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> |
cf264e13 | 25 | #include <linux/syscalls.h> |
1da177e4 LT |
26 | #include <linux/mman.h> |
27 | #include <linux/pagemap.h> | |
28 | #include <linux/file.h> | |
29 | #include <linux/uio.h> | |
cfcbfb13 | 30 | #include <linux/error-injection.h> |
1da177e4 LT |
31 | #include <linux/hash.h> |
32 | #include <linux/writeback.h> | |
53253383 | 33 | #include <linux/backing-dev.h> |
1da177e4 | 34 | #include <linux/pagevec.h> |
1da177e4 | 35 | #include <linux/security.h> |
44110fe3 | 36 | #include <linux/cpuset.h> |
00501b53 | 37 | #include <linux/hugetlb.h> |
8a9f3ccd | 38 | #include <linux/memcontrol.h> |
c7df8ad2 | 39 | #include <linux/shmem_fs.h> |
f1820361 | 40 | #include <linux/rmap.h> |
b1d29ba8 | 41 | #include <linux/delayacct.h> |
eb414681 | 42 | #include <linux/psi.h> |
d0e6a582 | 43 | #include <linux/ramfs.h> |
b9306a79 | 44 | #include <linux/page_idle.h> |
ffa65753 | 45 | #include <linux/migrate.h> |
07073eb0 DH |
46 | #include <linux/pipe_fs_i.h> |
47 | #include <linux/splice.h> | |
f9ce0be7 | 48 | #include <asm/pgalloc.h> |
de591a82 | 49 | #include <asm/tlbflush.h> |
0f8053a5 NP |
50 | #include "internal.h" |
51 | ||
fe0bfaaf RJ |
52 | #define CREATE_TRACE_POINTS |
53 | #include <trace/events/filemap.h> | |
54 | ||
1da177e4 | 55 | /* |
1da177e4 LT |
56 | * FIXME: remove all knowledge of the buffer layer from the core VM |
57 | */ | |
148f948b | 58 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 59 | |
1da177e4 LT |
60 | #include <asm/mman.h> |
61 | ||
cf264e13 NP |
62 | #include "swap.h" |
63 | ||
1da177e4 LT |
64 | /* |
65 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
66 | * though. | |
67 | * | |
68 | * Shared mappings now work. 15.8.1995 Bruno. | |
69 | * | |
70 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
71 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
72 | * | |
73 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
74 | */ | |
75 | ||
76 | /* | |
77 | * Lock ordering: | |
78 | * | |
c8c06efa | 79 | * ->i_mmap_rwsem (truncate_pagecache) |
e621900a | 80 | * ->private_lock (__free_pte->block_dirty_folio) |
5d337b91 | 81 | * ->swap_lock (exclusive_swap_page, others) |
b93b0163 | 82 | * ->i_pages lock |
1da177e4 | 83 | * |
9608703e | 84 | * ->i_rwsem |
730633f0 JK |
85 | * ->invalidate_lock (acquired by fs in truncate path) |
86 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) | |
1da177e4 | 87 | * |
c1e8d7c6 | 88 | * ->mmap_lock |
c8c06efa | 89 | * ->i_mmap_rwsem |
b8072f09 | 90 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
b93b0163 | 91 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
1da177e4 | 92 | * |
c1e8d7c6 | 93 | * ->mmap_lock |
730633f0 JK |
94 | * ->invalidate_lock (filemap_fault) |
95 | * ->lock_page (filemap_fault, access_process_vm) | |
1da177e4 | 96 | * |
9608703e | 97 | * ->i_rwsem (generic_perform_write) |
bb523b40 | 98 | * ->mmap_lock (fault_in_readable->do_page_fault) |
1da177e4 | 99 | * |
f758eeab | 100 | * bdi->wb.list_lock |
a66979ab | 101 | * sb_lock (fs/fs-writeback.c) |
b93b0163 | 102 | * ->i_pages lock (__sync_single_inode) |
1da177e4 | 103 | * |
c8c06efa | 104 | * ->i_mmap_rwsem |
0503ea8f | 105 | * ->anon_vma.lock (vma_merge) |
1da177e4 LT |
106 | * |
107 | * ->anon_vma.lock | |
b8072f09 | 108 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 109 | * |
b8072f09 | 110 | * ->page_table_lock or pte_lock |
5d337b91 | 111 | * ->swap_lock (try_to_unmap_one) |
1da177e4 | 112 | * ->private_lock (try_to_unmap_one) |
b93b0163 | 113 | * ->i_pages lock (try_to_unmap_one) |
15b44736 HD |
114 | * ->lruvec->lru_lock (follow_page->mark_page_accessed) |
115 | * ->lruvec->lru_lock (check_pte_range->isolate_lru_page) | |
1da177e4 | 116 | * ->private_lock (page_remove_rmap->set_page_dirty) |
b93b0163 | 117 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
f758eeab | 118 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 119 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 120 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 121 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 122 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
e621900a | 123 | * ->private_lock (zap_pte_range->block_dirty_folio) |
1da177e4 | 124 | * |
c8c06efa | 125 | * ->i_mmap_rwsem |
9a3c531d | 126 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
127 | */ |
128 | ||
5c024e6a | 129 | static void page_cache_delete(struct address_space *mapping, |
a548b615 | 130 | struct folio *folio, void *shadow) |
91b0abe3 | 131 | { |
a548b615 MWO |
132 | XA_STATE(xas, &mapping->i_pages, folio->index); |
133 | long nr = 1; | |
c70b647d | 134 | |
5c024e6a | 135 | mapping_set_update(&xas, mapping); |
c70b647d | 136 | |
5c024e6a | 137 | /* hugetlb pages are represented by a single entry in the xarray */ |
a548b615 MWO |
138 | if (!folio_test_hugetlb(folio)) { |
139 | xas_set_order(&xas, folio->index, folio_order(folio)); | |
140 | nr = folio_nr_pages(folio); | |
5c024e6a | 141 | } |
91b0abe3 | 142 | |
a548b615 | 143 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
449dd698 | 144 | |
5c024e6a MW |
145 | xas_store(&xas, shadow); |
146 | xas_init_marks(&xas); | |
d3798ae8 | 147 | |
a548b615 | 148 | folio->mapping = NULL; |
2300638b | 149 | /* Leave page->index set: truncation lookup relies upon it */ |
d3798ae8 | 150 | mapping->nrpages -= nr; |
91b0abe3 JW |
151 | } |
152 | ||
621db488 MWO |
153 | static void filemap_unaccount_folio(struct address_space *mapping, |
154 | struct folio *folio) | |
1da177e4 | 155 | { |
621db488 | 156 | long nr; |
1da177e4 | 157 | |
621db488 MWO |
158 | VM_BUG_ON_FOLIO(folio_mapped(folio), folio); |
159 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) { | |
06b241f3 | 160 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", |
621db488 MWO |
161 | current->comm, folio_pfn(folio)); |
162 | dump_page(&folio->page, "still mapped when deleted"); | |
06b241f3 HD |
163 | dump_stack(); |
164 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
165 | ||
85207ad8 HD |
166 | if (mapping_exiting(mapping) && !folio_test_large(folio)) { |
167 | int mapcount = page_mapcount(&folio->page); | |
168 | ||
169 | if (folio_ref_count(folio) >= mapcount + 2) { | |
170 | /* | |
171 | * All vmas have already been torn down, so it's | |
172 | * a good bet that actually the page is unmapped | |
173 | * and we'd rather not leak it: if we're wrong, | |
174 | * another bad page check should catch it later. | |
175 | */ | |
176 | page_mapcount_reset(&folio->page); | |
177 | folio_ref_sub(folio, mapcount); | |
178 | } | |
06b241f3 HD |
179 | } |
180 | } | |
181 | ||
621db488 MWO |
182 | /* hugetlb folios do not participate in page cache accounting. */ |
183 | if (folio_test_hugetlb(folio)) | |
5ecc4d85 | 184 | return; |
09612fa6 | 185 | |
621db488 | 186 | nr = folio_nr_pages(folio); |
5ecc4d85 | 187 | |
621db488 MWO |
188 | __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr); |
189 | if (folio_test_swapbacked(folio)) { | |
190 | __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr); | |
191 | if (folio_test_pmd_mappable(folio)) | |
192 | __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr); | |
193 | } else if (folio_test_pmd_mappable(folio)) { | |
194 | __lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr); | |
09d91cda | 195 | filemap_nr_thps_dec(mapping); |
800d8c63 | 196 | } |
5ecc4d85 JK |
197 | |
198 | /* | |
621db488 MWO |
199 | * At this point folio must be either written or cleaned by |
200 | * truncate. Dirty folio here signals a bug and loss of | |
566d3362 | 201 | * unwritten data - on ordinary filesystems. |
5ecc4d85 | 202 | * |
566d3362 HD |
203 | * But it's harmless on in-memory filesystems like tmpfs; and can |
204 | * occur when a driver which did get_user_pages() sets page dirty | |
205 | * before putting it, while the inode is being finally evicted. | |
206 | * | |
207 | * Below fixes dirty accounting after removing the folio entirely | |
621db488 MWO |
208 | * but leaves the dirty flag set: it has no effect for truncated |
209 | * folio and anyway will be cleared before returning folio to | |
5ecc4d85 JK |
210 | * buddy allocator. |
211 | */ | |
566d3362 HD |
212 | if (WARN_ON_ONCE(folio_test_dirty(folio) && |
213 | mapping_can_writeback(mapping))) | |
214 | folio_account_cleaned(folio, inode_to_wb(mapping->host)); | |
5ecc4d85 JK |
215 | } |
216 | ||
217 | /* | |
218 | * Delete a page from the page cache and free it. Caller has to make | |
219 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 220 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 | 221 | */ |
452e9e69 | 222 | void __filemap_remove_folio(struct folio *folio, void *shadow) |
5ecc4d85 | 223 | { |
452e9e69 | 224 | struct address_space *mapping = folio->mapping; |
5ecc4d85 | 225 | |
a0580c6f | 226 | trace_mm_filemap_delete_from_page_cache(folio); |
621db488 | 227 | filemap_unaccount_folio(mapping, folio); |
a548b615 | 228 | page_cache_delete(mapping, folio, shadow); |
1da177e4 LT |
229 | } |
230 | ||
78f42660 | 231 | void filemap_free_folio(struct address_space *mapping, struct folio *folio) |
59c66c5f | 232 | { |
d2329aa0 | 233 | void (*free_folio)(struct folio *); |
3abb28e2 | 234 | int refs = 1; |
59c66c5f | 235 | |
d2329aa0 MWO |
236 | free_folio = mapping->a_ops->free_folio; |
237 | if (free_folio) | |
238 | free_folio(folio); | |
59c66c5f | 239 | |
3abb28e2 MWO |
240 | if (folio_test_large(folio) && !folio_test_hugetlb(folio)) |
241 | refs = folio_nr_pages(folio); | |
242 | folio_put_refs(folio, refs); | |
59c66c5f JK |
243 | } |
244 | ||
702cfbf9 | 245 | /** |
452e9e69 MWO |
246 | * filemap_remove_folio - Remove folio from page cache. |
247 | * @folio: The folio. | |
702cfbf9 | 248 | * |
452e9e69 MWO |
249 | * This must be called only on folios that are locked and have been |
250 | * verified to be in the page cache. It will never put the folio into | |
251 | * the free list because the caller has a reference on the page. | |
702cfbf9 | 252 | */ |
452e9e69 | 253 | void filemap_remove_folio(struct folio *folio) |
1da177e4 | 254 | { |
452e9e69 | 255 | struct address_space *mapping = folio->mapping; |
1da177e4 | 256 | |
452e9e69 | 257 | BUG_ON(!folio_test_locked(folio)); |
51b8c1fe | 258 | spin_lock(&mapping->host->i_lock); |
30472509 | 259 | xa_lock_irq(&mapping->i_pages); |
452e9e69 | 260 | __filemap_remove_folio(folio, NULL); |
30472509 | 261 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
262 | if (mapping_shrinkable(mapping)) |
263 | inode_add_lru(mapping->host); | |
264 | spin_unlock(&mapping->host->i_lock); | |
6072d13c | 265 | |
452e9e69 | 266 | filemap_free_folio(mapping, folio); |
97cecb5a | 267 | } |
97cecb5a | 268 | |
aa65c29c | 269 | /* |
51dcbdac MWO |
270 | * page_cache_delete_batch - delete several folios from page cache |
271 | * @mapping: the mapping to which folios belong | |
272 | * @fbatch: batch of folios to delete | |
aa65c29c | 273 | * |
51dcbdac MWO |
274 | * The function walks over mapping->i_pages and removes folios passed in |
275 | * @fbatch from the mapping. The function expects @fbatch to be sorted | |
276 | * by page index and is optimised for it to be dense. | |
277 | * It tolerates holes in @fbatch (mapping entries at those indices are not | |
278 | * modified). | |
aa65c29c | 279 | * |
b93b0163 | 280 | * The function expects the i_pages lock to be held. |
aa65c29c | 281 | */ |
ef8e5717 | 282 | static void page_cache_delete_batch(struct address_space *mapping, |
51dcbdac | 283 | struct folio_batch *fbatch) |
aa65c29c | 284 | { |
51dcbdac | 285 | XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index); |
6b24ca4a | 286 | long total_pages = 0; |
4101196b | 287 | int i = 0; |
1afd7ae5 | 288 | struct folio *folio; |
aa65c29c | 289 | |
ef8e5717 | 290 | mapping_set_update(&xas, mapping); |
1afd7ae5 | 291 | xas_for_each(&xas, folio, ULONG_MAX) { |
51dcbdac | 292 | if (i >= folio_batch_count(fbatch)) |
aa65c29c | 293 | break; |
4101196b MWO |
294 | |
295 | /* A swap/dax/shadow entry got inserted? Skip it. */ | |
1afd7ae5 | 296 | if (xa_is_value(folio)) |
aa65c29c | 297 | continue; |
4101196b MWO |
298 | /* |
299 | * A page got inserted in our range? Skip it. We have our | |
300 | * pages locked so they are protected from being removed. | |
301 | * If we see a page whose index is higher than ours, it | |
302 | * means our page has been removed, which shouldn't be | |
303 | * possible because we're holding the PageLock. | |
304 | */ | |
51dcbdac | 305 | if (folio != fbatch->folios[i]) { |
1afd7ae5 | 306 | VM_BUG_ON_FOLIO(folio->index > |
51dcbdac | 307 | fbatch->folios[i]->index, folio); |
4101196b MWO |
308 | continue; |
309 | } | |
310 | ||
1afd7ae5 | 311 | WARN_ON_ONCE(!folio_test_locked(folio)); |
4101196b | 312 | |
6b24ca4a | 313 | folio->mapping = NULL; |
51dcbdac | 314 | /* Leave folio->index set: truncation lookup relies on it */ |
4101196b | 315 | |
6b24ca4a | 316 | i++; |
ef8e5717 | 317 | xas_store(&xas, NULL); |
6b24ca4a | 318 | total_pages += folio_nr_pages(folio); |
aa65c29c JK |
319 | } |
320 | mapping->nrpages -= total_pages; | |
321 | } | |
322 | ||
323 | void delete_from_page_cache_batch(struct address_space *mapping, | |
51dcbdac | 324 | struct folio_batch *fbatch) |
aa65c29c JK |
325 | { |
326 | int i; | |
aa65c29c | 327 | |
51dcbdac | 328 | if (!folio_batch_count(fbatch)) |
aa65c29c JK |
329 | return; |
330 | ||
51b8c1fe | 331 | spin_lock(&mapping->host->i_lock); |
30472509 | 332 | xa_lock_irq(&mapping->i_pages); |
51dcbdac MWO |
333 | for (i = 0; i < folio_batch_count(fbatch); i++) { |
334 | struct folio *folio = fbatch->folios[i]; | |
aa65c29c | 335 | |
a0580c6f MWO |
336 | trace_mm_filemap_delete_from_page_cache(folio); |
337 | filemap_unaccount_folio(mapping, folio); | |
aa65c29c | 338 | } |
51dcbdac | 339 | page_cache_delete_batch(mapping, fbatch); |
30472509 | 340 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
341 | if (mapping_shrinkable(mapping)) |
342 | inode_add_lru(mapping->host); | |
343 | spin_unlock(&mapping->host->i_lock); | |
aa65c29c | 344 | |
51dcbdac MWO |
345 | for (i = 0; i < folio_batch_count(fbatch); i++) |
346 | filemap_free_folio(mapping, fbatch->folios[i]); | |
aa65c29c JK |
347 | } |
348 | ||
d72d9e2a | 349 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
350 | { |
351 | int ret = 0; | |
352 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
353 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
354 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 355 | ret = -ENOSPC; |
7fcbbaf1 JA |
356 | if (test_bit(AS_EIO, &mapping->flags) && |
357 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
358 | ret = -EIO; |
359 | return ret; | |
360 | } | |
d72d9e2a | 361 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 362 | |
76341cab JL |
363 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
364 | { | |
365 | /* Check for outstanding write errors */ | |
366 | if (test_bit(AS_EIO, &mapping->flags)) | |
367 | return -EIO; | |
368 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
369 | return -ENOSPC; | |
370 | return 0; | |
371 | } | |
372 | ||
5a798493 JB |
373 | /** |
374 | * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range | |
375 | * @mapping: address space structure to write | |
376 | * @wbc: the writeback_control controlling the writeout | |
377 | * | |
378 | * Call writepages on the mapping using the provided wbc to control the | |
379 | * writeout. | |
380 | * | |
381 | * Return: %0 on success, negative error code otherwise. | |
382 | */ | |
383 | int filemap_fdatawrite_wbc(struct address_space *mapping, | |
384 | struct writeback_control *wbc) | |
385 | { | |
386 | int ret; | |
387 | ||
388 | if (!mapping_can_writeback(mapping) || | |
389 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
390 | return 0; | |
391 | ||
392 | wbc_attach_fdatawrite_inode(wbc, mapping->host); | |
393 | ret = do_writepages(mapping, wbc); | |
394 | wbc_detach_inode(wbc); | |
395 | return ret; | |
396 | } | |
397 | EXPORT_SYMBOL(filemap_fdatawrite_wbc); | |
398 | ||
1da177e4 | 399 | /** |
485bb99b | 400 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
401 | * @mapping: address space structure to write |
402 | * @start: offset in bytes where the range starts | |
469eb4d0 | 403 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 404 | * @sync_mode: enable synchronous operation |
1da177e4 | 405 | * |
485bb99b RD |
406 | * Start writeback against all of a mapping's dirty pages that lie |
407 | * within the byte offsets <start, end> inclusive. | |
408 | * | |
1da177e4 | 409 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 410 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
411 | * these two operations is that if a dirty page/buffer is encountered, it must |
412 | * be waited upon, and not just skipped over. | |
a862f68a MR |
413 | * |
414 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 415 | */ |
ebcf28e1 AM |
416 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
417 | loff_t end, int sync_mode) | |
1da177e4 | 418 | { |
1da177e4 LT |
419 | struct writeback_control wbc = { |
420 | .sync_mode = sync_mode, | |
05fe478d | 421 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
422 | .range_start = start, |
423 | .range_end = end, | |
1da177e4 LT |
424 | }; |
425 | ||
5a798493 | 426 | return filemap_fdatawrite_wbc(mapping, &wbc); |
1da177e4 LT |
427 | } |
428 | ||
429 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
430 | int sync_mode) | |
431 | { | |
111ebb6e | 432 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
433 | } |
434 | ||
435 | int filemap_fdatawrite(struct address_space *mapping) | |
436 | { | |
437 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
438 | } | |
439 | EXPORT_SYMBOL(filemap_fdatawrite); | |
440 | ||
f4c0a0fd | 441 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 442 | loff_t end) |
1da177e4 LT |
443 | { |
444 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
445 | } | |
f4c0a0fd | 446 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 447 | |
485bb99b RD |
448 | /** |
449 | * filemap_flush - mostly a non-blocking flush | |
450 | * @mapping: target address_space | |
451 | * | |
1da177e4 LT |
452 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
453 | * purposes - I/O may not be started against all dirty pages. | |
a862f68a MR |
454 | * |
455 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 LT |
456 | */ |
457 | int filemap_flush(struct address_space *mapping) | |
458 | { | |
459 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
460 | } | |
461 | EXPORT_SYMBOL(filemap_flush); | |
462 | ||
7fc9e472 GR |
463 | /** |
464 | * filemap_range_has_page - check if a page exists in range. | |
465 | * @mapping: address space within which to check | |
466 | * @start_byte: offset in bytes where the range starts | |
467 | * @end_byte: offset in bytes where the range ends (inclusive) | |
468 | * | |
469 | * Find at least one page in the range supplied, usually used to check if | |
470 | * direct writing in this range will trigger a writeback. | |
a862f68a MR |
471 | * |
472 | * Return: %true if at least one page exists in the specified range, | |
473 | * %false otherwise. | |
7fc9e472 GR |
474 | */ |
475 | bool filemap_range_has_page(struct address_space *mapping, | |
476 | loff_t start_byte, loff_t end_byte) | |
477 | { | |
eff3b364 | 478 | struct folio *folio; |
8fa8e538 MW |
479 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
480 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
7fc9e472 GR |
481 | |
482 | if (end_byte < start_byte) | |
483 | return false; | |
484 | ||
8fa8e538 MW |
485 | rcu_read_lock(); |
486 | for (;;) { | |
eff3b364 MWO |
487 | folio = xas_find(&xas, max); |
488 | if (xas_retry(&xas, folio)) | |
8fa8e538 MW |
489 | continue; |
490 | /* Shadow entries don't count */ | |
eff3b364 | 491 | if (xa_is_value(folio)) |
8fa8e538 MW |
492 | continue; |
493 | /* | |
494 | * We don't need to try to pin this page; we're about to | |
495 | * release the RCU lock anyway. It is enough to know that | |
496 | * there was a page here recently. | |
497 | */ | |
498 | break; | |
499 | } | |
500 | rcu_read_unlock(); | |
7fc9e472 | 501 | |
eff3b364 | 502 | return folio != NULL; |
7fc9e472 GR |
503 | } |
504 | EXPORT_SYMBOL(filemap_range_has_page); | |
505 | ||
5e8fcc1a | 506 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 507 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 508 | { |
09cbfeaf KS |
509 | pgoff_t index = start_byte >> PAGE_SHIFT; |
510 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
6817ef51 VMO |
511 | struct folio_batch fbatch; |
512 | unsigned nr_folios; | |
513 | ||
514 | folio_batch_init(&fbatch); | |
1da177e4 | 515 | |
312e9d2f | 516 | while (index <= end) { |
1da177e4 LT |
517 | unsigned i; |
518 | ||
6817ef51 VMO |
519 | nr_folios = filemap_get_folios_tag(mapping, &index, end, |
520 | PAGECACHE_TAG_WRITEBACK, &fbatch); | |
521 | ||
522 | if (!nr_folios) | |
312e9d2f JK |
523 | break; |
524 | ||
6817ef51 VMO |
525 | for (i = 0; i < nr_folios; i++) { |
526 | struct folio *folio = fbatch.folios[i]; | |
1da177e4 | 527 | |
6817ef51 VMO |
528 | folio_wait_writeback(folio); |
529 | folio_clear_error(folio); | |
1da177e4 | 530 | } |
6817ef51 | 531 | folio_batch_release(&fbatch); |
1da177e4 LT |
532 | cond_resched(); |
533 | } | |
aa750fd7 JN |
534 | } |
535 | ||
536 | /** | |
537 | * filemap_fdatawait_range - wait for writeback to complete | |
538 | * @mapping: address space structure to wait for | |
539 | * @start_byte: offset in bytes where the range starts | |
540 | * @end_byte: offset in bytes where the range ends (inclusive) | |
541 | * | |
542 | * Walk the list of under-writeback pages of the given address space | |
543 | * in the given range and wait for all of them. Check error status of | |
544 | * the address space and return it. | |
545 | * | |
546 | * Since the error status of the address space is cleared by this function, | |
547 | * callers are responsible for checking the return value and handling and/or | |
548 | * reporting the error. | |
a862f68a MR |
549 | * |
550 | * Return: error status of the address space. | |
aa750fd7 JN |
551 | */ |
552 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
553 | loff_t end_byte) | |
554 | { | |
5e8fcc1a JL |
555 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
556 | return filemap_check_errors(mapping); | |
1da177e4 | 557 | } |
d3bccb6f JK |
558 | EXPORT_SYMBOL(filemap_fdatawait_range); |
559 | ||
aa0bfcd9 RZ |
560 | /** |
561 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete | |
562 | * @mapping: address space structure to wait for | |
563 | * @start_byte: offset in bytes where the range starts | |
564 | * @end_byte: offset in bytes where the range ends (inclusive) | |
565 | * | |
566 | * Walk the list of under-writeback pages of the given address space in the | |
567 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), | |
568 | * this function does not clear error status of the address space. | |
569 | * | |
570 | * Use this function if callers don't handle errors themselves. Expected | |
571 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
572 | * fsfreeze(8) | |
573 | */ | |
574 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, | |
575 | loff_t start_byte, loff_t end_byte) | |
576 | { | |
577 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
578 | return filemap_check_and_keep_errors(mapping); | |
579 | } | |
580 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); | |
581 | ||
a823e458 JL |
582 | /** |
583 | * file_fdatawait_range - wait for writeback to complete | |
584 | * @file: file pointing to address space structure to wait for | |
585 | * @start_byte: offset in bytes where the range starts | |
586 | * @end_byte: offset in bytes where the range ends (inclusive) | |
587 | * | |
588 | * Walk the list of under-writeback pages of the address space that file | |
589 | * refers to, in the given range and wait for all of them. Check error | |
590 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
591 | * | |
592 | * Since the error status of the file is advanced by this function, | |
593 | * callers are responsible for checking the return value and handling and/or | |
594 | * reporting the error. | |
a862f68a MR |
595 | * |
596 | * Return: error status of the address space vs. the file->f_wb_err cursor. | |
a823e458 JL |
597 | */ |
598 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
599 | { | |
600 | struct address_space *mapping = file->f_mapping; | |
601 | ||
602 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
603 | return file_check_and_advance_wb_err(file); | |
604 | } | |
605 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 606 | |
aa750fd7 JN |
607 | /** |
608 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
609 | * @mapping: address space structure to wait for | |
610 | * | |
611 | * Walk the list of under-writeback pages of the given address space | |
612 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
613 | * does not clear error status of the address space. | |
614 | * | |
615 | * Use this function if callers don't handle errors themselves. Expected | |
616 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
617 | * fsfreeze(8) | |
a862f68a MR |
618 | * |
619 | * Return: error status of the address space. | |
aa750fd7 | 620 | */ |
76341cab | 621 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 622 | { |
ffb959bb | 623 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 624 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 625 | } |
76341cab | 626 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 627 | |
875d91b1 | 628 | /* Returns true if writeback might be needed or already in progress. */ |
9326c9b2 | 629 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 630 | { |
875d91b1 | 631 | return mapping->nrpages; |
1da177e4 | 632 | } |
1da177e4 | 633 | |
4bdcd1dd JA |
634 | bool filemap_range_has_writeback(struct address_space *mapping, |
635 | loff_t start_byte, loff_t end_byte) | |
f8ee8909 JA |
636 | { |
637 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); | |
638 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
b05f41a1 | 639 | struct folio *folio; |
f8ee8909 JA |
640 | |
641 | if (end_byte < start_byte) | |
642 | return false; | |
643 | ||
644 | rcu_read_lock(); | |
b05f41a1 VMO |
645 | xas_for_each(&xas, folio, max) { |
646 | if (xas_retry(&xas, folio)) | |
f8ee8909 | 647 | continue; |
b05f41a1 | 648 | if (xa_is_value(folio)) |
f8ee8909 | 649 | continue; |
b05f41a1 VMO |
650 | if (folio_test_dirty(folio) || folio_test_locked(folio) || |
651 | folio_test_writeback(folio)) | |
f8ee8909 JA |
652 | break; |
653 | } | |
654 | rcu_read_unlock(); | |
b05f41a1 | 655 | return folio != NULL; |
63135aa3 | 656 | } |
4bdcd1dd | 657 | EXPORT_SYMBOL_GPL(filemap_range_has_writeback); |
63135aa3 | 658 | |
485bb99b RD |
659 | /** |
660 | * filemap_write_and_wait_range - write out & wait on a file range | |
661 | * @mapping: the address_space for the pages | |
662 | * @lstart: offset in bytes where the range starts | |
663 | * @lend: offset in bytes where the range ends (inclusive) | |
664 | * | |
469eb4d0 AM |
665 | * Write out and wait upon file offsets lstart->lend, inclusive. |
666 | * | |
0e056eb5 | 667 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 | 668 | * that this function can be used to write to the very end-of-file (end = -1). |
a862f68a MR |
669 | * |
670 | * Return: error status of the address space. | |
469eb4d0 | 671 | */ |
1da177e4 LT |
672 | int filemap_write_and_wait_range(struct address_space *mapping, |
673 | loff_t lstart, loff_t lend) | |
674 | { | |
ccac11da | 675 | int err = 0, err2; |
1da177e4 | 676 | |
feeb9b26 BF |
677 | if (lend < lstart) |
678 | return 0; | |
679 | ||
9326c9b2 | 680 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
681 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
682 | WB_SYNC_ALL); | |
ddf8f376 IW |
683 | /* |
684 | * Even if the above returned error, the pages may be | |
685 | * written partially (e.g. -ENOSPC), so we wait for it. | |
686 | * But the -EIO is special case, it may indicate the worst | |
687 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
688 | */ | |
ccac11da ML |
689 | if (err != -EIO) |
690 | __filemap_fdatawait_range(mapping, lstart, lend); | |
1da177e4 | 691 | } |
ccac11da ML |
692 | err2 = filemap_check_errors(mapping); |
693 | if (!err) | |
694 | err = err2; | |
28fd1298 | 695 | return err; |
1da177e4 | 696 | } |
f6995585 | 697 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 698 | |
5660e13d JL |
699 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
700 | { | |
3acdfd28 | 701 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
702 | |
703 | trace_filemap_set_wb_err(mapping, eseq); | |
704 | } | |
705 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
706 | ||
707 | /** | |
708 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
709 | * and advance wb_err to current one | |
710 | * @file: struct file on which the error is being reported | |
711 | * | |
712 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
713 | * want to report any writeback errors that occurred since the last fsync (or | |
714 | * since the file was opened if there haven't been any). | |
715 | * | |
716 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
717 | * then just quickly return 0. The file is all caught up. | |
718 | * | |
719 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
720 | * it and try to swap it into place. If it works, or another task beat us | |
721 | * to it with the new value, then update the f_wb_err and return the error | |
722 | * portion. The error at this point must be reported via proper channels | |
723 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
724 | * | |
725 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
726 | * value is protected by the f_lock since we must ensure that it reflects | |
727 | * the latest value swapped in for this file descriptor. | |
a862f68a MR |
728 | * |
729 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
730 | */ |
731 | int file_check_and_advance_wb_err(struct file *file) | |
732 | { | |
733 | int err = 0; | |
734 | errseq_t old = READ_ONCE(file->f_wb_err); | |
735 | struct address_space *mapping = file->f_mapping; | |
736 | ||
737 | /* Locklessly handle the common case where nothing has changed */ | |
738 | if (errseq_check(&mapping->wb_err, old)) { | |
739 | /* Something changed, must use slow path */ | |
740 | spin_lock(&file->f_lock); | |
741 | old = file->f_wb_err; | |
742 | err = errseq_check_and_advance(&mapping->wb_err, | |
743 | &file->f_wb_err); | |
744 | trace_file_check_and_advance_wb_err(file, old); | |
745 | spin_unlock(&file->f_lock); | |
746 | } | |
f4e222c5 JL |
747 | |
748 | /* | |
749 | * We're mostly using this function as a drop in replacement for | |
750 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
751 | * that the legacy code would have had on these flags. | |
752 | */ | |
753 | clear_bit(AS_EIO, &mapping->flags); | |
754 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
755 | return err; |
756 | } | |
757 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
758 | ||
759 | /** | |
760 | * file_write_and_wait_range - write out & wait on a file range | |
761 | * @file: file pointing to address_space with pages | |
762 | * @lstart: offset in bytes where the range starts | |
763 | * @lend: offset in bytes where the range ends (inclusive) | |
764 | * | |
765 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
766 | * | |
767 | * Note that @lend is inclusive (describes the last byte to be written) so | |
768 | * that this function can be used to write to the very end-of-file (end = -1). | |
769 | * | |
770 | * After writing out and waiting on the data, we check and advance the | |
771 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
a862f68a MR |
772 | * |
773 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
774 | */ |
775 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
776 | { | |
777 | int err = 0, err2; | |
778 | struct address_space *mapping = file->f_mapping; | |
779 | ||
feeb9b26 BF |
780 | if (lend < lstart) |
781 | return 0; | |
782 | ||
9326c9b2 | 783 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
784 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
785 | WB_SYNC_ALL); | |
786 | /* See comment of filemap_write_and_wait() */ | |
787 | if (err != -EIO) | |
788 | __filemap_fdatawait_range(mapping, lstart, lend); | |
789 | } | |
790 | err2 = file_check_and_advance_wb_err(file); | |
791 | if (!err) | |
792 | err = err2; | |
793 | return err; | |
794 | } | |
795 | EXPORT_SYMBOL(file_write_and_wait_range); | |
796 | ||
ef6a3c63 | 797 | /** |
3720dd6d VMO |
798 | * replace_page_cache_folio - replace a pagecache folio with a new one |
799 | * @old: folio to be replaced | |
800 | * @new: folio to replace with | |
801 | * | |
802 | * This function replaces a folio in the pagecache with a new one. On | |
803 | * success it acquires the pagecache reference for the new folio and | |
804 | * drops it for the old folio. Both the old and new folios must be | |
805 | * locked. This function does not add the new folio to the LRU, the | |
ef6a3c63 MS |
806 | * caller must do that. |
807 | * | |
74d60958 | 808 | * The remove + add is atomic. This function cannot fail. |
ef6a3c63 | 809 | */ |
3720dd6d | 810 | void replace_page_cache_folio(struct folio *old, struct folio *new) |
ef6a3c63 | 811 | { |
74d60958 | 812 | struct address_space *mapping = old->mapping; |
d2329aa0 | 813 | void (*free_folio)(struct folio *) = mapping->a_ops->free_folio; |
74d60958 MW |
814 | pgoff_t offset = old->index; |
815 | XA_STATE(xas, &mapping->i_pages, offset); | |
ef6a3c63 | 816 | |
3720dd6d VMO |
817 | VM_BUG_ON_FOLIO(!folio_test_locked(old), old); |
818 | VM_BUG_ON_FOLIO(!folio_test_locked(new), new); | |
819 | VM_BUG_ON_FOLIO(new->mapping, new); | |
ef6a3c63 | 820 | |
3720dd6d | 821 | folio_get(new); |
74d60958 MW |
822 | new->mapping = mapping; |
823 | new->index = offset; | |
ef6a3c63 | 824 | |
3720dd6d | 825 | mem_cgroup_migrate(old, new); |
0d1c2072 | 826 | |
30472509 | 827 | xas_lock_irq(&xas); |
74d60958 | 828 | xas_store(&xas, new); |
4165b9b4 | 829 | |
74d60958 MW |
830 | old->mapping = NULL; |
831 | /* hugetlb pages do not participate in page cache accounting. */ | |
3720dd6d VMO |
832 | if (!folio_test_hugetlb(old)) |
833 | __lruvec_stat_sub_folio(old, NR_FILE_PAGES); | |
834 | if (!folio_test_hugetlb(new)) | |
835 | __lruvec_stat_add_folio(new, NR_FILE_PAGES); | |
836 | if (folio_test_swapbacked(old)) | |
837 | __lruvec_stat_sub_folio(old, NR_SHMEM); | |
838 | if (folio_test_swapbacked(new)) | |
839 | __lruvec_stat_add_folio(new, NR_SHMEM); | |
30472509 | 840 | xas_unlock_irq(&xas); |
d2329aa0 | 841 | if (free_folio) |
3720dd6d VMO |
842 | free_folio(old); |
843 | folio_put(old); | |
ef6a3c63 | 844 | } |
3720dd6d | 845 | EXPORT_SYMBOL_GPL(replace_page_cache_folio); |
ef6a3c63 | 846 | |
9dd3d069 MWO |
847 | noinline int __filemap_add_folio(struct address_space *mapping, |
848 | struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp) | |
1da177e4 | 849 | { |
9dd3d069 MWO |
850 | XA_STATE(xas, &mapping->i_pages, index); |
851 | int huge = folio_test_hugetlb(folio); | |
da74240e | 852 | bool charged = false; |
d68eccad | 853 | long nr = 1; |
e286781d | 854 | |
9dd3d069 MWO |
855 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
856 | VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio); | |
74d60958 | 857 | mapping_set_update(&xas, mapping); |
e286781d | 858 | |
3fea5a49 | 859 | if (!huge) { |
d68eccad | 860 | int error = mem_cgroup_charge(folio, NULL, gfp); |
9dd3d069 | 861 | VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio); |
3fea5a49 | 862 | if (error) |
d68eccad | 863 | return error; |
da74240e | 864 | charged = true; |
d68eccad MWO |
865 | xas_set_order(&xas, index, folio_order(folio)); |
866 | nr = folio_nr_pages(folio); | |
3fea5a49 JW |
867 | } |
868 | ||
198b62f8 | 869 | gfp &= GFP_RECLAIM_MASK; |
d68eccad MWO |
870 | folio_ref_add(folio, nr); |
871 | folio->mapping = mapping; | |
872 | folio->index = xas.xa_index; | |
198b62f8 | 873 | |
74d60958 | 874 | do { |
198b62f8 MWO |
875 | unsigned int order = xa_get_order(xas.xa, xas.xa_index); |
876 | void *entry, *old = NULL; | |
877 | ||
9dd3d069 | 878 | if (order > folio_order(folio)) |
198b62f8 MWO |
879 | xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index), |
880 | order, gfp); | |
74d60958 | 881 | xas_lock_irq(&xas); |
198b62f8 MWO |
882 | xas_for_each_conflict(&xas, entry) { |
883 | old = entry; | |
884 | if (!xa_is_value(entry)) { | |
885 | xas_set_err(&xas, -EEXIST); | |
886 | goto unlock; | |
887 | } | |
888 | } | |
889 | ||
890 | if (old) { | |
891 | if (shadowp) | |
892 | *shadowp = old; | |
893 | /* entry may have been split before we acquired lock */ | |
894 | order = xa_get_order(xas.xa, xas.xa_index); | |
9dd3d069 | 895 | if (order > folio_order(folio)) { |
d68eccad MWO |
896 | /* How to handle large swap entries? */ |
897 | BUG_ON(shmem_mapping(mapping)); | |
198b62f8 MWO |
898 | xas_split(&xas, old, order); |
899 | xas_reset(&xas); | |
900 | } | |
901 | } | |
902 | ||
9dd3d069 | 903 | xas_store(&xas, folio); |
74d60958 MW |
904 | if (xas_error(&xas)) |
905 | goto unlock; | |
906 | ||
d68eccad | 907 | mapping->nrpages += nr; |
74d60958 MW |
908 | |
909 | /* hugetlb pages do not participate in page cache accounting */ | |
d68eccad MWO |
910 | if (!huge) { |
911 | __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr); | |
912 | if (folio_test_pmd_mappable(folio)) | |
913 | __lruvec_stat_mod_folio(folio, | |
914 | NR_FILE_THPS, nr); | |
915 | } | |
74d60958 MW |
916 | unlock: |
917 | xas_unlock_irq(&xas); | |
198b62f8 | 918 | } while (xas_nomem(&xas, gfp)); |
74d60958 | 919 | |
d68eccad | 920 | if (xas_error(&xas)) |
74d60958 | 921 | goto error; |
4165b9b4 | 922 | |
a0580c6f | 923 | trace_mm_filemap_add_to_page_cache(folio); |
66a0c8ee | 924 | return 0; |
74d60958 | 925 | error: |
d68eccad MWO |
926 | if (charged) |
927 | mem_cgroup_uncharge(folio); | |
9dd3d069 | 928 | folio->mapping = NULL; |
66a0c8ee | 929 | /* Leave page->index set: truncation relies upon it */ |
d68eccad MWO |
930 | folio_put_refs(folio, nr); |
931 | return xas_error(&xas); | |
1da177e4 | 932 | } |
9dd3d069 | 933 | ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO); |
a528910e | 934 | |
9dd3d069 MWO |
935 | int filemap_add_folio(struct address_space *mapping, struct folio *folio, |
936 | pgoff_t index, gfp_t gfp) | |
1da177e4 | 937 | { |
a528910e | 938 | void *shadow = NULL; |
4f98a2fe RR |
939 | int ret; |
940 | ||
9dd3d069 MWO |
941 | __folio_set_locked(folio); |
942 | ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow); | |
a528910e | 943 | if (unlikely(ret)) |
9dd3d069 | 944 | __folio_clear_locked(folio); |
a528910e JW |
945 | else { |
946 | /* | |
9dd3d069 | 947 | * The folio might have been evicted from cache only |
a528910e | 948 | * recently, in which case it should be activated like |
9dd3d069 MWO |
949 | * any other repeatedly accessed folio. |
950 | * The exception is folios getting rewritten; evicting other | |
f0281a00 RR |
951 | * data from the working set, only to cache data that will |
952 | * get overwritten with something else, is a waste of memory. | |
a528910e | 953 | */ |
9dd3d069 MWO |
954 | WARN_ON_ONCE(folio_test_active(folio)); |
955 | if (!(gfp & __GFP_WRITE) && shadow) | |
956 | workingset_refault(folio, shadow); | |
957 | folio_add_lru(folio); | |
a528910e | 958 | } |
1da177e4 LT |
959 | return ret; |
960 | } | |
9dd3d069 | 961 | EXPORT_SYMBOL_GPL(filemap_add_folio); |
1da177e4 | 962 | |
44110fe3 | 963 | #ifdef CONFIG_NUMA |
bb3c579e | 964 | struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order) |
44110fe3 | 965 | { |
c0ff7453 | 966 | int n; |
bb3c579e | 967 | struct folio *folio; |
c0ff7453 | 968 | |
44110fe3 | 969 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
970 | unsigned int cpuset_mems_cookie; |
971 | do { | |
d26914d1 | 972 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 973 | n = cpuset_mem_spread_node(); |
bb3c579e MWO |
974 | folio = __folio_alloc_node(gfp, order, n); |
975 | } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie)); | |
cc9a6c87 | 976 | |
bb3c579e | 977 | return folio; |
44110fe3 | 978 | } |
bb3c579e | 979 | return folio_alloc(gfp, order); |
44110fe3 | 980 | } |
bb3c579e | 981 | EXPORT_SYMBOL(filemap_alloc_folio); |
44110fe3 PJ |
982 | #endif |
983 | ||
7506ae6a JK |
984 | /* |
985 | * filemap_invalidate_lock_two - lock invalidate_lock for two mappings | |
986 | * | |
987 | * Lock exclusively invalidate_lock of any passed mapping that is not NULL. | |
988 | * | |
989 | * @mapping1: the first mapping to lock | |
990 | * @mapping2: the second mapping to lock | |
991 | */ | |
992 | void filemap_invalidate_lock_two(struct address_space *mapping1, | |
993 | struct address_space *mapping2) | |
994 | { | |
995 | if (mapping1 > mapping2) | |
996 | swap(mapping1, mapping2); | |
997 | if (mapping1) | |
998 | down_write(&mapping1->invalidate_lock); | |
999 | if (mapping2 && mapping1 != mapping2) | |
1000 | down_write_nested(&mapping2->invalidate_lock, 1); | |
1001 | } | |
1002 | EXPORT_SYMBOL(filemap_invalidate_lock_two); | |
1003 | ||
1004 | /* | |
1005 | * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings | |
1006 | * | |
1007 | * Unlock exclusive invalidate_lock of any passed mapping that is not NULL. | |
1008 | * | |
1009 | * @mapping1: the first mapping to unlock | |
1010 | * @mapping2: the second mapping to unlock | |
1011 | */ | |
1012 | void filemap_invalidate_unlock_two(struct address_space *mapping1, | |
1013 | struct address_space *mapping2) | |
1014 | { | |
1015 | if (mapping1) | |
1016 | up_write(&mapping1->invalidate_lock); | |
1017 | if (mapping2 && mapping1 != mapping2) | |
1018 | up_write(&mapping2->invalidate_lock); | |
1019 | } | |
1020 | EXPORT_SYMBOL(filemap_invalidate_unlock_two); | |
1021 | ||
1da177e4 LT |
1022 | /* |
1023 | * In order to wait for pages to become available there must be | |
1024 | * waitqueues associated with pages. By using a hash table of | |
1025 | * waitqueues where the bucket discipline is to maintain all | |
1026 | * waiters on the same queue and wake all when any of the pages | |
1027 | * become available, and for the woken contexts to check to be | |
1028 | * sure the appropriate page became available, this saves space | |
1029 | * at a cost of "thundering herd" phenomena during rare hash | |
1030 | * collisions. | |
1031 | */ | |
62906027 NP |
1032 | #define PAGE_WAIT_TABLE_BITS 8 |
1033 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
df4d4f12 | 1034 | static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; |
62906027 | 1035 | |
df4d4f12 | 1036 | static wait_queue_head_t *folio_waitqueue(struct folio *folio) |
1da177e4 | 1037 | { |
df4d4f12 | 1038 | return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 1039 | } |
1da177e4 | 1040 | |
62906027 | 1041 | void __init pagecache_init(void) |
1da177e4 | 1042 | { |
62906027 | 1043 | int i; |
1da177e4 | 1044 | |
62906027 | 1045 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
df4d4f12 | 1046 | init_waitqueue_head(&folio_wait_table[i]); |
62906027 NP |
1047 | |
1048 | page_writeback_init(); | |
1da177e4 | 1049 | } |
1da177e4 | 1050 | |
5ef64cc8 LT |
1051 | /* |
1052 | * The page wait code treats the "wait->flags" somewhat unusually, because | |
5868ec26 | 1053 | * we have multiple different kinds of waits, not just the usual "exclusive" |
5ef64cc8 LT |
1054 | * one. |
1055 | * | |
1056 | * We have: | |
1057 | * | |
1058 | * (a) no special bits set: | |
1059 | * | |
1060 | * We're just waiting for the bit to be released, and when a waker | |
1061 | * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up, | |
1062 | * and remove it from the wait queue. | |
1063 | * | |
1064 | * Simple and straightforward. | |
1065 | * | |
1066 | * (b) WQ_FLAG_EXCLUSIVE: | |
1067 | * | |
1068 | * The waiter is waiting to get the lock, and only one waiter should | |
1069 | * be woken up to avoid any thundering herd behavior. We'll set the | |
1070 | * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue. | |
1071 | * | |
1072 | * This is the traditional exclusive wait. | |
1073 | * | |
5868ec26 | 1074 | * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: |
5ef64cc8 LT |
1075 | * |
1076 | * The waiter is waiting to get the bit, and additionally wants the | |
1077 | * lock to be transferred to it for fair lock behavior. If the lock | |
1078 | * cannot be taken, we stop walking the wait queue without waking | |
1079 | * the waiter. | |
1080 | * | |
1081 | * This is the "fair lock handoff" case, and in addition to setting | |
1082 | * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see | |
1083 | * that it now has the lock. | |
1084 | */ | |
ac6424b9 | 1085 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 1086 | { |
5ef64cc8 | 1087 | unsigned int flags; |
62906027 NP |
1088 | struct wait_page_key *key = arg; |
1089 | struct wait_page_queue *wait_page | |
1090 | = container_of(wait, struct wait_page_queue, wait); | |
1091 | ||
cdc8fcb4 | 1092 | if (!wake_page_match(wait_page, key)) |
62906027 | 1093 | return 0; |
3510ca20 | 1094 | |
9a1ea439 | 1095 | /* |
5ef64cc8 LT |
1096 | * If it's a lock handoff wait, we get the bit for it, and |
1097 | * stop walking (and do not wake it up) if we can't. | |
9a1ea439 | 1098 | */ |
5ef64cc8 LT |
1099 | flags = wait->flags; |
1100 | if (flags & WQ_FLAG_EXCLUSIVE) { | |
df4d4f12 | 1101 | if (test_bit(key->bit_nr, &key->folio->flags)) |
2a9127fc | 1102 | return -1; |
5ef64cc8 | 1103 | if (flags & WQ_FLAG_CUSTOM) { |
df4d4f12 | 1104 | if (test_and_set_bit(key->bit_nr, &key->folio->flags)) |
5ef64cc8 LT |
1105 | return -1; |
1106 | flags |= WQ_FLAG_DONE; | |
1107 | } | |
2a9127fc | 1108 | } |
f62e00cc | 1109 | |
5ef64cc8 LT |
1110 | /* |
1111 | * We are holding the wait-queue lock, but the waiter that | |
1112 | * is waiting for this will be checking the flags without | |
1113 | * any locking. | |
1114 | * | |
1115 | * So update the flags atomically, and wake up the waiter | |
1116 | * afterwards to avoid any races. This store-release pairs | |
101c0bf6 | 1117 | * with the load-acquire in folio_wait_bit_common(). |
5ef64cc8 LT |
1118 | */ |
1119 | smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN); | |
2a9127fc LT |
1120 | wake_up_state(wait->private, mode); |
1121 | ||
1122 | /* | |
1123 | * Ok, we have successfully done what we're waiting for, | |
1124 | * and we can unconditionally remove the wait entry. | |
1125 | * | |
5ef64cc8 LT |
1126 | * Note that this pairs with the "finish_wait()" in the |
1127 | * waiter, and has to be the absolute last thing we do. | |
1128 | * After this list_del_init(&wait->entry) the wait entry | |
2a9127fc LT |
1129 | * might be de-allocated and the process might even have |
1130 | * exited. | |
2a9127fc | 1131 | */ |
c6fe44d9 | 1132 | list_del_init_careful(&wait->entry); |
5ef64cc8 | 1133 | return (flags & WQ_FLAG_EXCLUSIVE) != 0; |
f62e00cc KM |
1134 | } |
1135 | ||
6974d7c9 | 1136 | static void folio_wake_bit(struct folio *folio, int bit_nr) |
cbbce822 | 1137 | { |
df4d4f12 | 1138 | wait_queue_head_t *q = folio_waitqueue(folio); |
62906027 NP |
1139 | struct wait_page_key key; |
1140 | unsigned long flags; | |
11a19c7b | 1141 | wait_queue_entry_t bookmark; |
cbbce822 | 1142 | |
df4d4f12 | 1143 | key.folio = folio; |
62906027 NP |
1144 | key.bit_nr = bit_nr; |
1145 | key.page_match = 0; | |
1146 | ||
11a19c7b TC |
1147 | bookmark.flags = 0; |
1148 | bookmark.private = NULL; | |
1149 | bookmark.func = NULL; | |
1150 | INIT_LIST_HEAD(&bookmark.entry); | |
1151 | ||
62906027 | 1152 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1153 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1154 | ||
1155 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1156 | /* | |
1157 | * Take a breather from holding the lock, | |
1158 | * allow pages that finish wake up asynchronously | |
1159 | * to acquire the lock and remove themselves | |
1160 | * from wait queue | |
1161 | */ | |
1162 | spin_unlock_irqrestore(&q->lock, flags); | |
1163 | cpu_relax(); | |
1164 | spin_lock_irqsave(&q->lock, flags); | |
1165 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1166 | } | |
1167 | ||
62906027 | 1168 | /* |
bb43b14b HD |
1169 | * It's possible to miss clearing waiters here, when we woke our page |
1170 | * waiters, but the hashed waitqueue has waiters for other pages on it. | |
1171 | * That's okay, it's a rare case. The next waker will clear it. | |
62906027 | 1172 | * |
bb43b14b HD |
1173 | * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE, |
1174 | * other), the flag may be cleared in the course of freeing the page; | |
1175 | * but that is not required for correctness. | |
62906027 | 1176 | */ |
bb43b14b | 1177 | if (!waitqueue_active(q) || !key.page_match) |
6974d7c9 | 1178 | folio_clear_waiters(folio); |
bb43b14b | 1179 | |
62906027 NP |
1180 | spin_unlock_irqrestore(&q->lock, flags); |
1181 | } | |
74d81bfa | 1182 | |
4268b480 | 1183 | static void folio_wake(struct folio *folio, int bit) |
74d81bfa | 1184 | { |
4268b480 | 1185 | if (!folio_test_waiters(folio)) |
74d81bfa | 1186 | return; |
6974d7c9 | 1187 | folio_wake_bit(folio, bit); |
74d81bfa | 1188 | } |
62906027 | 1189 | |
9a1ea439 | 1190 | /* |
101c0bf6 | 1191 | * A choice of three behaviors for folio_wait_bit_common(): |
9a1ea439 HD |
1192 | */ |
1193 | enum behavior { | |
1194 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like | |
7c23c782 | 1195 | * __folio_lock() waiting on then setting PG_locked. |
9a1ea439 HD |
1196 | */ |
1197 | SHARED, /* Hold ref to page and check the bit when woken, like | |
9f2b04a2 | 1198 | * folio_wait_writeback() waiting on PG_writeback. |
9a1ea439 HD |
1199 | */ |
1200 | DROP, /* Drop ref to page before wait, no check when woken, | |
9f2b04a2 | 1201 | * like folio_put_wait_locked() on PG_locked. |
9a1ea439 HD |
1202 | */ |
1203 | }; | |
1204 | ||
2a9127fc | 1205 | /* |
101c0bf6 | 1206 | * Attempt to check (or get) the folio flag, and mark us done |
5ef64cc8 | 1207 | * if successful. |
2a9127fc | 1208 | */ |
101c0bf6 | 1209 | static inline bool folio_trylock_flag(struct folio *folio, int bit_nr, |
2a9127fc LT |
1210 | struct wait_queue_entry *wait) |
1211 | { | |
1212 | if (wait->flags & WQ_FLAG_EXCLUSIVE) { | |
101c0bf6 | 1213 | if (test_and_set_bit(bit_nr, &folio->flags)) |
2a9127fc | 1214 | return false; |
101c0bf6 | 1215 | } else if (test_bit(bit_nr, &folio->flags)) |
2a9127fc LT |
1216 | return false; |
1217 | ||
5ef64cc8 | 1218 | wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE; |
2a9127fc LT |
1219 | return true; |
1220 | } | |
1221 | ||
5ef64cc8 LT |
1222 | /* How many times do we accept lock stealing from under a waiter? */ |
1223 | int sysctl_page_lock_unfairness = 5; | |
1224 | ||
101c0bf6 MWO |
1225 | static inline int folio_wait_bit_common(struct folio *folio, int bit_nr, |
1226 | int state, enum behavior behavior) | |
62906027 | 1227 | { |
df4d4f12 | 1228 | wait_queue_head_t *q = folio_waitqueue(folio); |
5ef64cc8 | 1229 | int unfairness = sysctl_page_lock_unfairness; |
62906027 | 1230 | struct wait_page_queue wait_page; |
ac6424b9 | 1231 | wait_queue_entry_t *wait = &wait_page.wait; |
b1d29ba8 | 1232 | bool thrashing = false; |
eb414681 | 1233 | unsigned long pflags; |
aa1cf99b | 1234 | bool in_thrashing; |
62906027 | 1235 | |
eb414681 | 1236 | if (bit_nr == PG_locked && |
101c0bf6 | 1237 | !folio_test_uptodate(folio) && folio_test_workingset(folio)) { |
aa1cf99b | 1238 | delayacct_thrashing_start(&in_thrashing); |
eb414681 | 1239 | psi_memstall_enter(&pflags); |
b1d29ba8 JW |
1240 | thrashing = true; |
1241 | } | |
1242 | ||
62906027 NP |
1243 | init_wait(wait); |
1244 | wait->func = wake_page_function; | |
df4d4f12 | 1245 | wait_page.folio = folio; |
62906027 NP |
1246 | wait_page.bit_nr = bit_nr; |
1247 | ||
5ef64cc8 LT |
1248 | repeat: |
1249 | wait->flags = 0; | |
1250 | if (behavior == EXCLUSIVE) { | |
1251 | wait->flags = WQ_FLAG_EXCLUSIVE; | |
1252 | if (--unfairness < 0) | |
1253 | wait->flags |= WQ_FLAG_CUSTOM; | |
1254 | } | |
1255 | ||
2a9127fc LT |
1256 | /* |
1257 | * Do one last check whether we can get the | |
1258 | * page bit synchronously. | |
1259 | * | |
101c0bf6 | 1260 | * Do the folio_set_waiters() marking before that |
2a9127fc LT |
1261 | * to let any waker we _just_ missed know they |
1262 | * need to wake us up (otherwise they'll never | |
1263 | * even go to the slow case that looks at the | |
1264 | * page queue), and add ourselves to the wait | |
1265 | * queue if we need to sleep. | |
1266 | * | |
1267 | * This part needs to be done under the queue | |
1268 | * lock to avoid races. | |
1269 | */ | |
1270 | spin_lock_irq(&q->lock); | |
101c0bf6 MWO |
1271 | folio_set_waiters(folio); |
1272 | if (!folio_trylock_flag(folio, bit_nr, wait)) | |
2a9127fc LT |
1273 | __add_wait_queue_entry_tail(q, wait); |
1274 | spin_unlock_irq(&q->lock); | |
62906027 | 1275 | |
2a9127fc LT |
1276 | /* |
1277 | * From now on, all the logic will be based on | |
5ef64cc8 LT |
1278 | * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to |
1279 | * see whether the page bit testing has already | |
1280 | * been done by the wake function. | |
2a9127fc | 1281 | * |
101c0bf6 | 1282 | * We can drop our reference to the folio. |
2a9127fc LT |
1283 | */ |
1284 | if (behavior == DROP) | |
101c0bf6 | 1285 | folio_put(folio); |
62906027 | 1286 | |
5ef64cc8 LT |
1287 | /* |
1288 | * Note that until the "finish_wait()", or until | |
1289 | * we see the WQ_FLAG_WOKEN flag, we need to | |
1290 | * be very careful with the 'wait->flags', because | |
1291 | * we may race with a waker that sets them. | |
1292 | */ | |
2a9127fc | 1293 | for (;;) { |
5ef64cc8 LT |
1294 | unsigned int flags; |
1295 | ||
62906027 NP |
1296 | set_current_state(state); |
1297 | ||
5ef64cc8 LT |
1298 | /* Loop until we've been woken or interrupted */ |
1299 | flags = smp_load_acquire(&wait->flags); | |
1300 | if (!(flags & WQ_FLAG_WOKEN)) { | |
1301 | if (signal_pending_state(state, current)) | |
1302 | break; | |
1303 | ||
1304 | io_schedule(); | |
1305 | continue; | |
1306 | } | |
1307 | ||
1308 | /* If we were non-exclusive, we're done */ | |
1309 | if (behavior != EXCLUSIVE) | |
a8b169af | 1310 | break; |
9a1ea439 | 1311 | |
5ef64cc8 LT |
1312 | /* If the waker got the lock for us, we're done */ |
1313 | if (flags & WQ_FLAG_DONE) | |
9a1ea439 | 1314 | break; |
2a9127fc | 1315 | |
5ef64cc8 LT |
1316 | /* |
1317 | * Otherwise, if we're getting the lock, we need to | |
1318 | * try to get it ourselves. | |
1319 | * | |
1320 | * And if that fails, we'll have to retry this all. | |
1321 | */ | |
101c0bf6 | 1322 | if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0)))) |
5ef64cc8 LT |
1323 | goto repeat; |
1324 | ||
1325 | wait->flags |= WQ_FLAG_DONE; | |
1326 | break; | |
62906027 NP |
1327 | } |
1328 | ||
5ef64cc8 LT |
1329 | /* |
1330 | * If a signal happened, this 'finish_wait()' may remove the last | |
101c0bf6 | 1331 | * waiter from the wait-queues, but the folio waiters bit will remain |
5ef64cc8 LT |
1332 | * set. That's ok. The next wakeup will take care of it, and trying |
1333 | * to do it here would be difficult and prone to races. | |
1334 | */ | |
62906027 NP |
1335 | finish_wait(q, wait); |
1336 | ||
eb414681 | 1337 | if (thrashing) { |
aa1cf99b | 1338 | delayacct_thrashing_end(&in_thrashing); |
eb414681 JW |
1339 | psi_memstall_leave(&pflags); |
1340 | } | |
b1d29ba8 | 1341 | |
62906027 | 1342 | /* |
5ef64cc8 LT |
1343 | * NOTE! The wait->flags weren't stable until we've done the |
1344 | * 'finish_wait()', and we could have exited the loop above due | |
1345 | * to a signal, and had a wakeup event happen after the signal | |
1346 | * test but before the 'finish_wait()'. | |
1347 | * | |
1348 | * So only after the finish_wait() can we reliably determine | |
1349 | * if we got woken up or not, so we can now figure out the final | |
1350 | * return value based on that state without races. | |
1351 | * | |
1352 | * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive | |
1353 | * waiter, but an exclusive one requires WQ_FLAG_DONE. | |
62906027 | 1354 | */ |
5ef64cc8 LT |
1355 | if (behavior == EXCLUSIVE) |
1356 | return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR; | |
62906027 | 1357 | |
2a9127fc | 1358 | return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR; |
62906027 NP |
1359 | } |
1360 | ||
ffa65753 AP |
1361 | #ifdef CONFIG_MIGRATION |
1362 | /** | |
1363 | * migration_entry_wait_on_locked - Wait for a migration entry to be removed | |
1364 | * @entry: migration swap entry. | |
1365 | * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required | |
1366 | * for pte entries, pass NULL for pmd entries. | |
1367 | * @ptl: already locked ptl. This function will drop the lock. | |
1368 | * | |
1369 | * Wait for a migration entry referencing the given page to be removed. This is | |
1370 | * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except | |
1371 | * this can be called without taking a reference on the page. Instead this | |
1372 | * should be called while holding the ptl for the migration entry referencing | |
1373 | * the page. | |
1374 | * | |
1375 | * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock(). | |
1376 | * | |
1377 | * This follows the same logic as folio_wait_bit_common() so see the comments | |
1378 | * there. | |
1379 | */ | |
1380 | void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep, | |
1381 | spinlock_t *ptl) | |
1382 | { | |
1383 | struct wait_page_queue wait_page; | |
1384 | wait_queue_entry_t *wait = &wait_page.wait; | |
1385 | bool thrashing = false; | |
ffa65753 | 1386 | unsigned long pflags; |
aa1cf99b | 1387 | bool in_thrashing; |
ffa65753 AP |
1388 | wait_queue_head_t *q; |
1389 | struct folio *folio = page_folio(pfn_swap_entry_to_page(entry)); | |
1390 | ||
1391 | q = folio_waitqueue(folio); | |
1392 | if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) { | |
aa1cf99b | 1393 | delayacct_thrashing_start(&in_thrashing); |
ffa65753 AP |
1394 | psi_memstall_enter(&pflags); |
1395 | thrashing = true; | |
1396 | } | |
1397 | ||
1398 | init_wait(wait); | |
1399 | wait->func = wake_page_function; | |
1400 | wait_page.folio = folio; | |
1401 | wait_page.bit_nr = PG_locked; | |
1402 | wait->flags = 0; | |
1403 | ||
1404 | spin_lock_irq(&q->lock); | |
1405 | folio_set_waiters(folio); | |
1406 | if (!folio_trylock_flag(folio, PG_locked, wait)) | |
1407 | __add_wait_queue_entry_tail(q, wait); | |
1408 | spin_unlock_irq(&q->lock); | |
1409 | ||
1410 | /* | |
1411 | * If a migration entry exists for the page the migration path must hold | |
1412 | * a valid reference to the page, and it must take the ptl to remove the | |
1413 | * migration entry. So the page is valid until the ptl is dropped. | |
1414 | */ | |
1415 | if (ptep) | |
1416 | pte_unmap_unlock(ptep, ptl); | |
1417 | else | |
1418 | spin_unlock(ptl); | |
1419 | ||
1420 | for (;;) { | |
1421 | unsigned int flags; | |
1422 | ||
1423 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1424 | ||
1425 | /* Loop until we've been woken or interrupted */ | |
1426 | flags = smp_load_acquire(&wait->flags); | |
1427 | if (!(flags & WQ_FLAG_WOKEN)) { | |
1428 | if (signal_pending_state(TASK_UNINTERRUPTIBLE, current)) | |
1429 | break; | |
1430 | ||
1431 | io_schedule(); | |
1432 | continue; | |
1433 | } | |
1434 | break; | |
1435 | } | |
1436 | ||
1437 | finish_wait(q, wait); | |
1438 | ||
1439 | if (thrashing) { | |
aa1cf99b | 1440 | delayacct_thrashing_end(&in_thrashing); |
ffa65753 AP |
1441 | psi_memstall_leave(&pflags); |
1442 | } | |
1443 | } | |
1444 | #endif | |
1445 | ||
101c0bf6 | 1446 | void folio_wait_bit(struct folio *folio, int bit_nr) |
62906027 | 1447 | { |
101c0bf6 | 1448 | folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
62906027 | 1449 | } |
101c0bf6 | 1450 | EXPORT_SYMBOL(folio_wait_bit); |
62906027 | 1451 | |
101c0bf6 | 1452 | int folio_wait_bit_killable(struct folio *folio, int bit_nr) |
62906027 | 1453 | { |
101c0bf6 | 1454 | return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED); |
cbbce822 | 1455 | } |
101c0bf6 | 1456 | EXPORT_SYMBOL(folio_wait_bit_killable); |
cbbce822 | 1457 | |
9a1ea439 | 1458 | /** |
9f2b04a2 MWO |
1459 | * folio_put_wait_locked - Drop a reference and wait for it to be unlocked |
1460 | * @folio: The folio to wait for. | |
48054625 | 1461 | * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc). |
9a1ea439 | 1462 | * |
9f2b04a2 | 1463 | * The caller should hold a reference on @folio. They expect the page to |
9a1ea439 | 1464 | * become unlocked relatively soon, but do not wish to hold up migration |
9f2b04a2 | 1465 | * (for example) by holding the reference while waiting for the folio to |
9a1ea439 | 1466 | * come unlocked. After this function returns, the caller should not |
9f2b04a2 | 1467 | * dereference @folio. |
48054625 | 1468 | * |
9f2b04a2 | 1469 | * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal. |
9a1ea439 | 1470 | */ |
c195c321 | 1471 | static int folio_put_wait_locked(struct folio *folio, int state) |
9a1ea439 | 1472 | { |
9f2b04a2 | 1473 | return folio_wait_bit_common(folio, PG_locked, state, DROP); |
9a1ea439 HD |
1474 | } |
1475 | ||
385e1ca5 | 1476 | /** |
df4d4f12 MWO |
1477 | * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue |
1478 | * @folio: Folio defining the wait queue of interest | |
697f619f | 1479 | * @waiter: Waiter to add to the queue |
385e1ca5 | 1480 | * |
df4d4f12 | 1481 | * Add an arbitrary @waiter to the wait queue for the nominated @folio. |
385e1ca5 | 1482 | */ |
df4d4f12 | 1483 | void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter) |
385e1ca5 | 1484 | { |
df4d4f12 | 1485 | wait_queue_head_t *q = folio_waitqueue(folio); |
385e1ca5 DH |
1486 | unsigned long flags; |
1487 | ||
1488 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1489 | __add_wait_queue_entry_tail(q, waiter); |
df4d4f12 | 1490 | folio_set_waiters(folio); |
385e1ca5 DH |
1491 | spin_unlock_irqrestore(&q->lock, flags); |
1492 | } | |
df4d4f12 | 1493 | EXPORT_SYMBOL_GPL(folio_add_wait_queue); |
385e1ca5 | 1494 | |
b91e1302 LT |
1495 | #ifndef clear_bit_unlock_is_negative_byte |
1496 | ||
1497 | /* | |
1498 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1499 | * | |
1500 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1501 | * test the sign bit at the same time. But if the architecture does | |
1502 | * not support that special operation, we just do this all by hand | |
1503 | * instead. | |
1504 | * | |
1505 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
ffceeb62 | 1506 | * being cleared, but a memory barrier should be unnecessary since it is |
b91e1302 LT |
1507 | * in the same byte as PG_locked. |
1508 | */ | |
1509 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1510 | { | |
1511 | clear_bit_unlock(nr, mem); | |
1512 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1513 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1514 | } |
1515 | ||
1516 | #endif | |
1517 | ||
1da177e4 | 1518 | /** |
4e136428 MWO |
1519 | * folio_unlock - Unlock a locked folio. |
1520 | * @folio: The folio. | |
1521 | * | |
1522 | * Unlocks the folio and wakes up any thread sleeping on the page lock. | |
1523 | * | |
1524 | * Context: May be called from interrupt or process context. May not be | |
1525 | * called from NMI context. | |
1da177e4 | 1526 | */ |
4e136428 | 1527 | void folio_unlock(struct folio *folio) |
1da177e4 | 1528 | { |
4e136428 | 1529 | /* Bit 7 allows x86 to check the byte's sign bit */ |
b91e1302 | 1530 | BUILD_BUG_ON(PG_waiters != 7); |
4e136428 MWO |
1531 | BUILD_BUG_ON(PG_locked > 7); |
1532 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); | |
1533 | if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0))) | |
6974d7c9 | 1534 | folio_wake_bit(folio, PG_locked); |
1da177e4 | 1535 | } |
4e136428 | 1536 | EXPORT_SYMBOL(folio_unlock); |
1da177e4 | 1537 | |
73e10ded | 1538 | /** |
b47393f8 MWO |
1539 | * folio_end_private_2 - Clear PG_private_2 and wake any waiters. |
1540 | * @folio: The folio. | |
73e10ded | 1541 | * |
b47393f8 MWO |
1542 | * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for |
1543 | * it. The folio reference held for PG_private_2 being set is released. | |
73e10ded | 1544 | * |
b47393f8 MWO |
1545 | * This is, for example, used when a netfs folio is being written to a local |
1546 | * disk cache, thereby allowing writes to the cache for the same folio to be | |
73e10ded DH |
1547 | * serialised. |
1548 | */ | |
b47393f8 | 1549 | void folio_end_private_2(struct folio *folio) |
73e10ded | 1550 | { |
6974d7c9 MWO |
1551 | VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio); |
1552 | clear_bit_unlock(PG_private_2, folio_flags(folio, 0)); | |
1553 | folio_wake_bit(folio, PG_private_2); | |
1554 | folio_put(folio); | |
73e10ded | 1555 | } |
b47393f8 | 1556 | EXPORT_SYMBOL(folio_end_private_2); |
73e10ded DH |
1557 | |
1558 | /** | |
b47393f8 MWO |
1559 | * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio. |
1560 | * @folio: The folio to wait on. | |
73e10ded | 1561 | * |
b47393f8 | 1562 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio. |
73e10ded | 1563 | */ |
b47393f8 | 1564 | void folio_wait_private_2(struct folio *folio) |
73e10ded | 1565 | { |
101c0bf6 MWO |
1566 | while (folio_test_private_2(folio)) |
1567 | folio_wait_bit(folio, PG_private_2); | |
73e10ded | 1568 | } |
b47393f8 | 1569 | EXPORT_SYMBOL(folio_wait_private_2); |
73e10ded DH |
1570 | |
1571 | /** | |
b47393f8 MWO |
1572 | * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio. |
1573 | * @folio: The folio to wait on. | |
73e10ded | 1574 | * |
b47393f8 | 1575 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a |
73e10ded DH |
1576 | * fatal signal is received by the calling task. |
1577 | * | |
1578 | * Return: | |
1579 | * - 0 if successful. | |
1580 | * - -EINTR if a fatal signal was encountered. | |
1581 | */ | |
b47393f8 | 1582 | int folio_wait_private_2_killable(struct folio *folio) |
73e10ded DH |
1583 | { |
1584 | int ret = 0; | |
1585 | ||
101c0bf6 MWO |
1586 | while (folio_test_private_2(folio)) { |
1587 | ret = folio_wait_bit_killable(folio, PG_private_2); | |
73e10ded DH |
1588 | if (ret < 0) |
1589 | break; | |
1590 | } | |
1591 | ||
1592 | return ret; | |
1593 | } | |
b47393f8 | 1594 | EXPORT_SYMBOL(folio_wait_private_2_killable); |
73e10ded | 1595 | |
485bb99b | 1596 | /** |
4268b480 MWO |
1597 | * folio_end_writeback - End writeback against a folio. |
1598 | * @folio: The folio. | |
1da177e4 | 1599 | */ |
4268b480 | 1600 | void folio_end_writeback(struct folio *folio) |
1da177e4 | 1601 | { |
888cf2db | 1602 | /* |
4268b480 MWO |
1603 | * folio_test_clear_reclaim() could be used here but it is an |
1604 | * atomic operation and overkill in this particular case. Failing | |
1605 | * to shuffle a folio marked for immediate reclaim is too mild | |
1606 | * a gain to justify taking an atomic operation penalty at the | |
1607 | * end of every folio writeback. | |
888cf2db | 1608 | */ |
4268b480 MWO |
1609 | if (folio_test_reclaim(folio)) { |
1610 | folio_clear_reclaim(folio); | |
575ced1c | 1611 | folio_rotate_reclaimable(folio); |
888cf2db | 1612 | } |
ac6aadb2 | 1613 | |
073861ed | 1614 | /* |
4268b480 | 1615 | * Writeback does not hold a folio reference of its own, relying |
073861ed | 1616 | * on truncation to wait for the clearing of PG_writeback. |
4268b480 MWO |
1617 | * But here we must make sure that the folio is not freed and |
1618 | * reused before the folio_wake(). | |
073861ed | 1619 | */ |
4268b480 | 1620 | folio_get(folio); |
269ccca3 | 1621 | if (!__folio_end_writeback(folio)) |
ac6aadb2 MS |
1622 | BUG(); |
1623 | ||
4e857c58 | 1624 | smp_mb__after_atomic(); |
4268b480 | 1625 | folio_wake(folio, PG_writeback); |
512b7931 | 1626 | acct_reclaim_writeback(folio); |
4268b480 | 1627 | folio_put(folio); |
1da177e4 | 1628 | } |
4268b480 | 1629 | EXPORT_SYMBOL(folio_end_writeback); |
1da177e4 | 1630 | |
485bb99b | 1631 | /** |
7c23c782 MWO |
1632 | * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it. |
1633 | * @folio: The folio to lock | |
1da177e4 | 1634 | */ |
7c23c782 | 1635 | void __folio_lock(struct folio *folio) |
1da177e4 | 1636 | { |
101c0bf6 | 1637 | folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE, |
9a1ea439 | 1638 | EXCLUSIVE); |
1da177e4 | 1639 | } |
7c23c782 | 1640 | EXPORT_SYMBOL(__folio_lock); |
1da177e4 | 1641 | |
af7f29d9 | 1642 | int __folio_lock_killable(struct folio *folio) |
2687a356 | 1643 | { |
101c0bf6 | 1644 | return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE, |
9a1ea439 | 1645 | EXCLUSIVE); |
2687a356 | 1646 | } |
af7f29d9 | 1647 | EXPORT_SYMBOL_GPL(__folio_lock_killable); |
2687a356 | 1648 | |
ffdc8dab | 1649 | static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait) |
dd3e6d50 | 1650 | { |
df4d4f12 | 1651 | struct wait_queue_head *q = folio_waitqueue(folio); |
f32b5dd7 MWO |
1652 | int ret = 0; |
1653 | ||
df4d4f12 | 1654 | wait->folio = folio; |
f32b5dd7 MWO |
1655 | wait->bit_nr = PG_locked; |
1656 | ||
1657 | spin_lock_irq(&q->lock); | |
1658 | __add_wait_queue_entry_tail(q, &wait->wait); | |
ffdc8dab MWO |
1659 | folio_set_waiters(folio); |
1660 | ret = !folio_trylock(folio); | |
f32b5dd7 MWO |
1661 | /* |
1662 | * If we were successful now, we know we're still on the | |
1663 | * waitqueue as we're still under the lock. This means it's | |
1664 | * safe to remove and return success, we know the callback | |
1665 | * isn't going to trigger. | |
1666 | */ | |
1667 | if (!ret) | |
1668 | __remove_wait_queue(q, &wait->wait); | |
1669 | else | |
1670 | ret = -EIOCBQUEUED; | |
1671 | spin_unlock_irq(&q->lock); | |
1672 | return ret; | |
dd3e6d50 JA |
1673 | } |
1674 | ||
9a95f3cf PC |
1675 | /* |
1676 | * Return values: | |
9138e47e MWO |
1677 | * true - folio is locked; mmap_lock is still held. |
1678 | * false - folio is not locked. | |
3e4e28c5 | 1679 | * mmap_lock has been released (mmap_read_unlock(), unless flags had both |
9a95f3cf | 1680 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
c1e8d7c6 | 1681 | * which case mmap_lock is still held. |
9a95f3cf | 1682 | * |
9138e47e MWO |
1683 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return true |
1684 | * with the folio locked and the mmap_lock unperturbed. | |
9a95f3cf | 1685 | */ |
9138e47e | 1686 | bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm, |
d065bd81 ML |
1687 | unsigned int flags) |
1688 | { | |
4064b982 | 1689 | if (fault_flag_allow_retry_first(flags)) { |
37b23e05 | 1690 | /* |
c1e8d7c6 | 1691 | * CAUTION! In this case, mmap_lock is not released |
37b23e05 KM |
1692 | * even though return 0. |
1693 | */ | |
1694 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
9138e47e | 1695 | return false; |
37b23e05 | 1696 | |
d8ed45c5 | 1697 | mmap_read_unlock(mm); |
37b23e05 | 1698 | if (flags & FAULT_FLAG_KILLABLE) |
6baa8d60 | 1699 | folio_wait_locked_killable(folio); |
37b23e05 | 1700 | else |
6baa8d60 | 1701 | folio_wait_locked(folio); |
9138e47e | 1702 | return false; |
800bca7c HL |
1703 | } |
1704 | if (flags & FAULT_FLAG_KILLABLE) { | |
9138e47e | 1705 | bool ret; |
37b23e05 | 1706 | |
af7f29d9 | 1707 | ret = __folio_lock_killable(folio); |
800bca7c HL |
1708 | if (ret) { |
1709 | mmap_read_unlock(mm); | |
9138e47e | 1710 | return false; |
800bca7c HL |
1711 | } |
1712 | } else { | |
af7f29d9 | 1713 | __folio_lock(folio); |
d065bd81 | 1714 | } |
800bca7c | 1715 | |
9138e47e | 1716 | return true; |
d065bd81 ML |
1717 | } |
1718 | ||
e7b563bb | 1719 | /** |
0d3f9296 MW |
1720 | * page_cache_next_miss() - Find the next gap in the page cache. |
1721 | * @mapping: Mapping. | |
1722 | * @index: Index. | |
1723 | * @max_scan: Maximum range to search. | |
e7b563bb | 1724 | * |
0d3f9296 MW |
1725 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1726 | * gap with the lowest index. | |
e7b563bb | 1727 | * |
0d3f9296 MW |
1728 | * This function may be called under the rcu_read_lock. However, this will |
1729 | * not atomically search a snapshot of the cache at a single point in time. | |
1730 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1731 | * created at index 10, page_cache_next_miss covering both indices may | |
1732 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1733 | * |
0d3f9296 MW |
1734 | * Return: The index of the gap if found, otherwise an index outside the |
1735 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1736 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1737 | */ |
0d3f9296 | 1738 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1739 | pgoff_t index, unsigned long max_scan) |
1740 | { | |
0d3f9296 | 1741 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1742 | |
0d3f9296 MW |
1743 | while (max_scan--) { |
1744 | void *entry = xas_next(&xas); | |
1745 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1746 | break; |
0d3f9296 | 1747 | if (xas.xa_index == 0) |
e7b563bb JW |
1748 | break; |
1749 | } | |
1750 | ||
0d3f9296 | 1751 | return xas.xa_index; |
e7b563bb | 1752 | } |
0d3f9296 | 1753 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1754 | |
1755 | /** | |
2346a560 | 1756 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
0d3f9296 MW |
1757 | * @mapping: Mapping. |
1758 | * @index: Index. | |
1759 | * @max_scan: Maximum range to search. | |
e7b563bb | 1760 | * |
0d3f9296 MW |
1761 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1762 | * gap with the highest index. | |
e7b563bb | 1763 | * |
0d3f9296 MW |
1764 | * This function may be called under the rcu_read_lock. However, this will |
1765 | * not atomically search a snapshot of the cache at a single point in time. | |
1766 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1767 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1768 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1769 | * |
0d3f9296 MW |
1770 | * Return: The index of the gap if found, otherwise an index outside the |
1771 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1772 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1773 | */ |
0d3f9296 | 1774 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1775 | pgoff_t index, unsigned long max_scan) |
1776 | { | |
0d3f9296 | 1777 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1778 | |
0d3f9296 MW |
1779 | while (max_scan--) { |
1780 | void *entry = xas_prev(&xas); | |
1781 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1782 | break; |
0d3f9296 | 1783 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1784 | break; |
1785 | } | |
1786 | ||
0d3f9296 | 1787 | return xas.xa_index; |
e7b563bb | 1788 | } |
0d3f9296 | 1789 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1790 | |
020853b6 MWO |
1791 | /* |
1792 | * Lockless page cache protocol: | |
1793 | * On the lookup side: | |
1794 | * 1. Load the folio from i_pages | |
1795 | * 2. Increment the refcount if it's not zero | |
1796 | * 3. If the folio is not found by xas_reload(), put the refcount and retry | |
1797 | * | |
1798 | * On the removal side: | |
1799 | * A. Freeze the page (by zeroing the refcount if nobody else has a reference) | |
1800 | * B. Remove the page from i_pages | |
1801 | * C. Return the page to the page allocator | |
1802 | * | |
1803 | * This means that any page may have its reference count temporarily | |
1804 | * increased by a speculative page cache (or fast GUP) lookup as it can | |
1805 | * be allocated by another user before the RCU grace period expires. | |
1806 | * Because the refcount temporarily acquired here may end up being the | |
1807 | * last refcount on the page, any page allocation must be freeable by | |
1808 | * folio_put(). | |
1809 | */ | |
1810 | ||
44835d20 | 1811 | /* |
263e721e | 1812 | * filemap_get_entry - Get a page cache entry. |
485bb99b | 1813 | * @mapping: the address_space to search |
a6de4b48 | 1814 | * @index: The page cache index. |
0cd6144a | 1815 | * |
bca65eea MWO |
1816 | * Looks up the page cache entry at @mapping & @index. If it is a folio, |
1817 | * it is returned with an increased refcount. If it is a shadow entry | |
1818 | * of a previously evicted folio, or a swap entry from shmem/tmpfs, | |
1819 | * it is returned without further action. | |
485bb99b | 1820 | * |
bca65eea | 1821 | * Return: The folio, swap or shadow entry, %NULL if nothing is found. |
1da177e4 | 1822 | */ |
263e721e | 1823 | void *filemap_get_entry(struct address_space *mapping, pgoff_t index) |
1da177e4 | 1824 | { |
a6de4b48 | 1825 | XA_STATE(xas, &mapping->i_pages, index); |
bca65eea | 1826 | struct folio *folio; |
1da177e4 | 1827 | |
a60637c8 NP |
1828 | rcu_read_lock(); |
1829 | repeat: | |
4c7472c0 | 1830 | xas_reset(&xas); |
bca65eea MWO |
1831 | folio = xas_load(&xas); |
1832 | if (xas_retry(&xas, folio)) | |
4c7472c0 MW |
1833 | goto repeat; |
1834 | /* | |
1835 | * A shadow entry of a recently evicted page, or a swap entry from | |
1836 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1837 | */ | |
bca65eea | 1838 | if (!folio || xa_is_value(folio)) |
4c7472c0 | 1839 | goto out; |
83929372 | 1840 | |
bca65eea | 1841 | if (!folio_try_get_rcu(folio)) |
4c7472c0 | 1842 | goto repeat; |
83929372 | 1843 | |
bca65eea MWO |
1844 | if (unlikely(folio != xas_reload(&xas))) { |
1845 | folio_put(folio); | |
4c7472c0 | 1846 | goto repeat; |
a60637c8 | 1847 | } |
27d20fdd | 1848 | out: |
a60637c8 NP |
1849 | rcu_read_unlock(); |
1850 | ||
bca65eea | 1851 | return folio; |
1da177e4 | 1852 | } |
1da177e4 | 1853 | |
0cd6144a | 1854 | /** |
3f0c6a07 | 1855 | * __filemap_get_folio - Find and get a reference to a folio. |
2294b32e MWO |
1856 | * @mapping: The address_space to search. |
1857 | * @index: The page index. | |
3f0c6a07 MWO |
1858 | * @fgp_flags: %FGP flags modify how the folio is returned. |
1859 | * @gfp: Memory allocation flags to use if %FGP_CREAT is specified. | |
1da177e4 | 1860 | * |
2294b32e | 1861 | * Looks up the page cache entry at @mapping & @index. |
0cd6144a | 1862 | * |
2294b32e | 1863 | * @fgp_flags can be zero or more of these flags: |
0e056eb5 | 1864 | * |
3f0c6a07 MWO |
1865 | * * %FGP_ACCESSED - The folio will be marked accessed. |
1866 | * * %FGP_LOCK - The folio is returned locked. | |
2294b32e | 1867 | * * %FGP_CREAT - If no page is present then a new page is allocated using |
3f0c6a07 | 1868 | * @gfp and added to the page cache and the VM's LRU list. |
2294b32e MWO |
1869 | * The page is returned locked and with an increased refcount. |
1870 | * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the | |
1871 | * page is already in cache. If the page was allocated, unlock it before | |
1872 | * returning so the caller can do the same dance. | |
3f0c6a07 MWO |
1873 | * * %FGP_WRITE - The page will be written to by the caller. |
1874 | * * %FGP_NOFS - __GFP_FS will get cleared in gfp. | |
1875 | * * %FGP_NOWAIT - Don't get blocked by page lock. | |
b27652d9 | 1876 | * * %FGP_STABLE - Wait for the folio to be stable (finished writeback) |
1da177e4 | 1877 | * |
2294b32e MWO |
1878 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even |
1879 | * if the %GFP flags specified for %FGP_CREAT are atomic. | |
1da177e4 | 1880 | * |
2457aec6 | 1881 | * If there is a page cache page, it is returned with an increased refcount. |
a862f68a | 1882 | * |
66dabbb6 | 1883 | * Return: The found folio or an ERR_PTR() otherwise. |
1da177e4 | 1884 | */ |
3f0c6a07 MWO |
1885 | struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index, |
1886 | int fgp_flags, gfp_t gfp) | |
1da177e4 | 1887 | { |
3f0c6a07 | 1888 | struct folio *folio; |
2457aec6 | 1889 | |
1da177e4 | 1890 | repeat: |
263e721e | 1891 | folio = filemap_get_entry(mapping, index); |
48c9d113 | 1892 | if (xa_is_value(folio)) |
3f0c6a07 | 1893 | folio = NULL; |
3f0c6a07 | 1894 | if (!folio) |
2457aec6 MG |
1895 | goto no_page; |
1896 | ||
1897 | if (fgp_flags & FGP_LOCK) { | |
1898 | if (fgp_flags & FGP_NOWAIT) { | |
3f0c6a07 MWO |
1899 | if (!folio_trylock(folio)) { |
1900 | folio_put(folio); | |
66dabbb6 | 1901 | return ERR_PTR(-EAGAIN); |
2457aec6 MG |
1902 | } |
1903 | } else { | |
3f0c6a07 | 1904 | folio_lock(folio); |
2457aec6 MG |
1905 | } |
1906 | ||
1907 | /* Has the page been truncated? */ | |
3f0c6a07 MWO |
1908 | if (unlikely(folio->mapping != mapping)) { |
1909 | folio_unlock(folio); | |
1910 | folio_put(folio); | |
2457aec6 MG |
1911 | goto repeat; |
1912 | } | |
3f0c6a07 | 1913 | VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); |
2457aec6 MG |
1914 | } |
1915 | ||
c16eb000 | 1916 | if (fgp_flags & FGP_ACCESSED) |
3f0c6a07 | 1917 | folio_mark_accessed(folio); |
b9306a79 YS |
1918 | else if (fgp_flags & FGP_WRITE) { |
1919 | /* Clear idle flag for buffer write */ | |
3f0c6a07 MWO |
1920 | if (folio_test_idle(folio)) |
1921 | folio_clear_idle(folio); | |
b9306a79 | 1922 | } |
2457aec6 | 1923 | |
b27652d9 MWO |
1924 | if (fgp_flags & FGP_STABLE) |
1925 | folio_wait_stable(folio); | |
2457aec6 | 1926 | no_page: |
3f0c6a07 | 1927 | if (!folio && (fgp_flags & FGP_CREAT)) { |
2457aec6 | 1928 | int err; |
f56753ac | 1929 | if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping)) |
3f0c6a07 | 1930 | gfp |= __GFP_WRITE; |
45f87de5 | 1931 | if (fgp_flags & FGP_NOFS) |
3f0c6a07 | 1932 | gfp &= ~__GFP_FS; |
0dd316ba JA |
1933 | if (fgp_flags & FGP_NOWAIT) { |
1934 | gfp &= ~GFP_KERNEL; | |
1935 | gfp |= GFP_NOWAIT | __GFP_NOWARN; | |
1936 | } | |
2457aec6 | 1937 | |
3f0c6a07 MWO |
1938 | folio = filemap_alloc_folio(gfp, 0); |
1939 | if (!folio) | |
66dabbb6 | 1940 | return ERR_PTR(-ENOMEM); |
2457aec6 | 1941 | |
a75d4c33 | 1942 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
2457aec6 MG |
1943 | fgp_flags |= FGP_LOCK; |
1944 | ||
eb39d618 | 1945 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1946 | if (fgp_flags & FGP_ACCESSED) |
3f0c6a07 | 1947 | __folio_set_referenced(folio); |
2457aec6 | 1948 | |
3f0c6a07 | 1949 | err = filemap_add_folio(mapping, folio, index, gfp); |
eb2be189 | 1950 | if (unlikely(err)) { |
3f0c6a07 MWO |
1951 | folio_put(folio); |
1952 | folio = NULL; | |
eb2be189 NP |
1953 | if (err == -EEXIST) |
1954 | goto repeat; | |
1da177e4 | 1955 | } |
a75d4c33 JB |
1956 | |
1957 | /* | |
3f0c6a07 MWO |
1958 | * filemap_add_folio locks the page, and for mmap |
1959 | * we expect an unlocked page. | |
a75d4c33 | 1960 | */ |
3f0c6a07 MWO |
1961 | if (folio && (fgp_flags & FGP_FOR_MMAP)) |
1962 | folio_unlock(folio); | |
1da177e4 | 1963 | } |
2457aec6 | 1964 | |
66dabbb6 CH |
1965 | if (!folio) |
1966 | return ERR_PTR(-ENOENT); | |
3f0c6a07 | 1967 | return folio; |
1da177e4 | 1968 | } |
3f0c6a07 | 1969 | EXPORT_SYMBOL(__filemap_get_folio); |
1da177e4 | 1970 | |
f5e6429a | 1971 | static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max, |
c7bad633 MWO |
1972 | xa_mark_t mark) |
1973 | { | |
f5e6429a | 1974 | struct folio *folio; |
c7bad633 MWO |
1975 | |
1976 | retry: | |
1977 | if (mark == XA_PRESENT) | |
f5e6429a | 1978 | folio = xas_find(xas, max); |
c7bad633 | 1979 | else |
f5e6429a | 1980 | folio = xas_find_marked(xas, max, mark); |
c7bad633 | 1981 | |
f5e6429a | 1982 | if (xas_retry(xas, folio)) |
c7bad633 MWO |
1983 | goto retry; |
1984 | /* | |
1985 | * A shadow entry of a recently evicted page, a swap | |
1986 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1987 | * without attempting to raise page count. | |
1988 | */ | |
f5e6429a MWO |
1989 | if (!folio || xa_is_value(folio)) |
1990 | return folio; | |
c7bad633 | 1991 | |
f5e6429a | 1992 | if (!folio_try_get_rcu(folio)) |
c7bad633 MWO |
1993 | goto reset; |
1994 | ||
f5e6429a MWO |
1995 | if (unlikely(folio != xas_reload(xas))) { |
1996 | folio_put(folio); | |
c7bad633 MWO |
1997 | goto reset; |
1998 | } | |
1999 | ||
f5e6429a | 2000 | return folio; |
c7bad633 MWO |
2001 | reset: |
2002 | xas_reset(xas); | |
2003 | goto retry; | |
2004 | } | |
2005 | ||
0cd6144a JW |
2006 | /** |
2007 | * find_get_entries - gang pagecache lookup | |
2008 | * @mapping: The address_space to search | |
2009 | * @start: The starting page cache index | |
ca122fe4 | 2010 | * @end: The final page index (inclusive). |
0e499ed3 | 2011 | * @fbatch: Where the resulting entries are placed. |
0cd6144a JW |
2012 | * @indices: The cache indices corresponding to the entries in @entries |
2013 | * | |
cf2039af | 2014 | * find_get_entries() will search for and return a batch of entries in |
0e499ed3 MWO |
2015 | * the mapping. The entries are placed in @fbatch. find_get_entries() |
2016 | * takes a reference on any actual folios it returns. | |
0cd6144a | 2017 | * |
0e499ed3 MWO |
2018 | * The entries have ascending indexes. The indices may not be consecutive |
2019 | * due to not-present entries or large folios. | |
0cd6144a | 2020 | * |
0e499ed3 | 2021 | * Any shadow entries of evicted folios, or swap entries from |
139b6a6f | 2022 | * shmem/tmpfs, are included in the returned array. |
0cd6144a | 2023 | * |
0e499ed3 | 2024 | * Return: The number of entries which were found. |
0cd6144a | 2025 | */ |
9fb6beea | 2026 | unsigned find_get_entries(struct address_space *mapping, pgoff_t *start, |
0e499ed3 | 2027 | pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices) |
0cd6144a | 2028 | { |
9fb6beea | 2029 | XA_STATE(xas, &mapping->i_pages, *start); |
f5e6429a | 2030 | struct folio *folio; |
0cd6144a JW |
2031 | |
2032 | rcu_read_lock(); | |
f5e6429a | 2033 | while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) { |
0e499ed3 MWO |
2034 | indices[fbatch->nr] = xas.xa_index; |
2035 | if (!folio_batch_add(fbatch, folio)) | |
0cd6144a JW |
2036 | break; |
2037 | } | |
2038 | rcu_read_unlock(); | |
cf2039af | 2039 | |
9fb6beea VMO |
2040 | if (folio_batch_count(fbatch)) { |
2041 | unsigned long nr = 1; | |
2042 | int idx = folio_batch_count(fbatch) - 1; | |
2043 | ||
2044 | folio = fbatch->folios[idx]; | |
2045 | if (!xa_is_value(folio) && !folio_test_hugetlb(folio)) | |
2046 | nr = folio_nr_pages(folio); | |
2047 | *start = indices[idx] + nr; | |
2048 | } | |
0e499ed3 | 2049 | return folio_batch_count(fbatch); |
0cd6144a JW |
2050 | } |
2051 | ||
5c211ba2 MWO |
2052 | /** |
2053 | * find_lock_entries - Find a batch of pagecache entries. | |
2054 | * @mapping: The address_space to search. | |
2055 | * @start: The starting page cache index. | |
2056 | * @end: The final page index (inclusive). | |
51dcbdac MWO |
2057 | * @fbatch: Where the resulting entries are placed. |
2058 | * @indices: The cache indices of the entries in @fbatch. | |
5c211ba2 MWO |
2059 | * |
2060 | * find_lock_entries() will return a batch of entries from @mapping. | |
f5e6429a MWO |
2061 | * Swap, shadow and DAX entries are included. Folios are returned |
2062 | * locked and with an incremented refcount. Folios which are locked | |
2063 | * by somebody else or under writeback are skipped. Folios which are | |
2064 | * partially outside the range are not returned. | |
5c211ba2 MWO |
2065 | * |
2066 | * The entries have ascending indexes. The indices may not be consecutive | |
f5e6429a MWO |
2067 | * due to not-present entries, large folios, folios which could not be |
2068 | * locked or folios under writeback. | |
5c211ba2 MWO |
2069 | * |
2070 | * Return: The number of entries which were found. | |
2071 | */ | |
3392ca12 | 2072 | unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start, |
51dcbdac | 2073 | pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices) |
5c211ba2 | 2074 | { |
3392ca12 | 2075 | XA_STATE(xas, &mapping->i_pages, *start); |
f5e6429a | 2076 | struct folio *folio; |
5c211ba2 MWO |
2077 | |
2078 | rcu_read_lock(); | |
f5e6429a MWO |
2079 | while ((folio = find_get_entry(&xas, end, XA_PRESENT))) { |
2080 | if (!xa_is_value(folio)) { | |
3392ca12 | 2081 | if (folio->index < *start) |
5c211ba2 | 2082 | goto put; |
f5e6429a | 2083 | if (folio->index + folio_nr_pages(folio) - 1 > end) |
5c211ba2 | 2084 | goto put; |
f5e6429a | 2085 | if (!folio_trylock(folio)) |
5c211ba2 | 2086 | goto put; |
f5e6429a MWO |
2087 | if (folio->mapping != mapping || |
2088 | folio_test_writeback(folio)) | |
5c211ba2 | 2089 | goto unlock; |
f5e6429a MWO |
2090 | VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index), |
2091 | folio); | |
5c211ba2 | 2092 | } |
51dcbdac MWO |
2093 | indices[fbatch->nr] = xas.xa_index; |
2094 | if (!folio_batch_add(fbatch, folio)) | |
5c211ba2 | 2095 | break; |
6b24ca4a | 2096 | continue; |
5c211ba2 | 2097 | unlock: |
f5e6429a | 2098 | folio_unlock(folio); |
5c211ba2 | 2099 | put: |
f5e6429a | 2100 | folio_put(folio); |
5c211ba2 MWO |
2101 | } |
2102 | rcu_read_unlock(); | |
2103 | ||
3392ca12 VMO |
2104 | if (folio_batch_count(fbatch)) { |
2105 | unsigned long nr = 1; | |
2106 | int idx = folio_batch_count(fbatch) - 1; | |
2107 | ||
2108 | folio = fbatch->folios[idx]; | |
2109 | if (!xa_is_value(folio) && !folio_test_hugetlb(folio)) | |
2110 | nr = folio_nr_pages(folio); | |
2111 | *start = indices[idx] + nr; | |
2112 | } | |
51dcbdac | 2113 | return folio_batch_count(fbatch); |
5c211ba2 MWO |
2114 | } |
2115 | ||
1da177e4 | 2116 | /** |
be0ced5e | 2117 | * filemap_get_folios - Get a batch of folios |
1da177e4 LT |
2118 | * @mapping: The address_space to search |
2119 | * @start: The starting page index | |
b947cee4 | 2120 | * @end: The final page index (inclusive) |
be0ced5e | 2121 | * @fbatch: The batch to fill. |
1da177e4 | 2122 | * |
be0ced5e MWO |
2123 | * Search for and return a batch of folios in the mapping starting at |
2124 | * index @start and up to index @end (inclusive). The folios are returned | |
2125 | * in @fbatch with an elevated reference count. | |
1da177e4 | 2126 | * |
be0ced5e MWO |
2127 | * The first folio may start before @start; if it does, it will contain |
2128 | * @start. The final folio may extend beyond @end; if it does, it will | |
2129 | * contain @end. The folios have ascending indices. There may be gaps | |
2130 | * between the folios if there are indices which have no folio in the | |
2131 | * page cache. If folios are added to or removed from the page cache | |
2132 | * while this is running, they may or may not be found by this call. | |
1da177e4 | 2133 | * |
be0ced5e MWO |
2134 | * Return: The number of folios which were found. |
2135 | * We also update @start to index the next folio for the traversal. | |
1da177e4 | 2136 | */ |
be0ced5e MWO |
2137 | unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start, |
2138 | pgoff_t end, struct folio_batch *fbatch) | |
1da177e4 | 2139 | { |
fd1b3cee | 2140 | XA_STATE(xas, &mapping->i_pages, *start); |
f5e6429a | 2141 | struct folio *folio; |
a60637c8 NP |
2142 | |
2143 | rcu_read_lock(); | |
be0ced5e | 2144 | while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) { |
fd1b3cee | 2145 | /* Skip over shadow, swap and DAX entries */ |
f5e6429a | 2146 | if (xa_is_value(folio)) |
8079b1c8 | 2147 | continue; |
be0ced5e MWO |
2148 | if (!folio_batch_add(fbatch, folio)) { |
2149 | unsigned long nr = folio_nr_pages(folio); | |
a60637c8 | 2150 | |
be0ced5e MWO |
2151 | if (folio_test_hugetlb(folio)) |
2152 | nr = 1; | |
2153 | *start = folio->index + nr; | |
b947cee4 JK |
2154 | goto out; |
2155 | } | |
a60637c8 | 2156 | } |
5b280c0c | 2157 | |
b947cee4 JK |
2158 | /* |
2159 | * We come here when there is no page beyond @end. We take care to not | |
2160 | * overflow the index @start as it confuses some of the callers. This | |
fd1b3cee | 2161 | * breaks the iteration when there is a page at index -1 but that is |
b947cee4 JK |
2162 | * already broken anyway. |
2163 | */ | |
2164 | if (end == (pgoff_t)-1) | |
2165 | *start = (pgoff_t)-1; | |
2166 | else | |
2167 | *start = end + 1; | |
2168 | out: | |
a60637c8 | 2169 | rcu_read_unlock(); |
d72dc8a2 | 2170 | |
be0ced5e MWO |
2171 | return folio_batch_count(fbatch); |
2172 | } | |
2173 | EXPORT_SYMBOL(filemap_get_folios); | |
2174 | ||
6b24ca4a MWO |
2175 | static inline |
2176 | bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max) | |
2177 | { | |
2178 | if (!folio_test_large(folio) || folio_test_hugetlb(folio)) | |
2179 | return false; | |
2180 | if (index >= max) | |
2181 | return false; | |
2182 | return index < folio->index + folio_nr_pages(folio) - 1; | |
1da177e4 LT |
2183 | } |
2184 | ||
ebf43500 | 2185 | /** |
35b47146 | 2186 | * filemap_get_folios_contig - Get a batch of contiguous folios |
ebf43500 | 2187 | * @mapping: The address_space to search |
35b47146 VMO |
2188 | * @start: The starting page index |
2189 | * @end: The final page index (inclusive) | |
2190 | * @fbatch: The batch to fill | |
ebf43500 | 2191 | * |
35b47146 VMO |
2192 | * filemap_get_folios_contig() works exactly like filemap_get_folios(), |
2193 | * except the returned folios are guaranteed to be contiguous. This may | |
2194 | * not return all contiguous folios if the batch gets filled up. | |
ebf43500 | 2195 | * |
35b47146 VMO |
2196 | * Return: The number of folios found. |
2197 | * Also update @start to be positioned for traversal of the next folio. | |
ebf43500 | 2198 | */ |
35b47146 VMO |
2199 | |
2200 | unsigned filemap_get_folios_contig(struct address_space *mapping, | |
2201 | pgoff_t *start, pgoff_t end, struct folio_batch *fbatch) | |
ebf43500 | 2202 | { |
35b47146 VMO |
2203 | XA_STATE(xas, &mapping->i_pages, *start); |
2204 | unsigned long nr; | |
e1c37722 | 2205 | struct folio *folio; |
a60637c8 NP |
2206 | |
2207 | rcu_read_lock(); | |
35b47146 VMO |
2208 | |
2209 | for (folio = xas_load(&xas); folio && xas.xa_index <= end; | |
2210 | folio = xas_next(&xas)) { | |
e1c37722 | 2211 | if (xas_retry(&xas, folio)) |
3ece58a2 MW |
2212 | continue; |
2213 | /* | |
2214 | * If the entry has been swapped out, we can stop looking. | |
2215 | * No current caller is looking for DAX entries. | |
2216 | */ | |
e1c37722 | 2217 | if (xa_is_value(folio)) |
35b47146 | 2218 | goto update_start; |
ebf43500 | 2219 | |
e1c37722 | 2220 | if (!folio_try_get_rcu(folio)) |
3ece58a2 | 2221 | goto retry; |
83929372 | 2222 | |
e1c37722 | 2223 | if (unlikely(folio != xas_reload(&xas))) |
35b47146 | 2224 | goto put_folio; |
a60637c8 | 2225 | |
35b47146 VMO |
2226 | if (!folio_batch_add(fbatch, folio)) { |
2227 | nr = folio_nr_pages(folio); | |
2228 | ||
2229 | if (folio_test_hugetlb(folio)) | |
2230 | nr = 1; | |
2231 | *start = folio->index + nr; | |
2232 | goto out; | |
6b24ca4a | 2233 | } |
3ece58a2 | 2234 | continue; |
35b47146 | 2235 | put_folio: |
e1c37722 | 2236 | folio_put(folio); |
35b47146 | 2237 | |
3ece58a2 MW |
2238 | retry: |
2239 | xas_reset(&xas); | |
ebf43500 | 2240 | } |
35b47146 VMO |
2241 | |
2242 | update_start: | |
2243 | nr = folio_batch_count(fbatch); | |
2244 | ||
2245 | if (nr) { | |
2246 | folio = fbatch->folios[nr - 1]; | |
2247 | if (folio_test_hugetlb(folio)) | |
2248 | *start = folio->index + 1; | |
2249 | else | |
2250 | *start = folio->index + folio_nr_pages(folio); | |
2251 | } | |
2252 | out: | |
a60637c8 | 2253 | rcu_read_unlock(); |
35b47146 | 2254 | return folio_batch_count(fbatch); |
ebf43500 | 2255 | } |
35b47146 | 2256 | EXPORT_SYMBOL(filemap_get_folios_contig); |
ebf43500 | 2257 | |
485bb99b | 2258 | /** |
247f9e1f VMO |
2259 | * filemap_get_folios_tag - Get a batch of folios matching @tag |
2260 | * @mapping: The address_space to search | |
2261 | * @start: The starting page index | |
2262 | * @end: The final page index (inclusive) | |
2263 | * @tag: The tag index | |
2264 | * @fbatch: The batch to fill | |
485bb99b | 2265 | * |
247f9e1f | 2266 | * Same as filemap_get_folios(), but only returning folios tagged with @tag. |
a862f68a | 2267 | * |
247f9e1f VMO |
2268 | * Return: The number of folios found. |
2269 | * Also update @start to index the next folio for traversal. | |
1da177e4 | 2270 | */ |
247f9e1f VMO |
2271 | unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start, |
2272 | pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch) | |
1da177e4 | 2273 | { |
247f9e1f | 2274 | XA_STATE(xas, &mapping->i_pages, *start); |
f5e6429a | 2275 | struct folio *folio; |
a60637c8 NP |
2276 | |
2277 | rcu_read_lock(); | |
247f9e1f | 2278 | while ((folio = find_get_entry(&xas, end, tag)) != NULL) { |
a6906972 MW |
2279 | /* |
2280 | * Shadow entries should never be tagged, but this iteration | |
2281 | * is lockless so there is a window for page reclaim to evict | |
247f9e1f | 2282 | * a page we saw tagged. Skip over it. |
a6906972 | 2283 | */ |
f5e6429a | 2284 | if (xa_is_value(folio)) |
139b6a6f | 2285 | continue; |
247f9e1f VMO |
2286 | if (!folio_batch_add(fbatch, folio)) { |
2287 | unsigned long nr = folio_nr_pages(folio); | |
a60637c8 | 2288 | |
247f9e1f VMO |
2289 | if (folio_test_hugetlb(folio)) |
2290 | nr = 1; | |
2291 | *start = folio->index + nr; | |
72b045ae JK |
2292 | goto out; |
2293 | } | |
a60637c8 | 2294 | } |
72b045ae | 2295 | /* |
247f9e1f VMO |
2296 | * We come here when there is no page beyond @end. We take care to not |
2297 | * overflow the index @start as it confuses some of the callers. This | |
2298 | * breaks the iteration when there is a page at index -1 but that is | |
2299 | * already broke anyway. | |
72b045ae JK |
2300 | */ |
2301 | if (end == (pgoff_t)-1) | |
247f9e1f | 2302 | *start = (pgoff_t)-1; |
72b045ae | 2303 | else |
247f9e1f | 2304 | *start = end + 1; |
72b045ae | 2305 | out: |
a60637c8 | 2306 | rcu_read_unlock(); |
1da177e4 | 2307 | |
247f9e1f | 2308 | return folio_batch_count(fbatch); |
1da177e4 | 2309 | } |
247f9e1f | 2310 | EXPORT_SYMBOL(filemap_get_folios_tag); |
1da177e4 | 2311 | |
76d42bd9 WF |
2312 | /* |
2313 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
2314 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
2315 | * | |
2316 | * ---R__________________________________________B__________ | |
2317 | * ^ reading here ^ bad block(assume 4k) | |
2318 | * | |
2319 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
2320 | * => failing the whole request => read(R) => read(R+1) => | |
2321 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
2322 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
2323 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
2324 | * | |
2325 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
2326 | */ | |
0f8e2db4 | 2327 | static void shrink_readahead_size_eio(struct file_ra_state *ra) |
76d42bd9 | 2328 | { |
76d42bd9 | 2329 | ra->ra_pages /= 4; |
76d42bd9 WF |
2330 | } |
2331 | ||
cbd59c48 | 2332 | /* |
25d6a23e | 2333 | * filemap_get_read_batch - Get a batch of folios for read |
cbd59c48 | 2334 | * |
25d6a23e MWO |
2335 | * Get a batch of folios which represent a contiguous range of bytes in |
2336 | * the file. No exceptional entries will be returned. If @index is in | |
2337 | * the middle of a folio, the entire folio will be returned. The last | |
2338 | * folio in the batch may have the readahead flag set or the uptodate flag | |
2339 | * clear so that the caller can take the appropriate action. | |
cbd59c48 MWO |
2340 | */ |
2341 | static void filemap_get_read_batch(struct address_space *mapping, | |
25d6a23e | 2342 | pgoff_t index, pgoff_t max, struct folio_batch *fbatch) |
cbd59c48 MWO |
2343 | { |
2344 | XA_STATE(xas, &mapping->i_pages, index); | |
bdb72932 | 2345 | struct folio *folio; |
cbd59c48 MWO |
2346 | |
2347 | rcu_read_lock(); | |
bdb72932 MWO |
2348 | for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) { |
2349 | if (xas_retry(&xas, folio)) | |
cbd59c48 | 2350 | continue; |
bdb72932 | 2351 | if (xas.xa_index > max || xa_is_value(folio)) |
cbd59c48 | 2352 | break; |
cb995f4e MWO |
2353 | if (xa_is_sibling(folio)) |
2354 | break; | |
bdb72932 | 2355 | if (!folio_try_get_rcu(folio)) |
cbd59c48 MWO |
2356 | goto retry; |
2357 | ||
bdb72932 | 2358 | if (unlikely(folio != xas_reload(&xas))) |
25d6a23e | 2359 | goto put_folio; |
cbd59c48 | 2360 | |
25d6a23e | 2361 | if (!folio_batch_add(fbatch, folio)) |
cbd59c48 | 2362 | break; |
bdb72932 | 2363 | if (!folio_test_uptodate(folio)) |
cbd59c48 | 2364 | break; |
bdb72932 | 2365 | if (folio_test_readahead(folio)) |
cbd59c48 | 2366 | break; |
6b24ca4a | 2367 | xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1); |
cbd59c48 | 2368 | continue; |
25d6a23e | 2369 | put_folio: |
bdb72932 | 2370 | folio_put(folio); |
cbd59c48 MWO |
2371 | retry: |
2372 | xas_reset(&xas); | |
2373 | } | |
2374 | rcu_read_unlock(); | |
2375 | } | |
2376 | ||
290e1a32 | 2377 | static int filemap_read_folio(struct file *file, filler_t filler, |
9d427b4e | 2378 | struct folio *folio) |
723ef24b | 2379 | { |
17604240 CH |
2380 | bool workingset = folio_test_workingset(folio); |
2381 | unsigned long pflags; | |
723ef24b KO |
2382 | int error; |
2383 | ||
723ef24b | 2384 | /* |
68430303 | 2385 | * A previous I/O error may have been due to temporary failures, |
7e0a1265 | 2386 | * eg. multipath errors. PG_error will be set again if read_folio |
68430303 | 2387 | * fails. |
723ef24b | 2388 | */ |
9d427b4e | 2389 | folio_clear_error(folio); |
17604240 | 2390 | |
723ef24b | 2391 | /* Start the actual read. The read will unlock the page. */ |
17604240 CH |
2392 | if (unlikely(workingset)) |
2393 | psi_memstall_enter(&pflags); | |
290e1a32 | 2394 | error = filler(file, folio); |
17604240 CH |
2395 | if (unlikely(workingset)) |
2396 | psi_memstall_leave(&pflags); | |
68430303 MWO |
2397 | if (error) |
2398 | return error; | |
723ef24b | 2399 | |
9d427b4e | 2400 | error = folio_wait_locked_killable(folio); |
68430303 MWO |
2401 | if (error) |
2402 | return error; | |
9d427b4e | 2403 | if (folio_test_uptodate(folio)) |
aa1ec2f6 | 2404 | return 0; |
290e1a32 MWO |
2405 | if (file) |
2406 | shrink_readahead_size_eio(&file->f_ra); | |
aa1ec2f6 | 2407 | return -EIO; |
723ef24b KO |
2408 | } |
2409 | ||
fce70da3 | 2410 | static bool filemap_range_uptodate(struct address_space *mapping, |
dd5b9d00 DH |
2411 | loff_t pos, size_t count, struct folio *folio, |
2412 | bool need_uptodate) | |
fce70da3 | 2413 | { |
2fa4eeb8 | 2414 | if (folio_test_uptodate(folio)) |
fce70da3 MWO |
2415 | return true; |
2416 | /* pipes can't handle partially uptodate pages */ | |
dd5b9d00 | 2417 | if (need_uptodate) |
fce70da3 MWO |
2418 | return false; |
2419 | if (!mapping->a_ops->is_partially_uptodate) | |
2420 | return false; | |
2fa4eeb8 | 2421 | if (mapping->host->i_blkbits >= folio_shift(folio)) |
fce70da3 MWO |
2422 | return false; |
2423 | ||
2fa4eeb8 MWO |
2424 | if (folio_pos(folio) > pos) { |
2425 | count -= folio_pos(folio) - pos; | |
fce70da3 MWO |
2426 | pos = 0; |
2427 | } else { | |
2fa4eeb8 | 2428 | pos -= folio_pos(folio); |
fce70da3 MWO |
2429 | } |
2430 | ||
2e7e80f7 | 2431 | return mapping->a_ops->is_partially_uptodate(folio, pos, count); |
fce70da3 MWO |
2432 | } |
2433 | ||
4612aeef | 2434 | static int filemap_update_page(struct kiocb *iocb, |
dd5b9d00 DH |
2435 | struct address_space *mapping, size_t count, |
2436 | struct folio *folio, bool need_uptodate) | |
723ef24b | 2437 | { |
723ef24b KO |
2438 | int error; |
2439 | ||
730633f0 JK |
2440 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2441 | if (!filemap_invalidate_trylock_shared(mapping)) | |
2442 | return -EAGAIN; | |
2443 | } else { | |
2444 | filemap_invalidate_lock_shared(mapping); | |
2445 | } | |
2446 | ||
ffdc8dab | 2447 | if (!folio_trylock(folio)) { |
730633f0 | 2448 | error = -EAGAIN; |
87d1d7b6 | 2449 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) |
730633f0 | 2450 | goto unlock_mapping; |
87d1d7b6 | 2451 | if (!(iocb->ki_flags & IOCB_WAITQ)) { |
730633f0 | 2452 | filemap_invalidate_unlock_shared(mapping); |
9f2b04a2 MWO |
2453 | /* |
2454 | * This is where we usually end up waiting for a | |
2455 | * previously submitted readahead to finish. | |
2456 | */ | |
2457 | folio_put_wait_locked(folio, TASK_KILLABLE); | |
4612aeef | 2458 | return AOP_TRUNCATED_PAGE; |
bd8a1f36 | 2459 | } |
ffdc8dab | 2460 | error = __folio_lock_async(folio, iocb->ki_waitq); |
87d1d7b6 | 2461 | if (error) |
730633f0 | 2462 | goto unlock_mapping; |
723ef24b | 2463 | } |
723ef24b | 2464 | |
730633f0 | 2465 | error = AOP_TRUNCATED_PAGE; |
ffdc8dab | 2466 | if (!folio->mapping) |
730633f0 | 2467 | goto unlock; |
723ef24b | 2468 | |
fce70da3 | 2469 | error = 0; |
dd5b9d00 DH |
2470 | if (filemap_range_uptodate(mapping, iocb->ki_pos, count, folio, |
2471 | need_uptodate)) | |
fce70da3 MWO |
2472 | goto unlock; |
2473 | ||
2474 | error = -EAGAIN; | |
2475 | if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ)) | |
2476 | goto unlock; | |
2477 | ||
290e1a32 MWO |
2478 | error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio, |
2479 | folio); | |
730633f0 | 2480 | goto unlock_mapping; |
fce70da3 | 2481 | unlock: |
ffdc8dab | 2482 | folio_unlock(folio); |
730633f0 JK |
2483 | unlock_mapping: |
2484 | filemap_invalidate_unlock_shared(mapping); | |
2485 | if (error == AOP_TRUNCATED_PAGE) | |
ffdc8dab | 2486 | folio_put(folio); |
fce70da3 | 2487 | return error; |
723ef24b KO |
2488 | } |
2489 | ||
a5d4ad09 | 2490 | static int filemap_create_folio(struct file *file, |
f253e185 | 2491 | struct address_space *mapping, pgoff_t index, |
25d6a23e | 2492 | struct folio_batch *fbatch) |
723ef24b | 2493 | { |
a5d4ad09 | 2494 | struct folio *folio; |
723ef24b KO |
2495 | int error; |
2496 | ||
a5d4ad09 MWO |
2497 | folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0); |
2498 | if (!folio) | |
f253e185 | 2499 | return -ENOMEM; |
723ef24b | 2500 | |
730633f0 | 2501 | /* |
a5d4ad09 MWO |
2502 | * Protect against truncate / hole punch. Grabbing invalidate_lock |
2503 | * here assures we cannot instantiate and bring uptodate new | |
2504 | * pagecache folios after evicting page cache during truncate | |
2505 | * and before actually freeing blocks. Note that we could | |
2506 | * release invalidate_lock after inserting the folio into | |
2507 | * the page cache as the locked folio would then be enough to | |
2508 | * synchronize with hole punching. But there are code paths | |
2509 | * such as filemap_update_page() filling in partially uptodate | |
704528d8 | 2510 | * pages or ->readahead() that need to hold invalidate_lock |
a5d4ad09 MWO |
2511 | * while mapping blocks for IO so let's hold the lock here as |
2512 | * well to keep locking rules simple. | |
730633f0 JK |
2513 | */ |
2514 | filemap_invalidate_lock_shared(mapping); | |
a5d4ad09 | 2515 | error = filemap_add_folio(mapping, folio, index, |
f253e185 MWO |
2516 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
2517 | if (error == -EEXIST) | |
2518 | error = AOP_TRUNCATED_PAGE; | |
2519 | if (error) | |
2520 | goto error; | |
2521 | ||
290e1a32 | 2522 | error = filemap_read_folio(file, mapping->a_ops->read_folio, folio); |
f253e185 MWO |
2523 | if (error) |
2524 | goto error; | |
2525 | ||
730633f0 | 2526 | filemap_invalidate_unlock_shared(mapping); |
25d6a23e | 2527 | folio_batch_add(fbatch, folio); |
f253e185 MWO |
2528 | return 0; |
2529 | error: | |
730633f0 | 2530 | filemap_invalidate_unlock_shared(mapping); |
a5d4ad09 | 2531 | folio_put(folio); |
f253e185 | 2532 | return error; |
723ef24b KO |
2533 | } |
2534 | ||
5963fe03 | 2535 | static int filemap_readahead(struct kiocb *iocb, struct file *file, |
65bca53b | 2536 | struct address_space *mapping, struct folio *folio, |
5963fe03 MWO |
2537 | pgoff_t last_index) |
2538 | { | |
65bca53b MWO |
2539 | DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index); |
2540 | ||
5963fe03 MWO |
2541 | if (iocb->ki_flags & IOCB_NOIO) |
2542 | return -EAGAIN; | |
65bca53b | 2543 | page_cache_async_ra(&ractl, folio, last_index - folio->index); |
5963fe03 MWO |
2544 | return 0; |
2545 | } | |
2546 | ||
dd5b9d00 DH |
2547 | static int filemap_get_pages(struct kiocb *iocb, size_t count, |
2548 | struct folio_batch *fbatch, bool need_uptodate) | |
06c04442 KO |
2549 | { |
2550 | struct file *filp = iocb->ki_filp; | |
2551 | struct address_space *mapping = filp->f_mapping; | |
2552 | struct file_ra_state *ra = &filp->f_ra; | |
2553 | pgoff_t index = iocb->ki_pos >> PAGE_SHIFT; | |
cbd59c48 | 2554 | pgoff_t last_index; |
65bca53b | 2555 | struct folio *folio; |
cbd59c48 | 2556 | int err = 0; |
06c04442 | 2557 | |
5956592c | 2558 | /* "last_index" is the index of the page beyond the end of the read */ |
dd5b9d00 | 2559 | last_index = DIV_ROUND_UP(iocb->ki_pos + count, PAGE_SIZE); |
2642fca6 | 2560 | retry: |
06c04442 KO |
2561 | if (fatal_signal_pending(current)) |
2562 | return -EINTR; | |
2563 | ||
5956592c | 2564 | filemap_get_read_batch(mapping, index, last_index - 1, fbatch); |
25d6a23e | 2565 | if (!folio_batch_count(fbatch)) { |
2642fca6 MWO |
2566 | if (iocb->ki_flags & IOCB_NOIO) |
2567 | return -EAGAIN; | |
2568 | page_cache_sync_readahead(mapping, ra, filp, index, | |
2569 | last_index - index); | |
5956592c | 2570 | filemap_get_read_batch(mapping, index, last_index - 1, fbatch); |
2642fca6 | 2571 | } |
25d6a23e | 2572 | if (!folio_batch_count(fbatch)) { |
f253e185 MWO |
2573 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ)) |
2574 | return -EAGAIN; | |
a5d4ad09 | 2575 | err = filemap_create_folio(filp, mapping, |
25d6a23e | 2576 | iocb->ki_pos >> PAGE_SHIFT, fbatch); |
f253e185 | 2577 | if (err == AOP_TRUNCATED_PAGE) |
2642fca6 | 2578 | goto retry; |
f253e185 MWO |
2579 | return err; |
2580 | } | |
06c04442 | 2581 | |
25d6a23e | 2582 | folio = fbatch->folios[folio_batch_count(fbatch) - 1]; |
65bca53b MWO |
2583 | if (folio_test_readahead(folio)) { |
2584 | err = filemap_readahead(iocb, filp, mapping, folio, last_index); | |
2642fca6 MWO |
2585 | if (err) |
2586 | goto err; | |
2587 | } | |
65bca53b | 2588 | if (!folio_test_uptodate(folio)) { |
25d6a23e MWO |
2589 | if ((iocb->ki_flags & IOCB_WAITQ) && |
2590 | folio_batch_count(fbatch) > 1) | |
2642fca6 | 2591 | iocb->ki_flags |= IOCB_NOWAIT; |
dd5b9d00 DH |
2592 | err = filemap_update_page(iocb, mapping, count, folio, |
2593 | need_uptodate); | |
2642fca6 MWO |
2594 | if (err) |
2595 | goto err; | |
06c04442 KO |
2596 | } |
2597 | ||
2642fca6 | 2598 | return 0; |
cbd59c48 | 2599 | err: |
2642fca6 | 2600 | if (err < 0) |
65bca53b | 2601 | folio_put(folio); |
25d6a23e | 2602 | if (likely(--fbatch->nr)) |
ff993ba1 | 2603 | return 0; |
4612aeef | 2604 | if (err == AOP_TRUNCATED_PAGE) |
2642fca6 MWO |
2605 | goto retry; |
2606 | return err; | |
06c04442 KO |
2607 | } |
2608 | ||
5ccc944d MWO |
2609 | static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio) |
2610 | { | |
2611 | unsigned int shift = folio_shift(folio); | |
2612 | ||
2613 | return (pos1 >> shift == pos2 >> shift); | |
2614 | } | |
2615 | ||
485bb99b | 2616 | /** |
87fa0f3e CH |
2617 | * filemap_read - Read data from the page cache. |
2618 | * @iocb: The iocb to read. | |
2619 | * @iter: Destination for the data. | |
2620 | * @already_read: Number of bytes already read by the caller. | |
485bb99b | 2621 | * |
87fa0f3e | 2622 | * Copies data from the page cache. If the data is not currently present, |
7e0a1265 | 2623 | * uses the readahead and read_folio address_space operations to fetch it. |
1da177e4 | 2624 | * |
87fa0f3e CH |
2625 | * Return: Total number of bytes copied, including those already read by |
2626 | * the caller. If an error happens before any bytes are copied, returns | |
2627 | * a negative error number. | |
1da177e4 | 2628 | */ |
87fa0f3e CH |
2629 | ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter, |
2630 | ssize_t already_read) | |
1da177e4 | 2631 | { |
47c27bc4 | 2632 | struct file *filp = iocb->ki_filp; |
06c04442 | 2633 | struct file_ra_state *ra = &filp->f_ra; |
36e78914 | 2634 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2635 | struct inode *inode = mapping->host; |
25d6a23e | 2636 | struct folio_batch fbatch; |
ff993ba1 | 2637 | int i, error = 0; |
06c04442 KO |
2638 | bool writably_mapped; |
2639 | loff_t isize, end_offset; | |
1da177e4 | 2640 | |
723ef24b | 2641 | if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2642 | return 0; |
3644e2d2 KO |
2643 | if (unlikely(!iov_iter_count(iter))) |
2644 | return 0; | |
2645 | ||
c2a9737f | 2646 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
25d6a23e | 2647 | folio_batch_init(&fbatch); |
c2a9737f | 2648 | |
06c04442 | 2649 | do { |
1da177e4 | 2650 | cond_resched(); |
5abf186a | 2651 | |
723ef24b | 2652 | /* |
06c04442 KO |
2653 | * If we've already successfully copied some data, then we |
2654 | * can no longer safely return -EIOCBQUEUED. Hence mark | |
2655 | * an async read NOWAIT at that point. | |
723ef24b | 2656 | */ |
87fa0f3e | 2657 | if ((iocb->ki_flags & IOCB_WAITQ) && already_read) |
723ef24b KO |
2658 | iocb->ki_flags |= IOCB_NOWAIT; |
2659 | ||
8c8387ee DH |
2660 | if (unlikely(iocb->ki_pos >= i_size_read(inode))) |
2661 | break; | |
2662 | ||
dd5b9d00 DH |
2663 | error = filemap_get_pages(iocb, iter->count, &fbatch, |
2664 | iov_iter_is_pipe(iter)); | |
ff993ba1 | 2665 | if (error < 0) |
06c04442 | 2666 | break; |
1da177e4 | 2667 | |
06c04442 KO |
2668 | /* |
2669 | * i_size must be checked after we know the pages are Uptodate. | |
2670 | * | |
2671 | * Checking i_size after the check allows us to calculate | |
2672 | * the correct value for "nr", which means the zero-filled | |
2673 | * part of the page is not copied back to userspace (unless | |
2674 | * another truncate extends the file - this is desired though). | |
2675 | */ | |
2676 | isize = i_size_read(inode); | |
2677 | if (unlikely(iocb->ki_pos >= isize)) | |
25d6a23e | 2678 | goto put_folios; |
06c04442 KO |
2679 | end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count); |
2680 | ||
06c04442 KO |
2681 | /* |
2682 | * Once we start copying data, we don't want to be touching any | |
2683 | * cachelines that might be contended: | |
2684 | */ | |
2685 | writably_mapped = mapping_writably_mapped(mapping); | |
2686 | ||
2687 | /* | |
5ccc944d | 2688 | * When a read accesses the same folio several times, only |
06c04442 KO |
2689 | * mark it as accessed the first time. |
2690 | */ | |
5ccc944d MWO |
2691 | if (!pos_same_folio(iocb->ki_pos, ra->prev_pos - 1, |
2692 | fbatch.folios[0])) | |
25d6a23e | 2693 | folio_mark_accessed(fbatch.folios[0]); |
06c04442 | 2694 | |
25d6a23e MWO |
2695 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
2696 | struct folio *folio = fbatch.folios[i]; | |
d996fc7f MWO |
2697 | size_t fsize = folio_size(folio); |
2698 | size_t offset = iocb->ki_pos & (fsize - 1); | |
cbd59c48 | 2699 | size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos, |
d996fc7f | 2700 | fsize - offset); |
cbd59c48 | 2701 | size_t copied; |
06c04442 | 2702 | |
d996fc7f | 2703 | if (end_offset < folio_pos(folio)) |
cbd59c48 MWO |
2704 | break; |
2705 | if (i > 0) | |
d996fc7f | 2706 | folio_mark_accessed(folio); |
06c04442 | 2707 | /* |
d996fc7f MWO |
2708 | * If users can be writing to this folio using arbitrary |
2709 | * virtual addresses, take care of potential aliasing | |
2710 | * before reading the folio on the kernel side. | |
06c04442 | 2711 | */ |
d996fc7f MWO |
2712 | if (writably_mapped) |
2713 | flush_dcache_folio(folio); | |
06c04442 | 2714 | |
d996fc7f | 2715 | copied = copy_folio_to_iter(folio, offset, bytes, iter); |
06c04442 | 2716 | |
87fa0f3e | 2717 | already_read += copied; |
06c04442 KO |
2718 | iocb->ki_pos += copied; |
2719 | ra->prev_pos = iocb->ki_pos; | |
2720 | ||
2721 | if (copied < bytes) { | |
2722 | error = -EFAULT; | |
2723 | break; | |
2724 | } | |
1da177e4 | 2725 | } |
25d6a23e MWO |
2726 | put_folios: |
2727 | for (i = 0; i < folio_batch_count(&fbatch); i++) | |
2728 | folio_put(fbatch.folios[i]); | |
2729 | folio_batch_init(&fbatch); | |
06c04442 | 2730 | } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error); |
1da177e4 | 2731 | |
0c6aa263 | 2732 | file_accessed(filp); |
06c04442 | 2733 | |
87fa0f3e | 2734 | return already_read ? already_read : error; |
1da177e4 | 2735 | } |
87fa0f3e | 2736 | EXPORT_SYMBOL_GPL(filemap_read); |
1da177e4 | 2737 | |
485bb99b | 2738 | /** |
6abd2322 | 2739 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2740 | * @iocb: kernel I/O control block |
6abd2322 | 2741 | * @iter: destination for the data read |
485bb99b | 2742 | * |
6abd2322 | 2743 | * This is the "read_iter()" routine for all filesystems |
1da177e4 | 2744 | * that can use the page cache directly. |
41da51bc AG |
2745 | * |
2746 | * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall | |
2747 | * be returned when no data can be read without waiting for I/O requests | |
2748 | * to complete; it doesn't prevent readahead. | |
2749 | * | |
2750 | * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O | |
2751 | * requests shall be made for the read or for readahead. When no data | |
2752 | * can be read, -EAGAIN shall be returned. When readahead would be | |
2753 | * triggered, a partial, possibly empty read shall be returned. | |
2754 | * | |
a862f68a MR |
2755 | * Return: |
2756 | * * number of bytes copied, even for partial reads | |
41da51bc | 2757 | * * negative error code (or 0 if IOCB_NOIO) if nothing was read |
1da177e4 LT |
2758 | */ |
2759 | ssize_t | |
ed978a81 | 2760 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2761 | { |
e7080a43 | 2762 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2763 | ssize_t retval = 0; |
e7080a43 NS |
2764 | |
2765 | if (!count) | |
826ea860 | 2766 | return 0; /* skip atime */ |
1da177e4 | 2767 | |
2ba48ce5 | 2768 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2769 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2770 | struct address_space *mapping = file->f_mapping; |
2771 | struct inode *inode = mapping->host; | |
1da177e4 | 2772 | |
6be96d3a | 2773 | if (iocb->ki_flags & IOCB_NOWAIT) { |
7a60d6d7 JA |
2774 | if (filemap_range_needs_writeback(mapping, iocb->ki_pos, |
2775 | iocb->ki_pos + count - 1)) | |
6be96d3a GR |
2776 | return -EAGAIN; |
2777 | } else { | |
2778 | retval = filemap_write_and_wait_range(mapping, | |
2779 | iocb->ki_pos, | |
2780 | iocb->ki_pos + count - 1); | |
2781 | if (retval < 0) | |
826ea860 | 2782 | return retval; |
6be96d3a | 2783 | } |
d8d3d94b | 2784 | |
0d5b0cf2 CH |
2785 | file_accessed(file); |
2786 | ||
5ecda137 | 2787 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2788 | if (retval >= 0) { |
c64fb5c7 | 2789 | iocb->ki_pos += retval; |
5ecda137 | 2790 | count -= retval; |
9fe55eea | 2791 | } |
ab2125df PB |
2792 | if (retval != -EIOCBQUEUED) |
2793 | iov_iter_revert(iter, count - iov_iter_count(iter)); | |
66f998f6 | 2794 | |
9fe55eea SW |
2795 | /* |
2796 | * Btrfs can have a short DIO read if we encounter | |
2797 | * compressed extents, so if there was an error, or if | |
2798 | * we've already read everything we wanted to, or if | |
2799 | * there was a short read because we hit EOF, go ahead | |
2800 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2801 | * the rest of the read. Buffered reads will not work for |
2802 | * DAX files, so don't bother trying. | |
9fe55eea | 2803 | */ |
61d0017e JA |
2804 | if (retval < 0 || !count || IS_DAX(inode)) |
2805 | return retval; | |
2806 | if (iocb->ki_pos >= i_size_read(inode)) | |
826ea860 | 2807 | return retval; |
1da177e4 LT |
2808 | } |
2809 | ||
826ea860 | 2810 | return filemap_read(iocb, iter, retval); |
1da177e4 | 2811 | } |
ed978a81 | 2812 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2813 | |
07073eb0 DH |
2814 | /* |
2815 | * Splice subpages from a folio into a pipe. | |
2816 | */ | |
2817 | size_t splice_folio_into_pipe(struct pipe_inode_info *pipe, | |
2818 | struct folio *folio, loff_t fpos, size_t size) | |
2819 | { | |
2820 | struct page *page; | |
2821 | size_t spliced = 0, offset = offset_in_folio(folio, fpos); | |
2822 | ||
2823 | page = folio_page(folio, offset / PAGE_SIZE); | |
2824 | size = min(size, folio_size(folio) - offset); | |
2825 | offset %= PAGE_SIZE; | |
2826 | ||
2827 | while (spliced < size && | |
2828 | !pipe_full(pipe->head, pipe->tail, pipe->max_usage)) { | |
2829 | struct pipe_buffer *buf = pipe_head_buf(pipe); | |
2830 | size_t part = min_t(size_t, PAGE_SIZE - offset, size - spliced); | |
2831 | ||
2832 | *buf = (struct pipe_buffer) { | |
2833 | .ops = &page_cache_pipe_buf_ops, | |
2834 | .page = page, | |
2835 | .offset = offset, | |
2836 | .len = part, | |
2837 | }; | |
2838 | folio_get(folio); | |
2839 | pipe->head++; | |
2840 | page++; | |
2841 | spliced += part; | |
2842 | offset = 0; | |
2843 | } | |
2844 | ||
2845 | return spliced; | |
2846 | } | |
2847 | ||
2848 | /* | |
2849 | * Splice folios from the pagecache of a buffered (ie. non-O_DIRECT) file into | |
2850 | * a pipe. | |
2851 | */ | |
2852 | ssize_t filemap_splice_read(struct file *in, loff_t *ppos, | |
2853 | struct pipe_inode_info *pipe, | |
2854 | size_t len, unsigned int flags) | |
2855 | { | |
2856 | struct folio_batch fbatch; | |
2857 | struct kiocb iocb; | |
2858 | size_t total_spliced = 0, used, npages; | |
2859 | loff_t isize, end_offset; | |
2860 | bool writably_mapped; | |
2861 | int i, error = 0; | |
2862 | ||
2863 | init_sync_kiocb(&iocb, in); | |
2864 | iocb.ki_pos = *ppos; | |
2865 | ||
2866 | /* Work out how much data we can actually add into the pipe */ | |
2867 | used = pipe_occupancy(pipe->head, pipe->tail); | |
2868 | npages = max_t(ssize_t, pipe->max_usage - used, 0); | |
2869 | len = min_t(size_t, len, npages * PAGE_SIZE); | |
2870 | ||
2871 | folio_batch_init(&fbatch); | |
2872 | ||
2873 | do { | |
2874 | cond_resched(); | |
2875 | ||
2876 | if (*ppos >= i_size_read(file_inode(in))) | |
2877 | break; | |
2878 | ||
2879 | iocb.ki_pos = *ppos; | |
2880 | error = filemap_get_pages(&iocb, len, &fbatch, true); | |
2881 | if (error < 0) | |
2882 | break; | |
2883 | ||
2884 | /* | |
2885 | * i_size must be checked after we know the pages are Uptodate. | |
2886 | * | |
2887 | * Checking i_size after the check allows us to calculate | |
2888 | * the correct value for "nr", which means the zero-filled | |
2889 | * part of the page is not copied back to userspace (unless | |
2890 | * another truncate extends the file - this is desired though). | |
2891 | */ | |
2892 | isize = i_size_read(file_inode(in)); | |
2893 | if (unlikely(*ppos >= isize)) | |
2894 | break; | |
2895 | end_offset = min_t(loff_t, isize, *ppos + len); | |
2896 | ||
2897 | /* | |
2898 | * Once we start copying data, we don't want to be touching any | |
2899 | * cachelines that might be contended: | |
2900 | */ | |
2901 | writably_mapped = mapping_writably_mapped(in->f_mapping); | |
2902 | ||
2903 | for (i = 0; i < folio_batch_count(&fbatch); i++) { | |
2904 | struct folio *folio = fbatch.folios[i]; | |
2905 | size_t n; | |
2906 | ||
2907 | if (folio_pos(folio) >= end_offset) | |
2908 | goto out; | |
2909 | folio_mark_accessed(folio); | |
2910 | ||
2911 | /* | |
2912 | * If users can be writing to this folio using arbitrary | |
2913 | * virtual addresses, take care of potential aliasing | |
2914 | * before reading the folio on the kernel side. | |
2915 | */ | |
2916 | if (writably_mapped) | |
2917 | flush_dcache_folio(folio); | |
2918 | ||
2919 | n = min_t(loff_t, len, isize - *ppos); | |
2920 | n = splice_folio_into_pipe(pipe, folio, *ppos, n); | |
2921 | if (!n) | |
2922 | goto out; | |
2923 | len -= n; | |
2924 | total_spliced += n; | |
2925 | *ppos += n; | |
2926 | in->f_ra.prev_pos = *ppos; | |
2927 | if (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) | |
2928 | goto out; | |
2929 | } | |
2930 | ||
2931 | folio_batch_release(&fbatch); | |
2932 | } while (len); | |
2933 | ||
2934 | out: | |
2935 | folio_batch_release(&fbatch); | |
2936 | file_accessed(in); | |
2937 | ||
2938 | return total_spliced ? total_spliced : error; | |
2939 | } | |
7c8e01eb | 2940 | EXPORT_SYMBOL(filemap_splice_read); |
07073eb0 | 2941 | |
f5e6429a MWO |
2942 | static inline loff_t folio_seek_hole_data(struct xa_state *xas, |
2943 | struct address_space *mapping, struct folio *folio, | |
54fa39ac | 2944 | loff_t start, loff_t end, bool seek_data) |
41139aa4 | 2945 | { |
54fa39ac MWO |
2946 | const struct address_space_operations *ops = mapping->a_ops; |
2947 | size_t offset, bsz = i_blocksize(mapping->host); | |
2948 | ||
f5e6429a | 2949 | if (xa_is_value(folio) || folio_test_uptodate(folio)) |
54fa39ac MWO |
2950 | return seek_data ? start : end; |
2951 | if (!ops->is_partially_uptodate) | |
2952 | return seek_data ? end : start; | |
2953 | ||
2954 | xas_pause(xas); | |
2955 | rcu_read_unlock(); | |
f5e6429a MWO |
2956 | folio_lock(folio); |
2957 | if (unlikely(folio->mapping != mapping)) | |
54fa39ac MWO |
2958 | goto unlock; |
2959 | ||
f5e6429a | 2960 | offset = offset_in_folio(folio, start) & ~(bsz - 1); |
54fa39ac MWO |
2961 | |
2962 | do { | |
2e7e80f7 | 2963 | if (ops->is_partially_uptodate(folio, offset, bsz) == |
f5e6429a | 2964 | seek_data) |
54fa39ac MWO |
2965 | break; |
2966 | start = (start + bsz) & ~(bsz - 1); | |
2967 | offset += bsz; | |
f5e6429a | 2968 | } while (offset < folio_size(folio)); |
54fa39ac | 2969 | unlock: |
f5e6429a | 2970 | folio_unlock(folio); |
54fa39ac MWO |
2971 | rcu_read_lock(); |
2972 | return start; | |
41139aa4 MWO |
2973 | } |
2974 | ||
f5e6429a | 2975 | static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio) |
41139aa4 | 2976 | { |
f5e6429a | 2977 | if (xa_is_value(folio)) |
41139aa4 | 2978 | return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index); |
f5e6429a | 2979 | return folio_size(folio); |
41139aa4 MWO |
2980 | } |
2981 | ||
2982 | /** | |
2983 | * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache. | |
2984 | * @mapping: Address space to search. | |
2985 | * @start: First byte to consider. | |
2986 | * @end: Limit of search (exclusive). | |
2987 | * @whence: Either SEEK_HOLE or SEEK_DATA. | |
2988 | * | |
2989 | * If the page cache knows which blocks contain holes and which blocks | |
2990 | * contain data, your filesystem can use this function to implement | |
2991 | * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are | |
2992 | * entirely memory-based such as tmpfs, and filesystems which support | |
2993 | * unwritten extents. | |
2994 | * | |
f0953a1b | 2995 | * Return: The requested offset on success, or -ENXIO if @whence specifies |
41139aa4 MWO |
2996 | * SEEK_DATA and there is no data after @start. There is an implicit hole |
2997 | * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start | |
2998 | * and @end contain data. | |
2999 | */ | |
3000 | loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start, | |
3001 | loff_t end, int whence) | |
3002 | { | |
3003 | XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT); | |
ed98b015 | 3004 | pgoff_t max = (end - 1) >> PAGE_SHIFT; |
41139aa4 | 3005 | bool seek_data = (whence == SEEK_DATA); |
f5e6429a | 3006 | struct folio *folio; |
41139aa4 MWO |
3007 | |
3008 | if (end <= start) | |
3009 | return -ENXIO; | |
3010 | ||
3011 | rcu_read_lock(); | |
f5e6429a | 3012 | while ((folio = find_get_entry(&xas, max, XA_PRESENT))) { |
ed98b015 | 3013 | loff_t pos = (u64)xas.xa_index << PAGE_SHIFT; |
f5e6429a | 3014 | size_t seek_size; |
41139aa4 MWO |
3015 | |
3016 | if (start < pos) { | |
3017 | if (!seek_data) | |
3018 | goto unlock; | |
3019 | start = pos; | |
3020 | } | |
3021 | ||
f5e6429a MWO |
3022 | seek_size = seek_folio_size(&xas, folio); |
3023 | pos = round_up((u64)pos + 1, seek_size); | |
3024 | start = folio_seek_hole_data(&xas, mapping, folio, start, pos, | |
54fa39ac MWO |
3025 | seek_data); |
3026 | if (start < pos) | |
41139aa4 | 3027 | goto unlock; |
ed98b015 HD |
3028 | if (start >= end) |
3029 | break; | |
3030 | if (seek_size > PAGE_SIZE) | |
3031 | xas_set(&xas, pos >> PAGE_SHIFT); | |
f5e6429a MWO |
3032 | if (!xa_is_value(folio)) |
3033 | folio_put(folio); | |
41139aa4 | 3034 | } |
41139aa4 | 3035 | if (seek_data) |
ed98b015 | 3036 | start = -ENXIO; |
41139aa4 MWO |
3037 | unlock: |
3038 | rcu_read_unlock(); | |
f5e6429a MWO |
3039 | if (folio && !xa_is_value(folio)) |
3040 | folio_put(folio); | |
41139aa4 MWO |
3041 | if (start > end) |
3042 | return end; | |
3043 | return start; | |
3044 | } | |
3045 | ||
1da177e4 | 3046 | #ifdef CONFIG_MMU |
1da177e4 | 3047 | #define MMAP_LOTSAMISS (100) |
6b4c9f44 | 3048 | /* |
e292e6d6 | 3049 | * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock |
6b4c9f44 | 3050 | * @vmf - the vm_fault for this fault. |
e292e6d6 | 3051 | * @folio - the folio to lock. |
6b4c9f44 JB |
3052 | * @fpin - the pointer to the file we may pin (or is already pinned). |
3053 | * | |
e292e6d6 MWO |
3054 | * This works similar to lock_folio_or_retry in that it can drop the |
3055 | * mmap_lock. It differs in that it actually returns the folio locked | |
3056 | * if it returns 1 and 0 if it couldn't lock the folio. If we did have | |
3057 | * to drop the mmap_lock then fpin will point to the pinned file and | |
3058 | * needs to be fput()'ed at a later point. | |
6b4c9f44 | 3059 | */ |
e292e6d6 | 3060 | static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio, |
6b4c9f44 JB |
3061 | struct file **fpin) |
3062 | { | |
7c23c782 | 3063 | if (folio_trylock(folio)) |
6b4c9f44 JB |
3064 | return 1; |
3065 | ||
8b0f9fa2 LT |
3066 | /* |
3067 | * NOTE! This will make us return with VM_FAULT_RETRY, but with | |
c1e8d7c6 | 3068 | * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT |
8b0f9fa2 LT |
3069 | * is supposed to work. We have way too many special cases.. |
3070 | */ | |
6b4c9f44 JB |
3071 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
3072 | return 0; | |
3073 | ||
3074 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); | |
3075 | if (vmf->flags & FAULT_FLAG_KILLABLE) { | |
af7f29d9 | 3076 | if (__folio_lock_killable(folio)) { |
6b4c9f44 | 3077 | /* |
c1e8d7c6 | 3078 | * We didn't have the right flags to drop the mmap_lock, |
6b4c9f44 JB |
3079 | * but all fault_handlers only check for fatal signals |
3080 | * if we return VM_FAULT_RETRY, so we need to drop the | |
c1e8d7c6 | 3081 | * mmap_lock here and return 0 if we don't have a fpin. |
6b4c9f44 JB |
3082 | */ |
3083 | if (*fpin == NULL) | |
d8ed45c5 | 3084 | mmap_read_unlock(vmf->vma->vm_mm); |
6b4c9f44 JB |
3085 | return 0; |
3086 | } | |
3087 | } else | |
7c23c782 MWO |
3088 | __folio_lock(folio); |
3089 | ||
6b4c9f44 JB |
3090 | return 1; |
3091 | } | |
3092 | ||
ef00e08e | 3093 | /* |
6b4c9f44 JB |
3094 | * Synchronous readahead happens when we don't even find a page in the page |
3095 | * cache at all. We don't want to perform IO under the mmap sem, so if we have | |
3096 | * to drop the mmap sem we return the file that was pinned in order for us to do | |
3097 | * that. If we didn't pin a file then we return NULL. The file that is | |
3098 | * returned needs to be fput()'ed when we're done with it. | |
ef00e08e | 3099 | */ |
6b4c9f44 | 3100 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
ef00e08e | 3101 | { |
2a1180f1 JB |
3102 | struct file *file = vmf->vma->vm_file; |
3103 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 3104 | struct address_space *mapping = file->f_mapping; |
fcd9ae4f | 3105 | DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff); |
6b4c9f44 | 3106 | struct file *fpin = NULL; |
dcfa24ba | 3107 | unsigned long vm_flags = vmf->vma->vm_flags; |
e630bfac | 3108 | unsigned int mmap_miss; |
ef00e08e | 3109 | |
4687fdbb MWO |
3110 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3111 | /* Use the readahead code, even if readahead is disabled */ | |
dcfa24ba | 3112 | if (vm_flags & VM_HUGEPAGE) { |
4687fdbb MWO |
3113 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
3114 | ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1); | |
3115 | ra->size = HPAGE_PMD_NR; | |
3116 | /* | |
3117 | * Fetch two PMD folios, so we get the chance to actually | |
3118 | * readahead, unless we've been told not to. | |
3119 | */ | |
dcfa24ba | 3120 | if (!(vm_flags & VM_RAND_READ)) |
4687fdbb MWO |
3121 | ra->size *= 2; |
3122 | ra->async_size = HPAGE_PMD_NR; | |
3123 | page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER); | |
3124 | return fpin; | |
3125 | } | |
3126 | #endif | |
3127 | ||
ef00e08e | 3128 | /* If we don't want any read-ahead, don't bother */ |
dcfa24ba | 3129 | if (vm_flags & VM_RAND_READ) |
6b4c9f44 | 3130 | return fpin; |
275b12bf | 3131 | if (!ra->ra_pages) |
6b4c9f44 | 3132 | return fpin; |
ef00e08e | 3133 | |
dcfa24ba | 3134 | if (vm_flags & VM_SEQ_READ) { |
6b4c9f44 | 3135 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
fcd9ae4f | 3136 | page_cache_sync_ra(&ractl, ra->ra_pages); |
6b4c9f44 | 3137 | return fpin; |
ef00e08e LT |
3138 | } |
3139 | ||
207d04ba | 3140 | /* Avoid banging the cache line if not needed */ |
e630bfac KS |
3141 | mmap_miss = READ_ONCE(ra->mmap_miss); |
3142 | if (mmap_miss < MMAP_LOTSAMISS * 10) | |
3143 | WRITE_ONCE(ra->mmap_miss, ++mmap_miss); | |
ef00e08e LT |
3144 | |
3145 | /* | |
3146 | * Do we miss much more than hit in this file? If so, | |
3147 | * stop bothering with read-ahead. It will only hurt. | |
3148 | */ | |
e630bfac | 3149 | if (mmap_miss > MMAP_LOTSAMISS) |
6b4c9f44 | 3150 | return fpin; |
ef00e08e | 3151 | |
d30a1100 WF |
3152 | /* |
3153 | * mmap read-around | |
3154 | */ | |
6b4c9f44 | 3155 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
db660d46 | 3156 | ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2); |
600e19af RG |
3157 | ra->size = ra->ra_pages; |
3158 | ra->async_size = ra->ra_pages / 4; | |
db660d46 | 3159 | ractl._index = ra->start; |
56a4d67c | 3160 | page_cache_ra_order(&ractl, ra, 0); |
6b4c9f44 | 3161 | return fpin; |
ef00e08e LT |
3162 | } |
3163 | ||
3164 | /* | |
3165 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
6b4c9f44 | 3166 | * so we want to possibly extend the readahead further. We return the file that |
c1e8d7c6 | 3167 | * was pinned if we have to drop the mmap_lock in order to do IO. |
ef00e08e | 3168 | */ |
6b4c9f44 | 3169 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
79598ced | 3170 | struct folio *folio) |
ef00e08e | 3171 | { |
2a1180f1 JB |
3172 | struct file *file = vmf->vma->vm_file; |
3173 | struct file_ra_state *ra = &file->f_ra; | |
79598ced | 3174 | DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff); |
6b4c9f44 | 3175 | struct file *fpin = NULL; |
e630bfac | 3176 | unsigned int mmap_miss; |
ef00e08e LT |
3177 | |
3178 | /* If we don't want any read-ahead, don't bother */ | |
5c72feee | 3179 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
6b4c9f44 | 3180 | return fpin; |
79598ced | 3181 | |
e630bfac KS |
3182 | mmap_miss = READ_ONCE(ra->mmap_miss); |
3183 | if (mmap_miss) | |
3184 | WRITE_ONCE(ra->mmap_miss, --mmap_miss); | |
79598ced MWO |
3185 | |
3186 | if (folio_test_readahead(folio)) { | |
6b4c9f44 | 3187 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
79598ced | 3188 | page_cache_async_ra(&ractl, folio, ra->ra_pages); |
6b4c9f44 JB |
3189 | } |
3190 | return fpin; | |
ef00e08e LT |
3191 | } |
3192 | ||
485bb99b | 3193 | /** |
54cb8821 | 3194 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 3195 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 3196 | * |
54cb8821 | 3197 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
3198 | * mapped memory region to read in file data during a page fault. |
3199 | * | |
3200 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
3201 | * it in the page cache, and handles the special cases reasonably without | |
3202 | * having a lot of duplicated code. | |
9a95f3cf | 3203 | * |
c1e8d7c6 | 3204 | * vma->vm_mm->mmap_lock must be held on entry. |
9a95f3cf | 3205 | * |
c1e8d7c6 | 3206 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock |
e292e6d6 | 3207 | * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap(). |
9a95f3cf | 3208 | * |
c1e8d7c6 | 3209 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock |
9a95f3cf PC |
3210 | * has not been released. |
3211 | * | |
3212 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
a862f68a MR |
3213 | * |
3214 | * Return: bitwise-OR of %VM_FAULT_ codes. | |
1da177e4 | 3215 | */ |
2bcd6454 | 3216 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
3217 | { |
3218 | int error; | |
11bac800 | 3219 | struct file *file = vmf->vma->vm_file; |
6b4c9f44 | 3220 | struct file *fpin = NULL; |
1da177e4 | 3221 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3222 | struct inode *inode = mapping->host; |
e292e6d6 MWO |
3223 | pgoff_t max_idx, index = vmf->pgoff; |
3224 | struct folio *folio; | |
2bcd6454 | 3225 | vm_fault_t ret = 0; |
730633f0 | 3226 | bool mapping_locked = false; |
1da177e4 | 3227 | |
e292e6d6 MWO |
3228 | max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3229 | if (unlikely(index >= max_idx)) | |
5307cc1a | 3230 | return VM_FAULT_SIGBUS; |
1da177e4 | 3231 | |
1da177e4 | 3232 | /* |
49426420 | 3233 | * Do we have something in the page cache already? |
1da177e4 | 3234 | */ |
e292e6d6 | 3235 | folio = filemap_get_folio(mapping, index); |
66dabbb6 | 3236 | if (likely(!IS_ERR(folio))) { |
1da177e4 | 3237 | /* |
730633f0 JK |
3238 | * We found the page, so try async readahead before waiting for |
3239 | * the lock. | |
1da177e4 | 3240 | */ |
730633f0 | 3241 | if (!(vmf->flags & FAULT_FLAG_TRIED)) |
79598ced | 3242 | fpin = do_async_mmap_readahead(vmf, folio); |
e292e6d6 | 3243 | if (unlikely(!folio_test_uptodate(folio))) { |
730633f0 JK |
3244 | filemap_invalidate_lock_shared(mapping); |
3245 | mapping_locked = true; | |
3246 | } | |
3247 | } else { | |
ef00e08e | 3248 | /* No page in the page cache at all */ |
ef00e08e | 3249 | count_vm_event(PGMAJFAULT); |
2262185c | 3250 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e | 3251 | ret = VM_FAULT_MAJOR; |
6b4c9f44 | 3252 | fpin = do_sync_mmap_readahead(vmf); |
ef00e08e | 3253 | retry_find: |
730633f0 | 3254 | /* |
e292e6d6 | 3255 | * See comment in filemap_create_folio() why we need |
730633f0 JK |
3256 | * invalidate_lock |
3257 | */ | |
3258 | if (!mapping_locked) { | |
3259 | filemap_invalidate_lock_shared(mapping); | |
3260 | mapping_locked = true; | |
3261 | } | |
e292e6d6 | 3262 | folio = __filemap_get_folio(mapping, index, |
a75d4c33 JB |
3263 | FGP_CREAT|FGP_FOR_MMAP, |
3264 | vmf->gfp_mask); | |
66dabbb6 | 3265 | if (IS_ERR(folio)) { |
6b4c9f44 JB |
3266 | if (fpin) |
3267 | goto out_retry; | |
730633f0 | 3268 | filemap_invalidate_unlock_shared(mapping); |
e520e932 | 3269 | return VM_FAULT_OOM; |
6b4c9f44 | 3270 | } |
1da177e4 LT |
3271 | } |
3272 | ||
e292e6d6 | 3273 | if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin)) |
6b4c9f44 | 3274 | goto out_retry; |
b522c94d ML |
3275 | |
3276 | /* Did it get truncated? */ | |
e292e6d6 MWO |
3277 | if (unlikely(folio->mapping != mapping)) { |
3278 | folio_unlock(folio); | |
3279 | folio_put(folio); | |
b522c94d ML |
3280 | goto retry_find; |
3281 | } | |
e292e6d6 | 3282 | VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); |
b522c94d | 3283 | |
1da177e4 | 3284 | /* |
d00806b1 NP |
3285 | * We have a locked page in the page cache, now we need to check |
3286 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 3287 | */ |
e292e6d6 | 3288 | if (unlikely(!folio_test_uptodate(folio))) { |
730633f0 JK |
3289 | /* |
3290 | * The page was in cache and uptodate and now it is not. | |
3291 | * Strange but possible since we didn't hold the page lock all | |
3292 | * the time. Let's drop everything get the invalidate lock and | |
3293 | * try again. | |
3294 | */ | |
3295 | if (!mapping_locked) { | |
e292e6d6 MWO |
3296 | folio_unlock(folio); |
3297 | folio_put(folio); | |
730633f0 JK |
3298 | goto retry_find; |
3299 | } | |
1da177e4 | 3300 | goto page_not_uptodate; |
730633f0 | 3301 | } |
1da177e4 | 3302 | |
6b4c9f44 | 3303 | /* |
c1e8d7c6 | 3304 | * We've made it this far and we had to drop our mmap_lock, now is the |
6b4c9f44 JB |
3305 | * time to return to the upper layer and have it re-find the vma and |
3306 | * redo the fault. | |
3307 | */ | |
3308 | if (fpin) { | |
e292e6d6 | 3309 | folio_unlock(folio); |
6b4c9f44 JB |
3310 | goto out_retry; |
3311 | } | |
730633f0 JK |
3312 | if (mapping_locked) |
3313 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 | 3314 | |
ef00e08e LT |
3315 | /* |
3316 | * Found the page and have a reference on it. | |
3317 | * We must recheck i_size under page lock. | |
3318 | */ | |
e292e6d6 MWO |
3319 | max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3320 | if (unlikely(index >= max_idx)) { | |
3321 | folio_unlock(folio); | |
3322 | folio_put(folio); | |
5307cc1a | 3323 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
3324 | } |
3325 | ||
e292e6d6 | 3326 | vmf->page = folio_file_page(folio, index); |
83c54070 | 3327 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 3328 | |
1da177e4 | 3329 | page_not_uptodate: |
1da177e4 LT |
3330 | /* |
3331 | * Umm, take care of errors if the page isn't up-to-date. | |
3332 | * Try to re-read it _once_. We do this synchronously, | |
3333 | * because there really aren't any performance issues here | |
3334 | * and we need to check for errors. | |
3335 | */ | |
6b4c9f44 | 3336 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
290e1a32 | 3337 | error = filemap_read_folio(file, mapping->a_ops->read_folio, folio); |
6b4c9f44 JB |
3338 | if (fpin) |
3339 | goto out_retry; | |
e292e6d6 | 3340 | folio_put(folio); |
d00806b1 NP |
3341 | |
3342 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 3343 | goto retry_find; |
730633f0 | 3344 | filemap_invalidate_unlock_shared(mapping); |
1da177e4 | 3345 | |
d0217ac0 | 3346 | return VM_FAULT_SIGBUS; |
6b4c9f44 JB |
3347 | |
3348 | out_retry: | |
3349 | /* | |
c1e8d7c6 | 3350 | * We dropped the mmap_lock, we need to return to the fault handler to |
6b4c9f44 JB |
3351 | * re-find the vma and come back and find our hopefully still populated |
3352 | * page. | |
3353 | */ | |
38a55db9 | 3354 | if (!IS_ERR(folio)) |
e292e6d6 | 3355 | folio_put(folio); |
730633f0 JK |
3356 | if (mapping_locked) |
3357 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 JB |
3358 | if (fpin) |
3359 | fput(fpin); | |
3360 | return ret | VM_FAULT_RETRY; | |
54cb8821 NP |
3361 | } |
3362 | EXPORT_SYMBOL(filemap_fault); | |
3363 | ||
8808ecab MWO |
3364 | static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio, |
3365 | pgoff_t start) | |
f1820361 | 3366 | { |
f9ce0be7 KS |
3367 | struct mm_struct *mm = vmf->vma->vm_mm; |
3368 | ||
3369 | /* Huge page is mapped? No need to proceed. */ | |
3370 | if (pmd_trans_huge(*vmf->pmd)) { | |
8808ecab MWO |
3371 | folio_unlock(folio); |
3372 | folio_put(folio); | |
f9ce0be7 KS |
3373 | return true; |
3374 | } | |
3375 | ||
8808ecab MWO |
3376 | if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) { |
3377 | struct page *page = folio_file_page(folio, start); | |
e0f43fa5 YS |
3378 | vm_fault_t ret = do_set_pmd(vmf, page); |
3379 | if (!ret) { | |
3380 | /* The page is mapped successfully, reference consumed. */ | |
8808ecab | 3381 | folio_unlock(folio); |
e0f43fa5 | 3382 | return true; |
f9ce0be7 | 3383 | } |
f9ce0be7 KS |
3384 | } |
3385 | ||
03c4f204 QZ |
3386 | if (pmd_none(*vmf->pmd)) |
3387 | pmd_install(mm, vmf->pmd, &vmf->prealloc_pte); | |
f9ce0be7 KS |
3388 | |
3389 | /* See comment in handle_pte_fault() */ | |
3390 | if (pmd_devmap_trans_unstable(vmf->pmd)) { | |
8808ecab MWO |
3391 | folio_unlock(folio); |
3392 | folio_put(folio); | |
f9ce0be7 KS |
3393 | return true; |
3394 | } | |
3395 | ||
3396 | return false; | |
3397 | } | |
3398 | ||
820b05e9 | 3399 | static struct folio *next_uptodate_page(struct folio *folio, |
f9ce0be7 KS |
3400 | struct address_space *mapping, |
3401 | struct xa_state *xas, pgoff_t end_pgoff) | |
3402 | { | |
3403 | unsigned long max_idx; | |
3404 | ||
3405 | do { | |
9184a307 | 3406 | if (!folio) |
f9ce0be7 | 3407 | return NULL; |
9184a307 | 3408 | if (xas_retry(xas, folio)) |
f9ce0be7 | 3409 | continue; |
9184a307 | 3410 | if (xa_is_value(folio)) |
f9ce0be7 | 3411 | continue; |
9184a307 | 3412 | if (folio_test_locked(folio)) |
f9ce0be7 | 3413 | continue; |
9184a307 | 3414 | if (!folio_try_get_rcu(folio)) |
f9ce0be7 KS |
3415 | continue; |
3416 | /* Has the page moved or been split? */ | |
9184a307 | 3417 | if (unlikely(folio != xas_reload(xas))) |
f9ce0be7 | 3418 | goto skip; |
9184a307 | 3419 | if (!folio_test_uptodate(folio) || folio_test_readahead(folio)) |
f9ce0be7 | 3420 | goto skip; |
9184a307 | 3421 | if (!folio_trylock(folio)) |
f9ce0be7 | 3422 | goto skip; |
9184a307 | 3423 | if (folio->mapping != mapping) |
f9ce0be7 | 3424 | goto unlock; |
9184a307 | 3425 | if (!folio_test_uptodate(folio)) |
f9ce0be7 KS |
3426 | goto unlock; |
3427 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); | |
3428 | if (xas->xa_index >= max_idx) | |
3429 | goto unlock; | |
820b05e9 | 3430 | return folio; |
f9ce0be7 | 3431 | unlock: |
9184a307 | 3432 | folio_unlock(folio); |
f9ce0be7 | 3433 | skip: |
9184a307 MWO |
3434 | folio_put(folio); |
3435 | } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL); | |
f9ce0be7 KS |
3436 | |
3437 | return NULL; | |
3438 | } | |
3439 | ||
820b05e9 | 3440 | static inline struct folio *first_map_page(struct address_space *mapping, |
f9ce0be7 KS |
3441 | struct xa_state *xas, |
3442 | pgoff_t end_pgoff) | |
3443 | { | |
3444 | return next_uptodate_page(xas_find(xas, end_pgoff), | |
3445 | mapping, xas, end_pgoff); | |
3446 | } | |
3447 | ||
820b05e9 | 3448 | static inline struct folio *next_map_page(struct address_space *mapping, |
f9ce0be7 KS |
3449 | struct xa_state *xas, |
3450 | pgoff_t end_pgoff) | |
3451 | { | |
3452 | return next_uptodate_page(xas_next_entry(xas, end_pgoff), | |
3453 | mapping, xas, end_pgoff); | |
3454 | } | |
3455 | ||
3456 | vm_fault_t filemap_map_pages(struct vm_fault *vmf, | |
3457 | pgoff_t start_pgoff, pgoff_t end_pgoff) | |
3458 | { | |
3459 | struct vm_area_struct *vma = vmf->vma; | |
3460 | struct file *file = vma->vm_file; | |
f1820361 | 3461 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 3462 | pgoff_t last_pgoff = start_pgoff; |
9d3af4b4 | 3463 | unsigned long addr; |
070e807c | 3464 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
820b05e9 MWO |
3465 | struct folio *folio; |
3466 | struct page *page; | |
e630bfac | 3467 | unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss); |
f9ce0be7 | 3468 | vm_fault_t ret = 0; |
f1820361 KS |
3469 | |
3470 | rcu_read_lock(); | |
820b05e9 MWO |
3471 | folio = first_map_page(mapping, &xas, end_pgoff); |
3472 | if (!folio) | |
f9ce0be7 | 3473 | goto out; |
f1820361 | 3474 | |
8808ecab | 3475 | if (filemap_map_pmd(vmf, folio, start_pgoff)) { |
f9ce0be7 KS |
3476 | ret = VM_FAULT_NOPAGE; |
3477 | goto out; | |
3478 | } | |
f1820361 | 3479 | |
9d3af4b4 WD |
3480 | addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
3481 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); | |
f9ce0be7 | 3482 | do { |
6b24ca4a | 3483 | again: |
820b05e9 | 3484 | page = folio_file_page(folio, xas.xa_index); |
f9ce0be7 | 3485 | if (PageHWPoison(page)) |
f1820361 KS |
3486 | goto unlock; |
3487 | ||
e630bfac KS |
3488 | if (mmap_miss > 0) |
3489 | mmap_miss--; | |
7267ec00 | 3490 | |
9d3af4b4 | 3491 | addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
f9ce0be7 | 3492 | vmf->pte += xas.xa_index - last_pgoff; |
070e807c | 3493 | last_pgoff = xas.xa_index; |
f9ce0be7 | 3494 | |
5c041f5d PX |
3495 | /* |
3496 | * NOTE: If there're PTE markers, we'll leave them to be | |
3497 | * handled in the specific fault path, and it'll prohibit the | |
3498 | * fault-around logic. | |
3499 | */ | |
f9ce0be7 | 3500 | if (!pte_none(*vmf->pte)) |
7267ec00 | 3501 | goto unlock; |
f9ce0be7 | 3502 | |
46bdb427 | 3503 | /* We're about to handle the fault */ |
9d3af4b4 | 3504 | if (vmf->address == addr) |
46bdb427 | 3505 | ret = VM_FAULT_NOPAGE; |
46bdb427 | 3506 | |
9d3af4b4 | 3507 | do_set_pte(vmf, page, addr); |
f9ce0be7 | 3508 | /* no need to invalidate: a not-present page won't be cached */ |
9d3af4b4 | 3509 | update_mmu_cache(vma, addr, vmf->pte); |
6b24ca4a MWO |
3510 | if (folio_more_pages(folio, xas.xa_index, end_pgoff)) { |
3511 | xas.xa_index++; | |
3512 | folio_ref_inc(folio); | |
3513 | goto again; | |
3514 | } | |
820b05e9 | 3515 | folio_unlock(folio); |
f9ce0be7 | 3516 | continue; |
f1820361 | 3517 | unlock: |
6b24ca4a MWO |
3518 | if (folio_more_pages(folio, xas.xa_index, end_pgoff)) { |
3519 | xas.xa_index++; | |
3520 | goto again; | |
3521 | } | |
820b05e9 MWO |
3522 | folio_unlock(folio); |
3523 | folio_put(folio); | |
3524 | } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL); | |
f9ce0be7 KS |
3525 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3526 | out: | |
f1820361 | 3527 | rcu_read_unlock(); |
e630bfac | 3528 | WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss); |
f9ce0be7 | 3529 | return ret; |
f1820361 KS |
3530 | } |
3531 | EXPORT_SYMBOL(filemap_map_pages); | |
3532 | ||
2bcd6454 | 3533 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 | 3534 | { |
5df1a672 | 3535 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
960ea971 | 3536 | struct folio *folio = page_folio(vmf->page); |
2bcd6454 | 3537 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 3538 | |
5df1a672 | 3539 | sb_start_pagefault(mapping->host->i_sb); |
11bac800 | 3540 | file_update_time(vmf->vma->vm_file); |
960ea971 MWO |
3541 | folio_lock(folio); |
3542 | if (folio->mapping != mapping) { | |
3543 | folio_unlock(folio); | |
4fcf1c62 JK |
3544 | ret = VM_FAULT_NOPAGE; |
3545 | goto out; | |
3546 | } | |
14da9200 | 3547 | /* |
960ea971 | 3548 | * We mark the folio dirty already here so that when freeze is in |
14da9200 | 3549 | * progress, we are guaranteed that writeback during freezing will |
960ea971 | 3550 | * see the dirty folio and writeprotect it again. |
14da9200 | 3551 | */ |
960ea971 MWO |
3552 | folio_mark_dirty(folio); |
3553 | folio_wait_stable(folio); | |
4fcf1c62 | 3554 | out: |
5df1a672 | 3555 | sb_end_pagefault(mapping->host->i_sb); |
4fcf1c62 JK |
3556 | return ret; |
3557 | } | |
4fcf1c62 | 3558 | |
f0f37e2f | 3559 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 3560 | .fault = filemap_fault, |
f1820361 | 3561 | .map_pages = filemap_map_pages, |
4fcf1c62 | 3562 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
3563 | }; |
3564 | ||
3565 | /* This is used for a general mmap of a disk file */ | |
3566 | ||
68d68ff6 | 3567 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3568 | { |
3569 | struct address_space *mapping = file->f_mapping; | |
3570 | ||
7e0a1265 | 3571 | if (!mapping->a_ops->read_folio) |
1da177e4 LT |
3572 | return -ENOEXEC; |
3573 | file_accessed(file); | |
3574 | vma->vm_ops = &generic_file_vm_ops; | |
3575 | return 0; | |
3576 | } | |
1da177e4 LT |
3577 | |
3578 | /* | |
3579 | * This is for filesystems which do not implement ->writepage. | |
3580 | */ | |
3581 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
3582 | { | |
3583 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
3584 | return -EINVAL; | |
3585 | return generic_file_mmap(file, vma); | |
3586 | } | |
3587 | #else | |
4b96a37d | 3588 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
45397228 | 3589 | { |
4b96a37d | 3590 | return VM_FAULT_SIGBUS; |
45397228 | 3591 | } |
68d68ff6 | 3592 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3593 | { |
3594 | return -ENOSYS; | |
3595 | } | |
68d68ff6 | 3596 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3597 | { |
3598 | return -ENOSYS; | |
3599 | } | |
3600 | #endif /* CONFIG_MMU */ | |
3601 | ||
45397228 | 3602 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
3603 | EXPORT_SYMBOL(generic_file_mmap); |
3604 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
3605 | ||
539a3322 | 3606 | static struct folio *do_read_cache_folio(struct address_space *mapping, |
e9b5b23e | 3607 | pgoff_t index, filler_t filler, struct file *file, gfp_t gfp) |
67f9fd91 | 3608 | { |
539a3322 | 3609 | struct folio *folio; |
1da177e4 | 3610 | int err; |
07950008 MWO |
3611 | |
3612 | if (!filler) | |
3613 | filler = mapping->a_ops->read_folio; | |
1da177e4 | 3614 | repeat: |
539a3322 | 3615 | folio = filemap_get_folio(mapping, index); |
66dabbb6 | 3616 | if (IS_ERR(folio)) { |
539a3322 MWO |
3617 | folio = filemap_alloc_folio(gfp, 0); |
3618 | if (!folio) | |
eb2be189 | 3619 | return ERR_PTR(-ENOMEM); |
539a3322 | 3620 | err = filemap_add_folio(mapping, folio, index, gfp); |
eb2be189 | 3621 | if (unlikely(err)) { |
539a3322 | 3622 | folio_put(folio); |
eb2be189 NP |
3623 | if (err == -EEXIST) |
3624 | goto repeat; | |
22ecdb4f | 3625 | /* Presumably ENOMEM for xarray node */ |
1da177e4 LT |
3626 | return ERR_PTR(err); |
3627 | } | |
32b63529 | 3628 | |
9bc3e869 | 3629 | goto filler; |
32b63529 | 3630 | } |
539a3322 | 3631 | if (folio_test_uptodate(folio)) |
1da177e4 LT |
3632 | goto out; |
3633 | ||
81f4c03b MWO |
3634 | if (!folio_trylock(folio)) { |
3635 | folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE); | |
3636 | goto repeat; | |
3637 | } | |
ebded027 | 3638 | |
81f4c03b | 3639 | /* Folio was truncated from mapping */ |
539a3322 MWO |
3640 | if (!folio->mapping) { |
3641 | folio_unlock(folio); | |
3642 | folio_put(folio); | |
32b63529 | 3643 | goto repeat; |
1da177e4 | 3644 | } |
ebded027 MG |
3645 | |
3646 | /* Someone else locked and filled the page in a very small window */ | |
539a3322 MWO |
3647 | if (folio_test_uptodate(folio)) { |
3648 | folio_unlock(folio); | |
1da177e4 LT |
3649 | goto out; |
3650 | } | |
faffdfa0 | 3651 | |
9bc3e869 | 3652 | filler: |
290e1a32 | 3653 | err = filemap_read_folio(file, filler, folio); |
1dfa24a4 | 3654 | if (err) { |
9bc3e869 | 3655 | folio_put(folio); |
1dfa24a4 MWO |
3656 | if (err == AOP_TRUNCATED_PAGE) |
3657 | goto repeat; | |
9bc3e869 MWO |
3658 | return ERR_PTR(err); |
3659 | } | |
32b63529 | 3660 | |
c855ff37 | 3661 | out: |
539a3322 MWO |
3662 | folio_mark_accessed(folio); |
3663 | return folio; | |
6fe6900e | 3664 | } |
0531b2aa LT |
3665 | |
3666 | /** | |
e9b5b23e MWO |
3667 | * read_cache_folio - Read into page cache, fill it if needed. |
3668 | * @mapping: The address_space to read from. | |
3669 | * @index: The index to read. | |
3670 | * @filler: Function to perform the read, or NULL to use aops->read_folio(). | |
3671 | * @file: Passed to filler function, may be NULL if not required. | |
0531b2aa | 3672 | * |
e9b5b23e MWO |
3673 | * Read one page into the page cache. If it succeeds, the folio returned |
3674 | * will contain @index, but it may not be the first page of the folio. | |
a862f68a | 3675 | * |
e9b5b23e MWO |
3676 | * If the filler function returns an error, it will be returned to the |
3677 | * caller. | |
730633f0 | 3678 | * |
e9b5b23e MWO |
3679 | * Context: May sleep. Expects mapping->invalidate_lock to be held. |
3680 | * Return: An uptodate folio on success, ERR_PTR() on failure. | |
0531b2aa | 3681 | */ |
539a3322 | 3682 | struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index, |
e9b5b23e | 3683 | filler_t filler, struct file *file) |
539a3322 | 3684 | { |
e9b5b23e | 3685 | return do_read_cache_folio(mapping, index, filler, file, |
539a3322 MWO |
3686 | mapping_gfp_mask(mapping)); |
3687 | } | |
3688 | EXPORT_SYMBOL(read_cache_folio); | |
3689 | ||
3e629597 MWO |
3690 | /** |
3691 | * mapping_read_folio_gfp - Read into page cache, using specified allocation flags. | |
3692 | * @mapping: The address_space for the folio. | |
3693 | * @index: The index that the allocated folio will contain. | |
3694 | * @gfp: The page allocator flags to use if allocating. | |
3695 | * | |
3696 | * This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with | |
3697 | * any new memory allocations done using the specified allocation flags. | |
3698 | * | |
3699 | * The most likely error from this function is EIO, but ENOMEM is | |
3700 | * possible and so is EINTR. If ->read_folio returns another error, | |
3701 | * that will be returned to the caller. | |
3702 | * | |
3703 | * The function expects mapping->invalidate_lock to be already held. | |
3704 | * | |
3705 | * Return: Uptodate folio on success, ERR_PTR() on failure. | |
3706 | */ | |
3707 | struct folio *mapping_read_folio_gfp(struct address_space *mapping, | |
3708 | pgoff_t index, gfp_t gfp) | |
3709 | { | |
3710 | return do_read_cache_folio(mapping, index, NULL, NULL, gfp); | |
3711 | } | |
3712 | EXPORT_SYMBOL(mapping_read_folio_gfp); | |
3713 | ||
539a3322 | 3714 | static struct page *do_read_cache_page(struct address_space *mapping, |
e9b5b23e | 3715 | pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp) |
539a3322 MWO |
3716 | { |
3717 | struct folio *folio; | |
3718 | ||
e9b5b23e | 3719 | folio = do_read_cache_folio(mapping, index, filler, file, gfp); |
539a3322 MWO |
3720 | if (IS_ERR(folio)) |
3721 | return &folio->page; | |
3722 | return folio_file_page(folio, index); | |
3723 | } | |
3724 | ||
67f9fd91 | 3725 | struct page *read_cache_page(struct address_space *mapping, |
e9b5b23e | 3726 | pgoff_t index, filler_t *filler, struct file *file) |
0531b2aa | 3727 | { |
e9b5b23e | 3728 | return do_read_cache_page(mapping, index, filler, file, |
d322a8e5 | 3729 | mapping_gfp_mask(mapping)); |
0531b2aa | 3730 | } |
67f9fd91 | 3731 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
3732 | |
3733 | /** | |
3734 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
3735 | * @mapping: the page's address_space | |
3736 | * @index: the page index | |
3737 | * @gfp: the page allocator flags to use if allocating | |
3738 | * | |
3739 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 3740 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
3741 | * |
3742 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3743 | * |
730633f0 JK |
3744 | * The function expects mapping->invalidate_lock to be already held. |
3745 | * | |
a862f68a | 3746 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa LT |
3747 | */ |
3748 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
3749 | pgoff_t index, | |
3750 | gfp_t gfp) | |
3751 | { | |
6c45b454 | 3752 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
0531b2aa LT |
3753 | } |
3754 | EXPORT_SYMBOL(read_cache_page_gfp); | |
3755 | ||
a92853b6 KK |
3756 | /* |
3757 | * Warn about a page cache invalidation failure during a direct I/O write. | |
3758 | */ | |
3759 | void dio_warn_stale_pagecache(struct file *filp) | |
3760 | { | |
3761 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); | |
3762 | char pathname[128]; | |
a92853b6 KK |
3763 | char *path; |
3764 | ||
5df1a672 | 3765 | errseq_set(&filp->f_mapping->wb_err, -EIO); |
a92853b6 KK |
3766 | if (__ratelimit(&_rs)) { |
3767 | path = file_path(filp, pathname, sizeof(pathname)); | |
3768 | if (IS_ERR(path)) | |
3769 | path = "(unknown)"; | |
3770 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); | |
3771 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, | |
3772 | current->comm); | |
3773 | } | |
3774 | } | |
3775 | ||
1da177e4 | 3776 | ssize_t |
1af5bb49 | 3777 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3778 | { |
3779 | struct file *file = iocb->ki_filp; | |
3780 | struct address_space *mapping = file->f_mapping; | |
3781 | struct inode *inode = mapping->host; | |
1af5bb49 | 3782 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3783 | ssize_t written; |
a969e903 CH |
3784 | size_t write_len; |
3785 | pgoff_t end; | |
1da177e4 | 3786 | |
0c949334 | 3787 | write_len = iov_iter_count(from); |
09cbfeaf | 3788 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3789 | |
6be96d3a GR |
3790 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3791 | /* If there are pages to writeback, return */ | |
5df1a672 | 3792 | if (filemap_range_has_page(file->f_mapping, pos, |
35f12f0f | 3793 | pos + write_len - 1)) |
6be96d3a GR |
3794 | return -EAGAIN; |
3795 | } else { | |
3796 | written = filemap_write_and_wait_range(mapping, pos, | |
3797 | pos + write_len - 1); | |
3798 | if (written) | |
3799 | goto out; | |
3800 | } | |
a969e903 CH |
3801 | |
3802 | /* | |
3803 | * After a write we want buffered reads to be sure to go to disk to get | |
3804 | * the new data. We invalidate clean cached page from the region we're | |
3805 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3806 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3807 | */ |
55635ba7 | 3808 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3809 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3810 | /* |
3811 | * If a page can not be invalidated, return 0 to fall back | |
3812 | * to buffered write. | |
3813 | */ | |
3814 | if (written) { | |
3815 | if (written == -EBUSY) | |
3816 | return 0; | |
3817 | goto out; | |
a969e903 CH |
3818 | } |
3819 | ||
639a93a5 | 3820 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3821 | |
3822 | /* | |
3823 | * Finally, try again to invalidate clean pages which might have been | |
3824 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3825 | * if the source of the write was an mmap'ed region of the file | |
3826 | * we're writing. Either one is a pretty crazy thing to do, | |
3827 | * so we don't support it 100%. If this invalidation | |
3828 | * fails, tough, the write still worked... | |
332391a9 LC |
3829 | * |
3830 | * Most of the time we do not need this since dio_complete() will do | |
3831 | * the invalidation for us. However there are some file systems that | |
3832 | * do not end up with dio_complete() being called, so let's not break | |
80c1fe90 KK |
3833 | * them by removing it completely. |
3834 | * | |
9266a140 KK |
3835 | * Noticeable example is a blkdev_direct_IO(). |
3836 | * | |
80c1fe90 | 3837 | * Skip invalidation for async writes or if mapping has no pages. |
a969e903 | 3838 | */ |
9266a140 KK |
3839 | if (written > 0 && mapping->nrpages && |
3840 | invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) | |
3841 | dio_warn_stale_pagecache(file); | |
a969e903 | 3842 | |
1da177e4 | 3843 | if (written > 0) { |
0116651c | 3844 | pos += written; |
639a93a5 | 3845 | write_len -= written; |
0116651c NK |
3846 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3847 | i_size_write(inode, pos); | |
1da177e4 LT |
3848 | mark_inode_dirty(inode); |
3849 | } | |
5cb6c6c7 | 3850 | iocb->ki_pos = pos; |
1da177e4 | 3851 | } |
ab2125df PB |
3852 | if (written != -EIOCBQUEUED) |
3853 | iov_iter_revert(from, write_len - iov_iter_count(from)); | |
a969e903 | 3854 | out: |
1da177e4 LT |
3855 | return written; |
3856 | } | |
3857 | EXPORT_SYMBOL(generic_file_direct_write); | |
3858 | ||
800ba295 | 3859 | ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i) |
afddba49 | 3860 | { |
800ba295 MWO |
3861 | struct file *file = iocb->ki_filp; |
3862 | loff_t pos = iocb->ki_pos; | |
afddba49 NP |
3863 | struct address_space *mapping = file->f_mapping; |
3864 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3865 | long status = 0; | |
3866 | ssize_t written = 0; | |
674b892e | 3867 | |
afddba49 NP |
3868 | do { |
3869 | struct page *page; | |
afddba49 NP |
3870 | unsigned long offset; /* Offset into pagecache page */ |
3871 | unsigned long bytes; /* Bytes to write to page */ | |
3872 | size_t copied; /* Bytes copied from user */ | |
1468c6f4 | 3873 | void *fsdata = NULL; |
afddba49 | 3874 | |
09cbfeaf KS |
3875 | offset = (pos & (PAGE_SIZE - 1)); |
3876 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3877 | iov_iter_count(i)); |
3878 | ||
3879 | again: | |
00a3d660 LT |
3880 | /* |
3881 | * Bring in the user page that we will copy from _first_. | |
3882 | * Otherwise there's a nasty deadlock on copying from the | |
3883 | * same page as we're writing to, without it being marked | |
3884 | * up-to-date. | |
00a3d660 | 3885 | */ |
631f871f | 3886 | if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) { |
00a3d660 LT |
3887 | status = -EFAULT; |
3888 | break; | |
3889 | } | |
3890 | ||
296291cd JK |
3891 | if (fatal_signal_pending(current)) { |
3892 | status = -EINTR; | |
3893 | break; | |
3894 | } | |
3895 | ||
9d6b0cd7 | 3896 | status = a_ops->write_begin(file, mapping, pos, bytes, |
afddba49 | 3897 | &page, &fsdata); |
2457aec6 | 3898 | if (unlikely(status < 0)) |
afddba49 NP |
3899 | break; |
3900 | ||
931e80e4 | 3901 | if (mapping_writably_mapped(mapping)) |
3902 | flush_dcache_page(page); | |
00a3d660 | 3903 | |
f0b65f39 | 3904 | copied = copy_page_from_iter_atomic(page, offset, bytes, i); |
afddba49 NP |
3905 | flush_dcache_page(page); |
3906 | ||
3907 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3908 | page, fsdata); | |
f0b65f39 AV |
3909 | if (unlikely(status != copied)) { |
3910 | iov_iter_revert(i, copied - max(status, 0L)); | |
3911 | if (unlikely(status < 0)) | |
3912 | break; | |
3913 | } | |
afddba49 NP |
3914 | cond_resched(); |
3915 | ||
bc1bb416 | 3916 | if (unlikely(status == 0)) { |
afddba49 | 3917 | /* |
bc1bb416 AV |
3918 | * A short copy made ->write_end() reject the |
3919 | * thing entirely. Might be memory poisoning | |
3920 | * halfway through, might be a race with munmap, | |
3921 | * might be severe memory pressure. | |
afddba49 | 3922 | */ |
bc1bb416 AV |
3923 | if (copied) |
3924 | bytes = copied; | |
afddba49 NP |
3925 | goto again; |
3926 | } | |
f0b65f39 AV |
3927 | pos += status; |
3928 | written += status; | |
afddba49 NP |
3929 | |
3930 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3931 | } while (iov_iter_count(i)); |
3932 | ||
3933 | return written ? written : status; | |
3934 | } | |
3b93f911 | 3935 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3936 | |
e4dd9de3 | 3937 | /** |
8174202b | 3938 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3939 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3940 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3941 | * |
3942 | * This function does all the work needed for actually writing data to a | |
3943 | * file. It does all basic checks, removes SUID from the file, updates | |
3944 | * modification times and calls proper subroutines depending on whether we | |
3945 | * do direct IO or a standard buffered write. | |
3946 | * | |
9608703e | 3947 | * It expects i_rwsem to be grabbed unless we work on a block device or similar |
e4dd9de3 JK |
3948 | * object which does not need locking at all. |
3949 | * | |
3950 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3951 | * A caller has to handle it. This is mainly due to the fact that we want to | |
9608703e | 3952 | * avoid syncing under i_rwsem. |
a862f68a MR |
3953 | * |
3954 | * Return: | |
3955 | * * number of bytes written, even for truncated writes | |
3956 | * * negative error code if no data has been written at all | |
e4dd9de3 | 3957 | */ |
8174202b | 3958 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3959 | { |
3960 | struct file *file = iocb->ki_filp; | |
68d68ff6 | 3961 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3962 | struct inode *inode = mapping->host; |
3b93f911 | 3963 | ssize_t written = 0; |
1da177e4 | 3964 | ssize_t err; |
3b93f911 | 3965 | ssize_t status; |
1da177e4 | 3966 | |
5fa8e0a1 | 3967 | err = file_remove_privs(file); |
1da177e4 LT |
3968 | if (err) |
3969 | goto out; | |
3970 | ||
c3b2da31 JB |
3971 | err = file_update_time(file); |
3972 | if (err) | |
3973 | goto out; | |
1da177e4 | 3974 | |
2ba48ce5 | 3975 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3976 | loff_t pos, endbyte; |
fb5527e6 | 3977 | |
1af5bb49 | 3978 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3979 | /* |
fbbbad4b MW |
3980 | * If the write stopped short of completing, fall back to |
3981 | * buffered writes. Some filesystems do this for writes to | |
3982 | * holes, for example. For DAX files, a buffered write will | |
3983 | * not succeed (even if it did, DAX does not handle dirty | |
3984 | * page-cache pages correctly). | |
1da177e4 | 3985 | */ |
0b8def9d | 3986 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3987 | goto out; |
3988 | ||
800ba295 MWO |
3989 | pos = iocb->ki_pos; |
3990 | status = generic_perform_write(iocb, from); | |
fb5527e6 | 3991 | /* |
3b93f911 | 3992 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3993 | * then we want to return the number of bytes which were |
3994 | * direct-written, or the error code if that was zero. Note | |
3995 | * that this differs from normal direct-io semantics, which | |
3996 | * will return -EFOO even if some bytes were written. | |
3997 | */ | |
60bb4529 | 3998 | if (unlikely(status < 0)) { |
3b93f911 | 3999 | err = status; |
fb5527e6 JM |
4000 | goto out; |
4001 | } | |
fb5527e6 JM |
4002 | /* |
4003 | * We need to ensure that the page cache pages are written to | |
4004 | * disk and invalidated to preserve the expected O_DIRECT | |
4005 | * semantics. | |
4006 | */ | |
3b93f911 | 4007 | endbyte = pos + status - 1; |
0b8def9d | 4008 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 4009 | if (err == 0) { |
0b8def9d | 4010 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 4011 | written += status; |
fb5527e6 | 4012 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
4013 | pos >> PAGE_SHIFT, |
4014 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
4015 | } else { |
4016 | /* | |
4017 | * We don't know how much we wrote, so just return | |
4018 | * the number of bytes which were direct-written | |
4019 | */ | |
4020 | } | |
4021 | } else { | |
800ba295 | 4022 | written = generic_perform_write(iocb, from); |
0b8def9d AV |
4023 | if (likely(written > 0)) |
4024 | iocb->ki_pos += written; | |
fb5527e6 | 4025 | } |
1da177e4 | 4026 | out: |
1da177e4 LT |
4027 | return written ? written : err; |
4028 | } | |
8174202b | 4029 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 4030 | |
e4dd9de3 | 4031 | /** |
8174202b | 4032 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 4033 | * @iocb: IO state structure |
8174202b | 4034 | * @from: iov_iter with data to write |
e4dd9de3 | 4035 | * |
8174202b | 4036 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 | 4037 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
9608703e | 4038 | * and acquires i_rwsem as needed. |
a862f68a MR |
4039 | * Return: |
4040 | * * negative error code if no data has been written at all of | |
4041 | * vfs_fsync_range() failed for a synchronous write | |
4042 | * * number of bytes written, even for truncated writes | |
e4dd9de3 | 4043 | */ |
8174202b | 4044 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
4045 | { |
4046 | struct file *file = iocb->ki_filp; | |
148f948b | 4047 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 4048 | ssize_t ret; |
1da177e4 | 4049 | |
5955102c | 4050 | inode_lock(inode); |
3309dd04 AV |
4051 | ret = generic_write_checks(iocb, from); |
4052 | if (ret > 0) | |
5f380c7f | 4053 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 4054 | inode_unlock(inode); |
1da177e4 | 4055 | |
e2592217 CH |
4056 | if (ret > 0) |
4057 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
4058 | return ret; |
4059 | } | |
8174202b | 4060 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 4061 | |
cf9a2ae8 | 4062 | /** |
82c50f8b MWO |
4063 | * filemap_release_folio() - Release fs-specific metadata on a folio. |
4064 | * @folio: The folio which the kernel is trying to free. | |
4065 | * @gfp: Memory allocation flags (and I/O mode). | |
cf9a2ae8 | 4066 | * |
82c50f8b MWO |
4067 | * The address_space is trying to release any data attached to a folio |
4068 | * (presumably at folio->private). | |
cf9a2ae8 | 4069 | * |
82c50f8b MWO |
4070 | * This will also be called if the private_2 flag is set on a page, |
4071 | * indicating that the folio has other metadata associated with it. | |
266cf658 | 4072 | * |
82c50f8b MWO |
4073 | * The @gfp argument specifies whether I/O may be performed to release |
4074 | * this page (__GFP_IO), and whether the call may block | |
4075 | * (__GFP_RECLAIM & __GFP_FS). | |
cf9a2ae8 | 4076 | * |
82c50f8b | 4077 | * Return: %true if the release was successful, otherwise %false. |
cf9a2ae8 | 4078 | */ |
82c50f8b | 4079 | bool filemap_release_folio(struct folio *folio, gfp_t gfp) |
cf9a2ae8 | 4080 | { |
82c50f8b | 4081 | struct address_space * const mapping = folio->mapping; |
cf9a2ae8 | 4082 | |
82c50f8b MWO |
4083 | BUG_ON(!folio_test_locked(folio)); |
4084 | if (folio_test_writeback(folio)) | |
4085 | return false; | |
cf9a2ae8 | 4086 | |
fa29000b MWO |
4087 | if (mapping && mapping->a_ops->release_folio) |
4088 | return mapping->a_ops->release_folio(folio, gfp); | |
68189fef | 4089 | return try_to_free_buffers(folio); |
cf9a2ae8 | 4090 | } |
82c50f8b | 4091 | EXPORT_SYMBOL(filemap_release_folio); |
cf264e13 NP |
4092 | |
4093 | #ifdef CONFIG_CACHESTAT_SYSCALL | |
4094 | /** | |
4095 | * filemap_cachestat() - compute the page cache statistics of a mapping | |
4096 | * @mapping: The mapping to compute the statistics for. | |
4097 | * @first_index: The starting page cache index. | |
4098 | * @last_index: The final page index (inclusive). | |
4099 | * @cs: the cachestat struct to write the result to. | |
4100 | * | |
4101 | * This will query the page cache statistics of a mapping in the | |
4102 | * page range of [first_index, last_index] (inclusive). The statistics | |
4103 | * queried include: number of dirty pages, number of pages marked for | |
4104 | * writeback, and the number of (recently) evicted pages. | |
4105 | */ | |
4106 | static void filemap_cachestat(struct address_space *mapping, | |
4107 | pgoff_t first_index, pgoff_t last_index, struct cachestat *cs) | |
4108 | { | |
4109 | XA_STATE(xas, &mapping->i_pages, first_index); | |
4110 | struct folio *folio; | |
4111 | ||
4112 | rcu_read_lock(); | |
4113 | xas_for_each(&xas, folio, last_index) { | |
4114 | unsigned long nr_pages; | |
4115 | pgoff_t folio_first_index, folio_last_index; | |
4116 | ||
4117 | if (xas_retry(&xas, folio)) | |
4118 | continue; | |
4119 | ||
4120 | if (xa_is_value(folio)) { | |
4121 | /* page is evicted */ | |
4122 | void *shadow = (void *)folio; | |
4123 | bool workingset; /* not used */ | |
4124 | int order = xa_get_order(xas.xa, xas.xa_index); | |
4125 | ||
4126 | nr_pages = 1 << order; | |
4127 | folio_first_index = round_down(xas.xa_index, 1 << order); | |
4128 | folio_last_index = folio_first_index + nr_pages - 1; | |
4129 | ||
4130 | /* Folios might straddle the range boundaries, only count covered pages */ | |
4131 | if (folio_first_index < first_index) | |
4132 | nr_pages -= first_index - folio_first_index; | |
4133 | ||
4134 | if (folio_last_index > last_index) | |
4135 | nr_pages -= folio_last_index - last_index; | |
4136 | ||
4137 | cs->nr_evicted += nr_pages; | |
4138 | ||
4139 | #ifdef CONFIG_SWAP /* implies CONFIG_MMU */ | |
4140 | if (shmem_mapping(mapping)) { | |
4141 | /* shmem file - in swap cache */ | |
4142 | swp_entry_t swp = radix_to_swp_entry(folio); | |
4143 | ||
4144 | shadow = get_shadow_from_swap_cache(swp); | |
4145 | } | |
4146 | #endif | |
4147 | if (workingset_test_recent(shadow, true, &workingset)) | |
4148 | cs->nr_recently_evicted += nr_pages; | |
4149 | ||
4150 | goto resched; | |
4151 | } | |
4152 | ||
4153 | nr_pages = folio_nr_pages(folio); | |
4154 | folio_first_index = folio_pgoff(folio); | |
4155 | folio_last_index = folio_first_index + nr_pages - 1; | |
4156 | ||
4157 | /* Folios might straddle the range boundaries, only count covered pages */ | |
4158 | if (folio_first_index < first_index) | |
4159 | nr_pages -= first_index - folio_first_index; | |
4160 | ||
4161 | if (folio_last_index > last_index) | |
4162 | nr_pages -= folio_last_index - last_index; | |
4163 | ||
4164 | /* page is in cache */ | |
4165 | cs->nr_cache += nr_pages; | |
4166 | ||
4167 | if (folio_test_dirty(folio)) | |
4168 | cs->nr_dirty += nr_pages; | |
4169 | ||
4170 | if (folio_test_writeback(folio)) | |
4171 | cs->nr_writeback += nr_pages; | |
4172 | ||
4173 | resched: | |
4174 | if (need_resched()) { | |
4175 | xas_pause(&xas); | |
4176 | cond_resched_rcu(); | |
4177 | } | |
4178 | } | |
4179 | rcu_read_unlock(); | |
4180 | } | |
4181 | ||
4182 | /* | |
4183 | * The cachestat(2) system call. | |
4184 | * | |
4185 | * cachestat() returns the page cache statistics of a file in the | |
4186 | * bytes range specified by `off` and `len`: number of cached pages, | |
4187 | * number of dirty pages, number of pages marked for writeback, | |
4188 | * number of evicted pages, and number of recently evicted pages. | |
4189 | * | |
4190 | * An evicted page is a page that is previously in the page cache | |
4191 | * but has been evicted since. A page is recently evicted if its last | |
4192 | * eviction was recent enough that its reentry to the cache would | |
4193 | * indicate that it is actively being used by the system, and that | |
4194 | * there is memory pressure on the system. | |
4195 | * | |
4196 | * `off` and `len` must be non-negative integers. If `len` > 0, | |
4197 | * the queried range is [`off`, `off` + `len`]. If `len` == 0, | |
4198 | * we will query in the range from `off` to the end of the file. | |
4199 | * | |
4200 | * The `flags` argument is unused for now, but is included for future | |
4201 | * extensibility. User should pass 0 (i.e no flag specified). | |
4202 | * | |
4203 | * Currently, hugetlbfs is not supported. | |
4204 | * | |
4205 | * Because the status of a page can change after cachestat() checks it | |
4206 | * but before it returns to the application, the returned values may | |
4207 | * contain stale information. | |
4208 | * | |
4209 | * return values: | |
4210 | * zero - success | |
4211 | * -EFAULT - cstat or cstat_range points to an illegal address | |
4212 | * -EINVAL - invalid flags | |
4213 | * -EBADF - invalid file descriptor | |
4214 | * -EOPNOTSUPP - file descriptor is of a hugetlbfs file | |
4215 | */ | |
4216 | SYSCALL_DEFINE4(cachestat, unsigned int, fd, | |
4217 | struct cachestat_range __user *, cstat_range, | |
4218 | struct cachestat __user *, cstat, unsigned int, flags) | |
4219 | { | |
4220 | struct fd f = fdget(fd); | |
4221 | struct address_space *mapping; | |
4222 | struct cachestat_range csr; | |
4223 | struct cachestat cs; | |
4224 | pgoff_t first_index, last_index; | |
4225 | ||
4226 | if (!f.file) | |
4227 | return -EBADF; | |
4228 | ||
4229 | if (copy_from_user(&csr, cstat_range, | |
4230 | sizeof(struct cachestat_range))) { | |
4231 | fdput(f); | |
4232 | return -EFAULT; | |
4233 | } | |
4234 | ||
4235 | /* hugetlbfs is not supported */ | |
4236 | if (is_file_hugepages(f.file)) { | |
4237 | fdput(f); | |
4238 | return -EOPNOTSUPP; | |
4239 | } | |
4240 | ||
4241 | if (flags != 0) { | |
4242 | fdput(f); | |
4243 | return -EINVAL; | |
4244 | } | |
4245 | ||
4246 | first_index = csr.off >> PAGE_SHIFT; | |
4247 | last_index = | |
4248 | csr.len == 0 ? ULONG_MAX : (csr.off + csr.len - 1) >> PAGE_SHIFT; | |
4249 | memset(&cs, 0, sizeof(struct cachestat)); | |
4250 | mapping = f.file->f_mapping; | |
4251 | filemap_cachestat(mapping, first_index, last_index, &cs); | |
4252 | fdput(f); | |
4253 | ||
4254 | if (copy_to_user(cstat, &cs, sizeof(struct cachestat))) | |
4255 | return -EFAULT; | |
4256 | ||
4257 | return 0; | |
4258 | } | |
4259 | #endif /* CONFIG_CACHESTAT_SYSCALL */ |