]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/filemap.c | |
4 | * | |
5 | * Copyright (C) 1994-1999 Linus Torvalds | |
6 | */ | |
7 | ||
8 | /* | |
9 | * This file handles the generic file mmap semantics used by | |
10 | * most "normal" filesystems (but you don't /have/ to use this: | |
11 | * the NFS filesystem used to do this differently, for example) | |
12 | */ | |
b95f1b31 | 13 | #include <linux/export.h> |
1da177e4 | 14 | #include <linux/compiler.h> |
f9fe48be | 15 | #include <linux/dax.h> |
1da177e4 | 16 | #include <linux/fs.h> |
3f07c014 | 17 | #include <linux/sched/signal.h> |
c22ce143 | 18 | #include <linux/uaccess.h> |
c59ede7b | 19 | #include <linux/capability.h> |
1da177e4 | 20 | #include <linux/kernel_stat.h> |
5a0e3ad6 | 21 | #include <linux/gfp.h> |
1da177e4 LT |
22 | #include <linux/mm.h> |
23 | #include <linux/swap.h> | |
24 | #include <linux/mman.h> | |
25 | #include <linux/pagemap.h> | |
26 | #include <linux/file.h> | |
27 | #include <linux/uio.h> | |
cfcbfb13 | 28 | #include <linux/error-injection.h> |
1da177e4 LT |
29 | #include <linux/hash.h> |
30 | #include <linux/writeback.h> | |
53253383 | 31 | #include <linux/backing-dev.h> |
1da177e4 | 32 | #include <linux/pagevec.h> |
1da177e4 | 33 | #include <linux/security.h> |
44110fe3 | 34 | #include <linux/cpuset.h> |
00501b53 | 35 | #include <linux/hugetlb.h> |
8a9f3ccd | 36 | #include <linux/memcontrol.h> |
c515e1fd | 37 | #include <linux/cleancache.h> |
c7df8ad2 | 38 | #include <linux/shmem_fs.h> |
f1820361 | 39 | #include <linux/rmap.h> |
b1d29ba8 | 40 | #include <linux/delayacct.h> |
eb414681 | 41 | #include <linux/psi.h> |
d0e6a582 | 42 | #include <linux/ramfs.h> |
b9306a79 | 43 | #include <linux/page_idle.h> |
f9ce0be7 | 44 | #include <asm/pgalloc.h> |
de591a82 | 45 | #include <asm/tlbflush.h> |
0f8053a5 NP |
46 | #include "internal.h" |
47 | ||
fe0bfaaf RJ |
48 | #define CREATE_TRACE_POINTS |
49 | #include <trace/events/filemap.h> | |
50 | ||
1da177e4 | 51 | /* |
1da177e4 LT |
52 | * FIXME: remove all knowledge of the buffer layer from the core VM |
53 | */ | |
148f948b | 54 | #include <linux/buffer_head.h> /* for try_to_free_buffers */ |
1da177e4 | 55 | |
1da177e4 LT |
56 | #include <asm/mman.h> |
57 | ||
58 | /* | |
59 | * Shared mappings implemented 30.11.1994. It's not fully working yet, | |
60 | * though. | |
61 | * | |
62 | * Shared mappings now work. 15.8.1995 Bruno. | |
63 | * | |
64 | * finished 'unifying' the page and buffer cache and SMP-threaded the | |
65 | * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> | |
66 | * | |
67 | * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> | |
68 | */ | |
69 | ||
70 | /* | |
71 | * Lock ordering: | |
72 | * | |
c8c06efa | 73 | * ->i_mmap_rwsem (truncate_pagecache) |
1da177e4 | 74 | * ->private_lock (__free_pte->__set_page_dirty_buffers) |
5d337b91 | 75 | * ->swap_lock (exclusive_swap_page, others) |
b93b0163 | 76 | * ->i_pages lock |
1da177e4 | 77 | * |
9608703e | 78 | * ->i_rwsem |
730633f0 JK |
79 | * ->invalidate_lock (acquired by fs in truncate path) |
80 | * ->i_mmap_rwsem (truncate->unmap_mapping_range) | |
1da177e4 | 81 | * |
c1e8d7c6 | 82 | * ->mmap_lock |
c8c06efa | 83 | * ->i_mmap_rwsem |
b8072f09 | 84 | * ->page_table_lock or pte_lock (various, mainly in memory.c) |
b93b0163 | 85 | * ->i_pages lock (arch-dependent flush_dcache_mmap_lock) |
1da177e4 | 86 | * |
c1e8d7c6 | 87 | * ->mmap_lock |
730633f0 JK |
88 | * ->invalidate_lock (filemap_fault) |
89 | * ->lock_page (filemap_fault, access_process_vm) | |
1da177e4 | 90 | * |
9608703e | 91 | * ->i_rwsem (generic_perform_write) |
bb523b40 | 92 | * ->mmap_lock (fault_in_readable->do_page_fault) |
1da177e4 | 93 | * |
f758eeab | 94 | * bdi->wb.list_lock |
a66979ab | 95 | * sb_lock (fs/fs-writeback.c) |
b93b0163 | 96 | * ->i_pages lock (__sync_single_inode) |
1da177e4 | 97 | * |
c8c06efa | 98 | * ->i_mmap_rwsem |
1da177e4 LT |
99 | * ->anon_vma.lock (vma_adjust) |
100 | * | |
101 | * ->anon_vma.lock | |
b8072f09 | 102 | * ->page_table_lock or pte_lock (anon_vma_prepare and various) |
1da177e4 | 103 | * |
b8072f09 | 104 | * ->page_table_lock or pte_lock |
5d337b91 | 105 | * ->swap_lock (try_to_unmap_one) |
1da177e4 | 106 | * ->private_lock (try_to_unmap_one) |
b93b0163 | 107 | * ->i_pages lock (try_to_unmap_one) |
15b44736 HD |
108 | * ->lruvec->lru_lock (follow_page->mark_page_accessed) |
109 | * ->lruvec->lru_lock (check_pte_range->isolate_lru_page) | |
1da177e4 | 110 | * ->private_lock (page_remove_rmap->set_page_dirty) |
b93b0163 | 111 | * ->i_pages lock (page_remove_rmap->set_page_dirty) |
f758eeab | 112 | * bdi.wb->list_lock (page_remove_rmap->set_page_dirty) |
250df6ed | 113 | * ->inode->i_lock (page_remove_rmap->set_page_dirty) |
81f8c3a4 | 114 | * ->memcg->move_lock (page_remove_rmap->lock_page_memcg) |
f758eeab | 115 | * bdi.wb->list_lock (zap_pte_range->set_page_dirty) |
250df6ed | 116 | * ->inode->i_lock (zap_pte_range->set_page_dirty) |
1da177e4 LT |
117 | * ->private_lock (zap_pte_range->__set_page_dirty_buffers) |
118 | * | |
c8c06efa | 119 | * ->i_mmap_rwsem |
9a3c531d | 120 | * ->tasklist_lock (memory_failure, collect_procs_ao) |
1da177e4 LT |
121 | */ |
122 | ||
5c024e6a | 123 | static void page_cache_delete(struct address_space *mapping, |
a548b615 | 124 | struct folio *folio, void *shadow) |
91b0abe3 | 125 | { |
a548b615 MWO |
126 | XA_STATE(xas, &mapping->i_pages, folio->index); |
127 | long nr = 1; | |
c70b647d | 128 | |
5c024e6a | 129 | mapping_set_update(&xas, mapping); |
c70b647d | 130 | |
5c024e6a | 131 | /* hugetlb pages are represented by a single entry in the xarray */ |
a548b615 MWO |
132 | if (!folio_test_hugetlb(folio)) { |
133 | xas_set_order(&xas, folio->index, folio_order(folio)); | |
134 | nr = folio_nr_pages(folio); | |
5c024e6a | 135 | } |
91b0abe3 | 136 | |
a548b615 MWO |
137 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
138 | VM_BUG_ON_FOLIO(nr != 1 && shadow, folio); | |
449dd698 | 139 | |
5c024e6a MW |
140 | xas_store(&xas, shadow); |
141 | xas_init_marks(&xas); | |
d3798ae8 | 142 | |
a548b615 | 143 | folio->mapping = NULL; |
2300638b | 144 | /* Leave page->index set: truncation lookup relies upon it */ |
d3798ae8 | 145 | mapping->nrpages -= nr; |
91b0abe3 JW |
146 | } |
147 | ||
621db488 MWO |
148 | static void filemap_unaccount_folio(struct address_space *mapping, |
149 | struct folio *folio) | |
1da177e4 | 150 | { |
621db488 | 151 | long nr; |
1da177e4 | 152 | |
c515e1fd DM |
153 | /* |
154 | * if we're uptodate, flush out into the cleancache, otherwise | |
155 | * invalidate any existing cleancache entries. We can't leave | |
156 | * stale data around in the cleancache once our page is gone | |
157 | */ | |
621db488 MWO |
158 | if (folio_test_uptodate(folio) && folio_test_mappedtodisk(folio)) |
159 | cleancache_put_page(&folio->page); | |
c515e1fd | 160 | else |
621db488 | 161 | cleancache_invalidate_page(mapping, &folio->page); |
c515e1fd | 162 | |
621db488 MWO |
163 | VM_BUG_ON_FOLIO(folio_mapped(folio), folio); |
164 | if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) { | |
06b241f3 HD |
165 | int mapcount; |
166 | ||
167 | pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n", | |
621db488 MWO |
168 | current->comm, folio_pfn(folio)); |
169 | dump_page(&folio->page, "still mapped when deleted"); | |
06b241f3 HD |
170 | dump_stack(); |
171 | add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE); | |
172 | ||
621db488 | 173 | mapcount = page_mapcount(&folio->page); |
06b241f3 | 174 | if (mapping_exiting(mapping) && |
621db488 | 175 | folio_ref_count(folio) >= mapcount + 2) { |
06b241f3 HD |
176 | /* |
177 | * All vmas have already been torn down, so it's | |
621db488 | 178 | * a good bet that actually the folio is unmapped, |
06b241f3 HD |
179 | * and we'd prefer not to leak it: if we're wrong, |
180 | * some other bad page check should catch it later. | |
181 | */ | |
621db488 MWO |
182 | page_mapcount_reset(&folio->page); |
183 | folio_ref_sub(folio, mapcount); | |
06b241f3 HD |
184 | } |
185 | } | |
186 | ||
621db488 MWO |
187 | /* hugetlb folios do not participate in page cache accounting. */ |
188 | if (folio_test_hugetlb(folio)) | |
5ecc4d85 | 189 | return; |
09612fa6 | 190 | |
621db488 | 191 | nr = folio_nr_pages(folio); |
5ecc4d85 | 192 | |
621db488 MWO |
193 | __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr); |
194 | if (folio_test_swapbacked(folio)) { | |
195 | __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr); | |
196 | if (folio_test_pmd_mappable(folio)) | |
197 | __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr); | |
198 | } else if (folio_test_pmd_mappable(folio)) { | |
199 | __lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr); | |
09d91cda | 200 | filemap_nr_thps_dec(mapping); |
800d8c63 | 201 | } |
5ecc4d85 JK |
202 | |
203 | /* | |
621db488 MWO |
204 | * At this point folio must be either written or cleaned by |
205 | * truncate. Dirty folio here signals a bug and loss of | |
5ecc4d85 JK |
206 | * unwritten data. |
207 | * | |
621db488 MWO |
208 | * This fixes dirty accounting after removing the folio entirely |
209 | * but leaves the dirty flag set: it has no effect for truncated | |
210 | * folio and anyway will be cleared before returning folio to | |
5ecc4d85 JK |
211 | * buddy allocator. |
212 | */ | |
621db488 MWO |
213 | if (WARN_ON_ONCE(folio_test_dirty(folio))) |
214 | folio_account_cleaned(folio, mapping, | |
215 | inode_to_wb(mapping->host)); | |
5ecc4d85 JK |
216 | } |
217 | ||
218 | /* | |
219 | * Delete a page from the page cache and free it. Caller has to make | |
220 | * sure the page is locked and that nobody else uses it - or that usage | |
b93b0163 | 221 | * is safe. The caller must hold the i_pages lock. |
5ecc4d85 | 222 | */ |
452e9e69 | 223 | void __filemap_remove_folio(struct folio *folio, void *shadow) |
5ecc4d85 | 224 | { |
452e9e69 | 225 | struct address_space *mapping = folio->mapping; |
5ecc4d85 | 226 | |
a0580c6f | 227 | trace_mm_filemap_delete_from_page_cache(folio); |
621db488 | 228 | filemap_unaccount_folio(mapping, folio); |
a548b615 | 229 | page_cache_delete(mapping, folio, shadow); |
1da177e4 LT |
230 | } |
231 | ||
78f42660 | 232 | void filemap_free_folio(struct address_space *mapping, struct folio *folio) |
59c66c5f JK |
233 | { |
234 | void (*freepage)(struct page *); | |
235 | ||
236 | freepage = mapping->a_ops->freepage; | |
237 | if (freepage) | |
452e9e69 | 238 | freepage(&folio->page); |
59c66c5f | 239 | |
452e9e69 MWO |
240 | if (folio_test_large(folio) && !folio_test_hugetlb(folio)) { |
241 | folio_ref_sub(folio, folio_nr_pages(folio)); | |
242 | VM_BUG_ON_FOLIO(folio_ref_count(folio) <= 0, folio); | |
59c66c5f | 243 | } else { |
452e9e69 | 244 | folio_put(folio); |
59c66c5f JK |
245 | } |
246 | } | |
247 | ||
702cfbf9 | 248 | /** |
452e9e69 MWO |
249 | * filemap_remove_folio - Remove folio from page cache. |
250 | * @folio: The folio. | |
702cfbf9 | 251 | * |
452e9e69 MWO |
252 | * This must be called only on folios that are locked and have been |
253 | * verified to be in the page cache. It will never put the folio into | |
254 | * the free list because the caller has a reference on the page. | |
702cfbf9 | 255 | */ |
452e9e69 | 256 | void filemap_remove_folio(struct folio *folio) |
1da177e4 | 257 | { |
452e9e69 | 258 | struct address_space *mapping = folio->mapping; |
1da177e4 | 259 | |
452e9e69 | 260 | BUG_ON(!folio_test_locked(folio)); |
51b8c1fe | 261 | spin_lock(&mapping->host->i_lock); |
30472509 | 262 | xa_lock_irq(&mapping->i_pages); |
452e9e69 | 263 | __filemap_remove_folio(folio, NULL); |
30472509 | 264 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
265 | if (mapping_shrinkable(mapping)) |
266 | inode_add_lru(mapping->host); | |
267 | spin_unlock(&mapping->host->i_lock); | |
6072d13c | 268 | |
452e9e69 | 269 | filemap_free_folio(mapping, folio); |
97cecb5a | 270 | } |
97cecb5a | 271 | |
aa65c29c | 272 | /* |
ef8e5717 | 273 | * page_cache_delete_batch - delete several pages from page cache |
aa65c29c JK |
274 | * @mapping: the mapping to which pages belong |
275 | * @pvec: pagevec with pages to delete | |
276 | * | |
b93b0163 | 277 | * The function walks over mapping->i_pages and removes pages passed in @pvec |
4101196b MWO |
278 | * from the mapping. The function expects @pvec to be sorted by page index |
279 | * and is optimised for it to be dense. | |
b93b0163 | 280 | * It tolerates holes in @pvec (mapping entries at those indices are not |
aa65c29c | 281 | * modified). The function expects only THP head pages to be present in the |
4101196b | 282 | * @pvec. |
aa65c29c | 283 | * |
b93b0163 | 284 | * The function expects the i_pages lock to be held. |
aa65c29c | 285 | */ |
ef8e5717 | 286 | static void page_cache_delete_batch(struct address_space *mapping, |
aa65c29c JK |
287 | struct pagevec *pvec) |
288 | { | |
ef8e5717 | 289 | XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index); |
aa65c29c | 290 | int total_pages = 0; |
4101196b | 291 | int i = 0; |
1afd7ae5 | 292 | struct folio *folio; |
aa65c29c | 293 | |
ef8e5717 | 294 | mapping_set_update(&xas, mapping); |
1afd7ae5 | 295 | xas_for_each(&xas, folio, ULONG_MAX) { |
4101196b | 296 | if (i >= pagevec_count(pvec)) |
aa65c29c | 297 | break; |
4101196b MWO |
298 | |
299 | /* A swap/dax/shadow entry got inserted? Skip it. */ | |
1afd7ae5 | 300 | if (xa_is_value(folio)) |
aa65c29c | 301 | continue; |
4101196b MWO |
302 | /* |
303 | * A page got inserted in our range? Skip it. We have our | |
304 | * pages locked so they are protected from being removed. | |
305 | * If we see a page whose index is higher than ours, it | |
306 | * means our page has been removed, which shouldn't be | |
307 | * possible because we're holding the PageLock. | |
308 | */ | |
1afd7ae5 MWO |
309 | if (&folio->page != pvec->pages[i]) { |
310 | VM_BUG_ON_FOLIO(folio->index > | |
311 | pvec->pages[i]->index, folio); | |
4101196b MWO |
312 | continue; |
313 | } | |
314 | ||
1afd7ae5 | 315 | WARN_ON_ONCE(!folio_test_locked(folio)); |
4101196b | 316 | |
1afd7ae5 MWO |
317 | if (folio->index == xas.xa_index) |
318 | folio->mapping = NULL; | |
4101196b MWO |
319 | /* Leave page->index set: truncation lookup relies on it */ |
320 | ||
321 | /* | |
322 | * Move to the next page in the vector if this is a regular | |
323 | * page or the index is of the last sub-page of this compound | |
324 | * page. | |
325 | */ | |
1afd7ae5 | 326 | if (folio->index + folio_nr_pages(folio) - 1 == xas.xa_index) |
aa65c29c | 327 | i++; |
ef8e5717 | 328 | xas_store(&xas, NULL); |
aa65c29c JK |
329 | total_pages++; |
330 | } | |
331 | mapping->nrpages -= total_pages; | |
332 | } | |
333 | ||
334 | void delete_from_page_cache_batch(struct address_space *mapping, | |
335 | struct pagevec *pvec) | |
336 | { | |
337 | int i; | |
aa65c29c JK |
338 | |
339 | if (!pagevec_count(pvec)) | |
340 | return; | |
341 | ||
51b8c1fe | 342 | spin_lock(&mapping->host->i_lock); |
30472509 | 343 | xa_lock_irq(&mapping->i_pages); |
aa65c29c | 344 | for (i = 0; i < pagevec_count(pvec); i++) { |
a0580c6f | 345 | struct folio *folio = page_folio(pvec->pages[i]); |
aa65c29c | 346 | |
a0580c6f MWO |
347 | trace_mm_filemap_delete_from_page_cache(folio); |
348 | filemap_unaccount_folio(mapping, folio); | |
aa65c29c | 349 | } |
ef8e5717 | 350 | page_cache_delete_batch(mapping, pvec); |
30472509 | 351 | xa_unlock_irq(&mapping->i_pages); |
51b8c1fe JW |
352 | if (mapping_shrinkable(mapping)) |
353 | inode_add_lru(mapping->host); | |
354 | spin_unlock(&mapping->host->i_lock); | |
aa65c29c JK |
355 | |
356 | for (i = 0; i < pagevec_count(pvec); i++) | |
452e9e69 | 357 | filemap_free_folio(mapping, page_folio(pvec->pages[i])); |
aa65c29c JK |
358 | } |
359 | ||
d72d9e2a | 360 | int filemap_check_errors(struct address_space *mapping) |
865ffef3 DM |
361 | { |
362 | int ret = 0; | |
363 | /* Check for outstanding write errors */ | |
7fcbbaf1 JA |
364 | if (test_bit(AS_ENOSPC, &mapping->flags) && |
365 | test_and_clear_bit(AS_ENOSPC, &mapping->flags)) | |
865ffef3 | 366 | ret = -ENOSPC; |
7fcbbaf1 JA |
367 | if (test_bit(AS_EIO, &mapping->flags) && |
368 | test_and_clear_bit(AS_EIO, &mapping->flags)) | |
865ffef3 DM |
369 | ret = -EIO; |
370 | return ret; | |
371 | } | |
d72d9e2a | 372 | EXPORT_SYMBOL(filemap_check_errors); |
865ffef3 | 373 | |
76341cab JL |
374 | static int filemap_check_and_keep_errors(struct address_space *mapping) |
375 | { | |
376 | /* Check for outstanding write errors */ | |
377 | if (test_bit(AS_EIO, &mapping->flags)) | |
378 | return -EIO; | |
379 | if (test_bit(AS_ENOSPC, &mapping->flags)) | |
380 | return -ENOSPC; | |
381 | return 0; | |
382 | } | |
383 | ||
5a798493 JB |
384 | /** |
385 | * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range | |
386 | * @mapping: address space structure to write | |
387 | * @wbc: the writeback_control controlling the writeout | |
388 | * | |
389 | * Call writepages on the mapping using the provided wbc to control the | |
390 | * writeout. | |
391 | * | |
392 | * Return: %0 on success, negative error code otherwise. | |
393 | */ | |
394 | int filemap_fdatawrite_wbc(struct address_space *mapping, | |
395 | struct writeback_control *wbc) | |
396 | { | |
397 | int ret; | |
398 | ||
399 | if (!mapping_can_writeback(mapping) || | |
400 | !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
401 | return 0; | |
402 | ||
403 | wbc_attach_fdatawrite_inode(wbc, mapping->host); | |
404 | ret = do_writepages(mapping, wbc); | |
405 | wbc_detach_inode(wbc); | |
406 | return ret; | |
407 | } | |
408 | EXPORT_SYMBOL(filemap_fdatawrite_wbc); | |
409 | ||
1da177e4 | 410 | /** |
485bb99b | 411 | * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range |
67be2dd1 MW |
412 | * @mapping: address space structure to write |
413 | * @start: offset in bytes where the range starts | |
469eb4d0 | 414 | * @end: offset in bytes where the range ends (inclusive) |
67be2dd1 | 415 | * @sync_mode: enable synchronous operation |
1da177e4 | 416 | * |
485bb99b RD |
417 | * Start writeback against all of a mapping's dirty pages that lie |
418 | * within the byte offsets <start, end> inclusive. | |
419 | * | |
1da177e4 | 420 | * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as |
485bb99b | 421 | * opposed to a regular memory cleansing writeback. The difference between |
1da177e4 LT |
422 | * these two operations is that if a dirty page/buffer is encountered, it must |
423 | * be waited upon, and not just skipped over. | |
a862f68a MR |
424 | * |
425 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 | 426 | */ |
ebcf28e1 AM |
427 | int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
428 | loff_t end, int sync_mode) | |
1da177e4 | 429 | { |
1da177e4 LT |
430 | struct writeback_control wbc = { |
431 | .sync_mode = sync_mode, | |
05fe478d | 432 | .nr_to_write = LONG_MAX, |
111ebb6e OH |
433 | .range_start = start, |
434 | .range_end = end, | |
1da177e4 LT |
435 | }; |
436 | ||
5a798493 | 437 | return filemap_fdatawrite_wbc(mapping, &wbc); |
1da177e4 LT |
438 | } |
439 | ||
440 | static inline int __filemap_fdatawrite(struct address_space *mapping, | |
441 | int sync_mode) | |
442 | { | |
111ebb6e | 443 | return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); |
1da177e4 LT |
444 | } |
445 | ||
446 | int filemap_fdatawrite(struct address_space *mapping) | |
447 | { | |
448 | return __filemap_fdatawrite(mapping, WB_SYNC_ALL); | |
449 | } | |
450 | EXPORT_SYMBOL(filemap_fdatawrite); | |
451 | ||
f4c0a0fd | 452 | int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, |
ebcf28e1 | 453 | loff_t end) |
1da177e4 LT |
454 | { |
455 | return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); | |
456 | } | |
f4c0a0fd | 457 | EXPORT_SYMBOL(filemap_fdatawrite_range); |
1da177e4 | 458 | |
485bb99b RD |
459 | /** |
460 | * filemap_flush - mostly a non-blocking flush | |
461 | * @mapping: target address_space | |
462 | * | |
1da177e4 LT |
463 | * This is a mostly non-blocking flush. Not suitable for data-integrity |
464 | * purposes - I/O may not be started against all dirty pages. | |
a862f68a MR |
465 | * |
466 | * Return: %0 on success, negative error code otherwise. | |
1da177e4 LT |
467 | */ |
468 | int filemap_flush(struct address_space *mapping) | |
469 | { | |
470 | return __filemap_fdatawrite(mapping, WB_SYNC_NONE); | |
471 | } | |
472 | EXPORT_SYMBOL(filemap_flush); | |
473 | ||
7fc9e472 GR |
474 | /** |
475 | * filemap_range_has_page - check if a page exists in range. | |
476 | * @mapping: address space within which to check | |
477 | * @start_byte: offset in bytes where the range starts | |
478 | * @end_byte: offset in bytes where the range ends (inclusive) | |
479 | * | |
480 | * Find at least one page in the range supplied, usually used to check if | |
481 | * direct writing in this range will trigger a writeback. | |
a862f68a MR |
482 | * |
483 | * Return: %true if at least one page exists in the specified range, | |
484 | * %false otherwise. | |
7fc9e472 GR |
485 | */ |
486 | bool filemap_range_has_page(struct address_space *mapping, | |
487 | loff_t start_byte, loff_t end_byte) | |
488 | { | |
f7b68046 | 489 | struct page *page; |
8fa8e538 MW |
490 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); |
491 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
7fc9e472 GR |
492 | |
493 | if (end_byte < start_byte) | |
494 | return false; | |
495 | ||
8fa8e538 MW |
496 | rcu_read_lock(); |
497 | for (;;) { | |
498 | page = xas_find(&xas, max); | |
499 | if (xas_retry(&xas, page)) | |
500 | continue; | |
501 | /* Shadow entries don't count */ | |
502 | if (xa_is_value(page)) | |
503 | continue; | |
504 | /* | |
505 | * We don't need to try to pin this page; we're about to | |
506 | * release the RCU lock anyway. It is enough to know that | |
507 | * there was a page here recently. | |
508 | */ | |
509 | break; | |
510 | } | |
511 | rcu_read_unlock(); | |
7fc9e472 | 512 | |
8fa8e538 | 513 | return page != NULL; |
7fc9e472 GR |
514 | } |
515 | EXPORT_SYMBOL(filemap_range_has_page); | |
516 | ||
5e8fcc1a | 517 | static void __filemap_fdatawait_range(struct address_space *mapping, |
aa750fd7 | 518 | loff_t start_byte, loff_t end_byte) |
1da177e4 | 519 | { |
09cbfeaf KS |
520 | pgoff_t index = start_byte >> PAGE_SHIFT; |
521 | pgoff_t end = end_byte >> PAGE_SHIFT; | |
1da177e4 LT |
522 | struct pagevec pvec; |
523 | int nr_pages; | |
1da177e4 | 524 | |
94004ed7 | 525 | if (end_byte < start_byte) |
5e8fcc1a | 526 | return; |
1da177e4 | 527 | |
86679820 | 528 | pagevec_init(&pvec); |
312e9d2f | 529 | while (index <= end) { |
1da177e4 LT |
530 | unsigned i; |
531 | ||
312e9d2f | 532 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, |
67fd707f | 533 | end, PAGECACHE_TAG_WRITEBACK); |
312e9d2f JK |
534 | if (!nr_pages) |
535 | break; | |
536 | ||
1da177e4 LT |
537 | for (i = 0; i < nr_pages; i++) { |
538 | struct page *page = pvec.pages[i]; | |
539 | ||
1da177e4 | 540 | wait_on_page_writeback(page); |
5e8fcc1a | 541 | ClearPageError(page); |
1da177e4 LT |
542 | } |
543 | pagevec_release(&pvec); | |
544 | cond_resched(); | |
545 | } | |
aa750fd7 JN |
546 | } |
547 | ||
548 | /** | |
549 | * filemap_fdatawait_range - wait for writeback to complete | |
550 | * @mapping: address space structure to wait for | |
551 | * @start_byte: offset in bytes where the range starts | |
552 | * @end_byte: offset in bytes where the range ends (inclusive) | |
553 | * | |
554 | * Walk the list of under-writeback pages of the given address space | |
555 | * in the given range and wait for all of them. Check error status of | |
556 | * the address space and return it. | |
557 | * | |
558 | * Since the error status of the address space is cleared by this function, | |
559 | * callers are responsible for checking the return value and handling and/or | |
560 | * reporting the error. | |
a862f68a MR |
561 | * |
562 | * Return: error status of the address space. | |
aa750fd7 JN |
563 | */ |
564 | int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, | |
565 | loff_t end_byte) | |
566 | { | |
5e8fcc1a JL |
567 | __filemap_fdatawait_range(mapping, start_byte, end_byte); |
568 | return filemap_check_errors(mapping); | |
1da177e4 | 569 | } |
d3bccb6f JK |
570 | EXPORT_SYMBOL(filemap_fdatawait_range); |
571 | ||
aa0bfcd9 RZ |
572 | /** |
573 | * filemap_fdatawait_range_keep_errors - wait for writeback to complete | |
574 | * @mapping: address space structure to wait for | |
575 | * @start_byte: offset in bytes where the range starts | |
576 | * @end_byte: offset in bytes where the range ends (inclusive) | |
577 | * | |
578 | * Walk the list of under-writeback pages of the given address space in the | |
579 | * given range and wait for all of them. Unlike filemap_fdatawait_range(), | |
580 | * this function does not clear error status of the address space. | |
581 | * | |
582 | * Use this function if callers don't handle errors themselves. Expected | |
583 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
584 | * fsfreeze(8) | |
585 | */ | |
586 | int filemap_fdatawait_range_keep_errors(struct address_space *mapping, | |
587 | loff_t start_byte, loff_t end_byte) | |
588 | { | |
589 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
590 | return filemap_check_and_keep_errors(mapping); | |
591 | } | |
592 | EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors); | |
593 | ||
a823e458 JL |
594 | /** |
595 | * file_fdatawait_range - wait for writeback to complete | |
596 | * @file: file pointing to address space structure to wait for | |
597 | * @start_byte: offset in bytes where the range starts | |
598 | * @end_byte: offset in bytes where the range ends (inclusive) | |
599 | * | |
600 | * Walk the list of under-writeback pages of the address space that file | |
601 | * refers to, in the given range and wait for all of them. Check error | |
602 | * status of the address space vs. the file->f_wb_err cursor and return it. | |
603 | * | |
604 | * Since the error status of the file is advanced by this function, | |
605 | * callers are responsible for checking the return value and handling and/or | |
606 | * reporting the error. | |
a862f68a MR |
607 | * |
608 | * Return: error status of the address space vs. the file->f_wb_err cursor. | |
a823e458 JL |
609 | */ |
610 | int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte) | |
611 | { | |
612 | struct address_space *mapping = file->f_mapping; | |
613 | ||
614 | __filemap_fdatawait_range(mapping, start_byte, end_byte); | |
615 | return file_check_and_advance_wb_err(file); | |
616 | } | |
617 | EXPORT_SYMBOL(file_fdatawait_range); | |
d3bccb6f | 618 | |
aa750fd7 JN |
619 | /** |
620 | * filemap_fdatawait_keep_errors - wait for writeback without clearing errors | |
621 | * @mapping: address space structure to wait for | |
622 | * | |
623 | * Walk the list of under-writeback pages of the given address space | |
624 | * and wait for all of them. Unlike filemap_fdatawait(), this function | |
625 | * does not clear error status of the address space. | |
626 | * | |
627 | * Use this function if callers don't handle errors themselves. Expected | |
628 | * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2), | |
629 | * fsfreeze(8) | |
a862f68a MR |
630 | * |
631 | * Return: error status of the address space. | |
aa750fd7 | 632 | */ |
76341cab | 633 | int filemap_fdatawait_keep_errors(struct address_space *mapping) |
aa750fd7 | 634 | { |
ffb959bb | 635 | __filemap_fdatawait_range(mapping, 0, LLONG_MAX); |
76341cab | 636 | return filemap_check_and_keep_errors(mapping); |
aa750fd7 | 637 | } |
76341cab | 638 | EXPORT_SYMBOL(filemap_fdatawait_keep_errors); |
aa750fd7 | 639 | |
875d91b1 | 640 | /* Returns true if writeback might be needed or already in progress. */ |
9326c9b2 | 641 | static bool mapping_needs_writeback(struct address_space *mapping) |
1da177e4 | 642 | { |
875d91b1 | 643 | return mapping->nrpages; |
1da177e4 | 644 | } |
1da177e4 | 645 | |
f8ee8909 JA |
646 | static bool filemap_range_has_writeback(struct address_space *mapping, |
647 | loff_t start_byte, loff_t end_byte) | |
648 | { | |
649 | XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT); | |
650 | pgoff_t max = end_byte >> PAGE_SHIFT; | |
651 | struct page *page; | |
652 | ||
653 | if (end_byte < start_byte) | |
654 | return false; | |
655 | ||
656 | rcu_read_lock(); | |
657 | xas_for_each(&xas, page, max) { | |
658 | if (xas_retry(&xas, page)) | |
659 | continue; | |
660 | if (xa_is_value(page)) | |
661 | continue; | |
662 | if (PageDirty(page) || PageLocked(page) || PageWriteback(page)) | |
663 | break; | |
664 | } | |
665 | rcu_read_unlock(); | |
666 | return page != NULL; | |
667 | ||
668 | } | |
669 | ||
63135aa3 JA |
670 | /** |
671 | * filemap_range_needs_writeback - check if range potentially needs writeback | |
672 | * @mapping: address space within which to check | |
673 | * @start_byte: offset in bytes where the range starts | |
674 | * @end_byte: offset in bytes where the range ends (inclusive) | |
675 | * | |
676 | * Find at least one page in the range supplied, usually used to check if | |
677 | * direct writing in this range will trigger a writeback. Used by O_DIRECT | |
678 | * read/write with IOCB_NOWAIT, to see if the caller needs to do | |
679 | * filemap_write_and_wait_range() before proceeding. | |
680 | * | |
681 | * Return: %true if the caller should do filemap_write_and_wait_range() before | |
682 | * doing O_DIRECT to a page in this range, %false otherwise. | |
683 | */ | |
684 | bool filemap_range_needs_writeback(struct address_space *mapping, | |
685 | loff_t start_byte, loff_t end_byte) | |
686 | { | |
63135aa3 JA |
687 | if (!mapping_needs_writeback(mapping)) |
688 | return false; | |
689 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && | |
690 | !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK)) | |
691 | return false; | |
f8ee8909 | 692 | return filemap_range_has_writeback(mapping, start_byte, end_byte); |
63135aa3 JA |
693 | } |
694 | EXPORT_SYMBOL_GPL(filemap_range_needs_writeback); | |
695 | ||
485bb99b RD |
696 | /** |
697 | * filemap_write_and_wait_range - write out & wait on a file range | |
698 | * @mapping: the address_space for the pages | |
699 | * @lstart: offset in bytes where the range starts | |
700 | * @lend: offset in bytes where the range ends (inclusive) | |
701 | * | |
469eb4d0 AM |
702 | * Write out and wait upon file offsets lstart->lend, inclusive. |
703 | * | |
0e056eb5 | 704 | * Note that @lend is inclusive (describes the last byte to be written) so |
469eb4d0 | 705 | * that this function can be used to write to the very end-of-file (end = -1). |
a862f68a MR |
706 | * |
707 | * Return: error status of the address space. | |
469eb4d0 | 708 | */ |
1da177e4 LT |
709 | int filemap_write_and_wait_range(struct address_space *mapping, |
710 | loff_t lstart, loff_t lend) | |
711 | { | |
28fd1298 | 712 | int err = 0; |
1da177e4 | 713 | |
9326c9b2 | 714 | if (mapping_needs_writeback(mapping)) { |
28fd1298 OH |
715 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
716 | WB_SYNC_ALL); | |
ddf8f376 IW |
717 | /* |
718 | * Even if the above returned error, the pages may be | |
719 | * written partially (e.g. -ENOSPC), so we wait for it. | |
720 | * But the -EIO is special case, it may indicate the worst | |
721 | * thing (e.g. bug) happened, so we avoid waiting for it. | |
722 | */ | |
28fd1298 | 723 | if (err != -EIO) { |
94004ed7 CH |
724 | int err2 = filemap_fdatawait_range(mapping, |
725 | lstart, lend); | |
28fd1298 OH |
726 | if (!err) |
727 | err = err2; | |
cbeaf951 JL |
728 | } else { |
729 | /* Clear any previously stored errors */ | |
730 | filemap_check_errors(mapping); | |
28fd1298 | 731 | } |
865ffef3 DM |
732 | } else { |
733 | err = filemap_check_errors(mapping); | |
1da177e4 | 734 | } |
28fd1298 | 735 | return err; |
1da177e4 | 736 | } |
f6995585 | 737 | EXPORT_SYMBOL(filemap_write_and_wait_range); |
1da177e4 | 738 | |
5660e13d JL |
739 | void __filemap_set_wb_err(struct address_space *mapping, int err) |
740 | { | |
3acdfd28 | 741 | errseq_t eseq = errseq_set(&mapping->wb_err, err); |
5660e13d JL |
742 | |
743 | trace_filemap_set_wb_err(mapping, eseq); | |
744 | } | |
745 | EXPORT_SYMBOL(__filemap_set_wb_err); | |
746 | ||
747 | /** | |
748 | * file_check_and_advance_wb_err - report wb error (if any) that was previously | |
749 | * and advance wb_err to current one | |
750 | * @file: struct file on which the error is being reported | |
751 | * | |
752 | * When userland calls fsync (or something like nfsd does the equivalent), we | |
753 | * want to report any writeback errors that occurred since the last fsync (or | |
754 | * since the file was opened if there haven't been any). | |
755 | * | |
756 | * Grab the wb_err from the mapping. If it matches what we have in the file, | |
757 | * then just quickly return 0. The file is all caught up. | |
758 | * | |
759 | * If it doesn't match, then take the mapping value, set the "seen" flag in | |
760 | * it and try to swap it into place. If it works, or another task beat us | |
761 | * to it with the new value, then update the f_wb_err and return the error | |
762 | * portion. The error at this point must be reported via proper channels | |
763 | * (a'la fsync, or NFS COMMIT operation, etc.). | |
764 | * | |
765 | * While we handle mapping->wb_err with atomic operations, the f_wb_err | |
766 | * value is protected by the f_lock since we must ensure that it reflects | |
767 | * the latest value swapped in for this file descriptor. | |
a862f68a MR |
768 | * |
769 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
770 | */ |
771 | int file_check_and_advance_wb_err(struct file *file) | |
772 | { | |
773 | int err = 0; | |
774 | errseq_t old = READ_ONCE(file->f_wb_err); | |
775 | struct address_space *mapping = file->f_mapping; | |
776 | ||
777 | /* Locklessly handle the common case where nothing has changed */ | |
778 | if (errseq_check(&mapping->wb_err, old)) { | |
779 | /* Something changed, must use slow path */ | |
780 | spin_lock(&file->f_lock); | |
781 | old = file->f_wb_err; | |
782 | err = errseq_check_and_advance(&mapping->wb_err, | |
783 | &file->f_wb_err); | |
784 | trace_file_check_and_advance_wb_err(file, old); | |
785 | spin_unlock(&file->f_lock); | |
786 | } | |
f4e222c5 JL |
787 | |
788 | /* | |
789 | * We're mostly using this function as a drop in replacement for | |
790 | * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect | |
791 | * that the legacy code would have had on these flags. | |
792 | */ | |
793 | clear_bit(AS_EIO, &mapping->flags); | |
794 | clear_bit(AS_ENOSPC, &mapping->flags); | |
5660e13d JL |
795 | return err; |
796 | } | |
797 | EXPORT_SYMBOL(file_check_and_advance_wb_err); | |
798 | ||
799 | /** | |
800 | * file_write_and_wait_range - write out & wait on a file range | |
801 | * @file: file pointing to address_space with pages | |
802 | * @lstart: offset in bytes where the range starts | |
803 | * @lend: offset in bytes where the range ends (inclusive) | |
804 | * | |
805 | * Write out and wait upon file offsets lstart->lend, inclusive. | |
806 | * | |
807 | * Note that @lend is inclusive (describes the last byte to be written) so | |
808 | * that this function can be used to write to the very end-of-file (end = -1). | |
809 | * | |
810 | * After writing out and waiting on the data, we check and advance the | |
811 | * f_wb_err cursor to the latest value, and return any errors detected there. | |
a862f68a MR |
812 | * |
813 | * Return: %0 on success, negative error code otherwise. | |
5660e13d JL |
814 | */ |
815 | int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend) | |
816 | { | |
817 | int err = 0, err2; | |
818 | struct address_space *mapping = file->f_mapping; | |
819 | ||
9326c9b2 | 820 | if (mapping_needs_writeback(mapping)) { |
5660e13d JL |
821 | err = __filemap_fdatawrite_range(mapping, lstart, lend, |
822 | WB_SYNC_ALL); | |
823 | /* See comment of filemap_write_and_wait() */ | |
824 | if (err != -EIO) | |
825 | __filemap_fdatawait_range(mapping, lstart, lend); | |
826 | } | |
827 | err2 = file_check_and_advance_wb_err(file); | |
828 | if (!err) | |
829 | err = err2; | |
830 | return err; | |
831 | } | |
832 | EXPORT_SYMBOL(file_write_and_wait_range); | |
833 | ||
ef6a3c63 MS |
834 | /** |
835 | * replace_page_cache_page - replace a pagecache page with a new one | |
836 | * @old: page to be replaced | |
837 | * @new: page to replace with | |
ef6a3c63 MS |
838 | * |
839 | * This function replaces a page in the pagecache with a new one. On | |
840 | * success it acquires the pagecache reference for the new page and | |
841 | * drops it for the old page. Both the old and new pages must be | |
842 | * locked. This function does not add the new page to the LRU, the | |
843 | * caller must do that. | |
844 | * | |
74d60958 | 845 | * The remove + add is atomic. This function cannot fail. |
ef6a3c63 | 846 | */ |
1f7ef657 | 847 | void replace_page_cache_page(struct page *old, struct page *new) |
ef6a3c63 | 848 | { |
d21bba2b MWO |
849 | struct folio *fold = page_folio(old); |
850 | struct folio *fnew = page_folio(new); | |
74d60958 MW |
851 | struct address_space *mapping = old->mapping; |
852 | void (*freepage)(struct page *) = mapping->a_ops->freepage; | |
853 | pgoff_t offset = old->index; | |
854 | XA_STATE(xas, &mapping->i_pages, offset); | |
ef6a3c63 | 855 | |
309381fe SL |
856 | VM_BUG_ON_PAGE(!PageLocked(old), old); |
857 | VM_BUG_ON_PAGE(!PageLocked(new), new); | |
858 | VM_BUG_ON_PAGE(new->mapping, new); | |
ef6a3c63 | 859 | |
74d60958 MW |
860 | get_page(new); |
861 | new->mapping = mapping; | |
862 | new->index = offset; | |
ef6a3c63 | 863 | |
d21bba2b | 864 | mem_cgroup_migrate(fold, fnew); |
0d1c2072 | 865 | |
30472509 | 866 | xas_lock_irq(&xas); |
74d60958 | 867 | xas_store(&xas, new); |
4165b9b4 | 868 | |
74d60958 MW |
869 | old->mapping = NULL; |
870 | /* hugetlb pages do not participate in page cache accounting. */ | |
871 | if (!PageHuge(old)) | |
0d1c2072 | 872 | __dec_lruvec_page_state(old, NR_FILE_PAGES); |
74d60958 | 873 | if (!PageHuge(new)) |
0d1c2072 | 874 | __inc_lruvec_page_state(new, NR_FILE_PAGES); |
74d60958 | 875 | if (PageSwapBacked(old)) |
0d1c2072 | 876 | __dec_lruvec_page_state(old, NR_SHMEM); |
74d60958 | 877 | if (PageSwapBacked(new)) |
0d1c2072 | 878 | __inc_lruvec_page_state(new, NR_SHMEM); |
30472509 | 879 | xas_unlock_irq(&xas); |
74d60958 MW |
880 | if (freepage) |
881 | freepage(old); | |
882 | put_page(old); | |
ef6a3c63 MS |
883 | } |
884 | EXPORT_SYMBOL_GPL(replace_page_cache_page); | |
885 | ||
9dd3d069 MWO |
886 | noinline int __filemap_add_folio(struct address_space *mapping, |
887 | struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp) | |
1da177e4 | 888 | { |
9dd3d069 MWO |
889 | XA_STATE(xas, &mapping->i_pages, index); |
890 | int huge = folio_test_hugetlb(folio); | |
e286781d | 891 | int error; |
da74240e | 892 | bool charged = false; |
e286781d | 893 | |
9dd3d069 MWO |
894 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
895 | VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio); | |
74d60958 | 896 | mapping_set_update(&xas, mapping); |
e286781d | 897 | |
9dd3d069 MWO |
898 | folio_get(folio); |
899 | folio->mapping = mapping; | |
900 | folio->index = index; | |
66a0c8ee | 901 | |
3fea5a49 | 902 | if (!huge) { |
9dd3d069 MWO |
903 | error = mem_cgroup_charge(folio, NULL, gfp); |
904 | VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio); | |
3fea5a49 JW |
905 | if (error) |
906 | goto error; | |
da74240e | 907 | charged = true; |
3fea5a49 JW |
908 | } |
909 | ||
198b62f8 MWO |
910 | gfp &= GFP_RECLAIM_MASK; |
911 | ||
74d60958 | 912 | do { |
198b62f8 MWO |
913 | unsigned int order = xa_get_order(xas.xa, xas.xa_index); |
914 | void *entry, *old = NULL; | |
915 | ||
9dd3d069 | 916 | if (order > folio_order(folio)) |
198b62f8 MWO |
917 | xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index), |
918 | order, gfp); | |
74d60958 | 919 | xas_lock_irq(&xas); |
198b62f8 MWO |
920 | xas_for_each_conflict(&xas, entry) { |
921 | old = entry; | |
922 | if (!xa_is_value(entry)) { | |
923 | xas_set_err(&xas, -EEXIST); | |
924 | goto unlock; | |
925 | } | |
926 | } | |
927 | ||
928 | if (old) { | |
929 | if (shadowp) | |
930 | *shadowp = old; | |
931 | /* entry may have been split before we acquired lock */ | |
932 | order = xa_get_order(xas.xa, xas.xa_index); | |
9dd3d069 | 933 | if (order > folio_order(folio)) { |
198b62f8 MWO |
934 | xas_split(&xas, old, order); |
935 | xas_reset(&xas); | |
936 | } | |
937 | } | |
938 | ||
9dd3d069 | 939 | xas_store(&xas, folio); |
74d60958 MW |
940 | if (xas_error(&xas)) |
941 | goto unlock; | |
942 | ||
74d60958 MW |
943 | mapping->nrpages++; |
944 | ||
945 | /* hugetlb pages do not participate in page cache accounting */ | |
946 | if (!huge) | |
9dd3d069 | 947 | __lruvec_stat_add_folio(folio, NR_FILE_PAGES); |
74d60958 MW |
948 | unlock: |
949 | xas_unlock_irq(&xas); | |
198b62f8 | 950 | } while (xas_nomem(&xas, gfp)); |
74d60958 | 951 | |
3fea5a49 JW |
952 | if (xas_error(&xas)) { |
953 | error = xas_error(&xas); | |
da74240e | 954 | if (charged) |
9dd3d069 | 955 | mem_cgroup_uncharge(folio); |
74d60958 | 956 | goto error; |
3fea5a49 | 957 | } |
4165b9b4 | 958 | |
a0580c6f | 959 | trace_mm_filemap_add_to_page_cache(folio); |
66a0c8ee | 960 | return 0; |
74d60958 | 961 | error: |
9dd3d069 | 962 | folio->mapping = NULL; |
66a0c8ee | 963 | /* Leave page->index set: truncation relies upon it */ |
9dd3d069 | 964 | folio_put(folio); |
3fea5a49 | 965 | return error; |
1da177e4 | 966 | } |
9dd3d069 | 967 | ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO); |
a528910e JW |
968 | |
969 | /** | |
970 | * add_to_page_cache_locked - add a locked page to the pagecache | |
971 | * @page: page to add | |
972 | * @mapping: the page's address_space | |
973 | * @offset: page index | |
974 | * @gfp_mask: page allocation mode | |
975 | * | |
976 | * This function is used to add a page to the pagecache. It must be locked. | |
977 | * This function does not add the page to the LRU. The caller must do that. | |
a862f68a MR |
978 | * |
979 | * Return: %0 on success, negative error code otherwise. | |
a528910e JW |
980 | */ |
981 | int add_to_page_cache_locked(struct page *page, struct address_space *mapping, | |
982 | pgoff_t offset, gfp_t gfp_mask) | |
983 | { | |
9dd3d069 | 984 | return __filemap_add_folio(mapping, page_folio(page), offset, |
a528910e JW |
985 | gfp_mask, NULL); |
986 | } | |
e286781d | 987 | EXPORT_SYMBOL(add_to_page_cache_locked); |
1da177e4 | 988 | |
9dd3d069 MWO |
989 | int filemap_add_folio(struct address_space *mapping, struct folio *folio, |
990 | pgoff_t index, gfp_t gfp) | |
1da177e4 | 991 | { |
a528910e | 992 | void *shadow = NULL; |
4f98a2fe RR |
993 | int ret; |
994 | ||
9dd3d069 MWO |
995 | __folio_set_locked(folio); |
996 | ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow); | |
a528910e | 997 | if (unlikely(ret)) |
9dd3d069 | 998 | __folio_clear_locked(folio); |
a528910e JW |
999 | else { |
1000 | /* | |
9dd3d069 | 1001 | * The folio might have been evicted from cache only |
a528910e | 1002 | * recently, in which case it should be activated like |
9dd3d069 MWO |
1003 | * any other repeatedly accessed folio. |
1004 | * The exception is folios getting rewritten; evicting other | |
f0281a00 RR |
1005 | * data from the working set, only to cache data that will |
1006 | * get overwritten with something else, is a waste of memory. | |
a528910e | 1007 | */ |
9dd3d069 MWO |
1008 | WARN_ON_ONCE(folio_test_active(folio)); |
1009 | if (!(gfp & __GFP_WRITE) && shadow) | |
1010 | workingset_refault(folio, shadow); | |
1011 | folio_add_lru(folio); | |
a528910e | 1012 | } |
1da177e4 LT |
1013 | return ret; |
1014 | } | |
9dd3d069 | 1015 | EXPORT_SYMBOL_GPL(filemap_add_folio); |
1da177e4 | 1016 | |
44110fe3 | 1017 | #ifdef CONFIG_NUMA |
bb3c579e | 1018 | struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order) |
44110fe3 | 1019 | { |
c0ff7453 | 1020 | int n; |
bb3c579e | 1021 | struct folio *folio; |
c0ff7453 | 1022 | |
44110fe3 | 1023 | if (cpuset_do_page_mem_spread()) { |
cc9a6c87 MG |
1024 | unsigned int cpuset_mems_cookie; |
1025 | do { | |
d26914d1 | 1026 | cpuset_mems_cookie = read_mems_allowed_begin(); |
cc9a6c87 | 1027 | n = cpuset_mem_spread_node(); |
bb3c579e MWO |
1028 | folio = __folio_alloc_node(gfp, order, n); |
1029 | } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie)); | |
cc9a6c87 | 1030 | |
bb3c579e | 1031 | return folio; |
44110fe3 | 1032 | } |
bb3c579e | 1033 | return folio_alloc(gfp, order); |
44110fe3 | 1034 | } |
bb3c579e | 1035 | EXPORT_SYMBOL(filemap_alloc_folio); |
44110fe3 PJ |
1036 | #endif |
1037 | ||
7506ae6a JK |
1038 | /* |
1039 | * filemap_invalidate_lock_two - lock invalidate_lock for two mappings | |
1040 | * | |
1041 | * Lock exclusively invalidate_lock of any passed mapping that is not NULL. | |
1042 | * | |
1043 | * @mapping1: the first mapping to lock | |
1044 | * @mapping2: the second mapping to lock | |
1045 | */ | |
1046 | void filemap_invalidate_lock_two(struct address_space *mapping1, | |
1047 | struct address_space *mapping2) | |
1048 | { | |
1049 | if (mapping1 > mapping2) | |
1050 | swap(mapping1, mapping2); | |
1051 | if (mapping1) | |
1052 | down_write(&mapping1->invalidate_lock); | |
1053 | if (mapping2 && mapping1 != mapping2) | |
1054 | down_write_nested(&mapping2->invalidate_lock, 1); | |
1055 | } | |
1056 | EXPORT_SYMBOL(filemap_invalidate_lock_two); | |
1057 | ||
1058 | /* | |
1059 | * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings | |
1060 | * | |
1061 | * Unlock exclusive invalidate_lock of any passed mapping that is not NULL. | |
1062 | * | |
1063 | * @mapping1: the first mapping to unlock | |
1064 | * @mapping2: the second mapping to unlock | |
1065 | */ | |
1066 | void filemap_invalidate_unlock_two(struct address_space *mapping1, | |
1067 | struct address_space *mapping2) | |
1068 | { | |
1069 | if (mapping1) | |
1070 | up_write(&mapping1->invalidate_lock); | |
1071 | if (mapping2 && mapping1 != mapping2) | |
1072 | up_write(&mapping2->invalidate_lock); | |
1073 | } | |
1074 | EXPORT_SYMBOL(filemap_invalidate_unlock_two); | |
1075 | ||
1da177e4 LT |
1076 | /* |
1077 | * In order to wait for pages to become available there must be | |
1078 | * waitqueues associated with pages. By using a hash table of | |
1079 | * waitqueues where the bucket discipline is to maintain all | |
1080 | * waiters on the same queue and wake all when any of the pages | |
1081 | * become available, and for the woken contexts to check to be | |
1082 | * sure the appropriate page became available, this saves space | |
1083 | * at a cost of "thundering herd" phenomena during rare hash | |
1084 | * collisions. | |
1085 | */ | |
62906027 NP |
1086 | #define PAGE_WAIT_TABLE_BITS 8 |
1087 | #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS) | |
df4d4f12 | 1088 | static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned; |
62906027 | 1089 | |
df4d4f12 | 1090 | static wait_queue_head_t *folio_waitqueue(struct folio *folio) |
1da177e4 | 1091 | { |
df4d4f12 | 1092 | return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)]; |
1da177e4 | 1093 | } |
1da177e4 | 1094 | |
62906027 | 1095 | void __init pagecache_init(void) |
1da177e4 | 1096 | { |
62906027 | 1097 | int i; |
1da177e4 | 1098 | |
62906027 | 1099 | for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++) |
df4d4f12 | 1100 | init_waitqueue_head(&folio_wait_table[i]); |
62906027 NP |
1101 | |
1102 | page_writeback_init(); | |
1da177e4 | 1103 | } |
1da177e4 | 1104 | |
5ef64cc8 LT |
1105 | /* |
1106 | * The page wait code treats the "wait->flags" somewhat unusually, because | |
5868ec26 | 1107 | * we have multiple different kinds of waits, not just the usual "exclusive" |
5ef64cc8 LT |
1108 | * one. |
1109 | * | |
1110 | * We have: | |
1111 | * | |
1112 | * (a) no special bits set: | |
1113 | * | |
1114 | * We're just waiting for the bit to be released, and when a waker | |
1115 | * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up, | |
1116 | * and remove it from the wait queue. | |
1117 | * | |
1118 | * Simple and straightforward. | |
1119 | * | |
1120 | * (b) WQ_FLAG_EXCLUSIVE: | |
1121 | * | |
1122 | * The waiter is waiting to get the lock, and only one waiter should | |
1123 | * be woken up to avoid any thundering herd behavior. We'll set the | |
1124 | * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue. | |
1125 | * | |
1126 | * This is the traditional exclusive wait. | |
1127 | * | |
5868ec26 | 1128 | * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM: |
5ef64cc8 LT |
1129 | * |
1130 | * The waiter is waiting to get the bit, and additionally wants the | |
1131 | * lock to be transferred to it for fair lock behavior. If the lock | |
1132 | * cannot be taken, we stop walking the wait queue without waking | |
1133 | * the waiter. | |
1134 | * | |
1135 | * This is the "fair lock handoff" case, and in addition to setting | |
1136 | * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see | |
1137 | * that it now has the lock. | |
1138 | */ | |
ac6424b9 | 1139 | static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg) |
f62e00cc | 1140 | { |
5ef64cc8 | 1141 | unsigned int flags; |
62906027 NP |
1142 | struct wait_page_key *key = arg; |
1143 | struct wait_page_queue *wait_page | |
1144 | = container_of(wait, struct wait_page_queue, wait); | |
1145 | ||
cdc8fcb4 | 1146 | if (!wake_page_match(wait_page, key)) |
62906027 | 1147 | return 0; |
3510ca20 | 1148 | |
9a1ea439 | 1149 | /* |
5ef64cc8 LT |
1150 | * If it's a lock handoff wait, we get the bit for it, and |
1151 | * stop walking (and do not wake it up) if we can't. | |
9a1ea439 | 1152 | */ |
5ef64cc8 LT |
1153 | flags = wait->flags; |
1154 | if (flags & WQ_FLAG_EXCLUSIVE) { | |
df4d4f12 | 1155 | if (test_bit(key->bit_nr, &key->folio->flags)) |
2a9127fc | 1156 | return -1; |
5ef64cc8 | 1157 | if (flags & WQ_FLAG_CUSTOM) { |
df4d4f12 | 1158 | if (test_and_set_bit(key->bit_nr, &key->folio->flags)) |
5ef64cc8 LT |
1159 | return -1; |
1160 | flags |= WQ_FLAG_DONE; | |
1161 | } | |
2a9127fc | 1162 | } |
f62e00cc | 1163 | |
5ef64cc8 LT |
1164 | /* |
1165 | * We are holding the wait-queue lock, but the waiter that | |
1166 | * is waiting for this will be checking the flags without | |
1167 | * any locking. | |
1168 | * | |
1169 | * So update the flags atomically, and wake up the waiter | |
1170 | * afterwards to avoid any races. This store-release pairs | |
101c0bf6 | 1171 | * with the load-acquire in folio_wait_bit_common(). |
5ef64cc8 LT |
1172 | */ |
1173 | smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN); | |
2a9127fc LT |
1174 | wake_up_state(wait->private, mode); |
1175 | ||
1176 | /* | |
1177 | * Ok, we have successfully done what we're waiting for, | |
1178 | * and we can unconditionally remove the wait entry. | |
1179 | * | |
5ef64cc8 LT |
1180 | * Note that this pairs with the "finish_wait()" in the |
1181 | * waiter, and has to be the absolute last thing we do. | |
1182 | * After this list_del_init(&wait->entry) the wait entry | |
2a9127fc LT |
1183 | * might be de-allocated and the process might even have |
1184 | * exited. | |
2a9127fc | 1185 | */ |
c6fe44d9 | 1186 | list_del_init_careful(&wait->entry); |
5ef64cc8 | 1187 | return (flags & WQ_FLAG_EXCLUSIVE) != 0; |
f62e00cc KM |
1188 | } |
1189 | ||
6974d7c9 | 1190 | static void folio_wake_bit(struct folio *folio, int bit_nr) |
cbbce822 | 1191 | { |
df4d4f12 | 1192 | wait_queue_head_t *q = folio_waitqueue(folio); |
62906027 NP |
1193 | struct wait_page_key key; |
1194 | unsigned long flags; | |
11a19c7b | 1195 | wait_queue_entry_t bookmark; |
cbbce822 | 1196 | |
df4d4f12 | 1197 | key.folio = folio; |
62906027 NP |
1198 | key.bit_nr = bit_nr; |
1199 | key.page_match = 0; | |
1200 | ||
11a19c7b TC |
1201 | bookmark.flags = 0; |
1202 | bookmark.private = NULL; | |
1203 | bookmark.func = NULL; | |
1204 | INIT_LIST_HEAD(&bookmark.entry); | |
1205 | ||
62906027 | 1206 | spin_lock_irqsave(&q->lock, flags); |
11a19c7b TC |
1207 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); |
1208 | ||
1209 | while (bookmark.flags & WQ_FLAG_BOOKMARK) { | |
1210 | /* | |
1211 | * Take a breather from holding the lock, | |
1212 | * allow pages that finish wake up asynchronously | |
1213 | * to acquire the lock and remove themselves | |
1214 | * from wait queue | |
1215 | */ | |
1216 | spin_unlock_irqrestore(&q->lock, flags); | |
1217 | cpu_relax(); | |
1218 | spin_lock_irqsave(&q->lock, flags); | |
1219 | __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark); | |
1220 | } | |
1221 | ||
62906027 NP |
1222 | /* |
1223 | * It is possible for other pages to have collided on the waitqueue | |
1224 | * hash, so in that case check for a page match. That prevents a long- | |
1225 | * term waiter | |
1226 | * | |
1227 | * It is still possible to miss a case here, when we woke page waiters | |
1228 | * and removed them from the waitqueue, but there are still other | |
1229 | * page waiters. | |
1230 | */ | |
1231 | if (!waitqueue_active(q) || !key.page_match) { | |
6974d7c9 | 1232 | folio_clear_waiters(folio); |
62906027 NP |
1233 | /* |
1234 | * It's possible to miss clearing Waiters here, when we woke | |
1235 | * our page waiters, but the hashed waitqueue has waiters for | |
1236 | * other pages on it. | |
1237 | * | |
1238 | * That's okay, it's a rare case. The next waker will clear it. | |
1239 | */ | |
1240 | } | |
1241 | spin_unlock_irqrestore(&q->lock, flags); | |
1242 | } | |
74d81bfa | 1243 | |
4268b480 | 1244 | static void folio_wake(struct folio *folio, int bit) |
74d81bfa | 1245 | { |
4268b480 | 1246 | if (!folio_test_waiters(folio)) |
74d81bfa | 1247 | return; |
6974d7c9 | 1248 | folio_wake_bit(folio, bit); |
74d81bfa | 1249 | } |
62906027 | 1250 | |
9a1ea439 | 1251 | /* |
101c0bf6 | 1252 | * A choice of three behaviors for folio_wait_bit_common(): |
9a1ea439 HD |
1253 | */ |
1254 | enum behavior { | |
1255 | EXCLUSIVE, /* Hold ref to page and take the bit when woken, like | |
7c23c782 | 1256 | * __folio_lock() waiting on then setting PG_locked. |
9a1ea439 HD |
1257 | */ |
1258 | SHARED, /* Hold ref to page and check the bit when woken, like | |
9f2b04a2 | 1259 | * folio_wait_writeback() waiting on PG_writeback. |
9a1ea439 HD |
1260 | */ |
1261 | DROP, /* Drop ref to page before wait, no check when woken, | |
9f2b04a2 | 1262 | * like folio_put_wait_locked() on PG_locked. |
9a1ea439 HD |
1263 | */ |
1264 | }; | |
1265 | ||
2a9127fc | 1266 | /* |
101c0bf6 | 1267 | * Attempt to check (or get) the folio flag, and mark us done |
5ef64cc8 | 1268 | * if successful. |
2a9127fc | 1269 | */ |
101c0bf6 | 1270 | static inline bool folio_trylock_flag(struct folio *folio, int bit_nr, |
2a9127fc LT |
1271 | struct wait_queue_entry *wait) |
1272 | { | |
1273 | if (wait->flags & WQ_FLAG_EXCLUSIVE) { | |
101c0bf6 | 1274 | if (test_and_set_bit(bit_nr, &folio->flags)) |
2a9127fc | 1275 | return false; |
101c0bf6 | 1276 | } else if (test_bit(bit_nr, &folio->flags)) |
2a9127fc LT |
1277 | return false; |
1278 | ||
5ef64cc8 | 1279 | wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE; |
2a9127fc LT |
1280 | return true; |
1281 | } | |
1282 | ||
5ef64cc8 LT |
1283 | /* How many times do we accept lock stealing from under a waiter? */ |
1284 | int sysctl_page_lock_unfairness = 5; | |
1285 | ||
101c0bf6 MWO |
1286 | static inline int folio_wait_bit_common(struct folio *folio, int bit_nr, |
1287 | int state, enum behavior behavior) | |
62906027 | 1288 | { |
df4d4f12 | 1289 | wait_queue_head_t *q = folio_waitqueue(folio); |
5ef64cc8 | 1290 | int unfairness = sysctl_page_lock_unfairness; |
62906027 | 1291 | struct wait_page_queue wait_page; |
ac6424b9 | 1292 | wait_queue_entry_t *wait = &wait_page.wait; |
b1d29ba8 | 1293 | bool thrashing = false; |
9a1ea439 | 1294 | bool delayacct = false; |
eb414681 | 1295 | unsigned long pflags; |
62906027 | 1296 | |
eb414681 | 1297 | if (bit_nr == PG_locked && |
101c0bf6 MWO |
1298 | !folio_test_uptodate(folio) && folio_test_workingset(folio)) { |
1299 | if (!folio_test_swapbacked(folio)) { | |
eb414681 | 1300 | delayacct_thrashing_start(); |
9a1ea439 HD |
1301 | delayacct = true; |
1302 | } | |
eb414681 | 1303 | psi_memstall_enter(&pflags); |
b1d29ba8 JW |
1304 | thrashing = true; |
1305 | } | |
1306 | ||
62906027 NP |
1307 | init_wait(wait); |
1308 | wait->func = wake_page_function; | |
df4d4f12 | 1309 | wait_page.folio = folio; |
62906027 NP |
1310 | wait_page.bit_nr = bit_nr; |
1311 | ||
5ef64cc8 LT |
1312 | repeat: |
1313 | wait->flags = 0; | |
1314 | if (behavior == EXCLUSIVE) { | |
1315 | wait->flags = WQ_FLAG_EXCLUSIVE; | |
1316 | if (--unfairness < 0) | |
1317 | wait->flags |= WQ_FLAG_CUSTOM; | |
1318 | } | |
1319 | ||
2a9127fc LT |
1320 | /* |
1321 | * Do one last check whether we can get the | |
1322 | * page bit synchronously. | |
1323 | * | |
101c0bf6 | 1324 | * Do the folio_set_waiters() marking before that |
2a9127fc LT |
1325 | * to let any waker we _just_ missed know they |
1326 | * need to wake us up (otherwise they'll never | |
1327 | * even go to the slow case that looks at the | |
1328 | * page queue), and add ourselves to the wait | |
1329 | * queue if we need to sleep. | |
1330 | * | |
1331 | * This part needs to be done under the queue | |
1332 | * lock to avoid races. | |
1333 | */ | |
1334 | spin_lock_irq(&q->lock); | |
101c0bf6 MWO |
1335 | folio_set_waiters(folio); |
1336 | if (!folio_trylock_flag(folio, bit_nr, wait)) | |
2a9127fc LT |
1337 | __add_wait_queue_entry_tail(q, wait); |
1338 | spin_unlock_irq(&q->lock); | |
62906027 | 1339 | |
2a9127fc LT |
1340 | /* |
1341 | * From now on, all the logic will be based on | |
5ef64cc8 LT |
1342 | * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to |
1343 | * see whether the page bit testing has already | |
1344 | * been done by the wake function. | |
2a9127fc | 1345 | * |
101c0bf6 | 1346 | * We can drop our reference to the folio. |
2a9127fc LT |
1347 | */ |
1348 | if (behavior == DROP) | |
101c0bf6 | 1349 | folio_put(folio); |
62906027 | 1350 | |
5ef64cc8 LT |
1351 | /* |
1352 | * Note that until the "finish_wait()", or until | |
1353 | * we see the WQ_FLAG_WOKEN flag, we need to | |
1354 | * be very careful with the 'wait->flags', because | |
1355 | * we may race with a waker that sets them. | |
1356 | */ | |
2a9127fc | 1357 | for (;;) { |
5ef64cc8 LT |
1358 | unsigned int flags; |
1359 | ||
62906027 NP |
1360 | set_current_state(state); |
1361 | ||
5ef64cc8 LT |
1362 | /* Loop until we've been woken or interrupted */ |
1363 | flags = smp_load_acquire(&wait->flags); | |
1364 | if (!(flags & WQ_FLAG_WOKEN)) { | |
1365 | if (signal_pending_state(state, current)) | |
1366 | break; | |
1367 | ||
1368 | io_schedule(); | |
1369 | continue; | |
1370 | } | |
1371 | ||
1372 | /* If we were non-exclusive, we're done */ | |
1373 | if (behavior != EXCLUSIVE) | |
a8b169af | 1374 | break; |
9a1ea439 | 1375 | |
5ef64cc8 LT |
1376 | /* If the waker got the lock for us, we're done */ |
1377 | if (flags & WQ_FLAG_DONE) | |
9a1ea439 | 1378 | break; |
2a9127fc | 1379 | |
5ef64cc8 LT |
1380 | /* |
1381 | * Otherwise, if we're getting the lock, we need to | |
1382 | * try to get it ourselves. | |
1383 | * | |
1384 | * And if that fails, we'll have to retry this all. | |
1385 | */ | |
101c0bf6 | 1386 | if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0)))) |
5ef64cc8 LT |
1387 | goto repeat; |
1388 | ||
1389 | wait->flags |= WQ_FLAG_DONE; | |
1390 | break; | |
62906027 NP |
1391 | } |
1392 | ||
5ef64cc8 LT |
1393 | /* |
1394 | * If a signal happened, this 'finish_wait()' may remove the last | |
101c0bf6 | 1395 | * waiter from the wait-queues, but the folio waiters bit will remain |
5ef64cc8 LT |
1396 | * set. That's ok. The next wakeup will take care of it, and trying |
1397 | * to do it here would be difficult and prone to races. | |
1398 | */ | |
62906027 NP |
1399 | finish_wait(q, wait); |
1400 | ||
eb414681 | 1401 | if (thrashing) { |
9a1ea439 | 1402 | if (delayacct) |
eb414681 JW |
1403 | delayacct_thrashing_end(); |
1404 | psi_memstall_leave(&pflags); | |
1405 | } | |
b1d29ba8 | 1406 | |
62906027 | 1407 | /* |
5ef64cc8 LT |
1408 | * NOTE! The wait->flags weren't stable until we've done the |
1409 | * 'finish_wait()', and we could have exited the loop above due | |
1410 | * to a signal, and had a wakeup event happen after the signal | |
1411 | * test but before the 'finish_wait()'. | |
1412 | * | |
1413 | * So only after the finish_wait() can we reliably determine | |
1414 | * if we got woken up or not, so we can now figure out the final | |
1415 | * return value based on that state without races. | |
1416 | * | |
1417 | * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive | |
1418 | * waiter, but an exclusive one requires WQ_FLAG_DONE. | |
62906027 | 1419 | */ |
5ef64cc8 LT |
1420 | if (behavior == EXCLUSIVE) |
1421 | return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR; | |
62906027 | 1422 | |
2a9127fc | 1423 | return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR; |
62906027 NP |
1424 | } |
1425 | ||
101c0bf6 | 1426 | void folio_wait_bit(struct folio *folio, int bit_nr) |
62906027 | 1427 | { |
101c0bf6 | 1428 | folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED); |
62906027 | 1429 | } |
101c0bf6 | 1430 | EXPORT_SYMBOL(folio_wait_bit); |
62906027 | 1431 | |
101c0bf6 | 1432 | int folio_wait_bit_killable(struct folio *folio, int bit_nr) |
62906027 | 1433 | { |
101c0bf6 | 1434 | return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED); |
cbbce822 | 1435 | } |
101c0bf6 | 1436 | EXPORT_SYMBOL(folio_wait_bit_killable); |
cbbce822 | 1437 | |
9a1ea439 | 1438 | /** |
9f2b04a2 MWO |
1439 | * folio_put_wait_locked - Drop a reference and wait for it to be unlocked |
1440 | * @folio: The folio to wait for. | |
48054625 | 1441 | * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc). |
9a1ea439 | 1442 | * |
9f2b04a2 | 1443 | * The caller should hold a reference on @folio. They expect the page to |
9a1ea439 | 1444 | * become unlocked relatively soon, but do not wish to hold up migration |
9f2b04a2 | 1445 | * (for example) by holding the reference while waiting for the folio to |
9a1ea439 | 1446 | * come unlocked. After this function returns, the caller should not |
9f2b04a2 | 1447 | * dereference @folio. |
48054625 | 1448 | * |
9f2b04a2 | 1449 | * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal. |
9a1ea439 | 1450 | */ |
9f2b04a2 | 1451 | int folio_put_wait_locked(struct folio *folio, int state) |
9a1ea439 | 1452 | { |
9f2b04a2 | 1453 | return folio_wait_bit_common(folio, PG_locked, state, DROP); |
9a1ea439 HD |
1454 | } |
1455 | ||
385e1ca5 | 1456 | /** |
df4d4f12 MWO |
1457 | * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue |
1458 | * @folio: Folio defining the wait queue of interest | |
697f619f | 1459 | * @waiter: Waiter to add to the queue |
385e1ca5 | 1460 | * |
df4d4f12 | 1461 | * Add an arbitrary @waiter to the wait queue for the nominated @folio. |
385e1ca5 | 1462 | */ |
df4d4f12 | 1463 | void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter) |
385e1ca5 | 1464 | { |
df4d4f12 | 1465 | wait_queue_head_t *q = folio_waitqueue(folio); |
385e1ca5 DH |
1466 | unsigned long flags; |
1467 | ||
1468 | spin_lock_irqsave(&q->lock, flags); | |
9c3a815f | 1469 | __add_wait_queue_entry_tail(q, waiter); |
df4d4f12 | 1470 | folio_set_waiters(folio); |
385e1ca5 DH |
1471 | spin_unlock_irqrestore(&q->lock, flags); |
1472 | } | |
df4d4f12 | 1473 | EXPORT_SYMBOL_GPL(folio_add_wait_queue); |
385e1ca5 | 1474 | |
b91e1302 LT |
1475 | #ifndef clear_bit_unlock_is_negative_byte |
1476 | ||
1477 | /* | |
1478 | * PG_waiters is the high bit in the same byte as PG_lock. | |
1479 | * | |
1480 | * On x86 (and on many other architectures), we can clear PG_lock and | |
1481 | * test the sign bit at the same time. But if the architecture does | |
1482 | * not support that special operation, we just do this all by hand | |
1483 | * instead. | |
1484 | * | |
1485 | * The read of PG_waiters has to be after (or concurrently with) PG_locked | |
ffceeb62 | 1486 | * being cleared, but a memory barrier should be unnecessary since it is |
b91e1302 LT |
1487 | * in the same byte as PG_locked. |
1488 | */ | |
1489 | static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem) | |
1490 | { | |
1491 | clear_bit_unlock(nr, mem); | |
1492 | /* smp_mb__after_atomic(); */ | |
98473f9f | 1493 | return test_bit(PG_waiters, mem); |
b91e1302 LT |
1494 | } |
1495 | ||
1496 | #endif | |
1497 | ||
1da177e4 | 1498 | /** |
4e136428 MWO |
1499 | * folio_unlock - Unlock a locked folio. |
1500 | * @folio: The folio. | |
1501 | * | |
1502 | * Unlocks the folio and wakes up any thread sleeping on the page lock. | |
1503 | * | |
1504 | * Context: May be called from interrupt or process context. May not be | |
1505 | * called from NMI context. | |
1da177e4 | 1506 | */ |
4e136428 | 1507 | void folio_unlock(struct folio *folio) |
1da177e4 | 1508 | { |
4e136428 | 1509 | /* Bit 7 allows x86 to check the byte's sign bit */ |
b91e1302 | 1510 | BUILD_BUG_ON(PG_waiters != 7); |
4e136428 MWO |
1511 | BUILD_BUG_ON(PG_locked > 7); |
1512 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); | |
1513 | if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0))) | |
6974d7c9 | 1514 | folio_wake_bit(folio, PG_locked); |
1da177e4 | 1515 | } |
4e136428 | 1516 | EXPORT_SYMBOL(folio_unlock); |
1da177e4 | 1517 | |
73e10ded | 1518 | /** |
b47393f8 MWO |
1519 | * folio_end_private_2 - Clear PG_private_2 and wake any waiters. |
1520 | * @folio: The folio. | |
73e10ded | 1521 | * |
b47393f8 MWO |
1522 | * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for |
1523 | * it. The folio reference held for PG_private_2 being set is released. | |
73e10ded | 1524 | * |
b47393f8 MWO |
1525 | * This is, for example, used when a netfs folio is being written to a local |
1526 | * disk cache, thereby allowing writes to the cache for the same folio to be | |
73e10ded DH |
1527 | * serialised. |
1528 | */ | |
b47393f8 | 1529 | void folio_end_private_2(struct folio *folio) |
73e10ded | 1530 | { |
6974d7c9 MWO |
1531 | VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio); |
1532 | clear_bit_unlock(PG_private_2, folio_flags(folio, 0)); | |
1533 | folio_wake_bit(folio, PG_private_2); | |
1534 | folio_put(folio); | |
73e10ded | 1535 | } |
b47393f8 | 1536 | EXPORT_SYMBOL(folio_end_private_2); |
73e10ded DH |
1537 | |
1538 | /** | |
b47393f8 MWO |
1539 | * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio. |
1540 | * @folio: The folio to wait on. | |
73e10ded | 1541 | * |
b47393f8 | 1542 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio. |
73e10ded | 1543 | */ |
b47393f8 | 1544 | void folio_wait_private_2(struct folio *folio) |
73e10ded | 1545 | { |
101c0bf6 MWO |
1546 | while (folio_test_private_2(folio)) |
1547 | folio_wait_bit(folio, PG_private_2); | |
73e10ded | 1548 | } |
b47393f8 | 1549 | EXPORT_SYMBOL(folio_wait_private_2); |
73e10ded DH |
1550 | |
1551 | /** | |
b47393f8 MWO |
1552 | * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio. |
1553 | * @folio: The folio to wait on. | |
73e10ded | 1554 | * |
b47393f8 | 1555 | * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a |
73e10ded DH |
1556 | * fatal signal is received by the calling task. |
1557 | * | |
1558 | * Return: | |
1559 | * - 0 if successful. | |
1560 | * - -EINTR if a fatal signal was encountered. | |
1561 | */ | |
b47393f8 | 1562 | int folio_wait_private_2_killable(struct folio *folio) |
73e10ded DH |
1563 | { |
1564 | int ret = 0; | |
1565 | ||
101c0bf6 MWO |
1566 | while (folio_test_private_2(folio)) { |
1567 | ret = folio_wait_bit_killable(folio, PG_private_2); | |
73e10ded DH |
1568 | if (ret < 0) |
1569 | break; | |
1570 | } | |
1571 | ||
1572 | return ret; | |
1573 | } | |
b47393f8 | 1574 | EXPORT_SYMBOL(folio_wait_private_2_killable); |
73e10ded | 1575 | |
485bb99b | 1576 | /** |
4268b480 MWO |
1577 | * folio_end_writeback - End writeback against a folio. |
1578 | * @folio: The folio. | |
1da177e4 | 1579 | */ |
4268b480 | 1580 | void folio_end_writeback(struct folio *folio) |
1da177e4 | 1581 | { |
888cf2db | 1582 | /* |
4268b480 MWO |
1583 | * folio_test_clear_reclaim() could be used here but it is an |
1584 | * atomic operation and overkill in this particular case. Failing | |
1585 | * to shuffle a folio marked for immediate reclaim is too mild | |
1586 | * a gain to justify taking an atomic operation penalty at the | |
1587 | * end of every folio writeback. | |
888cf2db | 1588 | */ |
4268b480 MWO |
1589 | if (folio_test_reclaim(folio)) { |
1590 | folio_clear_reclaim(folio); | |
575ced1c | 1591 | folio_rotate_reclaimable(folio); |
888cf2db | 1592 | } |
ac6aadb2 | 1593 | |
073861ed | 1594 | /* |
4268b480 | 1595 | * Writeback does not hold a folio reference of its own, relying |
073861ed | 1596 | * on truncation to wait for the clearing of PG_writeback. |
4268b480 MWO |
1597 | * But here we must make sure that the folio is not freed and |
1598 | * reused before the folio_wake(). | |
073861ed | 1599 | */ |
4268b480 | 1600 | folio_get(folio); |
269ccca3 | 1601 | if (!__folio_end_writeback(folio)) |
ac6aadb2 MS |
1602 | BUG(); |
1603 | ||
4e857c58 | 1604 | smp_mb__after_atomic(); |
4268b480 | 1605 | folio_wake(folio, PG_writeback); |
512b7931 | 1606 | acct_reclaim_writeback(folio); |
4268b480 | 1607 | folio_put(folio); |
1da177e4 | 1608 | } |
4268b480 | 1609 | EXPORT_SYMBOL(folio_end_writeback); |
1da177e4 | 1610 | |
57d99845 MW |
1611 | /* |
1612 | * After completing I/O on a page, call this routine to update the page | |
1613 | * flags appropriately | |
1614 | */ | |
c11f0c0b | 1615 | void page_endio(struct page *page, bool is_write, int err) |
57d99845 | 1616 | { |
c11f0c0b | 1617 | if (!is_write) { |
57d99845 MW |
1618 | if (!err) { |
1619 | SetPageUptodate(page); | |
1620 | } else { | |
1621 | ClearPageUptodate(page); | |
1622 | SetPageError(page); | |
1623 | } | |
1624 | unlock_page(page); | |
abf54548 | 1625 | } else { |
57d99845 | 1626 | if (err) { |
dd8416c4 MK |
1627 | struct address_space *mapping; |
1628 | ||
57d99845 | 1629 | SetPageError(page); |
dd8416c4 MK |
1630 | mapping = page_mapping(page); |
1631 | if (mapping) | |
1632 | mapping_set_error(mapping, err); | |
57d99845 MW |
1633 | } |
1634 | end_page_writeback(page); | |
1635 | } | |
1636 | } | |
1637 | EXPORT_SYMBOL_GPL(page_endio); | |
1638 | ||
485bb99b | 1639 | /** |
7c23c782 MWO |
1640 | * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it. |
1641 | * @folio: The folio to lock | |
1da177e4 | 1642 | */ |
7c23c782 | 1643 | void __folio_lock(struct folio *folio) |
1da177e4 | 1644 | { |
101c0bf6 | 1645 | folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE, |
9a1ea439 | 1646 | EXCLUSIVE); |
1da177e4 | 1647 | } |
7c23c782 | 1648 | EXPORT_SYMBOL(__folio_lock); |
1da177e4 | 1649 | |
af7f29d9 | 1650 | int __folio_lock_killable(struct folio *folio) |
2687a356 | 1651 | { |
101c0bf6 | 1652 | return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE, |
9a1ea439 | 1653 | EXCLUSIVE); |
2687a356 | 1654 | } |
af7f29d9 | 1655 | EXPORT_SYMBOL_GPL(__folio_lock_killable); |
2687a356 | 1656 | |
ffdc8dab | 1657 | static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait) |
dd3e6d50 | 1658 | { |
df4d4f12 | 1659 | struct wait_queue_head *q = folio_waitqueue(folio); |
f32b5dd7 MWO |
1660 | int ret = 0; |
1661 | ||
df4d4f12 | 1662 | wait->folio = folio; |
f32b5dd7 MWO |
1663 | wait->bit_nr = PG_locked; |
1664 | ||
1665 | spin_lock_irq(&q->lock); | |
1666 | __add_wait_queue_entry_tail(q, &wait->wait); | |
ffdc8dab MWO |
1667 | folio_set_waiters(folio); |
1668 | ret = !folio_trylock(folio); | |
f32b5dd7 MWO |
1669 | /* |
1670 | * If we were successful now, we know we're still on the | |
1671 | * waitqueue as we're still under the lock. This means it's | |
1672 | * safe to remove and return success, we know the callback | |
1673 | * isn't going to trigger. | |
1674 | */ | |
1675 | if (!ret) | |
1676 | __remove_wait_queue(q, &wait->wait); | |
1677 | else | |
1678 | ret = -EIOCBQUEUED; | |
1679 | spin_unlock_irq(&q->lock); | |
1680 | return ret; | |
dd3e6d50 JA |
1681 | } |
1682 | ||
9a95f3cf PC |
1683 | /* |
1684 | * Return values: | |
9138e47e MWO |
1685 | * true - folio is locked; mmap_lock is still held. |
1686 | * false - folio is not locked. | |
3e4e28c5 | 1687 | * mmap_lock has been released (mmap_read_unlock(), unless flags had both |
9a95f3cf | 1688 | * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in |
c1e8d7c6 | 1689 | * which case mmap_lock is still held. |
9a95f3cf | 1690 | * |
9138e47e MWO |
1691 | * If neither ALLOW_RETRY nor KILLABLE are set, will always return true |
1692 | * with the folio locked and the mmap_lock unperturbed. | |
9a95f3cf | 1693 | */ |
9138e47e | 1694 | bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm, |
d065bd81 ML |
1695 | unsigned int flags) |
1696 | { | |
4064b982 | 1697 | if (fault_flag_allow_retry_first(flags)) { |
37b23e05 | 1698 | /* |
c1e8d7c6 | 1699 | * CAUTION! In this case, mmap_lock is not released |
37b23e05 KM |
1700 | * even though return 0. |
1701 | */ | |
1702 | if (flags & FAULT_FLAG_RETRY_NOWAIT) | |
9138e47e | 1703 | return false; |
37b23e05 | 1704 | |
d8ed45c5 | 1705 | mmap_read_unlock(mm); |
37b23e05 | 1706 | if (flags & FAULT_FLAG_KILLABLE) |
6baa8d60 | 1707 | folio_wait_locked_killable(folio); |
37b23e05 | 1708 | else |
6baa8d60 | 1709 | folio_wait_locked(folio); |
9138e47e | 1710 | return false; |
800bca7c HL |
1711 | } |
1712 | if (flags & FAULT_FLAG_KILLABLE) { | |
9138e47e | 1713 | bool ret; |
37b23e05 | 1714 | |
af7f29d9 | 1715 | ret = __folio_lock_killable(folio); |
800bca7c HL |
1716 | if (ret) { |
1717 | mmap_read_unlock(mm); | |
9138e47e | 1718 | return false; |
800bca7c HL |
1719 | } |
1720 | } else { | |
af7f29d9 | 1721 | __folio_lock(folio); |
d065bd81 | 1722 | } |
800bca7c | 1723 | |
9138e47e | 1724 | return true; |
d065bd81 ML |
1725 | } |
1726 | ||
e7b563bb | 1727 | /** |
0d3f9296 MW |
1728 | * page_cache_next_miss() - Find the next gap in the page cache. |
1729 | * @mapping: Mapping. | |
1730 | * @index: Index. | |
1731 | * @max_scan: Maximum range to search. | |
e7b563bb | 1732 | * |
0d3f9296 MW |
1733 | * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the |
1734 | * gap with the lowest index. | |
e7b563bb | 1735 | * |
0d3f9296 MW |
1736 | * This function may be called under the rcu_read_lock. However, this will |
1737 | * not atomically search a snapshot of the cache at a single point in time. | |
1738 | * For example, if a gap is created at index 5, then subsequently a gap is | |
1739 | * created at index 10, page_cache_next_miss covering both indices may | |
1740 | * return 10 if called under the rcu_read_lock. | |
e7b563bb | 1741 | * |
0d3f9296 MW |
1742 | * Return: The index of the gap if found, otherwise an index outside the |
1743 | * range specified (in which case 'return - index >= max_scan' will be true). | |
1744 | * In the rare case of index wrap-around, 0 will be returned. | |
e7b563bb | 1745 | */ |
0d3f9296 | 1746 | pgoff_t page_cache_next_miss(struct address_space *mapping, |
e7b563bb JW |
1747 | pgoff_t index, unsigned long max_scan) |
1748 | { | |
0d3f9296 | 1749 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1750 | |
0d3f9296 MW |
1751 | while (max_scan--) { |
1752 | void *entry = xas_next(&xas); | |
1753 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1754 | break; |
0d3f9296 | 1755 | if (xas.xa_index == 0) |
e7b563bb JW |
1756 | break; |
1757 | } | |
1758 | ||
0d3f9296 | 1759 | return xas.xa_index; |
e7b563bb | 1760 | } |
0d3f9296 | 1761 | EXPORT_SYMBOL(page_cache_next_miss); |
e7b563bb JW |
1762 | |
1763 | /** | |
2346a560 | 1764 | * page_cache_prev_miss() - Find the previous gap in the page cache. |
0d3f9296 MW |
1765 | * @mapping: Mapping. |
1766 | * @index: Index. | |
1767 | * @max_scan: Maximum range to search. | |
e7b563bb | 1768 | * |
0d3f9296 MW |
1769 | * Search the range [max(index - max_scan + 1, 0), index] for the |
1770 | * gap with the highest index. | |
e7b563bb | 1771 | * |
0d3f9296 MW |
1772 | * This function may be called under the rcu_read_lock. However, this will |
1773 | * not atomically search a snapshot of the cache at a single point in time. | |
1774 | * For example, if a gap is created at index 10, then subsequently a gap is | |
1775 | * created at index 5, page_cache_prev_miss() covering both indices may | |
1776 | * return 5 if called under the rcu_read_lock. | |
e7b563bb | 1777 | * |
0d3f9296 MW |
1778 | * Return: The index of the gap if found, otherwise an index outside the |
1779 | * range specified (in which case 'index - return >= max_scan' will be true). | |
1780 | * In the rare case of wrap-around, ULONG_MAX will be returned. | |
e7b563bb | 1781 | */ |
0d3f9296 | 1782 | pgoff_t page_cache_prev_miss(struct address_space *mapping, |
e7b563bb JW |
1783 | pgoff_t index, unsigned long max_scan) |
1784 | { | |
0d3f9296 | 1785 | XA_STATE(xas, &mapping->i_pages, index); |
e7b563bb | 1786 | |
0d3f9296 MW |
1787 | while (max_scan--) { |
1788 | void *entry = xas_prev(&xas); | |
1789 | if (!entry || xa_is_value(entry)) | |
e7b563bb | 1790 | break; |
0d3f9296 | 1791 | if (xas.xa_index == ULONG_MAX) |
e7b563bb JW |
1792 | break; |
1793 | } | |
1794 | ||
0d3f9296 | 1795 | return xas.xa_index; |
e7b563bb | 1796 | } |
0d3f9296 | 1797 | EXPORT_SYMBOL(page_cache_prev_miss); |
e7b563bb | 1798 | |
020853b6 MWO |
1799 | /* |
1800 | * Lockless page cache protocol: | |
1801 | * On the lookup side: | |
1802 | * 1. Load the folio from i_pages | |
1803 | * 2. Increment the refcount if it's not zero | |
1804 | * 3. If the folio is not found by xas_reload(), put the refcount and retry | |
1805 | * | |
1806 | * On the removal side: | |
1807 | * A. Freeze the page (by zeroing the refcount if nobody else has a reference) | |
1808 | * B. Remove the page from i_pages | |
1809 | * C. Return the page to the page allocator | |
1810 | * | |
1811 | * This means that any page may have its reference count temporarily | |
1812 | * increased by a speculative page cache (or fast GUP) lookup as it can | |
1813 | * be allocated by another user before the RCU grace period expires. | |
1814 | * Because the refcount temporarily acquired here may end up being the | |
1815 | * last refcount on the page, any page allocation must be freeable by | |
1816 | * folio_put(). | |
1817 | */ | |
1818 | ||
44835d20 | 1819 | /* |
bc5a3011 | 1820 | * mapping_get_entry - Get a page cache entry. |
485bb99b | 1821 | * @mapping: the address_space to search |
a6de4b48 | 1822 | * @index: The page cache index. |
0cd6144a | 1823 | * |
bca65eea MWO |
1824 | * Looks up the page cache entry at @mapping & @index. If it is a folio, |
1825 | * it is returned with an increased refcount. If it is a shadow entry | |
1826 | * of a previously evicted folio, or a swap entry from shmem/tmpfs, | |
1827 | * it is returned without further action. | |
485bb99b | 1828 | * |
bca65eea | 1829 | * Return: The folio, swap or shadow entry, %NULL if nothing is found. |
1da177e4 | 1830 | */ |
bca65eea | 1831 | static void *mapping_get_entry(struct address_space *mapping, pgoff_t index) |
1da177e4 | 1832 | { |
a6de4b48 | 1833 | XA_STATE(xas, &mapping->i_pages, index); |
bca65eea | 1834 | struct folio *folio; |
1da177e4 | 1835 | |
a60637c8 NP |
1836 | rcu_read_lock(); |
1837 | repeat: | |
4c7472c0 | 1838 | xas_reset(&xas); |
bca65eea MWO |
1839 | folio = xas_load(&xas); |
1840 | if (xas_retry(&xas, folio)) | |
4c7472c0 MW |
1841 | goto repeat; |
1842 | /* | |
1843 | * A shadow entry of a recently evicted page, or a swap entry from | |
1844 | * shmem/tmpfs. Return it without attempting to raise page count. | |
1845 | */ | |
bca65eea | 1846 | if (!folio || xa_is_value(folio)) |
4c7472c0 | 1847 | goto out; |
83929372 | 1848 | |
bca65eea | 1849 | if (!folio_try_get_rcu(folio)) |
4c7472c0 | 1850 | goto repeat; |
83929372 | 1851 | |
bca65eea MWO |
1852 | if (unlikely(folio != xas_reload(&xas))) { |
1853 | folio_put(folio); | |
4c7472c0 | 1854 | goto repeat; |
a60637c8 | 1855 | } |
27d20fdd | 1856 | out: |
a60637c8 NP |
1857 | rcu_read_unlock(); |
1858 | ||
bca65eea | 1859 | return folio; |
1da177e4 | 1860 | } |
1da177e4 | 1861 | |
0cd6144a | 1862 | /** |
3f0c6a07 | 1863 | * __filemap_get_folio - Find and get a reference to a folio. |
2294b32e MWO |
1864 | * @mapping: The address_space to search. |
1865 | * @index: The page index. | |
3f0c6a07 MWO |
1866 | * @fgp_flags: %FGP flags modify how the folio is returned. |
1867 | * @gfp: Memory allocation flags to use if %FGP_CREAT is specified. | |
1da177e4 | 1868 | * |
2294b32e | 1869 | * Looks up the page cache entry at @mapping & @index. |
0cd6144a | 1870 | * |
2294b32e | 1871 | * @fgp_flags can be zero or more of these flags: |
0e056eb5 | 1872 | * |
3f0c6a07 MWO |
1873 | * * %FGP_ACCESSED - The folio will be marked accessed. |
1874 | * * %FGP_LOCK - The folio is returned locked. | |
44835d20 | 1875 | * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it |
3f0c6a07 | 1876 | * instead of allocating a new folio to replace it. |
2294b32e | 1877 | * * %FGP_CREAT - If no page is present then a new page is allocated using |
3f0c6a07 | 1878 | * @gfp and added to the page cache and the VM's LRU list. |
2294b32e MWO |
1879 | * The page is returned locked and with an increased refcount. |
1880 | * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the | |
1881 | * page is already in cache. If the page was allocated, unlock it before | |
1882 | * returning so the caller can do the same dance. | |
3f0c6a07 MWO |
1883 | * * %FGP_WRITE - The page will be written to by the caller. |
1884 | * * %FGP_NOFS - __GFP_FS will get cleared in gfp. | |
1885 | * * %FGP_NOWAIT - Don't get blocked by page lock. | |
b27652d9 | 1886 | * * %FGP_STABLE - Wait for the folio to be stable (finished writeback) |
1da177e4 | 1887 | * |
2294b32e MWO |
1888 | * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even |
1889 | * if the %GFP flags specified for %FGP_CREAT are atomic. | |
1da177e4 | 1890 | * |
2457aec6 | 1891 | * If there is a page cache page, it is returned with an increased refcount. |
a862f68a | 1892 | * |
3f0c6a07 | 1893 | * Return: The found folio or %NULL otherwise. |
1da177e4 | 1894 | */ |
3f0c6a07 MWO |
1895 | struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index, |
1896 | int fgp_flags, gfp_t gfp) | |
1da177e4 | 1897 | { |
3f0c6a07 | 1898 | struct folio *folio; |
2457aec6 | 1899 | |
1da177e4 | 1900 | repeat: |
3f0c6a07 MWO |
1901 | folio = mapping_get_entry(mapping, index); |
1902 | if (xa_is_value(folio)) { | |
44835d20 | 1903 | if (fgp_flags & FGP_ENTRY) |
3f0c6a07 MWO |
1904 | return folio; |
1905 | folio = NULL; | |
44835d20 | 1906 | } |
3f0c6a07 | 1907 | if (!folio) |
2457aec6 MG |
1908 | goto no_page; |
1909 | ||
1910 | if (fgp_flags & FGP_LOCK) { | |
1911 | if (fgp_flags & FGP_NOWAIT) { | |
3f0c6a07 MWO |
1912 | if (!folio_trylock(folio)) { |
1913 | folio_put(folio); | |
2457aec6 MG |
1914 | return NULL; |
1915 | } | |
1916 | } else { | |
3f0c6a07 | 1917 | folio_lock(folio); |
2457aec6 MG |
1918 | } |
1919 | ||
1920 | /* Has the page been truncated? */ | |
3f0c6a07 MWO |
1921 | if (unlikely(folio->mapping != mapping)) { |
1922 | folio_unlock(folio); | |
1923 | folio_put(folio); | |
2457aec6 MG |
1924 | goto repeat; |
1925 | } | |
3f0c6a07 | 1926 | VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); |
2457aec6 MG |
1927 | } |
1928 | ||
c16eb000 | 1929 | if (fgp_flags & FGP_ACCESSED) |
3f0c6a07 | 1930 | folio_mark_accessed(folio); |
b9306a79 YS |
1931 | else if (fgp_flags & FGP_WRITE) { |
1932 | /* Clear idle flag for buffer write */ | |
3f0c6a07 MWO |
1933 | if (folio_test_idle(folio)) |
1934 | folio_clear_idle(folio); | |
b9306a79 | 1935 | } |
2457aec6 | 1936 | |
b27652d9 MWO |
1937 | if (fgp_flags & FGP_STABLE) |
1938 | folio_wait_stable(folio); | |
2457aec6 | 1939 | no_page: |
3f0c6a07 | 1940 | if (!folio && (fgp_flags & FGP_CREAT)) { |
2457aec6 | 1941 | int err; |
f56753ac | 1942 | if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping)) |
3f0c6a07 | 1943 | gfp |= __GFP_WRITE; |
45f87de5 | 1944 | if (fgp_flags & FGP_NOFS) |
3f0c6a07 | 1945 | gfp &= ~__GFP_FS; |
2457aec6 | 1946 | |
3f0c6a07 MWO |
1947 | folio = filemap_alloc_folio(gfp, 0); |
1948 | if (!folio) | |
eb2be189 | 1949 | return NULL; |
2457aec6 | 1950 | |
a75d4c33 | 1951 | if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP)))) |
2457aec6 MG |
1952 | fgp_flags |= FGP_LOCK; |
1953 | ||
eb39d618 | 1954 | /* Init accessed so avoid atomic mark_page_accessed later */ |
2457aec6 | 1955 | if (fgp_flags & FGP_ACCESSED) |
3f0c6a07 | 1956 | __folio_set_referenced(folio); |
2457aec6 | 1957 | |
3f0c6a07 | 1958 | err = filemap_add_folio(mapping, folio, index, gfp); |
eb2be189 | 1959 | if (unlikely(err)) { |
3f0c6a07 MWO |
1960 | folio_put(folio); |
1961 | folio = NULL; | |
eb2be189 NP |
1962 | if (err == -EEXIST) |
1963 | goto repeat; | |
1da177e4 | 1964 | } |
a75d4c33 JB |
1965 | |
1966 | /* | |
3f0c6a07 MWO |
1967 | * filemap_add_folio locks the page, and for mmap |
1968 | * we expect an unlocked page. | |
a75d4c33 | 1969 | */ |
3f0c6a07 MWO |
1970 | if (folio && (fgp_flags & FGP_FOR_MMAP)) |
1971 | folio_unlock(folio); | |
1da177e4 | 1972 | } |
2457aec6 | 1973 | |
3f0c6a07 | 1974 | return folio; |
1da177e4 | 1975 | } |
3f0c6a07 | 1976 | EXPORT_SYMBOL(__filemap_get_folio); |
1da177e4 | 1977 | |
f5e6429a | 1978 | static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max, |
c7bad633 MWO |
1979 | xa_mark_t mark) |
1980 | { | |
f5e6429a | 1981 | struct folio *folio; |
c7bad633 MWO |
1982 | |
1983 | retry: | |
1984 | if (mark == XA_PRESENT) | |
f5e6429a | 1985 | folio = xas_find(xas, max); |
c7bad633 | 1986 | else |
f5e6429a | 1987 | folio = xas_find_marked(xas, max, mark); |
c7bad633 | 1988 | |
f5e6429a | 1989 | if (xas_retry(xas, folio)) |
c7bad633 MWO |
1990 | goto retry; |
1991 | /* | |
1992 | * A shadow entry of a recently evicted page, a swap | |
1993 | * entry from shmem/tmpfs or a DAX entry. Return it | |
1994 | * without attempting to raise page count. | |
1995 | */ | |
f5e6429a MWO |
1996 | if (!folio || xa_is_value(folio)) |
1997 | return folio; | |
c7bad633 | 1998 | |
f5e6429a | 1999 | if (!folio_try_get_rcu(folio)) |
c7bad633 MWO |
2000 | goto reset; |
2001 | ||
f5e6429a MWO |
2002 | if (unlikely(folio != xas_reload(xas))) { |
2003 | folio_put(folio); | |
c7bad633 MWO |
2004 | goto reset; |
2005 | } | |
2006 | ||
f5e6429a | 2007 | return folio; |
c7bad633 MWO |
2008 | reset: |
2009 | xas_reset(xas); | |
2010 | goto retry; | |
2011 | } | |
2012 | ||
0cd6144a JW |
2013 | /** |
2014 | * find_get_entries - gang pagecache lookup | |
2015 | * @mapping: The address_space to search | |
2016 | * @start: The starting page cache index | |
ca122fe4 | 2017 | * @end: The final page index (inclusive). |
cf2039af | 2018 | * @pvec: Where the resulting entries are placed. |
0cd6144a JW |
2019 | * @indices: The cache indices corresponding to the entries in @entries |
2020 | * | |
cf2039af MWO |
2021 | * find_get_entries() will search for and return a batch of entries in |
2022 | * the mapping. The entries are placed in @pvec. find_get_entries() | |
2023 | * takes a reference on any actual pages it returns. | |
0cd6144a JW |
2024 | * |
2025 | * The search returns a group of mapping-contiguous page cache entries | |
2026 | * with ascending indexes. There may be holes in the indices due to | |
2027 | * not-present pages. | |
2028 | * | |
139b6a6f JW |
2029 | * Any shadow entries of evicted pages, or swap entries from |
2030 | * shmem/tmpfs, are included in the returned array. | |
0cd6144a | 2031 | * |
71725ed1 HD |
2032 | * If it finds a Transparent Huge Page, head or tail, find_get_entries() |
2033 | * stops at that page: the caller is likely to have a better way to handle | |
2034 | * the compound page as a whole, and then skip its extent, than repeatedly | |
2035 | * calling find_get_entries() to return all its tails. | |
2036 | * | |
a862f68a | 2037 | * Return: the number of pages and shadow entries which were found. |
0cd6144a | 2038 | */ |
ca122fe4 | 2039 | unsigned find_get_entries(struct address_space *mapping, pgoff_t start, |
cf2039af | 2040 | pgoff_t end, struct pagevec *pvec, pgoff_t *indices) |
0cd6144a | 2041 | { |
f280bf09 | 2042 | XA_STATE(xas, &mapping->i_pages, start); |
f5e6429a | 2043 | struct folio *folio; |
0cd6144a | 2044 | unsigned int ret = 0; |
cf2039af | 2045 | unsigned nr_entries = PAGEVEC_SIZE; |
0cd6144a JW |
2046 | |
2047 | rcu_read_lock(); | |
f5e6429a MWO |
2048 | while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) { |
2049 | struct page *page = &folio->page; | |
71725ed1 HD |
2050 | /* |
2051 | * Terminate early on finding a THP, to allow the caller to | |
2052 | * handle it all at once; but continue if this is hugetlbfs. | |
2053 | */ | |
f5e6429a MWO |
2054 | if (!xa_is_value(folio) && folio_test_large(folio) && |
2055 | !folio_test_hugetlb(folio)) { | |
2056 | page = folio_file_page(folio, xas.xa_index); | |
71725ed1 HD |
2057 | nr_entries = ret + 1; |
2058 | } | |
c7bad633 | 2059 | |
f280bf09 | 2060 | indices[ret] = xas.xa_index; |
cf2039af | 2061 | pvec->pages[ret] = page; |
0cd6144a JW |
2062 | if (++ret == nr_entries) |
2063 | break; | |
2064 | } | |
2065 | rcu_read_unlock(); | |
cf2039af MWO |
2066 | |
2067 | pvec->nr = ret; | |
0cd6144a JW |
2068 | return ret; |
2069 | } | |
2070 | ||
5c211ba2 MWO |
2071 | /** |
2072 | * find_lock_entries - Find a batch of pagecache entries. | |
2073 | * @mapping: The address_space to search. | |
2074 | * @start: The starting page cache index. | |
2075 | * @end: The final page index (inclusive). | |
2076 | * @pvec: Where the resulting entries are placed. | |
2077 | * @indices: The cache indices of the entries in @pvec. | |
2078 | * | |
2079 | * find_lock_entries() will return a batch of entries from @mapping. | |
f5e6429a MWO |
2080 | * Swap, shadow and DAX entries are included. Folios are returned |
2081 | * locked and with an incremented refcount. Folios which are locked | |
2082 | * by somebody else or under writeback are skipped. Folios which are | |
2083 | * partially outside the range are not returned. | |
5c211ba2 MWO |
2084 | * |
2085 | * The entries have ascending indexes. The indices may not be consecutive | |
f5e6429a MWO |
2086 | * due to not-present entries, large folios, folios which could not be |
2087 | * locked or folios under writeback. | |
5c211ba2 MWO |
2088 | * |
2089 | * Return: The number of entries which were found. | |
2090 | */ | |
2091 | unsigned find_lock_entries(struct address_space *mapping, pgoff_t start, | |
2092 | pgoff_t end, struct pagevec *pvec, pgoff_t *indices) | |
2093 | { | |
2094 | XA_STATE(xas, &mapping->i_pages, start); | |
f5e6429a | 2095 | struct folio *folio; |
5c211ba2 MWO |
2096 | |
2097 | rcu_read_lock(); | |
f5e6429a MWO |
2098 | while ((folio = find_get_entry(&xas, end, XA_PRESENT))) { |
2099 | if (!xa_is_value(folio)) { | |
2100 | if (folio->index < start) | |
5c211ba2 | 2101 | goto put; |
f5e6429a | 2102 | if (folio->index + folio_nr_pages(folio) - 1 > end) |
5c211ba2 | 2103 | goto put; |
f5e6429a | 2104 | if (!folio_trylock(folio)) |
5c211ba2 | 2105 | goto put; |
f5e6429a MWO |
2106 | if (folio->mapping != mapping || |
2107 | folio_test_writeback(folio)) | |
5c211ba2 | 2108 | goto unlock; |
f5e6429a MWO |
2109 | VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index), |
2110 | folio); | |
5c211ba2 MWO |
2111 | } |
2112 | indices[pvec->nr] = xas.xa_index; | |
f5e6429a | 2113 | if (!pagevec_add(pvec, &folio->page)) |
5c211ba2 MWO |
2114 | break; |
2115 | goto next; | |
2116 | unlock: | |
f5e6429a | 2117 | folio_unlock(folio); |
5c211ba2 | 2118 | put: |
f5e6429a | 2119 | folio_put(folio); |
5c211ba2 | 2120 | next: |
f5e6429a MWO |
2121 | if (!xa_is_value(folio) && folio_test_large(folio)) { |
2122 | xas_set(&xas, folio->index + folio_nr_pages(folio)); | |
2123 | /* Did we wrap on 32-bit? */ | |
2124 | if (!xas.xa_index) | |
2d11e738 HD |
2125 | break; |
2126 | } | |
5c211ba2 MWO |
2127 | } |
2128 | rcu_read_unlock(); | |
2129 | ||
2130 | return pagevec_count(pvec); | |
2131 | } | |
2132 | ||
1da177e4 | 2133 | /** |
b947cee4 | 2134 | * find_get_pages_range - gang pagecache lookup |
1da177e4 LT |
2135 | * @mapping: The address_space to search |
2136 | * @start: The starting page index | |
b947cee4 | 2137 | * @end: The final page index (inclusive) |
1da177e4 LT |
2138 | * @nr_pages: The maximum number of pages |
2139 | * @pages: Where the resulting pages are placed | |
2140 | * | |
b947cee4 JK |
2141 | * find_get_pages_range() will search for and return a group of up to @nr_pages |
2142 | * pages in the mapping starting at index @start and up to index @end | |
2143 | * (inclusive). The pages are placed at @pages. find_get_pages_range() takes | |
2144 | * a reference against the returned pages. | |
1da177e4 LT |
2145 | * |
2146 | * The search returns a group of mapping-contiguous pages with ascending | |
2147 | * indexes. There may be holes in the indices due to not-present pages. | |
d72dc8a2 | 2148 | * We also update @start to index the next page for the traversal. |
1da177e4 | 2149 | * |
a862f68a MR |
2150 | * Return: the number of pages which were found. If this number is |
2151 | * smaller than @nr_pages, the end of specified range has been | |
b947cee4 | 2152 | * reached. |
1da177e4 | 2153 | */ |
b947cee4 JK |
2154 | unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start, |
2155 | pgoff_t end, unsigned int nr_pages, | |
2156 | struct page **pages) | |
1da177e4 | 2157 | { |
fd1b3cee | 2158 | XA_STATE(xas, &mapping->i_pages, *start); |
f5e6429a | 2159 | struct folio *folio; |
0fc9d104 KK |
2160 | unsigned ret = 0; |
2161 | ||
2162 | if (unlikely(!nr_pages)) | |
2163 | return 0; | |
a60637c8 NP |
2164 | |
2165 | rcu_read_lock(); | |
f5e6429a | 2166 | while ((folio = find_get_entry(&xas, end, XA_PRESENT))) { |
fd1b3cee | 2167 | /* Skip over shadow, swap and DAX entries */ |
f5e6429a | 2168 | if (xa_is_value(folio)) |
8079b1c8 | 2169 | continue; |
a60637c8 | 2170 | |
f5e6429a | 2171 | pages[ret] = folio_file_page(folio, xas.xa_index); |
b947cee4 | 2172 | if (++ret == nr_pages) { |
5d3ee42f | 2173 | *start = xas.xa_index + 1; |
b947cee4 JK |
2174 | goto out; |
2175 | } | |
a60637c8 | 2176 | } |
5b280c0c | 2177 | |
b947cee4 JK |
2178 | /* |
2179 | * We come here when there is no page beyond @end. We take care to not | |
2180 | * overflow the index @start as it confuses some of the callers. This | |
fd1b3cee | 2181 | * breaks the iteration when there is a page at index -1 but that is |
b947cee4 JK |
2182 | * already broken anyway. |
2183 | */ | |
2184 | if (end == (pgoff_t)-1) | |
2185 | *start = (pgoff_t)-1; | |
2186 | else | |
2187 | *start = end + 1; | |
2188 | out: | |
a60637c8 | 2189 | rcu_read_unlock(); |
d72dc8a2 | 2190 | |
1da177e4 LT |
2191 | return ret; |
2192 | } | |
2193 | ||
ebf43500 JA |
2194 | /** |
2195 | * find_get_pages_contig - gang contiguous pagecache lookup | |
2196 | * @mapping: The address_space to search | |
2197 | * @index: The starting page index | |
2198 | * @nr_pages: The maximum number of pages | |
2199 | * @pages: Where the resulting pages are placed | |
2200 | * | |
2201 | * find_get_pages_contig() works exactly like find_get_pages(), except | |
2202 | * that the returned number of pages are guaranteed to be contiguous. | |
2203 | * | |
a862f68a | 2204 | * Return: the number of pages which were found. |
ebf43500 JA |
2205 | */ |
2206 | unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, | |
2207 | unsigned int nr_pages, struct page **pages) | |
2208 | { | |
3ece58a2 | 2209 | XA_STATE(xas, &mapping->i_pages, index); |
e1c37722 | 2210 | struct folio *folio; |
0fc9d104 KK |
2211 | unsigned int ret = 0; |
2212 | ||
2213 | if (unlikely(!nr_pages)) | |
2214 | return 0; | |
a60637c8 NP |
2215 | |
2216 | rcu_read_lock(); | |
e1c37722 MWO |
2217 | for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) { |
2218 | if (xas_retry(&xas, folio)) | |
3ece58a2 MW |
2219 | continue; |
2220 | /* | |
2221 | * If the entry has been swapped out, we can stop looking. | |
2222 | * No current caller is looking for DAX entries. | |
2223 | */ | |
e1c37722 | 2224 | if (xa_is_value(folio)) |
8079b1c8 | 2225 | break; |
ebf43500 | 2226 | |
e1c37722 | 2227 | if (!folio_try_get_rcu(folio)) |
3ece58a2 | 2228 | goto retry; |
83929372 | 2229 | |
e1c37722 | 2230 | if (unlikely(folio != xas_reload(&xas))) |
3ece58a2 | 2231 | goto put_page; |
a60637c8 | 2232 | |
e1c37722 | 2233 | pages[ret] = &folio->page; |
0fc9d104 KK |
2234 | if (++ret == nr_pages) |
2235 | break; | |
3ece58a2 MW |
2236 | continue; |
2237 | put_page: | |
e1c37722 | 2238 | folio_put(folio); |
3ece58a2 MW |
2239 | retry: |
2240 | xas_reset(&xas); | |
ebf43500 | 2241 | } |
a60637c8 NP |
2242 | rcu_read_unlock(); |
2243 | return ret; | |
ebf43500 | 2244 | } |
ef71c15c | 2245 | EXPORT_SYMBOL(find_get_pages_contig); |
ebf43500 | 2246 | |
485bb99b | 2247 | /** |
c49f50d1 | 2248 | * find_get_pages_range_tag - Find and return head pages matching @tag. |
485bb99b RD |
2249 | * @mapping: the address_space to search |
2250 | * @index: the starting page index | |
72b045ae | 2251 | * @end: The final page index (inclusive) |
485bb99b RD |
2252 | * @tag: the tag index |
2253 | * @nr_pages: the maximum number of pages | |
2254 | * @pages: where the resulting pages are placed | |
2255 | * | |
c49f50d1 MWO |
2256 | * Like find_get_pages(), except we only return head pages which are tagged |
2257 | * with @tag. @index is updated to the index immediately after the last | |
2258 | * page we return, ready for the next iteration. | |
a862f68a MR |
2259 | * |
2260 | * Return: the number of pages which were found. | |
1da177e4 | 2261 | */ |
72b045ae | 2262 | unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index, |
a6906972 | 2263 | pgoff_t end, xa_mark_t tag, unsigned int nr_pages, |
72b045ae | 2264 | struct page **pages) |
1da177e4 | 2265 | { |
a6906972 | 2266 | XA_STATE(xas, &mapping->i_pages, *index); |
f5e6429a | 2267 | struct folio *folio; |
0fc9d104 KK |
2268 | unsigned ret = 0; |
2269 | ||
2270 | if (unlikely(!nr_pages)) | |
2271 | return 0; | |
a60637c8 NP |
2272 | |
2273 | rcu_read_lock(); | |
f5e6429a | 2274 | while ((folio = find_get_entry(&xas, end, tag))) { |
a6906972 MW |
2275 | /* |
2276 | * Shadow entries should never be tagged, but this iteration | |
2277 | * is lockless so there is a window for page reclaim to evict | |
2278 | * a page we saw tagged. Skip over it. | |
2279 | */ | |
f5e6429a | 2280 | if (xa_is_value(folio)) |
139b6a6f | 2281 | continue; |
a60637c8 | 2282 | |
f5e6429a | 2283 | pages[ret] = &folio->page; |
72b045ae | 2284 | if (++ret == nr_pages) { |
f5e6429a | 2285 | *index = folio->index + folio_nr_pages(folio); |
72b045ae JK |
2286 | goto out; |
2287 | } | |
a60637c8 | 2288 | } |
5b280c0c | 2289 | |
72b045ae | 2290 | /* |
a6906972 | 2291 | * We come here when we got to @end. We take care to not overflow the |
72b045ae | 2292 | * index @index as it confuses some of the callers. This breaks the |
a6906972 MW |
2293 | * iteration when there is a page at index -1 but that is already |
2294 | * broken anyway. | |
72b045ae JK |
2295 | */ |
2296 | if (end == (pgoff_t)-1) | |
2297 | *index = (pgoff_t)-1; | |
2298 | else | |
2299 | *index = end + 1; | |
2300 | out: | |
a60637c8 | 2301 | rcu_read_unlock(); |
1da177e4 | 2302 | |
1da177e4 LT |
2303 | return ret; |
2304 | } | |
72b045ae | 2305 | EXPORT_SYMBOL(find_get_pages_range_tag); |
1da177e4 | 2306 | |
76d42bd9 WF |
2307 | /* |
2308 | * CD/DVDs are error prone. When a medium error occurs, the driver may fail | |
2309 | * a _large_ part of the i/o request. Imagine the worst scenario: | |
2310 | * | |
2311 | * ---R__________________________________________B__________ | |
2312 | * ^ reading here ^ bad block(assume 4k) | |
2313 | * | |
2314 | * read(R) => miss => readahead(R...B) => media error => frustrating retries | |
2315 | * => failing the whole request => read(R) => read(R+1) => | |
2316 | * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => | |
2317 | * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => | |
2318 | * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... | |
2319 | * | |
2320 | * It is going insane. Fix it by quickly scaling down the readahead size. | |
2321 | */ | |
0f8e2db4 | 2322 | static void shrink_readahead_size_eio(struct file_ra_state *ra) |
76d42bd9 | 2323 | { |
76d42bd9 | 2324 | ra->ra_pages /= 4; |
76d42bd9 WF |
2325 | } |
2326 | ||
cbd59c48 MWO |
2327 | /* |
2328 | * filemap_get_read_batch - Get a batch of pages for read | |
2329 | * | |
2330 | * Get a batch of pages which represent a contiguous range of bytes | |
2331 | * in the file. No tail pages will be returned. If @index is in the | |
2332 | * middle of a THP, the entire THP will be returned. The last page in | |
2333 | * the batch may have Readahead set or be not Uptodate so that the | |
2334 | * caller can take the appropriate action. | |
2335 | */ | |
2336 | static void filemap_get_read_batch(struct address_space *mapping, | |
2337 | pgoff_t index, pgoff_t max, struct pagevec *pvec) | |
2338 | { | |
2339 | XA_STATE(xas, &mapping->i_pages, index); | |
bdb72932 | 2340 | struct folio *folio; |
cbd59c48 MWO |
2341 | |
2342 | rcu_read_lock(); | |
bdb72932 MWO |
2343 | for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) { |
2344 | if (xas_retry(&xas, folio)) | |
cbd59c48 | 2345 | continue; |
bdb72932 | 2346 | if (xas.xa_index > max || xa_is_value(folio)) |
cbd59c48 | 2347 | break; |
bdb72932 | 2348 | if (!folio_try_get_rcu(folio)) |
cbd59c48 MWO |
2349 | goto retry; |
2350 | ||
bdb72932 | 2351 | if (unlikely(folio != xas_reload(&xas))) |
cbd59c48 MWO |
2352 | goto put_page; |
2353 | ||
bdb72932 | 2354 | if (!pagevec_add(pvec, &folio->page)) |
cbd59c48 | 2355 | break; |
bdb72932 | 2356 | if (!folio_test_uptodate(folio)) |
cbd59c48 | 2357 | break; |
bdb72932 | 2358 | if (folio_test_readahead(folio)) |
cbd59c48 | 2359 | break; |
bdb72932 | 2360 | xas.xa_index = folio->index + folio_nr_pages(folio) - 1; |
cbd59c48 MWO |
2361 | xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK; |
2362 | continue; | |
2363 | put_page: | |
bdb72932 | 2364 | folio_put(folio); |
cbd59c48 MWO |
2365 | retry: |
2366 | xas_reset(&xas); | |
2367 | } | |
2368 | rcu_read_unlock(); | |
2369 | } | |
2370 | ||
9d427b4e MWO |
2371 | static int filemap_read_folio(struct file *file, struct address_space *mapping, |
2372 | struct folio *folio) | |
723ef24b | 2373 | { |
723ef24b KO |
2374 | int error; |
2375 | ||
723ef24b | 2376 | /* |
68430303 MWO |
2377 | * A previous I/O error may have been due to temporary failures, |
2378 | * eg. multipath errors. PG_error will be set again if readpage | |
2379 | * fails. | |
723ef24b | 2380 | */ |
9d427b4e | 2381 | folio_clear_error(folio); |
723ef24b | 2382 | /* Start the actual read. The read will unlock the page. */ |
9d427b4e | 2383 | error = mapping->a_ops->readpage(file, &folio->page); |
68430303 MWO |
2384 | if (error) |
2385 | return error; | |
723ef24b | 2386 | |
9d427b4e | 2387 | error = folio_wait_locked_killable(folio); |
68430303 MWO |
2388 | if (error) |
2389 | return error; | |
9d427b4e | 2390 | if (folio_test_uptodate(folio)) |
aa1ec2f6 | 2391 | return 0; |
aa1ec2f6 MWO |
2392 | shrink_readahead_size_eio(&file->f_ra); |
2393 | return -EIO; | |
723ef24b KO |
2394 | } |
2395 | ||
fce70da3 | 2396 | static bool filemap_range_uptodate(struct address_space *mapping, |
2fa4eeb8 | 2397 | loff_t pos, struct iov_iter *iter, struct folio *folio) |
fce70da3 MWO |
2398 | { |
2399 | int count; | |
2400 | ||
2fa4eeb8 | 2401 | if (folio_test_uptodate(folio)) |
fce70da3 MWO |
2402 | return true; |
2403 | /* pipes can't handle partially uptodate pages */ | |
2404 | if (iov_iter_is_pipe(iter)) | |
2405 | return false; | |
2406 | if (!mapping->a_ops->is_partially_uptodate) | |
2407 | return false; | |
2fa4eeb8 | 2408 | if (mapping->host->i_blkbits >= folio_shift(folio)) |
fce70da3 MWO |
2409 | return false; |
2410 | ||
2411 | count = iter->count; | |
2fa4eeb8 MWO |
2412 | if (folio_pos(folio) > pos) { |
2413 | count -= folio_pos(folio) - pos; | |
fce70da3 MWO |
2414 | pos = 0; |
2415 | } else { | |
2fa4eeb8 | 2416 | pos -= folio_pos(folio); |
fce70da3 MWO |
2417 | } |
2418 | ||
2fa4eeb8 | 2419 | return mapping->a_ops->is_partially_uptodate(&folio->page, pos, count); |
fce70da3 MWO |
2420 | } |
2421 | ||
4612aeef MWO |
2422 | static int filemap_update_page(struct kiocb *iocb, |
2423 | struct address_space *mapping, struct iov_iter *iter, | |
65bca53b | 2424 | struct folio *folio) |
723ef24b | 2425 | { |
723ef24b KO |
2426 | int error; |
2427 | ||
730633f0 JK |
2428 | if (iocb->ki_flags & IOCB_NOWAIT) { |
2429 | if (!filemap_invalidate_trylock_shared(mapping)) | |
2430 | return -EAGAIN; | |
2431 | } else { | |
2432 | filemap_invalidate_lock_shared(mapping); | |
2433 | } | |
2434 | ||
ffdc8dab | 2435 | if (!folio_trylock(folio)) { |
730633f0 | 2436 | error = -EAGAIN; |
87d1d7b6 | 2437 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO)) |
730633f0 | 2438 | goto unlock_mapping; |
87d1d7b6 | 2439 | if (!(iocb->ki_flags & IOCB_WAITQ)) { |
730633f0 | 2440 | filemap_invalidate_unlock_shared(mapping); |
9f2b04a2 MWO |
2441 | /* |
2442 | * This is where we usually end up waiting for a | |
2443 | * previously submitted readahead to finish. | |
2444 | */ | |
2445 | folio_put_wait_locked(folio, TASK_KILLABLE); | |
4612aeef | 2446 | return AOP_TRUNCATED_PAGE; |
bd8a1f36 | 2447 | } |
ffdc8dab | 2448 | error = __folio_lock_async(folio, iocb->ki_waitq); |
87d1d7b6 | 2449 | if (error) |
730633f0 | 2450 | goto unlock_mapping; |
723ef24b | 2451 | } |
723ef24b | 2452 | |
730633f0 | 2453 | error = AOP_TRUNCATED_PAGE; |
ffdc8dab | 2454 | if (!folio->mapping) |
730633f0 | 2455 | goto unlock; |
723ef24b | 2456 | |
fce70da3 | 2457 | error = 0; |
2fa4eeb8 | 2458 | if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, folio)) |
fce70da3 MWO |
2459 | goto unlock; |
2460 | ||
2461 | error = -EAGAIN; | |
2462 | if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ)) | |
2463 | goto unlock; | |
2464 | ||
9d427b4e | 2465 | error = filemap_read_folio(iocb->ki_filp, mapping, folio); |
730633f0 | 2466 | goto unlock_mapping; |
fce70da3 | 2467 | unlock: |
ffdc8dab | 2468 | folio_unlock(folio); |
730633f0 JK |
2469 | unlock_mapping: |
2470 | filemap_invalidate_unlock_shared(mapping); | |
2471 | if (error == AOP_TRUNCATED_PAGE) | |
ffdc8dab | 2472 | folio_put(folio); |
fce70da3 | 2473 | return error; |
723ef24b KO |
2474 | } |
2475 | ||
a5d4ad09 | 2476 | static int filemap_create_folio(struct file *file, |
f253e185 MWO |
2477 | struct address_space *mapping, pgoff_t index, |
2478 | struct pagevec *pvec) | |
723ef24b | 2479 | { |
a5d4ad09 | 2480 | struct folio *folio; |
723ef24b KO |
2481 | int error; |
2482 | ||
a5d4ad09 MWO |
2483 | folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0); |
2484 | if (!folio) | |
f253e185 | 2485 | return -ENOMEM; |
723ef24b | 2486 | |
730633f0 | 2487 | /* |
a5d4ad09 MWO |
2488 | * Protect against truncate / hole punch. Grabbing invalidate_lock |
2489 | * here assures we cannot instantiate and bring uptodate new | |
2490 | * pagecache folios after evicting page cache during truncate | |
2491 | * and before actually freeing blocks. Note that we could | |
2492 | * release invalidate_lock after inserting the folio into | |
2493 | * the page cache as the locked folio would then be enough to | |
2494 | * synchronize with hole punching. But there are code paths | |
2495 | * such as filemap_update_page() filling in partially uptodate | |
2496 | * pages or ->readpages() that need to hold invalidate_lock | |
2497 | * while mapping blocks for IO so let's hold the lock here as | |
2498 | * well to keep locking rules simple. | |
730633f0 JK |
2499 | */ |
2500 | filemap_invalidate_lock_shared(mapping); | |
a5d4ad09 | 2501 | error = filemap_add_folio(mapping, folio, index, |
f253e185 MWO |
2502 | mapping_gfp_constraint(mapping, GFP_KERNEL)); |
2503 | if (error == -EEXIST) | |
2504 | error = AOP_TRUNCATED_PAGE; | |
2505 | if (error) | |
2506 | goto error; | |
2507 | ||
a5d4ad09 | 2508 | error = filemap_read_folio(file, mapping, folio); |
f253e185 MWO |
2509 | if (error) |
2510 | goto error; | |
2511 | ||
730633f0 | 2512 | filemap_invalidate_unlock_shared(mapping); |
a5d4ad09 | 2513 | pagevec_add(pvec, &folio->page); |
f253e185 MWO |
2514 | return 0; |
2515 | error: | |
730633f0 | 2516 | filemap_invalidate_unlock_shared(mapping); |
a5d4ad09 | 2517 | folio_put(folio); |
f253e185 | 2518 | return error; |
723ef24b KO |
2519 | } |
2520 | ||
5963fe03 | 2521 | static int filemap_readahead(struct kiocb *iocb, struct file *file, |
65bca53b | 2522 | struct address_space *mapping, struct folio *folio, |
5963fe03 MWO |
2523 | pgoff_t last_index) |
2524 | { | |
65bca53b MWO |
2525 | DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index); |
2526 | ||
5963fe03 MWO |
2527 | if (iocb->ki_flags & IOCB_NOIO) |
2528 | return -EAGAIN; | |
65bca53b | 2529 | page_cache_async_ra(&ractl, folio, last_index - folio->index); |
5963fe03 MWO |
2530 | return 0; |
2531 | } | |
2532 | ||
3a6bae48 | 2533 | static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter, |
ff993ba1 | 2534 | struct pagevec *pvec) |
06c04442 KO |
2535 | { |
2536 | struct file *filp = iocb->ki_filp; | |
2537 | struct address_space *mapping = filp->f_mapping; | |
2538 | struct file_ra_state *ra = &filp->f_ra; | |
2539 | pgoff_t index = iocb->ki_pos >> PAGE_SHIFT; | |
cbd59c48 | 2540 | pgoff_t last_index; |
65bca53b | 2541 | struct folio *folio; |
cbd59c48 | 2542 | int err = 0; |
06c04442 | 2543 | |
cbd59c48 | 2544 | last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE); |
2642fca6 | 2545 | retry: |
06c04442 KO |
2546 | if (fatal_signal_pending(current)) |
2547 | return -EINTR; | |
2548 | ||
cbd59c48 | 2549 | filemap_get_read_batch(mapping, index, last_index, pvec); |
2642fca6 MWO |
2550 | if (!pagevec_count(pvec)) { |
2551 | if (iocb->ki_flags & IOCB_NOIO) | |
2552 | return -EAGAIN; | |
2553 | page_cache_sync_readahead(mapping, ra, filp, index, | |
2554 | last_index - index); | |
2555 | filemap_get_read_batch(mapping, index, last_index, pvec); | |
2556 | } | |
f253e185 MWO |
2557 | if (!pagevec_count(pvec)) { |
2558 | if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ)) | |
2559 | return -EAGAIN; | |
a5d4ad09 | 2560 | err = filemap_create_folio(filp, mapping, |
f253e185 MWO |
2561 | iocb->ki_pos >> PAGE_SHIFT, pvec); |
2562 | if (err == AOP_TRUNCATED_PAGE) | |
2642fca6 | 2563 | goto retry; |
f253e185 MWO |
2564 | return err; |
2565 | } | |
06c04442 | 2566 | |
65bca53b MWO |
2567 | folio = page_folio(pvec->pages[pagevec_count(pvec) - 1]); |
2568 | if (folio_test_readahead(folio)) { | |
2569 | err = filemap_readahead(iocb, filp, mapping, folio, last_index); | |
2642fca6 MWO |
2570 | if (err) |
2571 | goto err; | |
2572 | } | |
65bca53b | 2573 | if (!folio_test_uptodate(folio)) { |
2642fca6 MWO |
2574 | if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1) |
2575 | iocb->ki_flags |= IOCB_NOWAIT; | |
65bca53b | 2576 | err = filemap_update_page(iocb, mapping, iter, folio); |
2642fca6 MWO |
2577 | if (err) |
2578 | goto err; | |
06c04442 KO |
2579 | } |
2580 | ||
2642fca6 | 2581 | return 0; |
cbd59c48 | 2582 | err: |
2642fca6 | 2583 | if (err < 0) |
65bca53b | 2584 | folio_put(folio); |
2642fca6 | 2585 | if (likely(--pvec->nr)) |
ff993ba1 | 2586 | return 0; |
4612aeef | 2587 | if (err == AOP_TRUNCATED_PAGE) |
2642fca6 MWO |
2588 | goto retry; |
2589 | return err; | |
06c04442 KO |
2590 | } |
2591 | ||
485bb99b | 2592 | /** |
87fa0f3e CH |
2593 | * filemap_read - Read data from the page cache. |
2594 | * @iocb: The iocb to read. | |
2595 | * @iter: Destination for the data. | |
2596 | * @already_read: Number of bytes already read by the caller. | |
485bb99b | 2597 | * |
87fa0f3e CH |
2598 | * Copies data from the page cache. If the data is not currently present, |
2599 | * uses the readahead and readpage address_space operations to fetch it. | |
1da177e4 | 2600 | * |
87fa0f3e CH |
2601 | * Return: Total number of bytes copied, including those already read by |
2602 | * the caller. If an error happens before any bytes are copied, returns | |
2603 | * a negative error number. | |
1da177e4 | 2604 | */ |
87fa0f3e CH |
2605 | ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter, |
2606 | ssize_t already_read) | |
1da177e4 | 2607 | { |
47c27bc4 | 2608 | struct file *filp = iocb->ki_filp; |
06c04442 | 2609 | struct file_ra_state *ra = &filp->f_ra; |
36e78914 | 2610 | struct address_space *mapping = filp->f_mapping; |
1da177e4 | 2611 | struct inode *inode = mapping->host; |
ff993ba1 MWO |
2612 | struct pagevec pvec; |
2613 | int i, error = 0; | |
06c04442 KO |
2614 | bool writably_mapped; |
2615 | loff_t isize, end_offset; | |
1da177e4 | 2616 | |
723ef24b | 2617 | if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes)) |
d05c5f7b | 2618 | return 0; |
3644e2d2 KO |
2619 | if (unlikely(!iov_iter_count(iter))) |
2620 | return 0; | |
2621 | ||
c2a9737f | 2622 | iov_iter_truncate(iter, inode->i_sb->s_maxbytes); |
cbd59c48 | 2623 | pagevec_init(&pvec); |
c2a9737f | 2624 | |
06c04442 | 2625 | do { |
1da177e4 | 2626 | cond_resched(); |
5abf186a | 2627 | |
723ef24b | 2628 | /* |
06c04442 KO |
2629 | * If we've already successfully copied some data, then we |
2630 | * can no longer safely return -EIOCBQUEUED. Hence mark | |
2631 | * an async read NOWAIT at that point. | |
723ef24b | 2632 | */ |
87fa0f3e | 2633 | if ((iocb->ki_flags & IOCB_WAITQ) && already_read) |
723ef24b KO |
2634 | iocb->ki_flags |= IOCB_NOWAIT; |
2635 | ||
8c8387ee DH |
2636 | if (unlikely(iocb->ki_pos >= i_size_read(inode))) |
2637 | break; | |
2638 | ||
ff993ba1 MWO |
2639 | error = filemap_get_pages(iocb, iter, &pvec); |
2640 | if (error < 0) | |
06c04442 | 2641 | break; |
1da177e4 | 2642 | |
06c04442 KO |
2643 | /* |
2644 | * i_size must be checked after we know the pages are Uptodate. | |
2645 | * | |
2646 | * Checking i_size after the check allows us to calculate | |
2647 | * the correct value for "nr", which means the zero-filled | |
2648 | * part of the page is not copied back to userspace (unless | |
2649 | * another truncate extends the file - this is desired though). | |
2650 | */ | |
2651 | isize = i_size_read(inode); | |
2652 | if (unlikely(iocb->ki_pos >= isize)) | |
2653 | goto put_pages; | |
06c04442 KO |
2654 | end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count); |
2655 | ||
06c04442 KO |
2656 | /* |
2657 | * Once we start copying data, we don't want to be touching any | |
2658 | * cachelines that might be contended: | |
2659 | */ | |
2660 | writably_mapped = mapping_writably_mapped(mapping); | |
2661 | ||
2662 | /* | |
2663 | * When a sequential read accesses a page several times, only | |
2664 | * mark it as accessed the first time. | |
2665 | */ | |
2666 | if (iocb->ki_pos >> PAGE_SHIFT != | |
2667 | ra->prev_pos >> PAGE_SHIFT) | |
ff993ba1 | 2668 | mark_page_accessed(pvec.pages[0]); |
06c04442 | 2669 | |
ff993ba1 | 2670 | for (i = 0; i < pagevec_count(&pvec); i++) { |
d996fc7f MWO |
2671 | struct folio *folio = page_folio(pvec.pages[i]); |
2672 | size_t fsize = folio_size(folio); | |
2673 | size_t offset = iocb->ki_pos & (fsize - 1); | |
cbd59c48 | 2674 | size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos, |
d996fc7f | 2675 | fsize - offset); |
cbd59c48 | 2676 | size_t copied; |
06c04442 | 2677 | |
d996fc7f | 2678 | if (end_offset < folio_pos(folio)) |
cbd59c48 MWO |
2679 | break; |
2680 | if (i > 0) | |
d996fc7f | 2681 | folio_mark_accessed(folio); |
06c04442 | 2682 | /* |
d996fc7f MWO |
2683 | * If users can be writing to this folio using arbitrary |
2684 | * virtual addresses, take care of potential aliasing | |
2685 | * before reading the folio on the kernel side. | |
06c04442 | 2686 | */ |
d996fc7f MWO |
2687 | if (writably_mapped) |
2688 | flush_dcache_folio(folio); | |
06c04442 | 2689 | |
d996fc7f | 2690 | copied = copy_folio_to_iter(folio, offset, bytes, iter); |
06c04442 | 2691 | |
87fa0f3e | 2692 | already_read += copied; |
06c04442 KO |
2693 | iocb->ki_pos += copied; |
2694 | ra->prev_pos = iocb->ki_pos; | |
2695 | ||
2696 | if (copied < bytes) { | |
2697 | error = -EFAULT; | |
2698 | break; | |
2699 | } | |
1da177e4 | 2700 | } |
06c04442 | 2701 | put_pages: |
ff993ba1 MWO |
2702 | for (i = 0; i < pagevec_count(&pvec); i++) |
2703 | put_page(pvec.pages[i]); | |
cbd59c48 | 2704 | pagevec_reinit(&pvec); |
06c04442 | 2705 | } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error); |
1da177e4 | 2706 | |
0c6aa263 | 2707 | file_accessed(filp); |
06c04442 | 2708 | |
87fa0f3e | 2709 | return already_read ? already_read : error; |
1da177e4 | 2710 | } |
87fa0f3e | 2711 | EXPORT_SYMBOL_GPL(filemap_read); |
1da177e4 | 2712 | |
485bb99b | 2713 | /** |
6abd2322 | 2714 | * generic_file_read_iter - generic filesystem read routine |
485bb99b | 2715 | * @iocb: kernel I/O control block |
6abd2322 | 2716 | * @iter: destination for the data read |
485bb99b | 2717 | * |
6abd2322 | 2718 | * This is the "read_iter()" routine for all filesystems |
1da177e4 | 2719 | * that can use the page cache directly. |
41da51bc AG |
2720 | * |
2721 | * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall | |
2722 | * be returned when no data can be read without waiting for I/O requests | |
2723 | * to complete; it doesn't prevent readahead. | |
2724 | * | |
2725 | * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O | |
2726 | * requests shall be made for the read or for readahead. When no data | |
2727 | * can be read, -EAGAIN shall be returned. When readahead would be | |
2728 | * triggered, a partial, possibly empty read shall be returned. | |
2729 | * | |
a862f68a MR |
2730 | * Return: |
2731 | * * number of bytes copied, even for partial reads | |
41da51bc | 2732 | * * negative error code (or 0 if IOCB_NOIO) if nothing was read |
1da177e4 LT |
2733 | */ |
2734 | ssize_t | |
ed978a81 | 2735 | generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter) |
1da177e4 | 2736 | { |
e7080a43 | 2737 | size_t count = iov_iter_count(iter); |
47c27bc4 | 2738 | ssize_t retval = 0; |
e7080a43 NS |
2739 | |
2740 | if (!count) | |
826ea860 | 2741 | return 0; /* skip atime */ |
1da177e4 | 2742 | |
2ba48ce5 | 2743 | if (iocb->ki_flags & IOCB_DIRECT) { |
47c27bc4 | 2744 | struct file *file = iocb->ki_filp; |
ed978a81 AV |
2745 | struct address_space *mapping = file->f_mapping; |
2746 | struct inode *inode = mapping->host; | |
1da177e4 | 2747 | |
6be96d3a | 2748 | if (iocb->ki_flags & IOCB_NOWAIT) { |
7a60d6d7 JA |
2749 | if (filemap_range_needs_writeback(mapping, iocb->ki_pos, |
2750 | iocb->ki_pos + count - 1)) | |
6be96d3a GR |
2751 | return -EAGAIN; |
2752 | } else { | |
2753 | retval = filemap_write_and_wait_range(mapping, | |
2754 | iocb->ki_pos, | |
2755 | iocb->ki_pos + count - 1); | |
2756 | if (retval < 0) | |
826ea860 | 2757 | return retval; |
6be96d3a | 2758 | } |
d8d3d94b | 2759 | |
0d5b0cf2 CH |
2760 | file_accessed(file); |
2761 | ||
5ecda137 | 2762 | retval = mapping->a_ops->direct_IO(iocb, iter); |
c3a69024 | 2763 | if (retval >= 0) { |
c64fb5c7 | 2764 | iocb->ki_pos += retval; |
5ecda137 | 2765 | count -= retval; |
9fe55eea | 2766 | } |
ab2125df PB |
2767 | if (retval != -EIOCBQUEUED) |
2768 | iov_iter_revert(iter, count - iov_iter_count(iter)); | |
66f998f6 | 2769 | |
9fe55eea SW |
2770 | /* |
2771 | * Btrfs can have a short DIO read if we encounter | |
2772 | * compressed extents, so if there was an error, or if | |
2773 | * we've already read everything we wanted to, or if | |
2774 | * there was a short read because we hit EOF, go ahead | |
2775 | * and return. Otherwise fallthrough to buffered io for | |
fbbbad4b MW |
2776 | * the rest of the read. Buffered reads will not work for |
2777 | * DAX files, so don't bother trying. | |
9fe55eea | 2778 | */ |
61d0017e JA |
2779 | if (retval < 0 || !count || IS_DAX(inode)) |
2780 | return retval; | |
2781 | if (iocb->ki_pos >= i_size_read(inode)) | |
826ea860 | 2782 | return retval; |
1da177e4 LT |
2783 | } |
2784 | ||
826ea860 | 2785 | return filemap_read(iocb, iter, retval); |
1da177e4 | 2786 | } |
ed978a81 | 2787 | EXPORT_SYMBOL(generic_file_read_iter); |
1da177e4 | 2788 | |
f5e6429a MWO |
2789 | static inline loff_t folio_seek_hole_data(struct xa_state *xas, |
2790 | struct address_space *mapping, struct folio *folio, | |
54fa39ac | 2791 | loff_t start, loff_t end, bool seek_data) |
41139aa4 | 2792 | { |
54fa39ac MWO |
2793 | const struct address_space_operations *ops = mapping->a_ops; |
2794 | size_t offset, bsz = i_blocksize(mapping->host); | |
2795 | ||
f5e6429a | 2796 | if (xa_is_value(folio) || folio_test_uptodate(folio)) |
54fa39ac MWO |
2797 | return seek_data ? start : end; |
2798 | if (!ops->is_partially_uptodate) | |
2799 | return seek_data ? end : start; | |
2800 | ||
2801 | xas_pause(xas); | |
2802 | rcu_read_unlock(); | |
f5e6429a MWO |
2803 | folio_lock(folio); |
2804 | if (unlikely(folio->mapping != mapping)) | |
54fa39ac MWO |
2805 | goto unlock; |
2806 | ||
f5e6429a | 2807 | offset = offset_in_folio(folio, start) & ~(bsz - 1); |
54fa39ac MWO |
2808 | |
2809 | do { | |
f5e6429a MWO |
2810 | if (ops->is_partially_uptodate(&folio->page, offset, bsz) == |
2811 | seek_data) | |
54fa39ac MWO |
2812 | break; |
2813 | start = (start + bsz) & ~(bsz - 1); | |
2814 | offset += bsz; | |
f5e6429a | 2815 | } while (offset < folio_size(folio)); |
54fa39ac | 2816 | unlock: |
f5e6429a | 2817 | folio_unlock(folio); |
54fa39ac MWO |
2818 | rcu_read_lock(); |
2819 | return start; | |
41139aa4 MWO |
2820 | } |
2821 | ||
f5e6429a | 2822 | static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio) |
41139aa4 | 2823 | { |
f5e6429a | 2824 | if (xa_is_value(folio)) |
41139aa4 | 2825 | return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index); |
f5e6429a | 2826 | return folio_size(folio); |
41139aa4 MWO |
2827 | } |
2828 | ||
2829 | /** | |
2830 | * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache. | |
2831 | * @mapping: Address space to search. | |
2832 | * @start: First byte to consider. | |
2833 | * @end: Limit of search (exclusive). | |
2834 | * @whence: Either SEEK_HOLE or SEEK_DATA. | |
2835 | * | |
2836 | * If the page cache knows which blocks contain holes and which blocks | |
2837 | * contain data, your filesystem can use this function to implement | |
2838 | * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are | |
2839 | * entirely memory-based such as tmpfs, and filesystems which support | |
2840 | * unwritten extents. | |
2841 | * | |
f0953a1b | 2842 | * Return: The requested offset on success, or -ENXIO if @whence specifies |
41139aa4 MWO |
2843 | * SEEK_DATA and there is no data after @start. There is an implicit hole |
2844 | * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start | |
2845 | * and @end contain data. | |
2846 | */ | |
2847 | loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start, | |
2848 | loff_t end, int whence) | |
2849 | { | |
2850 | XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT); | |
ed98b015 | 2851 | pgoff_t max = (end - 1) >> PAGE_SHIFT; |
41139aa4 | 2852 | bool seek_data = (whence == SEEK_DATA); |
f5e6429a | 2853 | struct folio *folio; |
41139aa4 MWO |
2854 | |
2855 | if (end <= start) | |
2856 | return -ENXIO; | |
2857 | ||
2858 | rcu_read_lock(); | |
f5e6429a | 2859 | while ((folio = find_get_entry(&xas, max, XA_PRESENT))) { |
ed98b015 | 2860 | loff_t pos = (u64)xas.xa_index << PAGE_SHIFT; |
f5e6429a | 2861 | size_t seek_size; |
41139aa4 MWO |
2862 | |
2863 | if (start < pos) { | |
2864 | if (!seek_data) | |
2865 | goto unlock; | |
2866 | start = pos; | |
2867 | } | |
2868 | ||
f5e6429a MWO |
2869 | seek_size = seek_folio_size(&xas, folio); |
2870 | pos = round_up((u64)pos + 1, seek_size); | |
2871 | start = folio_seek_hole_data(&xas, mapping, folio, start, pos, | |
54fa39ac MWO |
2872 | seek_data); |
2873 | if (start < pos) | |
41139aa4 | 2874 | goto unlock; |
ed98b015 HD |
2875 | if (start >= end) |
2876 | break; | |
2877 | if (seek_size > PAGE_SIZE) | |
2878 | xas_set(&xas, pos >> PAGE_SHIFT); | |
f5e6429a MWO |
2879 | if (!xa_is_value(folio)) |
2880 | folio_put(folio); | |
41139aa4 | 2881 | } |
41139aa4 | 2882 | if (seek_data) |
ed98b015 | 2883 | start = -ENXIO; |
41139aa4 MWO |
2884 | unlock: |
2885 | rcu_read_unlock(); | |
f5e6429a MWO |
2886 | if (folio && !xa_is_value(folio)) |
2887 | folio_put(folio); | |
41139aa4 MWO |
2888 | if (start > end) |
2889 | return end; | |
2890 | return start; | |
2891 | } | |
2892 | ||
1da177e4 | 2893 | #ifdef CONFIG_MMU |
1da177e4 | 2894 | #define MMAP_LOTSAMISS (100) |
6b4c9f44 | 2895 | /* |
e292e6d6 | 2896 | * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock |
6b4c9f44 | 2897 | * @vmf - the vm_fault for this fault. |
e292e6d6 | 2898 | * @folio - the folio to lock. |
6b4c9f44 JB |
2899 | * @fpin - the pointer to the file we may pin (or is already pinned). |
2900 | * | |
e292e6d6 MWO |
2901 | * This works similar to lock_folio_or_retry in that it can drop the |
2902 | * mmap_lock. It differs in that it actually returns the folio locked | |
2903 | * if it returns 1 and 0 if it couldn't lock the folio. If we did have | |
2904 | * to drop the mmap_lock then fpin will point to the pinned file and | |
2905 | * needs to be fput()'ed at a later point. | |
6b4c9f44 | 2906 | */ |
e292e6d6 | 2907 | static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio, |
6b4c9f44 JB |
2908 | struct file **fpin) |
2909 | { | |
7c23c782 | 2910 | if (folio_trylock(folio)) |
6b4c9f44 JB |
2911 | return 1; |
2912 | ||
8b0f9fa2 LT |
2913 | /* |
2914 | * NOTE! This will make us return with VM_FAULT_RETRY, but with | |
c1e8d7c6 | 2915 | * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT |
8b0f9fa2 LT |
2916 | * is supposed to work. We have way too many special cases.. |
2917 | */ | |
6b4c9f44 JB |
2918 | if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT) |
2919 | return 0; | |
2920 | ||
2921 | *fpin = maybe_unlock_mmap_for_io(vmf, *fpin); | |
2922 | if (vmf->flags & FAULT_FLAG_KILLABLE) { | |
af7f29d9 | 2923 | if (__folio_lock_killable(folio)) { |
6b4c9f44 | 2924 | /* |
c1e8d7c6 | 2925 | * We didn't have the right flags to drop the mmap_lock, |
6b4c9f44 JB |
2926 | * but all fault_handlers only check for fatal signals |
2927 | * if we return VM_FAULT_RETRY, so we need to drop the | |
c1e8d7c6 | 2928 | * mmap_lock here and return 0 if we don't have a fpin. |
6b4c9f44 JB |
2929 | */ |
2930 | if (*fpin == NULL) | |
d8ed45c5 | 2931 | mmap_read_unlock(vmf->vma->vm_mm); |
6b4c9f44 JB |
2932 | return 0; |
2933 | } | |
2934 | } else | |
7c23c782 MWO |
2935 | __folio_lock(folio); |
2936 | ||
6b4c9f44 JB |
2937 | return 1; |
2938 | } | |
2939 | ||
ef00e08e | 2940 | /* |
6b4c9f44 JB |
2941 | * Synchronous readahead happens when we don't even find a page in the page |
2942 | * cache at all. We don't want to perform IO under the mmap sem, so if we have | |
2943 | * to drop the mmap sem we return the file that was pinned in order for us to do | |
2944 | * that. If we didn't pin a file then we return NULL. The file that is | |
2945 | * returned needs to be fput()'ed when we're done with it. | |
ef00e08e | 2946 | */ |
6b4c9f44 | 2947 | static struct file *do_sync_mmap_readahead(struct vm_fault *vmf) |
ef00e08e | 2948 | { |
2a1180f1 JB |
2949 | struct file *file = vmf->vma->vm_file; |
2950 | struct file_ra_state *ra = &file->f_ra; | |
ef00e08e | 2951 | struct address_space *mapping = file->f_mapping; |
fcd9ae4f | 2952 | DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff); |
6b4c9f44 | 2953 | struct file *fpin = NULL; |
e630bfac | 2954 | unsigned int mmap_miss; |
ef00e08e LT |
2955 | |
2956 | /* If we don't want any read-ahead, don't bother */ | |
2a1180f1 | 2957 | if (vmf->vma->vm_flags & VM_RAND_READ) |
6b4c9f44 | 2958 | return fpin; |
275b12bf | 2959 | if (!ra->ra_pages) |
6b4c9f44 | 2960 | return fpin; |
ef00e08e | 2961 | |
2a1180f1 | 2962 | if (vmf->vma->vm_flags & VM_SEQ_READ) { |
6b4c9f44 | 2963 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
fcd9ae4f | 2964 | page_cache_sync_ra(&ractl, ra->ra_pages); |
6b4c9f44 | 2965 | return fpin; |
ef00e08e LT |
2966 | } |
2967 | ||
207d04ba | 2968 | /* Avoid banging the cache line if not needed */ |
e630bfac KS |
2969 | mmap_miss = READ_ONCE(ra->mmap_miss); |
2970 | if (mmap_miss < MMAP_LOTSAMISS * 10) | |
2971 | WRITE_ONCE(ra->mmap_miss, ++mmap_miss); | |
ef00e08e LT |
2972 | |
2973 | /* | |
2974 | * Do we miss much more than hit in this file? If so, | |
2975 | * stop bothering with read-ahead. It will only hurt. | |
2976 | */ | |
e630bfac | 2977 | if (mmap_miss > MMAP_LOTSAMISS) |
6b4c9f44 | 2978 | return fpin; |
ef00e08e | 2979 | |
d30a1100 WF |
2980 | /* |
2981 | * mmap read-around | |
2982 | */ | |
6b4c9f44 | 2983 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
db660d46 | 2984 | ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2); |
600e19af RG |
2985 | ra->size = ra->ra_pages; |
2986 | ra->async_size = ra->ra_pages / 4; | |
db660d46 DH |
2987 | ractl._index = ra->start; |
2988 | do_page_cache_ra(&ractl, ra->size, ra->async_size); | |
6b4c9f44 | 2989 | return fpin; |
ef00e08e LT |
2990 | } |
2991 | ||
2992 | /* | |
2993 | * Asynchronous readahead happens when we find the page and PG_readahead, | |
6b4c9f44 | 2994 | * so we want to possibly extend the readahead further. We return the file that |
c1e8d7c6 | 2995 | * was pinned if we have to drop the mmap_lock in order to do IO. |
ef00e08e | 2996 | */ |
6b4c9f44 | 2997 | static struct file *do_async_mmap_readahead(struct vm_fault *vmf, |
79598ced | 2998 | struct folio *folio) |
ef00e08e | 2999 | { |
2a1180f1 JB |
3000 | struct file *file = vmf->vma->vm_file; |
3001 | struct file_ra_state *ra = &file->f_ra; | |
79598ced | 3002 | DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff); |
6b4c9f44 | 3003 | struct file *fpin = NULL; |
e630bfac | 3004 | unsigned int mmap_miss; |
ef00e08e LT |
3005 | |
3006 | /* If we don't want any read-ahead, don't bother */ | |
5c72feee | 3007 | if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages) |
6b4c9f44 | 3008 | return fpin; |
79598ced | 3009 | |
e630bfac KS |
3010 | mmap_miss = READ_ONCE(ra->mmap_miss); |
3011 | if (mmap_miss) | |
3012 | WRITE_ONCE(ra->mmap_miss, --mmap_miss); | |
79598ced MWO |
3013 | |
3014 | if (folio_test_readahead(folio)) { | |
6b4c9f44 | 3015 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
79598ced | 3016 | page_cache_async_ra(&ractl, folio, ra->ra_pages); |
6b4c9f44 JB |
3017 | } |
3018 | return fpin; | |
ef00e08e LT |
3019 | } |
3020 | ||
485bb99b | 3021 | /** |
54cb8821 | 3022 | * filemap_fault - read in file data for page fault handling |
d0217ac0 | 3023 | * @vmf: struct vm_fault containing details of the fault |
485bb99b | 3024 | * |
54cb8821 | 3025 | * filemap_fault() is invoked via the vma operations vector for a |
1da177e4 LT |
3026 | * mapped memory region to read in file data during a page fault. |
3027 | * | |
3028 | * The goto's are kind of ugly, but this streamlines the normal case of having | |
3029 | * it in the page cache, and handles the special cases reasonably without | |
3030 | * having a lot of duplicated code. | |
9a95f3cf | 3031 | * |
c1e8d7c6 | 3032 | * vma->vm_mm->mmap_lock must be held on entry. |
9a95f3cf | 3033 | * |
c1e8d7c6 | 3034 | * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock |
e292e6d6 | 3035 | * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap(). |
9a95f3cf | 3036 | * |
c1e8d7c6 | 3037 | * If our return value does not have VM_FAULT_RETRY set, the mmap_lock |
9a95f3cf PC |
3038 | * has not been released. |
3039 | * | |
3040 | * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set. | |
a862f68a MR |
3041 | * |
3042 | * Return: bitwise-OR of %VM_FAULT_ codes. | |
1da177e4 | 3043 | */ |
2bcd6454 | 3044 | vm_fault_t filemap_fault(struct vm_fault *vmf) |
1da177e4 LT |
3045 | { |
3046 | int error; | |
11bac800 | 3047 | struct file *file = vmf->vma->vm_file; |
6b4c9f44 | 3048 | struct file *fpin = NULL; |
1da177e4 | 3049 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3050 | struct inode *inode = mapping->host; |
e292e6d6 MWO |
3051 | pgoff_t max_idx, index = vmf->pgoff; |
3052 | struct folio *folio; | |
2bcd6454 | 3053 | vm_fault_t ret = 0; |
730633f0 | 3054 | bool mapping_locked = false; |
1da177e4 | 3055 | |
e292e6d6 MWO |
3056 | max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3057 | if (unlikely(index >= max_idx)) | |
5307cc1a | 3058 | return VM_FAULT_SIGBUS; |
1da177e4 | 3059 | |
1da177e4 | 3060 | /* |
49426420 | 3061 | * Do we have something in the page cache already? |
1da177e4 | 3062 | */ |
e292e6d6 MWO |
3063 | folio = filemap_get_folio(mapping, index); |
3064 | if (likely(folio)) { | |
1da177e4 | 3065 | /* |
730633f0 JK |
3066 | * We found the page, so try async readahead before waiting for |
3067 | * the lock. | |
1da177e4 | 3068 | */ |
730633f0 | 3069 | if (!(vmf->flags & FAULT_FLAG_TRIED)) |
79598ced | 3070 | fpin = do_async_mmap_readahead(vmf, folio); |
e292e6d6 | 3071 | if (unlikely(!folio_test_uptodate(folio))) { |
730633f0 JK |
3072 | filemap_invalidate_lock_shared(mapping); |
3073 | mapping_locked = true; | |
3074 | } | |
3075 | } else { | |
ef00e08e | 3076 | /* No page in the page cache at all */ |
ef00e08e | 3077 | count_vm_event(PGMAJFAULT); |
2262185c | 3078 | count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT); |
ef00e08e | 3079 | ret = VM_FAULT_MAJOR; |
6b4c9f44 | 3080 | fpin = do_sync_mmap_readahead(vmf); |
ef00e08e | 3081 | retry_find: |
730633f0 | 3082 | /* |
e292e6d6 | 3083 | * See comment in filemap_create_folio() why we need |
730633f0 JK |
3084 | * invalidate_lock |
3085 | */ | |
3086 | if (!mapping_locked) { | |
3087 | filemap_invalidate_lock_shared(mapping); | |
3088 | mapping_locked = true; | |
3089 | } | |
e292e6d6 | 3090 | folio = __filemap_get_folio(mapping, index, |
a75d4c33 JB |
3091 | FGP_CREAT|FGP_FOR_MMAP, |
3092 | vmf->gfp_mask); | |
e292e6d6 | 3093 | if (!folio) { |
6b4c9f44 JB |
3094 | if (fpin) |
3095 | goto out_retry; | |
730633f0 | 3096 | filemap_invalidate_unlock_shared(mapping); |
e520e932 | 3097 | return VM_FAULT_OOM; |
6b4c9f44 | 3098 | } |
1da177e4 LT |
3099 | } |
3100 | ||
e292e6d6 | 3101 | if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin)) |
6b4c9f44 | 3102 | goto out_retry; |
b522c94d ML |
3103 | |
3104 | /* Did it get truncated? */ | |
e292e6d6 MWO |
3105 | if (unlikely(folio->mapping != mapping)) { |
3106 | folio_unlock(folio); | |
3107 | folio_put(folio); | |
b522c94d ML |
3108 | goto retry_find; |
3109 | } | |
e292e6d6 | 3110 | VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio); |
b522c94d | 3111 | |
1da177e4 | 3112 | /* |
d00806b1 NP |
3113 | * We have a locked page in the page cache, now we need to check |
3114 | * that it's up-to-date. If not, it is going to be due to an error. | |
1da177e4 | 3115 | */ |
e292e6d6 | 3116 | if (unlikely(!folio_test_uptodate(folio))) { |
730633f0 JK |
3117 | /* |
3118 | * The page was in cache and uptodate and now it is not. | |
3119 | * Strange but possible since we didn't hold the page lock all | |
3120 | * the time. Let's drop everything get the invalidate lock and | |
3121 | * try again. | |
3122 | */ | |
3123 | if (!mapping_locked) { | |
e292e6d6 MWO |
3124 | folio_unlock(folio); |
3125 | folio_put(folio); | |
730633f0 JK |
3126 | goto retry_find; |
3127 | } | |
1da177e4 | 3128 | goto page_not_uptodate; |
730633f0 | 3129 | } |
1da177e4 | 3130 | |
6b4c9f44 | 3131 | /* |
c1e8d7c6 | 3132 | * We've made it this far and we had to drop our mmap_lock, now is the |
6b4c9f44 JB |
3133 | * time to return to the upper layer and have it re-find the vma and |
3134 | * redo the fault. | |
3135 | */ | |
3136 | if (fpin) { | |
e292e6d6 | 3137 | folio_unlock(folio); |
6b4c9f44 JB |
3138 | goto out_retry; |
3139 | } | |
730633f0 JK |
3140 | if (mapping_locked) |
3141 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 | 3142 | |
ef00e08e LT |
3143 | /* |
3144 | * Found the page and have a reference on it. | |
3145 | * We must recheck i_size under page lock. | |
3146 | */ | |
e292e6d6 MWO |
3147 | max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE); |
3148 | if (unlikely(index >= max_idx)) { | |
3149 | folio_unlock(folio); | |
3150 | folio_put(folio); | |
5307cc1a | 3151 | return VM_FAULT_SIGBUS; |
d00806b1 NP |
3152 | } |
3153 | ||
e292e6d6 | 3154 | vmf->page = folio_file_page(folio, index); |
83c54070 | 3155 | return ret | VM_FAULT_LOCKED; |
1da177e4 | 3156 | |
1da177e4 | 3157 | page_not_uptodate: |
1da177e4 LT |
3158 | /* |
3159 | * Umm, take care of errors if the page isn't up-to-date. | |
3160 | * Try to re-read it _once_. We do this synchronously, | |
3161 | * because there really aren't any performance issues here | |
3162 | * and we need to check for errors. | |
3163 | */ | |
6b4c9f44 | 3164 | fpin = maybe_unlock_mmap_for_io(vmf, fpin); |
e292e6d6 | 3165 | error = filemap_read_folio(file, mapping, folio); |
6b4c9f44 JB |
3166 | if (fpin) |
3167 | goto out_retry; | |
e292e6d6 | 3168 | folio_put(folio); |
d00806b1 NP |
3169 | |
3170 | if (!error || error == AOP_TRUNCATED_PAGE) | |
994fc28c | 3171 | goto retry_find; |
730633f0 | 3172 | filemap_invalidate_unlock_shared(mapping); |
1da177e4 | 3173 | |
d0217ac0 | 3174 | return VM_FAULT_SIGBUS; |
6b4c9f44 JB |
3175 | |
3176 | out_retry: | |
3177 | /* | |
c1e8d7c6 | 3178 | * We dropped the mmap_lock, we need to return to the fault handler to |
6b4c9f44 JB |
3179 | * re-find the vma and come back and find our hopefully still populated |
3180 | * page. | |
3181 | */ | |
e292e6d6 MWO |
3182 | if (folio) |
3183 | folio_put(folio); | |
730633f0 JK |
3184 | if (mapping_locked) |
3185 | filemap_invalidate_unlock_shared(mapping); | |
6b4c9f44 JB |
3186 | if (fpin) |
3187 | fput(fpin); | |
3188 | return ret | VM_FAULT_RETRY; | |
54cb8821 NP |
3189 | } |
3190 | EXPORT_SYMBOL(filemap_fault); | |
3191 | ||
f9ce0be7 | 3192 | static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page) |
f1820361 | 3193 | { |
f9ce0be7 KS |
3194 | struct mm_struct *mm = vmf->vma->vm_mm; |
3195 | ||
3196 | /* Huge page is mapped? No need to proceed. */ | |
3197 | if (pmd_trans_huge(*vmf->pmd)) { | |
3198 | unlock_page(page); | |
3199 | put_page(page); | |
3200 | return true; | |
3201 | } | |
3202 | ||
3203 | if (pmd_none(*vmf->pmd) && PageTransHuge(page)) { | |
e0f43fa5 YS |
3204 | vm_fault_t ret = do_set_pmd(vmf, page); |
3205 | if (!ret) { | |
3206 | /* The page is mapped successfully, reference consumed. */ | |
3207 | unlock_page(page); | |
3208 | return true; | |
f9ce0be7 | 3209 | } |
f9ce0be7 KS |
3210 | } |
3211 | ||
03c4f204 QZ |
3212 | if (pmd_none(*vmf->pmd)) |
3213 | pmd_install(mm, vmf->pmd, &vmf->prealloc_pte); | |
f9ce0be7 KS |
3214 | |
3215 | /* See comment in handle_pte_fault() */ | |
3216 | if (pmd_devmap_trans_unstable(vmf->pmd)) { | |
3217 | unlock_page(page); | |
3218 | put_page(page); | |
3219 | return true; | |
3220 | } | |
3221 | ||
3222 | return false; | |
3223 | } | |
3224 | ||
820b05e9 | 3225 | static struct folio *next_uptodate_page(struct folio *folio, |
f9ce0be7 KS |
3226 | struct address_space *mapping, |
3227 | struct xa_state *xas, pgoff_t end_pgoff) | |
3228 | { | |
3229 | unsigned long max_idx; | |
3230 | ||
3231 | do { | |
9184a307 | 3232 | if (!folio) |
f9ce0be7 | 3233 | return NULL; |
9184a307 | 3234 | if (xas_retry(xas, folio)) |
f9ce0be7 | 3235 | continue; |
9184a307 | 3236 | if (xa_is_value(folio)) |
f9ce0be7 | 3237 | continue; |
9184a307 | 3238 | if (folio_test_locked(folio)) |
f9ce0be7 | 3239 | continue; |
9184a307 | 3240 | if (!folio_try_get_rcu(folio)) |
f9ce0be7 KS |
3241 | continue; |
3242 | /* Has the page moved or been split? */ | |
9184a307 | 3243 | if (unlikely(folio != xas_reload(xas))) |
f9ce0be7 | 3244 | goto skip; |
9184a307 | 3245 | if (!folio_test_uptodate(folio) || folio_test_readahead(folio)) |
f9ce0be7 | 3246 | goto skip; |
9184a307 | 3247 | if (!folio_trylock(folio)) |
f9ce0be7 | 3248 | goto skip; |
9184a307 | 3249 | if (folio->mapping != mapping) |
f9ce0be7 | 3250 | goto unlock; |
9184a307 | 3251 | if (!folio_test_uptodate(folio)) |
f9ce0be7 KS |
3252 | goto unlock; |
3253 | max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE); | |
3254 | if (xas->xa_index >= max_idx) | |
3255 | goto unlock; | |
820b05e9 | 3256 | return folio; |
f9ce0be7 | 3257 | unlock: |
9184a307 | 3258 | folio_unlock(folio); |
f9ce0be7 | 3259 | skip: |
9184a307 MWO |
3260 | folio_put(folio); |
3261 | } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL); | |
f9ce0be7 KS |
3262 | |
3263 | return NULL; | |
3264 | } | |
3265 | ||
820b05e9 | 3266 | static inline struct folio *first_map_page(struct address_space *mapping, |
f9ce0be7 KS |
3267 | struct xa_state *xas, |
3268 | pgoff_t end_pgoff) | |
3269 | { | |
3270 | return next_uptodate_page(xas_find(xas, end_pgoff), | |
3271 | mapping, xas, end_pgoff); | |
3272 | } | |
3273 | ||
820b05e9 | 3274 | static inline struct folio *next_map_page(struct address_space *mapping, |
f9ce0be7 KS |
3275 | struct xa_state *xas, |
3276 | pgoff_t end_pgoff) | |
3277 | { | |
3278 | return next_uptodate_page(xas_next_entry(xas, end_pgoff), | |
3279 | mapping, xas, end_pgoff); | |
3280 | } | |
3281 | ||
3282 | vm_fault_t filemap_map_pages(struct vm_fault *vmf, | |
3283 | pgoff_t start_pgoff, pgoff_t end_pgoff) | |
3284 | { | |
3285 | struct vm_area_struct *vma = vmf->vma; | |
3286 | struct file *file = vma->vm_file; | |
f1820361 | 3287 | struct address_space *mapping = file->f_mapping; |
bae473a4 | 3288 | pgoff_t last_pgoff = start_pgoff; |
9d3af4b4 | 3289 | unsigned long addr; |
070e807c | 3290 | XA_STATE(xas, &mapping->i_pages, start_pgoff); |
820b05e9 MWO |
3291 | struct folio *folio; |
3292 | struct page *page; | |
e630bfac | 3293 | unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss); |
f9ce0be7 | 3294 | vm_fault_t ret = 0; |
f1820361 KS |
3295 | |
3296 | rcu_read_lock(); | |
820b05e9 MWO |
3297 | folio = first_map_page(mapping, &xas, end_pgoff); |
3298 | if (!folio) | |
f9ce0be7 | 3299 | goto out; |
f1820361 | 3300 | |
820b05e9 | 3301 | if (filemap_map_pmd(vmf, &folio->page)) { |
f9ce0be7 KS |
3302 | ret = VM_FAULT_NOPAGE; |
3303 | goto out; | |
3304 | } | |
f1820361 | 3305 | |
9d3af4b4 WD |
3306 | addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT); |
3307 | vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl); | |
f9ce0be7 | 3308 | do { |
820b05e9 | 3309 | page = folio_file_page(folio, xas.xa_index); |
f9ce0be7 | 3310 | if (PageHWPoison(page)) |
f1820361 KS |
3311 | goto unlock; |
3312 | ||
e630bfac KS |
3313 | if (mmap_miss > 0) |
3314 | mmap_miss--; | |
7267ec00 | 3315 | |
9d3af4b4 | 3316 | addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT; |
f9ce0be7 | 3317 | vmf->pte += xas.xa_index - last_pgoff; |
070e807c | 3318 | last_pgoff = xas.xa_index; |
f9ce0be7 KS |
3319 | |
3320 | if (!pte_none(*vmf->pte)) | |
7267ec00 | 3321 | goto unlock; |
f9ce0be7 | 3322 | |
46bdb427 | 3323 | /* We're about to handle the fault */ |
9d3af4b4 | 3324 | if (vmf->address == addr) |
46bdb427 | 3325 | ret = VM_FAULT_NOPAGE; |
46bdb427 | 3326 | |
9d3af4b4 | 3327 | do_set_pte(vmf, page, addr); |
f9ce0be7 | 3328 | /* no need to invalidate: a not-present page won't be cached */ |
9d3af4b4 | 3329 | update_mmu_cache(vma, addr, vmf->pte); |
820b05e9 | 3330 | folio_unlock(folio); |
f9ce0be7 | 3331 | continue; |
f1820361 | 3332 | unlock: |
820b05e9 MWO |
3333 | folio_unlock(folio); |
3334 | folio_put(folio); | |
3335 | } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL); | |
f9ce0be7 KS |
3336 | pte_unmap_unlock(vmf->pte, vmf->ptl); |
3337 | out: | |
f1820361 | 3338 | rcu_read_unlock(); |
e630bfac | 3339 | WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss); |
f9ce0be7 | 3340 | return ret; |
f1820361 KS |
3341 | } |
3342 | EXPORT_SYMBOL(filemap_map_pages); | |
3343 | ||
2bcd6454 | 3344 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
4fcf1c62 | 3345 | { |
5df1a672 | 3346 | struct address_space *mapping = vmf->vma->vm_file->f_mapping; |
960ea971 | 3347 | struct folio *folio = page_folio(vmf->page); |
2bcd6454 | 3348 | vm_fault_t ret = VM_FAULT_LOCKED; |
4fcf1c62 | 3349 | |
5df1a672 | 3350 | sb_start_pagefault(mapping->host->i_sb); |
11bac800 | 3351 | file_update_time(vmf->vma->vm_file); |
960ea971 MWO |
3352 | folio_lock(folio); |
3353 | if (folio->mapping != mapping) { | |
3354 | folio_unlock(folio); | |
4fcf1c62 JK |
3355 | ret = VM_FAULT_NOPAGE; |
3356 | goto out; | |
3357 | } | |
14da9200 | 3358 | /* |
960ea971 | 3359 | * We mark the folio dirty already here so that when freeze is in |
14da9200 | 3360 | * progress, we are guaranteed that writeback during freezing will |
960ea971 | 3361 | * see the dirty folio and writeprotect it again. |
14da9200 | 3362 | */ |
960ea971 MWO |
3363 | folio_mark_dirty(folio); |
3364 | folio_wait_stable(folio); | |
4fcf1c62 | 3365 | out: |
5df1a672 | 3366 | sb_end_pagefault(mapping->host->i_sb); |
4fcf1c62 JK |
3367 | return ret; |
3368 | } | |
4fcf1c62 | 3369 | |
f0f37e2f | 3370 | const struct vm_operations_struct generic_file_vm_ops = { |
54cb8821 | 3371 | .fault = filemap_fault, |
f1820361 | 3372 | .map_pages = filemap_map_pages, |
4fcf1c62 | 3373 | .page_mkwrite = filemap_page_mkwrite, |
1da177e4 LT |
3374 | }; |
3375 | ||
3376 | /* This is used for a general mmap of a disk file */ | |
3377 | ||
68d68ff6 | 3378 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3379 | { |
3380 | struct address_space *mapping = file->f_mapping; | |
3381 | ||
3382 | if (!mapping->a_ops->readpage) | |
3383 | return -ENOEXEC; | |
3384 | file_accessed(file); | |
3385 | vma->vm_ops = &generic_file_vm_ops; | |
3386 | return 0; | |
3387 | } | |
1da177e4 LT |
3388 | |
3389 | /* | |
3390 | * This is for filesystems which do not implement ->writepage. | |
3391 | */ | |
3392 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) | |
3393 | { | |
3394 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
3395 | return -EINVAL; | |
3396 | return generic_file_mmap(file, vma); | |
3397 | } | |
3398 | #else | |
4b96a37d | 3399 | vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf) |
45397228 | 3400 | { |
4b96a37d | 3401 | return VM_FAULT_SIGBUS; |
45397228 | 3402 | } |
68d68ff6 | 3403 | int generic_file_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3404 | { |
3405 | return -ENOSYS; | |
3406 | } | |
68d68ff6 | 3407 | int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) |
1da177e4 LT |
3408 | { |
3409 | return -ENOSYS; | |
3410 | } | |
3411 | #endif /* CONFIG_MMU */ | |
3412 | ||
45397228 | 3413 | EXPORT_SYMBOL(filemap_page_mkwrite); |
1da177e4 LT |
3414 | EXPORT_SYMBOL(generic_file_mmap); |
3415 | EXPORT_SYMBOL(generic_file_readonly_mmap); | |
3416 | ||
539a3322 MWO |
3417 | static struct folio *do_read_cache_folio(struct address_space *mapping, |
3418 | pgoff_t index, filler_t filler, void *data, gfp_t gfp) | |
67f9fd91 | 3419 | { |
539a3322 | 3420 | struct folio *folio; |
1da177e4 LT |
3421 | int err; |
3422 | repeat: | |
539a3322 MWO |
3423 | folio = filemap_get_folio(mapping, index); |
3424 | if (!folio) { | |
3425 | folio = filemap_alloc_folio(gfp, 0); | |
3426 | if (!folio) | |
eb2be189 | 3427 | return ERR_PTR(-ENOMEM); |
539a3322 | 3428 | err = filemap_add_folio(mapping, folio, index, gfp); |
eb2be189 | 3429 | if (unlikely(err)) { |
539a3322 | 3430 | folio_put(folio); |
eb2be189 NP |
3431 | if (err == -EEXIST) |
3432 | goto repeat; | |
22ecdb4f | 3433 | /* Presumably ENOMEM for xarray node */ |
1da177e4 LT |
3434 | return ERR_PTR(err); |
3435 | } | |
32b63529 MG |
3436 | |
3437 | filler: | |
6c45b454 | 3438 | if (filler) |
539a3322 | 3439 | err = filler(data, &folio->page); |
6c45b454 | 3440 | else |
539a3322 | 3441 | err = mapping->a_ops->readpage(data, &folio->page); |
6c45b454 | 3442 | |
1da177e4 | 3443 | if (err < 0) { |
539a3322 | 3444 | folio_put(folio); |
32b63529 | 3445 | return ERR_PTR(err); |
1da177e4 | 3446 | } |
1da177e4 | 3447 | |
539a3322 MWO |
3448 | folio_wait_locked(folio); |
3449 | if (!folio_test_uptodate(folio)) { | |
3450 | folio_put(folio); | |
3451 | return ERR_PTR(-EIO); | |
3452 | } | |
3453 | ||
32b63529 MG |
3454 | goto out; |
3455 | } | |
539a3322 | 3456 | if (folio_test_uptodate(folio)) |
1da177e4 LT |
3457 | goto out; |
3458 | ||
81f4c03b MWO |
3459 | if (!folio_trylock(folio)) { |
3460 | folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE); | |
3461 | goto repeat; | |
3462 | } | |
ebded027 | 3463 | |
81f4c03b | 3464 | /* Folio was truncated from mapping */ |
539a3322 MWO |
3465 | if (!folio->mapping) { |
3466 | folio_unlock(folio); | |
3467 | folio_put(folio); | |
32b63529 | 3468 | goto repeat; |
1da177e4 | 3469 | } |
ebded027 MG |
3470 | |
3471 | /* Someone else locked and filled the page in a very small window */ | |
539a3322 MWO |
3472 | if (folio_test_uptodate(folio)) { |
3473 | folio_unlock(folio); | |
1da177e4 LT |
3474 | goto out; |
3475 | } | |
faffdfa0 XT |
3476 | |
3477 | /* | |
3478 | * A previous I/O error may have been due to temporary | |
3479 | * failures. | |
3480 | * Clear page error before actual read, PG_error will be | |
3481 | * set again if read page fails. | |
3482 | */ | |
539a3322 | 3483 | folio_clear_error(folio); |
32b63529 MG |
3484 | goto filler; |
3485 | ||
c855ff37 | 3486 | out: |
539a3322 MWO |
3487 | folio_mark_accessed(folio); |
3488 | return folio; | |
6fe6900e | 3489 | } |
0531b2aa LT |
3490 | |
3491 | /** | |
539a3322 | 3492 | * read_cache_folio - read into page cache, fill it if needed |
0531b2aa LT |
3493 | * @mapping: the page's address_space |
3494 | * @index: the page index | |
3495 | * @filler: function to perform the read | |
5e5358e7 | 3496 | * @data: first arg to filler(data, page) function, often left as NULL |
0531b2aa | 3497 | * |
0531b2aa | 3498 | * Read into the page cache. If a page already exists, and PageUptodate() is |
67f9fd91 | 3499 | * not set, try to fill the page and wait for it to become unlocked. |
0531b2aa LT |
3500 | * |
3501 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3502 | * |
730633f0 JK |
3503 | * The function expects mapping->invalidate_lock to be already held. |
3504 | * | |
a862f68a | 3505 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa | 3506 | */ |
539a3322 MWO |
3507 | struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index, |
3508 | filler_t filler, void *data) | |
3509 | { | |
3510 | return do_read_cache_folio(mapping, index, filler, data, | |
3511 | mapping_gfp_mask(mapping)); | |
3512 | } | |
3513 | EXPORT_SYMBOL(read_cache_folio); | |
3514 | ||
3515 | static struct page *do_read_cache_page(struct address_space *mapping, | |
3516 | pgoff_t index, filler_t *filler, void *data, gfp_t gfp) | |
3517 | { | |
3518 | struct folio *folio; | |
3519 | ||
3520 | folio = do_read_cache_folio(mapping, index, filler, data, gfp); | |
3521 | if (IS_ERR(folio)) | |
3522 | return &folio->page; | |
3523 | return folio_file_page(folio, index); | |
3524 | } | |
3525 | ||
67f9fd91 | 3526 | struct page *read_cache_page(struct address_space *mapping, |
539a3322 | 3527 | pgoff_t index, filler_t *filler, void *data) |
0531b2aa | 3528 | { |
d322a8e5 CH |
3529 | return do_read_cache_page(mapping, index, filler, data, |
3530 | mapping_gfp_mask(mapping)); | |
0531b2aa | 3531 | } |
67f9fd91 | 3532 | EXPORT_SYMBOL(read_cache_page); |
0531b2aa LT |
3533 | |
3534 | /** | |
3535 | * read_cache_page_gfp - read into page cache, using specified page allocation flags. | |
3536 | * @mapping: the page's address_space | |
3537 | * @index: the page index | |
3538 | * @gfp: the page allocator flags to use if allocating | |
3539 | * | |
3540 | * This is the same as "read_mapping_page(mapping, index, NULL)", but with | |
e6f67b8c | 3541 | * any new page allocations done using the specified allocation flags. |
0531b2aa LT |
3542 | * |
3543 | * If the page does not get brought uptodate, return -EIO. | |
a862f68a | 3544 | * |
730633f0 JK |
3545 | * The function expects mapping->invalidate_lock to be already held. |
3546 | * | |
a862f68a | 3547 | * Return: up to date page on success, ERR_PTR() on failure. |
0531b2aa LT |
3548 | */ |
3549 | struct page *read_cache_page_gfp(struct address_space *mapping, | |
3550 | pgoff_t index, | |
3551 | gfp_t gfp) | |
3552 | { | |
6c45b454 | 3553 | return do_read_cache_page(mapping, index, NULL, NULL, gfp); |
0531b2aa LT |
3554 | } |
3555 | EXPORT_SYMBOL(read_cache_page_gfp); | |
3556 | ||
afddba49 NP |
3557 | int pagecache_write_begin(struct file *file, struct address_space *mapping, |
3558 | loff_t pos, unsigned len, unsigned flags, | |
3559 | struct page **pagep, void **fsdata) | |
3560 | { | |
3561 | const struct address_space_operations *aops = mapping->a_ops; | |
3562 | ||
4e02ed4b | 3563 | return aops->write_begin(file, mapping, pos, len, flags, |
afddba49 | 3564 | pagep, fsdata); |
afddba49 NP |
3565 | } |
3566 | EXPORT_SYMBOL(pagecache_write_begin); | |
3567 | ||
3568 | int pagecache_write_end(struct file *file, struct address_space *mapping, | |
3569 | loff_t pos, unsigned len, unsigned copied, | |
3570 | struct page *page, void *fsdata) | |
3571 | { | |
3572 | const struct address_space_operations *aops = mapping->a_ops; | |
afddba49 | 3573 | |
4e02ed4b | 3574 | return aops->write_end(file, mapping, pos, len, copied, page, fsdata); |
afddba49 NP |
3575 | } |
3576 | EXPORT_SYMBOL(pagecache_write_end); | |
3577 | ||
a92853b6 KK |
3578 | /* |
3579 | * Warn about a page cache invalidation failure during a direct I/O write. | |
3580 | */ | |
3581 | void dio_warn_stale_pagecache(struct file *filp) | |
3582 | { | |
3583 | static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST); | |
3584 | char pathname[128]; | |
a92853b6 KK |
3585 | char *path; |
3586 | ||
5df1a672 | 3587 | errseq_set(&filp->f_mapping->wb_err, -EIO); |
a92853b6 KK |
3588 | if (__ratelimit(&_rs)) { |
3589 | path = file_path(filp, pathname, sizeof(pathname)); | |
3590 | if (IS_ERR(path)) | |
3591 | path = "(unknown)"; | |
3592 | pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n"); | |
3593 | pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid, | |
3594 | current->comm); | |
3595 | } | |
3596 | } | |
3597 | ||
1da177e4 | 3598 | ssize_t |
1af5bb49 | 3599 | generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3600 | { |
3601 | struct file *file = iocb->ki_filp; | |
3602 | struct address_space *mapping = file->f_mapping; | |
3603 | struct inode *inode = mapping->host; | |
1af5bb49 | 3604 | loff_t pos = iocb->ki_pos; |
1da177e4 | 3605 | ssize_t written; |
a969e903 CH |
3606 | size_t write_len; |
3607 | pgoff_t end; | |
1da177e4 | 3608 | |
0c949334 | 3609 | write_len = iov_iter_count(from); |
09cbfeaf | 3610 | end = (pos + write_len - 1) >> PAGE_SHIFT; |
a969e903 | 3611 | |
6be96d3a GR |
3612 | if (iocb->ki_flags & IOCB_NOWAIT) { |
3613 | /* If there are pages to writeback, return */ | |
5df1a672 | 3614 | if (filemap_range_has_page(file->f_mapping, pos, |
35f12f0f | 3615 | pos + write_len - 1)) |
6be96d3a GR |
3616 | return -EAGAIN; |
3617 | } else { | |
3618 | written = filemap_write_and_wait_range(mapping, pos, | |
3619 | pos + write_len - 1); | |
3620 | if (written) | |
3621 | goto out; | |
3622 | } | |
a969e903 CH |
3623 | |
3624 | /* | |
3625 | * After a write we want buffered reads to be sure to go to disk to get | |
3626 | * the new data. We invalidate clean cached page from the region we're | |
3627 | * about to write. We do this *before* the write so that we can return | |
6ccfa806 | 3628 | * without clobbering -EIOCBQUEUED from ->direct_IO(). |
a969e903 | 3629 | */ |
55635ba7 | 3630 | written = invalidate_inode_pages2_range(mapping, |
09cbfeaf | 3631 | pos >> PAGE_SHIFT, end); |
55635ba7 AR |
3632 | /* |
3633 | * If a page can not be invalidated, return 0 to fall back | |
3634 | * to buffered write. | |
3635 | */ | |
3636 | if (written) { | |
3637 | if (written == -EBUSY) | |
3638 | return 0; | |
3639 | goto out; | |
a969e903 CH |
3640 | } |
3641 | ||
639a93a5 | 3642 | written = mapping->a_ops->direct_IO(iocb, from); |
a969e903 CH |
3643 | |
3644 | /* | |
3645 | * Finally, try again to invalidate clean pages which might have been | |
3646 | * cached by non-direct readahead, or faulted in by get_user_pages() | |
3647 | * if the source of the write was an mmap'ed region of the file | |
3648 | * we're writing. Either one is a pretty crazy thing to do, | |
3649 | * so we don't support it 100%. If this invalidation | |
3650 | * fails, tough, the write still worked... | |
332391a9 LC |
3651 | * |
3652 | * Most of the time we do not need this since dio_complete() will do | |
3653 | * the invalidation for us. However there are some file systems that | |
3654 | * do not end up with dio_complete() being called, so let's not break | |
80c1fe90 KK |
3655 | * them by removing it completely. |
3656 | * | |
9266a140 KK |
3657 | * Noticeable example is a blkdev_direct_IO(). |
3658 | * | |
80c1fe90 | 3659 | * Skip invalidation for async writes or if mapping has no pages. |
a969e903 | 3660 | */ |
9266a140 KK |
3661 | if (written > 0 && mapping->nrpages && |
3662 | invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end)) | |
3663 | dio_warn_stale_pagecache(file); | |
a969e903 | 3664 | |
1da177e4 | 3665 | if (written > 0) { |
0116651c | 3666 | pos += written; |
639a93a5 | 3667 | write_len -= written; |
0116651c NK |
3668 | if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { |
3669 | i_size_write(inode, pos); | |
1da177e4 LT |
3670 | mark_inode_dirty(inode); |
3671 | } | |
5cb6c6c7 | 3672 | iocb->ki_pos = pos; |
1da177e4 | 3673 | } |
ab2125df PB |
3674 | if (written != -EIOCBQUEUED) |
3675 | iov_iter_revert(from, write_len - iov_iter_count(from)); | |
a969e903 | 3676 | out: |
1da177e4 LT |
3677 | return written; |
3678 | } | |
3679 | EXPORT_SYMBOL(generic_file_direct_write); | |
3680 | ||
3b93f911 | 3681 | ssize_t generic_perform_write(struct file *file, |
afddba49 NP |
3682 | struct iov_iter *i, loff_t pos) |
3683 | { | |
3684 | struct address_space *mapping = file->f_mapping; | |
3685 | const struct address_space_operations *a_ops = mapping->a_ops; | |
3686 | long status = 0; | |
3687 | ssize_t written = 0; | |
674b892e NP |
3688 | unsigned int flags = 0; |
3689 | ||
afddba49 NP |
3690 | do { |
3691 | struct page *page; | |
afddba49 NP |
3692 | unsigned long offset; /* Offset into pagecache page */ |
3693 | unsigned long bytes; /* Bytes to write to page */ | |
3694 | size_t copied; /* Bytes copied from user */ | |
3695 | void *fsdata; | |
3696 | ||
09cbfeaf KS |
3697 | offset = (pos & (PAGE_SIZE - 1)); |
3698 | bytes = min_t(unsigned long, PAGE_SIZE - offset, | |
afddba49 NP |
3699 | iov_iter_count(i)); |
3700 | ||
3701 | again: | |
00a3d660 LT |
3702 | /* |
3703 | * Bring in the user page that we will copy from _first_. | |
3704 | * Otherwise there's a nasty deadlock on copying from the | |
3705 | * same page as we're writing to, without it being marked | |
3706 | * up-to-date. | |
00a3d660 | 3707 | */ |
a6294593 | 3708 | if (unlikely(fault_in_iov_iter_readable(i, bytes))) { |
00a3d660 LT |
3709 | status = -EFAULT; |
3710 | break; | |
3711 | } | |
3712 | ||
296291cd JK |
3713 | if (fatal_signal_pending(current)) { |
3714 | status = -EINTR; | |
3715 | break; | |
3716 | } | |
3717 | ||
674b892e | 3718 | status = a_ops->write_begin(file, mapping, pos, bytes, flags, |
afddba49 | 3719 | &page, &fsdata); |
2457aec6 | 3720 | if (unlikely(status < 0)) |
afddba49 NP |
3721 | break; |
3722 | ||
931e80e4 | 3723 | if (mapping_writably_mapped(mapping)) |
3724 | flush_dcache_page(page); | |
00a3d660 | 3725 | |
f0b65f39 | 3726 | copied = copy_page_from_iter_atomic(page, offset, bytes, i); |
afddba49 NP |
3727 | flush_dcache_page(page); |
3728 | ||
3729 | status = a_ops->write_end(file, mapping, pos, bytes, copied, | |
3730 | page, fsdata); | |
f0b65f39 AV |
3731 | if (unlikely(status != copied)) { |
3732 | iov_iter_revert(i, copied - max(status, 0L)); | |
3733 | if (unlikely(status < 0)) | |
3734 | break; | |
3735 | } | |
afddba49 NP |
3736 | cond_resched(); |
3737 | ||
bc1bb416 | 3738 | if (unlikely(status == 0)) { |
afddba49 | 3739 | /* |
bc1bb416 AV |
3740 | * A short copy made ->write_end() reject the |
3741 | * thing entirely. Might be memory poisoning | |
3742 | * halfway through, might be a race with munmap, | |
3743 | * might be severe memory pressure. | |
afddba49 | 3744 | */ |
bc1bb416 AV |
3745 | if (copied) |
3746 | bytes = copied; | |
afddba49 NP |
3747 | goto again; |
3748 | } | |
f0b65f39 AV |
3749 | pos += status; |
3750 | written += status; | |
afddba49 NP |
3751 | |
3752 | balance_dirty_pages_ratelimited(mapping); | |
afddba49 NP |
3753 | } while (iov_iter_count(i)); |
3754 | ||
3755 | return written ? written : status; | |
3756 | } | |
3b93f911 | 3757 | EXPORT_SYMBOL(generic_perform_write); |
1da177e4 | 3758 | |
e4dd9de3 | 3759 | /** |
8174202b | 3760 | * __generic_file_write_iter - write data to a file |
e4dd9de3 | 3761 | * @iocb: IO state structure (file, offset, etc.) |
8174202b | 3762 | * @from: iov_iter with data to write |
e4dd9de3 JK |
3763 | * |
3764 | * This function does all the work needed for actually writing data to a | |
3765 | * file. It does all basic checks, removes SUID from the file, updates | |
3766 | * modification times and calls proper subroutines depending on whether we | |
3767 | * do direct IO or a standard buffered write. | |
3768 | * | |
9608703e | 3769 | * It expects i_rwsem to be grabbed unless we work on a block device or similar |
e4dd9de3 JK |
3770 | * object which does not need locking at all. |
3771 | * | |
3772 | * This function does *not* take care of syncing data in case of O_SYNC write. | |
3773 | * A caller has to handle it. This is mainly due to the fact that we want to | |
9608703e | 3774 | * avoid syncing under i_rwsem. |
a862f68a MR |
3775 | * |
3776 | * Return: | |
3777 | * * number of bytes written, even for truncated writes | |
3778 | * * negative error code if no data has been written at all | |
e4dd9de3 | 3779 | */ |
8174202b | 3780 | ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3781 | { |
3782 | struct file *file = iocb->ki_filp; | |
68d68ff6 | 3783 | struct address_space *mapping = file->f_mapping; |
1da177e4 | 3784 | struct inode *inode = mapping->host; |
3b93f911 | 3785 | ssize_t written = 0; |
1da177e4 | 3786 | ssize_t err; |
3b93f911 | 3787 | ssize_t status; |
1da177e4 | 3788 | |
1da177e4 | 3789 | /* We can write back this queue in page reclaim */ |
de1414a6 | 3790 | current->backing_dev_info = inode_to_bdi(inode); |
5fa8e0a1 | 3791 | err = file_remove_privs(file); |
1da177e4 LT |
3792 | if (err) |
3793 | goto out; | |
3794 | ||
c3b2da31 JB |
3795 | err = file_update_time(file); |
3796 | if (err) | |
3797 | goto out; | |
1da177e4 | 3798 | |
2ba48ce5 | 3799 | if (iocb->ki_flags & IOCB_DIRECT) { |
0b8def9d | 3800 | loff_t pos, endbyte; |
fb5527e6 | 3801 | |
1af5bb49 | 3802 | written = generic_file_direct_write(iocb, from); |
1da177e4 | 3803 | /* |
fbbbad4b MW |
3804 | * If the write stopped short of completing, fall back to |
3805 | * buffered writes. Some filesystems do this for writes to | |
3806 | * holes, for example. For DAX files, a buffered write will | |
3807 | * not succeed (even if it did, DAX does not handle dirty | |
3808 | * page-cache pages correctly). | |
1da177e4 | 3809 | */ |
0b8def9d | 3810 | if (written < 0 || !iov_iter_count(from) || IS_DAX(inode)) |
fbbbad4b MW |
3811 | goto out; |
3812 | ||
0b8def9d | 3813 | status = generic_perform_write(file, from, pos = iocb->ki_pos); |
fb5527e6 | 3814 | /* |
3b93f911 | 3815 | * If generic_perform_write() returned a synchronous error |
fb5527e6 JM |
3816 | * then we want to return the number of bytes which were |
3817 | * direct-written, or the error code if that was zero. Note | |
3818 | * that this differs from normal direct-io semantics, which | |
3819 | * will return -EFOO even if some bytes were written. | |
3820 | */ | |
60bb4529 | 3821 | if (unlikely(status < 0)) { |
3b93f911 | 3822 | err = status; |
fb5527e6 JM |
3823 | goto out; |
3824 | } | |
fb5527e6 JM |
3825 | /* |
3826 | * We need to ensure that the page cache pages are written to | |
3827 | * disk and invalidated to preserve the expected O_DIRECT | |
3828 | * semantics. | |
3829 | */ | |
3b93f911 | 3830 | endbyte = pos + status - 1; |
0b8def9d | 3831 | err = filemap_write_and_wait_range(mapping, pos, endbyte); |
fb5527e6 | 3832 | if (err == 0) { |
0b8def9d | 3833 | iocb->ki_pos = endbyte + 1; |
3b93f911 | 3834 | written += status; |
fb5527e6 | 3835 | invalidate_mapping_pages(mapping, |
09cbfeaf KS |
3836 | pos >> PAGE_SHIFT, |
3837 | endbyte >> PAGE_SHIFT); | |
fb5527e6 JM |
3838 | } else { |
3839 | /* | |
3840 | * We don't know how much we wrote, so just return | |
3841 | * the number of bytes which were direct-written | |
3842 | */ | |
3843 | } | |
3844 | } else { | |
0b8def9d AV |
3845 | written = generic_perform_write(file, from, iocb->ki_pos); |
3846 | if (likely(written > 0)) | |
3847 | iocb->ki_pos += written; | |
fb5527e6 | 3848 | } |
1da177e4 LT |
3849 | out: |
3850 | current->backing_dev_info = NULL; | |
3851 | return written ? written : err; | |
3852 | } | |
8174202b | 3853 | EXPORT_SYMBOL(__generic_file_write_iter); |
e4dd9de3 | 3854 | |
e4dd9de3 | 3855 | /** |
8174202b | 3856 | * generic_file_write_iter - write data to a file |
e4dd9de3 | 3857 | * @iocb: IO state structure |
8174202b | 3858 | * @from: iov_iter with data to write |
e4dd9de3 | 3859 | * |
8174202b | 3860 | * This is a wrapper around __generic_file_write_iter() to be used by most |
e4dd9de3 | 3861 | * filesystems. It takes care of syncing the file in case of O_SYNC file |
9608703e | 3862 | * and acquires i_rwsem as needed. |
a862f68a MR |
3863 | * Return: |
3864 | * * negative error code if no data has been written at all of | |
3865 | * vfs_fsync_range() failed for a synchronous write | |
3866 | * * number of bytes written, even for truncated writes | |
e4dd9de3 | 3867 | */ |
8174202b | 3868 | ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 LT |
3869 | { |
3870 | struct file *file = iocb->ki_filp; | |
148f948b | 3871 | struct inode *inode = file->f_mapping->host; |
1da177e4 | 3872 | ssize_t ret; |
1da177e4 | 3873 | |
5955102c | 3874 | inode_lock(inode); |
3309dd04 AV |
3875 | ret = generic_write_checks(iocb, from); |
3876 | if (ret > 0) | |
5f380c7f | 3877 | ret = __generic_file_write_iter(iocb, from); |
5955102c | 3878 | inode_unlock(inode); |
1da177e4 | 3879 | |
e2592217 CH |
3880 | if (ret > 0) |
3881 | ret = generic_write_sync(iocb, ret); | |
1da177e4 LT |
3882 | return ret; |
3883 | } | |
8174202b | 3884 | EXPORT_SYMBOL(generic_file_write_iter); |
1da177e4 | 3885 | |
cf9a2ae8 | 3886 | /** |
82c50f8b MWO |
3887 | * filemap_release_folio() - Release fs-specific metadata on a folio. |
3888 | * @folio: The folio which the kernel is trying to free. | |
3889 | * @gfp: Memory allocation flags (and I/O mode). | |
cf9a2ae8 | 3890 | * |
82c50f8b MWO |
3891 | * The address_space is trying to release any data attached to a folio |
3892 | * (presumably at folio->private). | |
cf9a2ae8 | 3893 | * |
82c50f8b MWO |
3894 | * This will also be called if the private_2 flag is set on a page, |
3895 | * indicating that the folio has other metadata associated with it. | |
cf9a2ae8 | 3896 | * |
82c50f8b MWO |
3897 | * The @gfp argument specifies whether I/O may be performed to release |
3898 | * this page (__GFP_IO), and whether the call may block | |
3899 | * (__GFP_RECLAIM & __GFP_FS). | |
266cf658 | 3900 | * |
82c50f8b | 3901 | * Return: %true if the release was successful, otherwise %false. |
cf9a2ae8 | 3902 | */ |
82c50f8b | 3903 | bool filemap_release_folio(struct folio *folio, gfp_t gfp) |
cf9a2ae8 | 3904 | { |
82c50f8b | 3905 | struct address_space * const mapping = folio->mapping; |
cf9a2ae8 | 3906 | |
82c50f8b MWO |
3907 | BUG_ON(!folio_test_locked(folio)); |
3908 | if (folio_test_writeback(folio)) | |
3909 | return false; | |
cf9a2ae8 DH |
3910 | |
3911 | if (mapping && mapping->a_ops->releasepage) | |
82c50f8b MWO |
3912 | return mapping->a_ops->releasepage(&folio->page, gfp); |
3913 | return try_to_free_buffers(&folio->page); | |
cf9a2ae8 | 3914 | } |
82c50f8b | 3915 | EXPORT_SYMBOL(filemap_release_folio); |