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> |
0f8053a5 NP |
45 | #include "internal.h" |
46 | ||
fe0bfaaf RJ |
47 | #define CREATE_TRACE_POINTS |
48 | #include <trace/events/filemap.h> | |
49 | ||
1da177e4 | 50 | /* |
1da177e4 LT |
51 | * FIXME: remove all knowledge of the buffer layer from the core VM |
52 | */ | |
148f948b | 53 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 54 | |
1da177e4 LT |
55 | #include <asm/mman.h> |
56 | ||
57 | /* | |
58 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
59 | * though. | |
60 | * | |
61 | * Shared mappings now work. 15.8.1995 Bruno. | |
62 | * | |
63 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
64 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
65 | * | |
66 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
67 | */ | |
68 | ||
69 | /* | |
70 | * Lock ordering: | |
71 | * | |
c8c06efa | 72 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 73 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 | 74 | * ->swap_lock (exclusive_swap_page, others) |
b93b0163 | 75 | * ->i_pages lock |
1da177e4 | 76 | * |
1b1dcc1b | 77 | * ->i_mutex |
c8c06efa | 78 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) |
1da177e4 | 79 | * |
c1e8d7c6 | 80 | * ->mmap_lock |
c8c06efa | 81 | * ->i_mmap_rwsem |
b8072f09 | 82 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
b93b0163 | 83 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
1da177e4 | 84 | * |
c1e8d7c6 | 85 | * ->mmap_lock |
1da177e4 LT |
86 | * ->lock_page (access_process_vm) |
87 | * | |
ccad2365 | 88 | * ->i_mutex (generic_perform_write) |
c1e8d7c6 | 89 | * ->mmap_lock (fault_in_pages_readable->do_page_fault) |
1da177e4 | 90 | * |
f758eeab | 91 | * bdi->wb.list_lock |
a66979ab | 92 | * sb_lock (fs/fs-writeback.c) |
b93b0163 | 93 | * ->i_pages lock (__sync_single_inode) |
1da177e4 | 94 | * |
c8c06efa | 95 | * ->i_mmap_rwsem |
1da177e4 LT |
96 | * ->anon_vma.lock (vma_adjust) |
97 | * | |
98 | * ->anon_vma.lock | |
b8072f09 | 99 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 100 | * |
b8072f09 | 101 | * ->page_table_lock or pte_lock |
5d337b91 | 102 | * ->swap_lock (try_to_unmap_one) |
1da177e4 | 103 | * ->private_lock (try_to_unmap_one) |
b93b0163 | 104 | * ->i_pages lock (try_to_unmap_one) |
f4b7e272 AR |
105 | * ->pgdat->lru_lock (follow_page->mark_page_accessed) |
106 | * ->pgdat->lru_lock (check_pte_range->isolate_lru_page) | |
1da177e4 | 107 | * ->private_lock (page_remove_rmap->set_page_dirty) |
b93b0163 | 108 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
f758eeab | 109 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 110 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 111 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 112 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 113 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
114 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
115 | * | |
c8c06efa | 116 | * ->i_mmap_rwsem |
9a3c531d | 117 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
118 | */ |
119 | ||
5c024e6a | 120 | static void page_cache_delete(struct address_space *mapping, |
91b0abe3 JW |
121 | struct page *page, void *shadow) |
122 | { | |
5c024e6a MW |
123 | XA_STATE(xas, &mapping->i_pages, page->index); |
124 | unsigned int nr = 1; | |
c70b647d | 125 | |
5c024e6a | 126 | mapping_set_update(&xas, mapping); |
c70b647d | 127 | |
5c024e6a MW |
128 | /* hugetlb pages are represented by a single entry in the xarray */ |
129 | if (!PageHuge(page)) { | |
130 | xas_set_order(&xas, page->index, compound_order(page)); | |
d8c6546b | 131 | nr = compound_nr(page); |
5c024e6a | 132 | } |
91b0abe3 | 133 | |
83929372 KS |
134 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
135 | VM_BUG_ON_PAGE(PageTail(page), page); | |
136 | VM_BUG_ON_PAGE(nr != 1 && shadow, page); | |
449dd698 | 137 | |
5c024e6a MW |
138 | xas_store(&xas, shadow); |
139 | xas_init_marks(&xas); | |
d3798ae8 | 140 | |
2300638b JK |
141 | page->mapping = NULL; |
142 | /* Leave page->index set: truncation lookup relies upon it */ | |
143 | ||
d3798ae8 JW |
144 | if (shadow) { |
145 | mapping->nrexceptional += nr; | |
146 | /* | |
147 | * Make sure the nrexceptional update is committed before | |
148 | * the nrpages update so that final truncate racing | |
149 | * with reclaim does not see both counters 0 at the | |
150 | * same time and miss a shadow entry. | |
151 | */ | |
152 | smp_wmb(); | |
153 | } | |
154 | mapping->nrpages -= nr; | |
91b0abe3 JW |
155 | } |
156 | ||
5ecc4d85 JK |
157 | static void unaccount_page_cache_page(struct address_space *mapping, |
158 | struct page *page) | |
1da177e4 | 159 | { |
5ecc4d85 | 160 | int nr; |
1da177e4 | 161 | |
c515e1fd DM |
162 | /* |
163 | * if we're uptodate, flush out into the cleancache, otherwise | |
164 | * invalidate any existing cleancache entries. We can't leave | |
165 | * stale data around in the cleancache once our page is gone | |
166 | */ | |
167 | if (PageUptodate(page) && PageMappedToDisk(page)) | |
168 | cleancache_put_page(page); | |
169 | else | |
3167760f | 170 | cleancache_invalidate_page(mapping, page); |
c515e1fd | 171 | |
83929372 | 172 | VM_BUG_ON_PAGE(PageTail(page), page); |
06b241f3 HD |
173 | VM_BUG_ON_PAGE(page_mapped(page), page); |
174 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) { | |
175 | int mapcount; | |
176 | ||
177 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
178 | current->comm, page_to_pfn(page)); | |
179 | dump_page(page, "still mapped when deleted"); | |
180 | dump_stack(); | |
181 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
182 | ||
183 | mapcount = page_mapcount(page); | |
184 | if (mapping_exiting(mapping) && | |
185 | page_count(page) >= mapcount + 2) { | |
186 | /* | |
187 | * All vmas have already been torn down, so it's | |
188 | * a good bet that actually the page is unmapped, | |
189 | * and we'd prefer not to leak it: if we're wrong, | |
190 | * some other bad page check should catch it later. | |
191 | */ | |
192 | page_mapcount_reset(page); | |
6d061f9f | 193 | page_ref_sub(page, mapcount); |
06b241f3 HD |
194 | } |
195 | } | |
196 | ||
4165b9b4 | 197 | /* hugetlb pages do not participate in page cache accounting. */ |
5ecc4d85 JK |
198 | if (PageHuge(page)) |
199 | return; | |
09612fa6 | 200 | |
6c357848 | 201 | nr = thp_nr_pages(page); |
5ecc4d85 | 202 | |
0d1c2072 | 203 | __mod_lruvec_page_state(page, NR_FILE_PAGES, -nr); |
5ecc4d85 | 204 | if (PageSwapBacked(page)) { |
0d1c2072 | 205 | __mod_lruvec_page_state(page, NR_SHMEM, -nr); |
5ecc4d85 JK |
206 | if (PageTransHuge(page)) |
207 | __dec_node_page_state(page, NR_SHMEM_THPS); | |
99cb0dbd SL |
208 | } else if (PageTransHuge(page)) { |
209 | __dec_node_page_state(page, NR_FILE_THPS); | |
09d91cda | 210 | filemap_nr_thps_dec(mapping); |
800d8c63 | 211 | } |
5ecc4d85 JK |
212 | |
213 | /* | |
214 | * At this point page must be either written or cleaned by | |
215 | * truncate. Dirty page here signals a bug and loss of | |
216 | * unwritten data. | |
217 | * | |
218 | * This fixes dirty accounting after removing the page entirely | |
219 | * but leaves PageDirty set: it has no effect for truncated | |
220 | * page and anyway will be cleared before returning page into | |
221 | * buddy allocator. | |
222 | */ | |
223 | if (WARN_ON_ONCE(PageDirty(page))) | |
224 | account_page_cleaned(page, mapping, inode_to_wb(mapping->host)); | |
225 | } | |
226 | ||
227 | /* | |
228 | * Delete a page from the page cache and free it. Caller has to make | |
229 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 230 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 JK |
231 | */ |
232 | void __delete_from_page_cache(struct page *page, void *shadow) | |
233 | { | |
234 | struct address_space *mapping = page->mapping; | |
235 | ||
236 | trace_mm_filemap_delete_from_page_cache(page); | |
237 | ||
238 | unaccount_page_cache_page(mapping, page); | |
5c024e6a | 239 | page_cache_delete(mapping, page, shadow); |
1da177e4 LT |
240 | } |
241 | ||
59c66c5f JK |
242 | static void page_cache_free_page(struct address_space *mapping, |
243 | struct page *page) | |
244 | { | |
245 | void (*freepage)(struct page *); | |
246 | ||
247 | freepage = mapping->a_ops->freepage; | |
248 | if (freepage) | |
249 | freepage(page); | |
250 | ||
251 | if (PageTransHuge(page) && !PageHuge(page)) { | |
252 | page_ref_sub(page, HPAGE_PMD_NR); | |
253 | VM_BUG_ON_PAGE(page_count(page) <= 0, page); | |
254 | } else { | |
255 | put_page(page); | |
256 | } | |
257 | } | |
258 | ||
702cfbf9 MK |
259 | /** |
260 | * delete_from_page_cache - delete page from page cache | |
261 | * @page: the page which the kernel is trying to remove from page cache | |
262 | * | |
263 | * This must be called only on pages that have been verified to be in the page | |
264 | * cache and locked. It will never put the page into the free list, the caller | |
265 | * has a reference on the page. | |
266 | */ | |
267 | void delete_from_page_cache(struct page *page) | |
1da177e4 | 268 | { |
83929372 | 269 | struct address_space *mapping = page_mapping(page); |
c4843a75 | 270 | unsigned long flags; |
1da177e4 | 271 | |
cd7619d6 | 272 | BUG_ON(!PageLocked(page)); |
b93b0163 | 273 | xa_lock_irqsave(&mapping->i_pages, flags); |
62cccb8c | 274 | __delete_from_page_cache(page, NULL); |
b93b0163 | 275 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
6072d13c | 276 | |
59c66c5f | 277 | page_cache_free_page(mapping, page); |
97cecb5a MK |
278 | } |
279 | EXPORT_SYMBOL(delete_from_page_cache); | |
280 | ||
aa65c29c | 281 | /* |
ef8e5717 | 282 | * page_cache_delete_batch - delete several pages from page cache |
aa65c29c JK |
283 | * @mapping: the mapping to which pages belong |
284 | * @pvec: pagevec with pages to delete | |
285 | * | |
b93b0163 | 286 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
4101196b MWO |
287 | * from the mapping. The function expects @pvec to be sorted by page index |
288 | * and is optimised for it to be dense. | |
b93b0163 | 289 | * It tolerates holes in @pvec (mapping entries at those indices are not |
aa65c29c | 290 | * modified). The function expects only THP head pages to be present in the |
4101196b | 291 | * @pvec. |
aa65c29c | 292 | * |
b93b0163 | 293 | * The function expects the i_pages lock to be held. |
aa65c29c | 294 | */ |
ef8e5717 | 295 | static void page_cache_delete_batch(struct address_space *mapping, |
aa65c29c JK |
296 | struct pagevec *pvec) |
297 | { | |
ef8e5717 | 298 | XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index); |
aa65c29c | 299 | int total_pages = 0; |
4101196b | 300 | int i = 0; |
aa65c29c | 301 | struct page *page; |
aa65c29c | 302 | |
ef8e5717 MW |
303 | mapping_set_update(&xas, mapping); |
304 | xas_for_each(&xas, page, ULONG_MAX) { | |
4101196b | 305 | if (i >= pagevec_count(pvec)) |
aa65c29c | 306 | break; |
4101196b MWO |
307 | |
308 | /* A swap/dax/shadow entry got inserted? Skip it. */ | |
3159f943 | 309 | if (xa_is_value(page)) |
aa65c29c | 310 | continue; |
4101196b MWO |
311 | /* |
312 | * A page got inserted in our range? Skip it. We have our | |
313 | * pages locked so they are protected from being removed. | |
314 | * If we see a page whose index is higher than ours, it | |
315 | * means our page has been removed, which shouldn't be | |
316 | * possible because we're holding the PageLock. | |
317 | */ | |
318 | if (page != pvec->pages[i]) { | |
319 | VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index, | |
320 | page); | |
321 | continue; | |
322 | } | |
323 | ||
324 | WARN_ON_ONCE(!PageLocked(page)); | |
325 | ||
326 | if (page->index == xas.xa_index) | |
aa65c29c | 327 | page->mapping = NULL; |
4101196b MWO |
328 | /* Leave page->index set: truncation lookup relies on it */ |
329 | ||
330 | /* | |
331 | * Move to the next page in the vector if this is a regular | |
332 | * page or the index is of the last sub-page of this compound | |
333 | * page. | |
334 | */ | |
335 | if (page->index + compound_nr(page) - 1 == xas.xa_index) | |
aa65c29c | 336 | i++; |
ef8e5717 | 337 | xas_store(&xas, NULL); |
aa65c29c JK |
338 | total_pages++; |
339 | } | |
340 | mapping->nrpages -= total_pages; | |
341 | } | |
342 | ||
343 | void delete_from_page_cache_batch(struct address_space *mapping, | |
344 | struct pagevec *pvec) | |
345 | { | |
346 | int i; | |
347 | unsigned long flags; | |
348 | ||
349 | if (!pagevec_count(pvec)) | |
350 | return; | |
351 | ||
b93b0163 | 352 | xa_lock_irqsave(&mapping->i_pages, flags); |
aa65c29c JK |
353 | for (i = 0; i < pagevec_count(pvec); i++) { |
354 | trace_mm_filemap_delete_from_page_cache(pvec->pages[i]); | |
355 | ||
356 | unaccount_page_cache_page(mapping, pvec->pages[i]); | |
357 | } | |
ef8e5717 | 358 | page_cache_delete_batch(mapping, pvec); |
b93b0163 | 359 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
aa65c29c JK |
360 | |
361 | for (i = 0; i < pagevec_count(pvec); i++) | |
362 | page_cache_free_page(mapping, pvec->pages[i]); | |
363 | } | |
364 | ||
d72d9e2a | 365 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
366 | { |
367 | int ret = 0; | |
368 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
369 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
370 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 371 | ret = -ENOSPC; |
7fcbbaf1 JA |
372 | if (test_bit(AS_EIO, &mapping->flags) && |
373 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
374 | ret = -EIO; |
375 | return ret; | |
376 | } | |
d72d9e2a | 377 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 378 | |
76341cab JL |
379 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
380 | { | |
381 | /* Check for outstanding write errors */ | |
382 | if (test_bit(AS_EIO, &mapping->flags)) | |
383 | return -EIO; | |
384 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
385 | return -ENOSPC; | |
386 | return 0; | |
387 | } | |
388 | ||
1da177e4 | 389 | /** |
485bb99b | 390 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
391 | * @mapping: address space structure to write |
392 | * @start: offset in bytes where the range starts | |
469eb4d0 | 393 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 394 | * @sync_mode: enable synchronous operation |
1da177e4 | 395 | * |
485bb99b RD |
396 | * Start writeback against all of a mapping's dirty pages that lie |
397 | * within the byte offsets <start, end> inclusive. | |
398 | * | |
1da177e4 | 399 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 400 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
401 | * these two operations is that if a dirty page/buffer is encountered, it must |
402 | * be waited upon, and not just skipped over. | |
a862f68a MR |
403 | * |
404 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 405 | */ |
ebcf28e1 AM |
406 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
407 | loff_t end, int sync_mode) | |
1da177e4 LT |
408 | { |
409 | int ret; | |
410 | struct writeback_control wbc = { | |
411 | .sync_mode = sync_mode, | |
05fe478d | 412 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
413 | .range_start = start, |
414 | .range_end = end, | |
1da177e4 LT |
415 | }; |
416 | ||
c3aab9a0 KK |
417 | if (!mapping_cap_writeback_dirty(mapping) || |
418 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
1da177e4 LT |
419 | return 0; |
420 | ||
b16b1deb | 421 | wbc_attach_fdatawrite_inode(&wbc, mapping->host); |
1da177e4 | 422 | ret = do_writepages(mapping, &wbc); |
b16b1deb | 423 | wbc_detach_inode(&wbc); |
1da177e4 LT |
424 | return ret; |
425 | } | |
426 | ||
427 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
428 | int sync_mode) | |
429 | { | |
111ebb6e | 430 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
431 | } |
432 | ||
433 | int filemap_fdatawrite(struct address_space *mapping) | |
434 | { | |
435 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
436 | } | |
437 | EXPORT_SYMBOL(filemap_fdatawrite); | |
438 | ||
f4c0a0fd | 439 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 440 | loff_t end) |
1da177e4 LT |
441 | { |
442 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
443 | } | |
f4c0a0fd | 444 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 445 | |
485bb99b RD |
446 | /** |
447 | * filemap_flush - mostly a non-blocking flush | |
448 | * @mapping: target address_space | |
449 | * | |
1da177e4 LT |
450 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
451 | * purposes - I/O may not be started against all dirty pages. | |
a862f68a MR |
452 | * |
453 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 LT |
454 | */ |
455 | int filemap_flush(struct address_space *mapping) | |
456 | { | |
457 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
458 | } | |
459 | EXPORT_SYMBOL(filemap_flush); | |
460 | ||
7fc9e472 GR |
461 | /** |
462 | * filemap_range_has_page - check if a page exists in range. | |
463 | * @mapping: address space within which to check | |
464 | * @start_byte: offset in bytes where the range starts | |
465 | * @end_byte: offset in bytes where the range ends (inclusive) | |
466 | * | |
467 | * Find at least one page in the range supplied, usually used to check if | |
468 | * direct writing in this range will trigger a writeback. | |
a862f68a MR |
469 | * |
470 | * Return: %true if at least one page exists in the specified range, | |
471 | * %false otherwise. | |
7fc9e472 GR |
472 | */ |
473 | bool filemap_range_has_page(struct address_space *mapping, | |
474 | loff_t start_byte, loff_t end_byte) | |
475 | { | |
f7b68046 | 476 | struct page *page; |
8fa8e538 MW |
477 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
478 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
7fc9e472 GR |
479 | |
480 | if (end_byte < start_byte) | |
481 | return false; | |
482 | ||
8fa8e538 MW |
483 | rcu_read_lock(); |
484 | for (;;) { | |
485 | page = xas_find(&xas, max); | |
486 | if (xas_retry(&xas, page)) | |
487 | continue; | |
488 | /* Shadow entries don't count */ | |
489 | if (xa_is_value(page)) | |
490 | continue; | |
491 | /* | |
492 | * We don't need to try to pin this page; we're about to | |
493 | * release the RCU lock anyway. It is enough to know that | |
494 | * there was a page here recently. | |
495 | */ | |
496 | break; | |
497 | } | |
498 | rcu_read_unlock(); | |
7fc9e472 | 499 | |
8fa8e538 | 500 | return page != NULL; |
7fc9e472 GR |
501 | } |
502 | EXPORT_SYMBOL(filemap_range_has_page); | |
503 | ||
5e8fcc1a | 504 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 505 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 506 | { |
09cbfeaf KS |
507 | pgoff_t index = start_byte >> PAGE_SHIFT; |
508 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
509 | struct pagevec pvec; |
510 | int nr_pages; | |
1da177e4 | 511 | |
94004ed7 | 512 | if (end_byte < start_byte) |
5e8fcc1a | 513 | return; |
1da177e4 | 514 | |
86679820 | 515 | pagevec_init(&pvec); |
312e9d2f | 516 | while (index <= end) { |
1da177e4 LT |
517 | unsigned i; |
518 | ||
312e9d2f | 519 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 520 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
521 | if (!nr_pages) |
522 | break; | |
523 | ||
1da177e4 LT |
524 | for (i = 0; i < nr_pages; i++) { |
525 | struct page *page = pvec.pages[i]; | |
526 | ||
1da177e4 | 527 | wait_on_page_writeback(page); |
5e8fcc1a | 528 | ClearPageError(page); |
1da177e4 LT |
529 | } |
530 | pagevec_release(&pvec); | |
531 | cond_resched(); | |
532 | } | |
aa750fd7 JN |
533 | } |
534 | ||
535 | /** | |
536 | * filemap_fdatawait_range - wait for writeback to complete | |
537 | * @mapping: address space structure to wait for | |
538 | * @start_byte: offset in bytes where the range starts | |
539 | * @end_byte: offset in bytes where the range ends (inclusive) | |
540 | * | |
541 | * Walk the list of under-writeback pages of the given address space | |
542 | * in the given range and wait for all of them. Check error status of | |
543 | * the address space and return it. | |
544 | * | |
545 | * Since the error status of the address space is cleared by this function, | |
546 | * callers are responsible for checking the return value and handling and/or | |
547 | * reporting the error. | |
a862f68a MR |
548 | * |
549 | * Return: error status of the address space. | |
aa750fd7 JN |
550 | */ |
551 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
552 | loff_t end_byte) | |
553 | { | |
5e8fcc1a JL |
554 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
555 | return filemap_check_errors(mapping); | |
1da177e4 | 556 | } |
d3bccb6f JK |
557 | EXPORT_SYMBOL(filemap_fdatawait_range); |
558 | ||
aa0bfcd9 RZ |
559 | /** |
560 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete | |
561 | * @mapping: address space structure to wait for | |
562 | * @start_byte: offset in bytes where the range starts | |
563 | * @end_byte: offset in bytes where the range ends (inclusive) | |
564 | * | |
565 | * Walk the list of under-writeback pages of the given address space in the | |
566 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), | |
567 | * this function does not clear error status of the address space. | |
568 | * | |
569 | * Use this function if callers don't handle errors themselves. Expected | |
570 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
571 | * fsfreeze(8) | |
572 | */ | |
573 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, | |
574 | loff_t start_byte, loff_t end_byte) | |
575 | { | |
576 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
577 | return filemap_check_and_keep_errors(mapping); | |
578 | } | |
579 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); | |
580 | ||
a823e458 JL |
581 | /** |
582 | * file_fdatawait_range - wait for writeback to complete | |
583 | * @file: file pointing to address space structure to wait for | |
584 | * @start_byte: offset in bytes where the range starts | |
585 | * @end_byte: offset in bytes where the range ends (inclusive) | |
586 | * | |
587 | * Walk the list of under-writeback pages of the address space that file | |
588 | * refers to, in the given range and wait for all of them. Check error | |
589 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
590 | * | |
591 | * Since the error status of the file is advanced by this function, | |
592 | * callers are responsible for checking the return value and handling and/or | |
593 | * reporting the error. | |
a862f68a MR |
594 | * |
595 | * Return: error status of the address space vs. the file->f_wb_err cursor. | |
a823e458 JL |
596 | */ |
597 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
598 | { | |
599 | struct address_space *mapping = file->f_mapping; | |
600 | ||
601 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
602 | return file_check_and_advance_wb_err(file); | |
603 | } | |
604 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 605 | |
aa750fd7 JN |
606 | /** |
607 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
608 | * @mapping: address space structure to wait for | |
609 | * | |
610 | * Walk the list of under-writeback pages of the given address space | |
611 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
612 | * does not clear error status of the address space. | |
613 | * | |
614 | * Use this function if callers don't handle errors themselves. Expected | |
615 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
616 | * fsfreeze(8) | |
a862f68a MR |
617 | * |
618 | * Return: error status of the address space. | |
aa750fd7 | 619 | */ |
76341cab | 620 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 621 | { |
ffb959bb | 622 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 623 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 624 | } |
76341cab | 625 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 626 | |
875d91b1 | 627 | /* Returns true if writeback might be needed or already in progress. */ |
9326c9b2 | 628 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 629 | { |
875d91b1 KK |
630 | if (dax_mapping(mapping)) |
631 | return mapping->nrexceptional; | |
632 | ||
633 | return mapping->nrpages; | |
1da177e4 | 634 | } |
1da177e4 | 635 | |
485bb99b RD |
636 | /** |
637 | * filemap_write_and_wait_range - write out & wait on a file range | |
638 | * @mapping: the address_space for the pages | |
639 | * @lstart: offset in bytes where the range starts | |
640 | * @lend: offset in bytes where the range ends (inclusive) | |
641 | * | |
469eb4d0 AM |
642 | * Write out and wait upon file offsets lstart->lend, inclusive. |
643 | * | |
0e056eb5 | 644 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 | 645 | * that this function can be used to write to the very end-of-file (end = -1). |
a862f68a MR |
646 | * |
647 | * Return: error status of the address space. | |
469eb4d0 | 648 | */ |
1da177e4 LT |
649 | int filemap_write_and_wait_range(struct address_space *mapping, |
650 | loff_t lstart, loff_t lend) | |
651 | { | |
28fd1298 | 652 | int err = 0; |
1da177e4 | 653 | |
9326c9b2 | 654 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
655 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
656 | WB_SYNC_ALL); | |
ddf8f376 IW |
657 | /* |
658 | * Even if the above returned error, the pages may be | |
659 | * written partially (e.g. -ENOSPC), so we wait for it. | |
660 | * But the -EIO is special case, it may indicate the worst | |
661 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
662 | */ | |
28fd1298 | 663 | if (err != -EIO) { |
94004ed7 CH |
664 | int err2 = filemap_fdatawait_range(mapping, |
665 | lstart, lend); | |
28fd1298 OH |
666 | if (!err) |
667 | err = err2; | |
cbeaf951 JL |
668 | } else { |
669 | /* Clear any previously stored errors */ | |
670 | filemap_check_errors(mapping); | |
28fd1298 | 671 | } |
865ffef3 DM |
672 | } else { |
673 | err = filemap_check_errors(mapping); | |
1da177e4 | 674 | } |
28fd1298 | 675 | return err; |
1da177e4 | 676 | } |
f6995585 | 677 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 678 | |
5660e13d JL |
679 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
680 | { | |
3acdfd28 | 681 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
682 | |
683 | trace_filemap_set_wb_err(mapping, eseq); | |
684 | } | |
685 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
686 | ||
687 | /** | |
688 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
689 | * and advance wb_err to current one | |
690 | * @file: struct file on which the error is being reported | |
691 | * | |
692 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
693 | * want to report any writeback errors that occurred since the last fsync (or | |
694 | * since the file was opened if there haven't been any). | |
695 | * | |
696 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
697 | * then just quickly return 0. The file is all caught up. | |
698 | * | |
699 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
700 | * it and try to swap it into place. If it works, or another task beat us | |
701 | * to it with the new value, then update the f_wb_err and return the error | |
702 | * portion. The error at this point must be reported via proper channels | |
703 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
704 | * | |
705 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
706 | * value is protected by the f_lock since we must ensure that it reflects | |
707 | * the latest value swapped in for this file descriptor. | |
a862f68a MR |
708 | * |
709 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
710 | */ |
711 | int file_check_and_advance_wb_err(struct file *file) | |
712 | { | |
713 | int err = 0; | |
714 | errseq_t old = READ_ONCE(file->f_wb_err); | |
715 | struct address_space *mapping = file->f_mapping; | |
716 | ||
717 | /* Locklessly handle the common case where nothing has changed */ | |
718 | if (errseq_check(&mapping->wb_err, old)) { | |
719 | /* Something changed, must use slow path */ | |
720 | spin_lock(&file->f_lock); | |
721 | old = file->f_wb_err; | |
722 | err = errseq_check_and_advance(&mapping->wb_err, | |
723 | &file->f_wb_err); | |
724 | trace_file_check_and_advance_wb_err(file, old); | |
725 | spin_unlock(&file->f_lock); | |
726 | } | |
f4e222c5 JL |
727 | |
728 | /* | |
729 | * We're mostly using this function as a drop in replacement for | |
730 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
731 | * that the legacy code would have had on these flags. | |
732 | */ | |
733 | clear_bit(AS_EIO, &mapping->flags); | |
734 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
735 | return err; |
736 | } | |
737 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
738 | ||
739 | /** | |
740 | * file_write_and_wait_range - write out & wait on a file range | |
741 | * @file: file pointing to address_space with pages | |
742 | * @lstart: offset in bytes where the range starts | |
743 | * @lend: offset in bytes where the range ends (inclusive) | |
744 | * | |
745 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
746 | * | |
747 | * Note that @lend is inclusive (describes the last byte to be written) so | |
748 | * that this function can be used to write to the very end-of-file (end = -1). | |
749 | * | |
750 | * After writing out and waiting on the data, we check and advance the | |
751 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
a862f68a MR |
752 | * |
753 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
754 | */ |
755 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
756 | { | |
757 | int err = 0, err2; | |
758 | struct address_space *mapping = file->f_mapping; | |
759 | ||
9326c9b2 | 760 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
761 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
762 | WB_SYNC_ALL); | |
763 | /* See comment of filemap_write_and_wait() */ | |
764 | if (err != -EIO) | |
765 | __filemap_fdatawait_range(mapping, lstart, lend); | |
766 | } | |
767 | err2 = file_check_and_advance_wb_err(file); | |
768 | if (!err) | |
769 | err = err2; | |
770 | return err; | |
771 | } | |
772 | EXPORT_SYMBOL(file_write_and_wait_range); | |
773 | ||
ef6a3c63 MS |
774 | /** |
775 | * replace_page_cache_page - replace a pagecache page with a new one | |
776 | * @old: page to be replaced | |
777 | * @new: page to replace with | |
778 | * @gfp_mask: allocation mode | |
779 | * | |
780 | * This function replaces a page in the pagecache with a new one. On | |
781 | * success it acquires the pagecache reference for the new page and | |
782 | * drops it for the old page. Both the old and new pages must be | |
783 | * locked. This function does not add the new page to the LRU, the | |
784 | * caller must do that. | |
785 | * | |
74d60958 | 786 | * The remove + add is atomic. This function cannot fail. |
a862f68a MR |
787 | * |
788 | * Return: %0 | |
ef6a3c63 MS |
789 | */ |
790 | int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) | |
791 | { | |
74d60958 MW |
792 | struct address_space *mapping = old->mapping; |
793 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
794 | pgoff_t offset = old->index; | |
795 | XA_STATE(xas, &mapping->i_pages, offset); | |
796 | unsigned long flags; | |
ef6a3c63 | 797 | |
309381fe SL |
798 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
799 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
800 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 801 | |
74d60958 MW |
802 | get_page(new); |
803 | new->mapping = mapping; | |
804 | new->index = offset; | |
ef6a3c63 | 805 | |
0d1c2072 JW |
806 | mem_cgroup_migrate(old, new); |
807 | ||
74d60958 MW |
808 | xas_lock_irqsave(&xas, flags); |
809 | xas_store(&xas, new); | |
4165b9b4 | 810 | |
74d60958 MW |
811 | old->mapping = NULL; |
812 | /* hugetlb pages do not participate in page cache accounting. */ | |
813 | if (!PageHuge(old)) | |
0d1c2072 | 814 | __dec_lruvec_page_state(old, NR_FILE_PAGES); |
74d60958 | 815 | if (!PageHuge(new)) |
0d1c2072 | 816 | __inc_lruvec_page_state(new, NR_FILE_PAGES); |
74d60958 | 817 | if (PageSwapBacked(old)) |
0d1c2072 | 818 | __dec_lruvec_page_state(old, NR_SHMEM); |
74d60958 | 819 | if (PageSwapBacked(new)) |
0d1c2072 | 820 | __inc_lruvec_page_state(new, NR_SHMEM); |
74d60958 | 821 | xas_unlock_irqrestore(&xas, flags); |
74d60958 MW |
822 | if (freepage) |
823 | freepage(old); | |
824 | put_page(old); | |
ef6a3c63 | 825 | |
74d60958 | 826 | return 0; |
ef6a3c63 MS |
827 | } |
828 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
829 | ||
a528910e JW |
830 | static int __add_to_page_cache_locked(struct page *page, |
831 | struct address_space *mapping, | |
832 | pgoff_t offset, gfp_t gfp_mask, | |
833 | void **shadowp) | |
1da177e4 | 834 | { |
74d60958 | 835 | XA_STATE(xas, &mapping->i_pages, offset); |
00501b53 | 836 | int huge = PageHuge(page); |
e286781d | 837 | int error; |
74d60958 | 838 | void *old; |
e286781d | 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) { |
d9eb1ea2 | 849 | error = mem_cgroup_charge(page, current->mm, gfp_mask); |
3fea5a49 JW |
850 | if (error) |
851 | goto error; | |
852 | } | |
853 | ||
74d60958 MW |
854 | do { |
855 | xas_lock_irq(&xas); | |
856 | old = xas_load(&xas); | |
857 | if (old && !xa_is_value(old)) | |
858 | xas_set_err(&xas, -EEXIST); | |
859 | xas_store(&xas, page); | |
860 | if (xas_error(&xas)) | |
861 | goto unlock; | |
862 | ||
863 | if (xa_is_value(old)) { | |
864 | mapping->nrexceptional--; | |
865 | if (shadowp) | |
866 | *shadowp = old; | |
867 | } | |
868 | mapping->nrpages++; | |
869 | ||
870 | /* hugetlb pages do not participate in page cache accounting */ | |
871 | if (!huge) | |
0d1c2072 | 872 | __inc_lruvec_page_state(page, NR_FILE_PAGES); |
74d60958 MW |
873 | unlock: |
874 | xas_unlock_irq(&xas); | |
875 | } while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK)); | |
876 | ||
3fea5a49 JW |
877 | if (xas_error(&xas)) { |
878 | error = xas_error(&xas); | |
74d60958 | 879 | goto error; |
3fea5a49 | 880 | } |
4165b9b4 | 881 | |
66a0c8ee KS |
882 | trace_mm_filemap_add_to_page_cache(page); |
883 | return 0; | |
74d60958 | 884 | error: |
66a0c8ee KS |
885 | page->mapping = NULL; |
886 | /* Leave page->index set: truncation relies upon it */ | |
09cbfeaf | 887 | put_page(page); |
3fea5a49 | 888 | return error; |
1da177e4 | 889 | } |
cfcbfb13 | 890 | ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO); |
a528910e JW |
891 | |
892 | /** | |
893 | * add_to_page_cache_locked - add a locked page to the pagecache | |
894 | * @page: page to add | |
895 | * @mapping: the page's address_space | |
896 | * @offset: page index | |
897 | * @gfp_mask: page allocation mode | |
898 | * | |
899 | * This function is used to add a page to the pagecache. It must be locked. | |
900 | * This function does not add the page to the LRU. The caller must do that. | |
a862f68a MR |
901 | * |
902 | * Return: %0 on success, negative error code otherwise. | |
a528910e JW |
903 | */ |
904 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
905 | pgoff_t offset, gfp_t gfp_mask) | |
906 | { | |
907 | return __add_to_page_cache_locked(page, mapping, offset, | |
908 | gfp_mask, NULL); | |
909 | } | |
e286781d | 910 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 LT |
911 | |
912 | int add_to_page_cache_lru(struct page *page, struct address_space *mapping, | |
6daa0e28 | 913 | pgoff_t offset, gfp_t gfp_mask) |
1da177e4 | 914 | { |
a528910e | 915 | void *shadow = NULL; |
4f98a2fe RR |
916 | int ret; |
917 | ||
48c935ad | 918 | __SetPageLocked(page); |
a528910e JW |
919 | ret = __add_to_page_cache_locked(page, mapping, offset, |
920 | gfp_mask, &shadow); | |
921 | if (unlikely(ret)) | |
48c935ad | 922 | __ClearPageLocked(page); |
a528910e JW |
923 | else { |
924 | /* | |
925 | * The page might have been evicted from cache only | |
926 | * recently, in which case it should be activated like | |
927 | * any other repeatedly accessed page. | |
f0281a00 RR |
928 | * The exception is pages getting rewritten; evicting other |
929 | * data from the working set, only to cache data that will | |
930 | * get overwritten with something else, is a waste of memory. | |
a528910e | 931 | */ |
1899ad18 JW |
932 | WARN_ON_ONCE(PageActive(page)); |
933 | if (!(gfp_mask & __GFP_WRITE) && shadow) | |
934 | workingset_refault(page, shadow); | |
a528910e JW |
935 | lru_cache_add(page); |
936 | } | |
1da177e4 LT |
937 | return ret; |
938 | } | |
18bc0bbd | 939 | EXPORT_SYMBOL_GPL(add_to_page_cache_lru); |
1da177e4 | 940 | |
44110fe3 | 941 | #ifdef CONFIG_NUMA |
2ae88149 | 942 | struct page *__page_cache_alloc(gfp_t gfp) |
44110fe3 | 943 | { |
c0ff7453 MX |
944 | int n; |
945 | struct page *page; | |
946 | ||
44110fe3 | 947 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
948 | unsigned int cpuset_mems_cookie; |
949 | do { | |
d26914d1 | 950 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 951 | n = cpuset_mem_spread_node(); |
96db800f | 952 | page = __alloc_pages_node(n, gfp, 0); |
d26914d1 | 953 | } while (!page && read_mems_allowed_retry(cpuset_mems_cookie)); |
cc9a6c87 | 954 | |
c0ff7453 | 955 | return page; |
44110fe3 | 956 | } |
2ae88149 | 957 | return alloc_pages(gfp, 0); |
44110fe3 | 958 | } |
2ae88149 | 959 | EXPORT_SYMBOL(__page_cache_alloc); |
44110fe3 PJ |
960 | #endif |
961 | ||
1da177e4 LT |
962 | /* |
963 | * In order to wait for pages to become available there must be | |
964 | * waitqueues associated with pages. By using a hash table of | |
965 | * waitqueues where the bucket discipline is to maintain all | |
966 | * waiters on the same queue and wake all when any of the pages | |
967 | * become available, and for the woken contexts to check to be | |
968 | * sure the appropriate page became available, this saves space | |
969 | * at a cost of "thundering herd" phenomena during rare hash | |
970 | * collisions. | |
971 | */ | |
62906027 NP |
972 | #define PAGE_WAIT_TABLE_BITS 8 |
973 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
974 | static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; | |
975 | ||
976 | static wait_queue_head_t *page_waitqueue(struct page *page) | |
1da177e4 | 977 | { |
62906027 | 978 | return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 979 | } |
1da177e4 | 980 | |
62906027 | 981 | void __init pagecache_init(void) |
1da177e4 | 982 | { |
62906027 | 983 | int i; |
1da177e4 | 984 | |
62906027 NP |
985 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
986 | init_waitqueue_head(&page_wait_table[i]); | |
987 | ||
988 | page_writeback_init(); | |
1da177e4 | 989 | } |
1da177e4 | 990 | |
5ef64cc8 LT |
991 | /* |
992 | * The page wait code treats the "wait->flags" somewhat unusually, because | |
5868ec26 | 993 | * we have multiple different kinds of waits, not just the usual "exclusive" |
5ef64cc8 LT |
994 | * one. |
995 | * | |
996 | * We have: | |
997 | * | |
998 | * (a) no special bits set: | |
999 | * | |
1000 | * We're just waiting for the bit to be released, and when a waker | |
1001 | * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up, | |
1002 | * and remove it from the wait queue. | |
1003 | * | |
1004 | * Simple and straightforward. | |
1005 | * | |
1006 | * (b) WQ_FLAG_EXCLUSIVE: | |
1007 | * | |
1008 | * The waiter is waiting to get the lock, and only one waiter should | |
1009 | * be woken up to avoid any thundering herd behavior. We'll set the | |
1010 | * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue. | |
1011 | * | |
1012 | * This is the traditional exclusive wait. | |
1013 | * | |
5868ec26 | 1014 | * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: |
5ef64cc8 LT |
1015 | * |
1016 | * The waiter is waiting to get the bit, and additionally wants the | |
1017 | * lock to be transferred to it for fair lock behavior. If the lock | |
1018 | * cannot be taken, we stop walking the wait queue without waking | |
1019 | * the waiter. | |
1020 | * | |
1021 | * This is the "fair lock handoff" case, and in addition to setting | |
1022 | * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see | |
1023 | * that it now has the lock. | |
1024 | */ | |
ac6424b9 | 1025 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 1026 | { |
5ef64cc8 | 1027 | unsigned int flags; |
62906027 NP |
1028 | struct wait_page_key *key = arg; |
1029 | struct wait_page_queue *wait_page | |
1030 | = container_of(wait, struct wait_page_queue, wait); | |
1031 | ||
cdc8fcb4 | 1032 | if (!wake_page_match(wait_page, key)) |
62906027 | 1033 | return 0; |
3510ca20 | 1034 | |
9a1ea439 | 1035 | /* |
5ef64cc8 LT |
1036 | * If it's a lock handoff wait, we get the bit for it, and |
1037 | * stop walking (and do not wake it up) if we can't. | |
9a1ea439 | 1038 | */ |
5ef64cc8 LT |
1039 | flags = wait->flags; |
1040 | if (flags & WQ_FLAG_EXCLUSIVE) { | |
1041 | if (test_bit(key->bit_nr, &key->page->flags)) | |
2a9127fc | 1042 | return -1; |
5ef64cc8 LT |
1043 | if (flags & WQ_FLAG_CUSTOM) { |
1044 | if (test_and_set_bit(key->bit_nr, &key->page->flags)) | |
1045 | return -1; | |
1046 | flags |= WQ_FLAG_DONE; | |
1047 | } | |
2a9127fc | 1048 | } |
f62e00cc | 1049 | |
5ef64cc8 LT |
1050 | /* |
1051 | * We are holding the wait-queue lock, but the waiter that | |
1052 | * is waiting for this will be checking the flags without | |
1053 | * any locking. | |
1054 | * | |
1055 | * So update the flags atomically, and wake up the waiter | |
1056 | * afterwards to avoid any races. This store-release pairs | |
1057 | * with the load-acquire in wait_on_page_bit_common(). | |
1058 | */ | |
1059 | smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN); | |
2a9127fc LT |
1060 | wake_up_state(wait->private, mode); |
1061 | ||
1062 | /* | |
1063 | * Ok, we have successfully done what we're waiting for, | |
1064 | * and we can unconditionally remove the wait entry. | |
1065 | * | |
5ef64cc8 LT |
1066 | * Note that this pairs with the "finish_wait()" in the |
1067 | * waiter, and has to be the absolute last thing we do. | |
1068 | * After this list_del_init(&wait->entry) the wait entry | |
2a9127fc LT |
1069 | * might be de-allocated and the process might even have |
1070 | * exited. | |
2a9127fc | 1071 | */ |
c6fe44d9 | 1072 | list_del_init_careful(&wait->entry); |
5ef64cc8 | 1073 | return (flags & WQ_FLAG_EXCLUSIVE) != 0; |
f62e00cc KM |
1074 | } |
1075 | ||
74d81bfa | 1076 | static void wake_up_page_bit(struct page *page, int bit_nr) |
cbbce822 | 1077 | { |
62906027 NP |
1078 | wait_queue_head_t *q = page_waitqueue(page); |
1079 | struct wait_page_key key; | |
1080 | unsigned long flags; | |
11a19c7b | 1081 | wait_queue_entry_t bookmark; |
cbbce822 | 1082 | |
62906027 NP |
1083 | key.page = page; |
1084 | key.bit_nr = bit_nr; | |
1085 | key.page_match = 0; | |
1086 | ||
11a19c7b TC |
1087 | bookmark.flags = 0; |
1088 | bookmark.private = NULL; | |
1089 | bookmark.func = NULL; | |
1090 | INIT_LIST_HEAD(&bookmark.entry); | |
1091 | ||
62906027 | 1092 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1093 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1094 | ||
1095 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1096 | /* | |
1097 | * Take a breather from holding the lock, | |
1098 | * allow pages that finish wake up asynchronously | |
1099 | * to acquire the lock and remove themselves | |
1100 | * from wait queue | |
1101 | */ | |
1102 | spin_unlock_irqrestore(&q->lock, flags); | |
1103 | cpu_relax(); | |
1104 | spin_lock_irqsave(&q->lock, flags); | |
1105 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1106 | } | |
1107 | ||
62906027 NP |
1108 | /* |
1109 | * It is possible for other pages to have collided on the waitqueue | |
1110 | * hash, so in that case check for a page match. That prevents a long- | |
1111 | * term waiter | |
1112 | * | |
1113 | * It is still possible to miss a case here, when we woke page waiters | |
1114 | * and removed them from the waitqueue, but there are still other | |
1115 | * page waiters. | |
1116 | */ | |
1117 | if (!waitqueue_active(q) || !key.page_match) { | |
1118 | ClearPageWaiters(page); | |
1119 | /* | |
1120 | * It's possible to miss clearing Waiters here, when we woke | |
1121 | * our page waiters, but the hashed waitqueue has waiters for | |
1122 | * other pages on it. | |
1123 | * | |
1124 | * That's okay, it's a rare case. The next waker will clear it. | |
1125 | */ | |
1126 | } | |
1127 | spin_unlock_irqrestore(&q->lock, flags); | |
1128 | } | |
74d81bfa NP |
1129 | |
1130 | static void wake_up_page(struct page *page, int bit) | |
1131 | { | |
1132 | if (!PageWaiters(page)) | |
1133 | return; | |
1134 | wake_up_page_bit(page, bit); | |
1135 | } | |
62906027 | 1136 | |
9a1ea439 HD |
1137 | /* |
1138 | * A choice of three behaviors for wait_on_page_bit_common(): | |
1139 | */ | |
1140 | enum behavior { | |
1141 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like | |
1142 | * __lock_page() waiting on then setting PG_locked. | |
1143 | */ | |
1144 | SHARED, /* Hold ref to page and check the bit when woken, like | |
1145 | * wait_on_page_writeback() waiting on PG_writeback. | |
1146 | */ | |
1147 | DROP, /* Drop ref to page before wait, no check when woken, | |
1148 | * like put_and_wait_on_page_locked() on PG_locked. | |
1149 | */ | |
1150 | }; | |
1151 | ||
2a9127fc | 1152 | /* |
5ef64cc8 LT |
1153 | * Attempt to check (or get) the page bit, and mark us done |
1154 | * if successful. | |
2a9127fc LT |
1155 | */ |
1156 | static inline bool trylock_page_bit_common(struct page *page, int bit_nr, | |
1157 | struct wait_queue_entry *wait) | |
1158 | { | |
1159 | if (wait->flags & WQ_FLAG_EXCLUSIVE) { | |
1160 | if (test_and_set_bit(bit_nr, &page->flags)) | |
1161 | return false; | |
1162 | } else if (test_bit(bit_nr, &page->flags)) | |
1163 | return false; | |
1164 | ||
5ef64cc8 | 1165 | wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE; |
2a9127fc LT |
1166 | return true; |
1167 | } | |
1168 | ||
5ef64cc8 LT |
1169 | /* How many times do we accept lock stealing from under a waiter? */ |
1170 | int sysctl_page_lock_unfairness = 5; | |
1171 | ||
62906027 | 1172 | static inline int wait_on_page_bit_common(wait_queue_head_t *q, |
9a1ea439 | 1173 | struct page *page, int bit_nr, int state, enum behavior behavior) |
62906027 | 1174 | { |
5ef64cc8 | 1175 | int unfairness = sysctl_page_lock_unfairness; |
62906027 | 1176 | struct wait_page_queue wait_page; |
ac6424b9 | 1177 | wait_queue_entry_t *wait = &wait_page.wait; |
b1d29ba8 | 1178 | bool thrashing = false; |
9a1ea439 | 1179 | bool delayacct = false; |
eb414681 | 1180 | unsigned long pflags; |
62906027 | 1181 | |
eb414681 | 1182 | if (bit_nr == PG_locked && |
b1d29ba8 | 1183 | !PageUptodate(page) && PageWorkingset(page)) { |
9a1ea439 | 1184 | if (!PageSwapBacked(page)) { |
eb414681 | 1185 | delayacct_thrashing_start(); |
9a1ea439 HD |
1186 | delayacct = true; |
1187 | } | |
eb414681 | 1188 | psi_memstall_enter(&pflags); |
b1d29ba8 JW |
1189 | thrashing = true; |
1190 | } | |
1191 | ||
62906027 NP |
1192 | init_wait(wait); |
1193 | wait->func = wake_page_function; | |
1194 | wait_page.page = page; | |
1195 | wait_page.bit_nr = bit_nr; | |
1196 | ||
5ef64cc8 LT |
1197 | repeat: |
1198 | wait->flags = 0; | |
1199 | if (behavior == EXCLUSIVE) { | |
1200 | wait->flags = WQ_FLAG_EXCLUSIVE; | |
1201 | if (--unfairness < 0) | |
1202 | wait->flags |= WQ_FLAG_CUSTOM; | |
1203 | } | |
1204 | ||
2a9127fc LT |
1205 | /* |
1206 | * Do one last check whether we can get the | |
1207 | * page bit synchronously. | |
1208 | * | |
1209 | * Do the SetPageWaiters() marking before that | |
1210 | * to let any waker we _just_ missed know they | |
1211 | * need to wake us up (otherwise they'll never | |
1212 | * even go to the slow case that looks at the | |
1213 | * page queue), and add ourselves to the wait | |
1214 | * queue if we need to sleep. | |
1215 | * | |
1216 | * This part needs to be done under the queue | |
1217 | * lock to avoid races. | |
1218 | */ | |
1219 | spin_lock_irq(&q->lock); | |
1220 | SetPageWaiters(page); | |
1221 | if (!trylock_page_bit_common(page, bit_nr, wait)) | |
1222 | __add_wait_queue_entry_tail(q, wait); | |
1223 | spin_unlock_irq(&q->lock); | |
62906027 | 1224 | |
2a9127fc LT |
1225 | /* |
1226 | * From now on, all the logic will be based on | |
5ef64cc8 LT |
1227 | * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to |
1228 | * see whether the page bit testing has already | |
1229 | * been done by the wake function. | |
2a9127fc LT |
1230 | * |
1231 | * We can drop our reference to the page. | |
1232 | */ | |
1233 | if (behavior == DROP) | |
1234 | put_page(page); | |
62906027 | 1235 | |
5ef64cc8 LT |
1236 | /* |
1237 | * Note that until the "finish_wait()", or until | |
1238 | * we see the WQ_FLAG_WOKEN flag, we need to | |
1239 | * be very careful with the 'wait->flags', because | |
1240 | * we may race with a waker that sets them. | |
1241 | */ | |
2a9127fc | 1242 | for (;;) { |
5ef64cc8 LT |
1243 | unsigned int flags; |
1244 | ||
62906027 NP |
1245 | set_current_state(state); |
1246 | ||
5ef64cc8 LT |
1247 | /* Loop until we've been woken or interrupted */ |
1248 | flags = smp_load_acquire(&wait->flags); | |
1249 | if (!(flags & WQ_FLAG_WOKEN)) { | |
1250 | if (signal_pending_state(state, current)) | |
1251 | break; | |
1252 | ||
1253 | io_schedule(); | |
1254 | continue; | |
1255 | } | |
1256 | ||
1257 | /* If we were non-exclusive, we're done */ | |
1258 | if (behavior != EXCLUSIVE) | |
a8b169af | 1259 | break; |
9a1ea439 | 1260 | |
5ef64cc8 LT |
1261 | /* If the waker got the lock for us, we're done */ |
1262 | if (flags & WQ_FLAG_DONE) | |
9a1ea439 | 1263 | break; |
2a9127fc | 1264 | |
5ef64cc8 LT |
1265 | /* |
1266 | * Otherwise, if we're getting the lock, we need to | |
1267 | * try to get it ourselves. | |
1268 | * | |
1269 | * And if that fails, we'll have to retry this all. | |
1270 | */ | |
1271 | if (unlikely(test_and_set_bit(bit_nr, &page->flags))) | |
1272 | goto repeat; | |
1273 | ||
1274 | wait->flags |= WQ_FLAG_DONE; | |
1275 | break; | |
62906027 NP |
1276 | } |
1277 | ||
5ef64cc8 LT |
1278 | /* |
1279 | * If a signal happened, this 'finish_wait()' may remove the last | |
1280 | * waiter from the wait-queues, but the PageWaiters bit will remain | |
1281 | * set. That's ok. The next wakeup will take care of it, and trying | |
1282 | * to do it here would be difficult and prone to races. | |
1283 | */ | |
62906027 NP |
1284 | finish_wait(q, wait); |
1285 | ||
eb414681 | 1286 | if (thrashing) { |
9a1ea439 | 1287 | if (delayacct) |
eb414681 JW |
1288 | delayacct_thrashing_end(); |
1289 | psi_memstall_leave(&pflags); | |
1290 | } | |
b1d29ba8 | 1291 | |
62906027 | 1292 | /* |
5ef64cc8 LT |
1293 | * NOTE! The wait->flags weren't stable until we've done the |
1294 | * 'finish_wait()', and we could have exited the loop above due | |
1295 | * to a signal, and had a wakeup event happen after the signal | |
1296 | * test but before the 'finish_wait()'. | |
1297 | * | |
1298 | * So only after the finish_wait() can we reliably determine | |
1299 | * if we got woken up or not, so we can now figure out the final | |
1300 | * return value based on that state without races. | |
1301 | * | |
1302 | * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive | |
1303 | * waiter, but an exclusive one requires WQ_FLAG_DONE. | |
62906027 | 1304 | */ |
5ef64cc8 LT |
1305 | if (behavior == EXCLUSIVE) |
1306 | return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR; | |
62906027 | 1307 | |
2a9127fc | 1308 | return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR; |
62906027 NP |
1309 | } |
1310 | ||
1311 | void wait_on_page_bit(struct page *page, int bit_nr) | |
1312 | { | |
1313 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 | 1314 | wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
62906027 NP |
1315 | } |
1316 | EXPORT_SYMBOL(wait_on_page_bit); | |
1317 | ||
1318 | int wait_on_page_bit_killable(struct page *page, int bit_nr) | |
1319 | { | |
1320 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 | 1321 | return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED); |
cbbce822 | 1322 | } |
4343d008 | 1323 | EXPORT_SYMBOL(wait_on_page_bit_killable); |
cbbce822 | 1324 | |
dd3e6d50 JA |
1325 | static int __wait_on_page_locked_async(struct page *page, |
1326 | struct wait_page_queue *wait, bool set) | |
1327 | { | |
1328 | struct wait_queue_head *q = page_waitqueue(page); | |
1329 | int ret = 0; | |
1330 | ||
1331 | wait->page = page; | |
1332 | wait->bit_nr = PG_locked; | |
1333 | ||
1334 | spin_lock_irq(&q->lock); | |
1335 | __add_wait_queue_entry_tail(q, &wait->wait); | |
1336 | SetPageWaiters(page); | |
1337 | if (set) | |
1338 | ret = !trylock_page(page); | |
1339 | else | |
1340 | ret = PageLocked(page); | |
1341 | /* | |
1342 | * If we were succesful now, we know we're still on the | |
1343 | * waitqueue as we're still under the lock. This means it's | |
1344 | * safe to remove and return success, we know the callback | |
1345 | * isn't going to trigger. | |
1346 | */ | |
1347 | if (!ret) | |
1348 | __remove_wait_queue(q, &wait->wait); | |
1349 | else | |
1350 | ret = -EIOCBQUEUED; | |
1351 | spin_unlock_irq(&q->lock); | |
1352 | return ret; | |
1353 | } | |
1354 | ||
1a0a7853 JA |
1355 | static int wait_on_page_locked_async(struct page *page, |
1356 | struct wait_page_queue *wait) | |
1357 | { | |
1358 | if (!PageLocked(page)) | |
1359 | return 0; | |
1360 | return __wait_on_page_locked_async(compound_head(page), wait, false); | |
1361 | } | |
1362 | ||
9a1ea439 HD |
1363 | /** |
1364 | * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked | |
1365 | * @page: The page to wait for. | |
1366 | * | |
1367 | * The caller should hold a reference on @page. They expect the page to | |
1368 | * become unlocked relatively soon, but do not wish to hold up migration | |
1369 | * (for example) by holding the reference while waiting for the page to | |
1370 | * come unlocked. After this function returns, the caller should not | |
1371 | * dereference @page. | |
1372 | */ | |
1373 | void put_and_wait_on_page_locked(struct page *page) | |
1374 | { | |
1375 | wait_queue_head_t *q; | |
1376 | ||
1377 | page = compound_head(page); | |
1378 | q = page_waitqueue(page); | |
1379 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP); | |
1380 | } | |
1381 | ||
385e1ca5 DH |
1382 | /** |
1383 | * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue | |
697f619f RD |
1384 | * @page: Page defining the wait queue of interest |
1385 | * @waiter: Waiter to add to the queue | |
385e1ca5 DH |
1386 | * |
1387 | * Add an arbitrary @waiter to the wait queue for the nominated @page. | |
1388 | */ | |
ac6424b9 | 1389 | void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter) |
385e1ca5 DH |
1390 | { |
1391 | wait_queue_head_t *q = page_waitqueue(page); | |
1392 | unsigned long flags; | |
1393 | ||
1394 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1395 | __add_wait_queue_entry_tail(q, waiter); |
62906027 | 1396 | SetPageWaiters(page); |
385e1ca5 DH |
1397 | spin_unlock_irqrestore(&q->lock, flags); |
1398 | } | |
1399 | EXPORT_SYMBOL_GPL(add_page_wait_queue); | |
1400 | ||
b91e1302 LT |
1401 | #ifndef clear_bit_unlock_is_negative_byte |
1402 | ||
1403 | /* | |
1404 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1405 | * | |
1406 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1407 | * test the sign bit at the same time. But if the architecture does | |
1408 | * not support that special operation, we just do this all by hand | |
1409 | * instead. | |
1410 | * | |
1411 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
ffceeb62 | 1412 | * being cleared, but a memory barrier should be unnecessary since it is |
b91e1302 LT |
1413 | * in the same byte as PG_locked. |
1414 | */ | |
1415 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1416 | { | |
1417 | clear_bit_unlock(nr, mem); | |
1418 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1419 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1420 | } |
1421 | ||
1422 | #endif | |
1423 | ||
1da177e4 | 1424 | /** |
485bb99b | 1425 | * unlock_page - unlock a locked page |
1da177e4 LT |
1426 | * @page: the page |
1427 | * | |
1428 | * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). | |
1429 | * Also wakes sleepers in wait_on_page_writeback() because the wakeup | |
da3dae54 | 1430 | * mechanism between PageLocked pages and PageWriteback pages is shared. |
1da177e4 LT |
1431 | * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. |
1432 | * | |
b91e1302 LT |
1433 | * Note that this depends on PG_waiters being the sign bit in the byte |
1434 | * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to | |
1435 | * clear the PG_locked bit and test PG_waiters at the same time fairly | |
1436 | * portably (architectures that do LL/SC can test any bit, while x86 can | |
1437 | * test the sign bit). | |
1da177e4 | 1438 | */ |
920c7a5d | 1439 | void unlock_page(struct page *page) |
1da177e4 | 1440 | { |
b91e1302 | 1441 | BUILD_BUG_ON(PG_waiters != 7); |
48c935ad | 1442 | page = compound_head(page); |
309381fe | 1443 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
b91e1302 LT |
1444 | if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags)) |
1445 | wake_up_page_bit(page, PG_locked); | |
1da177e4 LT |
1446 | } |
1447 | EXPORT_SYMBOL(unlock_page); | |
1448 | ||
485bb99b RD |
1449 | /** |
1450 | * end_page_writeback - end writeback against a page | |
1451 | * @page: the page | |
1da177e4 LT |
1452 | */ |
1453 | void end_page_writeback(struct page *page) | |
1454 | { | |
888cf2db MG |
1455 | /* |
1456 | * TestClearPageReclaim could be used here but it is an atomic | |
1457 | * operation and overkill in this particular case. Failing to | |
1458 | * shuffle a page marked for immediate reclaim is too mild to | |
1459 | * justify taking an atomic operation penalty at the end of | |
1460 | * ever page writeback. | |
1461 | */ | |
1462 | if (PageReclaim(page)) { | |
1463 | ClearPageReclaim(page); | |
ac6aadb2 | 1464 | rotate_reclaimable_page(page); |
888cf2db | 1465 | } |
ac6aadb2 MS |
1466 | |
1467 | if (!test_clear_page_writeback(page)) | |
1468 | BUG(); | |
1469 | ||
4e857c58 | 1470 | smp_mb__after_atomic(); |
1da177e4 LT |
1471 | wake_up_page(page, PG_writeback); |
1472 | } | |
1473 | EXPORT_SYMBOL(end_page_writeback); | |
1474 | ||
57d99845 MW |
1475 | /* |
1476 | * After completing I/O on a page, call this routine to update the page | |
1477 | * flags appropriately | |
1478 | */ | |
c11f0c0b | 1479 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1480 | { |
c11f0c0b | 1481 | if (!is_write) { |
57d99845 MW |
1482 | if (!err) { |
1483 | SetPageUptodate(page); | |
1484 | } else { | |
1485 | ClearPageUptodate(page); | |
1486 | SetPageError(page); | |
1487 | } | |
1488 | unlock_page(page); | |
abf54548 | 1489 | } else { |
57d99845 | 1490 | if (err) { |
dd8416c4 MK |
1491 | struct address_space *mapping; |
1492 | ||
57d99845 | 1493 | SetPageError(page); |
dd8416c4 MK |
1494 | mapping = page_mapping(page); |
1495 | if (mapping) | |
1496 | mapping_set_error(mapping, err); | |
57d99845 MW |
1497 | } |
1498 | end_page_writeback(page); | |
1499 | } | |
1500 | } | |
1501 | EXPORT_SYMBOL_GPL(page_endio); | |
1502 | ||
485bb99b RD |
1503 | /** |
1504 | * __lock_page - get a lock on the page, assuming we need to sleep to get it | |
87066755 | 1505 | * @__page: the page to lock |
1da177e4 | 1506 | */ |
62906027 | 1507 | void __lock_page(struct page *__page) |
1da177e4 | 1508 | { |
62906027 NP |
1509 | struct page *page = compound_head(__page); |
1510 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 HD |
1511 | wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, |
1512 | EXCLUSIVE); | |
1da177e4 LT |
1513 | } |
1514 | EXPORT_SYMBOL(__lock_page); | |
1515 | ||
62906027 | 1516 | int __lock_page_killable(struct page *__page) |
2687a356 | 1517 | { |
62906027 NP |
1518 | struct page *page = compound_head(__page); |
1519 | wait_queue_head_t *q = page_waitqueue(page); | |
9a1ea439 HD |
1520 | return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, |
1521 | EXCLUSIVE); | |
2687a356 | 1522 | } |
18bc0bbd | 1523 | EXPORT_SYMBOL_GPL(__lock_page_killable); |
2687a356 | 1524 | |
dd3e6d50 JA |
1525 | int __lock_page_async(struct page *page, struct wait_page_queue *wait) |
1526 | { | |
1527 | return __wait_on_page_locked_async(page, wait, true); | |
1528 | } | |
1529 | ||
9a95f3cf PC |
1530 | /* |
1531 | * Return values: | |
c1e8d7c6 | 1532 | * 1 - page is locked; mmap_lock is still held. |
9a95f3cf | 1533 | * 0 - page is not locked. |
3e4e28c5 | 1534 | * mmap_lock has been released (mmap_read_unlock(), unless flags had both |
9a95f3cf | 1535 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
c1e8d7c6 | 1536 | * which case mmap_lock is still held. |
9a95f3cf PC |
1537 | * |
1538 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1 | |
c1e8d7c6 | 1539 | * with the page locked and the mmap_lock unperturbed. |
9a95f3cf | 1540 | */ |
d065bd81 ML |
1541 | int __lock_page_or_retry(struct page *page, struct mm_struct *mm, |
1542 | unsigned int flags) | |
1543 | { | |
4064b982 | 1544 | if (fault_flag_allow_retry_first(flags)) { |
37b23e05 | 1545 | /* |
c1e8d7c6 | 1546 | * CAUTION! In this case, mmap_lock is not released |
37b23e05 KM |
1547 | * even though return 0. |
1548 | */ | |
1549 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
1550 | return 0; | |
1551 | ||
d8ed45c5 | 1552 | mmap_read_unlock(mm); |
37b23e05 KM |
1553 | if (flags & FAULT_FLAG_KILLABLE) |
1554 | wait_on_page_locked_killable(page); | |
1555 | else | |
318b275f | 1556 | wait_on_page_locked(page); |
d065bd81 | 1557 | return 0; |
37b23e05 KM |
1558 | } else { |
1559 | if (flags & FAULT_FLAG_KILLABLE) { | |
1560 | int ret; | |
1561 | ||
1562 | ret = __lock_page_killable(page); | |
1563 | if (ret) { | |
d8ed45c5 | 1564 | mmap_read_unlock(mm); |
37b23e05 KM |
1565 | return 0; |
1566 | } | |
1567 | } else | |
1568 | __lock_page(page); | |
1569 | return 1; | |
d065bd81 ML |
1570 | } |
1571 | } | |
1572 | ||
e7b563bb | 1573 | /** |
0d3f9296 MW |
1574 | * page_cache_next_miss() - Find the next gap in the page cache. |
1575 | * @mapping: Mapping. | |
1576 | * @index: Index. | |
1577 | * @max_scan: Maximum range to search. | |
e7b563bb | 1578 | * |
0d3f9296 MW |
1579 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1580 | * gap with the lowest index. | |
e7b563bb | 1581 | * |
0d3f9296 MW |
1582 | * This function may be called under the rcu_read_lock. However, this will |
1583 | * not atomically search a snapshot of the cache at a single point in time. | |
1584 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1585 | * created at index 10, page_cache_next_miss covering both indices may | |
1586 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1587 | * |
0d3f9296 MW |
1588 | * Return: The index of the gap if found, otherwise an index outside the |
1589 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1590 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1591 | */ |
0d3f9296 | 1592 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1593 | pgoff_t index, unsigned long max_scan) |
1594 | { | |
0d3f9296 | 1595 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1596 | |
0d3f9296 MW |
1597 | while (max_scan--) { |
1598 | void *entry = xas_next(&xas); | |
1599 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1600 | break; |
0d3f9296 | 1601 | if (xas.xa_index == 0) |
e7b563bb JW |
1602 | break; |
1603 | } | |
1604 | ||
0d3f9296 | 1605 | return xas.xa_index; |
e7b563bb | 1606 | } |
0d3f9296 | 1607 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1608 | |
1609 | /** | |
2346a560 | 1610 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
0d3f9296 MW |
1611 | * @mapping: Mapping. |
1612 | * @index: Index. | |
1613 | * @max_scan: Maximum range to search. | |
e7b563bb | 1614 | * |
0d3f9296 MW |
1615 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1616 | * gap with the highest index. | |
e7b563bb | 1617 | * |
0d3f9296 MW |
1618 | * This function may be called under the rcu_read_lock. However, this will |
1619 | * not atomically search a snapshot of the cache at a single point in time. | |
1620 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1621 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1622 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1623 | * |
0d3f9296 MW |
1624 | * Return: The index of the gap if found, otherwise an index outside the |
1625 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1626 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1627 | */ |
0d3f9296 | 1628 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1629 | pgoff_t index, unsigned long max_scan) |
1630 | { | |
0d3f9296 | 1631 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1632 | |
0d3f9296 MW |
1633 | while (max_scan--) { |
1634 | void *entry = xas_prev(&xas); | |
1635 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1636 | break; |
0d3f9296 | 1637 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1638 | break; |
1639 | } | |
1640 | ||
0d3f9296 | 1641 | return xas.xa_index; |
e7b563bb | 1642 | } |
0d3f9296 | 1643 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1644 | |
485bb99b | 1645 | /** |
0cd6144a | 1646 | * find_get_entry - find and get a page cache entry |
485bb99b | 1647 | * @mapping: the address_space to search |
0cd6144a JW |
1648 | * @offset: the page cache index |
1649 | * | |
1650 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1651 | * page cache page, it is returned with an increased refcount. | |
485bb99b | 1652 | * |
139b6a6f JW |
1653 | * If the slot holds a shadow entry of a previously evicted page, or a |
1654 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a | 1655 | * |
a862f68a | 1656 | * Return: the found page or shadow entry, %NULL if nothing is found. |
1da177e4 | 1657 | */ |
0cd6144a | 1658 | struct page *find_get_entry(struct address_space *mapping, pgoff_t offset) |
1da177e4 | 1659 | { |
4c7472c0 | 1660 | XA_STATE(xas, &mapping->i_pages, offset); |
4101196b | 1661 | struct page *page; |
1da177e4 | 1662 | |
a60637c8 NP |
1663 | rcu_read_lock(); |
1664 | repeat: | |
4c7472c0 MW |
1665 | xas_reset(&xas); |
1666 | page = xas_load(&xas); | |
1667 | if (xas_retry(&xas, page)) | |
1668 | goto repeat; | |
1669 | /* | |
1670 | * A shadow entry of a recently evicted page, or a swap entry from | |
1671 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1672 | */ | |
1673 | if (!page || xa_is_value(page)) | |
1674 | goto out; | |
83929372 | 1675 | |
4101196b | 1676 | if (!page_cache_get_speculative(page)) |
4c7472c0 | 1677 | goto repeat; |
83929372 | 1678 | |
4c7472c0 | 1679 | /* |
4101196b | 1680 | * Has the page moved or been split? |
4c7472c0 MW |
1681 | * This is part of the lockless pagecache protocol. See |
1682 | * include/linux/pagemap.h for details. | |
1683 | */ | |
1684 | if (unlikely(page != xas_reload(&xas))) { | |
4101196b | 1685 | put_page(page); |
4c7472c0 | 1686 | goto repeat; |
a60637c8 | 1687 | } |
4101196b | 1688 | page = find_subpage(page, offset); |
27d20fdd | 1689 | out: |
a60637c8 NP |
1690 | rcu_read_unlock(); |
1691 | ||
1da177e4 LT |
1692 | return page; |
1693 | } | |
1da177e4 | 1694 | |
0cd6144a JW |
1695 | /** |
1696 | * find_lock_entry - locate, pin and lock a page cache entry | |
1697 | * @mapping: the address_space to search | |
1698 | * @offset: the page cache index | |
1699 | * | |
1700 | * Looks up the page cache slot at @mapping & @offset. If there is a | |
1701 | * page cache page, it is returned locked and with an increased | |
1702 | * refcount. | |
1703 | * | |
139b6a6f JW |
1704 | * If the slot holds a shadow entry of a previously evicted page, or a |
1705 | * swap entry from shmem/tmpfs, it is returned. | |
0cd6144a | 1706 | * |
0cd6144a | 1707 | * find_lock_entry() may sleep. |
a862f68a MR |
1708 | * |
1709 | * Return: the found page or shadow entry, %NULL if nothing is found. | |
0cd6144a JW |
1710 | */ |
1711 | struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset) | |
1da177e4 LT |
1712 | { |
1713 | struct page *page; | |
1714 | ||
1da177e4 | 1715 | repeat: |
0cd6144a | 1716 | page = find_get_entry(mapping, offset); |
4c7472c0 | 1717 | if (page && !xa_is_value(page)) { |
a60637c8 NP |
1718 | lock_page(page); |
1719 | /* Has the page been truncated? */ | |
83929372 | 1720 | if (unlikely(page_mapping(page) != mapping)) { |
a60637c8 | 1721 | unlock_page(page); |
09cbfeaf | 1722 | put_page(page); |
a60637c8 | 1723 | goto repeat; |
1da177e4 | 1724 | } |
83929372 | 1725 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
1da177e4 | 1726 | } |
1da177e4 LT |
1727 | return page; |
1728 | } | |
0cd6144a JW |
1729 | EXPORT_SYMBOL(find_lock_entry); |
1730 | ||
1731 | /** | |
2294b32e MWO |
1732 | * pagecache_get_page - Find and get a reference to a page. |
1733 | * @mapping: The address_space to search. | |
1734 | * @index: The page index. | |
1735 | * @fgp_flags: %FGP flags modify how the page is returned. | |
1736 | * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified. | |
1da177e4 | 1737 | * |
2294b32e | 1738 | * Looks up the page cache entry at @mapping & @index. |
0cd6144a | 1739 | * |
2294b32e | 1740 | * @fgp_flags can be zero or more of these flags: |
0e056eb5 | 1741 | * |
2294b32e MWO |
1742 | * * %FGP_ACCESSED - The page will be marked accessed. |
1743 | * * %FGP_LOCK - The page is returned locked. | |
1744 | * * %FGP_CREAT - If no page is present then a new page is allocated using | |
1745 | * @gfp_mask and added to the page cache and the VM's LRU list. | |
1746 | * The page is returned locked and with an increased refcount. | |
1747 | * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the | |
1748 | * page is already in cache. If the page was allocated, unlock it before | |
1749 | * returning so the caller can do the same dance. | |
605cad83 YS |
1750 | * * %FGP_WRITE - The page will be written |
1751 | * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask | |
1752 | * * %FGP_NOWAIT - Don't get blocked by page lock | |
1da177e4 | 1753 | * |
2294b32e MWO |
1754 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even |
1755 | * if the %GFP flags specified for %FGP_CREAT are atomic. | |
1da177e4 | 1756 | * |
2457aec6 | 1757 | * If there is a page cache page, it is returned with an increased refcount. |
a862f68a | 1758 | * |
2294b32e | 1759 | * Return: The found page or %NULL otherwise. |
1da177e4 | 1760 | */ |
2294b32e MWO |
1761 | struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index, |
1762 | int fgp_flags, gfp_t gfp_mask) | |
1da177e4 | 1763 | { |
eb2be189 | 1764 | struct page *page; |
2457aec6 | 1765 | |
1da177e4 | 1766 | repeat: |
2294b32e | 1767 | page = find_get_entry(mapping, index); |
3159f943 | 1768 | if (xa_is_value(page)) |
2457aec6 MG |
1769 | page = NULL; |
1770 | if (!page) | |
1771 | goto no_page; | |
1772 | ||
1773 | if (fgp_flags & FGP_LOCK) { | |
1774 | if (fgp_flags & FGP_NOWAIT) { | |
1775 | if (!trylock_page(page)) { | |
09cbfeaf | 1776 | put_page(page); |
2457aec6 MG |
1777 | return NULL; |
1778 | } | |
1779 | } else { | |
1780 | lock_page(page); | |
1781 | } | |
1782 | ||
1783 | /* Has the page been truncated? */ | |
31895438 | 1784 | if (unlikely(compound_head(page)->mapping != mapping)) { |
2457aec6 | 1785 | unlock_page(page); |
09cbfeaf | 1786 | put_page(page); |
2457aec6 MG |
1787 | goto repeat; |
1788 | } | |
2294b32e | 1789 | VM_BUG_ON_PAGE(page->index != index, page); |
2457aec6 MG |
1790 | } |
1791 | ||
c16eb000 | 1792 | if (fgp_flags & FGP_ACCESSED) |
2457aec6 | 1793 | mark_page_accessed(page); |
b9306a79 YS |
1794 | else if (fgp_flags & FGP_WRITE) { |
1795 | /* Clear idle flag for buffer write */ | |
1796 | if (page_is_idle(page)) | |
1797 | clear_page_idle(page); | |
1798 | } | |
2457aec6 MG |
1799 | |
1800 | no_page: | |
1801 | if (!page && (fgp_flags & FGP_CREAT)) { | |
1802 | int err; | |
1803 | if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping)) | |
45f87de5 MH |
1804 | gfp_mask |= __GFP_WRITE; |
1805 | if (fgp_flags & FGP_NOFS) | |
1806 | gfp_mask &= ~__GFP_FS; | |
2457aec6 | 1807 | |
45f87de5 | 1808 | page = __page_cache_alloc(gfp_mask); |
eb2be189 NP |
1809 | if (!page) |
1810 | return NULL; | |
2457aec6 | 1811 | |
a75d4c33 | 1812 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
2457aec6 MG |
1813 | fgp_flags |= FGP_LOCK; |
1814 | ||
eb39d618 | 1815 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1816 | if (fgp_flags & FGP_ACCESSED) |
eb39d618 | 1817 | __SetPageReferenced(page); |
2457aec6 | 1818 | |
2294b32e | 1819 | err = add_to_page_cache_lru(page, mapping, index, gfp_mask); |
eb2be189 | 1820 | if (unlikely(err)) { |
09cbfeaf | 1821 | put_page(page); |
eb2be189 NP |
1822 | page = NULL; |
1823 | if (err == -EEXIST) | |
1824 | goto repeat; | |
1da177e4 | 1825 | } |
a75d4c33 JB |
1826 | |
1827 | /* | |
1828 | * add_to_page_cache_lru locks the page, and for mmap we expect | |
1829 | * an unlocked page. | |
1830 | */ | |
1831 | if (page && (fgp_flags & FGP_FOR_MMAP)) | |
1832 | unlock_page(page); | |
1da177e4 | 1833 | } |
2457aec6 | 1834 | |
1da177e4 LT |
1835 | return page; |
1836 | } | |
2457aec6 | 1837 | EXPORT_SYMBOL(pagecache_get_page); |
1da177e4 | 1838 | |
0cd6144a JW |
1839 | /** |
1840 | * find_get_entries - gang pagecache lookup | |
1841 | * @mapping: The address_space to search | |
1842 | * @start: The starting page cache index | |
1843 | * @nr_entries: The maximum number of entries | |
1844 | * @entries: Where the resulting entries are placed | |
1845 | * @indices: The cache indices corresponding to the entries in @entries | |
1846 | * | |
1847 | * find_get_entries() will search for and return a group of up to | |
1848 | * @nr_entries entries in the mapping. The entries are placed at | |
1849 | * @entries. find_get_entries() takes a reference against any actual | |
1850 | * pages it returns. | |
1851 | * | |
1852 | * The search returns a group of mapping-contiguous page cache entries | |
1853 | * with ascending indexes. There may be holes in the indices due to | |
1854 | * not-present pages. | |
1855 | * | |
139b6a6f JW |
1856 | * Any shadow entries of evicted pages, or swap entries from |
1857 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a | 1858 | * |
71725ed1 HD |
1859 | * If it finds a Transparent Huge Page, head or tail, find_get_entries() |
1860 | * stops at that page: the caller is likely to have a better way to handle | |
1861 | * the compound page as a whole, and then skip its extent, than repeatedly | |
1862 | * calling find_get_entries() to return all its tails. | |
1863 | * | |
a862f68a | 1864 | * Return: the number of pages and shadow entries which were found. |
0cd6144a JW |
1865 | */ |
1866 | unsigned find_get_entries(struct address_space *mapping, | |
1867 | pgoff_t start, unsigned int nr_entries, | |
1868 | struct page **entries, pgoff_t *indices) | |
1869 | { | |
f280bf09 MW |
1870 | XA_STATE(xas, &mapping->i_pages, start); |
1871 | struct page *page; | |
0cd6144a | 1872 | unsigned int ret = 0; |
0cd6144a JW |
1873 | |
1874 | if (!nr_entries) | |
1875 | return 0; | |
1876 | ||
1877 | rcu_read_lock(); | |
f280bf09 | 1878 | xas_for_each(&xas, page, ULONG_MAX) { |
f280bf09 | 1879 | if (xas_retry(&xas, page)) |
0cd6144a | 1880 | continue; |
f280bf09 MW |
1881 | /* |
1882 | * A shadow entry of a recently evicted page, a swap | |
1883 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1884 | * without attempting to raise page count. | |
1885 | */ | |
1886 | if (xa_is_value(page)) | |
0cd6144a | 1887 | goto export; |
83929372 | 1888 | |
4101196b | 1889 | if (!page_cache_get_speculative(page)) |
f280bf09 | 1890 | goto retry; |
83929372 | 1891 | |
4101196b | 1892 | /* Has the page moved or been split? */ |
f280bf09 MW |
1893 | if (unlikely(page != xas_reload(&xas))) |
1894 | goto put_page; | |
1895 | ||
71725ed1 HD |
1896 | /* |
1897 | * Terminate early on finding a THP, to allow the caller to | |
1898 | * handle it all at once; but continue if this is hugetlbfs. | |
1899 | */ | |
1900 | if (PageTransHuge(page) && !PageHuge(page)) { | |
1901 | page = find_subpage(page, xas.xa_index); | |
1902 | nr_entries = ret + 1; | |
1903 | } | |
0cd6144a | 1904 | export: |
f280bf09 | 1905 | indices[ret] = xas.xa_index; |
0cd6144a JW |
1906 | entries[ret] = page; |
1907 | if (++ret == nr_entries) | |
1908 | break; | |
f280bf09 MW |
1909 | continue; |
1910 | put_page: | |
4101196b | 1911 | put_page(page); |
f280bf09 MW |
1912 | retry: |
1913 | xas_reset(&xas); | |
0cd6144a JW |
1914 | } |
1915 | rcu_read_unlock(); | |
1916 | return ret; | |
1917 | } | |
1918 | ||
1da177e4 | 1919 | /** |
b947cee4 | 1920 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
1921 | * @mapping: The address_space to search |
1922 | * @start: The starting page index | |
b947cee4 | 1923 | * @end: The final page index (inclusive) |
1da177e4 LT |
1924 | * @nr_pages: The maximum number of pages |
1925 | * @pages: Where the resulting pages are placed | |
1926 | * | |
b947cee4 JK |
1927 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
1928 | * pages in the mapping starting at index @start and up to index @end | |
1929 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
1930 | * a reference against the returned pages. | |
1da177e4 LT |
1931 | * |
1932 | * The search returns a group of mapping-contiguous pages with ascending | |
1933 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 1934 | * We also update @start to index the next page for the traversal. |
1da177e4 | 1935 | * |
a862f68a MR |
1936 | * Return: the number of pages which were found. If this number is |
1937 | * smaller than @nr_pages, the end of specified range has been | |
b947cee4 | 1938 | * reached. |
1da177e4 | 1939 | */ |
b947cee4 JK |
1940 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
1941 | pgoff_t end, unsigned int nr_pages, | |
1942 | struct page **pages) | |
1da177e4 | 1943 | { |
fd1b3cee MW |
1944 | XA_STATE(xas, &mapping->i_pages, *start); |
1945 | struct page *page; | |
0fc9d104 KK |
1946 | unsigned ret = 0; |
1947 | ||
1948 | if (unlikely(!nr_pages)) | |
1949 | return 0; | |
a60637c8 NP |
1950 | |
1951 | rcu_read_lock(); | |
fd1b3cee | 1952 | xas_for_each(&xas, page, end) { |
fd1b3cee | 1953 | if (xas_retry(&xas, page)) |
a60637c8 | 1954 | continue; |
fd1b3cee MW |
1955 | /* Skip over shadow, swap and DAX entries */ |
1956 | if (xa_is_value(page)) | |
8079b1c8 | 1957 | continue; |
a60637c8 | 1958 | |
4101196b | 1959 | if (!page_cache_get_speculative(page)) |
fd1b3cee | 1960 | goto retry; |
83929372 | 1961 | |
4101196b | 1962 | /* Has the page moved or been split? */ |
fd1b3cee MW |
1963 | if (unlikely(page != xas_reload(&xas))) |
1964 | goto put_page; | |
1da177e4 | 1965 | |
4101196b | 1966 | pages[ret] = find_subpage(page, xas.xa_index); |
b947cee4 | 1967 | if (++ret == nr_pages) { |
5d3ee42f | 1968 | *start = xas.xa_index + 1; |
b947cee4 JK |
1969 | goto out; |
1970 | } | |
fd1b3cee MW |
1971 | continue; |
1972 | put_page: | |
4101196b | 1973 | put_page(page); |
fd1b3cee MW |
1974 | retry: |
1975 | xas_reset(&xas); | |
a60637c8 | 1976 | } |
5b280c0c | 1977 | |
b947cee4 JK |
1978 | /* |
1979 | * We come here when there is no page beyond @end. We take care to not | |
1980 | * overflow the index @start as it confuses some of the callers. This | |
fd1b3cee | 1981 | * breaks the iteration when there is a page at index -1 but that is |
b947cee4 JK |
1982 | * already broken anyway. |
1983 | */ | |
1984 | if (end == (pgoff_t)-1) | |
1985 | *start = (pgoff_t)-1; | |
1986 | else | |
1987 | *start = end + 1; | |
1988 | out: | |
a60637c8 | 1989 | rcu_read_unlock(); |
d72dc8a2 | 1990 | |
1da177e4 LT |
1991 | return ret; |
1992 | } | |
1993 | ||
ebf43500 JA |
1994 | /** |
1995 | * find_get_pages_contig - gang contiguous pagecache lookup | |
1996 | * @mapping: The address_space to search | |
1997 | * @index: The starting page index | |
1998 | * @nr_pages: The maximum number of pages | |
1999 | * @pages: Where the resulting pages are placed | |
2000 | * | |
2001 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
2002 | * that the returned number of pages are guaranteed to be contiguous. | |
2003 | * | |
a862f68a | 2004 | * Return: the number of pages which were found. |
ebf43500 JA |
2005 | */ |
2006 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
2007 | unsigned int nr_pages, struct page **pages) | |
2008 | { | |
3ece58a2 MW |
2009 | XA_STATE(xas, &mapping->i_pages, index); |
2010 | struct page *page; | |
0fc9d104 KK |
2011 | unsigned int ret = 0; |
2012 | ||
2013 | if (unlikely(!nr_pages)) | |
2014 | return 0; | |
a60637c8 NP |
2015 | |
2016 | rcu_read_lock(); | |
3ece58a2 | 2017 | for (page = xas_load(&xas); page; page = xas_next(&xas)) { |
3ece58a2 MW |
2018 | if (xas_retry(&xas, page)) |
2019 | continue; | |
2020 | /* | |
2021 | * If the entry has been swapped out, we can stop looking. | |
2022 | * No current caller is looking for DAX entries. | |
2023 | */ | |
2024 | if (xa_is_value(page)) | |
8079b1c8 | 2025 | break; |
ebf43500 | 2026 | |
4101196b | 2027 | if (!page_cache_get_speculative(page)) |
3ece58a2 | 2028 | goto retry; |
83929372 | 2029 | |
4101196b | 2030 | /* Has the page moved or been split? */ |
3ece58a2 MW |
2031 | if (unlikely(page != xas_reload(&xas))) |
2032 | goto put_page; | |
a60637c8 | 2033 | |
4101196b | 2034 | pages[ret] = find_subpage(page, xas.xa_index); |
0fc9d104 KK |
2035 | if (++ret == nr_pages) |
2036 | break; | |
3ece58a2 MW |
2037 | continue; |
2038 | put_page: | |
4101196b | 2039 | put_page(page); |
3ece58a2 MW |
2040 | retry: |
2041 | xas_reset(&xas); | |
ebf43500 | 2042 | } |
a60637c8 NP |
2043 | rcu_read_unlock(); |
2044 | return ret; | |
ebf43500 | 2045 | } |
ef71c15c | 2046 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 2047 | |
485bb99b | 2048 | /** |
72b045ae | 2049 | * find_get_pages_range_tag - find and return pages in given range matching @tag |
485bb99b RD |
2050 | * @mapping: the address_space to search |
2051 | * @index: the starting page index | |
72b045ae | 2052 | * @end: The final page index (inclusive) |
485bb99b RD |
2053 | * @tag: the tag index |
2054 | * @nr_pages: the maximum number of pages | |
2055 | * @pages: where the resulting pages are placed | |
2056 | * | |
1da177e4 | 2057 | * Like find_get_pages, except we only return pages which are tagged with |
485bb99b | 2058 | * @tag. We update @index to index the next page for the traversal. |
a862f68a MR |
2059 | * |
2060 | * Return: the number of pages which were found. | |
1da177e4 | 2061 | */ |
72b045ae | 2062 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
a6906972 | 2063 | pgoff_t end, xa_mark_t tag, unsigned int nr_pages, |
72b045ae | 2064 | struct page **pages) |
1da177e4 | 2065 | { |
a6906972 MW |
2066 | XA_STATE(xas, &mapping->i_pages, *index); |
2067 | struct page *page; | |
0fc9d104 KK |
2068 | unsigned ret = 0; |
2069 | ||
2070 | if (unlikely(!nr_pages)) | |
2071 | return 0; | |
a60637c8 NP |
2072 | |
2073 | rcu_read_lock(); | |
a6906972 | 2074 | xas_for_each_marked(&xas, page, end, tag) { |
a6906972 | 2075 | if (xas_retry(&xas, page)) |
a60637c8 | 2076 | continue; |
a6906972 MW |
2077 | /* |
2078 | * Shadow entries should never be tagged, but this iteration | |
2079 | * is lockless so there is a window for page reclaim to evict | |
2080 | * a page we saw tagged. Skip over it. | |
2081 | */ | |
2082 | if (xa_is_value(page)) | |
139b6a6f | 2083 | continue; |
a60637c8 | 2084 | |
4101196b | 2085 | if (!page_cache_get_speculative(page)) |
a6906972 | 2086 | goto retry; |
a60637c8 | 2087 | |
4101196b | 2088 | /* Has the page moved or been split? */ |
a6906972 MW |
2089 | if (unlikely(page != xas_reload(&xas))) |
2090 | goto put_page; | |
a60637c8 | 2091 | |
4101196b | 2092 | pages[ret] = find_subpage(page, xas.xa_index); |
72b045ae | 2093 | if (++ret == nr_pages) { |
5d3ee42f | 2094 | *index = xas.xa_index + 1; |
72b045ae JK |
2095 | goto out; |
2096 | } | |
a6906972 MW |
2097 | continue; |
2098 | put_page: | |
4101196b | 2099 | put_page(page); |
a6906972 MW |
2100 | retry: |
2101 | xas_reset(&xas); | |
a60637c8 | 2102 | } |
5b280c0c | 2103 | |
72b045ae | 2104 | /* |
a6906972 | 2105 | * We come here when we got to @end. We take care to not overflow the |
72b045ae | 2106 | * index @index as it confuses some of the callers. This breaks the |
a6906972 MW |
2107 | * iteration when there is a page at index -1 but that is already |
2108 | * broken anyway. | |
72b045ae JK |
2109 | */ |
2110 | if (end == (pgoff_t)-1) | |
2111 | *index = (pgoff_t)-1; | |
2112 | else | |
2113 | *index = end + 1; | |
2114 | out: | |
a60637c8 | 2115 | rcu_read_unlock(); |
1da177e4 | 2116 | |
1da177e4 LT |
2117 | return ret; |
2118 | } | |
72b045ae | 2119 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 2120 | |
76d42bd9 WF |
2121 | /* |
2122 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
2123 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
2124 | * | |
2125 | * ---R__________________________________________B__________ | |
2126 | * ^ reading here ^ bad block(assume 4k) | |
2127 | * | |
2128 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
2129 | * => failing the whole request => read(R) => read(R+1) => | |
2130 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
2131 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
2132 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
2133 | * | |
2134 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
2135 | */ | |
0f8e2db4 | 2136 | static void shrink_readahead_size_eio(struct file_ra_state *ra) |
76d42bd9 | 2137 | { |
76d42bd9 | 2138 | ra->ra_pages /= 4; |
76d42bd9 WF |
2139 | } |
2140 | ||
485bb99b | 2141 | /** |
47c27bc4 CH |
2142 | * generic_file_buffered_read - generic file read routine |
2143 | * @iocb: the iocb to read | |
6e58e79d AV |
2144 | * @iter: data destination |
2145 | * @written: already copied | |
485bb99b | 2146 | * |
1da177e4 | 2147 | * This is a generic file read routine, and uses the |
485bb99b | 2148 | * mapping->a_ops->readpage() function for the actual low-level stuff. |
1da177e4 LT |
2149 | * |
2150 | * This is really ugly. But the goto's actually try to clarify some | |
2151 | * of the logic when it comes to error handling etc. | |
a862f68a MR |
2152 | * |
2153 | * Return: | |
2154 | * * total number of bytes copied, including those the were already @written | |
2155 | * * negative error code if nothing was copied | |
1da177e4 | 2156 | */ |
d85dc2e1 | 2157 | ssize_t generic_file_buffered_read(struct kiocb *iocb, |
6e58e79d | 2158 | struct iov_iter *iter, ssize_t written) |
1da177e4 | 2159 | { |
47c27bc4 | 2160 | struct file *filp = iocb->ki_filp; |
36e78914 | 2161 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2162 | struct inode *inode = mapping->host; |
36e78914 | 2163 | struct file_ra_state *ra = &filp->f_ra; |
47c27bc4 | 2164 | loff_t *ppos = &iocb->ki_pos; |
57f6b96c FW |
2165 | pgoff_t index; |
2166 | pgoff_t last_index; | |
2167 | pgoff_t prev_index; | |
2168 | unsigned long offset; /* offset into pagecache page */ | |
ec0f1637 | 2169 | unsigned int prev_offset; |
6e58e79d | 2170 | int error = 0; |
1da177e4 | 2171 | |
c2a9737f | 2172 | if (unlikely(*ppos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2173 | return 0; |
c2a9737f WF |
2174 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
2175 | ||
09cbfeaf KS |
2176 | index = *ppos >> PAGE_SHIFT; |
2177 | prev_index = ra->prev_pos >> PAGE_SHIFT; | |
2178 | prev_offset = ra->prev_pos & (PAGE_SIZE-1); | |
2179 | last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT; | |
2180 | offset = *ppos & ~PAGE_MASK; | |
1da177e4 | 2181 | |
1da177e4 LT |
2182 | for (;;) { |
2183 | struct page *page; | |
57f6b96c | 2184 | pgoff_t end_index; |
a32ea1e1 | 2185 | loff_t isize; |
1da177e4 LT |
2186 | unsigned long nr, ret; |
2187 | ||
1da177e4 | 2188 | cond_resched(); |
1da177e4 | 2189 | find_page: |
5abf186a MH |
2190 | if (fatal_signal_pending(current)) { |
2191 | error = -EINTR; | |
2192 | goto out; | |
2193 | } | |
2194 | ||
1da177e4 | 2195 | page = find_get_page(mapping, index); |
3ea89ee8 | 2196 | if (!page) { |
cdc8fcb4 | 2197 | if (iocb->ki_flags & IOCB_NOIO) |
3239d834 | 2198 | goto would_block; |
cf914a7d | 2199 | page_cache_sync_readahead(mapping, |
7ff81078 | 2200 | ra, filp, |
3ea89ee8 FW |
2201 | index, last_index - index); |
2202 | page = find_get_page(mapping, index); | |
2203 | if (unlikely(page == NULL)) | |
2204 | goto no_cached_page; | |
2205 | } | |
2206 | if (PageReadahead(page)) { | |
41da51bc AG |
2207 | if (iocb->ki_flags & IOCB_NOIO) { |
2208 | put_page(page); | |
2209 | goto out; | |
2210 | } | |
cf914a7d | 2211 | page_cache_async_readahead(mapping, |
7ff81078 | 2212 | ra, filp, page, |
3ea89ee8 | 2213 | index, last_index - index); |
1da177e4 | 2214 | } |
8ab22b9a | 2215 | if (!PageUptodate(page)) { |
ebded027 MG |
2216 | /* |
2217 | * See comment in do_read_cache_page on why | |
2218 | * wait_on_page_locked is used to avoid unnecessarily | |
2219 | * serialisations and why it's safe. | |
2220 | */ | |
1a0a7853 JA |
2221 | if (iocb->ki_flags & IOCB_WAITQ) { |
2222 | if (written) { | |
2223 | put_page(page); | |
2224 | goto out; | |
2225 | } | |
2226 | error = wait_on_page_locked_async(page, | |
2227 | iocb->ki_waitq); | |
2228 | } else { | |
2229 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
2230 | put_page(page); | |
2231 | goto would_block; | |
2232 | } | |
2233 | error = wait_on_page_locked_killable(page); | |
2234 | } | |
c4b209a4 BVA |
2235 | if (unlikely(error)) |
2236 | goto readpage_error; | |
ebded027 MG |
2237 | if (PageUptodate(page)) |
2238 | goto page_ok; | |
2239 | ||
09cbfeaf | 2240 | if (inode->i_blkbits == PAGE_SHIFT || |
8ab22b9a HH |
2241 | !mapping->a_ops->is_partially_uptodate) |
2242 | goto page_not_up_to_date; | |
6d6d36bc | 2243 | /* pipes can't handle partially uptodate pages */ |
00e23707 | 2244 | if (unlikely(iov_iter_is_pipe(iter))) |
6d6d36bc | 2245 | goto page_not_up_to_date; |
529ae9aa | 2246 | if (!trylock_page(page)) |
8ab22b9a | 2247 | goto page_not_up_to_date; |
8d056cb9 DH |
2248 | /* Did it get truncated before we got the lock? */ |
2249 | if (!page->mapping) | |
2250 | goto page_not_up_to_date_locked; | |
8ab22b9a | 2251 | if (!mapping->a_ops->is_partially_uptodate(page, |
6e58e79d | 2252 | offset, iter->count)) |
8ab22b9a HH |
2253 | goto page_not_up_to_date_locked; |
2254 | unlock_page(page); | |
2255 | } | |
1da177e4 | 2256 | page_ok: |
a32ea1e1 N |
2257 | /* |
2258 | * i_size must be checked after we know the page is Uptodate. | |
2259 | * | |
2260 | * Checking i_size after the check allows us to calculate | |
2261 | * the correct value for "nr", which means the zero-filled | |
2262 | * part of the page is not copied back to userspace (unless | |
2263 | * another truncate extends the file - this is desired though). | |
2264 | */ | |
2265 | ||
2266 | isize = i_size_read(inode); | |
09cbfeaf | 2267 | end_index = (isize - 1) >> PAGE_SHIFT; |
a32ea1e1 | 2268 | if (unlikely(!isize || index > end_index)) { |
09cbfeaf | 2269 | put_page(page); |
a32ea1e1 N |
2270 | goto out; |
2271 | } | |
2272 | ||
2273 | /* nr is the maximum number of bytes to copy from this page */ | |
09cbfeaf | 2274 | nr = PAGE_SIZE; |
a32ea1e1 | 2275 | if (index == end_index) { |
09cbfeaf | 2276 | nr = ((isize - 1) & ~PAGE_MASK) + 1; |
a32ea1e1 | 2277 | if (nr <= offset) { |
09cbfeaf | 2278 | put_page(page); |
a32ea1e1 N |
2279 | goto out; |
2280 | } | |
2281 | } | |
2282 | nr = nr - offset; | |
1da177e4 LT |
2283 | |
2284 | /* If users can be writing to this page using arbitrary | |
2285 | * virtual addresses, take care about potential aliasing | |
2286 | * before reading the page on the kernel side. | |
2287 | */ | |
2288 | if (mapping_writably_mapped(mapping)) | |
2289 | flush_dcache_page(page); | |
2290 | ||
2291 | /* | |
ec0f1637 JK |
2292 | * When a sequential read accesses a page several times, |
2293 | * only mark it as accessed the first time. | |
1da177e4 | 2294 | */ |
ec0f1637 | 2295 | if (prev_index != index || offset != prev_offset) |
1da177e4 LT |
2296 | mark_page_accessed(page); |
2297 | prev_index = index; | |
2298 | ||
2299 | /* | |
2300 | * Ok, we have the page, and it's up-to-date, so | |
2301 | * now we can copy it to user space... | |
1da177e4 | 2302 | */ |
6e58e79d AV |
2303 | |
2304 | ret = copy_page_to_iter(page, offset, nr, iter); | |
1da177e4 | 2305 | offset += ret; |
09cbfeaf KS |
2306 | index += offset >> PAGE_SHIFT; |
2307 | offset &= ~PAGE_MASK; | |
6ce745ed | 2308 | prev_offset = offset; |
1da177e4 | 2309 | |
09cbfeaf | 2310 | put_page(page); |
6e58e79d AV |
2311 | written += ret; |
2312 | if (!iov_iter_count(iter)) | |
2313 | goto out; | |
2314 | if (ret < nr) { | |
2315 | error = -EFAULT; | |
2316 | goto out; | |
2317 | } | |
2318 | continue; | |
1da177e4 LT |
2319 | |
2320 | page_not_up_to_date: | |
2321 | /* Get exclusive access to the page ... */ | |
1a0a7853 JA |
2322 | if (iocb->ki_flags & IOCB_WAITQ) |
2323 | error = lock_page_async(page, iocb->ki_waitq); | |
2324 | else | |
2325 | error = lock_page_killable(page); | |
85462323 ON |
2326 | if (unlikely(error)) |
2327 | goto readpage_error; | |
1da177e4 | 2328 | |
8ab22b9a | 2329 | page_not_up_to_date_locked: |
da6052f7 | 2330 | /* Did it get truncated before we got the lock? */ |
1da177e4 LT |
2331 | if (!page->mapping) { |
2332 | unlock_page(page); | |
09cbfeaf | 2333 | put_page(page); |
1da177e4 LT |
2334 | continue; |
2335 | } | |
2336 | ||
2337 | /* Did somebody else fill it already? */ | |
2338 | if (PageUptodate(page)) { | |
2339 | unlock_page(page); | |
2340 | goto page_ok; | |
2341 | } | |
2342 | ||
2343 | readpage: | |
cdc8fcb4 | 2344 | if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT)) { |
41da51bc AG |
2345 | unlock_page(page); |
2346 | put_page(page); | |
2347 | goto would_block; | |
2348 | } | |
91803b49 JM |
2349 | /* |
2350 | * A previous I/O error may have been due to temporary | |
2351 | * failures, eg. multipath errors. | |
2352 | * PG_error will be set again if readpage fails. | |
2353 | */ | |
2354 | ClearPageError(page); | |
1da177e4 LT |
2355 | /* Start the actual read. The read will unlock the page. */ |
2356 | error = mapping->a_ops->readpage(filp, page); | |
2357 | ||
994fc28c ZB |
2358 | if (unlikely(error)) { |
2359 | if (error == AOP_TRUNCATED_PAGE) { | |
09cbfeaf | 2360 | put_page(page); |
6e58e79d | 2361 | error = 0; |
994fc28c ZB |
2362 | goto find_page; |
2363 | } | |
1da177e4 | 2364 | goto readpage_error; |
994fc28c | 2365 | } |
1da177e4 LT |
2366 | |
2367 | if (!PageUptodate(page)) { | |
85462323 ON |
2368 | error = lock_page_killable(page); |
2369 | if (unlikely(error)) | |
2370 | goto readpage_error; | |
1da177e4 LT |
2371 | if (!PageUptodate(page)) { |
2372 | if (page->mapping == NULL) { | |
2373 | /* | |
2ecdc82e | 2374 | * invalidate_mapping_pages got it |
1da177e4 LT |
2375 | */ |
2376 | unlock_page(page); | |
09cbfeaf | 2377 | put_page(page); |
1da177e4 LT |
2378 | goto find_page; |
2379 | } | |
2380 | unlock_page(page); | |
0f8e2db4 | 2381 | shrink_readahead_size_eio(ra); |
85462323 ON |
2382 | error = -EIO; |
2383 | goto readpage_error; | |
1da177e4 LT |
2384 | } |
2385 | unlock_page(page); | |
2386 | } | |
2387 | ||
1da177e4 LT |
2388 | goto page_ok; |
2389 | ||
2390 | readpage_error: | |
2391 | /* UHHUH! A synchronous read error occurred. Report it */ | |
09cbfeaf | 2392 | put_page(page); |
1da177e4 LT |
2393 | goto out; |
2394 | ||
2395 | no_cached_page: | |
2396 | /* | |
2397 | * Ok, it wasn't cached, so we need to create a new | |
2398 | * page.. | |
2399 | */ | |
453f85d4 | 2400 | page = page_cache_alloc(mapping); |
eb2be189 | 2401 | if (!page) { |
6e58e79d | 2402 | error = -ENOMEM; |
eb2be189 | 2403 | goto out; |
1da177e4 | 2404 | } |
6afdb859 | 2405 | error = add_to_page_cache_lru(page, mapping, index, |
c62d2555 | 2406 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
1da177e4 | 2407 | if (error) { |
09cbfeaf | 2408 | put_page(page); |
6e58e79d AV |
2409 | if (error == -EEXIST) { |
2410 | error = 0; | |
1da177e4 | 2411 | goto find_page; |
6e58e79d | 2412 | } |
1da177e4 LT |
2413 | goto out; |
2414 | } | |
1da177e4 LT |
2415 | goto readpage; |
2416 | } | |
2417 | ||
3239d834 MT |
2418 | would_block: |
2419 | error = -EAGAIN; | |
1da177e4 | 2420 | out: |
7ff81078 | 2421 | ra->prev_pos = prev_index; |
09cbfeaf | 2422 | ra->prev_pos <<= PAGE_SHIFT; |
7ff81078 | 2423 | ra->prev_pos |= prev_offset; |
1da177e4 | 2424 | |
09cbfeaf | 2425 | *ppos = ((loff_t)index << PAGE_SHIFT) + offset; |
0c6aa263 | 2426 | file_accessed(filp); |
6e58e79d | 2427 | return written ? written : error; |
1da177e4 | 2428 | } |
d85dc2e1 | 2429 | EXPORT_SYMBOL_GPL(generic_file_buffered_read); |
1da177e4 | 2430 | |
485bb99b | 2431 | /** |
6abd2322 | 2432 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2433 | * @iocb: kernel I/O control block |
6abd2322 | 2434 | * @iter: destination for the data read |
485bb99b | 2435 | * |
6abd2322 | 2436 | * This is the "read_iter()" routine for all filesystems |
1da177e4 | 2437 | * that can use the page cache directly. |
41da51bc AG |
2438 | * |
2439 | * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall | |
2440 | * be returned when no data can be read without waiting for I/O requests | |
2441 | * to complete; it doesn't prevent readahead. | |
2442 | * | |
2443 | * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O | |
2444 | * requests shall be made for the read or for readahead. When no data | |
2445 | * can be read, -EAGAIN shall be returned. When readahead would be | |
2446 | * triggered, a partial, possibly empty read shall be returned. | |
2447 | * | |
a862f68a MR |
2448 | * Return: |
2449 | * * number of bytes copied, even for partial reads | |
41da51bc | 2450 | * * negative error code (or 0 if IOCB_NOIO) if nothing was read |
1da177e4 LT |
2451 | */ |
2452 | ssize_t | |
ed978a81 | 2453 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2454 | { |
e7080a43 | 2455 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2456 | ssize_t retval = 0; |
e7080a43 NS |
2457 | |
2458 | if (!count) | |
2459 | goto out; /* skip atime */ | |
1da177e4 | 2460 | |
2ba48ce5 | 2461 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2462 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2463 | struct address_space *mapping = file->f_mapping; |
2464 | struct inode *inode = mapping->host; | |
543ade1f | 2465 | loff_t size; |
1da177e4 | 2466 | |
1da177e4 | 2467 | size = i_size_read(inode); |
6be96d3a GR |
2468 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2469 | if (filemap_range_has_page(mapping, iocb->ki_pos, | |
2470 | iocb->ki_pos + count - 1)) | |
2471 | return -EAGAIN; | |
2472 | } else { | |
2473 | retval = filemap_write_and_wait_range(mapping, | |
2474 | iocb->ki_pos, | |
2475 | iocb->ki_pos + count - 1); | |
2476 | if (retval < 0) | |
2477 | goto out; | |
2478 | } | |
d8d3d94b | 2479 | |
0d5b0cf2 CH |
2480 | file_accessed(file); |
2481 | ||
5ecda137 | 2482 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2483 | if (retval >= 0) { |
c64fb5c7 | 2484 | iocb->ki_pos += retval; |
5ecda137 | 2485 | count -= retval; |
9fe55eea | 2486 | } |
5b47d59a | 2487 | iov_iter_revert(iter, count - iov_iter_count(iter)); |
66f998f6 | 2488 | |
9fe55eea SW |
2489 | /* |
2490 | * Btrfs can have a short DIO read if we encounter | |
2491 | * compressed extents, so if there was an error, or if | |
2492 | * we've already read everything we wanted to, or if | |
2493 | * there was a short read because we hit EOF, go ahead | |
2494 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2495 | * the rest of the read. Buffered reads will not work for |
2496 | * DAX files, so don't bother trying. | |
9fe55eea | 2497 | */ |
5ecda137 | 2498 | if (retval < 0 || !count || iocb->ki_pos >= size || |
0d5b0cf2 | 2499 | IS_DAX(inode)) |
9fe55eea | 2500 | goto out; |
1da177e4 LT |
2501 | } |
2502 | ||
47c27bc4 | 2503 | retval = generic_file_buffered_read(iocb, iter, retval); |
1da177e4 LT |
2504 | out: |
2505 | return retval; | |
2506 | } | |
ed978a81 | 2507 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2508 | |
1da177e4 | 2509 | #ifdef CONFIG_MMU |
1da177e4 | 2510 | #define MMAP_LOTSAMISS (100) |
6b4c9f44 | 2511 | /* |
c1e8d7c6 | 2512 | * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock |
6b4c9f44 JB |
2513 | * @vmf - the vm_fault for this fault. |
2514 | * @page - the page to lock. | |
2515 | * @fpin - the pointer to the file we may pin (or is already pinned). | |
2516 | * | |
c1e8d7c6 | 2517 | * This works similar to lock_page_or_retry in that it can drop the mmap_lock. |
6b4c9f44 | 2518 | * It differs in that it actually returns the page locked if it returns 1 and 0 |
c1e8d7c6 | 2519 | * if it couldn't lock the page. If we did have to drop the mmap_lock then fpin |
6b4c9f44 JB |
2520 | * will point to the pinned file and needs to be fput()'ed at a later point. |
2521 | */ | |
2522 | static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page, | |
2523 | struct file **fpin) | |
2524 | { | |
2525 | if (trylock_page(page)) | |
2526 | return 1; | |
2527 | ||
8b0f9fa2 LT |
2528 | /* |
2529 | * NOTE! This will make us return with VM_FAULT_RETRY, but with | |
c1e8d7c6 | 2530 | * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT |
8b0f9fa2 LT |
2531 | * is supposed to work. We have way too many special cases.. |
2532 | */ | |
6b4c9f44 JB |
2533 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
2534 | return 0; | |
2535 | ||
2536 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); | |
2537 | if (vmf->flags & FAULT_FLAG_KILLABLE) { | |
2538 | if (__lock_page_killable(page)) { | |
2539 | /* | |
c1e8d7c6 | 2540 | * We didn't have the right flags to drop the mmap_lock, |
6b4c9f44 JB |
2541 | * but all fault_handlers only check for fatal signals |
2542 | * if we return VM_FAULT_RETRY, so we need to drop the | |
c1e8d7c6 | 2543 | * mmap_lock here and return 0 if we don't have a fpin. |
6b4c9f44 JB |
2544 | */ |
2545 | if (*fpin == NULL) | |
d8ed45c5 | 2546 | mmap_read_unlock(vmf->vma->vm_mm); |
6b4c9f44 JB |
2547 | return 0; |
2548 | } | |
2549 | } else | |
2550 | __lock_page(page); | |
2551 | return 1; | |
2552 | } | |
2553 | ||
1da177e4 | 2554 | |
ef00e08e | 2555 | /* |
6b4c9f44 JB |
2556 | * Synchronous readahead happens when we don't even find a page in the page |
2557 | * cache at all. We don't want to perform IO under the mmap sem, so if we have | |
2558 | * to drop the mmap sem we return the file that was pinned in order for us to do | |
2559 | * that. If we didn't pin a file then we return NULL. The file that is | |
2560 | * returned needs to be fput()'ed when we're done with it. | |
ef00e08e | 2561 | */ |
6b4c9f44 | 2562 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
ef00e08e | 2563 | { |
2a1180f1 JB |
2564 | struct file *file = vmf->vma->vm_file; |
2565 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2566 | struct address_space *mapping = file->f_mapping; |
6b4c9f44 | 2567 | struct file *fpin = NULL; |
2a1180f1 | 2568 | pgoff_t offset = vmf->pgoff; |
e630bfac | 2569 | unsigned int mmap_miss; |
ef00e08e LT |
2570 | |
2571 | /* If we don't want any read-ahead, don't bother */ | |
2a1180f1 | 2572 | if (vmf->vma->vm_flags & VM_RAND_READ) |
6b4c9f44 | 2573 | return fpin; |
275b12bf | 2574 | if (!ra->ra_pages) |
6b4c9f44 | 2575 | return fpin; |
ef00e08e | 2576 | |
2a1180f1 | 2577 | if (vmf->vma->vm_flags & VM_SEQ_READ) { |
6b4c9f44 | 2578 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
7ffc59b4 WF |
2579 | page_cache_sync_readahead(mapping, ra, file, offset, |
2580 | ra->ra_pages); | |
6b4c9f44 | 2581 | return fpin; |
ef00e08e LT |
2582 | } |
2583 | ||
207d04ba | 2584 | /* Avoid banging the cache line if not needed */ |
e630bfac KS |
2585 | mmap_miss = READ_ONCE(ra->mmap_miss); |
2586 | if (mmap_miss < MMAP_LOTSAMISS * 10) | |
2587 | WRITE_ONCE(ra->mmap_miss, ++mmap_miss); | |
ef00e08e LT |
2588 | |
2589 | /* | |
2590 | * Do we miss much more than hit in this file? If so, | |
2591 | * stop bothering with read-ahead. It will only hurt. | |
2592 | */ | |
e630bfac | 2593 | if (mmap_miss > MMAP_LOTSAMISS) |
6b4c9f44 | 2594 | return fpin; |
ef00e08e | 2595 | |
d30a1100 WF |
2596 | /* |
2597 | * mmap read-around | |
2598 | */ | |
6b4c9f44 | 2599 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
600e19af RG |
2600 | ra->start = max_t(long, 0, offset - ra->ra_pages / 2); |
2601 | ra->size = ra->ra_pages; | |
2602 | ra->async_size = ra->ra_pages / 4; | |
275b12bf | 2603 | ra_submit(ra, mapping, file); |
6b4c9f44 | 2604 | return fpin; |
ef00e08e LT |
2605 | } |
2606 | ||
2607 | /* | |
2608 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
6b4c9f44 | 2609 | * so we want to possibly extend the readahead further. We return the file that |
c1e8d7c6 | 2610 | * was pinned if we have to drop the mmap_lock in order to do IO. |
ef00e08e | 2611 | */ |
6b4c9f44 JB |
2612 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
2613 | struct page *page) | |
ef00e08e | 2614 | { |
2a1180f1 JB |
2615 | struct file *file = vmf->vma->vm_file; |
2616 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2617 | struct address_space *mapping = file->f_mapping; |
6b4c9f44 | 2618 | struct file *fpin = NULL; |
e630bfac | 2619 | unsigned int mmap_miss; |
2a1180f1 | 2620 | pgoff_t offset = vmf->pgoff; |
ef00e08e LT |
2621 | |
2622 | /* If we don't want any read-ahead, don't bother */ | |
5c72feee | 2623 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
6b4c9f44 | 2624 | return fpin; |
e630bfac KS |
2625 | mmap_miss = READ_ONCE(ra->mmap_miss); |
2626 | if (mmap_miss) | |
2627 | WRITE_ONCE(ra->mmap_miss, --mmap_miss); | |
6b4c9f44 JB |
2628 | if (PageReadahead(page)) { |
2629 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); | |
2fad6f5d WF |
2630 | page_cache_async_readahead(mapping, ra, file, |
2631 | page, offset, ra->ra_pages); | |
6b4c9f44 JB |
2632 | } |
2633 | return fpin; | |
ef00e08e LT |
2634 | } |
2635 | ||
485bb99b | 2636 | /** |
54cb8821 | 2637 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 2638 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 2639 | * |
54cb8821 | 2640 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
2641 | * mapped memory region to read in file data during a page fault. |
2642 | * | |
2643 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
2644 | * it in the page cache, and handles the special cases reasonably without | |
2645 | * having a lot of duplicated code. | |
9a95f3cf | 2646 | * |
c1e8d7c6 | 2647 | * vma->vm_mm->mmap_lock must be held on entry. |
9a95f3cf | 2648 | * |
c1e8d7c6 | 2649 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock |
a4985833 | 2650 | * may be dropped before doing I/O or by lock_page_maybe_drop_mmap(). |
9a95f3cf | 2651 | * |
c1e8d7c6 | 2652 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock |
9a95f3cf PC |
2653 | * has not been released. |
2654 | * | |
2655 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
a862f68a MR |
2656 | * |
2657 | * Return: bitwise-OR of %VM_FAULT_ codes. | |
1da177e4 | 2658 | */ |
2bcd6454 | 2659 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
2660 | { |
2661 | int error; | |
11bac800 | 2662 | struct file *file = vmf->vma->vm_file; |
6b4c9f44 | 2663 | struct file *fpin = NULL; |
1da177e4 LT |
2664 | struct address_space *mapping = file->f_mapping; |
2665 | struct file_ra_state *ra = &file->f_ra; | |
2666 | struct inode *inode = mapping->host; | |
ef00e08e | 2667 | pgoff_t offset = vmf->pgoff; |
9ab2594f | 2668 | pgoff_t max_off; |
1da177e4 | 2669 | struct page *page; |
2bcd6454 | 2670 | vm_fault_t ret = 0; |
1da177e4 | 2671 | |
9ab2594f MW |
2672 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2673 | if (unlikely(offset >= max_off)) | |
5307cc1a | 2674 | return VM_FAULT_SIGBUS; |
1da177e4 | 2675 | |
1da177e4 | 2676 | /* |
49426420 | 2677 | * Do we have something in the page cache already? |
1da177e4 | 2678 | */ |
ef00e08e | 2679 | page = find_get_page(mapping, offset); |
45cac65b | 2680 | if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) { |
1da177e4 | 2681 | /* |
ef00e08e LT |
2682 | * We found the page, so try async readahead before |
2683 | * waiting for the lock. | |
1da177e4 | 2684 | */ |
6b4c9f44 | 2685 | fpin = do_async_mmap_readahead(vmf, page); |
45cac65b | 2686 | } else if (!page) { |
ef00e08e | 2687 | /* No page in the page cache at all */ |
ef00e08e | 2688 | count_vm_event(PGMAJFAULT); |
2262185c | 2689 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e | 2690 | ret = VM_FAULT_MAJOR; |
6b4c9f44 | 2691 | fpin = do_sync_mmap_readahead(vmf); |
ef00e08e | 2692 | retry_find: |
a75d4c33 JB |
2693 | page = pagecache_get_page(mapping, offset, |
2694 | FGP_CREAT|FGP_FOR_MMAP, | |
2695 | vmf->gfp_mask); | |
6b4c9f44 JB |
2696 | if (!page) { |
2697 | if (fpin) | |
2698 | goto out_retry; | |
e520e932 | 2699 | return VM_FAULT_OOM; |
6b4c9f44 | 2700 | } |
1da177e4 LT |
2701 | } |
2702 | ||
6b4c9f44 JB |
2703 | if (!lock_page_maybe_drop_mmap(vmf, page, &fpin)) |
2704 | goto out_retry; | |
b522c94d ML |
2705 | |
2706 | /* Did it get truncated? */ | |
585e5a7b | 2707 | if (unlikely(compound_head(page)->mapping != mapping)) { |
b522c94d ML |
2708 | unlock_page(page); |
2709 | put_page(page); | |
2710 | goto retry_find; | |
2711 | } | |
520e5ba4 | 2712 | VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page); |
b522c94d | 2713 | |
1da177e4 | 2714 | /* |
d00806b1 NP |
2715 | * We have a locked page in the page cache, now we need to check |
2716 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 2717 | */ |
d00806b1 | 2718 | if (unlikely(!PageUptodate(page))) |
1da177e4 LT |
2719 | goto page_not_uptodate; |
2720 | ||
6b4c9f44 | 2721 | /* |
c1e8d7c6 | 2722 | * We've made it this far and we had to drop our mmap_lock, now is the |
6b4c9f44 JB |
2723 | * time to return to the upper layer and have it re-find the vma and |
2724 | * redo the fault. | |
2725 | */ | |
2726 | if (fpin) { | |
2727 | unlock_page(page); | |
2728 | goto out_retry; | |
2729 | } | |
2730 | ||
ef00e08e LT |
2731 | /* |
2732 | * Found the page and have a reference on it. | |
2733 | * We must recheck i_size under page lock. | |
2734 | */ | |
9ab2594f MW |
2735 | max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
2736 | if (unlikely(offset >= max_off)) { | |
d00806b1 | 2737 | unlock_page(page); |
09cbfeaf | 2738 | put_page(page); |
5307cc1a | 2739 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
2740 | } |
2741 | ||
d0217ac0 | 2742 | vmf->page = page; |
83c54070 | 2743 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 2744 | |
1da177e4 | 2745 | page_not_uptodate: |
1da177e4 LT |
2746 | /* |
2747 | * Umm, take care of errors if the page isn't up-to-date. | |
2748 | * Try to re-read it _once_. We do this synchronously, | |
2749 | * because there really aren't any performance issues here | |
2750 | * and we need to check for errors. | |
2751 | */ | |
1da177e4 | 2752 | ClearPageError(page); |
6b4c9f44 | 2753 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
994fc28c | 2754 | error = mapping->a_ops->readpage(file, page); |
3ef0f720 MS |
2755 | if (!error) { |
2756 | wait_on_page_locked(page); | |
2757 | if (!PageUptodate(page)) | |
2758 | error = -EIO; | |
2759 | } | |
6b4c9f44 JB |
2760 | if (fpin) |
2761 | goto out_retry; | |
09cbfeaf | 2762 | put_page(page); |
d00806b1 NP |
2763 | |
2764 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 2765 | goto retry_find; |
1da177e4 | 2766 | |
0f8e2db4 | 2767 | shrink_readahead_size_eio(ra); |
d0217ac0 | 2768 | return VM_FAULT_SIGBUS; |
6b4c9f44 JB |
2769 | |
2770 | out_retry: | |
2771 | /* | |
c1e8d7c6 | 2772 | * We dropped the mmap_lock, we need to return to the fault handler to |
6b4c9f44 JB |
2773 | * re-find the vma and come back and find our hopefully still populated |
2774 | * page. | |
2775 | */ | |
2776 | if (page) | |
2777 | put_page(page); | |
2778 | if (fpin) | |
2779 | fput(fpin); | |
2780 | return ret | VM_FAULT_RETRY; | |
54cb8821 NP |
2781 | } |
2782 | EXPORT_SYMBOL(filemap_fault); | |
2783 | ||
82b0f8c3 | 2784 | void filemap_map_pages(struct vm_fault *vmf, |
bae473a4 | 2785 | pgoff_t start_pgoff, pgoff_t end_pgoff) |
f1820361 | 2786 | { |
82b0f8c3 | 2787 | struct file *file = vmf->vma->vm_file; |
f1820361 | 2788 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 2789 | pgoff_t last_pgoff = start_pgoff; |
9ab2594f | 2790 | unsigned long max_idx; |
070e807c | 2791 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
4101196b | 2792 | struct page *page; |
e630bfac | 2793 | unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss); |
f1820361 KS |
2794 | |
2795 | rcu_read_lock(); | |
070e807c MW |
2796 | xas_for_each(&xas, page, end_pgoff) { |
2797 | if (xas_retry(&xas, page)) | |
2798 | continue; | |
2799 | if (xa_is_value(page)) | |
2cf938aa | 2800 | goto next; |
f1820361 | 2801 | |
e0975b2a MH |
2802 | /* |
2803 | * Check for a locked page first, as a speculative | |
2804 | * reference may adversely influence page migration. | |
2805 | */ | |
4101196b | 2806 | if (PageLocked(page)) |
e0975b2a | 2807 | goto next; |
4101196b | 2808 | if (!page_cache_get_speculative(page)) |
070e807c | 2809 | goto next; |
f1820361 | 2810 | |
4101196b | 2811 | /* Has the page moved or been split? */ |
070e807c MW |
2812 | if (unlikely(page != xas_reload(&xas))) |
2813 | goto skip; | |
4101196b | 2814 | page = find_subpage(page, xas.xa_index); |
f1820361 KS |
2815 | |
2816 | if (!PageUptodate(page) || | |
2817 | PageReadahead(page) || | |
2818 | PageHWPoison(page)) | |
2819 | goto skip; | |
2820 | if (!trylock_page(page)) | |
2821 | goto skip; | |
2822 | ||
2823 | if (page->mapping != mapping || !PageUptodate(page)) | |
2824 | goto unlock; | |
2825 | ||
9ab2594f MW |
2826 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); |
2827 | if (page->index >= max_idx) | |
f1820361 KS |
2828 | goto unlock; |
2829 | ||
e630bfac KS |
2830 | if (mmap_miss > 0) |
2831 | mmap_miss--; | |
7267ec00 | 2832 | |
070e807c | 2833 | vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
82b0f8c3 | 2834 | if (vmf->pte) |
070e807c MW |
2835 | vmf->pte += xas.xa_index - last_pgoff; |
2836 | last_pgoff = xas.xa_index; | |
9d82c694 | 2837 | if (alloc_set_pte(vmf, page)) |
7267ec00 | 2838 | goto unlock; |
f1820361 KS |
2839 | unlock_page(page); |
2840 | goto next; | |
2841 | unlock: | |
2842 | unlock_page(page); | |
2843 | skip: | |
09cbfeaf | 2844 | put_page(page); |
f1820361 | 2845 | next: |
7267ec00 | 2846 | /* Huge page is mapped? No need to proceed. */ |
82b0f8c3 | 2847 | if (pmd_trans_huge(*vmf->pmd)) |
7267ec00 | 2848 | break; |
f1820361 KS |
2849 | } |
2850 | rcu_read_unlock(); | |
e630bfac | 2851 | WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss); |
f1820361 KS |
2852 | } |
2853 | EXPORT_SYMBOL(filemap_map_pages); | |
2854 | ||
2bcd6454 | 2855 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 JK |
2856 | { |
2857 | struct page *page = vmf->page; | |
11bac800 | 2858 | struct inode *inode = file_inode(vmf->vma->vm_file); |
2bcd6454 | 2859 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 2860 | |
14da9200 | 2861 | sb_start_pagefault(inode->i_sb); |
11bac800 | 2862 | file_update_time(vmf->vma->vm_file); |
4fcf1c62 JK |
2863 | lock_page(page); |
2864 | if (page->mapping != inode->i_mapping) { | |
2865 | unlock_page(page); | |
2866 | ret = VM_FAULT_NOPAGE; | |
2867 | goto out; | |
2868 | } | |
14da9200 JK |
2869 | /* |
2870 | * We mark the page dirty already here so that when freeze is in | |
2871 | * progress, we are guaranteed that writeback during freezing will | |
2872 | * see the dirty page and writeprotect it again. | |
2873 | */ | |
2874 | set_page_dirty(page); | |
1d1d1a76 | 2875 | wait_for_stable_page(page); |
4fcf1c62 | 2876 | out: |
14da9200 | 2877 | sb_end_pagefault(inode->i_sb); |
4fcf1c62 JK |
2878 | return ret; |
2879 | } | |
4fcf1c62 | 2880 | |
f0f37e2f | 2881 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 2882 | .fault = filemap_fault, |
f1820361 | 2883 | .map_pages = filemap_map_pages, |
4fcf1c62 | 2884 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
2885 | }; |
2886 | ||
2887 | /* This is used for a general mmap of a disk file */ | |
2888 | ||
2889 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) | |
2890 | { | |
2891 | struct address_space *mapping = file->f_mapping; | |
2892 | ||
2893 | if (!mapping->a_ops->readpage) | |
2894 | return -ENOEXEC; | |
2895 | file_accessed(file); | |
2896 | vma->vm_ops = &generic_file_vm_ops; | |
2897 | return 0; | |
2898 | } | |
1da177e4 LT |
2899 | |
2900 | /* | |
2901 | * This is for filesystems which do not implement ->writepage. | |
2902 | */ | |
2903 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
2904 | { | |
2905 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
2906 | return -EINVAL; | |
2907 | return generic_file_mmap(file, vma); | |
2908 | } | |
2909 | #else | |
4b96a37d | 2910 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
45397228 | 2911 | { |
4b96a37d | 2912 | return VM_FAULT_SIGBUS; |
45397228 | 2913 | } |
1da177e4 LT |
2914 | int generic_file_mmap(struct file * file, struct vm_area_struct * vma) |
2915 | { | |
2916 | return -ENOSYS; | |
2917 | } | |
2918 | int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) | |
2919 | { | |
2920 | return -ENOSYS; | |
2921 | } | |
2922 | #endif /* CONFIG_MMU */ | |
2923 | ||
45397228 | 2924 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
2925 | EXPORT_SYMBOL(generic_file_mmap); |
2926 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
2927 | ||
67f9fd91 SL |
2928 | static struct page *wait_on_page_read(struct page *page) |
2929 | { | |
2930 | if (!IS_ERR(page)) { | |
2931 | wait_on_page_locked(page); | |
2932 | if (!PageUptodate(page)) { | |
09cbfeaf | 2933 | put_page(page); |
67f9fd91 SL |
2934 | page = ERR_PTR(-EIO); |
2935 | } | |
2936 | } | |
2937 | return page; | |
2938 | } | |
2939 | ||
32b63529 | 2940 | static struct page *do_read_cache_page(struct address_space *mapping, |
57f6b96c | 2941 | pgoff_t index, |
5e5358e7 | 2942 | int (*filler)(void *, struct page *), |
0531b2aa LT |
2943 | void *data, |
2944 | gfp_t gfp) | |
1da177e4 | 2945 | { |
eb2be189 | 2946 | struct page *page; |
1da177e4 LT |
2947 | int err; |
2948 | repeat: | |
2949 | page = find_get_page(mapping, index); | |
2950 | if (!page) { | |
453f85d4 | 2951 | page = __page_cache_alloc(gfp); |
eb2be189 NP |
2952 | if (!page) |
2953 | return ERR_PTR(-ENOMEM); | |
e6f67b8c | 2954 | err = add_to_page_cache_lru(page, mapping, index, gfp); |
eb2be189 | 2955 | if (unlikely(err)) { |
09cbfeaf | 2956 | put_page(page); |
eb2be189 NP |
2957 | if (err == -EEXIST) |
2958 | goto repeat; | |
22ecdb4f | 2959 | /* Presumably ENOMEM for xarray node */ |
1da177e4 LT |
2960 | return ERR_PTR(err); |
2961 | } | |
32b63529 MG |
2962 | |
2963 | filler: | |
6c45b454 CH |
2964 | if (filler) |
2965 | err = filler(data, page); | |
2966 | else | |
2967 | err = mapping->a_ops->readpage(data, page); | |
2968 | ||
1da177e4 | 2969 | if (err < 0) { |
09cbfeaf | 2970 | put_page(page); |
32b63529 | 2971 | return ERR_PTR(err); |
1da177e4 | 2972 | } |
1da177e4 | 2973 | |
32b63529 MG |
2974 | page = wait_on_page_read(page); |
2975 | if (IS_ERR(page)) | |
2976 | return page; | |
2977 | goto out; | |
2978 | } | |
1da177e4 LT |
2979 | if (PageUptodate(page)) |
2980 | goto out; | |
2981 | ||
ebded027 MG |
2982 | /* |
2983 | * Page is not up to date and may be locked due one of the following | |
2984 | * case a: Page is being filled and the page lock is held | |
2985 | * case b: Read/write error clearing the page uptodate status | |
2986 | * case c: Truncation in progress (page locked) | |
2987 | * case d: Reclaim in progress | |
2988 | * | |
2989 | * Case a, the page will be up to date when the page is unlocked. | |
2990 | * There is no need to serialise on the page lock here as the page | |
2991 | * is pinned so the lock gives no additional protection. Even if the | |
ce89fddf | 2992 | * page is truncated, the data is still valid if PageUptodate as |
ebded027 MG |
2993 | * it's a race vs truncate race. |
2994 | * Case b, the page will not be up to date | |
2995 | * Case c, the page may be truncated but in itself, the data may still | |
2996 | * be valid after IO completes as it's a read vs truncate race. The | |
2997 | * operation must restart if the page is not uptodate on unlock but | |
2998 | * otherwise serialising on page lock to stabilise the mapping gives | |
2999 | * no additional guarantees to the caller as the page lock is | |
3000 | * released before return. | |
3001 | * Case d, similar to truncation. If reclaim holds the page lock, it | |
3002 | * will be a race with remove_mapping that determines if the mapping | |
3003 | * is valid on unlock but otherwise the data is valid and there is | |
3004 | * no need to serialise with page lock. | |
3005 | * | |
3006 | * As the page lock gives no additional guarantee, we optimistically | |
3007 | * wait on the page to be unlocked and check if it's up to date and | |
3008 | * use the page if it is. Otherwise, the page lock is required to | |
3009 | * distinguish between the different cases. The motivation is that we | |
3010 | * avoid spurious serialisations and wakeups when multiple processes | |
3011 | * wait on the same page for IO to complete. | |
3012 | */ | |
3013 | wait_on_page_locked(page); | |
3014 | if (PageUptodate(page)) | |
3015 | goto out; | |
3016 | ||
3017 | /* Distinguish between all the cases under the safety of the lock */ | |
1da177e4 | 3018 | lock_page(page); |
ebded027 MG |
3019 | |
3020 | /* Case c or d, restart the operation */ | |
1da177e4 LT |
3021 | if (!page->mapping) { |
3022 | unlock_page(page); | |
09cbfeaf | 3023 | put_page(page); |
32b63529 | 3024 | goto repeat; |
1da177e4 | 3025 | } |
ebded027 MG |
3026 | |
3027 | /* Someone else locked and filled the page in a very small window */ | |
1da177e4 LT |
3028 | if (PageUptodate(page)) { |
3029 | unlock_page(page); | |
3030 | goto out; | |
3031 | } | |
faffdfa0 XT |
3032 | |
3033 | /* | |
3034 | * A previous I/O error may have been due to temporary | |
3035 | * failures. | |
3036 | * Clear page error before actual read, PG_error will be | |
3037 | * set again if read page fails. | |
3038 | */ | |
3039 | ClearPageError(page); | |
32b63529 MG |
3040 | goto filler; |
3041 | ||
c855ff37 | 3042 | out: |
6fe6900e NP |
3043 | mark_page_accessed(page); |
3044 | return page; | |
3045 | } | |
0531b2aa LT |
3046 | |
3047 | /** | |
67f9fd91 | 3048 | * read_cache_page - read into page cache, fill it if needed |
0531b2aa LT |
3049 | * @mapping: the page's address_space |
3050 | * @index: the page index | |
3051 | * @filler: function to perform the read | |
5e5358e7 | 3052 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 3053 | * |
0531b2aa | 3054 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 3055 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
3056 | * |
3057 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a MR |
3058 | * |
3059 | * Return: up to date page on success, ERR_PTR() on failure. | |
0531b2aa | 3060 | */ |
67f9fd91 | 3061 | struct page *read_cache_page(struct address_space *mapping, |
0531b2aa | 3062 | pgoff_t index, |
5e5358e7 | 3063 | int (*filler)(void *, struct page *), |
0531b2aa LT |
3064 | void *data) |
3065 | { | |
d322a8e5 CH |
3066 | return do_read_cache_page(mapping, index, filler, data, |
3067 | mapping_gfp_mask(mapping)); | |
0531b2aa | 3068 | } |
67f9fd91 | 3069 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
3070 | |
3071 | /** | |
3072 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
3073 | * @mapping: the page's address_space | |
3074 | * @index: the page index | |
3075 | * @gfp: the page allocator flags to use if allocating | |
3076 | * | |
3077 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 3078 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
3079 | * |
3080 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a MR |
3081 | * |
3082 | * Return: up to date page on success, ERR_PTR() on failure. | |
0531b2aa LT |
3083 | */ |
3084 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
3085 | pgoff_t index, | |
3086 | gfp_t gfp) | |
3087 | { | |
6c45b454 | 3088 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
0531b2aa LT |
3089 | } |
3090 | EXPORT_SYMBOL(read_cache_page_gfp); | |
3091 | ||
9fd91a90 DW |
3092 | /* |
3093 | * Don't operate on ranges the page cache doesn't support, and don't exceed the | |
3094 | * LFS limits. If pos is under the limit it becomes a short access. If it | |
3095 | * exceeds the limit we return -EFBIG. | |
3096 | */ | |
9fd91a90 DW |
3097 | static int generic_write_check_limits(struct file *file, loff_t pos, |
3098 | loff_t *count) | |
3099 | { | |
646955cd AG |
3100 | struct inode *inode = file->f_mapping->host; |
3101 | loff_t max_size = inode->i_sb->s_maxbytes; | |
9fd91a90 DW |
3102 | loff_t limit = rlimit(RLIMIT_FSIZE); |
3103 | ||
3104 | if (limit != RLIM_INFINITY) { | |
3105 | if (pos >= limit) { | |
3106 | send_sig(SIGXFSZ, current, 0); | |
3107 | return -EFBIG; | |
3108 | } | |
3109 | *count = min(*count, limit - pos); | |
3110 | } | |
3111 | ||
646955cd AG |
3112 | if (!(file->f_flags & O_LARGEFILE)) |
3113 | max_size = MAX_NON_LFS; | |
3114 | ||
3115 | if (unlikely(pos >= max_size)) | |
3116 | return -EFBIG; | |
3117 | ||
3118 | *count = min(*count, max_size - pos); | |
3119 | ||
3120 | return 0; | |
9fd91a90 DW |
3121 | } |
3122 | ||
1da177e4 LT |
3123 | /* |
3124 | * Performs necessary checks before doing a write | |
3125 | * | |
485bb99b | 3126 | * Can adjust writing position or amount of bytes to write. |
1da177e4 LT |
3127 | * Returns appropriate error code that caller should return or |
3128 | * zero in case that write should be allowed. | |
3129 | */ | |
3309dd04 | 3130 | inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 3131 | { |
3309dd04 | 3132 | struct file *file = iocb->ki_filp; |
1da177e4 | 3133 | struct inode *inode = file->f_mapping->host; |
9fd91a90 DW |
3134 | loff_t count; |
3135 | int ret; | |
1da177e4 | 3136 | |
dc617f29 DW |
3137 | if (IS_SWAPFILE(inode)) |
3138 | return -ETXTBSY; | |
3139 | ||
3309dd04 AV |
3140 | if (!iov_iter_count(from)) |
3141 | return 0; | |
1da177e4 | 3142 | |
0fa6b005 | 3143 | /* FIXME: this is for backwards compatibility with 2.4 */ |
2ba48ce5 | 3144 | if (iocb->ki_flags & IOCB_APPEND) |
3309dd04 | 3145 | iocb->ki_pos = i_size_read(inode); |
1da177e4 | 3146 | |
6be96d3a GR |
3147 | if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT)) |
3148 | return -EINVAL; | |
3149 | ||
9fd91a90 DW |
3150 | count = iov_iter_count(from); |
3151 | ret = generic_write_check_limits(file, iocb->ki_pos, &count); | |
3152 | if (ret) | |
3153 | return ret; | |
1da177e4 | 3154 | |
9fd91a90 | 3155 | iov_iter_truncate(from, count); |
3309dd04 | 3156 | return iov_iter_count(from); |
1da177e4 LT |
3157 | } |
3158 | EXPORT_SYMBOL(generic_write_checks); | |
3159 | ||
1383a7ed DW |
3160 | /* |
3161 | * Performs necessary checks before doing a clone. | |
3162 | * | |
646955cd | 3163 | * Can adjust amount of bytes to clone via @req_count argument. |
1383a7ed DW |
3164 | * Returns appropriate error code that caller should return or |
3165 | * zero in case the clone should be allowed. | |
3166 | */ | |
3167 | int generic_remap_checks(struct file *file_in, loff_t pos_in, | |
3168 | struct file *file_out, loff_t pos_out, | |
42ec3d4c | 3169 | loff_t *req_count, unsigned int remap_flags) |
1383a7ed DW |
3170 | { |
3171 | struct inode *inode_in = file_in->f_mapping->host; | |
3172 | struct inode *inode_out = file_out->f_mapping->host; | |
3173 | uint64_t count = *req_count; | |
3174 | uint64_t bcount; | |
3175 | loff_t size_in, size_out; | |
3176 | loff_t bs = inode_out->i_sb->s_blocksize; | |
9fd91a90 | 3177 | int ret; |
1383a7ed DW |
3178 | |
3179 | /* The start of both ranges must be aligned to an fs block. */ | |
3180 | if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs)) | |
3181 | return -EINVAL; | |
3182 | ||
3183 | /* Ensure offsets don't wrap. */ | |
3184 | if (pos_in + count < pos_in || pos_out + count < pos_out) | |
3185 | return -EINVAL; | |
3186 | ||
3187 | size_in = i_size_read(inode_in); | |
3188 | size_out = i_size_read(inode_out); | |
3189 | ||
3190 | /* Dedupe requires both ranges to be within EOF. */ | |
3d28193e | 3191 | if ((remap_flags & REMAP_FILE_DEDUP) && |
1383a7ed DW |
3192 | (pos_in >= size_in || pos_in + count > size_in || |
3193 | pos_out >= size_out || pos_out + count > size_out)) | |
3194 | return -EINVAL; | |
3195 | ||
3196 | /* Ensure the infile range is within the infile. */ | |
3197 | if (pos_in >= size_in) | |
3198 | return -EINVAL; | |
3199 | count = min(count, size_in - (uint64_t)pos_in); | |
3200 | ||
9fd91a90 DW |
3201 | ret = generic_write_check_limits(file_out, pos_out, &count); |
3202 | if (ret) | |
3203 | return ret; | |
1da177e4 LT |
3204 | |
3205 | /* | |
1383a7ed DW |
3206 | * If the user wanted us to link to the infile's EOF, round up to the |
3207 | * next block boundary for this check. | |
3208 | * | |
3209 | * Otherwise, make sure the count is also block-aligned, having | |
3210 | * already confirmed the starting offsets' block alignment. | |
1da177e4 | 3211 | */ |
1383a7ed DW |
3212 | if (pos_in + count == size_in) { |
3213 | bcount = ALIGN(size_in, bs) - pos_in; | |
3214 | } else { | |
3215 | if (!IS_ALIGNED(count, bs)) | |
eca3654e | 3216 | count = ALIGN_DOWN(count, bs); |
1383a7ed | 3217 | bcount = count; |
1da177e4 LT |
3218 | } |
3219 | ||
1383a7ed DW |
3220 | /* Don't allow overlapped cloning within the same file. */ |
3221 | if (inode_in == inode_out && | |
3222 | pos_out + bcount > pos_in && | |
3223 | pos_out < pos_in + bcount) | |
3224 | return -EINVAL; | |
3225 | ||
1da177e4 | 3226 | /* |
eca3654e DW |
3227 | * We shortened the request but the caller can't deal with that, so |
3228 | * bounce the request back to userspace. | |
1da177e4 | 3229 | */ |
eca3654e | 3230 | if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN)) |
1383a7ed | 3231 | return -EINVAL; |
1da177e4 | 3232 | |
eca3654e | 3233 | *req_count = count; |
1383a7ed | 3234 | return 0; |
1da177e4 | 3235 | } |
1da177e4 | 3236 | |
a3171351 AG |
3237 | |
3238 | /* | |
3239 | * Performs common checks before doing a file copy/clone | |
3240 | * from @file_in to @file_out. | |
3241 | */ | |
3242 | int generic_file_rw_checks(struct file *file_in, struct file *file_out) | |
3243 | { | |
3244 | struct inode *inode_in = file_inode(file_in); | |
3245 | struct inode *inode_out = file_inode(file_out); | |
3246 | ||
3247 | /* Don't copy dirs, pipes, sockets... */ | |
3248 | if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode)) | |
3249 | return -EISDIR; | |
3250 | if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode)) | |
3251 | return -EINVAL; | |
3252 | ||
3253 | if (!(file_in->f_mode & FMODE_READ) || | |
3254 | !(file_out->f_mode & FMODE_WRITE) || | |
3255 | (file_out->f_flags & O_APPEND)) | |
3256 | return -EBADF; | |
3257 | ||
3258 | return 0; | |
3259 | } | |
3260 | ||
96e6e8f4 AG |
3261 | /* |
3262 | * Performs necessary checks before doing a file copy | |
3263 | * | |
3264 | * Can adjust amount of bytes to copy via @req_count argument. | |
3265 | * Returns appropriate error code that caller should return or | |
3266 | * zero in case the copy should be allowed. | |
3267 | */ | |
3268 | int generic_copy_file_checks(struct file *file_in, loff_t pos_in, | |
3269 | struct file *file_out, loff_t pos_out, | |
3270 | size_t *req_count, unsigned int flags) | |
3271 | { | |
3272 | struct inode *inode_in = file_inode(file_in); | |
3273 | struct inode *inode_out = file_inode(file_out); | |
3274 | uint64_t count = *req_count; | |
3275 | loff_t size_in; | |
3276 | int ret; | |
3277 | ||
3278 | ret = generic_file_rw_checks(file_in, file_out); | |
3279 | if (ret) | |
3280 | return ret; | |
3281 | ||
3282 | /* Don't touch certain kinds of inodes */ | |
3283 | if (IS_IMMUTABLE(inode_out)) | |
3284 | return -EPERM; | |
3285 | ||
3286 | if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out)) | |
3287 | return -ETXTBSY; | |
3288 | ||
3289 | /* Ensure offsets don't wrap. */ | |
3290 | if (pos_in + count < pos_in || pos_out + count < pos_out) | |
3291 | return -EOVERFLOW; | |
3292 | ||
3293 | /* Shorten the copy to EOF */ | |
3294 | size_in = i_size_read(inode_in); | |
3295 | if (pos_in >= size_in) | |
3296 | count = 0; | |
3297 | else | |
3298 | count = min(count, size_in - (uint64_t)pos_in); | |
3299 | ||
3300 | ret = generic_write_check_limits(file_out, pos_out, &count); | |
3301 | if (ret) | |
3302 | return ret; | |
3303 | ||
3304 | /* Don't allow overlapped copying within the same file. */ | |
3305 | if (inode_in == inode_out && | |
3306 | pos_out + count > pos_in && | |
3307 | pos_out < pos_in + count) | |
3308 | return -EINVAL; | |
3309 | ||
3310 | *req_count = count; | |
3311 | return 0; | |
3312 | } | |
3313 | ||
afddba49 NP |
3314 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
3315 | loff_t pos, unsigned len, unsigned flags, | |
3316 | struct page **pagep, void **fsdata) | |
3317 | { | |
3318 | const struct address_space_operations *aops = mapping->a_ops; | |
3319 | ||
4e02ed4b | 3320 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 3321 | pagep, fsdata); |
afddba49 NP |
3322 | } |
3323 | EXPORT_SYMBOL(pagecache_write_begin); | |
3324 | ||
3325 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
3326 | loff_t pos, unsigned len, unsigned copied, | |
3327 | struct page *page, void *fsdata) | |
3328 | { | |
3329 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 3330 | |
4e02ed4b | 3331 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
3332 | } |
3333 | EXPORT_SYMBOL(pagecache_write_end); | |
3334 | ||
a92853b6 KK |
3335 | /* |
3336 | * Warn about a page cache invalidation failure during a direct I/O write. | |
3337 | */ | |
3338 | void dio_warn_stale_pagecache(struct file *filp) | |
3339 | { | |
3340 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); | |
3341 | char pathname[128]; | |
3342 | struct inode *inode = file_inode(filp); | |
3343 | char *path; | |
3344 | ||
3345 | errseq_set(&inode->i_mapping->wb_err, -EIO); | |
3346 | if (__ratelimit(&_rs)) { | |
3347 | path = file_path(filp, pathname, sizeof(pathname)); | |
3348 | if (IS_ERR(path)) | |
3349 | path = "(unknown)"; | |
3350 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); | |
3351 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, | |
3352 | current->comm); | |
3353 | } | |
3354 | } | |
3355 | ||
1da177e4 | 3356 | ssize_t |
1af5bb49 | 3357 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3358 | { |
3359 | struct file *file = iocb->ki_filp; | |
3360 | struct address_space *mapping = file->f_mapping; | |
3361 | struct inode *inode = mapping->host; | |
1af5bb49 | 3362 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3363 | ssize_t written; |
a969e903 CH |
3364 | size_t write_len; |
3365 | pgoff_t end; | |
1da177e4 | 3366 | |
0c949334 | 3367 | write_len = iov_iter_count(from); |
09cbfeaf | 3368 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3369 | |
6be96d3a GR |
3370 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3371 | /* If there are pages to writeback, return */ | |
3372 | if (filemap_range_has_page(inode->i_mapping, pos, | |
35f12f0f | 3373 | pos + write_len - 1)) |
6be96d3a GR |
3374 | return -EAGAIN; |
3375 | } else { | |
3376 | written = filemap_write_and_wait_range(mapping, pos, | |
3377 | pos + write_len - 1); | |
3378 | if (written) | |
3379 | goto out; | |
3380 | } | |
a969e903 CH |
3381 | |
3382 | /* | |
3383 | * After a write we want buffered reads to be sure to go to disk to get | |
3384 | * the new data. We invalidate clean cached page from the region we're | |
3385 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3386 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3387 | */ |
55635ba7 | 3388 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3389 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3390 | /* |
3391 | * If a page can not be invalidated, return 0 to fall back | |
3392 | * to buffered write. | |
3393 | */ | |
3394 | if (written) { | |
3395 | if (written == -EBUSY) | |
3396 | return 0; | |
3397 | goto out; | |
a969e903 CH |
3398 | } |
3399 | ||
639a93a5 | 3400 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3401 | |
3402 | /* | |
3403 | * Finally, try again to invalidate clean pages which might have been | |
3404 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3405 | * if the source of the write was an mmap'ed region of the file | |
3406 | * we're writing. Either one is a pretty crazy thing to do, | |
3407 | * so we don't support it 100%. If this invalidation | |
3408 | * fails, tough, the write still worked... | |
332391a9 LC |
3409 | * |
3410 | * Most of the time we do not need this since dio_complete() will do | |
3411 | * the invalidation for us. However there are some file systems that | |
3412 | * do not end up with dio_complete() being called, so let's not break | |
80c1fe90 KK |
3413 | * them by removing it completely. |
3414 | * | |
9266a140 KK |
3415 | * Noticeable example is a blkdev_direct_IO(). |
3416 | * | |
80c1fe90 | 3417 | * Skip invalidation for async writes or if mapping has no pages. |
a969e903 | 3418 | */ |
9266a140 KK |
3419 | if (written > 0 && mapping->nrpages && |
3420 | invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) | |
3421 | dio_warn_stale_pagecache(file); | |
a969e903 | 3422 | |
1da177e4 | 3423 | if (written > 0) { |
0116651c | 3424 | pos += written; |
639a93a5 | 3425 | write_len -= written; |
0116651c NK |
3426 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3427 | i_size_write(inode, pos); | |
1da177e4 LT |
3428 | mark_inode_dirty(inode); |
3429 | } | |
5cb6c6c7 | 3430 | iocb->ki_pos = pos; |
1da177e4 | 3431 | } |
639a93a5 | 3432 | iov_iter_revert(from, write_len - iov_iter_count(from)); |
a969e903 | 3433 | out: |
1da177e4 LT |
3434 | return written; |
3435 | } | |
3436 | EXPORT_SYMBOL(generic_file_direct_write); | |
3437 | ||
eb2be189 NP |
3438 | /* |
3439 | * Find or create a page at the given pagecache position. Return the locked | |
3440 | * page. This function is specifically for buffered writes. | |
3441 | */ | |
54566b2c NP |
3442 | struct page *grab_cache_page_write_begin(struct address_space *mapping, |
3443 | pgoff_t index, unsigned flags) | |
eb2be189 | 3444 | { |
eb2be189 | 3445 | struct page *page; |
bbddabe2 | 3446 | int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT; |
0faa70cb | 3447 | |
54566b2c | 3448 | if (flags & AOP_FLAG_NOFS) |
2457aec6 MG |
3449 | fgp_flags |= FGP_NOFS; |
3450 | ||
3451 | page = pagecache_get_page(mapping, index, fgp_flags, | |
45f87de5 | 3452 | mapping_gfp_mask(mapping)); |
c585a267 | 3453 | if (page) |
2457aec6 | 3454 | wait_for_stable_page(page); |
eb2be189 | 3455 | |
eb2be189 NP |
3456 | return page; |
3457 | } | |
54566b2c | 3458 | EXPORT_SYMBOL(grab_cache_page_write_begin); |
eb2be189 | 3459 | |
3b93f911 | 3460 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3461 | struct iov_iter *i, loff_t pos) |
3462 | { | |
3463 | struct address_space *mapping = file->f_mapping; | |
3464 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3465 | long status = 0; | |
3466 | ssize_t written = 0; | |
674b892e NP |
3467 | unsigned int flags = 0; |
3468 | ||
afddba49 NP |
3469 | do { |
3470 | struct page *page; | |
afddba49 NP |
3471 | unsigned long offset; /* Offset into pagecache page */ |
3472 | unsigned long bytes; /* Bytes to write to page */ | |
3473 | size_t copied; /* Bytes copied from user */ | |
3474 | void *fsdata; | |
3475 | ||
09cbfeaf KS |
3476 | offset = (pos & (PAGE_SIZE - 1)); |
3477 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3478 | iov_iter_count(i)); |
3479 | ||
3480 | again: | |
00a3d660 LT |
3481 | /* |
3482 | * Bring in the user page that we will copy from _first_. | |
3483 | * Otherwise there's a nasty deadlock on copying from the | |
3484 | * same page as we're writing to, without it being marked | |
3485 | * up-to-date. | |
3486 | * | |
3487 | * Not only is this an optimisation, but it is also required | |
3488 | * to check that the address is actually valid, when atomic | |
3489 | * usercopies are used, below. | |
3490 | */ | |
3491 | if (unlikely(iov_iter_fault_in_readable(i, bytes))) { | |
3492 | status = -EFAULT; | |
3493 | break; | |
3494 | } | |
3495 | ||
296291cd JK |
3496 | if (fatal_signal_pending(current)) { |
3497 | status = -EINTR; | |
3498 | break; | |
3499 | } | |
3500 | ||
674b892e | 3501 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3502 | &page, &fsdata); |
2457aec6 | 3503 | if (unlikely(status < 0)) |
afddba49 NP |
3504 | break; |
3505 | ||
931e80e4 | 3506 | if (mapping_writably_mapped(mapping)) |
3507 | flush_dcache_page(page); | |
00a3d660 | 3508 | |
afddba49 | 3509 | copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); |
afddba49 NP |
3510 | flush_dcache_page(page); |
3511 | ||
3512 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3513 | page, fsdata); | |
3514 | if (unlikely(status < 0)) | |
3515 | break; | |
3516 | copied = status; | |
3517 | ||
3518 | cond_resched(); | |
3519 | ||
124d3b70 | 3520 | iov_iter_advance(i, copied); |
afddba49 NP |
3521 | if (unlikely(copied == 0)) { |
3522 | /* | |
3523 | * If we were unable to copy any data at all, we must | |
3524 | * fall back to a single segment length write. | |
3525 | * | |
3526 | * If we didn't fallback here, we could livelock | |
3527 | * because not all segments in the iov can be copied at | |
3528 | * once without a pagefault. | |
3529 | */ | |
09cbfeaf | 3530 | bytes = min_t(unsigned long, PAGE_SIZE - offset, |
afddba49 NP |
3531 | iov_iter_single_seg_count(i)); |
3532 | goto again; | |
3533 | } | |
afddba49 NP |
3534 | pos += copied; |
3535 | written += copied; | |
3536 | ||
3537 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3538 | } while (iov_iter_count(i)); |
3539 | ||
3540 | return written ? written : status; | |
3541 | } | |
3b93f911 | 3542 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3543 | |
e4dd9de3 | 3544 | /** |
8174202b | 3545 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3546 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3547 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3548 | * |
3549 | * This function does all the work needed for actually writing data to a | |
3550 | * file. It does all basic checks, removes SUID from the file, updates | |
3551 | * modification times and calls proper subroutines depending on whether we | |
3552 | * do direct IO or a standard buffered write. | |
3553 | * | |
3554 | * It expects i_mutex to be grabbed unless we work on a block device or similar | |
3555 | * object which does not need locking at all. | |
3556 | * | |
3557 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3558 | * A caller has to handle it. This is mainly due to the fact that we want to | |
3559 | * avoid syncing under i_mutex. | |
a862f68a MR |
3560 | * |
3561 | * Return: | |
3562 | * * number of bytes written, even for truncated writes | |
3563 | * * negative error code if no data has been written at all | |
e4dd9de3 | 3564 | */ |
8174202b | 3565 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3566 | { |
3567 | struct file *file = iocb->ki_filp; | |
fb5527e6 | 3568 | struct address_space * mapping = file->f_mapping; |
1da177e4 | 3569 | struct inode *inode = mapping->host; |
3b93f911 | 3570 | ssize_t written = 0; |
1da177e4 | 3571 | ssize_t err; |
3b93f911 | 3572 | ssize_t status; |
1da177e4 | 3573 | |
1da177e4 | 3574 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3575 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3576 | err = file_remove_privs(file); |
1da177e4 LT |
3577 | if (err) |
3578 | goto out; | |
3579 | ||
c3b2da31 JB |
3580 | err = file_update_time(file); |
3581 | if (err) | |
3582 | goto out; | |
1da177e4 | 3583 | |
2ba48ce5 | 3584 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3585 | loff_t pos, endbyte; |
fb5527e6 | 3586 | |
1af5bb49 | 3587 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3588 | /* |
fbbbad4b MW |
3589 | * If the write stopped short of completing, fall back to |
3590 | * buffered writes. Some filesystems do this for writes to | |
3591 | * holes, for example. For DAX files, a buffered write will | |
3592 | * not succeed (even if it did, DAX does not handle dirty | |
3593 | * page-cache pages correctly). | |
1da177e4 | 3594 | */ |
0b8def9d | 3595 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3596 | goto out; |
3597 | ||
0b8def9d | 3598 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3599 | /* |
3b93f911 | 3600 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3601 | * then we want to return the number of bytes which were |
3602 | * direct-written, or the error code if that was zero. Note | |
3603 | * that this differs from normal direct-io semantics, which | |
3604 | * will return -EFOO even if some bytes were written. | |
3605 | */ | |
60bb4529 | 3606 | if (unlikely(status < 0)) { |
3b93f911 | 3607 | err = status; |
fb5527e6 JM |
3608 | goto out; |
3609 | } | |
fb5527e6 JM |
3610 | /* |
3611 | * We need to ensure that the page cache pages are written to | |
3612 | * disk and invalidated to preserve the expected O_DIRECT | |
3613 | * semantics. | |
3614 | */ | |
3b93f911 | 3615 | endbyte = pos + status - 1; |
0b8def9d | 3616 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3617 | if (err == 0) { |
0b8def9d | 3618 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3619 | written += status; |
fb5527e6 | 3620 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3621 | pos >> PAGE_SHIFT, |
3622 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3623 | } else { |
3624 | /* | |
3625 | * We don't know how much we wrote, so just return | |
3626 | * the number of bytes which were direct-written | |
3627 | */ | |
3628 | } | |
3629 | } else { | |
0b8def9d AV |
3630 | written = generic_perform_write(file, from, iocb->ki_pos); |
3631 | if (likely(written > 0)) | |
3632 | iocb->ki_pos += written; | |
fb5527e6 | 3633 | } |
1da177e4 LT |
3634 | out: |
3635 | current->backing_dev_info = NULL; | |
3636 | return written ? written : err; | |
3637 | } | |
8174202b | 3638 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3639 | |
e4dd9de3 | 3640 | /** |
8174202b | 3641 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3642 | * @iocb: IO state structure |
8174202b | 3643 | * @from: iov_iter with data to write |
e4dd9de3 | 3644 | * |
8174202b | 3645 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 JK |
3646 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
3647 | * and acquires i_mutex as needed. | |
a862f68a MR |
3648 | * Return: |
3649 | * * negative error code if no data has been written at all of | |
3650 | * vfs_fsync_range() failed for a synchronous write | |
3651 | * * number of bytes written, even for truncated writes | |
e4dd9de3 | 3652 | */ |
8174202b | 3653 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3654 | { |
3655 | struct file *file = iocb->ki_filp; | |
148f948b | 3656 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3657 | ssize_t ret; |
1da177e4 | 3658 | |
5955102c | 3659 | inode_lock(inode); |
3309dd04 AV |
3660 | ret = generic_write_checks(iocb, from); |
3661 | if (ret > 0) | |
5f380c7f | 3662 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3663 | inode_unlock(inode); |
1da177e4 | 3664 | |
e2592217 CH |
3665 | if (ret > 0) |
3666 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3667 | return ret; |
3668 | } | |
8174202b | 3669 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3670 | |
cf9a2ae8 DH |
3671 | /** |
3672 | * try_to_release_page() - release old fs-specific metadata on a page | |
3673 | * | |
3674 | * @page: the page which the kernel is trying to free | |
3675 | * @gfp_mask: memory allocation flags (and I/O mode) | |
3676 | * | |
3677 | * The address_space is to try to release any data against the page | |
a862f68a | 3678 | * (presumably at page->private). |
cf9a2ae8 | 3679 | * |
266cf658 DH |
3680 | * This may also be called if PG_fscache is set on a page, indicating that the |
3681 | * page is known to the local caching routines. | |
3682 | * | |
cf9a2ae8 | 3683 | * The @gfp_mask argument specifies whether I/O may be performed to release |
71baba4b | 3684 | * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS). |
cf9a2ae8 | 3685 | * |
a862f68a | 3686 | * Return: %1 if the release was successful, otherwise return zero. |
cf9a2ae8 DH |
3687 | */ |
3688 | int try_to_release_page(struct page *page, gfp_t gfp_mask) | |
3689 | { | |
3690 | struct address_space * const mapping = page->mapping; | |
3691 | ||
3692 | BUG_ON(!PageLocked(page)); | |
3693 | if (PageWriteback(page)) | |
3694 | return 0; | |
3695 | ||
3696 | if (mapping && mapping->a_ops->releasepage) | |
3697 | return mapping->a_ops->releasepage(page, gfp_mask); | |
3698 | return try_to_free_buffers(page); | |
3699 | } | |
3700 | ||
3701 | EXPORT_SYMBOL(try_to_release_page); |