446 | struct inode *inode = mapping->host; |
06c04442 KO |
2447 | struct page *pages_onstack[PAGEVEC_SIZE], **pages = NULL; |
2448 | unsigned int nr_pages = min_t(unsigned int, 512, | |
2449 | ((iocb->ki_pos + iter->count + PAGE_SIZE - 1) >> PAGE_SHIFT) - | |
2450 | (iocb->ki_pos >> PAGE_SHIFT)); | |
2451 | int i, pg_nr, error = 0; | |
2452 | bool writably_mapped; | |
2453 | loff_t isize, end_offset; | |
1da177e4 | 2454 | |
723ef24b | 2455 | if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2456 | return 0; |
3644e2d2 KO |
2457 | if (unlikely(!iov_iter_count(iter))) |
2458 | return 0; | |
2459 | ||
c2a9737f WF |
2460 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
2461 | ||
06c04442 KO |
2462 | if (nr_pages > ARRAY_SIZE(pages_onstack)) |
2463 | pages = kmalloc_array(nr_pages, sizeof(void *), GFP_KERNEL); | |
13bd6914 | 2464 | |
06c04442 KO |
2465 | if (!pages) { |
2466 | pages = pages_onstack; | |
2467 | nr_pages = min_t(unsigned int, nr_pages, ARRAY_SIZE(pages_onstack)); | |
2468 | } | |
1da177e4 | 2469 | |
06c04442 | 2470 | do { |
1da177e4 | 2471 | cond_resched(); |
5abf186a | 2472 | |
723ef24b | 2473 | /* |
06c04442 KO |
2474 | * If we've already successfully copied some data, then we |
2475 | * can no longer safely return -EIOCBQUEUED. Hence mark | |
2476 | * an async read NOWAIT at that point. | |
723ef24b | 2477 | */ |
06c04442 | 2478 | if ((iocb->ki_flags & IOCB_WAITQ) && written) |
723ef24b KO |
2479 | iocb->ki_flags |= IOCB_NOWAIT; |
2480 | ||
06c04442 KO |
2481 | i = 0; |
2482 | pg_nr = generic_file_buffered_read_get_pages(iocb, iter, | |
2483 | pages, nr_pages); | |
2484 | if (pg_nr < 0) { | |
2485 | error = pg_nr; | |
2486 | break; | |
1da177e4 LT |
2487 | } |
2488 | ||
06c04442 KO |
2489 | /* |
2490 | * i_size must be checked after we know the pages are Uptodate. | |
2491 | * | |
2492 | * Checking i_size after the check allows us to calculate | |
2493 | * the correct value for "nr", which means the zero-filled | |
2494 | * part of the page is not copied back to userspace (unless | |
2495 | * another truncate extends the file - this is desired though). | |
2496 | */ | |
2497 | isize = i_size_read(inode); | |
2498 | if (unlikely(iocb->ki_pos >= isize)) | |
2499 | goto put_pages; | |
1da177e4 | 2500 | |
06c04442 KO |
2501 | end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count); |
2502 | ||
2503 | while ((iocb->ki_pos >> PAGE_SHIFT) + pg_nr > | |
2504 | (end_offset + PAGE_SIZE - 1) >> PAGE_SHIFT) | |
2505 | put_page(pages[--pg_nr]); | |
2506 | ||
2507 | /* | |
2508 | * Once we start copying data, we don't want to be touching any | |
2509 | * cachelines that might be contended: | |
2510 | */ | |
2511 | writably_mapped = mapping_writably_mapped(mapping); | |
2512 | ||
2513 | /* | |
2514 | * When a sequential read accesses a page several times, only | |
2515 | * mark it as accessed the first time. | |
2516 | */ | |
2517 | if (iocb->ki_pos >> PAGE_SHIFT != | |
2518 | ra->prev_pos >> PAGE_SHIFT) | |
2519 | mark_page_accessed(pages[0]); | |
2520 | for (i = 1; i < pg_nr; i++) | |
2521 | mark_page_accessed(pages[i]); | |
2522 | ||
2523 | for (i = 0; i < pg_nr; i++) { | |
2524 | unsigned int offset = iocb->ki_pos & ~PAGE_MASK; | |
2525 | unsigned int bytes = min_t(loff_t, end_offset - iocb->ki_pos, | |
2526 | PAGE_SIZE - offset); | |
2527 | unsigned int copied; | |
2528 | ||
2529 | /* | |
2530 | * If users can be writing to this page using arbitrary | |
2531 | * virtual addresses, take care about potential aliasing | |
2532 | * before reading the page on the kernel side. | |
2533 | */ | |
2534 | if (writably_mapped) | |
2535 | flush_dcache_page(pages[i]); | |
2536 | ||
2537 | copied = copy_page_to_iter(pages[i], offset, bytes, iter); | |
2538 | ||
2539 | written += copied; | |
2540 | iocb->ki_pos += copied; | |
2541 | ra->prev_pos = iocb->ki_pos; | |
2542 | ||
2543 | if (copied < bytes) { | |
2544 | error = -EFAULT; | |
2545 | break; | |
2546 | } | |
1da177e4 | 2547 | } |
06c04442 KO |
2548 | put_pages: |
2549 | for (i = 0; i < pg_nr; i++) | |
2550 | put_page(pages[i]); | |
2551 | } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error); | |
1da177e4 | 2552 | |
0c6aa263 | 2553 | file_accessed(filp); |
06c04442 KO |
2554 | |
2555 | if (pages != pages_onstack) | |
2556 | kfree(pages); | |
723ef24b | 2557 | |
6e58e79d | 2558 | return written ? written : error; |
1da177e4 | 2559 | } |
d85dc2e1 | 2560 | EXPORT_SYMBOL_GPL(generic_file_buffered_read); |
1da177e4 | 2561 | |
485bb99b | 2562 | /** |
6abd2322 | 2563 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2564 | * @iocb: kernel I/O control block |
6abd2322 | 2565 | * @iter: destination for the data read |
485bb99b | 2566 | * |
6abd2322 | 2567 | * This is the "read_iter()" routine for all filesystems |
1da177e4 | 2568 | * that can use the page cache directly. |
41da51bc AG |
2569 | * |
2570 | * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall | |
2571 | * be returned when no data can be read without waiting for I/O requests | |
2572 | * to complete; it doesn't prevent readahead. | |
2573 | * | |
2574 | * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O | |
2575 | * requests shall be made for the read or for readahead. When no data | |
2576 | * can be read, -EAGAIN shall be returned. When readahead would be | |
2577 | * triggered, a partial, possibly empty read shall be returned. | |
2578 | * | |
a862f68a MR |
2579 | * Return: |
2580 | * * number of bytes copied, even for partial reads | |
41da51bc | 2581 | * * negative error code (or 0 if IOCB_NOIO) if nothing was read |
1da177e4 LT |
2582 | */ |
2583 | ssize_t | |
ed978a81 | 2584 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2585 | { |
e7080a43 | 2586 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2587 | ssize_t retval = 0; |
e7080a43 NS |
2588 | |
2589 | if (!count) | |
2590 | goto out; /* skip atime */ | |
1da177e4 | 2591 | |
2ba48ce5 | 2592 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2593 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2594 | struct address_space *mapping = file->f_mapping; |
2595 | struct inode *inode = mapping->host; | |
543ade1f | 2596 | loff_t size; |
1da177e4 | 2597 | |
1da177e4 | 2598 | size = i_size_read(inode); |
6be96d3a GR |
2599 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2600 | if (filemap_range_has_page(mapping, iocb->ki_pos, | |
2601 | iocb->ki_pos + count - 1)) | |
2602 | return -EAGAIN; | |
2603 | } else { | |
2604 | retval = filemap_write_and_wait_range(mapping, | |
2605 | iocb->ki_pos, | |
2606 | iocb->ki_pos + count - 1); | |
2607 | if (retval < 0) | |
2608 | goto out; | |
2609 | } | |
d8d3d94b | 2610 | |
0d5b0cf2 CH |
2611 | file_accessed(file); |
2612 | ||
5ecda137 | 2613 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2614 | if (retval >= 0) { |
c64fb5c7 | 2615 | iocb->ki_pos += retval; |
5ecda137 | 2616 | count -= retval; |
9fe55eea | 2617 | } |
5b47d59a | 2618 | iov_iter_revert(iter, count - iov_iter_count(iter)); |
66f998f6 | 2619 | |
9fe55eea SW |
2620 | /* |
2621 | * Btrfs can have a short DIO read if we encounter | |
2622 | * compressed extents, so if there was an error, or if | |
2623 | * we've already read everything we wanted to, or if | |
2624 | * there was a short read because we hit EOF, go ahead | |
2625 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2626 | * the rest of the read. Buffered reads will not work for |
2627 | * DAX files, so don't bother trying. | |
9fe55eea | 2628 | */ |
5ecda137 | 2629 | if (retval < 0 || !count || iocb->ki_pos >= size || |
0d5b0cf2 | 2630 | IS_DAX(inode)) |
9fe55eea | 2631 | goto out; |
1da177e4 LT |
2632 | } |
2633 | ||
47c27bc4 | 2634 | retval = generic_file_buffered_read(iocb, iter, retval); |
1da177e4 LT |
2635 | out: |
2636 | return retval; | |
2637 | } | |
ed978a81 | 2638 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2639 | |
1da177e4 | 2640 | #ifdef CONFIG_MMU |
1da177e4 | 2641 | #define MMAP_LOTSAMISS (100) |
6b4c9f44 | 2642 | /* |
c1e8d7c6 | 2643 | * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock |
6b4c9f44 JB |
2644 | * @vmf - the vm_fault for this fault. |
2645 | * @page - the page to lock. | |
2646 | * @fpin - the pointer to the file we may pin (or is already pinned). | |
2647 | * | |
c1e8d7c6 | 2648 | * This works similar to lock_page_or_retry in that it can drop the mmap_lock. |
6b4c9f44 | 2649 | * It differs in that it actually returns the page locked if it returns 1 and 0 |
c1e8d7c6 | 2650 | * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin |
6b4c9f44 JB |
2651 | * will point to the pinned file and needs to be fput()'ed at a later point. |
2652 | */ | |
2653 | static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page, | |
2654 | struct file **fpin) | |
2655 | { | |
2656 | if (trylock_page(page)) | |
2657 | return 1; | |
2658 | ||
8b0f9fa2 LT |
2659 | /* |
2660 | * NOTE! This will make us return with VM_FAULT_RETRY, but with | |
c1e8d7c6 | 2661 | * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT |
8b0f9fa2 LT |
2662 | * is supposed to work. We have way too many special cases.. |
2663 | */ | |
6b4c9f44 JB |
2664 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
2665 | return 0; | |
2666 | ||
2667 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); | |
2668 | if (vmf->flags & FAULT_FLAG_KILLABLE) { | |
2669 | if (__lock_page_killable(page)) { | |
2670 | /* | |
c1e8d7c6 | 2671 | * We didn't have the right flags to drop the mmap_lock, |
6b4c9f44 JB |
2672 | * but all fault_handlers only check for fatal signals |
2673 | * if we return VM_FAULT_RETRY, so we need to drop the | |
c1e8d7c6 | 2674 | * mmap_lock here and return 0 if we don't have a fpin. |
6b4c9f44 JB |
2675 | */ |
2676 | if (*fpin == NULL) | |
d8ed45c5 | 2677 | mmap_read_unlock(vmf->vma->vm_mm); |
6b4c9f44 JB |
2678 | return 0; |
2679 | } | |
2680 | } else | |
2681 | __lock_page(page); | |
2682 | return 1; | |
2683 | } | |
2684 | ||
1da177e4 | 2685 | |
ef00e08e | 2686 | /* |
6b4c9f44 JB |
2687 | * Synchronous readahead happens when we don't even find a page in the page |
2688 | * cache at all. We don't want to perform IO under the mmap sem, so if we have | |
2689 | * to drop the mmap sem we return the file that was pinned in order for us to do | |
2690 | * that. If we didn't pin a file then we return NULL. The file that is | |
2691 | * returned needs to be fput()'ed when we're done with it. | |
ef00e08e | 2692 | */ |
6b4c9f44 | 2693 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
ef00e08e | 2694 | { |
2a1180f1 JB |
2695 | struct file *file = vmf->vma->vm_file; |
2696 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2697 | struct address_space *mapping = file->f_mapping; |
db660d46 | 2698 | DEFINE_READAHEAD(ractl, file, mapping, vmf->pgoff); |
6b4c9f44 | 2699 | struct file *fpin = NULL; |
e630bfac | 2700 | unsigned int mmap_miss; |
ef00e08e LT |
2701 | |
2702 | /* If we don't want any read-ahead, don't bother */ | |
2a1180f1 | 2703 | if (vmf->vma->vm_flags & VM_RAND_READ) |
6b4c9f44 | 2704 | return fpin; |
275b12bf | 2705 | if (!ra->ra_pages) |
6b4c9f44 | 2706 | return fpin; |
ef00e08e | 2707 | |
2a1180f1 | 2708 | if (vmf->vma->vm_flags & VM_SEQ_READ) { |
6b4c9f44 | 2709 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
db660d46 | 2710 | page_cache_sync_ra(&ractl, ra, ra->ra_pages); |
6b4c9f44 | 2711 | return fpin; |
ef00e08e LT |
2712 | } |
2713 | ||
207d04ba | 2714 | /* Avoid banging the cache line if not needed */ |
e630bfac KS |
2715 | mmap_miss = READ_ONCE(ra->mmap_miss); |
2716 | if (mmap_miss < MMAP_LOTSAMISS * 10) | |
2717 | WRITE_ONCE(ra->mmap_miss, ++mmap_miss); | |
ef00e08e LT |
2718 | |
2719 | /* | |
2720 | * Do we miss much more than hit in this file? If so, | |
2721 | * stop bothering with read-ahead. It will only hurt. | |
2722 | */ | |
e630bfac | 2723 | if (mmap_miss > MMAP_LOTSAMISS) |
6b4c9f44 | 2724 | return fpin; |
ef00e08e | 2725 | |
d30a1100 WF |
2726 | /* |
2727 | * mmap read-around | |
2728 | */ | |
6b4c9f44 | 2729 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
db660d46 | 2730 | ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2); |
600e19af RG |
2731 | ra->size = ra->ra_pages; |
2732 | ra->async_size = ra->ra_pages / 4; | |
db660d46 DH |
2733 | ractl._index = ra->start; |
2734 | do_page_cache_ra(&ractl, ra->size, ra->async_size); | |
6b4c9f44 | 2735 | return fpin; |
ef00e08e LT |
2736 | } |
2737 | ||
2738 | /* | |
2739 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
6b4c9f44 | 2740 | * so we want to possibly extend the readahead further. We return the file that |
c1e8d7c6 | 2741 | * was pinned if we have to drop the mmap_lock in order to do IO. |
ef00e08e | 2742 | */ |
6b4c9f44 JB |
2743 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
2744 | struct page *page) | |
ef00e08e | 2745 | { |
2a1180f1 JB |
2746 | struct file *file = vmf->vma->vm_file; |
2747 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2748 | struct address_space *mapping = file->f_mapping; |
6b4c9f44 | 2749 | struct file *fpin = NULL; |
e630bfac | 2750 | unsigned int mmap_miss; |
2a1180f1 | 2751 | pgoff_t offset = vmf->pgoff; |
ef00e08e LT |
2752 | |
2753 | /* If we don't want any read-ahead, don't bother */ | |
5c72feee | 2754 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
6b4c9f44 | 2755 | return fpin; |
e630bfac KS |
2756 | mmap_miss = READ_ONCE(ra->mmap_miss); |
2757 | if (mmap_miss) | |
2758 | WRITE_ONCE(ra->mmap_miss, --mmap_miss); | |
6b4c9f44 JB |
2759 | if (PageReadahead(page)) { |
2760 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); | |
2fad6f5d WF |
2761 | page_cache_async_readahead(mapping, ra, file, |
2762 | page, offset, ra->ra_pages); | |
6b4c9f44 JB |
2763 | } |
2764 | return fpin; | |
ef00e08e LT |
2765 | } |
2766 | ||
485bb99b | 2767 | /** |
54cb8821 | 2768 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 2769 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 2770 | * |
54cb8821 | 2771 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2772 | * mapped memory region to read in file data during a page fault. |
2773 | * | |
2774 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2775 | * it in the page cache, and handles the special cases reasonably without | |
2776 | * having a lot of duplicated code. | |
9a95f3cf | 2777 | * |
c1e8d7c6 | 2778 | * vma->vm_mm->mmap_lock must be held on entry. |
9a95f3cf | 2779 | * |
c1e8d7c6 | 2780 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock |
a4985833 | 2781 | * may be dropped before doing I/O or by lock_page_maybe_drop_mmap(). |
9a95f3cf | 2782 | * |
c1e8d7c6 | 2783 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock |
9a95f3cf PC |
2784 | * has not been released. |
2785 | * | |
2786 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
a862f68a MR |
2787 | * |
2788 | * Return: bitwise-OR of %VM_FAULT_ codes. | |
1da177e4 | 2789 | */ |
2bcd6454 | 2790 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
2791 | { |
2792 | int error; | |
11bac800 | 2793 | struct file *file = vmf->vma->vm_file; |
6b4c9f44 | 2794 | struct file *fpin = NULL; |
1da177e4 LT |
2795 | struct address_space *mapping = file->f_mapping; |
2796 | struct file_ra_state *ra = &file->f_ra; | |
2797 | struct inode *inode = mapping->host; | |
ef00e08e | 2798 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 2799 | pgoff_t max_off; |
1da177e4 | 2800 | struct page *page; |
2bcd6454 | 2801 | vm_fault_t ret = 0; |
1da177e4 | 2802 | |
9ab2594f MW |
2803 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2804 | if (unlikely(offset >= max_off)) | |
5307cc1a | 2805 | return VM_FAULT_SIGBUS; |
1da177e4 | 2806 | |
1da177e4 | 2807 | /* |
49426420 | 2808 | * Do we have something in the page cache already? |
1da177e4 | 2809 | */ |
ef00e08e | 2810 | page = find_get_page(mapping, offset); |
45cac65b | 2811 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2812 | /* |
ef00e08e LT |
2813 | * We found the page, so try async readahead before |
2814 | * waiting for the lock. | |
1da177e4 | 2815 | */ |
6b4c9f44 | 2816 | fpin = do_async_mmap_readahead(vmf, page); |
45cac65b | 2817 | } else if (!page) { |
ef00e08e | 2818 | /* No page in the page cache at all */ |
ef00e08e | 2819 | count_vm_event(PGMAJFAULT); |
2262185c | 2820 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e | 2821 | ret = VM_FAULT_MAJOR; |
6b4c9f44 | 2822 | fpin = do_sync_mmap_readahead(vmf); |
ef00e08e | 2823 | retry_find: |
a75d4c33 JB |
2824 | page = pagecache_get_page(mapping, offset, |
2825 | FGP_CREAT|FGP_FOR_MMAP, | |
2826 | vmf->gfp_mask); | |
6b4c9f44 JB |
2827 | if (!page) { |
2828 | if (fpin) | |
2829 | goto out_retry; | |
e520e932 | 2830 | return VM_FAULT_OOM; |
6b4c9f44 | 2831 | } |
1da177e4 LT |
2832 | } |
2833 | ||
6b4c9f44 JB |
2834 | if (!lock_page_maybe_drop_mmap(vmf, page, &fpin)) |
2835 | goto out_retry; | |
b522c94d ML |
2836 | |
2837 | /* Did it get truncated? */ | |
585e5a7b | 2838 | if (unlikely(compound_head(page)->mapping != mapping)) { |
b522c94d ML |
2839 | unlock_page(page); |
2840 | put_page(page); | |
2841 | goto retry_find; | |
2842 | } | |
520e5ba4 | 2843 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
b522c94d | 2844 | |
1da177e4 | 2845 | /* |
d00806b1 NP |
2846 | * We have a locked page in the page cache, now we need to check |
2847 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2848 | */ |
d00806b1 | 2849 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2850 | goto page_not_uptodate; |
2851 | ||
6b4c9f44 | 2852 | /* |
c1e8d7c6 | 2853 | * We've made it this far and we had to drop our mmap_lock, now is the |
6b4c9f44 JB |
2854 | * time to return to the upper layer and have it re-find the vma and |
2855 | * redo the fault. | |
2856 | */ | |
2857 | if (fpin) { | |
2858 | unlock_page(page); | |
2859 | goto out_retry; | |
2860 | } | |
2861 | ||
ef00e08e LT |
2862 | /* |
2863 | * Found the page and have a reference on it. | |
2864 | * We must recheck i_size under page lock. | |
2865 | */ | |
9ab2594f MW |
2866 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2867 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 2868 | unlock_page(page); |
09cbfeaf | 2869 | put_page(page); |
5307cc1a | 2870 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2871 | } |
2872 | ||
d0217ac0 | 2873 | vmf->page = page; |
83c54070 | 2874 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2875 | |
1da177e4 | 2876 | page_not_uptodate: |
1da177e4 LT |
2877 | /* |
2878 | * Umm, take care of errors if the page isn't up-to-date. | |
2879 | * Try to re-read it _once_. We do this synchronously, | |
2880 | * because there really aren't any performance issues here | |
2881 | * and we need to check for errors. | |
2882 | */ | |
1da177e4 | 2883 | ClearPageError(page); |
6b4c9f44 | 2884 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
994fc28c | 2885 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2886 | if (!error) { |
2887 | wait_on_page_locked(page); | |
2888 | if (!PageUptodate(page)) | |
2889 | error = -EIO; | |
2890 | } | |
6b4c9f44 JB |
2891 | if (fpin) |
2892 | goto out_retry; | |
09cbfeaf | 2893 | put_page(page); |
d00806b1 NP |
2894 | |
2895 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2896 | goto retry_find; |
1da177e4 | 2897 | |
0f8e2db4 | 2898 | shrink_readahead_size_eio(ra); |
d0217ac0 | 2899 | return VM_FAULT_SIGBUS; |
6b4c9f44 JB |
2900 | |
2901 | out_retry: | |
2902 | /* | |
c1e8d7c6 | 2903 | * We dropped the mmap_lock, we need to return to the fault handler to |
6b4c9f44 JB |
2904 | * re-find the vma and come back and find our hopefully still populated |
2905 | * page. | |
2906 | */ | |
2907 | if (page) | |
2908 | put_page(page); | |
2909 | if (fpin) | |
2910 | fput(fpin); | |
2911 | return ret | VM_FAULT_RETRY; | |
54cb8821 NP |
2912 | } |
2913 | EXPORT_SYMBOL(filemap_fault); | |
2914 | ||
f9ce0be7 | 2915 | static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page) |
f1820361 | 2916 | { |
f9ce0be7 KS |
2917 | struct mm_struct *mm = vmf->vma->vm_mm; |
2918 | ||
2919 | /* Huge page is mapped? No need to proceed. */ | |
2920 | if (pmd_trans_huge(*vmf->pmd)) { | |
2921 | unlock_page(page); | |
2922 | put_page(page); | |
2923 | return true; | |
2924 | } | |
2925 | ||
2926 | if (pmd_none(*vmf->pmd) && PageTransHuge(page)) { | |
2927 | vm_fault_t ret = do_set_pmd(vmf, page); | |
2928 | if (!ret) { | |
2929 | /* The page is mapped successfully, reference consumed. */ | |
2930 | unlock_page(page); | |
2931 | return true; | |
2932 | } | |
2933 | } | |
2934 | ||
2935 | if (pmd_none(*vmf->pmd)) { | |
2936 | vmf->ptl = pmd_lock(mm, vmf->pmd); | |
2937 | if (likely(pmd_none(*vmf->pmd))) { | |
2938 | mm_inc_nr_ptes(mm); | |
2939 | pmd_populate(mm, vmf->pmd, vmf->prealloc_pte); | |
2940 | vmf->prealloc_pte = NULL; | |
2941 | } | |
2942 | spin_unlock(vmf->ptl); | |
2943 | } | |
2944 | ||
2945 | /* See comment in handle_pte_fault() */ | |
2946 | if (pmd_devmap_trans_unstable(vmf->pmd)) { | |
2947 | unlock_page(page); | |
2948 | put_page(page); | |
2949 | return true; | |
2950 | } | |
2951 | ||
2952 | return false; | |
2953 | } | |
2954 | ||
2955 | static struct page *next_uptodate_page(struct page *page, | |
2956 | struct address_space *mapping, | |
2957 | struct xa_state *xas, pgoff_t end_pgoff) | |
2958 | { | |
2959 | unsigned long max_idx; | |
2960 | ||
2961 | do { | |
2962 | if (!page) | |
2963 | return NULL; | |
2964 | if (xas_retry(xas, page)) | |
2965 | continue; | |
2966 | if (xa_is_value(page)) | |
2967 | continue; | |
2968 | if (PageLocked(page)) | |
2969 | continue; | |
2970 | if (!page_cache_get_speculative(page)) | |
2971 | continue; | |
2972 | /* Has the page moved or been split? */ | |
2973 | if (unlikely(page != xas_reload(xas))) | |
2974 | goto skip; | |
2975 | if (!PageUptodate(page) || PageReadahead(page)) | |
2976 | goto skip; | |
2977 | if (PageHWPoison(page)) | |
2978 | goto skip; | |
2979 | if (!trylock_page(page)) | |
2980 | goto skip; | |
2981 | if (page->mapping != mapping) | |
2982 | goto unlock; | |
2983 | if (!PageUptodate(page)) | |
2984 | goto unlock; | |
2985 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); | |
2986 | if (xas->xa_index >= max_idx) | |
2987 | goto unlock; | |
2988 | return page; | |
2989 | unlock: | |
2990 | unlock_page(page); | |
2991 | skip: | |
2992 | put_page(page); | |
2993 | } while ((page = xas_next_entry(xas, end_pgoff)) != NULL); | |
2994 | ||
2995 | return NULL; | |
2996 | } | |
2997 | ||
2998 | static inline struct page *first_map_page(struct address_space *mapping, | |
2999 | struct xa_state *xas, | |
3000 | pgoff_t end_pgoff) | |
3001 | { | |
3002 | return next_uptodate_page(xas_find(xas, end_pgoff), | |
3003 | mapping, xas, end_pgoff); | |
3004 | } | |
3005 | ||
3006 | static inline struct page *next_map_page(struct address_space *mapping, | |
3007 | struct xa_state *xas, | |
3008 | pgoff_t end_pgoff) | |
3009 | { | |
3010 | return next_uptodate_page(xas_next_entry(xas, end_pgoff), | |
3011 | mapping, xas, end_pgoff); | |
3012 | } | |
3013 | ||
3014 | vm_fault_t filemap_map_pages(struct vm_fault *vmf, | |
3015 | pgoff_t start_pgoff, pgoff_t end_pgoff) | |
3016 | { | |
3017 | struct vm_area_struct *vma = vmf->vma; | |
3018 | struct file *file = vma->vm_file; | |
f1820361 | 3019 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 3020 | pgoff_t last_pgoff = start_pgoff; |
f9ce0be7 | 3021 | unsigned long address = vmf->address; |
46bdb427 | 3022 | unsigned long flags = vmf->flags; |
070e807c | 3023 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
27a83a60 | 3024 | struct page *head, *page; |
e630bfac | 3025 | unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss); |
f9ce0be7 | 3026 | vm_fault_t ret = 0; |
f1820361 KS |
3027 | |
3028 | rcu_read_lock(); | |
f9ce0be7 KS |
3029 | head = first_map_page(mapping, &xas, end_pgoff); |
3030 | if (!head) | |
3031 | goto out; | |
f1820361 | 3032 | |
f9ce0be7 KS |
3033 | if (filemap_map_pmd(vmf, head)) { |
3034 | ret = VM_FAULT_NOPAGE; | |
3035 | goto out; | |
3036 | } | |
f1820361 | 3037 | |
f9ce0be7 KS |
3038 | vmf->address = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
3039 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, vmf->address, &vmf->ptl); | |
3040 | do { | |
27a83a60 | 3041 | page = find_subpage(head, xas.xa_index); |
f9ce0be7 | 3042 | if (PageHWPoison(page)) |
f1820361 KS |
3043 | goto unlock; |
3044 | ||
e630bfac KS |
3045 | if (mmap_miss > 0) |
3046 | mmap_miss--; | |
7267ec00 | 3047 | |
070e807c | 3048 | vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
f9ce0be7 | 3049 | vmf->pte += xas.xa_index - last_pgoff; |
070e807c | 3050 | last_pgoff = xas.xa_index; |
f9ce0be7 KS |
3051 | |
3052 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3053 | goto unlock; |
f9ce0be7 | 3054 | |
46bdb427 WD |
3055 | /* We're about to handle the fault */ |
3056 | if (vmf->address == address) { | |
3057 | vmf->flags &= ~FAULT_FLAG_PREFAULT; | |
3058 | ret = VM_FAULT_NOPAGE; | |
3059 | } else { | |
3060 | vmf->flags |= FAULT_FLAG_PREFAULT; | |
3061 | } | |
3062 | ||
f9ce0be7 KS |
3063 | do_set_pte(vmf, page); |
3064 | /* no need to invalidate: a not-present page won't be cached */ | |
3065 | update_mmu_cache(vma, vmf->address, vmf->pte); | |
27a83a60 | 3066 | unlock_page(head); |
f9ce0be7 | 3067 | continue; |
f1820361 | 3068 | unlock: |
27a83a60 | 3069 | unlock_page(head); |
27a83a60 | 3070 | put_page(head); |
f9ce0be7 KS |
3071 | } while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL); |
3072 | pte_unmap_unlock(vmf->pte, vmf->ptl); | |
3073 | out: | |
f1820361 | 3074 | rcu_read_unlock(); |
46bdb427 | 3075 | vmf->flags = flags; |
f9ce0be7 | 3076 | vmf->address = address; |
e630bfac | 3077 | WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss); |
f9ce0be7 | 3078 | return ret; |
f1820361 KS |
3079 | } |
3080 | EXPORT_SYMBOL(filemap_map_pages); | |
3081 | ||
2bcd6454 | 3082 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 | 3083 | { |
5df1a672 | 3084 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
4fcf1c62 | 3085 | struct page *page = vmf->page; |
2bcd6454 | 3086 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 3087 | |
5df1a672 | 3088 | sb_start_pagefault(mapping->host->i_sb); |
11bac800 | 3089 | file_update_time(vmf->vma->vm_file); |
4fcf1c62 | 3090 | lock_page(page); |
5df1a672 | 3091 | if (page->mapping != mapping) { |
4fcf1c62 JK |
3092 | unlock_page(page); |
3093 | ret = VM_FAULT_NOPAGE; | |
3094 | goto out; | |
3095 | } | |
14da9200 JK |
3096 | /* |
3097 | * We mark the page dirty already here so that when freeze is in | |
3098 | * progress, we are guaranteed that writeback during freezing will | |
3099 | * see the dirty page and writeprotect it again. | |
3100 | */ | |
3101 | set_page_dirty(page); | |
1d1d1a76 | 3102 | wait_for_stable_page(page); |
4fcf1c62 | 3103 | out: |
5df1a672 | 3104 | sb_end_pagefault(mapping->host->i_sb); |
4fcf1c62 JK |
3105 | return ret; |
3106 | } | |
4fcf1c62 | 3107 | |
f0f37e2f | 3108 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 3109 | .fault = filemap_fault, |
f1820361 | 3110 | .map_pages = filemap_map_pages, |
4fcf1c62 | 3111 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
3112 | }; |
3113 | ||
3114 | /* This is used for a general mmap of a disk file */ | |
3115 | ||
3116 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
3117 | { | |
3118 | struct address_space *mapping = file->f_mapping; | |
3119 | ||
3120 | if (!mapping->a_ops->readpage) | |
3121 | return -ENOEXEC; | |
3122 | file_accessed(file); | |
3123 | vma->vm_ops = &generic_file_vm_ops; | |
3124 | return 0; | |
3125 | } | |
1da177e4 LT |
3126 | |
3127 | /* | |
3128 | * This is for filesystems which do not implement ->writepage. | |
3129 | */ | |
3130 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
3131 | { | |
3132 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
3133 | return -EINVAL; | |
3134 | return generic_file_mmap(file, vma); | |
3135 | } | |
3136 | #else | |
4b96a37d | 3137 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
45397228 | 3138 | { |
4b96a37d | 3139 | return VM_FAULT_SIGBUS; |
45397228 | 3140 | } |
1da177e4 LT |
3141 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) |
3142 | { | |
3143 | return -ENOSYS; | |
3144 | } | |
3145 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
3146 | { | |
3147 | return -ENOSYS; | |
3148 | } | |
3149 | #endif /* CONFIG_MMU */ | |
3150 | ||
45397228 | 3151 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
3152 | EXPORT_SYMBOL(generic_file_mmap); |
3153 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
3154 | ||
67f9fd91 SL |
3155 | static struct page *wait_on_page_read(struct page *page) |
3156 | { | |
3157 | if (!IS_ERR(page)) { | |
3158 | wait_on_page_locked(page); | |
3159 | if (!PageUptodate(page)) { | |
09cbfeaf | 3160 | put_page(page); |
67f9fd91 SL |
3161 | page = ERR_PTR(-EIO); |
3162 | } | |
3163 | } | |
3164 | return page; | |
3165 | } | |
3166 | ||
32b63529 | 3167 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 3168 | pgoff_t index, |
5e5358e7 | 3169 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3170 | void *data, |
3171 | gfp_t gfp) | |
1da177e4 | 3172 | { |
eb2be189 | 3173 | struct page *page; |
1da177e4 LT |
3174 | int err; |
3175 | repeat: | |
3176 | page = find_get_page(mapping, index); | |
3177 | if (!page) { | |
453f85d4 | 3178 | page = __page_cache_alloc(gfp); |
eb2be189 NP |
3179 | if (!page) |
3180 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 3181 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 3182 | if (unlikely(err)) { |
09cbfeaf | 3183 | put_page(page); |
eb2be189 NP |
3184 | if (err == -EEXIST) |
3185 | goto repeat; | |
22ecdb4f | 3186 | /* Presumably ENOMEM for xarray node */ |
1da177e4 LT |
3187 | return ERR_PTR(err); |
3188 | } | |
32b63529 MG |
3189 | |
3190 | filler: | |
6c45b454 CH |
3191 | if (filler) |
3192 | err = filler(data, page); | |
3193 | else | |
3194 | err = mapping->a_ops->readpage(data, page); | |
3195 | ||
1da177e4 | 3196 | if (err < 0) { |
09cbfeaf | 3197 | put_page(page); |
32b63529 | 3198 | return ERR_PTR(err); |
1da177e4 | 3199 | } |
1da177e4 | 3200 | |
32b63529 MG |
3201 | page = wait_on_page_read(page); |
3202 | if (IS_ERR(page)) | |
3203 | return page; | |
3204 | goto out; | |
3205 | } | |
1da177e4 LT |
3206 | if (PageUptodate(page)) |
3207 | goto out; | |
3208 | ||
ebded027 | 3209 | /* |
0e9aa675 | 3210 | * Page is not up to date and may be locked due to one of the following |
ebded027 MG |
3211 | * case a: Page is being filled and the page lock is held |
3212 | * case b: Read/write error clearing the page uptodate status | |
3213 | * case c: Truncation in progress (page locked) | |
3214 | * case d: Reclaim in progress | |
3215 | * | |
3216 | * Case a, the page will be up to date when the page is unlocked. | |
3217 | * There is no need to serialise on the page lock here as the page | |
3218 | * is pinned so the lock gives no additional protection. Even if the | |
ce89fddf | 3219 | * page is truncated, the data is still valid if PageUptodate as |
ebded027 MG |
3220 | * it's a race vs truncate race. |
3221 | * Case b, the page will not be up to date | |
3222 | * Case c, the page may be truncated but in itself, the data may still | |
3223 | * be valid after IO completes as it's a read vs truncate race. The | |
3224 | * operation must restart if the page is not uptodate on unlock but | |
3225 | * otherwise serialising on page lock to stabilise the mapping gives | |
3226 | * no additional guarantees to the caller as the page lock is | |
3227 | * released before return. | |
3228 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
3229 | * will be a race with remove_mapping that determines if the mapping | |
3230 | * is valid on unlock but otherwise the data is valid and there is | |
3231 | * no need to serialise with page lock. | |
3232 | * | |
3233 | * As the page lock gives no additional guarantee, we optimistically | |
3234 | * wait on the page to be unlocked and check if it's up to date and | |
3235 | * use the page if it is. Otherwise, the page lock is required to | |
3236 | * distinguish between the different cases. The motivation is that we | |
3237 | * avoid spurious serialisations and wakeups when multiple processes | |
3238 | * wait on the same page for IO to complete. | |
3239 | */ | |
3240 | wait_on_page_locked(page); | |
3241 | if (PageUptodate(page)) | |
3242 | goto out; | |
3243 | ||
3244 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 3245 | lock_page(page); |
ebded027 MG |
3246 | |
3247 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
3248 | if (!page->mapping) { |
3249 | unlock_page(page); | |
09cbfeaf | 3250 | put_page(page); |
32b63529 | 3251 | goto repeat; |
1da177e4 | 3252 | } |
ebded027 MG |
3253 | |
3254 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
3255 | if (PageUptodate(page)) { |
3256 | unlock_page(page); | |
3257 | goto out; | |
3258 | } | |
faffdfa0 XT |
3259 | |
3260 | /* | |
3261 | * A previous I/O error may have been due to temporary | |
3262 | * failures. | |
3263 | * Clear page error before actual read, PG_error will be | |
3264 | * set again if read page fails. | |
3265 | */ | |
3266 | ClearPageError(page); | |
32b63529 MG |
3267 | goto filler; |
3268 | ||
c855ff37 | 3269 | out: |
6fe6900e NP |
3270 | mark_page_accessed(page); |
3271 | return page; | |
3272 | } | |
0531b2aa LT |
3273 | |
3274 | /** | |
67f9fd91 | 3275 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
3276 | * @mapping: the page's address_space |
3277 | * @index: the page index | |
3278 | * @filler: function to perform the read | |
5e5358e7 | 3279 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 3280 | * |
0531b2aa | 3281 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 3282 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
3283 | * |
3284 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a MR |
3285 | * |
3286 | * Return: up to date page on success, ERR_PTR() on failure. | |
0531b2aa | 3287 | */ |
67f9fd91 | 3288 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 3289 | pgoff_t index, |
5e5358e7 | 3290 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3291 | void *data) |
3292 | { | |
d322a8e5 CH |
3293 | return do_read_cache_page(mapping, index, filler, data, |
3294 | mapping_gfp_mask(mapping)); | |
0531b2aa | 3295 | } |
67f9fd91 | 3296 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
3297 | |
3298 | /** | |
3299 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
3300 | * @mapping: the page's address_space | |
3301 | * @index: the page index | |
3302 | * @gfp: the page allocator flags to use if allocating | |
3303 | * | |
3304 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 3305 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
3306 | * |
3307 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a MR |
3308 | * |
3309 | * Return: up to date page on success, ERR_PTR() on failure. | |
0531b2aa LT |
3310 | */ |
3311 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
3312 | pgoff_t index, | |
3313 | gfp_t gfp) | |
3314 | { | |
6c45b454 | 3315 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
0531b2aa LT |
3316 | } |
3317 | EXPORT_SYMBOL(read_cache_page_gfp); | |
3318 | ||
afddba49 NP |
3319 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
3320 | loff_t pos, unsigned len, unsigned flags, | |
3321 | struct page **pagep, void **fsdata) | |
3322 | { | |
3323 | const struct address_space_operations *aops = mapping->a_ops; | |
3324 | ||
4e02ed4b | 3325 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 3326 | pagep, fsdata); |
afddba49 NP |
3327 | } |
3328 | EXPORT_SYMBOL(pagecache_write_begin); | |
3329 | ||
3330 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
3331 | loff_t pos, unsigned len, unsigned copied, | |
3332 | struct page *page, void *fsdata) | |
3333 | { | |
3334 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 3335 | |
4e02ed4b | 3336 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
3337 | } |
3338 | EXPORT_SYMBOL(pagecache_write_end); | |
3339 | ||
a92853b6 KK |
3340 | /* |
3341 | * Warn about a page cache invalidation failure during a direct I/O write. | |
3342 | */ | |
3343 | void dio_warn_stale_pagecache(struct file *filp) | |
3344 | { | |
3345 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); | |
3346 | char pathname[128]; | |
a92853b6 KK |
3347 | char *path; |
3348 | ||
5df1a672 | 3349 | errseq_set(&filp->f_mapping->wb_err, -EIO); |
a92853b6 KK |
3350 | if (__ratelimit(&_rs)) { |
3351 | path = file_path(filp, pathname, sizeof(pathname)); | |
3352 | if (IS_ERR(path)) | |
3353 | path = "(unknown)"; | |
3354 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); | |
3355 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, | |
3356 | current->comm); | |
3357 | } | |
3358 | } | |
3359 | ||
1da177e4 | 3360 | ssize_t |
1af5bb49 | 3361 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3362 | { |
3363 | struct file *file = iocb->ki_filp; | |
3364 | struct address_space *mapping = file->f_mapping; | |
3365 | struct inode *inode = mapping->host; | |
1af5bb49 | 3366 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3367 | ssize_t written; |
a969e903 CH |
3368 | size_t write_len; |
3369 | pgoff_t end; | |
1da177e4 | 3370 | |
0c949334 | 3371 | write_len = iov_iter_count(from); |
09cbfeaf | 3372 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3373 | |
6be96d3a GR |
3374 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3375 | /* If there are pages to writeback, return */ | |
5df1a672 | 3376 | if (filemap_range_has_page(file->f_mapping, pos, |
35f12f0f | 3377 | pos + write_len - 1)) |
6be96d3a GR |
3378 | return -EAGAIN; |
3379 | } else { | |
3380 | written = filemap_write_and_wait_range(mapping, pos, | |
3381 | pos + write_len - 1); | |
3382 | if (written) | |
3383 | goto out; | |
3384 | } | |
a969e903 CH |
3385 | |
3386 | /* | |
3387 | * After a write we want buffered reads to be sure to go to disk to get | |
3388 | * the new data. We invalidate clean cached page from the region we're | |
3389 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3390 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3391 | */ |
55635ba7 | 3392 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3393 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3394 | /* |
3395 | * If a page can not be invalidated, return 0 to fall back | |
3396 | * to buffered write. | |
3397 | */ | |
3398 | if (written) { | |
3399 | if (written == -EBUSY) | |
3400 | return 0; | |
3401 | goto out; | |
a969e903 CH |
3402 | } |
3403 | ||
639a93a5 | 3404 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3405 | |
3406 | /* | |
3407 | * Finally, try again to invalidate clean pages which might have been | |
3408 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3409 | * if the source of the write was an mmap'ed region of the file | |
3410 | * we're writing. Either one is a pretty crazy thing to do, | |
3411 | * so we don't support it 100%. If this invalidation | |
3412 | * fails, tough, the write still worked... | |
332391a9 LC |
3413 | * |
3414 | * Most of the time we do not need this since dio_complete() will do | |
3415 | * the invalidation for us. However there are some file systems that | |
3416 | * do not end up with dio_complete() being called, so let's not break | |
80c1fe90 KK |
3417 | * them by removing it completely. |
3418 | * | |
9266a140 KK |
3419 | * Noticeable example is a blkdev_direct_IO(). |
3420 | * | |
80c1fe90 | 3421 | * Skip invalidation for async writes or if mapping has no pages. |
a969e903 | 3422 | */ |
9266a140 KK |
3423 | if (written > 0 && mapping->nrpages && |
3424 | invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) | |
3425 | dio_warn_stale_pagecache(file); | |
a969e903 | 3426 | |
1da177e4 | 3427 | if (written > 0) { |
0116651c | 3428 | pos += written; |
639a93a5 | 3429 | write_len -= written; |
0116651c NK |
3430 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3431 | i_size_write(inode, pos); | |
1da177e4 LT |
3432 | mark_inode_dirty(inode); |
3433 | } | |
5cb6c6c7 | 3434 | iocb->ki_pos = pos; |
1da177e4 | 3435 | } |
639a93a5 | 3436 | iov_iter_revert(from, write_len - iov_iter_count(from)); |
a969e903 | 3437 | out: |
1da177e4 LT |
3438 | return written; |
3439 | } | |
3440 | EXPORT_SYMBOL(generic_file_direct_write); | |
3441 | ||
eb2be189 NP |
3442 | /* |
3443 | * Find or create a page at the given pagecache position. Return the locked | |
3444 | * page. This function is specifically for buffered writes. | |
3445 | */ | |
54566b2c NP |
3446 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
3447 | pgoff_t index, unsigned flags) | |
eb2be189 | 3448 | { |
eb2be189 | 3449 | struct page *page; |
bbddabe2 | 3450 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 3451 | |
54566b2c | 3452 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
3453 | fgp_flags |= FGP_NOFS; |
3454 | ||
3455 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 3456 | mapping_gfp_mask(mapping)); |
c585a267 | 3457 | if (page) |
2457aec6 | 3458 | wait_for_stable_page(page); |
eb2be189 | 3459 | |
eb2be189 NP |
3460 | return page; |
3461 | } | |
54566b2c | 3462 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 3463 | |
3b93f911 | 3464 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3465 | struct iov_iter *i, loff_t pos) |
3466 | { | |
3467 | struct address_space *mapping = file->f_mapping; | |
3468 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3469 | long status = 0; | |
3470 | ssize_t written = 0; | |
674b892e NP |
3471 | unsigned int flags = 0; |
3472 | ||
afddba49 NP |
3473 | do { |
3474 | struct page *page; | |
afddba49 NP |
3475 | unsigned long offset; /* Offset into pagecache page */ |
3476 | unsigned long bytes; /* Bytes to write to page */ | |
3477 | size_t copied; /* Bytes copied from user */ | |
3478 | void *fsdata; | |
3479 | ||
09cbfeaf KS |
3480 | offset = (pos & (PAGE_SIZE - 1)); |
3481 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3482 | iov_iter_count(i)); |
3483 | ||
3484 | again: | |
00a3d660 LT |
3485 | /* |
3486 | * Bring in the user page that we will copy from _first_. | |
3487 | * Otherwise there's a nasty deadlock on copying from the | |
3488 | * same page as we're writing to, without it being marked | |
3489 | * up-to-date. | |
3490 | * | |
3491 | * Not only is this an optimisation, but it is also required | |
3492 | * to check that the address is actually valid, when atomic | |
3493 | * usercopies are used, below. | |
3494 | */ | |
3495 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
3496 | status = -EFAULT; | |
3497 | break; | |
3498 | } | |
3499 | ||
296291cd JK |
3500 | if (fatal_signal_pending(current)) { |
3501 | status = -EINTR; | |
3502 | break; | |
3503 | } | |
3504 | ||
674b892e | 3505 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3506 | &page, &fsdata); |
2457aec6 | 3507 | if (unlikely(status < 0)) |
afddba49 NP |
3508 | break; |
3509 | ||
931e80e4 | 3510 | if (mapping_writably_mapped(mapping)) |
3511 | flush_dcache_page(page); | |
00a3d660 | 3512 | |
afddba49 | 3513 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
3514 | flush_dcache_page(page); |
3515 | ||
3516 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3517 | page, fsdata); | |
3518 | if (unlikely(status < 0)) | |
3519 | break; | |
3520 | copied = status; | |
3521 | ||
3522 | cond_resched(); | |
3523 | ||
124d3b70 | 3524 | iov_iter_advance(i, copied); |
afddba49 NP |
3525 | if (unlikely(copied == 0)) { |
3526 | /* | |
3527 | * If we were unable to copy any data at all, we must | |
3528 | * fall back to a single segment length write. | |
3529 | * | |
3530 | * If we didn't fallback here, we could livelock | |
3531 | * because not all segments in the iov can be copied at | |
3532 | * once without a pagefault. | |
3533 | */ | |
09cbfeaf | 3534 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
afddba49 NP |
3535 | iov_iter_single_seg_count(i)); |
3536 | goto again; | |
3537 | } | |
afddba49 NP |
3538 | pos += copied; |
3539 | written += copied; | |
3540 | ||
3541 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3542 | } while (iov_iter_count(i)); |
3543 | ||
3544 | return written ? written : status; | |
3545 | } | |
3b93f911 | 3546 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3547 | |
e4dd9de3 | 3548 | /** |
8174202b | 3549 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3550 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3551 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3552 | * |
3553 | * This function does all the work needed for actually writing data to a | |
3554 | * file. It does all basic checks, removes SUID from the file, updates | |
3555 | * modification times and calls proper subroutines depending on whether we | |
3556 | * do direct IO or a standard buffered write. | |
3557 | * | |
3558 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
3559 | * object which does not need locking at all. | |
3560 | * | |
3561 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3562 | * A caller has to handle it. This is mainly due to the fact that we want to | |
3563 | * avoid syncing under i_mutex. | |
a862f68a MR |
3564 | * |
3565 | * Return: | |
3566 | * * number of bytes written, even for truncated writes | |
3567 | * * negative error code if no data has been written at all | |
e4dd9de3 | 3568 | */ |
8174202b | 3569 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3570 | { |
3571 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 3572 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 3573 | struct inode *inode = mapping->host; |
3b93f911 | 3574 | ssize_t written = 0; |
1da177e4 | 3575 | ssize_t err; |
3b93f911 | 3576 | ssize_t status; |
1da177e4 | 3577 | |
1da177e4 | 3578 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3579 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3580 | err = file_remove_privs(file); |
1da177e4 LT |
3581 | if (err) |
3582 | goto out; | |
3583 | ||
c3b2da31 JB |
3584 | err = file_update_time(file); |
3585 | if (err) | |
3586 | goto out; | |
1da177e4 | 3587 | |
2ba48ce5 | 3588 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3589 | loff_t pos, endbyte; |
fb5527e6 | 3590 | |
1af5bb49 | 3591 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3592 | /* |
fbbbad4b MW |
3593 | * If the write stopped short of completing, fall back to |
3594 | * buffered writes. Some filesystems do this for writes to | |
3595 | * holes, for example. For DAX files, a buffered write will | |
3596 | * not succeed (even if it did, DAX does not handle dirty | |
3597 | * page-cache pages correctly). | |
1da177e4 | 3598 | */ |
0b8def9d | 3599 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3600 | goto out; |
3601 | ||
0b8def9d | 3602 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3603 | /* |
3b93f911 | 3604 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3605 | * then we want to return the number of bytes which were |
3606 | * direct-written, or the error code if that was zero. Note | |
3607 | * that this differs from normal direct-io semantics, which | |
3608 | * will return -EFOO even if some bytes were written. | |
3609 | */ | |
60bb4529 | 3610 | if (unlikely(status < 0)) { |
3b93f911 | 3611 | err = status; |
fb5527e6 JM |
3612 | goto out; |
3613 | } | |
fb5527e6 JM |
3614 | /* |
3615 | * We need to ensure that the page cache pages are written to | |
3616 | * disk and invalidated to preserve the expected O_DIRECT | |
3617 | * semantics. | |
3618 | */ | |
3b93f911 | 3619 | endbyte = pos + status - 1; |
0b8def9d | 3620 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3621 | if (err == 0) { |
0b8def9d | 3622 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3623 | written += status; |
fb5527e6 | 3624 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3625 | pos >> PAGE_SHIFT, |
3626 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3627 | } else { |
3628 | /* | |
3629 | * We don't know how much we wrote, so just return | |
3630 | * the number of bytes which were direct-written | |
3631 | */ | |
3632 | } | |
3633 | } else { | |
0b8def9d AV |
3634 | written = generic_perform_write(file, from, iocb->ki_pos); |
3635 | if (likely(written > 0)) | |
3636 | iocb->ki_pos += written; | |
fb5527e6 | 3637 | } |
1da177e4 LT |
3638 | out: |
3639 | current->backing_dev_info = NULL; | |
3640 | return written ? written : err; | |
3641 | } | |
8174202b | 3642 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3643 | |
e4dd9de3 | 3644 | /** |
8174202b | 3645 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3646 | * @iocb: IO state structure |
8174202b | 3647 | * @from: iov_iter with data to write |
e4dd9de3 | 3648 | * |
8174202b | 3649 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
3650 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
3651 | * and acquires i_mutex as needed. | |
a862f68a MR |
3652 | * Return: |
3653 | * * negative error code if no data has been written at all of | |
3654 | * vfs_fsync_range() failed for a synchronous write | |
3655 | * * number of bytes written, even for truncated writes | |
e4dd9de3 | 3656 | */ |
8174202b | 3657 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3658 | { |
3659 | struct file *file = iocb->ki_filp; | |
148f948b | 3660 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3661 | ssize_t ret; |
1da177e4 | 3662 | |
5955102c | 3663 | inode_lock(inode); |
3309dd04 AV |
3664 | ret = generic_write_checks(iocb, from); |
3665 | if (ret > 0) | |
5f380c7f | 3666 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3667 | inode_unlock(inode); |
1da177e4 | 3668 | |
e2592217 CH |
3669 | if (ret > 0) |
3670 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3671 | return ret; |
3672 | } | |
8174202b | 3673 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3674 | |
cf9a2ae8 DH |
3675 | /** |
3676 | * try_to_release_page() - release old fs-specific metadata on a page | |
3677 | * | |
3678 | * @page: the page which the kernel is trying to free | |
3679 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3680 | * | |
3681 | * The address_space is to try to release any data against the page | |
a862f68a | 3682 | * (presumably at page->private). |
cf9a2ae8 | 3683 | * |
266cf658 DH |
3684 | * This may also be called if PG_fscache is set on a page, indicating that the |
3685 | * page is known to the local caching routines. | |
3686 | * | |
cf9a2ae8 | 3687 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3688 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3689 | * |
a862f68a | 3690 | * Return: %1 if the release was successful, otherwise return zero. |
cf9a2ae8 DH |
3691 | */ |
3692 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3693 | { | |
3694 | struct address_space * const mapping = page->mapping; | |
3695 | ||
3696 | BUG_ON(!PageLocked(page)); | |
3697 | if (PageWriteback(page)) | |
3698 | return 0; | |
3699 | ||
3700 | if (mapping && mapping->a_ops->releasepage) | |
3701 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3702 | return try_to_free_buffers(page); | |
3703 | } | |
3704 | ||
3705 | EXPORT_SYMBOL(try_to_release_page); |