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> | |
ffa65753 | 24 | #include <linux/swapops.h> |
1da177e4 LT |
25 | #include <linux/mman.h> |
26 | #include <linux/pagemap.h> | |
27 | #include <linux/file.h> | |
28 | #include <linux/uio.h> | |
cfcbfb13 | 29 | #include <linux/error-injection.h> |
1da177e4 LT |
30 | #include <linux/hash.h> |
31 | #include <linux/writeback.h> | |
53253383 | 32 | #include <linux/backing-dev.h> |
1da177e4 | 33 | #include <linux/pagevec.h> |
1da177e4 | 34 | #include <linux/security.h> |
44110fe3 | 35 | #include <linux/cpuset.h> |
00501b53 | 36 | #include <linux/hugetlb.h> |
8a9f3ccd | 37 | #include <linux/memcontrol.h> |
c7df8ad2 | 38 | #include <linux/shmem_fs.h> |
f1820361 | 39 | #include <linux/rmap.h> |
b1d29ba8 | 40 | #include <linux/delayacct.h> |
eb414681 | 41 | #include <linux/psi.h> |
d0e6a582 | 42 | #include <linux/ramfs.h> |
b9306a79 | 43 | #include <linux/page_idle.h> |
ffa65753 | 44 | #include <linux/migrate.h> |
f9ce0be7 | 45 | #include <asm/pgalloc.h> |
de591a82 | 46 | #include <asm/tlbflush.h> |
0f8053a5 NP |
47 | #include "internal.h" |
48 | ||
fe0bfaaf RJ |
49 | #define CREATE_TRACE_POINTS |
50 | #include <trace/events/filemap.h> | |
51 | ||
1da177e4 | 52 | /* |
1da177e4 LT |
53 | * FIXME: remove all knowledge of the buffer layer from the core VM |
54 | */ | |
148f948b | 55 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 56 | |
1da177e4 LT |
57 | #include <asm/mman.h> |
58 | ||
59 | /* | |
60 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
61 | * though. | |
62 | * | |
63 | * Shared mappings now work. 15.8.1995 Bruno. | |
64 | * | |
65 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
66 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
67 | * | |
68 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
69 | */ | |
70 | ||
71 | /* | |
72 | * Lock ordering: | |
73 | * | |
c8c06efa | 74 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 75 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 | 76 | * ->swap_lock (exclusive_swap_page, others) |
b93b0163 | 77 | * ->i_pages lock |
1da177e4 | 78 | * |
9608703e | 79 | * ->i_rwsem |
730633f0 JK |
80 | * ->invalidate_lock (acquired by fs in truncate path) |
81 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) | |
1da177e4 | 82 | * |
c1e8d7c6 | 83 | * ->mmap_lock |
c8c06efa | 84 | * ->i_mmap_rwsem |
b8072f09 | 85 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
b93b0163 | 86 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
1da177e4 | 87 | * |
c1e8d7c6 | 88 | * ->mmap_lock |
730633f0 JK |
89 | * ->invalidate_lock (filemap_fault) |
90 | * ->lock_page (filemap_fault, access_process_vm) | |
1da177e4 | 91 | * |
9608703e | 92 | * ->i_rwsem (generic_perform_write) |
bb523b40 | 93 | * ->mmap_lock (fault_in_readable->do_page_fault) |
1da177e4 | 94 | * |
f758eeab | 95 | * bdi->wb.list_lock |
a66979ab | 96 | * sb_lock (fs/fs-writeback.c) |
b93b0163 | 97 | * ->i_pages lock (__sync_single_inode) |
1da177e4 | 98 | * |
c8c06efa | 99 | * ->i_mmap_rwsem |
1da177e4 LT |
100 | * ->anon_vma.lock (vma_adjust) |
101 | * | |
102 | * ->anon_vma.lock | |
b8072f09 | 103 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 104 | * |
b8072f09 | 105 | * ->page_table_lock or pte_lock |
5d337b91 | 106 | * ->swap_lock (try_to_unmap_one) |
1da177e4 | 107 | * ->private_lock (try_to_unmap_one) |
b93b0163 | 108 | * ->i_pages lock (try_to_unmap_one) |
15b44736 HD |
109 | * ->lruvec->lru_lock (follow_page->mark_page_accessed) |
110 | * ->lruvec->lru_lock (check_pte_range->isolate_lru_page) | |
1da177e4 | 111 | * ->private_lock (page_remove_rmap->set_page_dirty) |
b93b0163 | 112 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
f758eeab | 113 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 114 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 115 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 116 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 117 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
118 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
119 | * | |
c8c06efa | 120 | * ->i_mmap_rwsem |
9a3c531d | 121 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
122 | */ |
123 | ||
5c024e6a | 124 | static void page_cache_delete(struct address_space *mapping, |
a548b615 | 125 | struct folio *folio, void *shadow) |
91b0abe3 | 126 | { |
a548b615 MWO |
127 | XA_STATE(xas, &mapping->i_pages, folio->index); |
128 | long nr = 1; | |
c70b647d | 129 | |
5c024e6a | 130 | mapping_set_update(&xas, mapping); |
c70b647d | 131 | |
5c024e6a | 132 | /* hugetlb pages are represented by a single entry in the xarray */ |
a548b615 MWO |
133 | if (!folio_test_hugetlb(folio)) { |
134 | xas_set_order(&xas, folio->index, folio_order(folio)); | |
135 | nr = folio_nr_pages(folio); | |
5c024e6a | 136 | } |
91b0abe3 | 137 | |
a548b615 | 138 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
449dd698 | 139 | |
5c024e6a MW |
140 | xas_store(&xas, shadow); |
141 | xas_init_marks(&xas); | |
d3798ae8 | 142 | |
a548b615 | 143 | folio->mapping = NULL; |
2300638b | 144 | /* Leave page->index set: truncation lookup relies upon it */ |
d3798ae8 | 145 | mapping->nrpages -= nr; |
91b0abe3 JW |
146 | } |
147 | ||
621db488 MWO |
148 | static void filemap_unaccount_folio(struct address_space *mapping, |
149 | struct folio *folio) | |
1da177e4 | 150 | { |
621db488 | 151 | long nr; |
1da177e4 | 152 | |
621db488 MWO |
153 | VM_BUG_ON_FOLIO(folio_mapped(folio), folio); |
154 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) { | |
06b241f3 HD |
155 | int mapcount; |
156 | ||
157 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
621db488 MWO |
158 | current->comm, folio_pfn(folio)); |
159 | dump_page(&folio->page, "still mapped when deleted"); | |
06b241f3 HD |
160 | dump_stack(); |
161 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
162 | ||
621db488 | 163 | mapcount = page_mapcount(&folio->page); |
06b241f3 | 164 | if (mapping_exiting(mapping) && |
621db488 | 165 | folio_ref_count(folio) >= mapcount + 2) { |
06b241f3 HD |
166 | /* |
167 | * All vmas have already been torn down, so it's | |
621db488 | 168 | * a good bet that actually the folio is unmapped, |
06b241f3 HD |
169 | * and we'd prefer not to leak it: if we're wrong, |
170 | * some other bad page check should catch it later. | |
171 | */ | |
621db488 MWO |
172 | page_mapcount_reset(&folio->page); |
173 | folio_ref_sub(folio, mapcount); | |
06b241f3 HD |
174 | } |
175 | } | |
176 | ||
621db488 MWO |
177 | /* hugetlb folios do not participate in page cache accounting. */ |
178 | if (folio_test_hugetlb(folio)) | |
5ecc4d85 | 179 | return; |
09612fa6 | 180 | |
621db488 | 181 | nr = folio_nr_pages(folio); |
5ecc4d85 | 182 | |
621db488 MWO |
183 | __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr); |
184 | if (folio_test_swapbacked(folio)) { | |
185 | __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr); | |
186 | if (folio_test_pmd_mappable(folio)) | |
187 | __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr); | |
188 | } else if (folio_test_pmd_mappable(folio)) { | |
189 | __lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr); | |
09d91cda | 190 | filemap_nr_thps_dec(mapping); |
800d8c63 | 191 | } |
5ecc4d85 JK |
192 | |
193 | /* | |
621db488 MWO |
194 | * At this point folio must be either written or cleaned by |
195 | * truncate. Dirty folio here signals a bug and loss of | |
5ecc4d85 JK |
196 | * unwritten data. |
197 | * | |
621db488 MWO |
198 | * This fixes dirty accounting after removing the folio entirely |
199 | * but leaves the dirty flag set: it has no effect for truncated | |
200 | * folio and anyway will be cleared before returning folio to | |
5ecc4d85 JK |
201 | * buddy allocator. |
202 | */ | |
621db488 MWO |
203 | if (WARN_ON_ONCE(folio_test_dirty(folio))) |
204 | folio_account_cleaned(folio, mapping, | |
205 | inode_to_wb(mapping->host)); | |
5ecc4d85 JK |
206 | } |
207 | ||
208 | /* | |
209 | * Delete a page from the page cache and free it. Caller has to make | |
210 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 211 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 | 212 | */ |
452e9e69 | 213 | void __filemap_remove_folio(struct folio *folio, void *shadow) |
5ecc4d85 | 214 | { |
452e9e69 | 215 | struct address_space *mapping = folio->mapping; |
5ecc4d85 | 216 | |
a0580c6f | 217 | trace_mm_filemap_delete_from_page_cache(folio); |
621db488 | 218 | filemap_unaccount_folio(mapping, folio); |
a548b615 | 219 | page_cache_delete(mapping, folio, shadow); |
1da177e4 LT |
220 | } |
221 | ||
78f42660 | 222 | void filemap_free_folio(struct address_space *mapping, struct folio *folio) |
59c66c5f JK |
223 | { |
224 | void (*freepage)(struct page *); | |
3abb28e2 | 225 | int refs = 1; |
59c66c5f JK |
226 | |
227 | freepage = mapping->a_ops->freepage; | |
228 | if (freepage) | |
452e9e69 | 229 | freepage(&folio->page); |
59c66c5f | 230 | |
3abb28e2 MWO |
231 | if (folio_test_large(folio) && !folio_test_hugetlb(folio)) |
232 | refs = folio_nr_pages(folio); | |
233 | folio_put_refs(folio, refs); | |
59c66c5f JK |
234 | } |
235 | ||
702cfbf9 | 236 | /** |
452e9e69 MWO |
237 | * filemap_remove_folio - Remove folio from page cache. |
238 | * @folio: The folio. | |
702cfbf9 | 239 | * |
452e9e69 MWO |
240 | * This must be called only on folios that are locked and have been |
241 | * verified to be in the page cache. It will never put the folio into | |
242 | * the free list because the caller has a reference on the page. | |
702cfbf9 | 243 | */ |
452e9e69 | 244 | void filemap_remove_folio(struct folio *folio) |
1da177e4 | 245 | { |
452e9e69 | 246 | struct address_space *mapping = folio->mapping; |
1da177e4 | 247 | |
452e9e69 | 248 | BUG_ON(!folio_test_locked(folio)); |
51b8c1fe | 249 | spin_lock(&mapping->host->i_lock); |
30472509 | 250 | xa_lock_irq(&mapping->i_pages); |
452e9e69 | 251 | __filemap_remove_folio(folio, NULL); |
30472509 | 252 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
253 | if (mapping_shrinkable(mapping)) |
254 | inode_add_lru(mapping->host); | |
255 | spin_unlock(&mapping->host->i_lock); | |
6072d13c | 256 | |
452e9e69 | 257 | filemap_free_folio(mapping, folio); |
97cecb5a | 258 | } |
97cecb5a | 259 | |
aa65c29c | 260 | /* |
51dcbdac MWO |
261 | * page_cache_delete_batch - delete several folios from page cache |
262 | * @mapping: the mapping to which folios belong | |
263 | * @fbatch: batch of folios to delete | |
aa65c29c | 264 | * |
51dcbdac MWO |
265 | * The function walks over mapping->i_pages and removes folios passed in |
266 | * @fbatch from the mapping. The function expects @fbatch to be sorted | |
267 | * by page index and is optimised for it to be dense. | |
268 | * It tolerates holes in @fbatch (mapping entries at those indices are not | |
269 | * modified). | |
aa65c29c | 270 | * |
b93b0163 | 271 | * The function expects the i_pages lock to be held. |
aa65c29c | 272 | */ |
ef8e5717 | 273 | static void page_cache_delete_batch(struct address_space *mapping, |
51dcbdac | 274 | struct folio_batch *fbatch) |
aa65c29c | 275 | { |
51dcbdac | 276 | XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index); |
6b24ca4a | 277 | long total_pages = 0; |
4101196b | 278 | int i = 0; |
1afd7ae5 | 279 | struct folio *folio; |
aa65c29c | 280 | |
ef8e5717 | 281 | mapping_set_update(&xas, mapping); |
1afd7ae5 | 282 | xas_for_each(&xas, folio, ULONG_MAX) { |
51dcbdac | 283 | if (i >= folio_batch_count(fbatch)) |
aa65c29c | 284 | break; |
4101196b MWO |
285 | |
286 | /* A swap/dax/shadow entry got inserted? Skip it. */ | |
1afd7ae5 | 287 | if (xa_is_value(folio)) |
aa65c29c | 288 | continue; |
4101196b MWO |
289 | /* |
290 | * A page got inserted in our range? Skip it. We have our | |
291 | * pages locked so they are protected from being removed. | |
292 | * If we see a page whose index is higher than ours, it | |
293 | * means our page has been removed, which shouldn't be | |
294 | * possible because we're holding the PageLock. | |
295 | */ | |
51dcbdac | 296 | if (folio != fbatch->folios[i]) { |
1afd7ae5 | 297 | VM_BUG_ON_FOLIO(folio->index > |
51dcbdac | 298 | fbatch->folios[i]->index, folio); |
4101196b MWO |
299 | continue; |
300 | } | |
301 | ||
1afd7ae5 | 302 | WARN_ON_ONCE(!folio_test_locked(folio)); |
4101196b | 303 | |
6b24ca4a | 304 | folio->mapping = NULL; |
51dcbdac | 305 | /* Leave folio->index set: truncation lookup relies on it */ |
4101196b | 306 | |
6b24ca4a | 307 | i++; |
ef8e5717 | 308 | xas_store(&xas, NULL); |
6b24ca4a | 309 | total_pages += folio_nr_pages(folio); |
aa65c29c JK |
310 | } |
311 | mapping->nrpages -= total_pages; | |
312 | } | |
313 | ||
314 | void delete_from_page_cache_batch(struct address_space *mapping, | |
51dcbdac | 315 | struct folio_batch *fbatch) |
aa65c29c JK |
316 | { |
317 | int i; | |
aa65c29c | 318 | |
51dcbdac | 319 | if (!folio_batch_count(fbatch)) |
aa65c29c JK |
320 | return; |
321 | ||
51b8c1fe | 322 | spin_lock(&mapping->host->i_lock); |
30472509 | 323 | xa_lock_irq(&mapping->i_pages); |
51dcbdac MWO |
324 | for (i = 0; i < folio_batch_count(fbatch); i++) { |
325 | struct folio *folio = fbatch->folios[i]; | |
aa65c29c | 326 | |
a0580c6f MWO |
327 | trace_mm_filemap_delete_from_page_cache(folio); |
328 | filemap_unaccount_folio(mapping, folio); | |
aa65c29c | 329 | } |
51dcbdac | 330 | page_cache_delete_batch(mapping, fbatch); |
30472509 | 331 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
332 | if (mapping_shrinkable(mapping)) |
333 | inode_add_lru(mapping->host); | |
334 | spin_unlock(&mapping->host->i_lock); | |
aa65c29c | 335 | |
51dcbdac MWO |
336 | for (i = 0; i < folio_batch_count(fbatch); i++) |
337 | filemap_free_folio(mapping, fbatch->folios[i]); | |
aa65c29c JK |
338 | } |
339 | ||
d72d9e2a | 340 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
341 | { |
342 | int ret = 0; | |
343 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
344 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
345 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 346 | ret = -ENOSPC; |
7fcbbaf1 JA |
347 | if (test_bit(AS_EIO, &mapping->flags) && |
348 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
349 | ret = -EIO; |
350 | return ret; | |
351 | } | |
d72d9e2a | 352 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 353 | |
76341cab JL |
354 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
355 | { | |
356 | /* Check for outstanding write errors */ | |
357 | if (test_bit(AS_EIO, &mapping->flags)) | |
358 | return -EIO; | |
359 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
360 | return -ENOSPC; | |
361 | return 0; | |
362 | } | |
363 | ||
5a798493 JB |
364 | /** |
365 | * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range | |
366 | * @mapping: address space structure to write | |
367 | * @wbc: the writeback_control controlling the writeout | |
368 | * | |
369 | * Call writepages on the mapping using the provided wbc to control the | |
370 | * writeout. | |
371 | * | |
372 | * Return: %0 on success, negative error code otherwise. | |
373 | */ | |
374 | int filemap_fdatawrite_wbc(struct address_space *mapping, | |
375 | struct writeback_control *wbc) | |
376 | { | |
377 | int ret; | |
378 | ||
379 | if (!mapping_can_writeback(mapping) || | |
380 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
381 | return 0; | |
382 | ||
383 | wbc_attach_fdatawrite_inode(wbc, mapping->host); | |
384 | ret = do_writepages(mapping, wbc); | |
385 | wbc_detach_inode(wbc); | |
386 | return ret; | |
387 | } | |
388 | EXPORT_SYMBOL(filemap_fdatawrite_wbc); | |
389 | ||
1da177e4 | 390 | /** |
485bb99b | 391 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
392 | * @mapping: address space structure to write |
393 | * @start: offset in bytes where the range starts | |
469eb4d0 | 394 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 395 | * @sync_mode: enable synchronous operation |
1da177e4 | 396 | * |
485bb99b RD |
397 | * Start writeback against all of a mapping's dirty pages that lie |
398 | * within the byte offsets <start, end> inclusive. | |
399 | * | |
1da177e4 | 400 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 401 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
402 | * these two operations is that if a dirty page/buffer is encountered, it must |
403 | * be waited upon, and not just skipped over. | |
a862f68a MR |
404 | * |
405 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 406 | */ |
ebcf28e1 AM |
407 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
408 | loff_t end, int sync_mode) | |
1da177e4 | 409 | { |
1da177e4 LT |
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 | ||
5a798493 | 417 | return filemap_fdatawrite_wbc(mapping, &wbc); |
1da177e4 LT |
418 | } |
419 | ||
420 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
421 | int sync_mode) | |
422 | { | |
111ebb6e | 423 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
424 | } |
425 | ||
426 | int filemap_fdatawrite(struct address_space *mapping) | |
427 | { | |
428 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
429 | } | |
430 | EXPORT_SYMBOL(filemap_fdatawrite); | |
431 | ||
f4c0a0fd | 432 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 433 | loff_t end) |
1da177e4 LT |
434 | { |
435 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
436 | } | |
f4c0a0fd | 437 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 438 | |
485bb99b RD |
439 | /** |
440 | * filemap_flush - mostly a non-blocking flush | |
441 | * @mapping: target address_space | |
442 | * | |
1da177e4 LT |
443 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
444 | * purposes - I/O may not be started against all dirty pages. | |
a862f68a MR |
445 | * |
446 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 LT |
447 | */ |
448 | int filemap_flush(struct address_space *mapping) | |
449 | { | |
450 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
451 | } | |
452 | EXPORT_SYMBOL(filemap_flush); | |
453 | ||
7fc9e472 GR |
454 | /** |
455 | * filemap_range_has_page - check if a page exists in range. | |
456 | * @mapping: address space within which to check | |
457 | * @start_byte: offset in bytes where the range starts | |
458 | * @end_byte: offset in bytes where the range ends (inclusive) | |
459 | * | |
460 | * Find at least one page in the range supplied, usually used to check if | |
461 | * direct writing in this range will trigger a writeback. | |
a862f68a MR |
462 | * |
463 | * Return: %true if at least one page exists in the specified range, | |
464 | * %false otherwise. | |
7fc9e472 GR |
465 | */ |
466 | bool filemap_range_has_page(struct address_space *mapping, | |
467 | loff_t start_byte, loff_t end_byte) | |
468 | { | |
f7b68046 | 469 | struct page *page; |
8fa8e538 MW |
470 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
471 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
7fc9e472 GR |
472 | |
473 | if (end_byte < start_byte) | |
474 | return false; | |
475 | ||
8fa8e538 MW |
476 | rcu_read_lock(); |
477 | for (;;) { | |
478 | page = xas_find(&xas, max); | |
479 | if (xas_retry(&xas, page)) | |
480 | continue; | |
481 | /* Shadow entries don't count */ | |
482 | if (xa_is_value(page)) | |
483 | continue; | |
484 | /* | |
485 | * We don't need to try to pin this page; we're about to | |
486 | * release the RCU lock anyway. It is enough to know that | |
487 | * there was a page here recently. | |
488 | */ | |
489 | break; | |
490 | } | |
491 | rcu_read_unlock(); | |
7fc9e472 | 492 | |
8fa8e538 | 493 | return page != NULL; |
7fc9e472 GR |
494 | } |
495 | EXPORT_SYMBOL(filemap_range_has_page); | |
496 | ||
5e8fcc1a | 497 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 498 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 499 | { |
09cbfeaf KS |
500 | pgoff_t index = start_byte >> PAGE_SHIFT; |
501 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
502 | struct pagevec pvec; |
503 | int nr_pages; | |
1da177e4 | 504 | |
94004ed7 | 505 | if (end_byte < start_byte) |
5e8fcc1a | 506 | return; |
1da177e4 | 507 | |
86679820 | 508 | pagevec_init(&pvec); |
312e9d2f | 509 | while (index <= end) { |
1da177e4 LT |
510 | unsigned i; |
511 | ||
312e9d2f | 512 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 513 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
514 | if (!nr_pages) |
515 | break; | |
516 | ||
1da177e4 LT |
517 | for (i = 0; i < nr_pages; i++) { |
518 | struct page *page = pvec.pages[i]; | |
519 | ||
1da177e4 | 520 | wait_on_page_writeback(page); |
5e8fcc1a | 521 | ClearPageError(page); |
1da177e4 LT |
522 | } |
523 | pagevec_release(&pvec); | |
524 | cond_resched(); | |
525 | } | |
aa750fd7 JN |
526 | } |
527 | ||
528 | /** | |
529 | * filemap_fdatawait_range - wait for writeback to complete | |
530 | * @mapping: address space structure to wait for | |
531 | * @start_byte: offset in bytes where the range starts | |
532 | * @end_byte: offset in bytes where the range ends (inclusive) | |
533 | * | |
534 | * Walk the list of under-writeback pages of the given address space | |
535 | * in the given range and wait for all of them. Check error status of | |
536 | * the address space and return it. | |
537 | * | |
538 | * Since the error status of the address space is cleared by this function, | |
539 | * callers are responsible for checking the return value and handling and/or | |
540 | * reporting the error. | |
a862f68a MR |
541 | * |
542 | * Return: error status of the address space. | |
aa750fd7 JN |
543 | */ |
544 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
545 | loff_t end_byte) | |
546 | { | |
5e8fcc1a JL |
547 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
548 | return filemap_check_errors(mapping); | |
1da177e4 | 549 | } |
d3bccb6f JK |
550 | EXPORT_SYMBOL(filemap_fdatawait_range); |
551 | ||
aa0bfcd9 RZ |
552 | /** |
553 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete | |
554 | * @mapping: address space structure to wait for | |
555 | * @start_byte: offset in bytes where the range starts | |
556 | * @end_byte: offset in bytes where the range ends (inclusive) | |
557 | * | |
558 | * Walk the list of under-writeback pages of the given address space in the | |
559 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), | |
560 | * this function does not clear error status of the address space. | |
561 | * | |
562 | * Use this function if callers don't handle errors themselves. Expected | |
563 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
564 | * fsfreeze(8) | |
565 | */ | |
566 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, | |
567 | loff_t start_byte, loff_t end_byte) | |
568 | { | |
569 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
570 | return filemap_check_and_keep_errors(mapping); | |
571 | } | |
572 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); | |
573 | ||
a823e458 JL |
574 | /** |
575 | * file_fdatawait_range - wait for writeback to complete | |
576 | * @file: file pointing to address space structure to wait for | |
577 | * @start_byte: offset in bytes where the range starts | |
578 | * @end_byte: offset in bytes where the range ends (inclusive) | |
579 | * | |
580 | * Walk the list of under-writeback pages of the address space that file | |
581 | * refers to, in the given range and wait for all of them. Check error | |
582 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
583 | * | |
584 | * Since the error status of the file is advanced by this function, | |
585 | * callers are responsible for checking the return value and handling and/or | |
586 | * reporting the error. | |
a862f68a MR |
587 | * |
588 | * Return: error status of the address space vs. the file->f_wb_err cursor. | |
a823e458 JL |
589 | */ |
590 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
591 | { | |
592 | struct address_space *mapping = file->f_mapping; | |
593 | ||
594 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
595 | return file_check_and_advance_wb_err(file); | |
596 | } | |
597 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 598 | |
aa750fd7 JN |
599 | /** |
600 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
601 | * @mapping: address space structure to wait for | |
602 | * | |
603 | * Walk the list of under-writeback pages of the given address space | |
604 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
605 | * does not clear error status of the address space. | |
606 | * | |
607 | * Use this function if callers don't handle errors themselves. Expected | |
608 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
609 | * fsfreeze(8) | |
a862f68a MR |
610 | * |
611 | * Return: error status of the address space. | |
aa750fd7 | 612 | */ |
76341cab | 613 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 614 | { |
ffb959bb | 615 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 616 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 617 | } |
76341cab | 618 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 619 | |
875d91b1 | 620 | /* Returns true if writeback might be needed or already in progress. */ |
9326c9b2 | 621 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 622 | { |
875d91b1 | 623 | return mapping->nrpages; |
1da177e4 | 624 | } |
1da177e4 | 625 | |
4bdcd1dd JA |
626 | bool filemap_range_has_writeback(struct address_space *mapping, |
627 | loff_t start_byte, loff_t end_byte) | |
f8ee8909 JA |
628 | { |
629 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); | |
630 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
631 | struct page *page; | |
632 | ||
633 | if (end_byte < start_byte) | |
634 | return false; | |
635 | ||
636 | rcu_read_lock(); | |
637 | xas_for_each(&xas, page, max) { | |
638 | if (xas_retry(&xas, page)) | |
639 | continue; | |
640 | if (xa_is_value(page)) | |
641 | continue; | |
642 | if (PageDirty(page) || PageLocked(page) || PageWriteback(page)) | |
643 | break; | |
644 | } | |
645 | rcu_read_unlock(); | |
646 | return page != NULL; | |
63135aa3 | 647 | } |
4bdcd1dd | 648 | EXPORT_SYMBOL_GPL(filemap_range_has_writeback); |
63135aa3 | 649 | |
485bb99b RD |
650 | /** |
651 | * filemap_write_and_wait_range - write out & wait on a file range | |
652 | * @mapping: the address_space for the pages | |
653 | * @lstart: offset in bytes where the range starts | |
654 | * @lend: offset in bytes where the range ends (inclusive) | |
655 | * | |
469eb4d0 AM |
656 | * Write out and wait upon file offsets lstart->lend, inclusive. |
657 | * | |
0e056eb5 | 658 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 | 659 | * that this function can be used to write to the very end-of-file (end = -1). |
a862f68a MR |
660 | * |
661 | * Return: error status of the address space. | |
469eb4d0 | 662 | */ |
1da177e4 LT |
663 | int filemap_write_and_wait_range(struct address_space *mapping, |
664 | loff_t lstart, loff_t lend) | |
665 | { | |
28fd1298 | 666 | int err = 0; |
1da177e4 | 667 | |
9326c9b2 | 668 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
669 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
670 | WB_SYNC_ALL); | |
ddf8f376 IW |
671 | /* |
672 | * Even if the above returned error, the pages may be | |
673 | * written partially (e.g. -ENOSPC), so we wait for it. | |
674 | * But the -EIO is special case, it may indicate the worst | |
675 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
676 | */ | |
28fd1298 | 677 | if (err != -EIO) { |
94004ed7 CH |
678 | int err2 = filemap_fdatawait_range(mapping, |
679 | lstart, lend); | |
28fd1298 OH |
680 | if (!err) |
681 | err = err2; | |
cbeaf951 JL |
682 | } else { |
683 | /* Clear any previously stored errors */ | |
684 | filemap_check_errors(mapping); | |
28fd1298 | 685 | } |
865ffef3 DM |
686 | } else { |
687 | err = filemap_check_errors(mapping); | |
1da177e4 | 688 | } |
28fd1298 | 689 | return err; |
1da177e4 | 690 | } |
f6995585 | 691 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 692 | |
5660e13d JL |
693 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
694 | { | |
3acdfd28 | 695 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
696 | |
697 | trace_filemap_set_wb_err(mapping, eseq); | |
698 | } | |
699 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
700 | ||
701 | /** | |
702 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
703 | * and advance wb_err to current one | |
704 | * @file: struct file on which the error is being reported | |
705 | * | |
706 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
707 | * want to report any writeback errors that occurred since the last fsync (or | |
708 | * since the file was opened if there haven't been any). | |
709 | * | |
710 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
711 | * then just quickly return 0. The file is all caught up. | |
712 | * | |
713 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
714 | * it and try to swap it into place. If it works, or another task beat us | |
715 | * to it with the new value, then update the f_wb_err and return the error | |
716 | * portion. The error at this point must be reported via proper channels | |
717 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
718 | * | |
719 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
720 | * value is protected by the f_lock since we must ensure that it reflects | |
721 | * the latest value swapped in for this file descriptor. | |
a862f68a MR |
722 | * |
723 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
724 | */ |
725 | int file_check_and_advance_wb_err(struct file *file) | |
726 | { | |
727 | int err = 0; | |
728 | errseq_t old = READ_ONCE(file->f_wb_err); | |
729 | struct address_space *mapping = file->f_mapping; | |
730 | ||
731 | /* Locklessly handle the common case where nothing has changed */ | |
732 | if (errseq_check(&mapping->wb_err, old)) { | |
733 | /* Something changed, must use slow path */ | |
734 | spin_lock(&file->f_lock); | |
735 | old = file->f_wb_err; | |
736 | err = errseq_check_and_advance(&mapping->wb_err, | |
737 | &file->f_wb_err); | |
738 | trace_file_check_and_advance_wb_err(file, old); | |
739 | spin_unlock(&file->f_lock); | |
740 | } | |
f4e222c5 JL |
741 | |
742 | /* | |
743 | * We're mostly using this function as a drop in replacement for | |
744 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
745 | * that the legacy code would have had on these flags. | |
746 | */ | |
747 | clear_bit(AS_EIO, &mapping->flags); | |
748 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
749 | return err; |
750 | } | |
751 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
752 | ||
753 | /** | |
754 | * file_write_and_wait_range - write out & wait on a file range | |
755 | * @file: file pointing to address_space with pages | |
756 | * @lstart: offset in bytes where the range starts | |
757 | * @lend: offset in bytes where the range ends (inclusive) | |
758 | * | |
759 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
760 | * | |
761 | * Note that @lend is inclusive (describes the last byte to be written) so | |
762 | * that this function can be used to write to the very end-of-file (end = -1). | |
763 | * | |
764 | * After writing out and waiting on the data, we check and advance the | |
765 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
a862f68a MR |
766 | * |
767 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
768 | */ |
769 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
770 | { | |
771 | int err = 0, err2; | |
772 | struct address_space *mapping = file->f_mapping; | |
773 | ||
9326c9b2 | 774 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
775 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
776 | WB_SYNC_ALL); | |
777 | /* See comment of filemap_write_and_wait() */ | |
778 | if (err != -EIO) | |
779 | __filemap_fdatawait_range(mapping, lstart, lend); | |
780 | } | |
781 | err2 = file_check_and_advance_wb_err(file); | |
782 | if (!err) | |
783 | err = err2; | |
784 | return err; | |
785 | } | |
786 | EXPORT_SYMBOL(file_write_and_wait_range); | |
787 | ||
ef6a3c63 MS |
788 | /** |
789 | * replace_page_cache_page - replace a pagecache page with a new one | |
790 | * @old: page to be replaced | |
791 | * @new: page to replace with | |
ef6a3c63 MS |
792 | * |
793 | * This function replaces a page in the pagecache with a new one. On | |
794 | * success it acquires the pagecache reference for the new page and | |
795 | * drops it for the old page. Both the old and new pages must be | |
796 | * locked. This function does not add the new page to the LRU, the | |
797 | * caller must do that. | |
798 | * | |
74d60958 | 799 | * The remove + add is atomic. This function cannot fail. |
ef6a3c63 | 800 | */ |
1f7ef657 | 801 | void replace_page_cache_page(struct page *old, struct page *new) |
ef6a3c63 | 802 | { |
d21bba2b MWO |
803 | struct folio *fold = page_folio(old); |
804 | struct folio *fnew = page_folio(new); | |
74d60958 MW |
805 | struct address_space *mapping = old->mapping; |
806 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
807 | pgoff_t offset = old->index; | |
808 | XA_STATE(xas, &mapping->i_pages, offset); | |
ef6a3c63 | 809 | |
309381fe SL |
810 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
811 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
812 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 813 | |
74d60958 MW |
814 | get_page(new); |
815 | new->mapping = mapping; | |
816 | new->index = offset; | |
ef6a3c63 | 817 | |
d21bba2b | 818 | mem_cgroup_migrate(fold, fnew); |
0d1c2072 | 819 | |
30472509 | 820 | xas_lock_irq(&xas); |
74d60958 | 821 | xas_store(&xas, new); |
4165b9b4 | 822 | |
74d60958 MW |
823 | old->mapping = NULL; |
824 | /* hugetlb pages do not participate in page cache accounting. */ | |
825 | if (!PageHuge(old)) | |
0d1c2072 | 826 | __dec_lruvec_page_state(old, NR_FILE_PAGES); |
74d60958 | 827 | if (!PageHuge(new)) |
0d1c2072 | 828 | __inc_lruvec_page_state(new, NR_FILE_PAGES); |
74d60958 | 829 | if (PageSwapBacked(old)) |
0d1c2072 | 830 | __dec_lruvec_page_state(old, NR_SHMEM); |
74d60958 | 831 | if (PageSwapBacked(new)) |
0d1c2072 | 832 | __inc_lruvec_page_state(new, NR_SHMEM); |
30472509 | 833 | xas_unlock_irq(&xas); |
74d60958 MW |
834 | if (freepage) |
835 | freepage(old); | |
836 | put_page(old); | |
ef6a3c63 MS |
837 | } |
838 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
839 | ||
9dd3d069 MWO |
840 | noinline int __filemap_add_folio(struct address_space *mapping, |
841 | struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp) | |
1da177e4 | 842 | { |
9dd3d069 MWO |
843 | XA_STATE(xas, &mapping->i_pages, index); |
844 | int huge = folio_test_hugetlb(folio); | |
da74240e | 845 | bool charged = false; |
d68eccad | 846 | long nr = 1; |
e286781d | 847 | |
9dd3d069 MWO |
848 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
849 | VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio); | |
74d60958 | 850 | mapping_set_update(&xas, mapping); |
e286781d | 851 | |
3fea5a49 | 852 | if (!huge) { |
d68eccad | 853 | int error = mem_cgroup_charge(folio, NULL, gfp); |
9dd3d069 | 854 | VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio); |
3fea5a49 | 855 | if (error) |
d68eccad | 856 | return error; |
da74240e | 857 | charged = true; |
d68eccad MWO |
858 | xas_set_order(&xas, index, folio_order(folio)); |
859 | nr = folio_nr_pages(folio); | |
3fea5a49 JW |
860 | } |
861 | ||
198b62f8 | 862 | gfp &= GFP_RECLAIM_MASK; |
d68eccad MWO |
863 | folio_ref_add(folio, nr); |
864 | folio->mapping = mapping; | |
865 | folio->index = xas.xa_index; | |
198b62f8 | 866 | |
74d60958 | 867 | do { |
198b62f8 MWO |
868 | unsigned int order = xa_get_order(xas.xa, xas.xa_index); |
869 | void *entry, *old = NULL; | |
870 | ||
9dd3d069 | 871 | if (order > folio_order(folio)) |
198b62f8 MWO |
872 | xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index), |
873 | order, gfp); | |
74d60958 | 874 | xas_lock_irq(&xas); |
198b62f8 MWO |
875 | xas_for_each_conflict(&xas, entry) { |
876 | old = entry; | |
877 | if (!xa_is_value(entry)) { | |
878 | xas_set_err(&xas, -EEXIST); | |
879 | goto unlock; | |
880 | } | |
881 | } | |
882 | ||
883 | if (old) { | |
884 | if (shadowp) | |
885 | *shadowp = old; | |
886 | /* entry may have been split before we acquired lock */ | |
887 | order = xa_get_order(xas.xa, xas.xa_index); | |
9dd3d069 | 888 | if (order > folio_order(folio)) { |
d68eccad MWO |
889 | /* How to handle large swap entries? */ |
890 | BUG_ON(shmem_mapping(mapping)); | |
198b62f8 MWO |
891 | xas_split(&xas, old, order); |
892 | xas_reset(&xas); | |
893 | } | |
894 | } | |
895 | ||
9dd3d069 | 896 | xas_store(&xas, folio); |
74d60958 MW |
897 | if (xas_error(&xas)) |
898 | goto unlock; | |
899 | ||
d68eccad | 900 | mapping->nrpages += nr; |
74d60958 MW |
901 | |
902 | /* hugetlb pages do not participate in page cache accounting */ | |
d68eccad MWO |
903 | if (!huge) { |
904 | __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr); | |
905 | if (folio_test_pmd_mappable(folio)) | |
906 | __lruvec_stat_mod_folio(folio, | |
907 | NR_FILE_THPS, nr); | |
908 | } | |
74d60958 MW |
909 | unlock: |
910 | xas_unlock_irq(&xas); | |
198b62f8 | 911 | } while (xas_nomem(&xas, gfp)); |
74d60958 | 912 | |
d68eccad | 913 | if (xas_error(&xas)) |
74d60958 | 914 | goto error; |
4165b9b4 | 915 | |
a0580c6f | 916 | trace_mm_filemap_add_to_page_cache(folio); |
66a0c8ee | 917 | return 0; |
74d60958 | 918 | error: |
d68eccad MWO |
919 | if (charged) |
920 | mem_cgroup_uncharge(folio); | |
9dd3d069 | 921 | folio->mapping = NULL; |
66a0c8ee | 922 | /* Leave page->index set: truncation relies upon it */ |
d68eccad MWO |
923 | folio_put_refs(folio, nr); |
924 | return xas_error(&xas); | |
1da177e4 | 925 | } |
9dd3d069 | 926 | ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO); |
a528910e JW |
927 | |
928 | /** | |
929 | * add_to_page_cache_locked - add a locked page to the pagecache | |
930 | * @page: page to add | |
931 | * @mapping: the page's address_space | |
932 | * @offset: page index | |
933 | * @gfp_mask: page allocation mode | |
934 | * | |
935 | * This function is used to add a page to the pagecache. It must be locked. | |
936 | * This function does not add the page to the LRU. The caller must do that. | |
a862f68a MR |
937 | * |
938 | * Return: %0 on success, negative error code otherwise. | |
a528910e JW |
939 | */ |
940 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
941 | pgoff_t offset, gfp_t gfp_mask) | |
942 | { | |
9dd3d069 | 943 | return __filemap_add_folio(mapping, page_folio(page), offset, |
a528910e JW |
944 | gfp_mask, NULL); |
945 | } | |
e286781d | 946 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 | 947 | |
9dd3d069 MWO |
948 | int filemap_add_folio(struct address_space *mapping, struct folio *folio, |
949 | pgoff_t index, gfp_t gfp) | |
1da177e4 | 950 | { |
a528910e | 951 | void *shadow = NULL; |
4f98a2fe RR |
952 | int ret; |
953 | ||
9dd3d069 MWO |
954 | __folio_set_locked(folio); |
955 | ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow); | |
a528910e | 956 | if (unlikely(ret)) |
9dd3d069 | 957 | __folio_clear_locked(folio); |
a528910e JW |
958 | else { |
959 | /* | |
9dd3d069 | 960 | * The folio might have been evicted from cache only |
a528910e | 961 | * recently, in which case it should be activated like |
9dd3d069 MWO |
962 | * any other repeatedly accessed folio. |
963 | * The exception is folios getting rewritten; evicting other | |
f0281a00 RR |
964 | * data from the working set, only to cache data that will |
965 | * get overwritten with something else, is a waste of memory. | |
a528910e | 966 | */ |
9dd3d069 MWO |
967 | WARN_ON_ONCE(folio_test_active(folio)); |
968 | if (!(gfp & __GFP_WRITE) && shadow) | |
969 | workingset_refault(folio, shadow); | |
970 | folio_add_lru(folio); | |
a528910e | 971 | } |
1da177e4 LT |
972 | return ret; |
973 | } | |
9dd3d069 | 974 | EXPORT_SYMBOL_GPL(filemap_add_folio); |
1da177e4 | 975 | |
44110fe3 | 976 | #ifdef CONFIG_NUMA |
bb3c579e | 977 | struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order) |
44110fe3 | 978 | { |
c0ff7453 | 979 | int n; |
bb3c579e | 980 | struct folio *folio; |
c0ff7453 | 981 | |
44110fe3 | 982 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
983 | unsigned int cpuset_mems_cookie; |
984 | do { | |
d26914d1 | 985 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 986 | n = cpuset_mem_spread_node(); |
bb3c579e MWO |
987 | folio = __folio_alloc_node(gfp, order, n); |
988 | } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie)); | |
cc9a6c87 | 989 | |
bb3c579e | 990 | return folio; |
44110fe3 | 991 | } |
bb3c579e | 992 | return folio_alloc(gfp, order); |
44110fe3 | 993 | } |
bb3c579e | 994 | EXPORT_SYMBOL(filemap_alloc_folio); |
44110fe3 PJ |
995 | #endif |
996 | ||
7506ae6a JK |
997 | /* |
998 | * filemap_invalidate_lock_two - lock invalidate_lock for two mappings | |
999 | * | |
1000 | * Lock exclusively invalidate_lock of any passed mapping that is not NULL. | |
1001 | * | |
1002 | * @mapping1: the first mapping to lock | |
1003 | * @mapping2: the second mapping to lock | |
1004 | */ | |
1005 | void filemap_invalidate_lock_two(struct address_space *mapping1, | |
1006 | struct address_space *mapping2) | |
1007 | { | |
1008 | if (mapping1 > mapping2) | |
1009 | swap(mapping1, mapping2); | |
1010 | if (mapping1) | |
1011 | down_write(&mapping1->invalidate_lock); | |
1012 | if (mapping2 && mapping1 != mapping2) | |
1013 | down_write_nested(&mapping2->invalidate_lock, 1); | |
1014 | } | |
1015 | EXPORT_SYMBOL(filemap_invalidate_lock_two); | |
1016 | ||
1017 | /* | |
1018 | * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings | |
1019 | * | |
1020 | * Unlock exclusive invalidate_lock of any passed mapping that is not NULL. | |
1021 | * | |
1022 | * @mapping1: the first mapping to unlock | |
1023 | * @mapping2: the second mapping to unlock | |
1024 | */ | |
1025 | void filemap_invalidate_unlock_two(struct address_space *mapping1, | |
1026 | struct address_space *mapping2) | |
1027 | { | |
1028 | if (mapping1) | |
1029 | up_write(&mapping1->invalidate_lock); | |
1030 | if (mapping2 && mapping1 != mapping2) | |
1031 | up_write(&mapping2->invalidate_lock); | |
1032 | } | |
1033 | EXPORT_SYMBOL(filemap_invalidate_unlock_two); | |
1034 | ||
1da177e4 LT |
1035 | /* |
1036 | * In order to wait for pages to become available there must be | |
1037 | * waitqueues associated with pages. By using a hash table of | |
1038 | * waitqueues where the bucket discipline is to maintain all | |
1039 | * waiters on the same queue and wake all when any of the pages | |
1040 | * become available, and for the woken contexts to check to be | |
1041 | * sure the appropriate page became available, this saves space | |
1042 | * at a cost of "thundering herd" phenomena during rare hash | |
1043 | * collisions. | |
1044 | */ | |
62906027 NP |
1045 | #define PAGE_WAIT_TABLE_BITS 8 |
1046 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
df4d4f12 | 1047 | static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; |
62906027 | 1048 | |
df4d4f12 | 1049 | static wait_queue_head_t *folio_waitqueue(struct folio *folio) |
1da177e4 | 1050 | { |
df4d4f12 | 1051 | return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 1052 | } |
1da177e4 | 1053 | |
62906027 | 1054 | void __init pagecache_init(void) |
1da177e4 | 1055 | { |
62906027 | 1056 | int i; |
1da177e4 | 1057 | |
62906027 | 1058 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
df4d4f12 | 1059 | init_waitqueue_head(&folio_wait_table[i]); |
62906027 NP |
1060 | |
1061 | page_writeback_init(); | |
1da177e4 | 1062 | } |
1da177e4 | 1063 | |
5ef64cc8 LT |
1064 | /* |
1065 | * The page wait code treats the "wait->flags" somewhat unusually, because | |
5868ec26 | 1066 | * we have multiple different kinds of waits, not just the usual "exclusive" |
5ef64cc8 LT |
1067 | * one. |
1068 | * | |
1069 | * We have: | |
1070 | * | |
1071 | * (a) no special bits set: | |
1072 | * | |
1073 | * We're just waiting for the bit to be released, and when a waker | |
1074 | * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up, | |
1075 | * and remove it from the wait queue. | |
1076 | * | |
1077 | * Simple and straightforward. | |
1078 | * | |
1079 | * (b) WQ_FLAG_EXCLUSIVE: | |
1080 | * | |
1081 | * The waiter is waiting to get the lock, and only one waiter should | |
1082 | * be woken up to avoid any thundering herd behavior. We'll set the | |
1083 | * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue. | |
1084 | * | |
1085 | * This is the traditional exclusive wait. | |
1086 | * | |
5868ec26 | 1087 | * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: |
5ef64cc8 LT |
1088 | * |
1089 | * The waiter is waiting to get the bit, and additionally wants the | |
1090 | * lock to be transferred to it for fair lock behavior. If the lock | |
1091 | * cannot be taken, we stop walking the wait queue without waking | |
1092 | * the waiter. | |
1093 | * | |
1094 | * This is the "fair lock handoff" case, and in addition to setting | |
1095 | * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see | |
1096 | * that it now has the lock. | |
1097 | */ | |
ac6424b9 | 1098 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 1099 | { |
5ef64cc8 | 1100 | unsigned int flags; |
62906027 NP |
1101 | struct wait_page_key *key = arg; |
1102 | struct wait_page_queue *wait_page | |
1103 | = container_of(wait, struct wait_page_queue, wait); | |
1104 | ||
cdc8fcb4 | 1105 | if (!wake_page_match(wait_page, key)) |
62906027 | 1106 | return 0; |
3510ca20 | 1107 | |
9a1ea439 | 1108 | /* |
5ef64cc8 LT |
1109 | * If it's a lock handoff wait, we get the bit for it, and |
1110 | * stop walking (and do not wake it up) if we can't. | |
9a1ea439 | 1111 | */ |
5ef64cc8 LT |
1112 | flags = wait->flags; |
1113 | if (flags & WQ_FLAG_EXCLUSIVE) { | |
df4d4f12 | 1114 | if (test_bit(key->bit_nr, &key->folio->flags)) |
2a9127fc | 1115 | return -1; |
5ef64cc8 | 1116 | if (flags & WQ_FLAG_CUSTOM) { |
df4d4f12 | 1117 | if (test_and_set_bit(key->bit_nr, &key->folio->flags)) |
5ef64cc8 LT |
1118 | return -1; |
1119 | flags |= WQ_FLAG_DONE; | |
1120 | } | |
2a9127fc | 1121 | } |
f62e00cc | 1122 | |
5ef64cc8 LT |
1123 | /* |
1124 | * We are holding the wait-queue lock, but the waiter that | |
1125 | * is waiting for this will be checking the flags without | |
1126 | * any locking. | |
1127 | * | |
1128 | * So update the flags atomically, and wake up the waiter | |
1129 | * afterwards to avoid any races. This store-release pairs | |
101c0bf6 | 1130 | * with the load-acquire in folio_wait_bit_common(). |
5ef64cc8 LT |
1131 | */ |
1132 | smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN); | |
2a9127fc LT |
1133 | wake_up_state(wait->private, mode); |
1134 | ||
1135 | /* | |
1136 | * Ok, we have successfully done what we're waiting for, | |
1137 | * and we can unconditionally remove the wait entry. | |
1138 | * | |
5ef64cc8 LT |
1139 | * Note that this pairs with the "finish_wait()" in the |
1140 | * waiter, and has to be the absolute last thing we do. | |
1141 | * After this list_del_init(&wait->entry) the wait entry | |
2a9127fc LT |
1142 | * might be de-allocated and the process might even have |
1143 | * exited. | |
2a9127fc | 1144 | */ |
c6fe44d9 | 1145 | list_del_init_careful(&wait->entry); |
5ef64cc8 | 1146 | return (flags & WQ_FLAG_EXCLUSIVE) != 0; |
f62e00cc KM |
1147 | } |
1148 | ||
6974d7c9 | 1149 | static void folio_wake_bit(struct folio *folio, int bit_nr) |
cbbce822 | 1150 | { |
df4d4f12 | 1151 | wait_queue_head_t *q = folio_waitqueue(folio); |
62906027 NP |
1152 | struct wait_page_key key; |
1153 | unsigned long flags; | |
11a19c7b | 1154 | wait_queue_entry_t bookmark; |
cbbce822 | 1155 | |
df4d4f12 | 1156 | key.folio = folio; |
62906027 NP |
1157 | key.bit_nr = bit_nr; |
1158 | key.page_match = 0; | |
1159 | ||
11a19c7b TC |
1160 | bookmark.flags = 0; |
1161 | bookmark.private = NULL; | |
1162 | bookmark.func = NULL; | |
1163 | INIT_LIST_HEAD(&bookmark.entry); | |
1164 | ||
62906027 | 1165 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1166 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1167 | ||
1168 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1169 | /* | |
1170 | * Take a breather from holding the lock, | |
1171 | * allow pages that finish wake up asynchronously | |
1172 | * to acquire the lock and remove themselves | |
1173 | * from wait queue | |
1174 | */ | |
1175 | spin_unlock_irqrestore(&q->lock, flags); | |
1176 | cpu_relax(); | |
1177 | spin_lock_irqsave(&q->lock, flags); | |
1178 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1179 | } | |
1180 | ||
62906027 NP |
1181 | /* |
1182 | * It is possible for other pages to have collided on the waitqueue | |
1183 | * hash, so in that case check for a page match. That prevents a long- | |
1184 | * term waiter | |
1185 | * | |
1186 | * It is still possible to miss a case here, when we woke page waiters | |
1187 | * and removed them from the waitqueue, but there are still other | |
1188 | * page waiters. | |
1189 | */ | |
1190 | if (!waitqueue_active(q) || !key.page_match) { | |
6974d7c9 | 1191 | folio_clear_waiters(folio); |
62906027 NP |
1192 | /* |
1193 | * It's possible to miss clearing Waiters here, when we woke | |
1194 | * our page waiters, but the hashed waitqueue has waiters for | |
1195 | * other pages on it. | |
1196 | * | |
1197 | * That's okay, it's a rare case. The next waker will clear it. | |
1198 | */ | |
1199 | } | |
1200 | spin_unlock_irqrestore(&q->lock, flags); | |
1201 | } | |
74d81bfa | 1202 | |
4268b480 | 1203 | static void folio_wake(struct folio *folio, int bit) |
74d81bfa | 1204 | { |
4268b480 | 1205 | if (!folio_test_waiters(folio)) |
74d81bfa | 1206 | return; |
6974d7c9 | 1207 | folio_wake_bit(folio, bit); |
74d81bfa | 1208 | } |
62906027 | 1209 | |
9a1ea439 | 1210 | /* |
101c0bf6 | 1211 | * A choice of three behaviors for folio_wait_bit_common(): |
9a1ea439 HD |
1212 | */ |
1213 | enum behavior { | |
1214 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like | |
7c23c782 | 1215 | * __folio_lock() waiting on then setting PG_locked. |
9a1ea439 HD |
1216 | */ |
1217 | SHARED, /* Hold ref to page and check the bit when woken, like | |
9f2b04a2 | 1218 | * folio_wait_writeback() waiting on PG_writeback. |
9a1ea439 HD |
1219 | */ |
1220 | DROP, /* Drop ref to page before wait, no check when woken, | |
9f2b04a2 | 1221 | * like folio_put_wait_locked() on PG_locked. |
9a1ea439 HD |
1222 | */ |
1223 | }; | |
1224 | ||
2a9127fc | 1225 | /* |
101c0bf6 | 1226 | * Attempt to check (or get) the folio flag, and mark us done |
5ef64cc8 | 1227 | * if successful. |
2a9127fc | 1228 | */ |
101c0bf6 | 1229 | static inline bool folio_trylock_flag(struct folio *folio, int bit_nr, |
2a9127fc LT |
1230 | struct wait_queue_entry *wait) |
1231 | { | |
1232 | if (wait->flags & WQ_FLAG_EXCLUSIVE) { | |
101c0bf6 | 1233 | if (test_and_set_bit(bit_nr, &folio->flags)) |
2a9127fc | 1234 | return false; |
101c0bf6 | 1235 | } else if (test_bit(bit_nr, &folio->flags)) |
2a9127fc LT |
1236 | return false; |
1237 | ||
5ef64cc8 | 1238 | wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE; |
2a9127fc LT |
1239 | return true; |
1240 | } | |
1241 | ||
5ef64cc8 LT |
1242 | /* How many times do we accept lock stealing from under a waiter? */ |
1243 | int sysctl_page_lock_unfairness = 5; | |
1244 | ||
101c0bf6 MWO |
1245 | static inline int folio_wait_bit_common(struct folio *folio, int bit_nr, |
1246 | int state, enum behavior behavior) | |
62906027 | 1247 | { |
df4d4f12 | 1248 | wait_queue_head_t *q = folio_waitqueue(folio); |
5ef64cc8 | 1249 | int unfairness = sysctl_page_lock_unfairness; |
62906027 | 1250 | struct wait_page_queue wait_page; |
ac6424b9 | 1251 | wait_queue_entry_t *wait = &wait_page.wait; |
b1d29ba8 | 1252 | bool thrashing = false; |
9a1ea439 | 1253 | bool delayacct = false; |
eb414681 | 1254 | unsigned long pflags; |
62906027 | 1255 | |
eb414681 | 1256 | if (bit_nr == PG_locked && |
101c0bf6 MWO |
1257 | !folio_test_uptodate(folio) && folio_test_workingset(folio)) { |
1258 | if (!folio_test_swapbacked(folio)) { | |
eb414681 | 1259 | delayacct_thrashing_start(); |
9a1ea439 HD |
1260 | delayacct = true; |
1261 | } | |
eb414681 | 1262 | psi_memstall_enter(&pflags); |
b1d29ba8 JW |
1263 | thrashing = true; |
1264 | } | |
1265 | ||
62906027 NP |
1266 | init_wait(wait); |
1267 | wait->func = wake_page_function; | |
df4d4f12 | 1268 | wait_page.folio = folio; |
62906027 NP |
1269 | wait_page.bit_nr = bit_nr; |
1270 | ||
5ef64cc8 LT |
1271 | repeat: |
1272 | wait->flags = 0; | |
1273 | if (behavior == EXCLUSIVE) { | |
1274 | wait->flags = WQ_FLAG_EXCLUSIVE; | |
1275 | if (--unfairness < 0) | |
1276 | wait->flags |= WQ_FLAG_CUSTOM; | |
1277 | } | |
1278 | ||
2a9127fc LT |
1279 | /* |
1280 | * Do one last check whether we can get the | |
1281 | * page bit synchronously. | |
1282 | * | |
101c0bf6 | 1283 | * Do the folio_set_waiters() marking before that |
2a9127fc LT |
1284 | * to let any waker we _just_ missed know they |
1285 | * need to wake us up (otherwise they'll never | |
1286 | * even go to the slow case that looks at the | |
1287 | * page queue), and add ourselves to the wait | |
1288 | * queue if we need to sleep. | |
1289 | * | |
1290 | * This part needs to be done under the queue | |
1291 | * lock to avoid races. | |
1292 | */ | |
1293 | spin_lock_irq(&q->lock); | |
101c0bf6 MWO |
1294 | folio_set_waiters(folio); |
1295 | if (!folio_trylock_flag(folio, bit_nr, wait)) | |
2a9127fc LT |
1296 | __add_wait_queue_entry_tail(q, wait); |
1297 | spin_unlock_irq(&q->lock); | |
62906027 | 1298 | |
2a9127fc LT |
1299 | /* |
1300 | * From now on, all the logic will be based on | |
5ef64cc8 LT |
1301 | * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to |
1302 | * see whether the page bit testing has already | |
1303 | * been done by the wake function. | |
2a9127fc | 1304 | * |
101c0bf6 | 1305 | * We can drop our reference to the folio. |
2a9127fc LT |
1306 | */ |
1307 | if (behavior == DROP) | |
101c0bf6 | 1308 | folio_put(folio); |
62906027 | 1309 | |
5ef64cc8 LT |
1310 | /* |
1311 | * Note that until the "finish_wait()", or until | |
1312 | * we see the WQ_FLAG_WOKEN flag, we need to | |
1313 | * be very careful with the 'wait->flags', because | |
1314 | * we may race with a waker that sets them. | |
1315 | */ | |
2a9127fc | 1316 | for (;;) { |
5ef64cc8 LT |
1317 | unsigned int flags; |
1318 | ||
62906027 NP |
1319 | set_current_state(state); |
1320 | ||
5ef64cc8 LT |
1321 | /* Loop until we've been woken or interrupted */ |
1322 | flags = smp_load_acquire(&wait->flags); | |
1323 | if (!(flags & WQ_FLAG_WOKEN)) { | |
1324 | if (signal_pending_state(state, current)) | |
1325 | break; | |
1326 | ||
1327 | io_schedule(); | |
1328 | continue; | |
1329 | } | |
1330 | ||
1331 | /* If we were non-exclusive, we're done */ | |
1332 | if (behavior != EXCLUSIVE) | |
a8b169af | 1333 | break; |
9a1ea439 | 1334 | |
5ef64cc8 LT |
1335 | /* If the waker got the lock for us, we're done */ |
1336 | if (flags & WQ_FLAG_DONE) | |
9a1ea439 | 1337 | break; |
2a9127fc | 1338 | |
5ef64cc8 LT |
1339 | /* |
1340 | * Otherwise, if we're getting the lock, we need to | |
1341 | * try to get it ourselves. | |
1342 | * | |
1343 | * And if that fails, we'll have to retry this all. | |
1344 | */ | |
101c0bf6 | 1345 | if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0)))) |
5ef64cc8 LT |
1346 | goto repeat; |
1347 | ||
1348 | wait->flags |= WQ_FLAG_DONE; | |
1349 | break; | |
62906027 NP |
1350 | } |
1351 | ||
5ef64cc8 LT |
1352 | /* |
1353 | * If a signal happened, this 'finish_wait()' may remove the last | |
101c0bf6 | 1354 | * waiter from the wait-queues, but the folio waiters bit will remain |
5ef64cc8 LT |
1355 | * set. That's ok. The next wakeup will take care of it, and trying |
1356 | * to do it here would be difficult and prone to races. | |
1357 | */ | |
62906027 NP |
1358 | finish_wait(q, wait); |
1359 | ||
eb414681 | 1360 | if (thrashing) { |
9a1ea439 | 1361 | if (delayacct) |
eb414681 JW |
1362 | delayacct_thrashing_end(); |
1363 | psi_memstall_leave(&pflags); | |
1364 | } | |
b1d29ba8 | 1365 | |
62906027 | 1366 | /* |
5ef64cc8 LT |
1367 | * NOTE! The wait->flags weren't stable until we've done the |
1368 | * 'finish_wait()', and we could have exited the loop above due | |
1369 | * to a signal, and had a wakeup event happen after the signal | |
1370 | * test but before the 'finish_wait()'. | |
1371 | * | |
1372 | * So only after the finish_wait() can we reliably determine | |
1373 | * if we got woken up or not, so we can now figure out the final | |
1374 | * return value based on that state without races. | |
1375 | * | |
1376 | * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive | |
1377 | * waiter, but an exclusive one requires WQ_FLAG_DONE. | |
62906027 | 1378 | */ |
5ef64cc8 LT |
1379 | if (behavior == EXCLUSIVE) |
1380 | return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR; | |
62906027 | 1381 | |
2a9127fc | 1382 | return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR; |
62906027 NP |
1383 | } |
1384 | ||
ffa65753 AP |
1385 | #ifdef CONFIG_MIGRATION |
1386 | /** | |
1387 | * migration_entry_wait_on_locked - Wait for a migration entry to be removed | |
1388 | * @entry: migration swap entry. | |
1389 | * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required | |
1390 | * for pte entries, pass NULL for pmd entries. | |
1391 | * @ptl: already locked ptl. This function will drop the lock. | |
1392 | * | |
1393 | * Wait for a migration entry referencing the given page to be removed. This is | |
1394 | * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except | |
1395 | * this can be called without taking a reference on the page. Instead this | |
1396 | * should be called while holding the ptl for the migration entry referencing | |
1397 | * the page. | |
1398 | * | |
1399 | * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock(). | |
1400 | * | |
1401 | * This follows the same logic as folio_wait_bit_common() so see the comments | |
1402 | * there. | |
1403 | */ | |
1404 | void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep, | |
1405 | spinlock_t *ptl) | |
1406 | { | |
1407 | struct wait_page_queue wait_page; | |
1408 | wait_queue_entry_t *wait = &wait_page.wait; | |
1409 | bool thrashing = false; | |
1410 | bool delayacct = false; | |
1411 | unsigned long pflags; | |
1412 | wait_queue_head_t *q; | |
1413 | struct folio *folio = page_folio(pfn_swap_entry_to_page(entry)); | |
1414 | ||
1415 | q = folio_waitqueue(folio); | |
1416 | if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) { | |
1417 | if (!folio_test_swapbacked(folio)) { | |
1418 | delayacct_thrashing_start(); | |
1419 | delayacct = true; | |
1420 | } | |
1421 | psi_memstall_enter(&pflags); | |
1422 | thrashing = true; | |
1423 | } | |
1424 | ||
1425 | init_wait(wait); | |
1426 | wait->func = wake_page_function; | |
1427 | wait_page.folio = folio; | |
1428 | wait_page.bit_nr = PG_locked; | |
1429 | wait->flags = 0; | |
1430 | ||
1431 | spin_lock_irq(&q->lock); | |
1432 | folio_set_waiters(folio); | |
1433 | if (!folio_trylock_flag(folio, PG_locked, wait)) | |
1434 | __add_wait_queue_entry_tail(q, wait); | |
1435 | spin_unlock_irq(&q->lock); | |
1436 | ||
1437 | /* | |
1438 | * If a migration entry exists for the page the migration path must hold | |
1439 | * a valid reference to the page, and it must take the ptl to remove the | |
1440 | * migration entry. So the page is valid until the ptl is dropped. | |
1441 | */ | |
1442 | if (ptep) | |
1443 | pte_unmap_unlock(ptep, ptl); | |
1444 | else | |
1445 | spin_unlock(ptl); | |
1446 | ||
1447 | for (;;) { | |
1448 | unsigned int flags; | |
1449 | ||
1450 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1451 | ||
1452 | /* Loop until we've been woken or interrupted */ | |
1453 | flags = smp_load_acquire(&wait->flags); | |
1454 | if (!(flags & WQ_FLAG_WOKEN)) { | |
1455 | if (signal_pending_state(TASK_UNINTERRUPTIBLE, current)) | |
1456 | break; | |
1457 | ||
1458 | io_schedule(); | |
1459 | continue; | |
1460 | } | |
1461 | break; | |
1462 | } | |
1463 | ||
1464 | finish_wait(q, wait); | |
1465 | ||
1466 | if (thrashing) { | |
1467 | if (delayacct) | |
1468 | delayacct_thrashing_end(); | |
1469 | psi_memstall_leave(&pflags); | |
1470 | } | |
1471 | } | |
1472 | #endif | |
1473 | ||
101c0bf6 | 1474 | void folio_wait_bit(struct folio *folio, int bit_nr) |
62906027 | 1475 | { |
101c0bf6 | 1476 | folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
62906027 | 1477 | } |
101c0bf6 | 1478 | EXPORT_SYMBOL(folio_wait_bit); |
62906027 | 1479 | |
101c0bf6 | 1480 | int folio_wait_bit_killable(struct folio *folio, int bit_nr) |
62906027 | 1481 | { |
101c0bf6 | 1482 | return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED); |
cbbce822 | 1483 | } |
101c0bf6 | 1484 | EXPORT_SYMBOL(folio_wait_bit_killable); |
cbbce822 | 1485 | |
9a1ea439 | 1486 | /** |
9f2b04a2 MWO |
1487 | * folio_put_wait_locked - Drop a reference and wait for it to be unlocked |
1488 | * @folio: The folio to wait for. | |
48054625 | 1489 | * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc). |
9a1ea439 | 1490 | * |
9f2b04a2 | 1491 | * The caller should hold a reference on @folio. They expect the page to |
9a1ea439 | 1492 | * become unlocked relatively soon, but do not wish to hold up migration |
9f2b04a2 | 1493 | * (for example) by holding the reference while waiting for the folio to |
9a1ea439 | 1494 | * come unlocked. After this function returns, the caller should not |
9f2b04a2 | 1495 | * dereference @folio. |
48054625 | 1496 | * |
9f2b04a2 | 1497 | * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal. |
9a1ea439 | 1498 | */ |
9f2b04a2 | 1499 | int folio_put_wait_locked(struct folio *folio, int state) |
9a1ea439 | 1500 | { |
9f2b04a2 | 1501 | return folio_wait_bit_common(folio, PG_locked, state, DROP); |
9a1ea439 HD |
1502 | } |
1503 | ||
385e1ca5 | 1504 | /** |
df4d4f12 MWO |
1505 | * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue |
1506 | * @folio: Folio defining the wait queue of interest | |
697f619f | 1507 | * @waiter: Waiter to add to the queue |
385e1ca5 | 1508 | * |
df4d4f12 | 1509 | * Add an arbitrary @waiter to the wait queue for the nominated @folio. |
385e1ca5 | 1510 | */ |
df4d4f12 | 1511 | void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter) |
385e1ca5 | 1512 | { |
df4d4f12 | 1513 | wait_queue_head_t *q = folio_waitqueue(folio); |
385e1ca5 DH |
1514 | unsigned long flags; |
1515 | ||
1516 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1517 | __add_wait_queue_entry_tail(q, waiter); |
df4d4f12 | 1518 | folio_set_waiters(folio); |
385e1ca5 DH |
1519 | spin_unlock_irqrestore(&q->lock, flags); |
1520 | } | |
df4d4f12 | 1521 | EXPORT_SYMBOL_GPL(folio_add_wait_queue); |
385e1ca5 | 1522 | |
b91e1302 LT |
1523 | #ifndef clear_bit_unlock_is_negative_byte |
1524 | ||
1525 | /* | |
1526 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1527 | * | |
1528 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1529 | * test the sign bit at the same time. But if the architecture does | |
1530 | * not support that special operation, we just do this all by hand | |
1531 | * instead. | |
1532 | * | |
1533 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
ffceeb62 | 1534 | * being cleared, but a memory barrier should be unnecessary since it is |
b91e1302 LT |
1535 | * in the same byte as PG_locked. |
1536 | */ | |
1537 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1538 | { | |
1539 | clear_bit_unlock(nr, mem); | |
1540 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1541 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1542 | } |
1543 | ||
1544 | #endif | |
1545 | ||
1da177e4 | 1546 | /** |
4e136428 MWO |
1547 | * folio_unlock - Unlock a locked folio. |
1548 | * @folio: The folio. | |
1549 | * | |
1550 | * Unlocks the folio and wakes up any thread sleeping on the page lock. | |
1551 | * | |
1552 | * Context: May be called from interrupt or process context. May not be | |
1553 | * called from NMI context. | |
1da177e4 | 1554 | */ |
4e136428 | 1555 | void folio_unlock(struct folio *folio) |
1da177e4 | 1556 | { |
4e136428 | 1557 | /* Bit 7 allows x86 to check the byte's sign bit */ |
b91e1302 | 1558 | BUILD_BUG_ON(PG_waiters != 7); |
4e136428 MWO |
1559 | BUILD_BUG_ON(PG_locked > 7); |
1560 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); | |
1561 | if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0))) | |
6974d7c9 | 1562 | folio_wake_bit(folio, PG_locked); |
1da177e4 | 1563 | } |
4e136428 | 1564 | EXPORT_SYMBOL(folio_unlock); |
1da177e4 | 1565 | |
73e10ded | 1566 | /** |
b47393f8 MWO |
1567 | * folio_end_private_2 - Clear PG_private_2 and wake any waiters. |
1568 | * @folio: The folio. | |
73e10ded | 1569 | * |
b47393f8 MWO |
1570 | * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for |
1571 | * it. The folio reference held for PG_private_2 being set is released. | |
73e10ded | 1572 | * |
b47393f8 MWO |
1573 | * This is, for example, used when a netfs folio is being written to a local |
1574 | * disk cache, thereby allowing writes to the cache for the same folio to be | |
73e10ded DH |
1575 | * serialised. |
1576 | */ | |
b47393f8 | 1577 | void folio_end_private_2(struct folio *folio) |
73e10ded | 1578 | { |
6974d7c9 MWO |
1579 | VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio); |
1580 | clear_bit_unlock(PG_private_2, folio_flags(folio, 0)); | |
1581 | folio_wake_bit(folio, PG_private_2); | |
1582 | folio_put(folio); | |
73e10ded | 1583 | } |
b47393f8 | 1584 | EXPORT_SYMBOL(folio_end_private_2); |
73e10ded DH |
1585 | |
1586 | /** | |
b47393f8 MWO |
1587 | * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio. |
1588 | * @folio: The folio to wait on. | |
73e10ded | 1589 | * |
b47393f8 | 1590 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio. |
73e10ded | 1591 | */ |
b47393f8 | 1592 | void folio_wait_private_2(struct folio *folio) |
73e10ded | 1593 | { |
101c0bf6 MWO |
1594 | while (folio_test_private_2(folio)) |
1595 | folio_wait_bit(folio, PG_private_2); | |
73e10ded | 1596 | } |
b47393f8 | 1597 | EXPORT_SYMBOL(folio_wait_private_2); |
73e10ded DH |
1598 | |
1599 | /** | |
b47393f8 MWO |
1600 | * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio. |
1601 | * @folio: The folio to wait on. | |
73e10ded | 1602 | * |
b47393f8 | 1603 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a |
73e10ded DH |
1604 | * fatal signal is received by the calling task. |
1605 | * | |
1606 | * Return: | |
1607 | * - 0 if successful. | |
1608 | * - -EINTR if a fatal signal was encountered. | |
1609 | */ | |
b47393f8 | 1610 | int folio_wait_private_2_killable(struct folio *folio) |
73e10ded DH |
1611 | { |
1612 | int ret = 0; | |
1613 | ||
101c0bf6 MWO |
1614 | while (folio_test_private_2(folio)) { |
1615 | ret = folio_wait_bit_killable(folio, PG_private_2); | |
73e10ded DH |
1616 | if (ret < 0) |
1617 | break; | |
1618 | } | |
1619 | ||
1620 | return ret; | |
1621 | } | |
b47393f8 | 1622 | EXPORT_SYMBOL(folio_wait_private_2_killable); |
73e10ded | 1623 | |
485bb99b | 1624 | /** |
4268b480 MWO |
1625 | * folio_end_writeback - End writeback against a folio. |
1626 | * @folio: The folio. | |
1da177e4 | 1627 | */ |
4268b480 | 1628 | void folio_end_writeback(struct folio *folio) |
1da177e4 | 1629 | { |
888cf2db | 1630 | /* |
4268b480 MWO |
1631 | * folio_test_clear_reclaim() could be used here but it is an |
1632 | * atomic operation and overkill in this particular case. Failing | |
1633 | * to shuffle a folio marked for immediate reclaim is too mild | |
1634 | * a gain to justify taking an atomic operation penalty at the | |
1635 | * end of every folio writeback. | |
888cf2db | 1636 | */ |
4268b480 MWO |
1637 | if (folio_test_reclaim(folio)) { |
1638 | folio_clear_reclaim(folio); | |
575ced1c | 1639 | folio_rotate_reclaimable(folio); |
888cf2db | 1640 | } |
ac6aadb2 | 1641 | |
073861ed | 1642 | /* |
4268b480 | 1643 | * Writeback does not hold a folio reference of its own, relying |
073861ed | 1644 | * on truncation to wait for the clearing of PG_writeback. |
4268b480 MWO |
1645 | * But here we must make sure that the folio is not freed and |
1646 | * reused before the folio_wake(). | |
073861ed | 1647 | */ |
4268b480 | 1648 | folio_get(folio); |
269ccca3 | 1649 | if (!__folio_end_writeback(folio)) |
ac6aadb2 MS |
1650 | BUG(); |
1651 | ||
4e857c58 | 1652 | smp_mb__after_atomic(); |
4268b480 | 1653 | folio_wake(folio, PG_writeback); |
512b7931 | 1654 | acct_reclaim_writeback(folio); |
4268b480 | 1655 | folio_put(folio); |
1da177e4 | 1656 | } |
4268b480 | 1657 | EXPORT_SYMBOL(folio_end_writeback); |
1da177e4 | 1658 | |
57d99845 MW |
1659 | /* |
1660 | * After completing I/O on a page, call this routine to update the page | |
1661 | * flags appropriately | |
1662 | */ | |
c11f0c0b | 1663 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1664 | { |
c11f0c0b | 1665 | if (!is_write) { |
57d99845 MW |
1666 | if (!err) { |
1667 | SetPageUptodate(page); | |
1668 | } else { | |
1669 | ClearPageUptodate(page); | |
1670 | SetPageError(page); | |
1671 | } | |
1672 | unlock_page(page); | |
abf54548 | 1673 | } else { |
57d99845 | 1674 | if (err) { |
dd8416c4 MK |
1675 | struct address_space *mapping; |
1676 | ||
57d99845 | 1677 | SetPageError(page); |
dd8416c4 MK |
1678 | mapping = page_mapping(page); |
1679 | if (mapping) | |
1680 | mapping_set_error(mapping, err); | |
57d99845 MW |
1681 | } |
1682 | end_page_writeback(page); | |
1683 | } | |
1684 | } | |
1685 | EXPORT_SYMBOL_GPL(page_endio); | |
1686 | ||
485bb99b | 1687 | /** |
7c23c782 MWO |
1688 | * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it. |
1689 | * @folio: The folio to lock | |
1da177e4 | 1690 | */ |
7c23c782 | 1691 | void __folio_lock(struct folio *folio) |
1da177e4 | 1692 | { |
101c0bf6 | 1693 | folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE, |
9a1ea439 | 1694 | EXCLUSIVE); |
1da177e4 | 1695 | } |
7c23c782 | 1696 | EXPORT_SYMBOL(__folio_lock); |
1da177e4 | 1697 | |
af7f29d9 | 1698 | int __folio_lock_killable(struct folio *folio) |
2687a356 | 1699 | { |
101c0bf6 | 1700 | return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE, |
9a1ea439 | 1701 | EXCLUSIVE); |
2687a356 | 1702 | } |
af7f29d9 | 1703 | EXPORT_SYMBOL_GPL(__folio_lock_killable); |
2687a356 | 1704 | |
ffdc8dab | 1705 | static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait) |
dd3e6d50 | 1706 | { |
df4d4f12 | 1707 | struct wait_queue_head *q = folio_waitqueue(folio); |
f32b5dd7 MWO |
1708 | int ret = 0; |
1709 | ||
df4d4f12 | 1710 | wait->folio = folio; |
f32b5dd7 MWO |
1711 | wait->bit_nr = PG_locked; |
1712 | ||
1713 | spin_lock_irq(&q->lock); | |
1714 | __add_wait_queue_entry_tail(q, &wait->wait); | |
ffdc8dab MWO |
1715 | folio_set_waiters(folio); |
1716 | ret = !folio_trylock(folio); | |
f32b5dd7 MWO |
1717 | /* |
1718 | * If we were successful now, we know we're still on the | |
1719 | * waitqueue as we're still under the lock. This means it's | |
1720 | * safe to remove and return success, we know the callback | |
1721 | * isn't going to trigger. | |
1722 | */ | |
1723 | if (!ret) | |
1724 | __remove_wait_queue(q, &wait->wait); | |
1725 | else | |
1726 | ret = -EIOCBQUEUED; | |
1727 | spin_unlock_irq(&q->lock); | |
1728 | return ret; | |
dd3e6d50 JA |
1729 | } |
1730 | ||
9a95f3cf PC |
1731 | /* |
1732 | * Return values: | |
9138e47e MWO |
1733 | * true - folio is locked; mmap_lock is still held. |
1734 | * false - folio is not locked. | |
3e4e28c5 | 1735 | * mmap_lock has been released (mmap_read_unlock(), unless flags had both |
9a95f3cf | 1736 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
c1e8d7c6 | 1737 | * which case mmap_lock is still held. |
9a95f3cf | 1738 | * |
9138e47e MWO |
1739 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return true |
1740 | * with the folio locked and the mmap_lock unperturbed. | |
9a95f3cf | 1741 | */ |
9138e47e | 1742 | bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm, |
d065bd81 ML |
1743 | unsigned int flags) |
1744 | { | |
4064b982 | 1745 | if (fault_flag_allow_retry_first(flags)) { |
37b23e05 | 1746 | /* |
c1e8d7c6 | 1747 | * CAUTION! In this case, mmap_lock is not released |
37b23e05 KM |
1748 | * even though return 0. |
1749 | */ | |
1750 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
9138e47e | 1751 | return false; |
37b23e05 | 1752 | |
d8ed45c5 | 1753 | mmap_read_unlock(mm); |
37b23e05 | 1754 | if (flags & FAULT_FLAG_KILLABLE) |
6baa8d60 | 1755 | folio_wait_locked_killable(folio); |
37b23e05 | 1756 | else |
6baa8d60 | 1757 | folio_wait_locked(folio); |
9138e47e | 1758 | return false; |
800bca7c HL |
1759 | } |
1760 | if (flags & FAULT_FLAG_KILLABLE) { | |
9138e47e | 1761 | bool ret; |
37b23e05 | 1762 | |
af7f29d9 | 1763 | ret = __folio_lock_killable(folio); |
800bca7c HL |
1764 | if (ret) { |
1765 | mmap_read_unlock(mm); | |
9138e47e | 1766 | return false; |
800bca7c HL |
1767 | } |
1768 | } else { | |
af7f29d9 | 1769 | __folio_lock(folio); |
d065bd81 | 1770 | } |
800bca7c | 1771 | |
9138e47e | 1772 | return true; |
d065bd81 ML |
1773 | } |
1774 | ||
e7b563bb | 1775 | /** |
0d3f9296 MW |
1776 | * page_cache_next_miss() - Find the next gap in the page cache. |
1777 | * @mapping: Mapping. | |
1778 | * @index: Index. | |
1779 | * @max_scan: Maximum range to search. | |
e7b563bb | 1780 | * |
0d3f9296 MW |
1781 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1782 | * gap with the lowest index. | |
e7b563bb | 1783 | * |
0d3f9296 MW |
1784 | * This function may be called under the rcu_read_lock. However, this will |
1785 | * not atomically search a snapshot of the cache at a single point in time. | |
1786 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1787 | * created at index 10, page_cache_next_miss covering both indices may | |
1788 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1789 | * |
0d3f9296 MW |
1790 | * Return: The index of the gap if found, otherwise an index outside the |
1791 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1792 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1793 | */ |
0d3f9296 | 1794 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1795 | pgoff_t index, unsigned long max_scan) |
1796 | { | |
0d3f9296 | 1797 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1798 | |
0d3f9296 MW |
1799 | while (max_scan--) { |
1800 | void *entry = xas_next(&xas); | |
1801 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1802 | break; |
0d3f9296 | 1803 | if (xas.xa_index == 0) |
e7b563bb JW |
1804 | break; |
1805 | } | |
1806 | ||
0d3f9296 | 1807 | return xas.xa_index; |
e7b563bb | 1808 | } |
0d3f9296 | 1809 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1810 | |
1811 | /** | |
2346a560 | 1812 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
0d3f9296 MW |
1813 | * @mapping: Mapping. |
1814 | * @index: Index. | |
1815 | * @max_scan: Maximum range to search. | |
e7b563bb | 1816 | * |
0d3f9296 MW |
1817 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1818 | * gap with the highest index. | |
e7b563bb | 1819 | * |
0d3f9296 MW |
1820 | * This function may be called under the rcu_read_lock. However, this will |
1821 | * not atomically search a snapshot of the cache at a single point in time. | |
1822 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1823 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1824 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1825 | * |
0d3f9296 MW |
1826 | * Return: The index of the gap if found, otherwise an index outside the |
1827 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1828 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1829 | */ |
0d3f9296 | 1830 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1831 | pgoff_t index, unsigned long max_scan) |
1832 | { | |
0d3f9296 | 1833 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1834 | |
0d3f9296 MW |
1835 | while (max_scan--) { |
1836 | void *entry = xas_prev(&xas); | |
1837 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1838 | break; |
0d3f9296 | 1839 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1840 | break; |
1841 | } | |
1842 | ||
0d3f9296 | 1843 | return xas.xa_index; |
e7b563bb | 1844 | } |
0d3f9296 | 1845 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1846 | |
020853b6 MWO |
1847 | /* |
1848 | * Lockless page cache protocol: | |
1849 | * On the lookup side: | |
1850 | * 1. Load the folio from i_pages | |
1851 | * 2. Increment the refcount if it's not zero | |
1852 | * 3. If the folio is not found by xas_reload(), put the refcount and retry | |
1853 | * | |
1854 | * On the removal side: | |
1855 | * A. Freeze the page (by zeroing the refcount if nobody else has a reference) | |
1856 | * B. Remove the page from i_pages | |
1857 | * C. Return the page to the page allocator | |
1858 | * | |
1859 | * This means that any page may have its reference count temporarily | |
1860 | * increased by a speculative page cache (or fast GUP) lookup as it can | |
1861 | * be allocated by another user before the RCU grace period expires. | |
1862 | * Because the refcount temporarily acquired here may end up being the | |
1863 | * last refcount on the page, any page allocation must be freeable by | |
1864 | * folio_put(). | |
1865 | */ | |
1866 | ||
44835d20 | 1867 | /* |
bc5a3011 | 1868 | * mapping_get_entry - Get a page cache entry. |
485bb99b | 1869 | * @mapping: the address_space to search |
a6de4b48 | 1870 | * @index: The page cache index. |
0cd6144a | 1871 | * |
bca65eea MWO |
1872 | * Looks up the page cache entry at @mapping & @index. If it is a folio, |
1873 | * it is returned with an increased refcount. If it is a shadow entry | |
1874 | * of a previously evicted folio, or a swap entry from shmem/tmpfs, | |
1875 | * it is returned without further action. | |
485bb99b | 1876 | * |
bca65eea | 1877 | * Return: The folio, swap or shadow entry, %NULL if nothing is found. |
1da177e4 | 1878 | */ |
bca65eea | 1879 | static void *mapping_get_entry(struct address_space *mapping, pgoff_t index) |
1da177e4 | 1880 | { |
a6de4b48 | 1881 | XA_STATE(xas, &mapping->i_pages, index); |
bca65eea | 1882 | struct folio *folio; |
1da177e4 | 1883 | |
a60637c8 NP |
1884 | rcu_read_lock(); |
1885 | repeat: | |
4c7472c0 | 1886 | xas_reset(&xas); |
bca65eea MWO |
1887 | folio = xas_load(&xas); |
1888 | if (xas_retry(&xas, folio)) | |
4c7472c0 MW |
1889 | goto repeat; |
1890 | /* | |
1891 | * A shadow entry of a recently evicted page, or a swap entry from | |
1892 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1893 | */ | |
bca65eea | 1894 | if (!folio || xa_is_value(folio)) |
4c7472c0 | 1895 | goto out; |
83929372 | 1896 | |
bca65eea | 1897 | if (!folio_try_get_rcu(folio)) |
4c7472c0 | 1898 | goto repeat; |
83929372 | 1899 | |
bca65eea MWO |
1900 | if (unlikely(folio != xas_reload(&xas))) { |
1901 | folio_put(folio); | |
4c7472c0 | 1902 | goto repeat; |
a60637c8 | 1903 | } |
27d20fdd | 1904 | out: |
a60637c8 NP |
1905 | rcu_read_unlock(); |
1906 | ||
bca65eea | 1907 | return folio; |
1da177e4 | 1908 | } |
1da177e4 | 1909 | |
0cd6144a | 1910 | /** |
3f0c6a07 | 1911 | * __filemap_get_folio - Find and get a reference to a folio. |
2294b32e MWO |
1912 | * @mapping: The address_space to search. |
1913 | * @index: The page index. | |
3f0c6a07 MWO |
1914 | * @fgp_flags: %FGP flags modify how the folio is returned. |
1915 | * @gfp: Memory allocation flags to use if %FGP_CREAT is specified. | |
1da177e4 | 1916 | * |
2294b32e | 1917 | * Looks up the page cache entry at @mapping & @index. |
0cd6144a | 1918 | * |
2294b32e | 1919 | * @fgp_flags can be zero or more of these flags: |
0e056eb5 | 1920 | * |
3f0c6a07 MWO |
1921 | * * %FGP_ACCESSED - The folio will be marked accessed. |
1922 | * * %FGP_LOCK - The folio is returned locked. | |
44835d20 | 1923 | * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it |
3f0c6a07 | 1924 | * instead of allocating a new folio to replace it. |
2294b32e | 1925 | * * %FGP_CREAT - If no page is present then a new page is allocated using |
3f0c6a07 | 1926 | * @gfp and added to the page cache and the VM's LRU list. |
2294b32e MWO |
1927 | * The page is returned locked and with an increased refcount. |
1928 | * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the | |
1929 | * page is already in cache. If the page was allocated, unlock it before | |
1930 | * returning so the caller can do the same dance. | |
3f0c6a07 MWO |
1931 | * * %FGP_WRITE - The page will be written to by the caller. |
1932 | * * %FGP_NOFS - __GFP_FS will get cleared in gfp. | |
1933 | * * %FGP_NOWAIT - Don't get blocked by page lock. | |
b27652d9 | 1934 | * * %FGP_STABLE - Wait for the folio to be stable (finished writeback) |
1da177e4 | 1935 | * |
2294b32e MWO |
1936 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even |
1937 | * if the %GFP flags specified for %FGP_CREAT are atomic. | |
1da177e4 | 1938 | * |
2457aec6 | 1939 | * If there is a page cache page, it is returned with an increased refcount. |
a862f68a | 1940 | * |
3f0c6a07 | 1941 | * Return: The found folio or %NULL otherwise. |
1da177e4 | 1942 | */ |
3f0c6a07 MWO |
1943 | struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index, |
1944 | int fgp_flags, gfp_t gfp) | |
1da177e4 | 1945 | { |
3f0c6a07 | 1946 | struct folio *folio; |
2457aec6 | 1947 | |
1da177e4 | 1948 | repeat: |
3f0c6a07 MWO |
1949 | folio = mapping_get_entry(mapping, index); |
1950 | if (xa_is_value(folio)) { | |
44835d20 | 1951 | if (fgp_flags & FGP_ENTRY) |
3f0c6a07 MWO |
1952 | return folio; |
1953 | folio = NULL; | |
44835d20 | 1954 | } |
3f0c6a07 | 1955 | if (!folio) |
2457aec6 MG |
1956 | goto no_page; |
1957 | ||
1958 | if (fgp_flags & FGP_LOCK) { | |
1959 | if (fgp_flags & FGP_NOWAIT) { | |
3f0c6a07 MWO |
1960 | if (!folio_trylock(folio)) { |
1961 | folio_put(folio); | |
2457aec6 MG |
1962 | return NULL; |
1963 | } | |
1964 | } else { | |
3f0c6a07 | 1965 | folio_lock(folio); |
2457aec6 MG |
1966 | } |
1967 | ||
1968 | /* Has the page been truncated? */ | |
3f0c6a07 MWO |
1969 | if (unlikely(folio->mapping != mapping)) { |
1970 | folio_unlock(folio); | |
1971 | folio_put(folio); | |
2457aec6 MG |
1972 | goto repeat; |
1973 | } | |
3f0c6a07 | 1974 | VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); |
2457aec6 MG |
1975 | } |
1976 | ||
c16eb000 | 1977 | if (fgp_flags & FGP_ACCESSED) |
3f0c6a07 | 1978 | folio_mark_accessed(folio); |
b9306a79 YS |
1979 | else if (fgp_flags & FGP_WRITE) { |
1980 | /* Clear idle flag for buffer write */ | |
3f0c6a07 MWO |
1981 | if (folio_test_idle(folio)) |
1982 | folio_clear_idle(folio); | |
b9306a79 | 1983 | } |
2457aec6 | 1984 | |
b27652d9 MWO |
1985 | if (fgp_flags & FGP_STABLE) |
1986 | folio_wait_stable(folio); | |
2457aec6 | 1987 | no_page: |
3f0c6a07 | 1988 | if (!folio && (fgp_flags & FGP_CREAT)) { |
2457aec6 | 1989 | int err; |
f56753ac | 1990 | if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping)) |
3f0c6a07 | 1991 | gfp |= __GFP_WRITE; |
45f87de5 | 1992 | if (fgp_flags & FGP_NOFS) |
3f0c6a07 | 1993 | gfp &= ~__GFP_FS; |
2457aec6 | 1994 | |
3f0c6a07 MWO |
1995 | folio = filemap_alloc_folio(gfp, 0); |
1996 | if (!folio) | |
eb2be189 | 1997 | return NULL; |
2457aec6 | 1998 | |
a75d4c33 | 1999 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
2457aec6 MG |
2000 | fgp_flags |= FGP_LOCK; |
2001 | ||
eb39d618 | 2002 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 2003 | if (fgp_flags & FGP_ACCESSED) |
3f0c6a07 | 2004 | __folio_set_referenced(folio); |
2457aec6 | 2005 | |
3f0c6a07 | 2006 | err = filemap_add_folio(mapping, folio, index, gfp); |
eb2be189 | 2007 | if (unlikely(err)) { |
3f0c6a07 MWO |
2008 | folio_put(folio); |
2009 | folio = NULL; | |
eb2be189 NP |
2010 | if (err == -EEXIST) |
2011 | goto repeat; | |
1da177e4 | 2012 | } |
a75d4c33 JB |
2013 | |
2014 | /* | |
3f0c6a07 MWO |
2015 | * filemap_add_folio locks the page, and for mmap |
2016 | * we expect an unlocked page. | |
a75d4c33 | 2017 | */ |
3f0c6a07 MWO |
2018 | if (folio && (fgp_flags & FGP_FOR_MMAP)) |
2019 | folio_unlock(folio); | |
1da177e4 | 2020 | } |
2457aec6 | 2021 | |
3f0c6a07 | 2022 | return folio; |
1da177e4 | 2023 | } |
3f0c6a07 | 2024 | EXPORT_SYMBOL(__filemap_get_folio); |
1da177e4 | 2025 | |
f5e6429a | 2026 | static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max, |
c7bad633 MWO |
2027 | xa_mark_t mark) |
2028 | { | |
f5e6429a | 2029 | struct folio *folio; |
c7bad633 MWO |
2030 | |
2031 | retry: | |
2032 | if (mark == XA_PRESENT) | |
f5e6429a | 2033 | folio = xas_find(xas, max); |
c7bad633 | 2034 | else |
f5e6429a | 2035 | folio = xas_find_marked(xas, max, mark); |
c7bad633 | 2036 | |
f5e6429a | 2037 | if (xas_retry(xas, folio)) |
c7bad633 MWO |
2038 | goto retry; |
2039 | /* | |
2040 | * A shadow entry of a recently evicted page, a swap | |
2041 | * entry from shmem/tmpfs or a DAX entry. Return it | |
2042 | * without attempting to raise page count. | |
2043 | */ | |
f5e6429a MWO |
2044 | if (!folio || xa_is_value(folio)) |
2045 | return folio; | |
c7bad633 | 2046 | |
f5e6429a | 2047 | if (!folio_try_get_rcu(folio)) |
c7bad633 MWO |
2048 | goto reset; |
2049 | ||
f5e6429a MWO |
2050 | if (unlikely(folio != xas_reload(xas))) { |
2051 | folio_put(folio); | |
c7bad633 MWO |
2052 | goto reset; |
2053 | } | |
2054 | ||
f5e6429a | 2055 | return folio; |
c7bad633 MWO |
2056 | reset: |
2057 | xas_reset(xas); | |
2058 | goto retry; | |
2059 | } | |
2060 | ||
0cd6144a JW |
2061 | /** |
2062 | * find_get_entries - gang pagecache lookup | |
2063 | * @mapping: The address_space to search | |
2064 | * @start: The starting page cache index | |
ca122fe4 | 2065 | * @end: The final page index (inclusive). |
0e499ed3 | 2066 | * @fbatch: Where the resulting entries are placed. |
0cd6144a JW |
2067 | * @indices: The cache indices corresponding to the entries in @entries |
2068 | * | |
cf2039af | 2069 | * find_get_entries() will search for and return a batch of entries in |
0e499ed3 MWO |
2070 | * the mapping. The entries are placed in @fbatch. find_get_entries() |
2071 | * takes a reference on any actual folios it returns. | |
0cd6144a | 2072 | * |
0e499ed3 MWO |
2073 | * The entries have ascending indexes. The indices may not be consecutive |
2074 | * due to not-present entries or large folios. | |
0cd6144a | 2075 | * |
0e499ed3 | 2076 | * Any shadow entries of evicted folios, or swap entries from |
139b6a6f | 2077 | * shmem/tmpfs, are included in the returned array. |
0cd6144a | 2078 | * |
0e499ed3 | 2079 | * Return: The number of entries which were found. |
0cd6144a | 2080 | */ |
ca122fe4 | 2081 | unsigned find_get_entries(struct address_space *mapping, pgoff_t start, |
0e499ed3 | 2082 | pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices) |
0cd6144a | 2083 | { |
f280bf09 | 2084 | XA_STATE(xas, &mapping->i_pages, start); |
f5e6429a | 2085 | struct folio *folio; |
0cd6144a JW |
2086 | |
2087 | rcu_read_lock(); | |
f5e6429a | 2088 | while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) { |
0e499ed3 MWO |
2089 | indices[fbatch->nr] = xas.xa_index; |
2090 | if (!folio_batch_add(fbatch, folio)) | |
0cd6144a JW |
2091 | break; |
2092 | } | |
2093 | rcu_read_unlock(); | |
cf2039af | 2094 | |
0e499ed3 | 2095 | return folio_batch_count(fbatch); |
0cd6144a JW |
2096 | } |
2097 | ||
5c211ba2 MWO |
2098 | /** |
2099 | * find_lock_entries - Find a batch of pagecache entries. | |
2100 | * @mapping: The address_space to search. | |
2101 | * @start: The starting page cache index. | |
2102 | * @end: The final page index (inclusive). | |
51dcbdac MWO |
2103 | * @fbatch: Where the resulting entries are placed. |
2104 | * @indices: The cache indices of the entries in @fbatch. | |
5c211ba2 MWO |
2105 | * |
2106 | * find_lock_entries() will return a batch of entries from @mapping. | |
f5e6429a MWO |
2107 | * Swap, shadow and DAX entries are included. Folios are returned |
2108 | * locked and with an incremented refcount. Folios which are locked | |
2109 | * by somebody else or under writeback are skipped. Folios which are | |
2110 | * partially outside the range are not returned. | |
5c211ba2 MWO |
2111 | * |
2112 | * The entries have ascending indexes. The indices may not be consecutive | |
f5e6429a MWO |
2113 | * due to not-present entries, large folios, folios which could not be |
2114 | * locked or folios under writeback. | |
5c211ba2 MWO |
2115 | * |
2116 | * Return: The number of entries which were found. | |
2117 | */ | |
2118 | unsigned find_lock_entries(struct address_space *mapping, pgoff_t start, | |
51dcbdac | 2119 | pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices) |
5c211ba2 MWO |
2120 | { |
2121 | XA_STATE(xas, &mapping->i_pages, start); | |
f5e6429a | 2122 | struct folio *folio; |
5c211ba2 MWO |
2123 | |
2124 | rcu_read_lock(); | |
f5e6429a MWO |
2125 | while ((folio = find_get_entry(&xas, end, XA_PRESENT))) { |
2126 | if (!xa_is_value(folio)) { | |
2127 | if (folio->index < start) | |
5c211ba2 | 2128 | goto put; |
f5e6429a | 2129 | if (folio->index + folio_nr_pages(folio) - 1 > end) |
5c211ba2 | 2130 | goto put; |
f5e6429a | 2131 | if (!folio_trylock(folio)) |
5c211ba2 | 2132 | goto put; |
f5e6429a MWO |
2133 | if (folio->mapping != mapping || |
2134 | folio_test_writeback(folio)) | |
5c211ba2 | 2135 | goto unlock; |
f5e6429a MWO |
2136 | VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index), |
2137 | folio); | |
5c211ba2 | 2138 | } |
51dcbdac MWO |
2139 | indices[fbatch->nr] = xas.xa_index; |
2140 | if (!folio_batch_add(fbatch, folio)) | |
5c211ba2 | 2141 | break; |
6b24ca4a | 2142 | continue; |
5c211ba2 | 2143 | unlock: |
f5e6429a | 2144 | folio_unlock(folio); |
5c211ba2 | 2145 | put: |
f5e6429a | 2146 | folio_put(folio); |
5c211ba2 MWO |
2147 | } |
2148 | rcu_read_unlock(); | |
2149 | ||
51dcbdac | 2150 | return folio_batch_count(fbatch); |
5c211ba2 MWO |
2151 | } |
2152 | ||
6b24ca4a MWO |
2153 | static inline |
2154 | bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max) | |
2155 | { | |
2156 | if (!folio_test_large(folio) || folio_test_hugetlb(folio)) | |
2157 | return false; | |
2158 | if (index >= max) | |
2159 | return false; | |
2160 | return index < folio->index + folio_nr_pages(folio) - 1; | |
5c211ba2 MWO |
2161 | } |
2162 | ||
1da177e4 | 2163 | /** |
b947cee4 | 2164 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
2165 | * @mapping: The address_space to search |
2166 | * @start: The starting page index | |
b947cee4 | 2167 | * @end: The final page index (inclusive) |
1da177e4 LT |
2168 | * @nr_pages: The maximum number of pages |
2169 | * @pages: Where the resulting pages are placed | |
2170 | * | |
b947cee4 JK |
2171 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
2172 | * pages in the mapping starting at index @start and up to index @end | |
2173 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
2174 | * a reference against the returned pages. | |
1da177e4 LT |
2175 | * |
2176 | * The search returns a group of mapping-contiguous pages with ascending | |
2177 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 2178 | * We also update @start to index the next page for the traversal. |
1da177e4 | 2179 | * |
a862f68a MR |
2180 | * Return: the number of pages which were found. If this number is |
2181 | * smaller than @nr_pages, the end of specified range has been | |
b947cee4 | 2182 | * reached. |
1da177e4 | 2183 | */ |
b947cee4 JK |
2184 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
2185 | pgoff_t end, unsigned int nr_pages, | |
2186 | struct page **pages) | |
1da177e4 | 2187 | { |
fd1b3cee | 2188 | XA_STATE(xas, &mapping->i_pages, *start); |
f5e6429a | 2189 | struct folio *folio; |
0fc9d104 KK |
2190 | unsigned ret = 0; |
2191 | ||
2192 | if (unlikely(!nr_pages)) | |
2193 | return 0; | |
a60637c8 NP |
2194 | |
2195 | rcu_read_lock(); | |
f5e6429a | 2196 | while ((folio = find_get_entry(&xas, end, XA_PRESENT))) { |
fd1b3cee | 2197 | /* Skip over shadow, swap and DAX entries */ |
f5e6429a | 2198 | if (xa_is_value(folio)) |
8079b1c8 | 2199 | continue; |
a60637c8 | 2200 | |
6b24ca4a | 2201 | again: |
f5e6429a | 2202 | pages[ret] = folio_file_page(folio, xas.xa_index); |
b947cee4 | 2203 | if (++ret == nr_pages) { |
5d3ee42f | 2204 | *start = xas.xa_index + 1; |
b947cee4 JK |
2205 | goto out; |
2206 | } | |
6b24ca4a MWO |
2207 | if (folio_more_pages(folio, xas.xa_index, end)) { |
2208 | xas.xa_index++; | |
2209 | folio_ref_inc(folio); | |
2210 | goto again; | |
2211 | } | |
a60637c8 | 2212 | } |
5b280c0c | 2213 | |
b947cee4 JK |
2214 | /* |
2215 | * We come here when there is no page beyond @end. We take care to not | |
2216 | * overflow the index @start as it confuses some of the callers. This | |
fd1b3cee | 2217 | * breaks the iteration when there is a page at index -1 but that is |
b947cee4 JK |
2218 | * already broken anyway. |
2219 | */ | |
2220 | if (end == (pgoff_t)-1) | |
2221 | *start = (pgoff_t)-1; | |
2222 | else | |
2223 | *start = end + 1; | |
2224 | out: | |
a60637c8 | 2225 | rcu_read_unlock(); |
d72dc8a2 | 2226 | |
1da177e4 LT |
2227 | return ret; |
2228 | } | |
2229 | ||
ebf43500 JA |
2230 | /** |
2231 | * find_get_pages_contig - gang contiguous pagecache lookup | |
2232 | * @mapping: The address_space to search | |
2233 | * @index: The starting page index | |
2234 | * @nr_pages: The maximum number of pages | |
2235 | * @pages: Where the resulting pages are placed | |
2236 | * | |
2237 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
2238 | * that the returned number of pages are guaranteed to be contiguous. | |
2239 | * | |
a862f68a | 2240 | * Return: the number of pages which were found. |
ebf43500 JA |
2241 | */ |
2242 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
2243 | unsigned int nr_pages, struct page **pages) | |
2244 | { | |
3ece58a2 | 2245 | XA_STATE(xas, &mapping->i_pages, index); |
e1c37722 | 2246 | struct folio *folio; |
0fc9d104 KK |
2247 | unsigned int ret = 0; |
2248 | ||
2249 | if (unlikely(!nr_pages)) | |
2250 | return 0; | |
a60637c8 NP |
2251 | |
2252 | rcu_read_lock(); | |
e1c37722 MWO |
2253 | for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) { |
2254 | if (xas_retry(&xas, folio)) | |
3ece58a2 MW |
2255 | continue; |
2256 | /* | |
2257 | * If the entry has been swapped out, we can stop looking. | |
2258 | * No current caller is looking for DAX entries. | |
2259 | */ | |
e1c37722 | 2260 | if (xa_is_value(folio)) |
8079b1c8 | 2261 | break; |
ebf43500 | 2262 | |
e1c37722 | 2263 | if (!folio_try_get_rcu(folio)) |
3ece58a2 | 2264 | goto retry; |
83929372 | 2265 | |
e1c37722 | 2266 | if (unlikely(folio != xas_reload(&xas))) |
3ece58a2 | 2267 | goto put_page; |
a60637c8 | 2268 | |
6b24ca4a MWO |
2269 | again: |
2270 | pages[ret] = folio_file_page(folio, xas.xa_index); | |
0fc9d104 KK |
2271 | if (++ret == nr_pages) |
2272 | break; | |
6b24ca4a MWO |
2273 | if (folio_more_pages(folio, xas.xa_index, ULONG_MAX)) { |
2274 | xas.xa_index++; | |
2275 | folio_ref_inc(folio); | |
2276 | goto again; | |
2277 | } | |
3ece58a2 MW |
2278 | continue; |
2279 | put_page: | |
e1c37722 | 2280 | folio_put(folio); |
3ece58a2 MW |
2281 | retry: |
2282 | xas_reset(&xas); | |
ebf43500 | 2283 | } |
a60637c8 NP |
2284 | rcu_read_unlock(); |
2285 | return ret; | |
ebf43500 | 2286 | } |
ef71c15c | 2287 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 2288 | |
485bb99b | 2289 | /** |
c49f50d1 | 2290 | * find_get_pages_range_tag - Find and return head pages matching @tag. |
485bb99b RD |
2291 | * @mapping: the address_space to search |
2292 | * @index: the starting page index | |
72b045ae | 2293 | * @end: The final page index (inclusive) |
485bb99b RD |
2294 | * @tag: the tag index |
2295 | * @nr_pages: the maximum number of pages | |
2296 | * @pages: where the resulting pages are placed | |
2297 | * | |
c49f50d1 MWO |
2298 | * Like find_get_pages(), except we only return head pages which are tagged |
2299 | * with @tag. @index is updated to the index immediately after the last | |
2300 | * page we return, ready for the next iteration. | |
a862f68a MR |
2301 | * |
2302 | * Return: the number of pages which were found. | |
1da177e4 | 2303 | */ |
72b045ae | 2304 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
a6906972 | 2305 | pgoff_t end, xa_mark_t tag, unsigned int nr_pages, |
72b045ae | 2306 | struct page **pages) |
1da177e4 | 2307 | { |
a6906972 | 2308 | XA_STATE(xas, &mapping->i_pages, *index); |
f5e6429a | 2309 | struct folio *folio; |
0fc9d104 KK |
2310 | unsigned ret = 0; |
2311 | ||
2312 | if (unlikely(!nr_pages)) | |
2313 | return 0; | |
a60637c8 NP |
2314 | |
2315 | rcu_read_lock(); | |
f5e6429a | 2316 | while ((folio = find_get_entry(&xas, end, tag))) { |
a6906972 MW |
2317 | /* |
2318 | * Shadow entries should never be tagged, but this iteration | |
2319 | * is lockless so there is a window for page reclaim to evict | |
2320 | * a page we saw tagged. Skip over it. | |
2321 | */ | |
f5e6429a | 2322 | if (xa_is_value(folio)) |
139b6a6f | 2323 | continue; |
a60637c8 | 2324 | |
f5e6429a | 2325 | pages[ret] = &folio->page; |
72b045ae | 2326 | if (++ret == nr_pages) { |
f5e6429a | 2327 | *index = folio->index + folio_nr_pages(folio); |
72b045ae JK |
2328 | goto out; |
2329 | } | |
a60637c8 | 2330 | } |
5b280c0c | 2331 | |
72b045ae | 2332 | /* |
a6906972 | 2333 | * We come here when we got to @end. We take care to not overflow the |
72b045ae | 2334 | * index @index as it confuses some of the callers. This breaks the |
a6906972 MW |
2335 | * iteration when there is a page at index -1 but that is already |
2336 | * broken anyway. | |
72b045ae JK |
2337 | */ |
2338 | if (end == (pgoff_t)-1) | |
2339 | *index = (pgoff_t)-1; | |
2340 | else | |
2341 | *index = end + 1; | |
2342 | out: | |
a60637c8 | 2343 | rcu_read_unlock(); |
1da177e4 | 2344 | |
1da177e4 LT |
2345 | return ret; |
2346 | } | |
72b045ae | 2347 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 2348 | |
76d42bd9 WF |
2349 | /* |
2350 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
2351 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
2352 | * | |
2353 | * ---R__________________________________________B__________ | |
2354 | * ^ reading here ^ bad block(assume 4k) | |
2355 | * | |
2356 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
2357 | * => failing the whole request => read(R) => read(R+1) => | |
2358 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
2359 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
2360 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
2361 | * | |
2362 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
2363 | */ | |
0f8e2db4 | 2364 | static void shrink_readahead_size_eio(struct file_ra_state *ra) |
76d42bd9 | 2365 | { |
76d42bd9 | 2366 | ra->ra_pages /= 4; |
76d42bd9 WF |
2367 | } |
2368 | ||
cbd59c48 | 2369 | /* |
25d6a23e | 2370 | * filemap_get_read_batch - Get a batch of folios for read |
cbd59c48 | 2371 | * |
25d6a23e MWO |
2372 | * Get a batch of folios which represent a contiguous range of bytes in |
2373 | * the file. No exceptional entries will be returned. If @index is in | |
2374 | * the middle of a folio, the entire folio will be returned. The last | |
2375 | * folio in the batch may have the readahead flag set or the uptodate flag | |
2376 | * clear so that the caller can take the appropriate action. | |
cbd59c48 MWO |
2377 | */ |
2378 | static void filemap_get_read_batch(struct address_space *mapping, | |
25d6a23e | 2379 | pgoff_t index, pgoff_t max, struct folio_batch *fbatch) |
cbd59c48 MWO |
2380 | { |
2381 | XA_STATE(xas, &mapping->i_pages, index); | |
bdb72932 | 2382 | struct folio *folio; |
cbd59c48 MWO |
2383 | |
2384 | rcu_read_lock(); | |
bdb72932 MWO |
2385 | for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) { |
2386 | if (xas_retry(&xas, folio)) | |
cbd59c48 | 2387 | continue; |
bdb72932 | 2388 | if (xas.xa_index > max || xa_is_value(folio)) |
cbd59c48 | 2389 | break; |
bdb72932 | 2390 | if (!folio_try_get_rcu(folio)) |
cbd59c48 MWO |
2391 | goto retry; |
2392 | ||
bdb72932 | 2393 | if (unlikely(folio != xas_reload(&xas))) |
25d6a23e | 2394 | goto put_folio; |
cbd59c48 | 2395 | |
25d6a23e | 2396 | if (!folio_batch_add(fbatch, folio)) |
cbd59c48 | 2397 | break; |
bdb72932 | 2398 | if (!folio_test_uptodate(folio)) |
cbd59c48 | 2399 | break; |
bdb72932 | 2400 | if (folio_test_readahead(folio)) |
cbd59c48 | 2401 | break; |
6b24ca4a | 2402 | xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1); |
cbd59c48 | 2403 | continue; |
25d6a23e | 2404 | put_folio: |
bdb72932 | 2405 | folio_put(folio); |
cbd59c48 MWO |
2406 | retry: |
2407 | xas_reset(&xas); | |
2408 | } | |
2409 | rcu_read_unlock(); | |
2410 | } | |
2411 | ||
9d427b4e MWO |
2412 | static int filemap_read_folio(struct file *file, struct address_space *mapping, |
2413 | struct folio *folio) | |
723ef24b | 2414 | { |
723ef24b KO |
2415 | int error; |
2416 | ||
723ef24b | 2417 | /* |
68430303 MWO |
2418 | * A previous I/O error may have been due to temporary failures, |
2419 | * eg. multipath errors. PG_error will be set again if readpage | |
2420 | * fails. | |
723ef24b | 2421 | */ |
9d427b4e | 2422 | folio_clear_error(folio); |
723ef24b | 2423 | /* Start the actual read. The read will unlock the page. */ |
9d427b4e | 2424 | error = mapping->a_ops->readpage(file, &folio->page); |
68430303 MWO |
2425 | if (error) |
2426 | return error; | |
723ef24b | 2427 | |
9d427b4e | 2428 | error = folio_wait_locked_killable(folio); |
68430303 MWO |
2429 | if (error) |
2430 | return error; | |
9d427b4e | 2431 | if (folio_test_uptodate(folio)) |
aa1ec2f6 | 2432 | return 0; |
aa1ec2f6 MWO |
2433 | shrink_readahead_size_eio(&file->f_ra); |
2434 | return -EIO; | |
723ef24b KO |
2435 | } |
2436 | ||
fce70da3 | 2437 | static bool filemap_range_uptodate(struct address_space *mapping, |
2fa4eeb8 | 2438 | loff_t pos, struct iov_iter *iter, struct folio *folio) |
fce70da3 MWO |
2439 | { |
2440 | int count; | |
2441 | ||
2fa4eeb8 | 2442 | if (folio_test_uptodate(folio)) |
fce70da3 MWO |
2443 | return true; |
2444 | /* pipes can't handle partially uptodate pages */ | |
2445 | if (iov_iter_is_pipe(iter)) | |
2446 | return false; | |
2447 | if (!mapping->a_ops->is_partially_uptodate) | |
2448 | return false; | |
2fa4eeb8 | 2449 | if (mapping->host->i_blkbits >= folio_shift(folio)) |
fce70da3 MWO |
2450 | return false; |
2451 | ||
2452 | count = iter->count; | |
2fa4eeb8 MWO |
2453 | if (folio_pos(folio) > pos) { |
2454 | count -= folio_pos(folio) - pos; | |
fce70da3 MWO |
2455 | pos = 0; |
2456 | } else { | |
2fa4eeb8 | 2457 | pos -= folio_pos(folio); |
fce70da3 MWO |
2458 | } |
2459 | ||
2fa4eeb8 | 2460 | return mapping->a_ops->is_partially_uptodate(&folio->page, pos, count); |
fce70da3 MWO |
2461 | } |
2462 | ||
4612aeef MWO |
2463 | static int filemap_update_page(struct kiocb *iocb, |
2464 | struct address_space *mapping, struct iov_iter *iter, | |
65bca53b | 2465 | struct folio *folio) |
723ef24b | 2466 | { |
723ef24b KO |
2467 | int error; |
2468 | ||
730633f0 JK |
2469 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2470 | if (!filemap_invalidate_trylock_shared(mapping)) | |
2471 | return -EAGAIN; | |
2472 | } else { | |
2473 | filemap_invalidate_lock_shared(mapping); | |
2474 | } | |
2475 | ||
ffdc8dab | 2476 | if (!folio_trylock(folio)) { |
730633f0 | 2477 | error = -EAGAIN; |
87d1d7b6 | 2478 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) |
730633f0 | 2479 | goto unlock_mapping; |
87d1d7b6 | 2480 | if (!(iocb->ki_flags & IOCB_WAITQ)) { |
730633f0 | 2481 | filemap_invalidate_unlock_shared(mapping); |
9f2b04a2 MWO |
2482 | /* |
2483 | * This is where we usually end up waiting for a | |
2484 | * previously submitted readahead to finish. | |
2485 | */ | |
2486 | folio_put_wait_locked(folio, TASK_KILLABLE); | |
4612aeef | 2487 | return AOP_TRUNCATED_PAGE; |
bd8a1f36 | 2488 | } |
ffdc8dab | 2489 | error = __folio_lock_async(folio, iocb->ki_waitq); |
87d1d7b6 | 2490 | if (error) |
730633f0 | 2491 | goto unlock_mapping; |
723ef24b | 2492 | } |
723ef24b | 2493 | |
730633f0 | 2494 | error = AOP_TRUNCATED_PAGE; |
ffdc8dab | 2495 | if (!folio->mapping) |
730633f0 | 2496 | goto unlock; |
723ef24b | 2497 | |
fce70da3 | 2498 | error = 0; |
2fa4eeb8 | 2499 | if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, folio)) |
fce70da3 MWO |
2500 | goto unlock; |
2501 | ||
2502 | error = -EAGAIN; | |
2503 | if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ)) | |
2504 | goto unlock; | |
2505 | ||
9d427b4e | 2506 | error = filemap_read_folio(iocb->ki_filp, mapping, folio); |
730633f0 | 2507 | goto unlock_mapping; |
fce70da3 | 2508 | unlock: |
ffdc8dab | 2509 | folio_unlock(folio); |
730633f0 JK |
2510 | unlock_mapping: |
2511 | filemap_invalidate_unlock_shared(mapping); | |
2512 | if (error == AOP_TRUNCATED_PAGE) | |
ffdc8dab | 2513 | folio_put(folio); |
fce70da3 | 2514 | return error; |
723ef24b KO |
2515 | } |
2516 | ||
a5d4ad09 | 2517 | static int filemap_create_folio(struct file *file, |
f253e185 | 2518 | struct address_space *mapping, pgoff_t index, |
25d6a23e | 2519 | struct folio_batch *fbatch) |
723ef24b | 2520 | { |
a5d4ad09 | 2521 | struct folio *folio; |
723ef24b KO |
2522 | int error; |
2523 | ||
a5d4ad09 MWO |
2524 | folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0); |
2525 | if (!folio) | |
f253e185 | 2526 | return -ENOMEM; |
723ef24b | 2527 | |
730633f0 | 2528 | /* |
a5d4ad09 MWO |
2529 | * Protect against truncate / hole punch. Grabbing invalidate_lock |
2530 | * here assures we cannot instantiate and bring uptodate new | |
2531 | * pagecache folios after evicting page cache during truncate | |
2532 | * and before actually freeing blocks. Note that we could | |
2533 | * release invalidate_lock after inserting the folio into | |
2534 | * the page cache as the locked folio would then be enough to | |
2535 | * synchronize with hole punching. But there are code paths | |
2536 | * such as filemap_update_page() filling in partially uptodate | |
2537 | * pages or ->readpages() that need to hold invalidate_lock | |
2538 | * while mapping blocks for IO so let's hold the lock here as | |
2539 | * well to keep locking rules simple. | |
730633f0 JK |
2540 | */ |
2541 | filemap_invalidate_lock_shared(mapping); | |
a5d4ad09 | 2542 | error = filemap_add_folio(mapping, folio, index, |
f253e185 MWO |
2543 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
2544 | if (error == -EEXIST) | |
2545 | error = AOP_TRUNCATED_PAGE; | |
2546 | if (error) | |
2547 | goto error; | |
2548 | ||
a5d4ad09 | 2549 | error = filemap_read_folio(file, mapping, folio); |
f253e185 MWO |
2550 | if (error) |
2551 | goto error; | |
2552 | ||
730633f0 | 2553 | filemap_invalidate_unlock_shared(mapping); |
25d6a23e | 2554 | folio_batch_add(fbatch, folio); |
f253e185 MWO |
2555 | return 0; |
2556 | error: | |
730633f0 | 2557 | filemap_invalidate_unlock_shared(mapping); |
a5d4ad09 | 2558 | folio_put(folio); |
f253e185 | 2559 | return error; |
723ef24b KO |
2560 | } |
2561 | ||
5963fe03 | 2562 | static int filemap_readahead(struct kiocb *iocb, struct file *file, |
65bca53b | 2563 | struct address_space *mapping, struct folio *folio, |
5963fe03 MWO |
2564 | pgoff_t last_index) |
2565 | { | |
65bca53b MWO |
2566 | DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index); |
2567 | ||
5963fe03 MWO |
2568 | if (iocb->ki_flags & IOCB_NOIO) |
2569 | return -EAGAIN; | |
65bca53b | 2570 | page_cache_async_ra(&ractl, folio, last_index - folio->index); |
5963fe03 MWO |
2571 | return 0; |
2572 | } | |
2573 | ||
3a6bae48 | 2574 | static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter, |
25d6a23e | 2575 | struct folio_batch *fbatch) |
06c04442 KO |
2576 | { |
2577 | struct file *filp = iocb->ki_filp; | |
2578 | struct address_space *mapping = filp->f_mapping; | |
2579 | struct file_ra_state *ra = &filp->f_ra; | |
2580 | pgoff_t index = iocb->ki_pos >> PAGE_SHIFT; | |
cbd59c48 | 2581 | pgoff_t last_index; |
65bca53b | 2582 | struct folio *folio; |
cbd59c48 | 2583 | int err = 0; |
06c04442 | 2584 | |
cbd59c48 | 2585 | last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE); |
2642fca6 | 2586 | retry: |
06c04442 KO |
2587 | if (fatal_signal_pending(current)) |
2588 | return -EINTR; | |
2589 | ||
25d6a23e MWO |
2590 | filemap_get_read_batch(mapping, index, last_index, fbatch); |
2591 | if (!folio_batch_count(fbatch)) { | |
2642fca6 MWO |
2592 | if (iocb->ki_flags & IOCB_NOIO) |
2593 | return -EAGAIN; | |
2594 | page_cache_sync_readahead(mapping, ra, filp, index, | |
2595 | last_index - index); | |
25d6a23e | 2596 | filemap_get_read_batch(mapping, index, last_index, fbatch); |
2642fca6 | 2597 | } |
25d6a23e | 2598 | if (!folio_batch_count(fbatch)) { |
f253e185 MWO |
2599 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ)) |
2600 | return -EAGAIN; | |
a5d4ad09 | 2601 | err = filemap_create_folio(filp, mapping, |
25d6a23e | 2602 | iocb->ki_pos >> PAGE_SHIFT, fbatch); |
f253e185 | 2603 | if (err == AOP_TRUNCATED_PAGE) |
2642fca6 | 2604 | goto retry; |
f253e185 MWO |
2605 | return err; |
2606 | } | |
06c04442 | 2607 | |
25d6a23e | 2608 | folio = fbatch->folios[folio_batch_count(fbatch) - 1]; |
65bca53b MWO |
2609 | if (folio_test_readahead(folio)) { |
2610 | err = filemap_readahead(iocb, filp, mapping, folio, last_index); | |
2642fca6 MWO |
2611 | if (err) |
2612 | goto err; | |
2613 | } | |
65bca53b | 2614 | if (!folio_test_uptodate(folio)) { |
25d6a23e MWO |
2615 | if ((iocb->ki_flags & IOCB_WAITQ) && |
2616 | folio_batch_count(fbatch) > 1) | |
2642fca6 | 2617 | iocb->ki_flags |= IOCB_NOWAIT; |
65bca53b | 2618 | err = filemap_update_page(iocb, mapping, iter, folio); |
2642fca6 MWO |
2619 | if (err) |
2620 | goto err; | |
06c04442 KO |
2621 | } |
2622 | ||
2642fca6 | 2623 | return 0; |
cbd59c48 | 2624 | err: |
2642fca6 | 2625 | if (err < 0) |
65bca53b | 2626 | folio_put(folio); |
25d6a23e | 2627 | if (likely(--fbatch->nr)) |
ff993ba1 | 2628 | return 0; |
4612aeef | 2629 | if (err == AOP_TRUNCATED_PAGE) |
2642fca6 MWO |
2630 | goto retry; |
2631 | return err; | |
06c04442 KO |
2632 | } |
2633 | ||
485bb99b | 2634 | /** |
87fa0f3e CH |
2635 | * filemap_read - Read data from the page cache. |
2636 | * @iocb: The iocb to read. | |
2637 | * @iter: Destination for the data. | |
2638 | * @already_read: Number of bytes already read by the caller. | |
485bb99b | 2639 | * |
87fa0f3e CH |
2640 | * Copies data from the page cache. If the data is not currently present, |
2641 | * uses the readahead and readpage address_space operations to fetch it. | |
1da177e4 | 2642 | * |
87fa0f3e CH |
2643 | * Return: Total number of bytes copied, including those already read by |
2644 | * the caller. If an error happens before any bytes are copied, returns | |
2645 | * a negative error number. | |
1da177e4 | 2646 | */ |
87fa0f3e CH |
2647 | ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter, |
2648 | ssize_t already_read) | |
1da177e4 | 2649 | { |
47c27bc4 | 2650 | struct file *filp = iocb->ki_filp; |
06c04442 | 2651 | struct file_ra_state *ra = &filp->f_ra; |
36e78914 | 2652 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2653 | struct inode *inode = mapping->host; |
25d6a23e | 2654 | struct folio_batch fbatch; |
ff993ba1 | 2655 | int i, error = 0; |
06c04442 KO |
2656 | bool writably_mapped; |
2657 | loff_t isize, end_offset; | |
1da177e4 | 2658 | |
723ef24b | 2659 | if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2660 | return 0; |
3644e2d2 KO |
2661 | if (unlikely(!iov_iter_count(iter))) |
2662 | return 0; | |
2663 | ||
c2a9737f | 2664 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
25d6a23e | 2665 | folio_batch_init(&fbatch); |
c2a9737f | 2666 | |
06c04442 | 2667 | do { |
1da177e4 | 2668 | cond_resched(); |
5abf186a | 2669 | |
723ef24b | 2670 | /* |
06c04442 KO |
2671 | * If we've already successfully copied some data, then we |
2672 | * can no longer safely return -EIOCBQUEUED. Hence mark | |
2673 | * an async read NOWAIT at that point. | |
723ef24b | 2674 | */ |
87fa0f3e | 2675 | if ((iocb->ki_flags & IOCB_WAITQ) && already_read) |
723ef24b KO |
2676 | iocb->ki_flags |= IOCB_NOWAIT; |
2677 | ||
8c8387ee DH |
2678 | if (unlikely(iocb->ki_pos >= i_size_read(inode))) |
2679 | break; | |
2680 | ||
25d6a23e | 2681 | error = filemap_get_pages(iocb, iter, &fbatch); |
ff993ba1 | 2682 | if (error < 0) |
06c04442 | 2683 | break; |
1da177e4 | 2684 | |
06c04442 KO |
2685 | /* |
2686 | * i_size must be checked after we know the pages are Uptodate. | |
2687 | * | |
2688 | * Checking i_size after the check allows us to calculate | |
2689 | * the correct value for "nr", which means the zero-filled | |
2690 | * part of the page is not copied back to userspace (unless | |
2691 | * another truncate extends the file - this is desired though). | |
2692 | */ | |
2693 | isize = i_size_read(inode); | |
2694 | if (unlikely(iocb->ki_pos >= isize)) | |
25d6a23e | 2695 | goto put_folios; |
06c04442 KO |
2696 | end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count); |
2697 | ||
06c04442 KO |
2698 | /* |
2699 | * Once we start copying data, we don't want to be touching any | |
2700 | * cachelines that might be contended: | |
2701 | */ | |
2702 | writably_mapped = mapping_writably_mapped(mapping); | |
2703 | ||
2704 | /* | |
2705 | * When a sequential read accesses a page several times, only | |
2706 | * mark it as accessed the first time. | |
2707 | */ | |
2708 | if (iocb->ki_pos >> PAGE_SHIFT != | |
2709 | ra->prev_pos >> PAGE_SHIFT) | |
25d6a23e | 2710 | folio_mark_accessed(fbatch.folios[0]); |
06c04442 | 2711 | |
25d6a23e MWO |
2712 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
2713 | struct folio *folio = fbatch.folios[i]; | |
d996fc7f MWO |
2714 | size_t fsize = folio_size(folio); |
2715 | size_t offset = iocb->ki_pos & (fsize - 1); | |
cbd59c48 | 2716 | size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos, |
d996fc7f | 2717 | fsize - offset); |
cbd59c48 | 2718 | size_t copied; |
06c04442 | 2719 | |
d996fc7f | 2720 | if (end_offset < folio_pos(folio)) |
cbd59c48 MWO |
2721 | break; |
2722 | if (i > 0) | |
d996fc7f | 2723 | folio_mark_accessed(folio); |
06c04442 | 2724 | /* |
d996fc7f MWO |
2725 | * If users can be writing to this folio using arbitrary |
2726 | * virtual addresses, take care of potential aliasing | |
2727 | * before reading the folio on the kernel side. | |
06c04442 | 2728 | */ |
d996fc7f MWO |
2729 | if (writably_mapped) |
2730 | flush_dcache_folio(folio); | |
06c04442 | 2731 | |
d996fc7f | 2732 | copied = copy_folio_to_iter(folio, offset, bytes, iter); |
06c04442 | 2733 | |
87fa0f3e | 2734 | already_read += copied; |
06c04442 KO |
2735 | iocb->ki_pos += copied; |
2736 | ra->prev_pos = iocb->ki_pos; | |
2737 | ||
2738 | if (copied < bytes) { | |
2739 | error = -EFAULT; | |
2740 | break; | |
2741 | } | |
1da177e4 | 2742 | } |
25d6a23e MWO |
2743 | put_folios: |
2744 | for (i = 0; i < folio_batch_count(&fbatch); i++) | |
2745 | folio_put(fbatch.folios[i]); | |
2746 | folio_batch_init(&fbatch); | |
06c04442 | 2747 | } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error); |
1da177e4 | 2748 | |
0c6aa263 | 2749 | file_accessed(filp); |
06c04442 | 2750 | |
87fa0f3e | 2751 | return already_read ? already_read : error; |
1da177e4 | 2752 | } |
87fa0f3e | 2753 | EXPORT_SYMBOL_GPL(filemap_read); |
1da177e4 | 2754 | |
485bb99b | 2755 | /** |
6abd2322 | 2756 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2757 | * @iocb: kernel I/O control block |
6abd2322 | 2758 | * @iter: destination for the data read |
485bb99b | 2759 | * |
6abd2322 | 2760 | * This is the "read_iter()" routine for all filesystems |
1da177e4 | 2761 | * that can use the page cache directly. |
41da51bc AG |
2762 | * |
2763 | * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall | |
2764 | * be returned when no data can be read without waiting for I/O requests | |
2765 | * to complete; it doesn't prevent readahead. | |
2766 | * | |
2767 | * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O | |
2768 | * requests shall be made for the read or for readahead. When no data | |
2769 | * can be read, -EAGAIN shall be returned. When readahead would be | |
2770 | * triggered, a partial, possibly empty read shall be returned. | |
2771 | * | |
a862f68a MR |
2772 | * Return: |
2773 | * * number of bytes copied, even for partial reads | |
41da51bc | 2774 | * * negative error code (or 0 if IOCB_NOIO) if nothing was read |
1da177e4 LT |
2775 | */ |
2776 | ssize_t | |
ed978a81 | 2777 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2778 | { |
e7080a43 | 2779 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2780 | ssize_t retval = 0; |
e7080a43 NS |
2781 | |
2782 | if (!count) | |
826ea860 | 2783 | return 0; /* skip atime */ |
1da177e4 | 2784 | |
2ba48ce5 | 2785 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2786 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2787 | struct address_space *mapping = file->f_mapping; |
2788 | struct inode *inode = mapping->host; | |
1da177e4 | 2789 | |
6be96d3a | 2790 | if (iocb->ki_flags & IOCB_NOWAIT) { |
7a60d6d7 JA |
2791 | if (filemap_range_needs_writeback(mapping, iocb->ki_pos, |
2792 | iocb->ki_pos + count - 1)) | |
6be96d3a GR |
2793 | return -EAGAIN; |
2794 | } else { | |
2795 | retval = filemap_write_and_wait_range(mapping, | |
2796 | iocb->ki_pos, | |
2797 | iocb->ki_pos + count - 1); | |
2798 | if (retval < 0) | |
826ea860 | 2799 | return retval; |
6be96d3a | 2800 | } |
d8d3d94b | 2801 | |
0d5b0cf2 CH |
2802 | file_accessed(file); |
2803 | ||
5ecda137 | 2804 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2805 | if (retval >= 0) { |
c64fb5c7 | 2806 | iocb->ki_pos += retval; |
5ecda137 | 2807 | count -= retval; |
9fe55eea | 2808 | } |
ab2125df PB |
2809 | if (retval != -EIOCBQUEUED) |
2810 | iov_iter_revert(iter, count - iov_iter_count(iter)); | |
66f998f6 | 2811 | |
9fe55eea SW |
2812 | /* |
2813 | * Btrfs can have a short DIO read if we encounter | |
2814 | * compressed extents, so if there was an error, or if | |
2815 | * we've already read everything we wanted to, or if | |
2816 | * there was a short read because we hit EOF, go ahead | |
2817 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2818 | * the rest of the read. Buffered reads will not work for |
2819 | * DAX files, so don't bother trying. | |
9fe55eea | 2820 | */ |
61d0017e JA |
2821 | if (retval < 0 || !count || IS_DAX(inode)) |
2822 | return retval; | |
2823 | if (iocb->ki_pos >= i_size_read(inode)) | |
826ea860 | 2824 | return retval; |
1da177e4 LT |
2825 | } |
2826 | ||
826ea860 | 2827 | return filemap_read(iocb, iter, retval); |
1da177e4 | 2828 | } |
ed978a81 | 2829 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2830 | |
f5e6429a MWO |
2831 | static inline loff_t folio_seek_hole_data(struct xa_state *xas, |
2832 | struct address_space *mapping, struct folio *folio, | |
54fa39ac | 2833 | loff_t start, loff_t end, bool seek_data) |
41139aa4 | 2834 | { |
54fa39ac MWO |
2835 | const struct address_space_operations *ops = mapping->a_ops; |
2836 | size_t offset, bsz = i_blocksize(mapping->host); | |
2837 | ||
f5e6429a | 2838 | if (xa_is_value(folio) || folio_test_uptodate(folio)) |
54fa39ac MWO |
2839 | return seek_data ? start : end; |
2840 | if (!ops->is_partially_uptodate) | |
2841 | return seek_data ? end : start; | |
2842 | ||
2843 | xas_pause(xas); | |
2844 | rcu_read_unlock(); | |
f5e6429a MWO |
2845 | folio_lock(folio); |
2846 | if (unlikely(folio->mapping != mapping)) | |
54fa39ac MWO |
2847 | goto unlock; |
2848 | ||
f5e6429a | 2849 | offset = offset_in_folio(folio, start) & ~(bsz - 1); |
54fa39ac MWO |
2850 | |
2851 | do { | |
f5e6429a MWO |
2852 | if (ops->is_partially_uptodate(&folio->page, offset, bsz) == |
2853 | seek_data) | |
54fa39ac MWO |
2854 | break; |
2855 | start = (start + bsz) & ~(bsz - 1); | |
2856 | offset += bsz; | |
f5e6429a | 2857 | } while (offset < folio_size(folio)); |
54fa39ac | 2858 | unlock: |
f5e6429a | 2859 | folio_unlock(folio); |
54fa39ac MWO |
2860 | rcu_read_lock(); |
2861 | return start; | |
41139aa4 MWO |
2862 | } |
2863 | ||
f5e6429a | 2864 | static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio) |
41139aa4 | 2865 | { |
f5e6429a | 2866 | if (xa_is_value(folio)) |
41139aa4 | 2867 | return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index); |
f5e6429a | 2868 | return folio_size(folio); |
41139aa4 MWO |
2869 | } |
2870 | ||
2871 | /** | |
2872 | * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache. | |
2873 | * @mapping: Address space to search. | |
2874 | * @start: First byte to consider. | |
2875 | * @end: Limit of search (exclusive). | |
2876 | * @whence: Either SEEK_HOLE or SEEK_DATA. | |
2877 | * | |
2878 | * If the page cache knows which blocks contain holes and which blocks | |
2879 | * contain data, your filesystem can use this function to implement | |
2880 | * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are | |
2881 | * entirely memory-based such as tmpfs, and filesystems which support | |
2882 | * unwritten extents. | |
2883 | * | |
f0953a1b | 2884 | * Return: The requested offset on success, or -ENXIO if @whence specifies |
41139aa4 MWO |
2885 | * SEEK_DATA and there is no data after @start. There is an implicit hole |
2886 | * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start | |
2887 | * and @end contain data. | |
2888 | */ | |
2889 | loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start, | |
2890 | loff_t end, int whence) | |
2891 | { | |
2892 | XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT); | |
ed98b015 | 2893 | pgoff_t max = (end - 1) >> PAGE_SHIFT; |
41139aa4 | 2894 | bool seek_data = (whence == SEEK_DATA); |
f5e6429a | 2895 | struct folio *folio; |
41139aa4 MWO |
2896 | |
2897 | if (end <= start) | |
2898 | return -ENXIO; | |
2899 | ||
2900 | rcu_read_lock(); | |
f5e6429a | 2901 | while ((folio = find_get_entry(&xas, max, XA_PRESENT))) { |
ed98b015 | 2902 | loff_t pos = (u64)xas.xa_index << PAGE_SHIFT; |
f5e6429a | 2903 | size_t seek_size; |
41139aa4 MWO |
2904 | |
2905 | if (start < pos) { | |
2906 | if (!seek_data) | |
2907 | goto unlock; | |
2908 | start = pos; | |
2909 | } | |
2910 | ||
f5e6429a MWO |
2911 | seek_size = seek_folio_size(&xas, folio); |
2912 | pos = round_up((u64)pos + 1, seek_size); | |
2913 | start = folio_seek_hole_data(&xas, mapping, folio, start, pos, | |
54fa39ac MWO |
2914 | seek_data); |
2915 | if (start < pos) | |
41139aa4 | 2916 | goto unlock; |
ed98b015 HD |
2917 | if (start >= end) |
2918 | break; | |
2919 | if (seek_size > PAGE_SIZE) | |
2920 | xas_set(&xas, pos >> PAGE_SHIFT); | |
f5e6429a MWO |
2921 | if (!xa_is_value(folio)) |
2922 | folio_put(folio); | |
41139aa4 | 2923 | } |
41139aa4 | 2924 | if (seek_data) |
ed98b015 | 2925 | start = -ENXIO; |
41139aa4 MWO |
2926 | unlock: |
2927 | rcu_read_unlock(); | |
f5e6429a MWO |
2928 | if (folio && !xa_is_value(folio)) |
2929 | folio_put(folio); | |
41139aa4 MWO |
2930 | if (start > end) |
2931 | return end; | |
2932 | return start; | |
2933 | } | |
2934 | ||
1da177e4 | 2935 | #ifdef CONFIG_MMU |
1da177e4 | 2936 | #define MMAP_LOTSAMISS (100) |
6b4c9f44 | 2937 | /* |
e292e6d6 | 2938 | * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock |
6b4c9f44 | 2939 | * @vmf - the vm_fault for this fault. |
e292e6d6 | 2940 | * @folio - the folio to lock. |
6b4c9f44 JB |
2941 | * @fpin - the pointer to the file we may pin (or is already pinned). |
2942 | * | |
e292e6d6 MWO |
2943 | * This works similar to lock_folio_or_retry in that it can drop the |
2944 | * mmap_lock. It differs in that it actually returns the folio locked | |
2945 | * if it returns 1 and 0 if it couldn't lock the folio. If we did have | |
2946 | * to drop the mmap_lock then fpin will point to the pinned file and | |
2947 | * needs to be fput()'ed at a later point. | |
6b4c9f44 | 2948 | */ |
e292e6d6 | 2949 | static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio, |
6b4c9f44 JB |
2950 | struct file **fpin) |
2951 | { | |
7c23c782 | 2952 | if (folio_trylock(folio)) |
6b4c9f44 JB |
2953 | return 1; |
2954 | ||
8b0f9fa2 LT |
2955 | /* |
2956 | * NOTE! This will make us return with VM_FAULT_RETRY, but with | |
c1e8d7c6 | 2957 | * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT |
8b0f9fa2 LT |
2958 | * is supposed to work. We have way too many special cases.. |
2959 | */ | |
6b4c9f44 JB |
2960 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
2961 | return 0; | |
2962 | ||
2963 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); | |
2964 | if (vmf->flags & FAULT_FLAG_KILLABLE) { | |
af7f29d9 | 2965 | if (__folio_lock_killable(folio)) { |
6b4c9f44 | 2966 | /* |
c1e8d7c6 | 2967 | * We didn't have the right flags to drop the mmap_lock, |
6b4c9f44 JB |
2968 | * but all fault_handlers only check for fatal signals |
2969 | * if we return VM_FAULT_RETRY, so we need to drop the | |
c1e8d7c6 | 2970 | * mmap_lock here and return 0 if we don't have a fpin. |
6b4c9f44 JB |
2971 | */ |
2972 | if (*fpin == NULL) | |
d8ed45c5 | 2973 | mmap_read_unlock(vmf->vma->vm_mm); |
6b4c9f44 JB |
2974 | return 0; |
2975 | } | |
2976 | } else | |
7c23c782 MWO |
2977 | __folio_lock(folio); |
2978 | ||
6b4c9f44 JB |
2979 | return 1; |
2980 | } | |
2981 | ||
ef00e08e | 2982 | /* |
6b4c9f44 JB |
2983 | * Synchronous readahead happens when we don't even find a page in the page |
2984 | * cache at all. We don't want to perform IO under the mmap sem, so if we have | |
2985 | * to drop the mmap sem we return the file that was pinned in order for us to do | |
2986 | * that. If we didn't pin a file then we return NULL. The file that is | |
2987 | * returned needs to be fput()'ed when we're done with it. | |
ef00e08e | 2988 | */ |
6b4c9f44 | 2989 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
ef00e08e | 2990 | { |
2a1180f1 JB |
2991 | struct file *file = vmf->vma->vm_file; |
2992 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2993 | struct address_space *mapping = file->f_mapping; |
fcd9ae4f | 2994 | DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff); |
6b4c9f44 | 2995 | struct file *fpin = NULL; |
e630bfac | 2996 | unsigned int mmap_miss; |
ef00e08e LT |
2997 | |
2998 | /* If we don't want any read-ahead, don't bother */ | |
2a1180f1 | 2999 | if (vmf->vma->vm_flags & VM_RAND_READ) |
6b4c9f44 | 3000 | return fpin; |
275b12bf | 3001 | if (!ra->ra_pages) |
6b4c9f44 | 3002 | return fpin; |
ef00e08e | 3003 | |
2a1180f1 | 3004 | if (vmf->vma->vm_flags & VM_SEQ_READ) { |
6b4c9f44 | 3005 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
fcd9ae4f | 3006 | page_cache_sync_ra(&ractl, ra->ra_pages); |
6b4c9f44 | 3007 | return fpin; |
ef00e08e LT |
3008 | } |
3009 | ||
207d04ba | 3010 | /* Avoid banging the cache line if not needed */ |
e630bfac KS |
3011 | mmap_miss = READ_ONCE(ra->mmap_miss); |
3012 | if (mmap_miss < MMAP_LOTSAMISS * 10) | |
3013 | WRITE_ONCE(ra->mmap_miss, ++mmap_miss); | |
ef00e08e LT |
3014 | |
3015 | /* | |
3016 | * Do we miss much more than hit in this file? If so, | |
3017 | * stop bothering with read-ahead. It will only hurt. | |
3018 | */ | |
e630bfac | 3019 | if (mmap_miss > MMAP_LOTSAMISS) |
6b4c9f44 | 3020 | return fpin; |
ef00e08e | 3021 | |
d30a1100 WF |
3022 | /* |
3023 | * mmap read-around | |
3024 | */ | |
6b4c9f44 | 3025 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
db660d46 | 3026 | ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2); |
600e19af RG |
3027 | ra->size = ra->ra_pages; |
3028 | ra->async_size = ra->ra_pages / 4; | |
db660d46 DH |
3029 | ractl._index = ra->start; |
3030 | do_page_cache_ra(&ractl, ra->size, ra->async_size); | |
6b4c9f44 | 3031 | return fpin; |
ef00e08e LT |
3032 | } |
3033 | ||
3034 | /* | |
3035 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
6b4c9f44 | 3036 | * so we want to possibly extend the readahead further. We return the file that |
c1e8d7c6 | 3037 | * was pinned if we have to drop the mmap_lock in order to do IO. |
ef00e08e | 3038 | */ |
6b4c9f44 | 3039 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
79598ced | 3040 | struct folio *folio) |
ef00e08e | 3041 | { |
2a1180f1 JB |
3042 | struct file *file = vmf->vma->vm_file; |
3043 | struct file_ra_state *ra = &file->f_ra; | |
79598ced | 3044 | DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff); |
6b4c9f44 | 3045 | struct file *fpin = NULL; |
e630bfac | 3046 | unsigned int mmap_miss; |
ef00e08e LT |
3047 | |
3048 | /* If we don't want any read-ahead, don't bother */ | |
5c72feee | 3049 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
6b4c9f44 | 3050 | return fpin; |
79598ced | 3051 | |
e630bfac KS |
3052 | mmap_miss = READ_ONCE(ra->mmap_miss); |
3053 | if (mmap_miss) | |
3054 | WRITE_ONCE(ra->mmap_miss, --mmap_miss); | |
79598ced MWO |
3055 | |
3056 | if (folio_test_readahead(folio)) { | |
6b4c9f44 | 3057 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
79598ced | 3058 | page_cache_async_ra(&ractl, folio, ra->ra_pages); |
6b4c9f44 JB |
3059 | } |
3060 | return fpin; | |
ef00e08e LT |
3061 | } |
3062 | ||
485bb99b | 3063 | /** |
54cb8821 | 3064 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 3065 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 3066 | * |
54cb8821 | 3067 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
3068 | * mapped memory region to read in file data during a page fault. |
3069 | * | |
3070 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
3071 | * it in the page cache, and handles the special cases reasonably without | |
3072 | * having a lot of duplicated code. | |
9a95f3cf | 3073 | * |
c1e8d7c6 | 3074 | * vma->vm_mm->mmap_lock must be held on entry. |
9a95f3cf | 3075 | * |
c1e8d7c6 | 3076 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock |
e292e6d6 | 3077 | * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap(). |
9a95f3cf | 3078 | * |
c1e8d7c6 | 3079 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock |
9a95f3cf PC |
3080 | * has not been released. |
3081 | * | |
3082 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
a862f68a MR |
3083 | * |
3084 | * Return: bitwise-OR of %VM_FAULT_ codes. | |
1da177e4 | 3085 | */ |
2bcd6454 | 3086 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
3087 | { |
3088 | int error; | |
11bac800 | 3089 | struct file *file = vmf->vma->vm_file; |
6b4c9f44 | 3090 | struct file *fpin = NULL; |
1da177e4 | 3091 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3092 | struct inode *inode = mapping->host; |
e292e6d6 MWO |
3093 | pgoff_t max_idx, index = vmf->pgoff; |
3094 | struct folio *folio; | |
2bcd6454 | 3095 | vm_fault_t ret = 0; |
730633f0 | 3096 | bool mapping_locked = false; |
1da177e4 | 3097 | |
e292e6d6 MWO |
3098 | max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3099 | if (unlikely(index >= max_idx)) | |
5307cc1a | 3100 | return VM_FAULT_SIGBUS; |
1da177e4 | 3101 | |
1da177e4 | 3102 | /* |
49426420 | 3103 | * Do we have something in the page cache already? |
1da177e4 | 3104 | */ |
e292e6d6 MWO |
3105 | folio = filemap_get_folio(mapping, index); |
3106 | if (likely(folio)) { | |
1da177e4 | 3107 | /* |
730633f0 JK |
3108 | * We found the page, so try async readahead before waiting for |
3109 | * the lock. | |
1da177e4 | 3110 | */ |
730633f0 | 3111 | if (!(vmf->flags & FAULT_FLAG_TRIED)) |
79598ced | 3112 | fpin = do_async_mmap_readahead(vmf, folio); |
e292e6d6 | 3113 | if (unlikely(!folio_test_uptodate(folio))) { |
730633f0 JK |
3114 | filemap_invalidate_lock_shared(mapping); |
3115 | mapping_locked = true; | |
3116 | } | |
3117 | } else { | |
ef00e08e | 3118 | /* No page in the page cache at all */ |
ef00e08e | 3119 | count_vm_event(PGMAJFAULT); |
2262185c | 3120 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e | 3121 | ret = VM_FAULT_MAJOR; |
6b4c9f44 | 3122 | fpin = do_sync_mmap_readahead(vmf); |
ef00e08e | 3123 | retry_find: |
730633f0 | 3124 | /* |
e292e6d6 | 3125 | * See comment in filemap_create_folio() why we need |
730633f0 JK |
3126 | * invalidate_lock |
3127 | */ | |
3128 | if (!mapping_locked) { | |
3129 | filemap_invalidate_lock_shared(mapping); | |
3130 | mapping_locked = true; | |
3131 | } | |
e292e6d6 | 3132 | folio = __filemap_get_folio(mapping, index, |
a75d4c33 JB |
3133 | FGP_CREAT|FGP_FOR_MMAP, |
3134 | vmf->gfp_mask); | |
e292e6d6 | 3135 | if (!folio) { |
6b4c9f44 JB |
3136 | if (fpin) |
3137 | goto out_retry; | |
730633f0 | 3138 | filemap_invalidate_unlock_shared(mapping); |
e520e932 | 3139 | return VM_FAULT_OOM; |
6b4c9f44 | 3140 | } |
1da177e4 LT |
3141 | } |
3142 | ||
e292e6d6 | 3143 | if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin)) |
6b4c9f44 | 3144 | goto out_retry; |
b522c94d ML |
3145 | |
3146 | /* Did it get truncated? */ | |
e292e6d6 MWO |
3147 | if (unlikely(folio->mapping != mapping)) { |
3148 | folio_unlock(folio); | |
3149 | folio_put(folio); | |
b522c94d ML |
3150 | goto retry_find; |
3151 | } | |
e292e6d6 | 3152 | VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); |
b522c94d | 3153 | |
1da177e4 | 3154 | /* |
d00806b1 NP |
3155 | * We have a locked page in the page cache, now we need to check |
3156 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 3157 | */ |
e292e6d6 | 3158 | if (unlikely(!folio_test_uptodate(folio))) { |
730633f0 JK |
3159 | /* |
3160 | * The page was in cache and uptodate and now it is not. | |
3161 | * Strange but possible since we didn't hold the page lock all | |
3162 | * the time. Let's drop everything get the invalidate lock and | |
3163 | * try again. | |
3164 | */ | |
3165 | if (!mapping_locked) { | |
e292e6d6 MWO |
3166 | folio_unlock(folio); |
3167 | folio_put(folio); | |
730633f0 JK |
3168 | goto retry_find; |
3169 | } | |
1da177e4 | 3170 | goto page_not_uptodate; |
730633f0 | 3171 | } |
1da177e4 | 3172 | |
6b4c9f44 | 3173 | /* |
c1e8d7c6 | 3174 | * We've made it this far and we had to drop our mmap_lock, now is the |
6b4c9f44 JB |
3175 | * time to return to the upper layer and have it re-find the vma and |
3176 | * redo the fault. | |
3177 | */ | |
3178 | if (fpin) { | |
e292e6d6 | 3179 | folio_unlock(folio); |
6b4c9f44 JB |
3180 | goto out_retry; |
3181 | } | |
730633f0 JK |
3182 | if (mapping_locked) |
3183 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 | 3184 | |
ef00e08e LT |
3185 | /* |
3186 | * Found the page and have a reference on it. | |
3187 | * We must recheck i_size under page lock. | |
3188 | */ | |
e292e6d6 MWO |
3189 | max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3190 | if (unlikely(index >= max_idx)) { | |
3191 | folio_unlock(folio); | |
3192 | folio_put(folio); | |
5307cc1a | 3193 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
3194 | } |
3195 | ||
e292e6d6 | 3196 | vmf->page = folio_file_page(folio, index); |
83c54070 | 3197 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 3198 | |
1da177e4 | 3199 | page_not_uptodate: |
1da177e4 LT |
3200 | /* |
3201 | * Umm, take care of errors if the page isn't up-to-date. | |
3202 | * Try to re-read it _once_. We do this synchronously, | |
3203 | * because there really aren't any performance issues here | |
3204 | * and we need to check for errors. | |
3205 | */ | |
6b4c9f44 | 3206 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
e292e6d6 | 3207 | error = filemap_read_folio(file, mapping, folio); |
6b4c9f44 JB |
3208 | if (fpin) |
3209 | goto out_retry; | |
e292e6d6 | 3210 | folio_put(folio); |
d00806b1 NP |
3211 | |
3212 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 3213 | goto retry_find; |
730633f0 | 3214 | filemap_invalidate_unlock_shared(mapping); |
1da177e4 | 3215 | |
d0217ac0 | 3216 | return VM_FAULT_SIGBUS; |
6b4c9f44 JB |
3217 | |
3218 | out_retry: | |
3219 | /* | |
c1e8d7c6 | 3220 | * We dropped the mmap_lock, we need to return to the fault handler to |
6b4c9f44 JB |
3221 | * re-find the vma and come back and find our hopefully still populated |
3222 | * page. | |
3223 | */ | |
e292e6d6 MWO |
3224 | if (folio) |
3225 | folio_put(folio); | |
730633f0 JK |
3226 | if (mapping_locked) |
3227 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 JB |
3228 | if (fpin) |
3229 | fput(fpin); | |
3230 | return ret | VM_FAULT_RETRY; | |
54cb8821 NP |
3231 | } |
3232 | EXPORT_SYMBOL(filemap_fault); | |
3233 | ||
f9ce0be7 | 3234 | static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page) |
f1820361 | 3235 | { |
f9ce0be7 KS |
3236 | struct mm_struct *mm = vmf->vma->vm_mm; |
3237 | ||
3238 | /* Huge page is mapped? No need to proceed. */ | |
3239 | if (pmd_trans_huge(*vmf->pmd)) { | |
3240 | unlock_page(page); | |
3241 | put_page(page); | |
3242 | return true; | |
3243 | } | |
3244 | ||
3245 | if (pmd_none(*vmf->pmd) && PageTransHuge(page)) { | |
e0f43fa5 YS |
3246 | vm_fault_t ret = do_set_pmd(vmf, page); |
3247 | if (!ret) { | |
3248 | /* The page is mapped successfully, reference consumed. */ | |
3249 | unlock_page(page); | |
3250 | return true; | |
f9ce0be7 | 3251 | } |
f9ce0be7 KS |
3252 | } |
3253 | ||
03c4f204 QZ |
3254 | if (pmd_none(*vmf->pmd)) |
3255 | pmd_install(mm, vmf->pmd, &vmf->prealloc_pte); | |
f9ce0be7 KS |
3256 | |
3257 | /* See comment in handle_pte_fault() */ | |
3258 | if (pmd_devmap_trans_unstable(vmf->pmd)) { | |
3259 | unlock_page(page); | |
3260 | put_page(page); | |
3261 | return true; | |
3262 | } | |
3263 | ||
3264 | return false; | |
3265 | } | |
3266 | ||
820b05e9 | 3267 | static struct folio *next_uptodate_page(struct folio *folio, |
f9ce0be7 KS |
3268 | struct address_space *mapping, |
3269 | struct xa_state *xas, pgoff_t end_pgoff) | |
3270 | { | |
3271 | unsigned long max_idx; | |
3272 | ||
3273 | do { | |
9184a307 | 3274 | if (!folio) |
f9ce0be7 | 3275 | return NULL; |
9184a307 | 3276 | if (xas_retry(xas, folio)) |
f9ce0be7 | 3277 | continue; |
9184a307 | 3278 | if (xa_is_value(folio)) |
f9ce0be7 | 3279 | continue; |
9184a307 | 3280 | if (folio_test_locked(folio)) |
f9ce0be7 | 3281 | continue; |
9184a307 | 3282 | if (!folio_try_get_rcu(folio)) |
f9ce0be7 KS |
3283 | continue; |
3284 | /* Has the page moved or been split? */ | |
9184a307 | 3285 | if (unlikely(folio != xas_reload(xas))) |
f9ce0be7 | 3286 | goto skip; |
9184a307 | 3287 | if (!folio_test_uptodate(folio) || folio_test_readahead(folio)) |
f9ce0be7 | 3288 | goto skip; |
9184a307 | 3289 | if (!folio_trylock(folio)) |
f9ce0be7 | 3290 | goto skip; |
9184a307 | 3291 | if (folio->mapping != mapping) |
f9ce0be7 | 3292 | goto unlock; |
9184a307 | 3293 | if (!folio_test_uptodate(folio)) |
f9ce0be7 KS |
3294 | goto unlock; |
3295 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); | |
3296 | if (xas->xa_index >= max_idx) | |
3297 | goto unlock; | |
820b05e9 | 3298 | return folio; |
f9ce0be7 | 3299 | unlock: |
9184a307 | 3300 | folio_unlock(folio); |
f9ce0be7 | 3301 | skip: |
9184a307 MWO |
3302 | folio_put(folio); |
3303 | } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL); | |
f9ce0be7 KS |
3304 | |
3305 | return NULL; | |
3306 | } | |
3307 | ||
820b05e9 | 3308 | static inline struct folio *first_map_page(struct address_space *mapping, |
f9ce0be7 KS |
3309 | struct xa_state *xas, |
3310 | pgoff_t end_pgoff) | |
3311 | { | |
3312 | return next_uptodate_page(xas_find(xas, end_pgoff), | |
3313 | mapping, xas, end_pgoff); | |
3314 | } | |
3315 | ||
820b05e9 | 3316 | static inline struct folio *next_map_page(struct address_space *mapping, |
f9ce0be7 KS |
3317 | struct xa_state *xas, |
3318 | pgoff_t end_pgoff) | |
3319 | { | |
3320 | return next_uptodate_page(xas_next_entry(xas, end_pgoff), | |
3321 | mapping, xas, end_pgoff); | |
3322 | } | |
3323 | ||
3324 | vm_fault_t filemap_map_pages(struct vm_fault *vmf, | |
3325 | pgoff_t start_pgoff, pgoff_t end_pgoff) | |
3326 | { | |
3327 | struct vm_area_struct *vma = vmf->vma; | |
3328 | struct file *file = vma->vm_file; | |
f1820361 | 3329 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 3330 | pgoff_t last_pgoff = start_pgoff; |
9d3af4b4 | 3331 | unsigned long addr; |
070e807c | 3332 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
820b05e9 MWO |
3333 | struct folio *folio; |
3334 | struct page *page; | |
e630bfac | 3335 | unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss); |
f9ce0be7 | 3336 | vm_fault_t ret = 0; |
f1820361 KS |
3337 | |
3338 | rcu_read_lock(); | |
820b05e9 MWO |
3339 | folio = first_map_page(mapping, &xas, end_pgoff); |
3340 | if (!folio) | |
f9ce0be7 | 3341 | goto out; |
f1820361 | 3342 | |
820b05e9 | 3343 | if (filemap_map_pmd(vmf, &folio->page)) { |
f9ce0be7 KS |
3344 | ret = VM_FAULT_NOPAGE; |
3345 | goto out; | |
3346 | } | |
f1820361 | 3347 | |
9d3af4b4 WD |
3348 | addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
3349 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); | |
f9ce0be7 | 3350 | do { |
6b24ca4a | 3351 | again: |
820b05e9 | 3352 | page = folio_file_page(folio, xas.xa_index); |
f9ce0be7 | 3353 | if (PageHWPoison(page)) |
f1820361 KS |
3354 | goto unlock; |
3355 | ||
e630bfac KS |
3356 | if (mmap_miss > 0) |
3357 | mmap_miss--; | |
7267ec00 | 3358 | |
9d3af4b4 | 3359 | addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
f9ce0be7 | 3360 | vmf->pte += xas.xa_index - last_pgoff; |
070e807c | 3361 | last_pgoff = xas.xa_index; |
f9ce0be7 KS |
3362 | |
3363 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3364 | goto unlock; |
f9ce0be7 | 3365 | |
46bdb427 | 3366 | /* We're about to handle the fault */ |
9d3af4b4 | 3367 | if (vmf->address == addr) |
46bdb427 | 3368 | ret = VM_FAULT_NOPAGE; |
46bdb427 | 3369 | |
9d3af4b4 | 3370 | do_set_pte(vmf, page, addr); |
f9ce0be7 | 3371 | /* no need to invalidate: a not-present page won't be cached */ |
9d3af4b4 | 3372 | update_mmu_cache(vma, addr, vmf->pte); |
6b24ca4a MWO |
3373 | if (folio_more_pages(folio, xas.xa_index, end_pgoff)) { |
3374 | xas.xa_index++; | |
3375 | folio_ref_inc(folio); | |
3376 | goto again; | |
3377 | } | |
820b05e9 | 3378 | folio_unlock(folio); |
f9ce0be7 | 3379 | continue; |
f1820361 | 3380 | unlock: |
6b24ca4a MWO |
3381 | if (folio_more_pages(folio, xas.xa_index, end_pgoff)) { |
3382 | xas.xa_index++; | |
3383 | goto again; | |
3384 | } | |
820b05e9 MWO |
3385 | folio_unlock(folio); |
3386 | folio_put(folio); | |
3387 | } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL); | |
f9ce0be7 KS |
3388 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3389 | out: | |
f1820361 | 3390 | rcu_read_unlock(); |
e630bfac | 3391 | WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss); |
f9ce0be7 | 3392 | return ret; |
f1820361 KS |
3393 | } |
3394 | EXPORT_SYMBOL(filemap_map_pages); | |
3395 | ||
2bcd6454 | 3396 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 | 3397 | { |
5df1a672 | 3398 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
960ea971 | 3399 | struct folio *folio = page_folio(vmf->page); |
2bcd6454 | 3400 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 3401 | |
5df1a672 | 3402 | sb_start_pagefault(mapping->host->i_sb); |
11bac800 | 3403 | file_update_time(vmf->vma->vm_file); |
960ea971 MWO |
3404 | folio_lock(folio); |
3405 | if (folio->mapping != mapping) { | |
3406 | folio_unlock(folio); | |
4fcf1c62 JK |
3407 | ret = VM_FAULT_NOPAGE; |
3408 | goto out; | |
3409 | } | |
14da9200 | 3410 | /* |
960ea971 | 3411 | * We mark the folio dirty already here so that when freeze is in |
14da9200 | 3412 | * progress, we are guaranteed that writeback during freezing will |
960ea971 | 3413 | * see the dirty folio and writeprotect it again. |
14da9200 | 3414 | */ |
960ea971 MWO |
3415 | folio_mark_dirty(folio); |
3416 | folio_wait_stable(folio); | |
4fcf1c62 | 3417 | out: |
5df1a672 | 3418 | sb_end_pagefault(mapping->host->i_sb); |
4fcf1c62 JK |
3419 | return ret; |
3420 | } | |
4fcf1c62 | 3421 | |
f0f37e2f | 3422 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 3423 | .fault = filemap_fault, |
f1820361 | 3424 | .map_pages = filemap_map_pages, |
4fcf1c62 | 3425 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
3426 | }; |
3427 | ||
3428 | /* This is used for a general mmap of a disk file */ | |
3429 | ||
68d68ff6 | 3430 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3431 | { |
3432 | struct address_space *mapping = file->f_mapping; | |
3433 | ||
3434 | if (!mapping->a_ops->readpage) | |
3435 | return -ENOEXEC; | |
3436 | file_accessed(file); | |
3437 | vma->vm_ops = &generic_file_vm_ops; | |
3438 | return 0; | |
3439 | } | |
1da177e4 LT |
3440 | |
3441 | /* | |
3442 | * This is for filesystems which do not implement ->writepage. | |
3443 | */ | |
3444 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
3445 | { | |
3446 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
3447 | return -EINVAL; | |
3448 | return generic_file_mmap(file, vma); | |
3449 | } | |
3450 | #else | |
4b96a37d | 3451 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
45397228 | 3452 | { |
4b96a37d | 3453 | return VM_FAULT_SIGBUS; |
45397228 | 3454 | } |
68d68ff6 | 3455 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3456 | { |
3457 | return -ENOSYS; | |
3458 | } | |
68d68ff6 | 3459 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3460 | { |
3461 | return -ENOSYS; | |
3462 | } | |
3463 | #endif /* CONFIG_MMU */ | |
3464 | ||
45397228 | 3465 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
3466 | EXPORT_SYMBOL(generic_file_mmap); |
3467 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
3468 | ||
539a3322 MWO |
3469 | static struct folio *do_read_cache_folio(struct address_space *mapping, |
3470 | pgoff_t index, filler_t filler, void *data, gfp_t gfp) | |
67f9fd91 | 3471 | { |
539a3322 | 3472 | struct folio *folio; |
1da177e4 LT |
3473 | int err; |
3474 | repeat: | |
539a3322 MWO |
3475 | folio = filemap_get_folio(mapping, index); |
3476 | if (!folio) { | |
3477 | folio = filemap_alloc_folio(gfp, 0); | |
3478 | if (!folio) | |
eb2be189 | 3479 | return ERR_PTR(-ENOMEM); |
539a3322 | 3480 | err = filemap_add_folio(mapping, folio, index, gfp); |
eb2be189 | 3481 | if (unlikely(err)) { |
539a3322 | 3482 | folio_put(folio); |
eb2be189 NP |
3483 | if (err == -EEXIST) |
3484 | goto repeat; | |
22ecdb4f | 3485 | /* Presumably ENOMEM for xarray node */ |
1da177e4 LT |
3486 | return ERR_PTR(err); |
3487 | } | |
32b63529 MG |
3488 | |
3489 | filler: | |
6c45b454 | 3490 | if (filler) |
539a3322 | 3491 | err = filler(data, &folio->page); |
6c45b454 | 3492 | else |
539a3322 | 3493 | err = mapping->a_ops->readpage(data, &folio->page); |
6c45b454 | 3494 | |
1da177e4 | 3495 | if (err < 0) { |
539a3322 | 3496 | folio_put(folio); |
32b63529 | 3497 | return ERR_PTR(err); |
1da177e4 | 3498 | } |
1da177e4 | 3499 | |
539a3322 MWO |
3500 | folio_wait_locked(folio); |
3501 | if (!folio_test_uptodate(folio)) { | |
3502 | folio_put(folio); | |
3503 | return ERR_PTR(-EIO); | |
3504 | } | |
3505 | ||
32b63529 MG |
3506 | goto out; |
3507 | } | |
539a3322 | 3508 | if (folio_test_uptodate(folio)) |
1da177e4 LT |
3509 | goto out; |
3510 | ||
81f4c03b MWO |
3511 | if (!folio_trylock(folio)) { |
3512 | folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE); | |
3513 | goto repeat; | |
3514 | } | |
ebded027 | 3515 | |
81f4c03b | 3516 | /* Folio was truncated from mapping */ |
539a3322 MWO |
3517 | if (!folio->mapping) { |
3518 | folio_unlock(folio); | |
3519 | folio_put(folio); | |
32b63529 | 3520 | goto repeat; |
1da177e4 | 3521 | } |
ebded027 MG |
3522 | |
3523 | /* Someone else locked and filled the page in a very small window */ | |
539a3322 MWO |
3524 | if (folio_test_uptodate(folio)) { |
3525 | folio_unlock(folio); | |
1da177e4 LT |
3526 | goto out; |
3527 | } | |
faffdfa0 XT |
3528 | |
3529 | /* | |
3530 | * A previous I/O error may have been due to temporary | |
3531 | * failures. | |
3532 | * Clear page error before actual read, PG_error will be | |
3533 | * set again if read page fails. | |
3534 | */ | |
539a3322 | 3535 | folio_clear_error(folio); |
32b63529 MG |
3536 | goto filler; |
3537 | ||
c855ff37 | 3538 | out: |
539a3322 MWO |
3539 | folio_mark_accessed(folio); |
3540 | return folio; | |
6fe6900e | 3541 | } |
0531b2aa LT |
3542 | |
3543 | /** | |
539a3322 | 3544 | * read_cache_folio - read into page cache, fill it if needed |
0531b2aa LT |
3545 | * @mapping: the page's address_space |
3546 | * @index: the page index | |
3547 | * @filler: function to perform the read | |
5e5358e7 | 3548 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 3549 | * |
0531b2aa | 3550 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 3551 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
3552 | * |
3553 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3554 | * |
730633f0 JK |
3555 | * The function expects mapping->invalidate_lock to be already held. |
3556 | * | |
a862f68a | 3557 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa | 3558 | */ |
539a3322 MWO |
3559 | struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index, |
3560 | filler_t filler, void *data) | |
3561 | { | |
3562 | return do_read_cache_folio(mapping, index, filler, data, | |
3563 | mapping_gfp_mask(mapping)); | |
3564 | } | |
3565 | EXPORT_SYMBOL(read_cache_folio); | |
3566 | ||
3567 | static struct page *do_read_cache_page(struct address_space *mapping, | |
3568 | pgoff_t index, filler_t *filler, void *data, gfp_t gfp) | |
3569 | { | |
3570 | struct folio *folio; | |
3571 | ||
3572 | folio = do_read_cache_folio(mapping, index, filler, data, gfp); | |
3573 | if (IS_ERR(folio)) | |
3574 | return &folio->page; | |
3575 | return folio_file_page(folio, index); | |
3576 | } | |
3577 | ||
67f9fd91 | 3578 | struct page *read_cache_page(struct address_space *mapping, |
539a3322 | 3579 | pgoff_t index, filler_t *filler, void *data) |
0531b2aa | 3580 | { |
d322a8e5 CH |
3581 | return do_read_cache_page(mapping, index, filler, data, |
3582 | mapping_gfp_mask(mapping)); | |
0531b2aa | 3583 | } |
67f9fd91 | 3584 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
3585 | |
3586 | /** | |
3587 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
3588 | * @mapping: the page's address_space | |
3589 | * @index: the page index | |
3590 | * @gfp: the page allocator flags to use if allocating | |
3591 | * | |
3592 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 3593 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
3594 | * |
3595 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3596 | * |
730633f0 JK |
3597 | * The function expects mapping->invalidate_lock to be already held. |
3598 | * | |
a862f68a | 3599 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa LT |
3600 | */ |
3601 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
3602 | pgoff_t index, | |
3603 | gfp_t gfp) | |
3604 | { | |
6c45b454 | 3605 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
0531b2aa LT |
3606 | } |
3607 | EXPORT_SYMBOL(read_cache_page_gfp); | |
3608 | ||
afddba49 NP |
3609 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
3610 | loff_t pos, unsigned len, unsigned flags, | |
3611 | struct page **pagep, void **fsdata) | |
3612 | { | |
3613 | const struct address_space_operations *aops = mapping->a_ops; | |
3614 | ||
4e02ed4b | 3615 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 3616 | pagep, fsdata); |
afddba49 NP |
3617 | } |
3618 | EXPORT_SYMBOL(pagecache_write_begin); | |
3619 | ||
3620 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
3621 | loff_t pos, unsigned len, unsigned copied, | |
3622 | struct page *page, void *fsdata) | |
3623 | { | |
3624 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 3625 | |
4e02ed4b | 3626 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
3627 | } |
3628 | EXPORT_SYMBOL(pagecache_write_end); | |
3629 | ||
a92853b6 KK |
3630 | /* |
3631 | * Warn about a page cache invalidation failure during a direct I/O write. | |
3632 | */ | |
3633 | void dio_warn_stale_pagecache(struct file *filp) | |
3634 | { | |
3635 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); | |
3636 | char pathname[128]; | |
a92853b6 KK |
3637 | char *path; |
3638 | ||
5df1a672 | 3639 | errseq_set(&filp->f_mapping->wb_err, -EIO); |
a92853b6 KK |
3640 | if (__ratelimit(&_rs)) { |
3641 | path = file_path(filp, pathname, sizeof(pathname)); | |
3642 | if (IS_ERR(path)) | |
3643 | path = "(unknown)"; | |
3644 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); | |
3645 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, | |
3646 | current->comm); | |
3647 | } | |
3648 | } | |
3649 | ||
1da177e4 | 3650 | ssize_t |
1af5bb49 | 3651 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3652 | { |
3653 | struct file *file = iocb->ki_filp; | |
3654 | struct address_space *mapping = file->f_mapping; | |
3655 | struct inode *inode = mapping->host; | |
1af5bb49 | 3656 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3657 | ssize_t written; |
a969e903 CH |
3658 | size_t write_len; |
3659 | pgoff_t end; | |
1da177e4 | 3660 | |
0c949334 | 3661 | write_len = iov_iter_count(from); |
09cbfeaf | 3662 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3663 | |
6be96d3a GR |
3664 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3665 | /* If there are pages to writeback, return */ | |
5df1a672 | 3666 | if (filemap_range_has_page(file->f_mapping, pos, |
35f12f0f | 3667 | pos + write_len - 1)) |
6be96d3a GR |
3668 | return -EAGAIN; |
3669 | } else { | |
3670 | written = filemap_write_and_wait_range(mapping, pos, | |
3671 | pos + write_len - 1); | |
3672 | if (written) | |
3673 | goto out; | |
3674 | } | |
a969e903 CH |
3675 | |
3676 | /* | |
3677 | * After a write we want buffered reads to be sure to go to disk to get | |
3678 | * the new data. We invalidate clean cached page from the region we're | |
3679 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3680 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3681 | */ |
55635ba7 | 3682 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3683 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3684 | /* |
3685 | * If a page can not be invalidated, return 0 to fall back | |
3686 | * to buffered write. | |
3687 | */ | |
3688 | if (written) { | |
3689 | if (written == -EBUSY) | |
3690 | return 0; | |
3691 | goto out; | |
a969e903 CH |
3692 | } |
3693 | ||
639a93a5 | 3694 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3695 | |
3696 | /* | |
3697 | * Finally, try again to invalidate clean pages which might have been | |
3698 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3699 | * if the source of the write was an mmap'ed region of the file | |
3700 | * we're writing. Either one is a pretty crazy thing to do, | |
3701 | * so we don't support it 100%. If this invalidation | |
3702 | * fails, tough, the write still worked... | |
332391a9 LC |
3703 | * |
3704 | * Most of the time we do not need this since dio_complete() will do | |
3705 | * the invalidation for us. However there are some file systems that | |
3706 | * do not end up with dio_complete() being called, so let's not break | |
80c1fe90 KK |
3707 | * them by removing it completely. |
3708 | * | |
9266a140 KK |
3709 | * Noticeable example is a blkdev_direct_IO(). |
3710 | * | |
80c1fe90 | 3711 | * Skip invalidation for async writes or if mapping has no pages. |
a969e903 | 3712 | */ |
9266a140 KK |
3713 | if (written > 0 && mapping->nrpages && |
3714 | invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) | |
3715 | dio_warn_stale_pagecache(file); | |
a969e903 | 3716 | |
1da177e4 | 3717 | if (written > 0) { |
0116651c | 3718 | pos += written; |
639a93a5 | 3719 | write_len -= written; |
0116651c NK |
3720 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3721 | i_size_write(inode, pos); | |
1da177e4 LT |
3722 | mark_inode_dirty(inode); |
3723 | } | |
5cb6c6c7 | 3724 | iocb->ki_pos = pos; |
1da177e4 | 3725 | } |
ab2125df PB |
3726 | if (written != -EIOCBQUEUED) |
3727 | iov_iter_revert(from, write_len - iov_iter_count(from)); | |
a969e903 | 3728 | out: |
1da177e4 LT |
3729 | return written; |
3730 | } | |
3731 | EXPORT_SYMBOL(generic_file_direct_write); | |
3732 | ||
3b93f911 | 3733 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3734 | struct iov_iter *i, loff_t pos) |
3735 | { | |
3736 | struct address_space *mapping = file->f_mapping; | |
3737 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3738 | long status = 0; | |
3739 | ssize_t written = 0; | |
674b892e NP |
3740 | unsigned int flags = 0; |
3741 | ||
afddba49 NP |
3742 | do { |
3743 | struct page *page; | |
afddba49 NP |
3744 | unsigned long offset; /* Offset into pagecache page */ |
3745 | unsigned long bytes; /* Bytes to write to page */ | |
3746 | size_t copied; /* Bytes copied from user */ | |
3747 | void *fsdata; | |
3748 | ||
09cbfeaf KS |
3749 | offset = (pos & (PAGE_SIZE - 1)); |
3750 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3751 | iov_iter_count(i)); |
3752 | ||
3753 | again: | |
00a3d660 LT |
3754 | /* |
3755 | * Bring in the user page that we will copy from _first_. | |
3756 | * Otherwise there's a nasty deadlock on copying from the | |
3757 | * same page as we're writing to, without it being marked | |
3758 | * up-to-date. | |
00a3d660 | 3759 | */ |
a6294593 | 3760 | if (unlikely(fault_in_iov_iter_readable(i, bytes))) { |
00a3d660 LT |
3761 | status = -EFAULT; |
3762 | break; | |
3763 | } | |
3764 | ||
296291cd JK |
3765 | if (fatal_signal_pending(current)) { |
3766 | status = -EINTR; | |
3767 | break; | |
3768 | } | |
3769 | ||
674b892e | 3770 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3771 | &page, &fsdata); |
2457aec6 | 3772 | if (unlikely(status < 0)) |
afddba49 NP |
3773 | break; |
3774 | ||
931e80e4 | 3775 | if (mapping_writably_mapped(mapping)) |
3776 | flush_dcache_page(page); | |
00a3d660 | 3777 | |
f0b65f39 | 3778 | copied = copy_page_from_iter_atomic(page, offset, bytes, i); |
afddba49 NP |
3779 | flush_dcache_page(page); |
3780 | ||
3781 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3782 | page, fsdata); | |
f0b65f39 AV |
3783 | if (unlikely(status != copied)) { |
3784 | iov_iter_revert(i, copied - max(status, 0L)); | |
3785 | if (unlikely(status < 0)) | |
3786 | break; | |
3787 | } | |
afddba49 NP |
3788 | cond_resched(); |
3789 | ||
bc1bb416 | 3790 | if (unlikely(status == 0)) { |
afddba49 | 3791 | /* |
bc1bb416 AV |
3792 | * A short copy made ->write_end() reject the |
3793 | * thing entirely. Might be memory poisoning | |
3794 | * halfway through, might be a race with munmap, | |
3795 | * might be severe memory pressure. | |
afddba49 | 3796 | */ |
bc1bb416 AV |
3797 | if (copied) |
3798 | bytes = copied; | |
afddba49 NP |
3799 | goto again; |
3800 | } | |
f0b65f39 AV |
3801 | pos += status; |
3802 | written += status; | |
afddba49 NP |
3803 | |
3804 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3805 | } while (iov_iter_count(i)); |
3806 | ||
3807 | return written ? written : status; | |
3808 | } | |
3b93f911 | 3809 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3810 | |
e4dd9de3 | 3811 | /** |
8174202b | 3812 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3813 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3814 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3815 | * |
3816 | * This function does all the work needed for actually writing data to a | |
3817 | * file. It does all basic checks, removes SUID from the file, updates | |
3818 | * modification times and calls proper subroutines depending on whether we | |
3819 | * do direct IO or a standard buffered write. | |
3820 | * | |
9608703e | 3821 | * It expects i_rwsem to be grabbed unless we work on a block device or similar |
e4dd9de3 JK |
3822 | * object which does not need locking at all. |
3823 | * | |
3824 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3825 | * A caller has to handle it. This is mainly due to the fact that we want to | |
9608703e | 3826 | * avoid syncing under i_rwsem. |
a862f68a MR |
3827 | * |
3828 | * Return: | |
3829 | * * number of bytes written, even for truncated writes | |
3830 | * * negative error code if no data has been written at all | |
e4dd9de3 | 3831 | */ |
8174202b | 3832 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3833 | { |
3834 | struct file *file = iocb->ki_filp; | |
68d68ff6 | 3835 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3836 | struct inode *inode = mapping->host; |
3b93f911 | 3837 | ssize_t written = 0; |
1da177e4 | 3838 | ssize_t err; |
3b93f911 | 3839 | ssize_t status; |
1da177e4 | 3840 | |
1da177e4 | 3841 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3842 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3843 | err = file_remove_privs(file); |
1da177e4 LT |
3844 | if (err) |
3845 | goto out; | |
3846 | ||
c3b2da31 JB |
3847 | err = file_update_time(file); |
3848 | if (err) | |
3849 | goto out; | |
1da177e4 | 3850 | |
2ba48ce5 | 3851 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3852 | loff_t pos, endbyte; |
fb5527e6 | 3853 | |
1af5bb49 | 3854 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3855 | /* |
fbbbad4b MW |
3856 | * If the write stopped short of completing, fall back to |
3857 | * buffered writes. Some filesystems do this for writes to | |
3858 | * holes, for example. For DAX files, a buffered write will | |
3859 | * not succeed (even if it did, DAX does not handle dirty | |
3860 | * page-cache pages correctly). | |
1da177e4 | 3861 | */ |
0b8def9d | 3862 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3863 | goto out; |
3864 | ||
0b8def9d | 3865 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3866 | /* |
3b93f911 | 3867 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3868 | * then we want to return the number of bytes which were |
3869 | * direct-written, or the error code if that was zero. Note | |
3870 | * that this differs from normal direct-io semantics, which | |
3871 | * will return -EFOO even if some bytes were written. | |
3872 | */ | |
60bb4529 | 3873 | if (unlikely(status < 0)) { |
3b93f911 | 3874 | err = status; |
fb5527e6 JM |
3875 | goto out; |
3876 | } | |
fb5527e6 JM |
3877 | /* |
3878 | * We need to ensure that the page cache pages are written to | |
3879 | * disk and invalidated to preserve the expected O_DIRECT | |
3880 | * semantics. | |
3881 | */ | |
3b93f911 | 3882 | endbyte = pos + status - 1; |
0b8def9d | 3883 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3884 | if (err == 0) { |
0b8def9d | 3885 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3886 | written += status; |
fb5527e6 | 3887 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3888 | pos >> PAGE_SHIFT, |
3889 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3890 | } else { |
3891 | /* | |
3892 | * We don't know how much we wrote, so just return | |
3893 | * the number of bytes which were direct-written | |
3894 | */ | |
3895 | } | |
3896 | } else { | |
0b8def9d AV |
3897 | written = generic_perform_write(file, from, iocb->ki_pos); |
3898 | if (likely(written > 0)) | |
3899 | iocb->ki_pos += written; | |
fb5527e6 | 3900 | } |
1da177e4 LT |
3901 | out: |
3902 | current->backing_dev_info = NULL; | |
3903 | return written ? written : err; | |
3904 | } | |
8174202b | 3905 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3906 | |
e4dd9de3 | 3907 | /** |
8174202b | 3908 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3909 | * @iocb: IO state structure |
8174202b | 3910 | * @from: iov_iter with data to write |
e4dd9de3 | 3911 | * |
8174202b | 3912 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 | 3913 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
9608703e | 3914 | * and acquires i_rwsem as needed. |
a862f68a MR |
3915 | * Return: |
3916 | * * negative error code if no data has been written at all of | |
3917 | * vfs_fsync_range() failed for a synchronous write | |
3918 | * * number of bytes written, even for truncated writes | |
e4dd9de3 | 3919 | */ |
8174202b | 3920 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3921 | { |
3922 | struct file *file = iocb->ki_filp; | |
148f948b | 3923 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3924 | ssize_t ret; |
1da177e4 | 3925 | |
5955102c | 3926 | inode_lock(inode); |
3309dd04 AV |
3927 | ret = generic_write_checks(iocb, from); |
3928 | if (ret > 0) | |
5f380c7f | 3929 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3930 | inode_unlock(inode); |
1da177e4 | 3931 | |
e2592217 CH |
3932 | if (ret > 0) |
3933 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3934 | return ret; |
3935 | } | |
8174202b | 3936 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3937 | |
cf9a2ae8 | 3938 | /** |
82c50f8b MWO |
3939 | * filemap_release_folio() - Release fs-specific metadata on a folio. |
3940 | * @folio: The folio which the kernel is trying to free. | |
3941 | * @gfp: Memory allocation flags (and I/O mode). | |
cf9a2ae8 | 3942 | * |
82c50f8b MWO |
3943 | * The address_space is trying to release any data attached to a folio |
3944 | * (presumably at folio->private). | |
cf9a2ae8 | 3945 | * |
82c50f8b MWO |
3946 | * This will also be called if the private_2 flag is set on a page, |
3947 | * indicating that the folio has other metadata associated with it. | |
266cf658 | 3948 | * |
82c50f8b MWO |
3949 | * The @gfp argument specifies whether I/O may be performed to release |
3950 | * this page (__GFP_IO), and whether the call may block | |
3951 | * (__GFP_RECLAIM & __GFP_FS). | |
cf9a2ae8 | 3952 | * |
82c50f8b | 3953 | * Return: %true if the release was successful, otherwise %false. |
cf9a2ae8 | 3954 | */ |
82c50f8b | 3955 | bool filemap_release_folio(struct folio *folio, gfp_t gfp) |
cf9a2ae8 | 3956 | { |
82c50f8b | 3957 | struct address_space * const mapping = folio->mapping; |
cf9a2ae8 | 3958 | |
82c50f8b MWO |
3959 | BUG_ON(!folio_test_locked(folio)); |
3960 | if (folio_test_writeback(folio)) | |
3961 | return false; | |
cf9a2ae8 DH |
3962 | |
3963 | if (mapping && mapping->a_ops->releasepage) | |
82c50f8b MWO |
3964 | return mapping->a_ops->releasepage(&folio->page, gfp); |
3965 | return try_to_free_buffers(&folio->page); | |
cf9a2ae8 | 3966 | } |
82c50f8b | 3967 | EXPORT_SYMBOL(filemap_release_folio); |