Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
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 : */
13 : #include <linux/export.h>
14 : #include <linux/compiler.h>
15 : #include <linux/dax.h>
16 : #include <linux/fs.h>
17 : #include <linux/sched/signal.h>
18 : #include <linux/uaccess.h>
19 : #include <linux/capability.h>
20 : #include <linux/kernel_stat.h>
21 : #include <linux/gfp.h>
22 : #include <linux/mm.h>
23 : #include <linux/swap.h>
24 : #include <linux/swapops.h>
25 : #include <linux/mman.h>
26 : #include <linux/pagemap.h>
27 : #include <linux/file.h>
28 : #include <linux/uio.h>
29 : #include <linux/error-injection.h>
30 : #include <linux/hash.h>
31 : #include <linux/writeback.h>
32 : #include <linux/backing-dev.h>
33 : #include <linux/pagevec.h>
34 : #include <linux/security.h>
35 : #include <linux/cpuset.h>
36 : #include <linux/hugetlb.h>
37 : #include <linux/memcontrol.h>
38 : #include <linux/shmem_fs.h>
39 : #include <linux/rmap.h>
40 : #include <linux/delayacct.h>
41 : #include <linux/psi.h>
42 : #include <linux/ramfs.h>
43 : #include <linux/page_idle.h>
44 : #include <linux/migrate.h>
45 : #include <linux/pipe_fs_i.h>
46 : #include <linux/splice.h>
47 : #include <asm/pgalloc.h>
48 : #include <asm/tlbflush.h>
49 : #include "internal.h"
50 :
51 : #define CREATE_TRACE_POINTS
52 : #include <trace/events/filemap.h>
53 :
54 : /*
55 : * FIXME: remove all knowledge of the buffer layer from the core VM
56 : */
57 : #include <linux/buffer_head.h> /* for try_to_free_buffers */
58 :
59 : #include <asm/mman.h>
60 :
61 : /*
62 : * Shared mappings implemented 30.11.1994. It's not fully working yet,
63 : * though.
64 : *
65 : * Shared mappings now work. 15.8.1995 Bruno.
66 : *
67 : * finished 'unifying' the page and buffer cache and SMP-threaded the
68 : * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
69 : *
70 : * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
71 : */
72 :
73 : /*
74 : * Lock ordering:
75 : *
76 : * ->i_mmap_rwsem (truncate_pagecache)
77 : * ->private_lock (__free_pte->block_dirty_folio)
78 : * ->swap_lock (exclusive_swap_page, others)
79 : * ->i_pages lock
80 : *
81 : * ->i_rwsem
82 : * ->invalidate_lock (acquired by fs in truncate path)
83 : * ->i_mmap_rwsem (truncate->unmap_mapping_range)
84 : *
85 : * ->mmap_lock
86 : * ->i_mmap_rwsem
87 : * ->page_table_lock or pte_lock (various, mainly in memory.c)
88 : * ->i_pages lock (arch-dependent flush_dcache_mmap_lock)
89 : *
90 : * ->mmap_lock
91 : * ->invalidate_lock (filemap_fault)
92 : * ->lock_page (filemap_fault, access_process_vm)
93 : *
94 : * ->i_rwsem (generic_perform_write)
95 : * ->mmap_lock (fault_in_readable->do_page_fault)
96 : *
97 : * bdi->wb.list_lock
98 : * sb_lock (fs/fs-writeback.c)
99 : * ->i_pages lock (__sync_single_inode)
100 : *
101 : * ->i_mmap_rwsem
102 : * ->anon_vma.lock (vma_merge)
103 : *
104 : * ->anon_vma.lock
105 : * ->page_table_lock or pte_lock (anon_vma_prepare and various)
106 : *
107 : * ->page_table_lock or pte_lock
108 : * ->swap_lock (try_to_unmap_one)
109 : * ->private_lock (try_to_unmap_one)
110 : * ->i_pages lock (try_to_unmap_one)
111 : * ->lruvec->lru_lock (follow_page->mark_page_accessed)
112 : * ->lruvec->lru_lock (check_pte_range->isolate_lru_page)
113 : * ->private_lock (page_remove_rmap->set_page_dirty)
114 : * ->i_pages lock (page_remove_rmap->set_page_dirty)
115 : * bdi.wb->list_lock (page_remove_rmap->set_page_dirty)
116 : * ->inode->i_lock (page_remove_rmap->set_page_dirty)
117 : * ->memcg->move_lock (page_remove_rmap->lock_page_memcg)
118 : * bdi.wb->list_lock (zap_pte_range->set_page_dirty)
119 : * ->inode->i_lock (zap_pte_range->set_page_dirty)
120 : * ->private_lock (zap_pte_range->block_dirty_folio)
121 : *
122 : * ->i_mmap_rwsem
123 : * ->tasklist_lock (memory_failure, collect_procs_ao)
124 : */
125 :
126 0 : static void page_cache_delete(struct address_space *mapping,
127 : struct folio *folio, void *shadow)
128 : {
129 0 : XA_STATE(xas, &mapping->i_pages, folio->index);
130 0 : long nr = 1;
131 :
132 0 : mapping_set_update(&xas, mapping);
133 :
134 : /* hugetlb pages are represented by a single entry in the xarray */
135 0 : if (!folio_test_hugetlb(folio)) {
136 0 : xas_set_order(&xas, folio->index, folio_order(folio));
137 0 : nr = folio_nr_pages(folio);
138 : }
139 :
140 : VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
141 :
142 0 : xas_store(&xas, shadow);
143 0 : xas_init_marks(&xas);
144 :
145 0 : folio->mapping = NULL;
146 : /* Leave page->index set: truncation lookup relies upon it */
147 0 : mapping->nrpages -= nr;
148 0 : }
149 :
150 0 : static void filemap_unaccount_folio(struct address_space *mapping,
151 : struct folio *folio)
152 : {
153 : long nr;
154 :
155 : VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
156 0 : if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
157 0 : pr_alert("BUG: Bad page cache in process %s pfn:%05lx\n",
158 : current->comm, folio_pfn(folio));
159 0 : dump_page(&folio->page, "still mapped when deleted");
160 0 : dump_stack();
161 0 : add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
162 :
163 0 : if (mapping_exiting(mapping) && !folio_test_large(folio)) {
164 0 : int mapcount = page_mapcount(&folio->page);
165 :
166 0 : if (folio_ref_count(folio) >= mapcount + 2) {
167 : /*
168 : * All vmas have already been torn down, so it's
169 : * a good bet that actually the page is unmapped
170 : * and we'd rather not leak it: if we're wrong,
171 : * another bad page check should catch it later.
172 : */
173 0 : page_mapcount_reset(&folio->page);
174 : folio_ref_sub(folio, mapcount);
175 : }
176 : }
177 : }
178 :
179 : /* hugetlb folios do not participate in page cache accounting. */
180 0 : if (folio_test_hugetlb(folio))
181 : return;
182 :
183 0 : nr = folio_nr_pages(folio);
184 :
185 0 : __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
186 0 : if (folio_test_swapbacked(folio)) {
187 0 : __lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
188 0 : if (folio_test_pmd_mappable(folio))
189 : __lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
190 : } else if (folio_test_pmd_mappable(folio)) {
191 : __lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
192 : filemap_nr_thps_dec(mapping);
193 : }
194 :
195 : /*
196 : * At this point folio must be either written or cleaned by
197 : * truncate. Dirty folio here signals a bug and loss of
198 : * unwritten data - on ordinary filesystems.
199 : *
200 : * But it's harmless on in-memory filesystems like tmpfs; and can
201 : * occur when a driver which did get_user_pages() sets page dirty
202 : * before putting it, while the inode is being finally evicted.
203 : *
204 : * Below fixes dirty accounting after removing the folio entirely
205 : * but leaves the dirty flag set: it has no effect for truncated
206 : * folio and anyway will be cleared before returning folio to
207 : * buddy allocator.
208 : */
209 0 : if (WARN_ON_ONCE(folio_test_dirty(folio) &&
210 : mapping_can_writeback(mapping)))
211 0 : folio_account_cleaned(folio, inode_to_wb(mapping->host));
212 : }
213 :
214 : /*
215 : * Delete a page from the page cache and free it. Caller has to make
216 : * sure the page is locked and that nobody else uses it - or that usage
217 : * is safe. The caller must hold the i_pages lock.
218 : */
219 0 : void __filemap_remove_folio(struct folio *folio, void *shadow)
220 : {
221 0 : struct address_space *mapping = folio->mapping;
222 :
223 0 : trace_mm_filemap_delete_from_page_cache(folio);
224 0 : filemap_unaccount_folio(mapping, folio);
225 0 : page_cache_delete(mapping, folio, shadow);
226 0 : }
227 :
228 0 : void filemap_free_folio(struct address_space *mapping, struct folio *folio)
229 : {
230 : void (*free_folio)(struct folio *);
231 0 : int refs = 1;
232 :
233 0 : free_folio = mapping->a_ops->free_folio;
234 0 : if (free_folio)
235 0 : free_folio(folio);
236 :
237 0 : if (folio_test_large(folio) && !folio_test_hugetlb(folio))
238 0 : refs = folio_nr_pages(folio);
239 0 : folio_put_refs(folio, refs);
240 0 : }
241 :
242 : /**
243 : * filemap_remove_folio - Remove folio from page cache.
244 : * @folio: The folio.
245 : *
246 : * This must be called only on folios that are locked and have been
247 : * verified to be in the page cache. It will never put the folio into
248 : * the free list because the caller has a reference on the page.
249 : */
250 0 : void filemap_remove_folio(struct folio *folio)
251 : {
252 0 : struct address_space *mapping = folio->mapping;
253 :
254 0 : BUG_ON(!folio_test_locked(folio));
255 0 : spin_lock(&mapping->host->i_lock);
256 0 : xa_lock_irq(&mapping->i_pages);
257 0 : __filemap_remove_folio(folio, NULL);
258 0 : xa_unlock_irq(&mapping->i_pages);
259 0 : if (mapping_shrinkable(mapping))
260 0 : inode_add_lru(mapping->host);
261 0 : spin_unlock(&mapping->host->i_lock);
262 :
263 0 : filemap_free_folio(mapping, folio);
264 0 : }
265 :
266 : /*
267 : * page_cache_delete_batch - delete several folios from page cache
268 : * @mapping: the mapping to which folios belong
269 : * @fbatch: batch of folios to delete
270 : *
271 : * The function walks over mapping->i_pages and removes folios passed in
272 : * @fbatch from the mapping. The function expects @fbatch to be sorted
273 : * by page index and is optimised for it to be dense.
274 : * It tolerates holes in @fbatch (mapping entries at those indices are not
275 : * modified).
276 : *
277 : * The function expects the i_pages lock to be held.
278 : */
279 0 : static void page_cache_delete_batch(struct address_space *mapping,
280 : struct folio_batch *fbatch)
281 : {
282 0 : XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
283 0 : long total_pages = 0;
284 0 : int i = 0;
285 : struct folio *folio;
286 :
287 0 : mapping_set_update(&xas, mapping);
288 0 : xas_for_each(&xas, folio, ULONG_MAX) {
289 0 : if (i >= folio_batch_count(fbatch))
290 : break;
291 :
292 : /* A swap/dax/shadow entry got inserted? Skip it. */
293 0 : if (xa_is_value(folio))
294 0 : continue;
295 : /*
296 : * A page got inserted in our range? Skip it. We have our
297 : * pages locked so they are protected from being removed.
298 : * If we see a page whose index is higher than ours, it
299 : * means our page has been removed, which shouldn't be
300 : * possible because we're holding the PageLock.
301 : */
302 0 : if (folio != fbatch->folios[i]) {
303 : VM_BUG_ON_FOLIO(folio->index >
304 : fbatch->folios[i]->index, folio);
305 0 : continue;
306 : }
307 :
308 0 : WARN_ON_ONCE(!folio_test_locked(folio));
309 :
310 0 : folio->mapping = NULL;
311 : /* Leave folio->index set: truncation lookup relies on it */
312 :
313 0 : i++;
314 0 : xas_store(&xas, NULL);
315 0 : total_pages += folio_nr_pages(folio);
316 : }
317 0 : mapping->nrpages -= total_pages;
318 0 : }
319 :
320 0 : void delete_from_page_cache_batch(struct address_space *mapping,
321 : struct folio_batch *fbatch)
322 : {
323 : int i;
324 :
325 0 : if (!folio_batch_count(fbatch))
326 : return;
327 :
328 0 : spin_lock(&mapping->host->i_lock);
329 0 : xa_lock_irq(&mapping->i_pages);
330 0 : for (i = 0; i < folio_batch_count(fbatch); i++) {
331 0 : struct folio *folio = fbatch->folios[i];
332 :
333 0 : trace_mm_filemap_delete_from_page_cache(folio);
334 0 : filemap_unaccount_folio(mapping, folio);
335 : }
336 0 : page_cache_delete_batch(mapping, fbatch);
337 0 : xa_unlock_irq(&mapping->i_pages);
338 0 : if (mapping_shrinkable(mapping))
339 0 : inode_add_lru(mapping->host);
340 0 : spin_unlock(&mapping->host->i_lock);
341 :
342 0 : for (i = 0; i < folio_batch_count(fbatch); i++)
343 0 : filemap_free_folio(mapping, fbatch->folios[i]);
344 : }
345 :
346 0 : int filemap_check_errors(struct address_space *mapping)
347 : {
348 0 : int ret = 0;
349 : /* Check for outstanding write errors */
350 0 : if (test_bit(AS_ENOSPC, &mapping->flags) &&
351 0 : test_and_clear_bit(AS_ENOSPC, &mapping->flags))
352 0 : ret = -ENOSPC;
353 0 : if (test_bit(AS_EIO, &mapping->flags) &&
354 0 : test_and_clear_bit(AS_EIO, &mapping->flags))
355 0 : ret = -EIO;
356 0 : return ret;
357 : }
358 : EXPORT_SYMBOL(filemap_check_errors);
359 :
360 0 : static int filemap_check_and_keep_errors(struct address_space *mapping)
361 : {
362 : /* Check for outstanding write errors */
363 0 : if (test_bit(AS_EIO, &mapping->flags))
364 : return -EIO;
365 0 : if (test_bit(AS_ENOSPC, &mapping->flags))
366 : return -ENOSPC;
367 0 : return 0;
368 : }
369 :
370 : /**
371 : * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
372 : * @mapping: address space structure to write
373 : * @wbc: the writeback_control controlling the writeout
374 : *
375 : * Call writepages on the mapping using the provided wbc to control the
376 : * writeout.
377 : *
378 : * Return: %0 on success, negative error code otherwise.
379 : */
380 0 : int filemap_fdatawrite_wbc(struct address_space *mapping,
381 : struct writeback_control *wbc)
382 : {
383 : int ret;
384 :
385 0 : if (!mapping_can_writeback(mapping) ||
386 0 : !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
387 : return 0;
388 :
389 0 : wbc_attach_fdatawrite_inode(wbc, mapping->host);
390 0 : ret = do_writepages(mapping, wbc);
391 0 : wbc_detach_inode(wbc);
392 0 : return ret;
393 : }
394 : EXPORT_SYMBOL(filemap_fdatawrite_wbc);
395 :
396 : /**
397 : * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
398 : * @mapping: address space structure to write
399 : * @start: offset in bytes where the range starts
400 : * @end: offset in bytes where the range ends (inclusive)
401 : * @sync_mode: enable synchronous operation
402 : *
403 : * Start writeback against all of a mapping's dirty pages that lie
404 : * within the byte offsets <start, end> inclusive.
405 : *
406 : * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
407 : * opposed to a regular memory cleansing writeback. The difference between
408 : * these two operations is that if a dirty page/buffer is encountered, it must
409 : * be waited upon, and not just skipped over.
410 : *
411 : * Return: %0 on success, negative error code otherwise.
412 : */
413 0 : int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
414 : loff_t end, int sync_mode)
415 : {
416 0 : struct writeback_control wbc = {
417 : .sync_mode = sync_mode,
418 : .nr_to_write = LONG_MAX,
419 : .range_start = start,
420 : .range_end = end,
421 : };
422 :
423 0 : return filemap_fdatawrite_wbc(mapping, &wbc);
424 : }
425 :
426 : static inline int __filemap_fdatawrite(struct address_space *mapping,
427 : int sync_mode)
428 : {
429 0 : return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
430 : }
431 :
432 0 : int filemap_fdatawrite(struct address_space *mapping)
433 : {
434 0 : return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
435 : }
436 : EXPORT_SYMBOL(filemap_fdatawrite);
437 :
438 0 : int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
439 : loff_t end)
440 : {
441 0 : return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
442 : }
443 : EXPORT_SYMBOL(filemap_fdatawrite_range);
444 :
445 : /**
446 : * filemap_flush - mostly a non-blocking flush
447 : * @mapping: target address_space
448 : *
449 : * This is a mostly non-blocking flush. Not suitable for data-integrity
450 : * purposes - I/O may not be started against all dirty pages.
451 : *
452 : * Return: %0 on success, negative error code otherwise.
453 : */
454 0 : int filemap_flush(struct address_space *mapping)
455 : {
456 0 : return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
457 : }
458 : EXPORT_SYMBOL(filemap_flush);
459 :
460 : /**
461 : * filemap_range_has_page - check if a page exists in range.
462 : * @mapping: address space within which to check
463 : * @start_byte: offset in bytes where the range starts
464 : * @end_byte: offset in bytes where the range ends (inclusive)
465 : *
466 : * Find at least one page in the range supplied, usually used to check if
467 : * direct writing in this range will trigger a writeback.
468 : *
469 : * Return: %true if at least one page exists in the specified range,
470 : * %false otherwise.
471 : */
472 0 : bool filemap_range_has_page(struct address_space *mapping,
473 : loff_t start_byte, loff_t end_byte)
474 : {
475 : struct folio *folio;
476 0 : XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
477 0 : pgoff_t max = end_byte >> PAGE_SHIFT;
478 :
479 0 : if (end_byte < start_byte)
480 : return false;
481 :
482 : rcu_read_lock();
483 : for (;;) {
484 0 : folio = xas_find(&xas, max);
485 0 : if (xas_retry(&xas, folio))
486 0 : continue;
487 : /* Shadow entries don't count */
488 0 : if (xa_is_value(folio))
489 0 : continue;
490 : /*
491 : * We don't need to try to pin this page; we're about to
492 : * release the RCU lock anyway. It is enough to know that
493 : * there was a page here recently.
494 : */
495 : break;
496 : }
497 : rcu_read_unlock();
498 :
499 0 : return folio != NULL;
500 : }
501 : EXPORT_SYMBOL(filemap_range_has_page);
502 :
503 0 : static void __filemap_fdatawait_range(struct address_space *mapping,
504 : loff_t start_byte, loff_t end_byte)
505 : {
506 0 : pgoff_t index = start_byte >> PAGE_SHIFT;
507 0 : pgoff_t end = end_byte >> PAGE_SHIFT;
508 : struct folio_batch fbatch;
509 : unsigned nr_folios;
510 :
511 0 : folio_batch_init(&fbatch);
512 :
513 0 : while (index <= end) {
514 : unsigned i;
515 :
516 0 : nr_folios = filemap_get_folios_tag(mapping, &index, end,
517 : PAGECACHE_TAG_WRITEBACK, &fbatch);
518 :
519 0 : if (!nr_folios)
520 : break;
521 :
522 0 : for (i = 0; i < nr_folios; i++) {
523 0 : struct folio *folio = fbatch.folios[i];
524 :
525 0 : folio_wait_writeback(folio);
526 0 : folio_clear_error(folio);
527 : }
528 0 : folio_batch_release(&fbatch);
529 0 : cond_resched();
530 : }
531 0 : }
532 :
533 : /**
534 : * filemap_fdatawait_range - wait for writeback to complete
535 : * @mapping: address space structure to wait for
536 : * @start_byte: offset in bytes where the range starts
537 : * @end_byte: offset in bytes where the range ends (inclusive)
538 : *
539 : * Walk the list of under-writeback pages of the given address space
540 : * in the given range and wait for all of them. Check error status of
541 : * the address space and return it.
542 : *
543 : * Since the error status of the address space is cleared by this function,
544 : * callers are responsible for checking the return value and handling and/or
545 : * reporting the error.
546 : *
547 : * Return: error status of the address space.
548 : */
549 0 : int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
550 : loff_t end_byte)
551 : {
552 0 : __filemap_fdatawait_range(mapping, start_byte, end_byte);
553 0 : return filemap_check_errors(mapping);
554 : }
555 : EXPORT_SYMBOL(filemap_fdatawait_range);
556 :
557 : /**
558 : * filemap_fdatawait_range_keep_errors - wait for writeback to complete
559 : * @mapping: address space structure to wait for
560 : * @start_byte: offset in bytes where the range starts
561 : * @end_byte: offset in bytes where the range ends (inclusive)
562 : *
563 : * Walk the list of under-writeback pages of the given address space in the
564 : * given range and wait for all of them. Unlike filemap_fdatawait_range(),
565 : * this function does not clear error status of the address space.
566 : *
567 : * Use this function if callers don't handle errors themselves. Expected
568 : * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
569 : * fsfreeze(8)
570 : */
571 0 : int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
572 : loff_t start_byte, loff_t end_byte)
573 : {
574 0 : __filemap_fdatawait_range(mapping, start_byte, end_byte);
575 0 : return filemap_check_and_keep_errors(mapping);
576 : }
577 : EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
578 :
579 : /**
580 : * file_fdatawait_range - wait for writeback to complete
581 : * @file: file pointing to address space structure to wait for
582 : * @start_byte: offset in bytes where the range starts
583 : * @end_byte: offset in bytes where the range ends (inclusive)
584 : *
585 : * Walk the list of under-writeback pages of the address space that file
586 : * refers to, in the given range and wait for all of them. Check error
587 : * status of the address space vs. the file->f_wb_err cursor and return it.
588 : *
589 : * Since the error status of the file is advanced by this function,
590 : * callers are responsible for checking the return value and handling and/or
591 : * reporting the error.
592 : *
593 : * Return: error status of the address space vs. the file->f_wb_err cursor.
594 : */
595 0 : int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
596 : {
597 0 : struct address_space *mapping = file->f_mapping;
598 :
599 0 : __filemap_fdatawait_range(mapping, start_byte, end_byte);
600 0 : return file_check_and_advance_wb_err(file);
601 : }
602 : EXPORT_SYMBOL(file_fdatawait_range);
603 :
604 : /**
605 : * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
606 : * @mapping: address space structure to wait for
607 : *
608 : * Walk the list of under-writeback pages of the given address space
609 : * and wait for all of them. Unlike filemap_fdatawait(), this function
610 : * does not clear error status of the address space.
611 : *
612 : * Use this function if callers don't handle errors themselves. Expected
613 : * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
614 : * fsfreeze(8)
615 : *
616 : * Return: error status of the address space.
617 : */
618 0 : int filemap_fdatawait_keep_errors(struct address_space *mapping)
619 : {
620 0 : __filemap_fdatawait_range(mapping, 0, LLONG_MAX);
621 0 : return filemap_check_and_keep_errors(mapping);
622 : }
623 : EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
624 :
625 : /* Returns true if writeback might be needed or already in progress. */
626 : static bool mapping_needs_writeback(struct address_space *mapping)
627 : {
628 : return mapping->nrpages;
629 : }
630 :
631 0 : bool filemap_range_has_writeback(struct address_space *mapping,
632 : loff_t start_byte, loff_t end_byte)
633 : {
634 0 : XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
635 0 : pgoff_t max = end_byte >> PAGE_SHIFT;
636 : struct folio *folio;
637 :
638 0 : if (end_byte < start_byte)
639 : return false;
640 :
641 : rcu_read_lock();
642 0 : xas_for_each(&xas, folio, max) {
643 0 : if (xas_retry(&xas, folio))
644 0 : continue;
645 0 : if (xa_is_value(folio))
646 0 : continue;
647 0 : if (folio_test_dirty(folio) || folio_test_locked(folio) ||
648 0 : folio_test_writeback(folio))
649 : break;
650 : }
651 : rcu_read_unlock();
652 0 : return folio != NULL;
653 : }
654 : EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
655 :
656 : /**
657 : * filemap_write_and_wait_range - write out & wait on a file range
658 : * @mapping: the address_space for the pages
659 : * @lstart: offset in bytes where the range starts
660 : * @lend: offset in bytes where the range ends (inclusive)
661 : *
662 : * Write out and wait upon file offsets lstart->lend, inclusive.
663 : *
664 : * Note that @lend is inclusive (describes the last byte to be written) so
665 : * that this function can be used to write to the very end-of-file (end = -1).
666 : *
667 : * Return: error status of the address space.
668 : */
669 0 : int filemap_write_and_wait_range(struct address_space *mapping,
670 : loff_t lstart, loff_t lend)
671 : {
672 0 : int err = 0, err2;
673 :
674 0 : if (lend < lstart)
675 : return 0;
676 :
677 0 : if (mapping_needs_writeback(mapping)) {
678 0 : err = __filemap_fdatawrite_range(mapping, lstart, lend,
679 : WB_SYNC_ALL);
680 : /*
681 : * Even if the above returned error, the pages may be
682 : * written partially (e.g. -ENOSPC), so we wait for it.
683 : * But the -EIO is special case, it may indicate the worst
684 : * thing (e.g. bug) happened, so we avoid waiting for it.
685 : */
686 0 : if (err != -EIO)
687 0 : __filemap_fdatawait_range(mapping, lstart, lend);
688 : }
689 0 : err2 = filemap_check_errors(mapping);
690 0 : if (!err)
691 0 : err = err2;
692 : return err;
693 : }
694 : EXPORT_SYMBOL(filemap_write_and_wait_range);
695 :
696 0 : void __filemap_set_wb_err(struct address_space *mapping, int err)
697 : {
698 0 : errseq_t eseq = errseq_set(&mapping->wb_err, err);
699 :
700 0 : trace_filemap_set_wb_err(mapping, eseq);
701 0 : }
702 : EXPORT_SYMBOL(__filemap_set_wb_err);
703 :
704 : /**
705 : * file_check_and_advance_wb_err - report wb error (if any) that was previously
706 : * and advance wb_err to current one
707 : * @file: struct file on which the error is being reported
708 : *
709 : * When userland calls fsync (or something like nfsd does the equivalent), we
710 : * want to report any writeback errors that occurred since the last fsync (or
711 : * since the file was opened if there haven't been any).
712 : *
713 : * Grab the wb_err from the mapping. If it matches what we have in the file,
714 : * then just quickly return 0. The file is all caught up.
715 : *
716 : * If it doesn't match, then take the mapping value, set the "seen" flag in
717 : * it and try to swap it into place. If it works, or another task beat us
718 : * to it with the new value, then update the f_wb_err and return the error
719 : * portion. The error at this point must be reported via proper channels
720 : * (a'la fsync, or NFS COMMIT operation, etc.).
721 : *
722 : * While we handle mapping->wb_err with atomic operations, the f_wb_err
723 : * value is protected by the f_lock since we must ensure that it reflects
724 : * the latest value swapped in for this file descriptor.
725 : *
726 : * Return: %0 on success, negative error code otherwise.
727 : */
728 0 : int file_check_and_advance_wb_err(struct file *file)
729 : {
730 0 : int err = 0;
731 0 : errseq_t old = READ_ONCE(file->f_wb_err);
732 0 : struct address_space *mapping = file->f_mapping;
733 :
734 : /* Locklessly handle the common case where nothing has changed */
735 0 : if (errseq_check(&mapping->wb_err, old)) {
736 : /* Something changed, must use slow path */
737 0 : spin_lock(&file->f_lock);
738 0 : old = file->f_wb_err;
739 0 : err = errseq_check_and_advance(&mapping->wb_err,
740 : &file->f_wb_err);
741 0 : trace_file_check_and_advance_wb_err(file, old);
742 0 : spin_unlock(&file->f_lock);
743 : }
744 :
745 : /*
746 : * We're mostly using this function as a drop in replacement for
747 : * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
748 : * that the legacy code would have had on these flags.
749 : */
750 0 : clear_bit(AS_EIO, &mapping->flags);
751 0 : clear_bit(AS_ENOSPC, &mapping->flags);
752 0 : return err;
753 : }
754 : EXPORT_SYMBOL(file_check_and_advance_wb_err);
755 :
756 : /**
757 : * file_write_and_wait_range - write out & wait on a file range
758 : * @file: file pointing to address_space with pages
759 : * @lstart: offset in bytes where the range starts
760 : * @lend: offset in bytes where the range ends (inclusive)
761 : *
762 : * Write out and wait upon file offsets lstart->lend, inclusive.
763 : *
764 : * Note that @lend is inclusive (describes the last byte to be written) so
765 : * that this function can be used to write to the very end-of-file (end = -1).
766 : *
767 : * After writing out and waiting on the data, we check and advance the
768 : * f_wb_err cursor to the latest value, and return any errors detected there.
769 : *
770 : * Return: %0 on success, negative error code otherwise.
771 : */
772 0 : int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
773 : {
774 0 : int err = 0, err2;
775 0 : struct address_space *mapping = file->f_mapping;
776 :
777 0 : if (lend < lstart)
778 : return 0;
779 :
780 0 : if (mapping_needs_writeback(mapping)) {
781 0 : err = __filemap_fdatawrite_range(mapping, lstart, lend,
782 : WB_SYNC_ALL);
783 : /* See comment of filemap_write_and_wait() */
784 0 : if (err != -EIO)
785 0 : __filemap_fdatawait_range(mapping, lstart, lend);
786 : }
787 0 : err2 = file_check_and_advance_wb_err(file);
788 0 : if (!err)
789 0 : err = err2;
790 : return err;
791 : }
792 : EXPORT_SYMBOL(file_write_and_wait_range);
793 :
794 : /**
795 : * replace_page_cache_folio - replace a pagecache folio with a new one
796 : * @old: folio to be replaced
797 : * @new: folio to replace with
798 : *
799 : * This function replaces a folio in the pagecache with a new one. On
800 : * success it acquires the pagecache reference for the new folio and
801 : * drops it for the old folio. Both the old and new folios must be
802 : * locked. This function does not add the new folio to the LRU, the
803 : * caller must do that.
804 : *
805 : * The remove + add is atomic. This function cannot fail.
806 : */
807 0 : void replace_page_cache_folio(struct folio *old, struct folio *new)
808 : {
809 0 : struct address_space *mapping = old->mapping;
810 0 : void (*free_folio)(struct folio *) = mapping->a_ops->free_folio;
811 0 : pgoff_t offset = old->index;
812 0 : XA_STATE(xas, &mapping->i_pages, offset);
813 :
814 : VM_BUG_ON_FOLIO(!folio_test_locked(old), old);
815 : VM_BUG_ON_FOLIO(!folio_test_locked(new), new);
816 : VM_BUG_ON_FOLIO(new->mapping, new);
817 :
818 0 : folio_get(new);
819 0 : new->mapping = mapping;
820 0 : new->index = offset;
821 :
822 0 : mem_cgroup_migrate(old, new);
823 :
824 0 : xas_lock_irq(&xas);
825 0 : xas_store(&xas, new);
826 :
827 0 : old->mapping = NULL;
828 : /* hugetlb pages do not participate in page cache accounting. */
829 0 : if (!folio_test_hugetlb(old))
830 : __lruvec_stat_sub_folio(old, NR_FILE_PAGES);
831 0 : if (!folio_test_hugetlb(new))
832 : __lruvec_stat_add_folio(new, NR_FILE_PAGES);
833 0 : if (folio_test_swapbacked(old))
834 : __lruvec_stat_sub_folio(old, NR_SHMEM);
835 0 : if (folio_test_swapbacked(new))
836 : __lruvec_stat_add_folio(new, NR_SHMEM);
837 0 : xas_unlock_irq(&xas);
838 0 : if (free_folio)
839 0 : free_folio(old);
840 0 : folio_put(old);
841 0 : }
842 : EXPORT_SYMBOL_GPL(replace_page_cache_folio);
843 :
844 0 : noinline int __filemap_add_folio(struct address_space *mapping,
845 : struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
846 : {
847 0 : XA_STATE(xas, &mapping->i_pages, index);
848 0 : int huge = folio_test_hugetlb(folio);
849 0 : bool charged = false;
850 0 : long nr = 1;
851 :
852 : VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
853 : VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
854 0 : mapping_set_update(&xas, mapping);
855 :
856 : if (!huge) {
857 0 : int error = mem_cgroup_charge(folio, NULL, gfp);
858 : VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
859 : if (error)
860 : return error;
861 0 : charged = true;
862 0 : xas_set_order(&xas, index, folio_order(folio));
863 0 : nr = folio_nr_pages(folio);
864 : }
865 :
866 0 : gfp &= GFP_RECLAIM_MASK;
867 0 : folio_ref_add(folio, nr);
868 0 : folio->mapping = mapping;
869 0 : folio->index = xas.xa_index;
870 :
871 : do {
872 0 : unsigned int order = xa_get_order(xas.xa, xas.xa_index);
873 0 : void *entry, *old = NULL;
874 :
875 0 : if (order > folio_order(folio))
876 : xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
877 : order, gfp);
878 0 : xas_lock_irq(&xas);
879 0 : xas_for_each_conflict(&xas, entry) {
880 0 : old = entry;
881 0 : if (!xa_is_value(entry)) {
882 0 : xas_set_err(&xas, -EEXIST);
883 : goto unlock;
884 : }
885 : }
886 :
887 0 : if (old) {
888 0 : if (shadowp)
889 0 : *shadowp = old;
890 : /* entry may have been split before we acquired lock */
891 0 : order = xa_get_order(xas.xa, xas.xa_index);
892 : if (order > folio_order(folio)) {
893 : /* How to handle large swap entries? */
894 : BUG_ON(shmem_mapping(mapping));
895 : xas_split(&xas, old, order);
896 : xas_reset(&xas);
897 : }
898 : }
899 :
900 0 : xas_store(&xas, folio);
901 0 : if (xas_error(&xas))
902 : goto unlock;
903 :
904 0 : mapping->nrpages += nr;
905 :
906 : /* hugetlb pages do not participate in page cache accounting */
907 : if (!huge) {
908 0 : __lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
909 0 : if (folio_test_pmd_mappable(folio))
910 : __lruvec_stat_mod_folio(folio,
911 : NR_FILE_THPS, nr);
912 : }
913 : unlock:
914 0 : xas_unlock_irq(&xas);
915 0 : } while (xas_nomem(&xas, gfp));
916 :
917 0 : if (xas_error(&xas))
918 : goto error;
919 :
920 : trace_mm_filemap_add_to_page_cache(folio);
921 : return 0;
922 : error:
923 : if (charged)
924 : mem_cgroup_uncharge(folio);
925 0 : folio->mapping = NULL;
926 : /* Leave page->index set: truncation relies upon it */
927 0 : folio_put_refs(folio, nr);
928 0 : return xas_error(&xas);
929 : }
930 : ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
931 :
932 0 : int filemap_add_folio(struct address_space *mapping, struct folio *folio,
933 : pgoff_t index, gfp_t gfp)
934 : {
935 0 : void *shadow = NULL;
936 : int ret;
937 :
938 0 : __folio_set_locked(folio);
939 0 : ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
940 0 : if (unlikely(ret))
941 : __folio_clear_locked(folio);
942 : else {
943 : /*
944 : * The folio might have been evicted from cache only
945 : * recently, in which case it should be activated like
946 : * any other repeatedly accessed folio.
947 : * The exception is folios getting rewritten; evicting other
948 : * data from the working set, only to cache data that will
949 : * get overwritten with something else, is a waste of memory.
950 : */
951 0 : WARN_ON_ONCE(folio_test_active(folio));
952 0 : if (!(gfp & __GFP_WRITE) && shadow)
953 0 : workingset_refault(folio, shadow);
954 0 : folio_add_lru(folio);
955 : }
956 0 : return ret;
957 : }
958 : EXPORT_SYMBOL_GPL(filemap_add_folio);
959 :
960 : #ifdef CONFIG_NUMA
961 : struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
962 : {
963 : int n;
964 : struct folio *folio;
965 :
966 : if (cpuset_do_page_mem_spread()) {
967 : unsigned int cpuset_mems_cookie;
968 : do {
969 : cpuset_mems_cookie = read_mems_allowed_begin();
970 : n = cpuset_mem_spread_node();
971 : folio = __folio_alloc_node(gfp, order, n);
972 : } while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
973 :
974 : return folio;
975 : }
976 : return folio_alloc(gfp, order);
977 : }
978 : EXPORT_SYMBOL(filemap_alloc_folio);
979 : #endif
980 :
981 : /*
982 : * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
983 : *
984 : * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
985 : *
986 : * @mapping1: the first mapping to lock
987 : * @mapping2: the second mapping to lock
988 : */
989 0 : void filemap_invalidate_lock_two(struct address_space *mapping1,
990 : struct address_space *mapping2)
991 : {
992 0 : if (mapping1 > mapping2)
993 0 : swap(mapping1, mapping2);
994 0 : if (mapping1)
995 0 : down_write(&mapping1->invalidate_lock);
996 0 : if (mapping2 && mapping1 != mapping2)
997 0 : down_write_nested(&mapping2->invalidate_lock, 1);
998 0 : }
999 : EXPORT_SYMBOL(filemap_invalidate_lock_two);
1000 :
1001 : /*
1002 : * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1003 : *
1004 : * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1005 : *
1006 : * @mapping1: the first mapping to unlock
1007 : * @mapping2: the second mapping to unlock
1008 : */
1009 0 : void filemap_invalidate_unlock_two(struct address_space *mapping1,
1010 : struct address_space *mapping2)
1011 : {
1012 0 : if (mapping1)
1013 0 : up_write(&mapping1->invalidate_lock);
1014 0 : if (mapping2 && mapping1 != mapping2)
1015 0 : up_write(&mapping2->invalidate_lock);
1016 0 : }
1017 : EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1018 :
1019 : /*
1020 : * In order to wait for pages to become available there must be
1021 : * waitqueues associated with pages. By using a hash table of
1022 : * waitqueues where the bucket discipline is to maintain all
1023 : * waiters on the same queue and wake all when any of the pages
1024 : * become available, and for the woken contexts to check to be
1025 : * sure the appropriate page became available, this saves space
1026 : * at a cost of "thundering herd" phenomena during rare hash
1027 : * collisions.
1028 : */
1029 : #define PAGE_WAIT_TABLE_BITS 8
1030 : #define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1031 : static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1032 :
1033 : static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1034 : {
1035 0 : return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1036 : }
1037 :
1038 1 : void __init pagecache_init(void)
1039 : {
1040 : int i;
1041 :
1042 257 : for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1043 256 : init_waitqueue_head(&folio_wait_table[i]);
1044 :
1045 1 : page_writeback_init();
1046 1 : }
1047 :
1048 : /*
1049 : * The page wait code treats the "wait->flags" somewhat unusually, because
1050 : * we have multiple different kinds of waits, not just the usual "exclusive"
1051 : * one.
1052 : *
1053 : * We have:
1054 : *
1055 : * (a) no special bits set:
1056 : *
1057 : * We're just waiting for the bit to be released, and when a waker
1058 : * calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1059 : * and remove it from the wait queue.
1060 : *
1061 : * Simple and straightforward.
1062 : *
1063 : * (b) WQ_FLAG_EXCLUSIVE:
1064 : *
1065 : * The waiter is waiting to get the lock, and only one waiter should
1066 : * be woken up to avoid any thundering herd behavior. We'll set the
1067 : * WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1068 : *
1069 : * This is the traditional exclusive wait.
1070 : *
1071 : * (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1072 : *
1073 : * The waiter is waiting to get the bit, and additionally wants the
1074 : * lock to be transferred to it for fair lock behavior. If the lock
1075 : * cannot be taken, we stop walking the wait queue without waking
1076 : * the waiter.
1077 : *
1078 : * This is the "fair lock handoff" case, and in addition to setting
1079 : * WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1080 : * that it now has the lock.
1081 : */
1082 0 : static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1083 : {
1084 : unsigned int flags;
1085 0 : struct wait_page_key *key = arg;
1086 0 : struct wait_page_queue *wait_page
1087 0 : = container_of(wait, struct wait_page_queue, wait);
1088 :
1089 0 : if (!wake_page_match(wait_page, key))
1090 : return 0;
1091 :
1092 : /*
1093 : * If it's a lock handoff wait, we get the bit for it, and
1094 : * stop walking (and do not wake it up) if we can't.
1095 : */
1096 0 : flags = wait->flags;
1097 0 : if (flags & WQ_FLAG_EXCLUSIVE) {
1098 0 : if (test_bit(key->bit_nr, &key->folio->flags))
1099 : return -1;
1100 0 : if (flags & WQ_FLAG_CUSTOM) {
1101 0 : if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1102 : return -1;
1103 0 : flags |= WQ_FLAG_DONE;
1104 : }
1105 : }
1106 :
1107 : /*
1108 : * We are holding the wait-queue lock, but the waiter that
1109 : * is waiting for this will be checking the flags without
1110 : * any locking.
1111 : *
1112 : * So update the flags atomically, and wake up the waiter
1113 : * afterwards to avoid any races. This store-release pairs
1114 : * with the load-acquire in folio_wait_bit_common().
1115 : */
1116 0 : smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1117 0 : wake_up_state(wait->private, mode);
1118 :
1119 : /*
1120 : * Ok, we have successfully done what we're waiting for,
1121 : * and we can unconditionally remove the wait entry.
1122 : *
1123 : * Note that this pairs with the "finish_wait()" in the
1124 : * waiter, and has to be the absolute last thing we do.
1125 : * After this list_del_init(&wait->entry) the wait entry
1126 : * might be de-allocated and the process might even have
1127 : * exited.
1128 : */
1129 0 : list_del_init_careful(&wait->entry);
1130 0 : return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1131 : }
1132 :
1133 0 : static void folio_wake_bit(struct folio *folio, int bit_nr)
1134 : {
1135 0 : wait_queue_head_t *q = folio_waitqueue(folio);
1136 : struct wait_page_key key;
1137 : unsigned long flags;
1138 : wait_queue_entry_t bookmark;
1139 :
1140 0 : key.folio = folio;
1141 0 : key.bit_nr = bit_nr;
1142 0 : key.page_match = 0;
1143 :
1144 0 : bookmark.flags = 0;
1145 0 : bookmark.private = NULL;
1146 0 : bookmark.func = NULL;
1147 0 : INIT_LIST_HEAD(&bookmark.entry);
1148 :
1149 0 : spin_lock_irqsave(&q->lock, flags);
1150 0 : __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1151 :
1152 0 : while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1153 : /*
1154 : * Take a breather from holding the lock,
1155 : * allow pages that finish wake up asynchronously
1156 : * to acquire the lock and remove themselves
1157 : * from wait queue
1158 : */
1159 0 : spin_unlock_irqrestore(&q->lock, flags);
1160 : cpu_relax();
1161 0 : spin_lock_irqsave(&q->lock, flags);
1162 0 : __wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1163 : }
1164 :
1165 : /*
1166 : * It's possible to miss clearing waiters here, when we woke our page
1167 : * waiters, but the hashed waitqueue has waiters for other pages on it.
1168 : * That's okay, it's a rare case. The next waker will clear it.
1169 : *
1170 : * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
1171 : * other), the flag may be cleared in the course of freeing the page;
1172 : * but that is not required for correctness.
1173 : */
1174 0 : if (!waitqueue_active(q) || !key.page_match)
1175 : folio_clear_waiters(folio);
1176 :
1177 0 : spin_unlock_irqrestore(&q->lock, flags);
1178 0 : }
1179 :
1180 : static void folio_wake(struct folio *folio, int bit)
1181 : {
1182 0 : if (!folio_test_waiters(folio))
1183 : return;
1184 0 : folio_wake_bit(folio, bit);
1185 : }
1186 :
1187 : /*
1188 : * A choice of three behaviors for folio_wait_bit_common():
1189 : */
1190 : enum behavior {
1191 : EXCLUSIVE, /* Hold ref to page and take the bit when woken, like
1192 : * __folio_lock() waiting on then setting PG_locked.
1193 : */
1194 : SHARED, /* Hold ref to page and check the bit when woken, like
1195 : * folio_wait_writeback() waiting on PG_writeback.
1196 : */
1197 : DROP, /* Drop ref to page before wait, no check when woken,
1198 : * like folio_put_wait_locked() on PG_locked.
1199 : */
1200 : };
1201 :
1202 : /*
1203 : * Attempt to check (or get) the folio flag, and mark us done
1204 : * if successful.
1205 : */
1206 0 : static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1207 : struct wait_queue_entry *wait)
1208 : {
1209 0 : if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1210 0 : if (test_and_set_bit(bit_nr, &folio->flags))
1211 : return false;
1212 0 : } else if (test_bit(bit_nr, &folio->flags))
1213 : return false;
1214 :
1215 0 : wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1216 : return true;
1217 : }
1218 :
1219 : /* How many times do we accept lock stealing from under a waiter? */
1220 : int sysctl_page_lock_unfairness = 5;
1221 :
1222 0 : static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1223 : int state, enum behavior behavior)
1224 : {
1225 0 : wait_queue_head_t *q = folio_waitqueue(folio);
1226 0 : int unfairness = sysctl_page_lock_unfairness;
1227 : struct wait_page_queue wait_page;
1228 0 : wait_queue_entry_t *wait = &wait_page.wait;
1229 0 : bool thrashing = false;
1230 : unsigned long pflags;
1231 : bool in_thrashing;
1232 :
1233 0 : if (bit_nr == PG_locked &&
1234 0 : !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1235 : delayacct_thrashing_start(&in_thrashing);
1236 : psi_memstall_enter(&pflags);
1237 : thrashing = true;
1238 : }
1239 :
1240 0 : init_wait(wait);
1241 0 : wait->func = wake_page_function;
1242 0 : wait_page.folio = folio;
1243 0 : wait_page.bit_nr = bit_nr;
1244 :
1245 : repeat:
1246 0 : wait->flags = 0;
1247 0 : if (behavior == EXCLUSIVE) {
1248 0 : wait->flags = WQ_FLAG_EXCLUSIVE;
1249 0 : if (--unfairness < 0)
1250 0 : wait->flags |= WQ_FLAG_CUSTOM;
1251 : }
1252 :
1253 : /*
1254 : * Do one last check whether we can get the
1255 : * page bit synchronously.
1256 : *
1257 : * Do the folio_set_waiters() marking before that
1258 : * to let any waker we _just_ missed know they
1259 : * need to wake us up (otherwise they'll never
1260 : * even go to the slow case that looks at the
1261 : * page queue), and add ourselves to the wait
1262 : * queue if we need to sleep.
1263 : *
1264 : * This part needs to be done under the queue
1265 : * lock to avoid races.
1266 : */
1267 0 : spin_lock_irq(&q->lock);
1268 0 : folio_set_waiters(folio);
1269 0 : if (!folio_trylock_flag(folio, bit_nr, wait))
1270 : __add_wait_queue_entry_tail(q, wait);
1271 0 : spin_unlock_irq(&q->lock);
1272 :
1273 : /*
1274 : * From now on, all the logic will be based on
1275 : * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1276 : * see whether the page bit testing has already
1277 : * been done by the wake function.
1278 : *
1279 : * We can drop our reference to the folio.
1280 : */
1281 0 : if (behavior == DROP)
1282 : folio_put(folio);
1283 :
1284 : /*
1285 : * Note that until the "finish_wait()", or until
1286 : * we see the WQ_FLAG_WOKEN flag, we need to
1287 : * be very careful with the 'wait->flags', because
1288 : * we may race with a waker that sets them.
1289 : */
1290 0 : for (;;) {
1291 : unsigned int flags;
1292 :
1293 0 : set_current_state(state);
1294 :
1295 : /* Loop until we've been woken or interrupted */
1296 0 : flags = smp_load_acquire(&wait->flags);
1297 0 : if (!(flags & WQ_FLAG_WOKEN)) {
1298 0 : if (signal_pending_state(state, current))
1299 : break;
1300 :
1301 0 : io_schedule();
1302 0 : continue;
1303 : }
1304 :
1305 : /* If we were non-exclusive, we're done */
1306 0 : if (behavior != EXCLUSIVE)
1307 : break;
1308 :
1309 : /* If the waker got the lock for us, we're done */
1310 0 : if (flags & WQ_FLAG_DONE)
1311 : break;
1312 :
1313 : /*
1314 : * Otherwise, if we're getting the lock, we need to
1315 : * try to get it ourselves.
1316 : *
1317 : * And if that fails, we'll have to retry this all.
1318 : */
1319 0 : if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1320 : goto repeat;
1321 :
1322 0 : wait->flags |= WQ_FLAG_DONE;
1323 0 : break;
1324 : }
1325 :
1326 : /*
1327 : * If a signal happened, this 'finish_wait()' may remove the last
1328 : * waiter from the wait-queues, but the folio waiters bit will remain
1329 : * set. That's ok. The next wakeup will take care of it, and trying
1330 : * to do it here would be difficult and prone to races.
1331 : */
1332 0 : finish_wait(q, wait);
1333 :
1334 : if (thrashing) {
1335 : delayacct_thrashing_end(&in_thrashing);
1336 : psi_memstall_leave(&pflags);
1337 : }
1338 :
1339 : /*
1340 : * NOTE! The wait->flags weren't stable until we've done the
1341 : * 'finish_wait()', and we could have exited the loop above due
1342 : * to a signal, and had a wakeup event happen after the signal
1343 : * test but before the 'finish_wait()'.
1344 : *
1345 : * So only after the finish_wait() can we reliably determine
1346 : * if we got woken up or not, so we can now figure out the final
1347 : * return value based on that state without races.
1348 : *
1349 : * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1350 : * waiter, but an exclusive one requires WQ_FLAG_DONE.
1351 : */
1352 0 : if (behavior == EXCLUSIVE)
1353 0 : return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1354 :
1355 0 : return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1356 : }
1357 :
1358 : #ifdef CONFIG_MIGRATION
1359 : /**
1360 : * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1361 : * @entry: migration swap entry.
1362 : * @ptep: mapped pte pointer. Will return with the ptep unmapped. Only required
1363 : * for pte entries, pass NULL for pmd entries.
1364 : * @ptl: already locked ptl. This function will drop the lock.
1365 : *
1366 : * Wait for a migration entry referencing the given page to be removed. This is
1367 : * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1368 : * this can be called without taking a reference on the page. Instead this
1369 : * should be called while holding the ptl for the migration entry referencing
1370 : * the page.
1371 : *
1372 : * Returns after unmapping and unlocking the pte/ptl with pte_unmap_unlock().
1373 : *
1374 : * This follows the same logic as folio_wait_bit_common() so see the comments
1375 : * there.
1376 : */
1377 0 : void migration_entry_wait_on_locked(swp_entry_t entry, pte_t *ptep,
1378 : spinlock_t *ptl)
1379 : {
1380 : struct wait_page_queue wait_page;
1381 0 : wait_queue_entry_t *wait = &wait_page.wait;
1382 0 : bool thrashing = false;
1383 : unsigned long pflags;
1384 : bool in_thrashing;
1385 : wait_queue_head_t *q;
1386 0 : struct folio *folio = page_folio(pfn_swap_entry_to_page(entry));
1387 :
1388 0 : q = folio_waitqueue(folio);
1389 0 : if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1390 : delayacct_thrashing_start(&in_thrashing);
1391 : psi_memstall_enter(&pflags);
1392 : thrashing = true;
1393 : }
1394 :
1395 0 : init_wait(wait);
1396 0 : wait->func = wake_page_function;
1397 0 : wait_page.folio = folio;
1398 0 : wait_page.bit_nr = PG_locked;
1399 : wait->flags = 0;
1400 :
1401 0 : spin_lock_irq(&q->lock);
1402 0 : folio_set_waiters(folio);
1403 0 : if (!folio_trylock_flag(folio, PG_locked, wait))
1404 : __add_wait_queue_entry_tail(q, wait);
1405 0 : spin_unlock_irq(&q->lock);
1406 :
1407 : /*
1408 : * If a migration entry exists for the page the migration path must hold
1409 : * a valid reference to the page, and it must take the ptl to remove the
1410 : * migration entry. So the page is valid until the ptl is dropped.
1411 : */
1412 0 : if (ptep)
1413 : pte_unmap_unlock(ptep, ptl);
1414 : else
1415 : spin_unlock(ptl);
1416 :
1417 0 : for (;;) {
1418 : unsigned int flags;
1419 :
1420 0 : set_current_state(TASK_UNINTERRUPTIBLE);
1421 :
1422 : /* Loop until we've been woken or interrupted */
1423 0 : flags = smp_load_acquire(&wait->flags);
1424 0 : if (!(flags & WQ_FLAG_WOKEN)) {
1425 0 : if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1426 : break;
1427 :
1428 0 : io_schedule();
1429 0 : continue;
1430 : }
1431 : break;
1432 : }
1433 :
1434 0 : finish_wait(q, wait);
1435 :
1436 : if (thrashing) {
1437 : delayacct_thrashing_end(&in_thrashing);
1438 : psi_memstall_leave(&pflags);
1439 : }
1440 0 : }
1441 : #endif
1442 :
1443 0 : void folio_wait_bit(struct folio *folio, int bit_nr)
1444 : {
1445 0 : folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1446 0 : }
1447 : EXPORT_SYMBOL(folio_wait_bit);
1448 :
1449 0 : int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1450 : {
1451 0 : return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1452 : }
1453 : EXPORT_SYMBOL(folio_wait_bit_killable);
1454 :
1455 : /**
1456 : * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1457 : * @folio: The folio to wait for.
1458 : * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1459 : *
1460 : * The caller should hold a reference on @folio. They expect the page to
1461 : * become unlocked relatively soon, but do not wish to hold up migration
1462 : * (for example) by holding the reference while waiting for the folio to
1463 : * come unlocked. After this function returns, the caller should not
1464 : * dereference @folio.
1465 : *
1466 : * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1467 : */
1468 : static int folio_put_wait_locked(struct folio *folio, int state)
1469 : {
1470 0 : return folio_wait_bit_common(folio, PG_locked, state, DROP);
1471 : }
1472 :
1473 : /**
1474 : * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1475 : * @folio: Folio defining the wait queue of interest
1476 : * @waiter: Waiter to add to the queue
1477 : *
1478 : * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1479 : */
1480 0 : void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1481 : {
1482 0 : wait_queue_head_t *q = folio_waitqueue(folio);
1483 : unsigned long flags;
1484 :
1485 0 : spin_lock_irqsave(&q->lock, flags);
1486 0 : __add_wait_queue_entry_tail(q, waiter);
1487 0 : folio_set_waiters(folio);
1488 0 : spin_unlock_irqrestore(&q->lock, flags);
1489 0 : }
1490 : EXPORT_SYMBOL_GPL(folio_add_wait_queue);
1491 :
1492 : #ifndef clear_bit_unlock_is_negative_byte
1493 :
1494 : /*
1495 : * PG_waiters is the high bit in the same byte as PG_lock.
1496 : *
1497 : * On x86 (and on many other architectures), we can clear PG_lock and
1498 : * test the sign bit at the same time. But if the architecture does
1499 : * not support that special operation, we just do this all by hand
1500 : * instead.
1501 : *
1502 : * The read of PG_waiters has to be after (or concurrently with) PG_locked
1503 : * being cleared, but a memory barrier should be unnecessary since it is
1504 : * in the same byte as PG_locked.
1505 : */
1506 : static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1507 : {
1508 : clear_bit_unlock(nr, mem);
1509 : /* smp_mb__after_atomic(); */
1510 : return test_bit(PG_waiters, mem);
1511 : }
1512 :
1513 : #endif
1514 :
1515 : /**
1516 : * folio_unlock - Unlock a locked folio.
1517 : * @folio: The folio.
1518 : *
1519 : * Unlocks the folio and wakes up any thread sleeping on the page lock.
1520 : *
1521 : * Context: May be called from interrupt or process context. May not be
1522 : * called from NMI context.
1523 : */
1524 0 : void folio_unlock(struct folio *folio)
1525 : {
1526 : /* Bit 7 allows x86 to check the byte's sign bit */
1527 : BUILD_BUG_ON(PG_waiters != 7);
1528 : BUILD_BUG_ON(PG_locked > 7);
1529 : VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1530 0 : if (clear_bit_unlock_is_negative_byte(PG_locked, folio_flags(folio, 0)))
1531 0 : folio_wake_bit(folio, PG_locked);
1532 0 : }
1533 : EXPORT_SYMBOL(folio_unlock);
1534 :
1535 : /**
1536 : * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1537 : * @folio: The folio.
1538 : *
1539 : * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1540 : * it. The folio reference held for PG_private_2 being set is released.
1541 : *
1542 : * This is, for example, used when a netfs folio is being written to a local
1543 : * disk cache, thereby allowing writes to the cache for the same folio to be
1544 : * serialised.
1545 : */
1546 0 : void folio_end_private_2(struct folio *folio)
1547 : {
1548 : VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1549 0 : clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1550 0 : folio_wake_bit(folio, PG_private_2);
1551 0 : folio_put(folio);
1552 0 : }
1553 : EXPORT_SYMBOL(folio_end_private_2);
1554 :
1555 : /**
1556 : * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1557 : * @folio: The folio to wait on.
1558 : *
1559 : * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1560 : */
1561 0 : void folio_wait_private_2(struct folio *folio)
1562 : {
1563 0 : while (folio_test_private_2(folio))
1564 : folio_wait_bit(folio, PG_private_2);
1565 0 : }
1566 : EXPORT_SYMBOL(folio_wait_private_2);
1567 :
1568 : /**
1569 : * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1570 : * @folio: The folio to wait on.
1571 : *
1572 : * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1573 : * fatal signal is received by the calling task.
1574 : *
1575 : * Return:
1576 : * - 0 if successful.
1577 : * - -EINTR if a fatal signal was encountered.
1578 : */
1579 0 : int folio_wait_private_2_killable(struct folio *folio)
1580 : {
1581 0 : int ret = 0;
1582 :
1583 0 : while (folio_test_private_2(folio)) {
1584 0 : ret = folio_wait_bit_killable(folio, PG_private_2);
1585 0 : if (ret < 0)
1586 : break;
1587 : }
1588 :
1589 0 : return ret;
1590 : }
1591 : EXPORT_SYMBOL(folio_wait_private_2_killable);
1592 :
1593 : /**
1594 : * folio_end_writeback - End writeback against a folio.
1595 : * @folio: The folio.
1596 : */
1597 0 : void folio_end_writeback(struct folio *folio)
1598 : {
1599 : /*
1600 : * folio_test_clear_reclaim() could be used here but it is an
1601 : * atomic operation and overkill in this particular case. Failing
1602 : * to shuffle a folio marked for immediate reclaim is too mild
1603 : * a gain to justify taking an atomic operation penalty at the
1604 : * end of every folio writeback.
1605 : */
1606 0 : if (folio_test_reclaim(folio)) {
1607 0 : folio_clear_reclaim(folio);
1608 0 : folio_rotate_reclaimable(folio);
1609 : }
1610 :
1611 : /*
1612 : * Writeback does not hold a folio reference of its own, relying
1613 : * on truncation to wait for the clearing of PG_writeback.
1614 : * But here we must make sure that the folio is not freed and
1615 : * reused before the folio_wake().
1616 : */
1617 0 : folio_get(folio);
1618 0 : if (!__folio_end_writeback(folio))
1619 0 : BUG();
1620 :
1621 0 : smp_mb__after_atomic();
1622 0 : folio_wake(folio, PG_writeback);
1623 0 : acct_reclaim_writeback(folio);
1624 0 : folio_put(folio);
1625 0 : }
1626 : EXPORT_SYMBOL(folio_end_writeback);
1627 :
1628 : /*
1629 : * After completing I/O on a page, call this routine to update the page
1630 : * flags appropriately
1631 : */
1632 0 : void page_endio(struct page *page, bool is_write, int err)
1633 : {
1634 0 : struct folio *folio = page_folio(page);
1635 :
1636 0 : if (!is_write) {
1637 0 : if (!err) {
1638 : folio_mark_uptodate(folio);
1639 : } else {
1640 0 : folio_clear_uptodate(folio);
1641 : folio_set_error(folio);
1642 : }
1643 : folio_unlock(folio);
1644 : } else {
1645 0 : if (err) {
1646 : struct address_space *mapping;
1647 :
1648 0 : folio_set_error(folio);
1649 0 : mapping = folio_mapping(folio);
1650 0 : if (mapping)
1651 0 : mapping_set_error(mapping, err);
1652 : }
1653 0 : folio_end_writeback(folio);
1654 : }
1655 0 : }
1656 : EXPORT_SYMBOL_GPL(page_endio);
1657 :
1658 : /**
1659 : * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1660 : * @folio: The folio to lock
1661 : */
1662 0 : void __folio_lock(struct folio *folio)
1663 : {
1664 0 : folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
1665 : EXCLUSIVE);
1666 0 : }
1667 : EXPORT_SYMBOL(__folio_lock);
1668 :
1669 0 : int __folio_lock_killable(struct folio *folio)
1670 : {
1671 0 : return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
1672 : EXCLUSIVE);
1673 : }
1674 : EXPORT_SYMBOL_GPL(__folio_lock_killable);
1675 :
1676 0 : static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1677 : {
1678 0 : struct wait_queue_head *q = folio_waitqueue(folio);
1679 0 : int ret = 0;
1680 :
1681 0 : wait->folio = folio;
1682 0 : wait->bit_nr = PG_locked;
1683 :
1684 0 : spin_lock_irq(&q->lock);
1685 0 : __add_wait_queue_entry_tail(q, &wait->wait);
1686 0 : folio_set_waiters(folio);
1687 0 : ret = !folio_trylock(folio);
1688 : /*
1689 : * If we were successful now, we know we're still on the
1690 : * waitqueue as we're still under the lock. This means it's
1691 : * safe to remove and return success, we know the callback
1692 : * isn't going to trigger.
1693 : */
1694 0 : if (!ret)
1695 0 : __remove_wait_queue(q, &wait->wait);
1696 : else
1697 : ret = -EIOCBQUEUED;
1698 0 : spin_unlock_irq(&q->lock);
1699 0 : return ret;
1700 : }
1701 :
1702 : /*
1703 : * Return values:
1704 : * true - folio is locked; mmap_lock is still held.
1705 : * false - folio is not locked.
1706 : * mmap_lock has been released (mmap_read_unlock(), unless flags had both
1707 : * FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1708 : * which case mmap_lock is still held.
1709 : *
1710 : * If neither ALLOW_RETRY nor KILLABLE are set, will always return true
1711 : * with the folio locked and the mmap_lock unperturbed.
1712 : */
1713 0 : bool __folio_lock_or_retry(struct folio *folio, struct mm_struct *mm,
1714 : unsigned int flags)
1715 : {
1716 0 : if (fault_flag_allow_retry_first(flags)) {
1717 : /*
1718 : * CAUTION! In this case, mmap_lock is not released
1719 : * even though return 0.
1720 : */
1721 0 : if (flags & FAULT_FLAG_RETRY_NOWAIT)
1722 : return false;
1723 :
1724 0 : mmap_read_unlock(mm);
1725 0 : if (flags & FAULT_FLAG_KILLABLE)
1726 0 : folio_wait_locked_killable(folio);
1727 : else
1728 : folio_wait_locked(folio);
1729 : return false;
1730 : }
1731 0 : if (flags & FAULT_FLAG_KILLABLE) {
1732 : bool ret;
1733 :
1734 0 : ret = __folio_lock_killable(folio);
1735 0 : if (ret) {
1736 0 : mmap_read_unlock(mm);
1737 0 : return false;
1738 : }
1739 : } else {
1740 : __folio_lock(folio);
1741 : }
1742 :
1743 : return true;
1744 : }
1745 :
1746 : /**
1747 : * page_cache_next_miss() - Find the next gap in the page cache.
1748 : * @mapping: Mapping.
1749 : * @index: Index.
1750 : * @max_scan: Maximum range to search.
1751 : *
1752 : * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1753 : * gap with the lowest index.
1754 : *
1755 : * This function may be called under the rcu_read_lock. However, this will
1756 : * not atomically search a snapshot of the cache at a single point in time.
1757 : * For example, if a gap is created at index 5, then subsequently a gap is
1758 : * created at index 10, page_cache_next_miss covering both indices may
1759 : * return 10 if called under the rcu_read_lock.
1760 : *
1761 : * Return: The index of the gap if found, otherwise an index outside the
1762 : * range specified (in which case 'return - index >= max_scan' will be true).
1763 : * In the rare case of index wrap-around, 0 will be returned. 0 will also
1764 : * be returned if index == 0 and there is a gap at the index. We can not
1765 : * wrap-around if passed index == 0.
1766 : */
1767 0 : pgoff_t page_cache_next_miss(struct address_space *mapping,
1768 : pgoff_t index, unsigned long max_scan)
1769 : {
1770 0 : XA_STATE(xas, &mapping->i_pages, index);
1771 :
1772 0 : while (max_scan--) {
1773 0 : void *entry = xas_next(&xas);
1774 0 : if (!entry || xa_is_value(entry))
1775 0 : return xas.xa_index;
1776 0 : if (xas.xa_index == 0 && index != 0)
1777 : return xas.xa_index;
1778 : }
1779 :
1780 : /* No gaps in range and no wrap-around, return index beyond range */
1781 0 : return xas.xa_index + 1;
1782 : }
1783 : EXPORT_SYMBOL(page_cache_next_miss);
1784 :
1785 : /**
1786 : * page_cache_prev_miss() - Find the previous gap in the page cache.
1787 : * @mapping: Mapping.
1788 : * @index: Index.
1789 : * @max_scan: Maximum range to search.
1790 : *
1791 : * Search the range [max(index - max_scan + 1, 0), index] for the
1792 : * gap with the highest index.
1793 : *
1794 : * This function may be called under the rcu_read_lock. However, this will
1795 : * not atomically search a snapshot of the cache at a single point in time.
1796 : * For example, if a gap is created at index 10, then subsequently a gap is
1797 : * created at index 5, page_cache_prev_miss() covering both indices may
1798 : * return 5 if called under the rcu_read_lock.
1799 : *
1800 : * Return: The index of the gap if found, otherwise an index outside the
1801 : * range specified (in which case 'index - return >= max_scan' will be true).
1802 : * In the rare case of wrap-around, ULONG_MAX will be returned. ULONG_MAX
1803 : * will also be returned if index == ULONG_MAX and there is a gap at the
1804 : * index. We can not wrap-around if passed index == ULONG_MAX.
1805 : */
1806 0 : pgoff_t page_cache_prev_miss(struct address_space *mapping,
1807 : pgoff_t index, unsigned long max_scan)
1808 : {
1809 0 : XA_STATE(xas, &mapping->i_pages, index);
1810 :
1811 0 : while (max_scan--) {
1812 0 : void *entry = xas_prev(&xas);
1813 0 : if (!entry || xa_is_value(entry))
1814 0 : return xas.xa_index;
1815 0 : if (xas.xa_index == ULONG_MAX && index != ULONG_MAX)
1816 : return xas.xa_index;
1817 : }
1818 :
1819 : /* No gaps in range and no wrap-around, return index beyond range */
1820 0 : return xas.xa_index - 1;
1821 : }
1822 : EXPORT_SYMBOL(page_cache_prev_miss);
1823 :
1824 : /*
1825 : * Lockless page cache protocol:
1826 : * On the lookup side:
1827 : * 1. Load the folio from i_pages
1828 : * 2. Increment the refcount if it's not zero
1829 : * 3. If the folio is not found by xas_reload(), put the refcount and retry
1830 : *
1831 : * On the removal side:
1832 : * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1833 : * B. Remove the page from i_pages
1834 : * C. Return the page to the page allocator
1835 : *
1836 : * This means that any page may have its reference count temporarily
1837 : * increased by a speculative page cache (or fast GUP) lookup as it can
1838 : * be allocated by another user before the RCU grace period expires.
1839 : * Because the refcount temporarily acquired here may end up being the
1840 : * last refcount on the page, any page allocation must be freeable by
1841 : * folio_put().
1842 : */
1843 :
1844 : /*
1845 : * filemap_get_entry - Get a page cache entry.
1846 : * @mapping: the address_space to search
1847 : * @index: The page cache index.
1848 : *
1849 : * Looks up the page cache entry at @mapping & @index. If it is a folio,
1850 : * it is returned with an increased refcount. If it is a shadow entry
1851 : * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1852 : * it is returned without further action.
1853 : *
1854 : * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1855 : */
1856 0 : void *filemap_get_entry(struct address_space *mapping, pgoff_t index)
1857 : {
1858 0 : XA_STATE(xas, &mapping->i_pages, index);
1859 : struct folio *folio;
1860 :
1861 : rcu_read_lock();
1862 : repeat:
1863 0 : xas_reset(&xas);
1864 0 : folio = xas_load(&xas);
1865 0 : if (xas_retry(&xas, folio))
1866 : goto repeat;
1867 : /*
1868 : * A shadow entry of a recently evicted page, or a swap entry from
1869 : * shmem/tmpfs. Return it without attempting to raise page count.
1870 : */
1871 0 : if (!folio || xa_is_value(folio))
1872 : goto out;
1873 :
1874 0 : if (!folio_try_get_rcu(folio))
1875 : goto repeat;
1876 :
1877 0 : if (unlikely(folio != xas_reload(&xas))) {
1878 : folio_put(folio);
1879 : goto repeat;
1880 : }
1881 : out:
1882 : rcu_read_unlock();
1883 :
1884 0 : return folio;
1885 : }
1886 :
1887 : /**
1888 : * __filemap_get_folio - Find and get a reference to a folio.
1889 : * @mapping: The address_space to search.
1890 : * @index: The page index.
1891 : * @fgp_flags: %FGP flags modify how the folio is returned.
1892 : * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1893 : *
1894 : * Looks up the page cache entry at @mapping & @index.
1895 : *
1896 : * @fgp_flags can be zero or more of these flags:
1897 : *
1898 : * * %FGP_ACCESSED - The folio will be marked accessed.
1899 : * * %FGP_LOCK - The folio is returned locked.
1900 : * * %FGP_CREAT - If no page is present then a new page is allocated using
1901 : * @gfp and added to the page cache and the VM's LRU list.
1902 : * The page is returned locked and with an increased refcount.
1903 : * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1904 : * page is already in cache. If the page was allocated, unlock it before
1905 : * returning so the caller can do the same dance.
1906 : * * %FGP_WRITE - The page will be written to by the caller.
1907 : * * %FGP_NOFS - __GFP_FS will get cleared in gfp.
1908 : * * %FGP_NOWAIT - Don't get blocked by page lock.
1909 : * * %FGP_STABLE - Wait for the folio to be stable (finished writeback)
1910 : *
1911 : * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1912 : * if the %GFP flags specified for %FGP_CREAT are atomic.
1913 : *
1914 : * If there is a page cache page, it is returned with an increased refcount.
1915 : *
1916 : * Return: The found folio or an ERR_PTR() otherwise.
1917 : */
1918 0 : struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1919 : int fgp_flags, gfp_t gfp)
1920 : {
1921 : struct folio *folio;
1922 :
1923 : repeat:
1924 0 : folio = filemap_get_entry(mapping, index);
1925 0 : if (xa_is_value(folio))
1926 0 : folio = NULL;
1927 0 : if (!folio)
1928 : goto no_page;
1929 :
1930 0 : if (fgp_flags & FGP_LOCK) {
1931 0 : if (fgp_flags & FGP_NOWAIT) {
1932 0 : if (!folio_trylock(folio)) {
1933 : folio_put(folio);
1934 : return ERR_PTR(-EAGAIN);
1935 : }
1936 : } else {
1937 0 : folio_lock(folio);
1938 : }
1939 :
1940 : /* Has the page been truncated? */
1941 0 : if (unlikely(folio->mapping != mapping)) {
1942 0 : folio_unlock(folio);
1943 : folio_put(folio);
1944 : goto repeat;
1945 : }
1946 : VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1947 : }
1948 :
1949 0 : if (fgp_flags & FGP_ACCESSED)
1950 0 : folio_mark_accessed(folio);
1951 : else if (fgp_flags & FGP_WRITE) {
1952 : /* Clear idle flag for buffer write */
1953 : if (folio_test_idle(folio))
1954 : folio_clear_idle(folio);
1955 : }
1956 :
1957 0 : if (fgp_flags & FGP_STABLE)
1958 0 : folio_wait_stable(folio);
1959 : no_page:
1960 0 : if (!folio && (fgp_flags & FGP_CREAT)) {
1961 : int err;
1962 0 : if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1963 0 : gfp |= __GFP_WRITE;
1964 0 : if (fgp_flags & FGP_NOFS)
1965 0 : gfp &= ~__GFP_FS;
1966 0 : if (fgp_flags & FGP_NOWAIT) {
1967 0 : gfp &= ~GFP_KERNEL;
1968 0 : gfp |= GFP_NOWAIT | __GFP_NOWARN;
1969 : }
1970 :
1971 0 : folio = filemap_alloc_folio(gfp, 0);
1972 0 : if (!folio)
1973 : return ERR_PTR(-ENOMEM);
1974 :
1975 0 : if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1976 0 : fgp_flags |= FGP_LOCK;
1977 :
1978 : /* Init accessed so avoid atomic mark_page_accessed later */
1979 0 : if (fgp_flags & FGP_ACCESSED)
1980 : __folio_set_referenced(folio);
1981 :
1982 0 : err = filemap_add_folio(mapping, folio, index, gfp);
1983 0 : if (unlikely(err)) {
1984 0 : folio_put(folio);
1985 0 : folio = NULL;
1986 0 : if (err == -EEXIST)
1987 : goto repeat;
1988 : }
1989 :
1990 : /*
1991 : * filemap_add_folio locks the page, and for mmap
1992 : * we expect an unlocked page.
1993 : */
1994 0 : if (folio && (fgp_flags & FGP_FOR_MMAP))
1995 : folio_unlock(folio);
1996 : }
1997 :
1998 0 : if (!folio)
1999 : return ERR_PTR(-ENOENT);
2000 0 : return folio;
2001 : }
2002 : EXPORT_SYMBOL(__filemap_get_folio);
2003 :
2004 0 : static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
2005 : xa_mark_t mark)
2006 : {
2007 : struct folio *folio;
2008 :
2009 : retry:
2010 0 : if (mark == XA_PRESENT)
2011 0 : folio = xas_find(xas, max);
2012 : else
2013 0 : folio = xas_find_marked(xas, max, mark);
2014 :
2015 0 : if (xas_retry(xas, folio))
2016 : goto retry;
2017 : /*
2018 : * A shadow entry of a recently evicted page, a swap
2019 : * entry from shmem/tmpfs or a DAX entry. Return it
2020 : * without attempting to raise page count.
2021 : */
2022 0 : if (!folio || xa_is_value(folio))
2023 : return folio;
2024 :
2025 0 : if (!folio_try_get_rcu(folio))
2026 : goto reset;
2027 :
2028 0 : if (unlikely(folio != xas_reload(xas))) {
2029 : folio_put(folio);
2030 : goto reset;
2031 : }
2032 :
2033 : return folio;
2034 : reset:
2035 0 : xas_reset(xas);
2036 : goto retry;
2037 : }
2038 :
2039 : /**
2040 : * find_get_entries - gang pagecache lookup
2041 : * @mapping: The address_space to search
2042 : * @start: The starting page cache index
2043 : * @end: The final page index (inclusive).
2044 : * @fbatch: Where the resulting entries are placed.
2045 : * @indices: The cache indices corresponding to the entries in @entries
2046 : *
2047 : * find_get_entries() will search for and return a batch of entries in
2048 : * the mapping. The entries are placed in @fbatch. find_get_entries()
2049 : * takes a reference on any actual folios it returns.
2050 : *
2051 : * The entries have ascending indexes. The indices may not be consecutive
2052 : * due to not-present entries or large folios.
2053 : *
2054 : * Any shadow entries of evicted folios, or swap entries from
2055 : * shmem/tmpfs, are included in the returned array.
2056 : *
2057 : * Return: The number of entries which were found.
2058 : */
2059 0 : unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
2060 : pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2061 : {
2062 0 : XA_STATE(xas, &mapping->i_pages, *start);
2063 : struct folio *folio;
2064 :
2065 : rcu_read_lock();
2066 0 : while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2067 0 : indices[fbatch->nr] = xas.xa_index;
2068 0 : if (!folio_batch_add(fbatch, folio))
2069 : break;
2070 : }
2071 : rcu_read_unlock();
2072 :
2073 0 : if (folio_batch_count(fbatch)) {
2074 0 : unsigned long nr = 1;
2075 0 : int idx = folio_batch_count(fbatch) - 1;
2076 :
2077 0 : folio = fbatch->folios[idx];
2078 0 : if (!xa_is_value(folio) && !folio_test_hugetlb(folio))
2079 0 : nr = folio_nr_pages(folio);
2080 0 : *start = indices[idx] + nr;
2081 : }
2082 0 : return folio_batch_count(fbatch);
2083 : }
2084 :
2085 : /**
2086 : * find_lock_entries - Find a batch of pagecache entries.
2087 : * @mapping: The address_space to search.
2088 : * @start: The starting page cache index.
2089 : * @end: The final page index (inclusive).
2090 : * @fbatch: Where the resulting entries are placed.
2091 : * @indices: The cache indices of the entries in @fbatch.
2092 : *
2093 : * find_lock_entries() will return a batch of entries from @mapping.
2094 : * Swap, shadow and DAX entries are included. Folios are returned
2095 : * locked and with an incremented refcount. Folios which are locked
2096 : * by somebody else or under writeback are skipped. Folios which are
2097 : * partially outside the range are not returned.
2098 : *
2099 : * The entries have ascending indexes. The indices may not be consecutive
2100 : * due to not-present entries, large folios, folios which could not be
2101 : * locked or folios under writeback.
2102 : *
2103 : * Return: The number of entries which were found.
2104 : */
2105 0 : unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
2106 : pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
2107 : {
2108 0 : XA_STATE(xas, &mapping->i_pages, *start);
2109 : struct folio *folio;
2110 :
2111 : rcu_read_lock();
2112 0 : while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2113 0 : if (!xa_is_value(folio)) {
2114 0 : if (folio->index < *start)
2115 : goto put;
2116 0 : if (folio->index + folio_nr_pages(folio) - 1 > end)
2117 : goto put;
2118 0 : if (!folio_trylock(folio))
2119 : goto put;
2120 0 : if (folio->mapping != mapping ||
2121 0 : folio_test_writeback(folio))
2122 : goto unlock;
2123 : VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2124 : folio);
2125 : }
2126 0 : indices[fbatch->nr] = xas.xa_index;
2127 0 : if (!folio_batch_add(fbatch, folio))
2128 : break;
2129 0 : continue;
2130 : unlock:
2131 : folio_unlock(folio);
2132 : put:
2133 : folio_put(folio);
2134 : }
2135 : rcu_read_unlock();
2136 :
2137 0 : if (folio_batch_count(fbatch)) {
2138 0 : unsigned long nr = 1;
2139 0 : int idx = folio_batch_count(fbatch) - 1;
2140 :
2141 0 : folio = fbatch->folios[idx];
2142 0 : if (!xa_is_value(folio) && !folio_test_hugetlb(folio))
2143 0 : nr = folio_nr_pages(folio);
2144 0 : *start = indices[idx] + nr;
2145 : }
2146 0 : return folio_batch_count(fbatch);
2147 : }
2148 :
2149 : /**
2150 : * filemap_get_folios - Get a batch of folios
2151 : * @mapping: The address_space to search
2152 : * @start: The starting page index
2153 : * @end: The final page index (inclusive)
2154 : * @fbatch: The batch to fill.
2155 : *
2156 : * Search for and return a batch of folios in the mapping starting at
2157 : * index @start and up to index @end (inclusive). The folios are returned
2158 : * in @fbatch with an elevated reference count.
2159 : *
2160 : * The first folio may start before @start; if it does, it will contain
2161 : * @start. The final folio may extend beyond @end; if it does, it will
2162 : * contain @end. The folios have ascending indices. There may be gaps
2163 : * between the folios if there are indices which have no folio in the
2164 : * page cache. If folios are added to or removed from the page cache
2165 : * while this is running, they may or may not be found by this call.
2166 : *
2167 : * Return: The number of folios which were found.
2168 : * We also update @start to index the next folio for the traversal.
2169 : */
2170 0 : unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
2171 : pgoff_t end, struct folio_batch *fbatch)
2172 : {
2173 0 : XA_STATE(xas, &mapping->i_pages, *start);
2174 : struct folio *folio;
2175 :
2176 : rcu_read_lock();
2177 0 : while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2178 : /* Skip over shadow, swap and DAX entries */
2179 0 : if (xa_is_value(folio))
2180 0 : continue;
2181 0 : if (!folio_batch_add(fbatch, folio)) {
2182 0 : unsigned long nr = folio_nr_pages(folio);
2183 :
2184 0 : if (folio_test_hugetlb(folio))
2185 : nr = 1;
2186 0 : *start = folio->index + nr;
2187 0 : goto out;
2188 : }
2189 : }
2190 :
2191 : /*
2192 : * We come here when there is no page beyond @end. We take care to not
2193 : * overflow the index @start as it confuses some of the callers. This
2194 : * breaks the iteration when there is a page at index -1 but that is
2195 : * already broken anyway.
2196 : */
2197 0 : if (end == (pgoff_t)-1)
2198 0 : *start = (pgoff_t)-1;
2199 : else
2200 0 : *start = end + 1;
2201 : out:
2202 : rcu_read_unlock();
2203 :
2204 0 : return folio_batch_count(fbatch);
2205 : }
2206 : EXPORT_SYMBOL(filemap_get_folios);
2207 :
2208 : static inline
2209 : bool folio_more_pages(struct folio *folio, pgoff_t index, pgoff_t max)
2210 : {
2211 0 : if (!folio_test_large(folio) || folio_test_hugetlb(folio))
2212 : return false;
2213 0 : if (index >= max)
2214 : return false;
2215 0 : return index < folio->index + folio_nr_pages(folio) - 1;
2216 : }
2217 :
2218 : /**
2219 : * filemap_get_folios_contig - Get a batch of contiguous folios
2220 : * @mapping: The address_space to search
2221 : * @start: The starting page index
2222 : * @end: The final page index (inclusive)
2223 : * @fbatch: The batch to fill
2224 : *
2225 : * filemap_get_folios_contig() works exactly like filemap_get_folios(),
2226 : * except the returned folios are guaranteed to be contiguous. This may
2227 : * not return all contiguous folios if the batch gets filled up.
2228 : *
2229 : * Return: The number of folios found.
2230 : * Also update @start to be positioned for traversal of the next folio.
2231 : */
2232 :
2233 0 : unsigned filemap_get_folios_contig(struct address_space *mapping,
2234 : pgoff_t *start, pgoff_t end, struct folio_batch *fbatch)
2235 : {
2236 0 : XA_STATE(xas, &mapping->i_pages, *start);
2237 : unsigned long nr;
2238 : struct folio *folio;
2239 :
2240 : rcu_read_lock();
2241 :
2242 0 : for (folio = xas_load(&xas); folio && xas.xa_index <= end;
2243 0 : folio = xas_next(&xas)) {
2244 0 : if (xas_retry(&xas, folio))
2245 0 : continue;
2246 : /*
2247 : * If the entry has been swapped out, we can stop looking.
2248 : * No current caller is looking for DAX entries.
2249 : */
2250 0 : if (xa_is_value(folio))
2251 : goto update_start;
2252 :
2253 0 : if (!folio_try_get_rcu(folio))
2254 : goto retry;
2255 :
2256 0 : if (unlikely(folio != xas_reload(&xas)))
2257 : goto put_folio;
2258 :
2259 0 : if (!folio_batch_add(fbatch, folio)) {
2260 0 : nr = folio_nr_pages(folio);
2261 :
2262 0 : if (folio_test_hugetlb(folio))
2263 : nr = 1;
2264 0 : *start = folio->index + nr;
2265 0 : goto out;
2266 : }
2267 0 : continue;
2268 : put_folio:
2269 : folio_put(folio);
2270 :
2271 : retry:
2272 0 : xas_reset(&xas);
2273 : }
2274 :
2275 : update_start:
2276 0 : nr = folio_batch_count(fbatch);
2277 :
2278 0 : if (nr) {
2279 0 : folio = fbatch->folios[nr - 1];
2280 0 : if (folio_test_hugetlb(folio))
2281 : *start = folio->index + 1;
2282 : else
2283 0 : *start = folio->index + folio_nr_pages(folio);
2284 : }
2285 : out:
2286 : rcu_read_unlock();
2287 0 : return folio_batch_count(fbatch);
2288 : }
2289 : EXPORT_SYMBOL(filemap_get_folios_contig);
2290 :
2291 : /**
2292 : * filemap_get_folios_tag - Get a batch of folios matching @tag
2293 : * @mapping: The address_space to search
2294 : * @start: The starting page index
2295 : * @end: The final page index (inclusive)
2296 : * @tag: The tag index
2297 : * @fbatch: The batch to fill
2298 : *
2299 : * Same as filemap_get_folios(), but only returning folios tagged with @tag.
2300 : *
2301 : * Return: The number of folios found.
2302 : * Also update @start to index the next folio for traversal.
2303 : */
2304 0 : unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start,
2305 : pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch)
2306 : {
2307 0 : XA_STATE(xas, &mapping->i_pages, *start);
2308 : struct folio *folio;
2309 :
2310 : rcu_read_lock();
2311 0 : while ((folio = find_get_entry(&xas, end, tag)) != NULL) {
2312 : /*
2313 : * Shadow entries should never be tagged, but this iteration
2314 : * is lockless so there is a window for page reclaim to evict
2315 : * a page we saw tagged. Skip over it.
2316 : */
2317 0 : if (xa_is_value(folio))
2318 0 : continue;
2319 0 : if (!folio_batch_add(fbatch, folio)) {
2320 0 : unsigned long nr = folio_nr_pages(folio);
2321 :
2322 0 : if (folio_test_hugetlb(folio))
2323 : nr = 1;
2324 0 : *start = folio->index + nr;
2325 0 : goto out;
2326 : }
2327 : }
2328 : /*
2329 : * We come here when there is no page beyond @end. We take care to not
2330 : * overflow the index @start as it confuses some of the callers. This
2331 : * breaks the iteration when there is a page at index -1 but that is
2332 : * already broke anyway.
2333 : */
2334 0 : if (end == (pgoff_t)-1)
2335 0 : *start = (pgoff_t)-1;
2336 : else
2337 0 : *start = end + 1;
2338 : out:
2339 : rcu_read_unlock();
2340 :
2341 0 : return folio_batch_count(fbatch);
2342 : }
2343 : EXPORT_SYMBOL(filemap_get_folios_tag);
2344 :
2345 : /*
2346 : * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2347 : * a _large_ part of the i/o request. Imagine the worst scenario:
2348 : *
2349 : * ---R__________________________________________B__________
2350 : * ^ reading here ^ bad block(assume 4k)
2351 : *
2352 : * read(R) => miss => readahead(R...B) => media error => frustrating retries
2353 : * => failing the whole request => read(R) => read(R+1) =>
2354 : * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2355 : * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2356 : * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2357 : *
2358 : * It is going insane. Fix it by quickly scaling down the readahead size.
2359 : */
2360 : static void shrink_readahead_size_eio(struct file_ra_state *ra)
2361 : {
2362 0 : ra->ra_pages /= 4;
2363 : }
2364 :
2365 : /*
2366 : * filemap_get_read_batch - Get a batch of folios for read
2367 : *
2368 : * Get a batch of folios which represent a contiguous range of bytes in
2369 : * the file. No exceptional entries will be returned. If @index is in
2370 : * the middle of a folio, the entire folio will be returned. The last
2371 : * folio in the batch may have the readahead flag set or the uptodate flag
2372 : * clear so that the caller can take the appropriate action.
2373 : */
2374 0 : static void filemap_get_read_batch(struct address_space *mapping,
2375 : pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2376 : {
2377 0 : XA_STATE(xas, &mapping->i_pages, index);
2378 : struct folio *folio;
2379 :
2380 : rcu_read_lock();
2381 0 : for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2382 0 : if (xas_retry(&xas, folio))
2383 0 : continue;
2384 0 : if (xas.xa_index > max || xa_is_value(folio))
2385 : break;
2386 0 : if (xa_is_sibling(folio))
2387 : break;
2388 0 : if (!folio_try_get_rcu(folio))
2389 : goto retry;
2390 :
2391 0 : if (unlikely(folio != xas_reload(&xas)))
2392 : goto put_folio;
2393 :
2394 0 : if (!folio_batch_add(fbatch, folio))
2395 : break;
2396 0 : if (!folio_test_uptodate(folio))
2397 : break;
2398 0 : if (folio_test_readahead(folio))
2399 : break;
2400 0 : xas_advance(&xas, folio->index + folio_nr_pages(folio) - 1);
2401 0 : continue;
2402 : put_folio:
2403 : folio_put(folio);
2404 : retry:
2405 0 : xas_reset(&xas);
2406 : }
2407 : rcu_read_unlock();
2408 0 : }
2409 :
2410 0 : static int filemap_read_folio(struct file *file, filler_t filler,
2411 : struct folio *folio)
2412 : {
2413 0 : bool workingset = folio_test_workingset(folio);
2414 : unsigned long pflags;
2415 : int error;
2416 :
2417 : /*
2418 : * A previous I/O error may have been due to temporary failures,
2419 : * eg. multipath errors. PG_error will be set again if read_folio
2420 : * fails.
2421 : */
2422 0 : folio_clear_error(folio);
2423 :
2424 : /* Start the actual read. The read will unlock the page. */
2425 : if (unlikely(workingset))
2426 : psi_memstall_enter(&pflags);
2427 0 : error = filler(file, folio);
2428 : if (unlikely(workingset))
2429 : psi_memstall_leave(&pflags);
2430 0 : if (error)
2431 : return error;
2432 :
2433 0 : error = folio_wait_locked_killable(folio);
2434 0 : if (error)
2435 : return error;
2436 0 : if (folio_test_uptodate(folio))
2437 : return 0;
2438 0 : if (file)
2439 0 : shrink_readahead_size_eio(&file->f_ra);
2440 : return -EIO;
2441 : }
2442 :
2443 0 : static bool filemap_range_uptodate(struct address_space *mapping,
2444 : loff_t pos, size_t count, struct folio *folio,
2445 : bool need_uptodate)
2446 : {
2447 0 : if (folio_test_uptodate(folio))
2448 : return true;
2449 : /* pipes can't handle partially uptodate pages */
2450 0 : if (need_uptodate)
2451 : return false;
2452 0 : if (!mapping->a_ops->is_partially_uptodate)
2453 : return false;
2454 0 : if (mapping->host->i_blkbits >= folio_shift(folio))
2455 : return false;
2456 :
2457 0 : if (folio_pos(folio) > pos) {
2458 0 : count -= folio_pos(folio) - pos;
2459 0 : pos = 0;
2460 : } else {
2461 0 : pos -= folio_pos(folio);
2462 : }
2463 :
2464 0 : return mapping->a_ops->is_partially_uptodate(folio, pos, count);
2465 : }
2466 :
2467 0 : static int filemap_update_page(struct kiocb *iocb,
2468 : struct address_space *mapping, size_t count,
2469 : struct folio *folio, bool need_uptodate)
2470 : {
2471 : int error;
2472 :
2473 0 : if (iocb->ki_flags & IOCB_NOWAIT) {
2474 0 : if (!filemap_invalidate_trylock_shared(mapping))
2475 : return -EAGAIN;
2476 : } else {
2477 : filemap_invalidate_lock_shared(mapping);
2478 : }
2479 :
2480 0 : if (!folio_trylock(folio)) {
2481 0 : error = -EAGAIN;
2482 0 : if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2483 : goto unlock_mapping;
2484 0 : if (!(iocb->ki_flags & IOCB_WAITQ)) {
2485 0 : filemap_invalidate_unlock_shared(mapping);
2486 : /*
2487 : * This is where we usually end up waiting for a
2488 : * previously submitted readahead to finish.
2489 : */
2490 0 : folio_put_wait_locked(folio, TASK_KILLABLE);
2491 0 : return AOP_TRUNCATED_PAGE;
2492 : }
2493 0 : error = __folio_lock_async(folio, iocb->ki_waitq);
2494 0 : if (error)
2495 : goto unlock_mapping;
2496 : }
2497 :
2498 0 : error = AOP_TRUNCATED_PAGE;
2499 0 : if (!folio->mapping)
2500 : goto unlock;
2501 :
2502 0 : error = 0;
2503 0 : if (filemap_range_uptodate(mapping, iocb->ki_pos, count, folio,
2504 : need_uptodate))
2505 : goto unlock;
2506 :
2507 0 : error = -EAGAIN;
2508 0 : if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2509 : goto unlock;
2510 :
2511 0 : error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
2512 : folio);
2513 0 : goto unlock_mapping;
2514 : unlock:
2515 : folio_unlock(folio);
2516 : unlock_mapping:
2517 0 : filemap_invalidate_unlock_shared(mapping);
2518 0 : if (error == AOP_TRUNCATED_PAGE)
2519 : folio_put(folio);
2520 : return error;
2521 : }
2522 :
2523 0 : static int filemap_create_folio(struct file *file,
2524 : struct address_space *mapping, pgoff_t index,
2525 : struct folio_batch *fbatch)
2526 : {
2527 : struct folio *folio;
2528 : int error;
2529 :
2530 0 : folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2531 0 : if (!folio)
2532 : return -ENOMEM;
2533 :
2534 : /*
2535 : * Protect against truncate / hole punch. Grabbing invalidate_lock
2536 : * here assures we cannot instantiate and bring uptodate new
2537 : * pagecache folios after evicting page cache during truncate
2538 : * and before actually freeing blocks. Note that we could
2539 : * release invalidate_lock after inserting the folio into
2540 : * the page cache as the locked folio would then be enough to
2541 : * synchronize with hole punching. But there are code paths
2542 : * such as filemap_update_page() filling in partially uptodate
2543 : * pages or ->readahead() that need to hold invalidate_lock
2544 : * while mapping blocks for IO so let's hold the lock here as
2545 : * well to keep locking rules simple.
2546 : */
2547 0 : filemap_invalidate_lock_shared(mapping);
2548 0 : error = filemap_add_folio(mapping, folio, index,
2549 : mapping_gfp_constraint(mapping, GFP_KERNEL));
2550 0 : if (error == -EEXIST)
2551 0 : error = AOP_TRUNCATED_PAGE;
2552 0 : if (error)
2553 : goto error;
2554 :
2555 0 : error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
2556 0 : if (error)
2557 : goto error;
2558 :
2559 0 : filemap_invalidate_unlock_shared(mapping);
2560 0 : folio_batch_add(fbatch, folio);
2561 0 : return 0;
2562 : error:
2563 0 : filemap_invalidate_unlock_shared(mapping);
2564 : folio_put(folio);
2565 : return error;
2566 : }
2567 :
2568 0 : static int filemap_readahead(struct kiocb *iocb, struct file *file,
2569 : struct address_space *mapping, struct folio *folio,
2570 : pgoff_t last_index)
2571 : {
2572 0 : DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2573 :
2574 0 : if (iocb->ki_flags & IOCB_NOIO)
2575 : return -EAGAIN;
2576 0 : page_cache_async_ra(&ractl, folio, last_index - folio->index);
2577 : return 0;
2578 : }
2579 :
2580 0 : static int filemap_get_pages(struct kiocb *iocb, size_t count,
2581 : struct folio_batch *fbatch, bool need_uptodate)
2582 : {
2583 0 : struct file *filp = iocb->ki_filp;
2584 0 : struct address_space *mapping = filp->f_mapping;
2585 0 : struct file_ra_state *ra = &filp->f_ra;
2586 0 : pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2587 : pgoff_t last_index;
2588 : struct folio *folio;
2589 0 : int err = 0;
2590 :
2591 : /* "last_index" is the index of the page beyond the end of the read */
2592 0 : last_index = DIV_ROUND_UP(iocb->ki_pos + count, PAGE_SIZE);
2593 : retry:
2594 0 : if (fatal_signal_pending(current))
2595 : return -EINTR;
2596 :
2597 0 : filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2598 0 : if (!folio_batch_count(fbatch)) {
2599 0 : if (iocb->ki_flags & IOCB_NOIO)
2600 : return -EAGAIN;
2601 0 : page_cache_sync_readahead(mapping, ra, filp, index,
2602 : last_index - index);
2603 0 : filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2604 : }
2605 0 : if (!folio_batch_count(fbatch)) {
2606 0 : if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2607 : return -EAGAIN;
2608 0 : err = filemap_create_folio(filp, mapping,
2609 0 : iocb->ki_pos >> PAGE_SHIFT, fbatch);
2610 0 : if (err == AOP_TRUNCATED_PAGE)
2611 : goto retry;
2612 : return err;
2613 : }
2614 :
2615 0 : folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2616 0 : if (folio_test_readahead(folio)) {
2617 0 : err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2618 0 : if (err)
2619 : goto err;
2620 : }
2621 0 : if (!folio_test_uptodate(folio)) {
2622 0 : if ((iocb->ki_flags & IOCB_WAITQ) &&
2623 0 : folio_batch_count(fbatch) > 1)
2624 0 : iocb->ki_flags |= IOCB_NOWAIT;
2625 0 : err = filemap_update_page(iocb, mapping, count, folio,
2626 : need_uptodate);
2627 0 : if (err)
2628 : goto err;
2629 : }
2630 :
2631 : return 0;
2632 : err:
2633 0 : if (err < 0)
2634 : folio_put(folio);
2635 0 : if (likely(--fbatch->nr))
2636 : return 0;
2637 0 : if (err == AOP_TRUNCATED_PAGE)
2638 : goto retry;
2639 : return err;
2640 : }
2641 :
2642 : static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio)
2643 : {
2644 0 : unsigned int shift = folio_shift(folio);
2645 :
2646 0 : return (pos1 >> shift == pos2 >> shift);
2647 : }
2648 :
2649 : /**
2650 : * filemap_read - Read data from the page cache.
2651 : * @iocb: The iocb to read.
2652 : * @iter: Destination for the data.
2653 : * @already_read: Number of bytes already read by the caller.
2654 : *
2655 : * Copies data from the page cache. If the data is not currently present,
2656 : * uses the readahead and read_folio address_space operations to fetch it.
2657 : *
2658 : * Return: Total number of bytes copied, including those already read by
2659 : * the caller. If an error happens before any bytes are copied, returns
2660 : * a negative error number.
2661 : */
2662 0 : ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2663 : ssize_t already_read)
2664 : {
2665 0 : struct file *filp = iocb->ki_filp;
2666 0 : struct file_ra_state *ra = &filp->f_ra;
2667 0 : struct address_space *mapping = filp->f_mapping;
2668 0 : struct inode *inode = mapping->host;
2669 : struct folio_batch fbatch;
2670 0 : int i, error = 0;
2671 : bool writably_mapped;
2672 : loff_t isize, end_offset;
2673 :
2674 0 : if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2675 : return 0;
2676 0 : if (unlikely(!iov_iter_count(iter)))
2677 : return 0;
2678 :
2679 0 : iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2680 0 : folio_batch_init(&fbatch);
2681 :
2682 : do {
2683 0 : cond_resched();
2684 :
2685 : /*
2686 : * If we've already successfully copied some data, then we
2687 : * can no longer safely return -EIOCBQUEUED. Hence mark
2688 : * an async read NOWAIT at that point.
2689 : */
2690 0 : if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2691 0 : iocb->ki_flags |= IOCB_NOWAIT;
2692 :
2693 0 : if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2694 : break;
2695 :
2696 0 : error = filemap_get_pages(iocb, iter->count, &fbatch,
2697 0 : iov_iter_is_pipe(iter));
2698 0 : if (error < 0)
2699 : break;
2700 :
2701 : /*
2702 : * i_size must be checked after we know the pages are Uptodate.
2703 : *
2704 : * Checking i_size after the check allows us to calculate
2705 : * the correct value for "nr", which means the zero-filled
2706 : * part of the page is not copied back to userspace (unless
2707 : * another truncate extends the file - this is desired though).
2708 : */
2709 0 : isize = i_size_read(inode);
2710 0 : if (unlikely(iocb->ki_pos >= isize))
2711 : goto put_folios;
2712 0 : end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2713 :
2714 : /*
2715 : * Once we start copying data, we don't want to be touching any
2716 : * cachelines that might be contended:
2717 : */
2718 0 : writably_mapped = mapping_writably_mapped(mapping);
2719 :
2720 : /*
2721 : * When a read accesses the same folio several times, only
2722 : * mark it as accessed the first time.
2723 : */
2724 0 : if (!pos_same_folio(iocb->ki_pos, ra->prev_pos - 1,
2725 : fbatch.folios[0]))
2726 0 : folio_mark_accessed(fbatch.folios[0]);
2727 :
2728 0 : for (i = 0; i < folio_batch_count(&fbatch); i++) {
2729 0 : struct folio *folio = fbatch.folios[i];
2730 0 : size_t fsize = folio_size(folio);
2731 0 : size_t offset = iocb->ki_pos & (fsize - 1);
2732 0 : size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2733 : fsize - offset);
2734 : size_t copied;
2735 :
2736 0 : if (end_offset < folio_pos(folio))
2737 : break;
2738 0 : if (i > 0)
2739 0 : folio_mark_accessed(folio);
2740 : /*
2741 : * If users can be writing to this folio using arbitrary
2742 : * virtual addresses, take care of potential aliasing
2743 : * before reading the folio on the kernel side.
2744 : */
2745 : if (writably_mapped)
2746 : flush_dcache_folio(folio);
2747 :
2748 0 : copied = copy_folio_to_iter(folio, offset, bytes, iter);
2749 :
2750 0 : already_read += copied;
2751 0 : iocb->ki_pos += copied;
2752 0 : ra->prev_pos = iocb->ki_pos;
2753 :
2754 0 : if (copied < bytes) {
2755 : error = -EFAULT;
2756 : break;
2757 : }
2758 : }
2759 : put_folios:
2760 0 : for (i = 0; i < folio_batch_count(&fbatch); i++)
2761 0 : folio_put(fbatch.folios[i]);
2762 0 : folio_batch_init(&fbatch);
2763 0 : } while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2764 :
2765 0 : file_accessed(filp);
2766 :
2767 0 : return already_read ? already_read : error;
2768 : }
2769 : EXPORT_SYMBOL_GPL(filemap_read);
2770 :
2771 : /**
2772 : * generic_file_read_iter - generic filesystem read routine
2773 : * @iocb: kernel I/O control block
2774 : * @iter: destination for the data read
2775 : *
2776 : * This is the "read_iter()" routine for all filesystems
2777 : * that can use the page cache directly.
2778 : *
2779 : * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2780 : * be returned when no data can be read without waiting for I/O requests
2781 : * to complete; it doesn't prevent readahead.
2782 : *
2783 : * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2784 : * requests shall be made for the read or for readahead. When no data
2785 : * can be read, -EAGAIN shall be returned. When readahead would be
2786 : * triggered, a partial, possibly empty read shall be returned.
2787 : *
2788 : * Return:
2789 : * * number of bytes copied, even for partial reads
2790 : * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2791 : */
2792 : ssize_t
2793 0 : generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2794 : {
2795 0 : size_t count = iov_iter_count(iter);
2796 0 : ssize_t retval = 0;
2797 :
2798 0 : if (!count)
2799 : return 0; /* skip atime */
2800 :
2801 0 : if (iocb->ki_flags & IOCB_DIRECT) {
2802 0 : struct file *file = iocb->ki_filp;
2803 0 : struct address_space *mapping = file->f_mapping;
2804 0 : struct inode *inode = mapping->host;
2805 :
2806 0 : if (iocb->ki_flags & IOCB_NOWAIT) {
2807 0 : if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2808 0 : iocb->ki_pos + count - 1))
2809 : return -EAGAIN;
2810 : } else {
2811 0 : retval = filemap_write_and_wait_range(mapping,
2812 : iocb->ki_pos,
2813 0 : iocb->ki_pos + count - 1);
2814 0 : if (retval < 0)
2815 : return retval;
2816 : }
2817 :
2818 0 : file_accessed(file);
2819 :
2820 0 : retval = mapping->a_ops->direct_IO(iocb, iter);
2821 0 : if (retval >= 0) {
2822 0 : iocb->ki_pos += retval;
2823 0 : count -= retval;
2824 : }
2825 0 : if (retval != -EIOCBQUEUED)
2826 0 : iov_iter_revert(iter, count - iov_iter_count(iter));
2827 :
2828 : /*
2829 : * Btrfs can have a short DIO read if we encounter
2830 : * compressed extents, so if there was an error, or if
2831 : * we've already read everything we wanted to, or if
2832 : * there was a short read because we hit EOF, go ahead
2833 : * and return. Otherwise fallthrough to buffered io for
2834 : * the rest of the read. Buffered reads will not work for
2835 : * DAX files, so don't bother trying.
2836 : */
2837 0 : if (retval < 0 || !count || IS_DAX(inode))
2838 : return retval;
2839 0 : if (iocb->ki_pos >= i_size_read(inode))
2840 : return retval;
2841 : }
2842 :
2843 0 : return filemap_read(iocb, iter, retval);
2844 : }
2845 : EXPORT_SYMBOL(generic_file_read_iter);
2846 :
2847 : /*
2848 : * Splice subpages from a folio into a pipe.
2849 : */
2850 0 : size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
2851 : struct folio *folio, loff_t fpos, size_t size)
2852 : {
2853 : struct page *page;
2854 0 : size_t spliced = 0, offset = offset_in_folio(folio, fpos);
2855 :
2856 0 : page = folio_page(folio, offset / PAGE_SIZE);
2857 0 : size = min(size, folio_size(folio) - offset);
2858 0 : offset %= PAGE_SIZE;
2859 :
2860 0 : while (spliced < size &&
2861 0 : !pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2862 0 : struct pipe_buffer *buf = pipe_head_buf(pipe);
2863 0 : size_t part = min_t(size_t, PAGE_SIZE - offset, size - spliced);
2864 :
2865 0 : *buf = (struct pipe_buffer) {
2866 : .ops = &page_cache_pipe_buf_ops,
2867 : .page = page,
2868 : .offset = offset,
2869 : .len = part,
2870 : };
2871 0 : folio_get(folio);
2872 0 : pipe->head++;
2873 0 : page++;
2874 0 : spliced += part;
2875 0 : offset = 0;
2876 : }
2877 :
2878 0 : return spliced;
2879 : }
2880 :
2881 : /*
2882 : * Splice folios from the pagecache of a buffered (ie. non-O_DIRECT) file into
2883 : * a pipe.
2884 : */
2885 0 : ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
2886 : struct pipe_inode_info *pipe,
2887 : size_t len, unsigned int flags)
2888 : {
2889 : struct folio_batch fbatch;
2890 : struct kiocb iocb;
2891 0 : size_t total_spliced = 0, used, npages;
2892 : loff_t isize, end_offset;
2893 : bool writably_mapped;
2894 0 : int i, error = 0;
2895 :
2896 0 : init_sync_kiocb(&iocb, in);
2897 0 : iocb.ki_pos = *ppos;
2898 :
2899 : /* Work out how much data we can actually add into the pipe */
2900 0 : used = pipe_occupancy(pipe->head, pipe->tail);
2901 0 : npages = max_t(ssize_t, pipe->max_usage - used, 0);
2902 0 : len = min_t(size_t, len, npages * PAGE_SIZE);
2903 :
2904 0 : folio_batch_init(&fbatch);
2905 :
2906 : do {
2907 0 : cond_resched();
2908 :
2909 0 : if (*ppos >= i_size_read(file_inode(in)))
2910 : break;
2911 :
2912 0 : iocb.ki_pos = *ppos;
2913 0 : error = filemap_get_pages(&iocb, len, &fbatch, true);
2914 0 : if (error < 0)
2915 : break;
2916 :
2917 : /*
2918 : * i_size must be checked after we know the pages are Uptodate.
2919 : *
2920 : * Checking i_size after the check allows us to calculate
2921 : * the correct value for "nr", which means the zero-filled
2922 : * part of the page is not copied back to userspace (unless
2923 : * another truncate extends the file - this is desired though).
2924 : */
2925 0 : isize = i_size_read(file_inode(in));
2926 0 : if (unlikely(*ppos >= isize))
2927 : break;
2928 0 : end_offset = min_t(loff_t, isize, *ppos + len);
2929 :
2930 : /*
2931 : * Once we start copying data, we don't want to be touching any
2932 : * cachelines that might be contended:
2933 : */
2934 0 : writably_mapped = mapping_writably_mapped(in->f_mapping);
2935 :
2936 0 : for (i = 0; i < folio_batch_count(&fbatch); i++) {
2937 0 : struct folio *folio = fbatch.folios[i];
2938 : size_t n;
2939 :
2940 0 : if (folio_pos(folio) >= end_offset)
2941 : goto out;
2942 0 : folio_mark_accessed(folio);
2943 :
2944 : /*
2945 : * If users can be writing to this folio using arbitrary
2946 : * virtual addresses, take care of potential aliasing
2947 : * before reading the folio on the kernel side.
2948 : */
2949 : if (writably_mapped)
2950 : flush_dcache_folio(folio);
2951 :
2952 0 : n = min_t(loff_t, len, isize - *ppos);
2953 0 : n = splice_folio_into_pipe(pipe, folio, *ppos, n);
2954 0 : if (!n)
2955 : goto out;
2956 0 : len -= n;
2957 0 : total_spliced += n;
2958 0 : *ppos += n;
2959 0 : in->f_ra.prev_pos = *ppos;
2960 0 : if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
2961 : goto out;
2962 : }
2963 :
2964 0 : folio_batch_release(&fbatch);
2965 0 : } while (len);
2966 :
2967 : out:
2968 0 : folio_batch_release(&fbatch);
2969 0 : file_accessed(in);
2970 :
2971 0 : return total_spliced ? total_spliced : error;
2972 : }
2973 : EXPORT_SYMBOL(filemap_splice_read);
2974 :
2975 0 : static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2976 : struct address_space *mapping, struct folio *folio,
2977 : loff_t start, loff_t end, bool seek_data)
2978 : {
2979 0 : const struct address_space_operations *ops = mapping->a_ops;
2980 0 : size_t offset, bsz = i_blocksize(mapping->host);
2981 :
2982 0 : if (xa_is_value(folio) || folio_test_uptodate(folio))
2983 0 : return seek_data ? start : end;
2984 0 : if (!ops->is_partially_uptodate)
2985 0 : return seek_data ? end : start;
2986 :
2987 0 : xas_pause(xas);
2988 : rcu_read_unlock();
2989 0 : folio_lock(folio);
2990 0 : if (unlikely(folio->mapping != mapping))
2991 : goto unlock;
2992 :
2993 0 : offset = offset_in_folio(folio, start) & ~(bsz - 1);
2994 :
2995 : do {
2996 0 : if (ops->is_partially_uptodate(folio, offset, bsz) ==
2997 : seek_data)
2998 : break;
2999 0 : start = (start + bsz) & ~(bsz - 1);
3000 0 : offset += bsz;
3001 0 : } while (offset < folio_size(folio));
3002 : unlock:
3003 0 : folio_unlock(folio);
3004 : rcu_read_lock();
3005 0 : return start;
3006 : }
3007 :
3008 : static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
3009 : {
3010 0 : if (xa_is_value(folio))
3011 : return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
3012 0 : return folio_size(folio);
3013 : }
3014 :
3015 : /**
3016 : * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3017 : * @mapping: Address space to search.
3018 : * @start: First byte to consider.
3019 : * @end: Limit of search (exclusive).
3020 : * @whence: Either SEEK_HOLE or SEEK_DATA.
3021 : *
3022 : * If the page cache knows which blocks contain holes and which blocks
3023 : * contain data, your filesystem can use this function to implement
3024 : * SEEK_HOLE and SEEK_DATA. This is useful for filesystems which are
3025 : * entirely memory-based such as tmpfs, and filesystems which support
3026 : * unwritten extents.
3027 : *
3028 : * Return: The requested offset on success, or -ENXIO if @whence specifies
3029 : * SEEK_DATA and there is no data after @start. There is an implicit hole
3030 : * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
3031 : * and @end contain data.
3032 : */
3033 0 : loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
3034 : loff_t end, int whence)
3035 : {
3036 0 : XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
3037 0 : pgoff_t max = (end - 1) >> PAGE_SHIFT;
3038 0 : bool seek_data = (whence == SEEK_DATA);
3039 : struct folio *folio;
3040 :
3041 0 : if (end <= start)
3042 : return -ENXIO;
3043 :
3044 : rcu_read_lock();
3045 0 : while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
3046 0 : loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
3047 : size_t seek_size;
3048 :
3049 0 : if (start < pos) {
3050 0 : if (!seek_data)
3051 : goto unlock;
3052 : start = pos;
3053 : }
3054 :
3055 0 : seek_size = seek_folio_size(&xas, folio);
3056 0 : pos = round_up((u64)pos + 1, seek_size);
3057 0 : start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
3058 : seek_data);
3059 0 : if (start < pos)
3060 : goto unlock;
3061 0 : if (start >= end)
3062 : break;
3063 0 : if (seek_size > PAGE_SIZE)
3064 0 : xas_set(&xas, pos >> PAGE_SHIFT);
3065 0 : if (!xa_is_value(folio))
3066 : folio_put(folio);
3067 : }
3068 0 : if (seek_data)
3069 0 : start = -ENXIO;
3070 : unlock:
3071 : rcu_read_unlock();
3072 0 : if (folio && !xa_is_value(folio))
3073 : folio_put(folio);
3074 0 : if (start > end)
3075 : return end;
3076 0 : return start;
3077 : }
3078 :
3079 : #ifdef CONFIG_MMU
3080 : #define MMAP_LOTSAMISS (100)
3081 : /*
3082 : * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
3083 : * @vmf - the vm_fault for this fault.
3084 : * @folio - the folio to lock.
3085 : * @fpin - the pointer to the file we may pin (or is already pinned).
3086 : *
3087 : * This works similar to lock_folio_or_retry in that it can drop the
3088 : * mmap_lock. It differs in that it actually returns the folio locked
3089 : * if it returns 1 and 0 if it couldn't lock the folio. If we did have
3090 : * to drop the mmap_lock then fpin will point to the pinned file and
3091 : * needs to be fput()'ed at a later point.
3092 : */
3093 0 : static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
3094 : struct file **fpin)
3095 : {
3096 0 : if (folio_trylock(folio))
3097 : return 1;
3098 :
3099 : /*
3100 : * NOTE! This will make us return with VM_FAULT_RETRY, but with
3101 : * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
3102 : * is supposed to work. We have way too many special cases..
3103 : */
3104 0 : if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3105 : return 0;
3106 :
3107 0 : *fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3108 0 : if (vmf->flags & FAULT_FLAG_KILLABLE) {
3109 0 : if (__folio_lock_killable(folio)) {
3110 : /*
3111 : * We didn't have the right flags to drop the mmap_lock,
3112 : * but all fault_handlers only check for fatal signals
3113 : * if we return VM_FAULT_RETRY, so we need to drop the
3114 : * mmap_lock here and return 0 if we don't have a fpin.
3115 : */
3116 0 : if (*fpin == NULL)
3117 0 : mmap_read_unlock(vmf->vma->vm_mm);
3118 : return 0;
3119 : }
3120 : } else
3121 : __folio_lock(folio);
3122 :
3123 : return 1;
3124 : }
3125 :
3126 : /*
3127 : * Synchronous readahead happens when we don't even find a page in the page
3128 : * cache at all. We don't want to perform IO under the mmap sem, so if we have
3129 : * to drop the mmap sem we return the file that was pinned in order for us to do
3130 : * that. If we didn't pin a file then we return NULL. The file that is
3131 : * returned needs to be fput()'ed when we're done with it.
3132 : */
3133 0 : static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3134 : {
3135 0 : struct file *file = vmf->vma->vm_file;
3136 0 : struct file_ra_state *ra = &file->f_ra;
3137 0 : struct address_space *mapping = file->f_mapping;
3138 0 : DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
3139 0 : struct file *fpin = NULL;
3140 0 : unsigned long vm_flags = vmf->vma->vm_flags;
3141 : unsigned int mmap_miss;
3142 :
3143 : #ifdef CONFIG_TRANSPARENT_HUGEPAGE
3144 : /* Use the readahead code, even if readahead is disabled */
3145 : if (vm_flags & VM_HUGEPAGE) {
3146 : fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3147 : ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1);
3148 : ra->size = HPAGE_PMD_NR;
3149 : /*
3150 : * Fetch two PMD folios, so we get the chance to actually
3151 : * readahead, unless we've been told not to.
3152 : */
3153 : if (!(vm_flags & VM_RAND_READ))
3154 : ra->size *= 2;
3155 : ra->async_size = HPAGE_PMD_NR;
3156 : page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER);
3157 : return fpin;
3158 : }
3159 : #endif
3160 :
3161 : /* If we don't want any read-ahead, don't bother */
3162 0 : if (vm_flags & VM_RAND_READ)
3163 : return fpin;
3164 0 : if (!ra->ra_pages)
3165 : return fpin;
3166 :
3167 0 : if (vm_flags & VM_SEQ_READ) {
3168 0 : fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3169 0 : page_cache_sync_ra(&ractl, ra->ra_pages);
3170 0 : return fpin;
3171 : }
3172 :
3173 : /* Avoid banging the cache line if not needed */
3174 0 : mmap_miss = READ_ONCE(ra->mmap_miss);
3175 0 : if (mmap_miss < MMAP_LOTSAMISS * 10)
3176 0 : WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3177 :
3178 : /*
3179 : * Do we miss much more than hit in this file? If so,
3180 : * stop bothering with read-ahead. It will only hurt.
3181 : */
3182 0 : if (mmap_miss > MMAP_LOTSAMISS)
3183 : return fpin;
3184 :
3185 : /*
3186 : * mmap read-around
3187 : */
3188 0 : fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3189 0 : ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3190 0 : ra->size = ra->ra_pages;
3191 0 : ra->async_size = ra->ra_pages / 4;
3192 0 : ractl._index = ra->start;
3193 0 : page_cache_ra_order(&ractl, ra, 0);
3194 0 : return fpin;
3195 : }
3196 :
3197 : /*
3198 : * Asynchronous readahead happens when we find the page and PG_readahead,
3199 : * so we want to possibly extend the readahead further. We return the file that
3200 : * was pinned if we have to drop the mmap_lock in order to do IO.
3201 : */
3202 0 : static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3203 : struct folio *folio)
3204 : {
3205 0 : struct file *file = vmf->vma->vm_file;
3206 0 : struct file_ra_state *ra = &file->f_ra;
3207 0 : DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3208 0 : struct file *fpin = NULL;
3209 : unsigned int mmap_miss;
3210 :
3211 : /* If we don't want any read-ahead, don't bother */
3212 0 : if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3213 : return fpin;
3214 :
3215 0 : mmap_miss = READ_ONCE(ra->mmap_miss);
3216 0 : if (mmap_miss)
3217 0 : WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3218 :
3219 0 : if (folio_test_readahead(folio)) {
3220 0 : fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3221 0 : page_cache_async_ra(&ractl, folio, ra->ra_pages);
3222 : }
3223 : return fpin;
3224 : }
3225 :
3226 : /**
3227 : * filemap_fault - read in file data for page fault handling
3228 : * @vmf: struct vm_fault containing details of the fault
3229 : *
3230 : * filemap_fault() is invoked via the vma operations vector for a
3231 : * mapped memory region to read in file data during a page fault.
3232 : *
3233 : * The goto's are kind of ugly, but this streamlines the normal case of having
3234 : * it in the page cache, and handles the special cases reasonably without
3235 : * having a lot of duplicated code.
3236 : *
3237 : * vma->vm_mm->mmap_lock must be held on entry.
3238 : *
3239 : * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3240 : * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3241 : *
3242 : * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3243 : * has not been released.
3244 : *
3245 : * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3246 : *
3247 : * Return: bitwise-OR of %VM_FAULT_ codes.
3248 : */
3249 0 : vm_fault_t filemap_fault(struct vm_fault *vmf)
3250 : {
3251 : int error;
3252 0 : struct file *file = vmf->vma->vm_file;
3253 0 : struct file *fpin = NULL;
3254 0 : struct address_space *mapping = file->f_mapping;
3255 0 : struct inode *inode = mapping->host;
3256 0 : pgoff_t max_idx, index = vmf->pgoff;
3257 : struct folio *folio;
3258 0 : vm_fault_t ret = 0;
3259 0 : bool mapping_locked = false;
3260 :
3261 0 : max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3262 0 : if (unlikely(index >= max_idx))
3263 : return VM_FAULT_SIGBUS;
3264 :
3265 : /*
3266 : * Do we have something in the page cache already?
3267 : */
3268 0 : folio = filemap_get_folio(mapping, index);
3269 0 : if (likely(!IS_ERR(folio))) {
3270 : /*
3271 : * We found the page, so try async readahead before waiting for
3272 : * the lock.
3273 : */
3274 0 : if (!(vmf->flags & FAULT_FLAG_TRIED))
3275 0 : fpin = do_async_mmap_readahead(vmf, folio);
3276 0 : if (unlikely(!folio_test_uptodate(folio))) {
3277 0 : filemap_invalidate_lock_shared(mapping);
3278 0 : mapping_locked = true;
3279 : }
3280 : } else {
3281 : /* No page in the page cache at all */
3282 0 : count_vm_event(PGMAJFAULT);
3283 0 : count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3284 0 : ret = VM_FAULT_MAJOR;
3285 0 : fpin = do_sync_mmap_readahead(vmf);
3286 : retry_find:
3287 : /*
3288 : * See comment in filemap_create_folio() why we need
3289 : * invalidate_lock
3290 : */
3291 0 : if (!mapping_locked) {
3292 0 : filemap_invalidate_lock_shared(mapping);
3293 0 : mapping_locked = true;
3294 : }
3295 0 : folio = __filemap_get_folio(mapping, index,
3296 : FGP_CREAT|FGP_FOR_MMAP,
3297 : vmf->gfp_mask);
3298 0 : if (IS_ERR(folio)) {
3299 0 : if (fpin)
3300 : goto out_retry;
3301 0 : filemap_invalidate_unlock_shared(mapping);
3302 0 : return VM_FAULT_OOM;
3303 : }
3304 : }
3305 :
3306 0 : if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3307 : goto out_retry;
3308 :
3309 : /* Did it get truncated? */
3310 0 : if (unlikely(folio->mapping != mapping)) {
3311 0 : folio_unlock(folio);
3312 : folio_put(folio);
3313 : goto retry_find;
3314 : }
3315 : VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3316 :
3317 : /*
3318 : * We have a locked page in the page cache, now we need to check
3319 : * that it's up-to-date. If not, it is going to be due to an error.
3320 : */
3321 0 : if (unlikely(!folio_test_uptodate(folio))) {
3322 : /*
3323 : * The page was in cache and uptodate and now it is not.
3324 : * Strange but possible since we didn't hold the page lock all
3325 : * the time. Let's drop everything get the invalidate lock and
3326 : * try again.
3327 : */
3328 0 : if (!mapping_locked) {
3329 0 : folio_unlock(folio);
3330 : folio_put(folio);
3331 : goto retry_find;
3332 : }
3333 : goto page_not_uptodate;
3334 : }
3335 :
3336 : /*
3337 : * We've made it this far and we had to drop our mmap_lock, now is the
3338 : * time to return to the upper layer and have it re-find the vma and
3339 : * redo the fault.
3340 : */
3341 0 : if (fpin) {
3342 : folio_unlock(folio);
3343 : goto out_retry;
3344 : }
3345 0 : if (mapping_locked)
3346 : filemap_invalidate_unlock_shared(mapping);
3347 :
3348 : /*
3349 : * Found the page and have a reference on it.
3350 : * We must recheck i_size under page lock.
3351 : */
3352 0 : max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3353 0 : if (unlikely(index >= max_idx)) {
3354 0 : folio_unlock(folio);
3355 : folio_put(folio);
3356 : return VM_FAULT_SIGBUS;
3357 : }
3358 :
3359 0 : vmf->page = folio_file_page(folio, index);
3360 0 : return ret | VM_FAULT_LOCKED;
3361 :
3362 : page_not_uptodate:
3363 : /*
3364 : * Umm, take care of errors if the page isn't up-to-date.
3365 : * Try to re-read it _once_. We do this synchronously,
3366 : * because there really aren't any performance issues here
3367 : * and we need to check for errors.
3368 : */
3369 0 : fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3370 0 : error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
3371 0 : if (fpin)
3372 : goto out_retry;
3373 0 : folio_put(folio);
3374 :
3375 0 : if (!error || error == AOP_TRUNCATED_PAGE)
3376 : goto retry_find;
3377 0 : filemap_invalidate_unlock_shared(mapping);
3378 :
3379 0 : return VM_FAULT_SIGBUS;
3380 :
3381 : out_retry:
3382 : /*
3383 : * We dropped the mmap_lock, we need to return to the fault handler to
3384 : * re-find the vma and come back and find our hopefully still populated
3385 : * page.
3386 : */
3387 0 : if (!IS_ERR(folio))
3388 : folio_put(folio);
3389 0 : if (mapping_locked)
3390 : filemap_invalidate_unlock_shared(mapping);
3391 0 : if (fpin)
3392 0 : fput(fpin);
3393 0 : return ret | VM_FAULT_RETRY;
3394 : }
3395 : EXPORT_SYMBOL(filemap_fault);
3396 :
3397 : static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio,
3398 : pgoff_t start)
3399 : {
3400 0 : struct mm_struct *mm = vmf->vma->vm_mm;
3401 :
3402 : /* Huge page is mapped? No need to proceed. */
3403 0 : if (pmd_trans_huge(*vmf->pmd)) {
3404 : folio_unlock(folio);
3405 : folio_put(folio);
3406 : return true;
3407 : }
3408 :
3409 0 : if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) {
3410 : struct page *page = folio_file_page(folio, start);
3411 : vm_fault_t ret = do_set_pmd(vmf, page);
3412 : if (!ret) {
3413 : /* The page is mapped successfully, reference consumed. */
3414 : folio_unlock(folio);
3415 : return true;
3416 : }
3417 : }
3418 :
3419 0 : if (pmd_none(*vmf->pmd))
3420 0 : pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3421 :
3422 : /* See comment in handle_pte_fault() */
3423 0 : if (pmd_devmap_trans_unstable(vmf->pmd)) {
3424 : folio_unlock(folio);
3425 : folio_put(folio);
3426 : return true;
3427 : }
3428 :
3429 : return false;
3430 : }
3431 :
3432 0 : static struct folio *next_uptodate_page(struct folio *folio,
3433 : struct address_space *mapping,
3434 : struct xa_state *xas, pgoff_t end_pgoff)
3435 : {
3436 : unsigned long max_idx;
3437 :
3438 : do {
3439 0 : if (!folio)
3440 : return NULL;
3441 0 : if (xas_retry(xas, folio))
3442 0 : continue;
3443 0 : if (xa_is_value(folio))
3444 0 : continue;
3445 0 : if (folio_test_locked(folio))
3446 0 : continue;
3447 0 : if (!folio_try_get_rcu(folio))
3448 : continue;
3449 : /* Has the page moved or been split? */
3450 0 : if (unlikely(folio != xas_reload(xas)))
3451 : goto skip;
3452 0 : if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
3453 : goto skip;
3454 0 : if (!folio_trylock(folio))
3455 : goto skip;
3456 0 : if (folio->mapping != mapping)
3457 : goto unlock;
3458 0 : if (!folio_test_uptodate(folio))
3459 : goto unlock;
3460 0 : max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3461 0 : if (xas->xa_index >= max_idx)
3462 : goto unlock;
3463 : return folio;
3464 : unlock:
3465 : folio_unlock(folio);
3466 : skip:
3467 : folio_put(folio);
3468 0 : } while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3469 :
3470 : return NULL;
3471 : }
3472 :
3473 0 : static inline struct folio *first_map_page(struct address_space *mapping,
3474 : struct xa_state *xas,
3475 : pgoff_t end_pgoff)
3476 : {
3477 0 : return next_uptodate_page(xas_find(xas, end_pgoff),
3478 : mapping, xas, end_pgoff);
3479 : }
3480 :
3481 0 : static inline struct folio *next_map_page(struct address_space *mapping,
3482 : struct xa_state *xas,
3483 : pgoff_t end_pgoff)
3484 : {
3485 0 : return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3486 : mapping, xas, end_pgoff);
3487 : }
3488 :
3489 0 : vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3490 : pgoff_t start_pgoff, pgoff_t end_pgoff)
3491 : {
3492 0 : struct vm_area_struct *vma = vmf->vma;
3493 0 : struct file *file = vma->vm_file;
3494 0 : struct address_space *mapping = file->f_mapping;
3495 0 : pgoff_t last_pgoff = start_pgoff;
3496 : unsigned long addr;
3497 0 : XA_STATE(xas, &mapping->i_pages, start_pgoff);
3498 : struct folio *folio;
3499 : struct page *page;
3500 0 : unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3501 0 : vm_fault_t ret = 0;
3502 :
3503 : rcu_read_lock();
3504 0 : folio = first_map_page(mapping, &xas, end_pgoff);
3505 0 : if (!folio)
3506 : goto out;
3507 :
3508 0 : if (filemap_map_pmd(vmf, folio, start_pgoff)) {
3509 : ret = VM_FAULT_NOPAGE;
3510 : goto out;
3511 : }
3512 :
3513 0 : addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3514 0 : vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3515 : do {
3516 : again:
3517 0 : page = folio_file_page(folio, xas.xa_index);
3518 : if (PageHWPoison(page))
3519 : goto unlock;
3520 :
3521 0 : if (mmap_miss > 0)
3522 0 : mmap_miss--;
3523 :
3524 0 : addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3525 0 : vmf->pte += xas.xa_index - last_pgoff;
3526 0 : last_pgoff = xas.xa_index;
3527 :
3528 : /*
3529 : * NOTE: If there're PTE markers, we'll leave them to be
3530 : * handled in the specific fault path, and it'll prohibit the
3531 : * fault-around logic.
3532 : */
3533 0 : if (!pte_none(*vmf->pte))
3534 : goto unlock;
3535 :
3536 : /* We're about to handle the fault */
3537 0 : if (vmf->address == addr)
3538 0 : ret = VM_FAULT_NOPAGE;
3539 :
3540 0 : do_set_pte(vmf, page, addr);
3541 : /* no need to invalidate: a not-present page won't be cached */
3542 : update_mmu_cache(vma, addr, vmf->pte);
3543 0 : if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3544 0 : xas.xa_index++;
3545 : folio_ref_inc(folio);
3546 : goto again;
3547 : }
3548 0 : folio_unlock(folio);
3549 0 : continue;
3550 : unlock:
3551 0 : if (folio_more_pages(folio, xas.xa_index, end_pgoff)) {
3552 0 : xas.xa_index++;
3553 0 : goto again;
3554 : }
3555 0 : folio_unlock(folio);
3556 : folio_put(folio);
3557 0 : } while ((folio = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3558 0 : pte_unmap_unlock(vmf->pte, vmf->ptl);
3559 : out:
3560 : rcu_read_unlock();
3561 0 : WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3562 0 : return ret;
3563 : }
3564 : EXPORT_SYMBOL(filemap_map_pages);
3565 :
3566 0 : vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3567 : {
3568 0 : struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3569 0 : struct folio *folio = page_folio(vmf->page);
3570 0 : vm_fault_t ret = VM_FAULT_LOCKED;
3571 :
3572 0 : sb_start_pagefault(mapping->host->i_sb);
3573 0 : file_update_time(vmf->vma->vm_file);
3574 0 : folio_lock(folio);
3575 0 : if (folio->mapping != mapping) {
3576 : folio_unlock(folio);
3577 : ret = VM_FAULT_NOPAGE;
3578 : goto out;
3579 : }
3580 : /*
3581 : * We mark the folio dirty already here so that when freeze is in
3582 : * progress, we are guaranteed that writeback during freezing will
3583 : * see the dirty folio and writeprotect it again.
3584 : */
3585 0 : folio_mark_dirty(folio);
3586 0 : folio_wait_stable(folio);
3587 : out:
3588 0 : sb_end_pagefault(mapping->host->i_sb);
3589 0 : return ret;
3590 : }
3591 :
3592 : const struct vm_operations_struct generic_file_vm_ops = {
3593 : .fault = filemap_fault,
3594 : .map_pages = filemap_map_pages,
3595 : .page_mkwrite = filemap_page_mkwrite,
3596 : };
3597 :
3598 : /* This is used for a general mmap of a disk file */
3599 :
3600 0 : int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3601 : {
3602 0 : struct address_space *mapping = file->f_mapping;
3603 :
3604 0 : if (!mapping->a_ops->read_folio)
3605 : return -ENOEXEC;
3606 0 : file_accessed(file);
3607 0 : vma->vm_ops = &generic_file_vm_ops;
3608 0 : return 0;
3609 : }
3610 :
3611 : /*
3612 : * This is for filesystems which do not implement ->writepage.
3613 : */
3614 0 : int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3615 : {
3616 0 : if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3617 : return -EINVAL;
3618 : return generic_file_mmap(file, vma);
3619 : }
3620 : #else
3621 : vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3622 : {
3623 : return VM_FAULT_SIGBUS;
3624 : }
3625 : int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3626 : {
3627 : return -ENOSYS;
3628 : }
3629 : int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3630 : {
3631 : return -ENOSYS;
3632 : }
3633 : #endif /* CONFIG_MMU */
3634 :
3635 : EXPORT_SYMBOL(filemap_page_mkwrite);
3636 : EXPORT_SYMBOL(generic_file_mmap);
3637 : EXPORT_SYMBOL(generic_file_readonly_mmap);
3638 :
3639 0 : static struct folio *do_read_cache_folio(struct address_space *mapping,
3640 : pgoff_t index, filler_t filler, struct file *file, gfp_t gfp)
3641 : {
3642 : struct folio *folio;
3643 : int err;
3644 :
3645 0 : if (!filler)
3646 0 : filler = mapping->a_ops->read_folio;
3647 : repeat:
3648 0 : folio = filemap_get_folio(mapping, index);
3649 0 : if (IS_ERR(folio)) {
3650 0 : folio = filemap_alloc_folio(gfp, 0);
3651 0 : if (!folio)
3652 : return ERR_PTR(-ENOMEM);
3653 0 : err = filemap_add_folio(mapping, folio, index, gfp);
3654 0 : if (unlikely(err)) {
3655 0 : folio_put(folio);
3656 0 : if (err == -EEXIST)
3657 : goto repeat;
3658 : /* Presumably ENOMEM for xarray node */
3659 0 : return ERR_PTR(err);
3660 : }
3661 :
3662 : goto filler;
3663 : }
3664 0 : if (folio_test_uptodate(folio))
3665 : goto out;
3666 :
3667 0 : if (!folio_trylock(folio)) {
3668 : folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3669 : goto repeat;
3670 : }
3671 :
3672 : /* Folio was truncated from mapping */
3673 0 : if (!folio->mapping) {
3674 0 : folio_unlock(folio);
3675 : folio_put(folio);
3676 : goto repeat;
3677 : }
3678 :
3679 : /* Someone else locked and filled the page in a very small window */
3680 0 : if (folio_test_uptodate(folio)) {
3681 : folio_unlock(folio);
3682 : goto out;
3683 : }
3684 :
3685 : filler:
3686 0 : err = filemap_read_folio(file, filler, folio);
3687 0 : if (err) {
3688 0 : folio_put(folio);
3689 0 : if (err == AOP_TRUNCATED_PAGE)
3690 : goto repeat;
3691 0 : return ERR_PTR(err);
3692 : }
3693 :
3694 : out:
3695 0 : folio_mark_accessed(folio);
3696 0 : return folio;
3697 : }
3698 :
3699 : /**
3700 : * read_cache_folio - Read into page cache, fill it if needed.
3701 : * @mapping: The address_space to read from.
3702 : * @index: The index to read.
3703 : * @filler: Function to perform the read, or NULL to use aops->read_folio().
3704 : * @file: Passed to filler function, may be NULL if not required.
3705 : *
3706 : * Read one page into the page cache. If it succeeds, the folio returned
3707 : * will contain @index, but it may not be the first page of the folio.
3708 : *
3709 : * If the filler function returns an error, it will be returned to the
3710 : * caller.
3711 : *
3712 : * Context: May sleep. Expects mapping->invalidate_lock to be held.
3713 : * Return: An uptodate folio on success, ERR_PTR() on failure.
3714 : */
3715 0 : struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3716 : filler_t filler, struct file *file)
3717 : {
3718 0 : return do_read_cache_folio(mapping, index, filler, file,
3719 : mapping_gfp_mask(mapping));
3720 : }
3721 : EXPORT_SYMBOL(read_cache_folio);
3722 :
3723 : /**
3724 : * mapping_read_folio_gfp - Read into page cache, using specified allocation flags.
3725 : * @mapping: The address_space for the folio.
3726 : * @index: The index that the allocated folio will contain.
3727 : * @gfp: The page allocator flags to use if allocating.
3728 : *
3729 : * This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with
3730 : * any new memory allocations done using the specified allocation flags.
3731 : *
3732 : * The most likely error from this function is EIO, but ENOMEM is
3733 : * possible and so is EINTR. If ->read_folio returns another error,
3734 : * that will be returned to the caller.
3735 : *
3736 : * The function expects mapping->invalidate_lock to be already held.
3737 : *
3738 : * Return: Uptodate folio on success, ERR_PTR() on failure.
3739 : */
3740 0 : struct folio *mapping_read_folio_gfp(struct address_space *mapping,
3741 : pgoff_t index, gfp_t gfp)
3742 : {
3743 0 : return do_read_cache_folio(mapping, index, NULL, NULL, gfp);
3744 : }
3745 : EXPORT_SYMBOL(mapping_read_folio_gfp);
3746 :
3747 0 : static struct page *do_read_cache_page(struct address_space *mapping,
3748 : pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp)
3749 : {
3750 : struct folio *folio;
3751 :
3752 0 : folio = do_read_cache_folio(mapping, index, filler, file, gfp);
3753 0 : if (IS_ERR(folio))
3754 0 : return &folio->page;
3755 0 : return folio_file_page(folio, index);
3756 : }
3757 :
3758 0 : struct page *read_cache_page(struct address_space *mapping,
3759 : pgoff_t index, filler_t *filler, struct file *file)
3760 : {
3761 0 : return do_read_cache_page(mapping, index, filler, file,
3762 : mapping_gfp_mask(mapping));
3763 : }
3764 : EXPORT_SYMBOL(read_cache_page);
3765 :
3766 : /**
3767 : * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3768 : * @mapping: the page's address_space
3769 : * @index: the page index
3770 : * @gfp: the page allocator flags to use if allocating
3771 : *
3772 : * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3773 : * any new page allocations done using the specified allocation flags.
3774 : *
3775 : * If the page does not get brought uptodate, return -EIO.
3776 : *
3777 : * The function expects mapping->invalidate_lock to be already held.
3778 : *
3779 : * Return: up to date page on success, ERR_PTR() on failure.
3780 : */
3781 0 : struct page *read_cache_page_gfp(struct address_space *mapping,
3782 : pgoff_t index,
3783 : gfp_t gfp)
3784 : {
3785 0 : return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3786 : }
3787 : EXPORT_SYMBOL(read_cache_page_gfp);
3788 :
3789 : /*
3790 : * Warn about a page cache invalidation failure during a direct I/O write.
3791 : */
3792 0 : void dio_warn_stale_pagecache(struct file *filp)
3793 : {
3794 : static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3795 : char pathname[128];
3796 : char *path;
3797 :
3798 0 : errseq_set(&filp->f_mapping->wb_err, -EIO);
3799 0 : if (__ratelimit(&_rs)) {
3800 0 : path = file_path(filp, pathname, sizeof(pathname));
3801 0 : if (IS_ERR(path))
3802 0 : path = "(unknown)";
3803 0 : pr_crit("Page cache invalidation failure on direct I/O. Possible data corruption due to collision with buffered I/O!\n");
3804 0 : pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3805 : current->comm);
3806 : }
3807 0 : }
3808 :
3809 : ssize_t
3810 0 : generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3811 : {
3812 0 : struct file *file = iocb->ki_filp;
3813 0 : struct address_space *mapping = file->f_mapping;
3814 0 : struct inode *inode = mapping->host;
3815 0 : loff_t pos = iocb->ki_pos;
3816 : ssize_t written;
3817 : size_t write_len;
3818 : pgoff_t end;
3819 :
3820 0 : write_len = iov_iter_count(from);
3821 0 : end = (pos + write_len - 1) >> PAGE_SHIFT;
3822 :
3823 0 : if (iocb->ki_flags & IOCB_NOWAIT) {
3824 : /* If there are pages to writeback, return */
3825 0 : if (filemap_range_has_page(file->f_mapping, pos,
3826 : pos + write_len - 1))
3827 : return -EAGAIN;
3828 : } else {
3829 0 : written = filemap_write_and_wait_range(mapping, pos,
3830 : pos + write_len - 1);
3831 0 : if (written)
3832 : goto out;
3833 : }
3834 :
3835 : /*
3836 : * After a write we want buffered reads to be sure to go to disk to get
3837 : * the new data. We invalidate clean cached page from the region we're
3838 : * about to write. We do this *before* the write so that we can return
3839 : * without clobbering -EIOCBQUEUED from ->direct_IO().
3840 : */
3841 0 : written = invalidate_inode_pages2_range(mapping,
3842 0 : pos >> PAGE_SHIFT, end);
3843 : /*
3844 : * If a page can not be invalidated, return 0 to fall back
3845 : * to buffered write.
3846 : */
3847 0 : if (written) {
3848 0 : if (written == -EBUSY)
3849 : return 0;
3850 : goto out;
3851 : }
3852 :
3853 0 : written = mapping->a_ops->direct_IO(iocb, from);
3854 :
3855 : /*
3856 : * Finally, try again to invalidate clean pages which might have been
3857 : * cached by non-direct readahead, or faulted in by get_user_pages()
3858 : * if the source of the write was an mmap'ed region of the file
3859 : * we're writing. Either one is a pretty crazy thing to do,
3860 : * so we don't support it 100%. If this invalidation
3861 : * fails, tough, the write still worked...
3862 : *
3863 : * Most of the time we do not need this since dio_complete() will do
3864 : * the invalidation for us. However there are some file systems that
3865 : * do not end up with dio_complete() being called, so let's not break
3866 : * them by removing it completely.
3867 : *
3868 : * Noticeable example is a blkdev_direct_IO().
3869 : *
3870 : * Skip invalidation for async writes or if mapping has no pages.
3871 : */
3872 0 : if (written > 0 && mapping->nrpages &&
3873 0 : invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3874 0 : dio_warn_stale_pagecache(file);
3875 :
3876 0 : if (written > 0) {
3877 0 : pos += written;
3878 0 : write_len -= written;
3879 0 : if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3880 0 : i_size_write(inode, pos);
3881 : mark_inode_dirty(inode);
3882 : }
3883 0 : iocb->ki_pos = pos;
3884 : }
3885 0 : if (written != -EIOCBQUEUED)
3886 0 : iov_iter_revert(from, write_len - iov_iter_count(from));
3887 : out:
3888 : return written;
3889 : }
3890 : EXPORT_SYMBOL(generic_file_direct_write);
3891 :
3892 0 : ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i)
3893 : {
3894 0 : struct file *file = iocb->ki_filp;
3895 0 : loff_t pos = iocb->ki_pos;
3896 0 : struct address_space *mapping = file->f_mapping;
3897 0 : const struct address_space_operations *a_ops = mapping->a_ops;
3898 0 : long status = 0;
3899 0 : ssize_t written = 0;
3900 :
3901 : do {
3902 : struct page *page;
3903 : unsigned long offset; /* Offset into pagecache page */
3904 : unsigned long bytes; /* Bytes to write to page */
3905 : size_t copied; /* Bytes copied from user */
3906 0 : void *fsdata = NULL;
3907 :
3908 0 : offset = (pos & (PAGE_SIZE - 1));
3909 0 : bytes = min_t(unsigned long, PAGE_SIZE - offset,
3910 : iov_iter_count(i));
3911 :
3912 : again:
3913 : /*
3914 : * Bring in the user page that we will copy from _first_.
3915 : * Otherwise there's a nasty deadlock on copying from the
3916 : * same page as we're writing to, without it being marked
3917 : * up-to-date.
3918 : */
3919 0 : if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) {
3920 : status = -EFAULT;
3921 0 : break;
3922 : }
3923 :
3924 0 : if (fatal_signal_pending(current)) {
3925 : status = -EINTR;
3926 : break;
3927 : }
3928 :
3929 0 : status = a_ops->write_begin(file, mapping, pos, bytes,
3930 : &page, &fsdata);
3931 0 : if (unlikely(status < 0))
3932 : break;
3933 :
3934 0 : if (mapping_writably_mapped(mapping))
3935 : flush_dcache_page(page);
3936 :
3937 0 : copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3938 0 : flush_dcache_page(page);
3939 :
3940 0 : status = a_ops->write_end(file, mapping, pos, bytes, copied,
3941 : page, fsdata);
3942 0 : if (unlikely(status != copied)) {
3943 0 : iov_iter_revert(i, copied - max(status, 0L));
3944 0 : if (unlikely(status < 0))
3945 : break;
3946 : }
3947 0 : cond_resched();
3948 :
3949 0 : if (unlikely(status == 0)) {
3950 : /*
3951 : * A short copy made ->write_end() reject the
3952 : * thing entirely. Might be memory poisoning
3953 : * halfway through, might be a race with munmap,
3954 : * might be severe memory pressure.
3955 : */
3956 0 : if (copied)
3957 0 : bytes = copied;
3958 : goto again;
3959 : }
3960 0 : pos += status;
3961 0 : written += status;
3962 :
3963 0 : balance_dirty_pages_ratelimited(mapping);
3964 0 : } while (iov_iter_count(i));
3965 :
3966 0 : return written ? written : status;
3967 : }
3968 : EXPORT_SYMBOL(generic_perform_write);
3969 :
3970 : /**
3971 : * __generic_file_write_iter - write data to a file
3972 : * @iocb: IO state structure (file, offset, etc.)
3973 : * @from: iov_iter with data to write
3974 : *
3975 : * This function does all the work needed for actually writing data to a
3976 : * file. It does all basic checks, removes SUID from the file, updates
3977 : * modification times and calls proper subroutines depending on whether we
3978 : * do direct IO or a standard buffered write.
3979 : *
3980 : * It expects i_rwsem to be grabbed unless we work on a block device or similar
3981 : * object which does not need locking at all.
3982 : *
3983 : * This function does *not* take care of syncing data in case of O_SYNC write.
3984 : * A caller has to handle it. This is mainly due to the fact that we want to
3985 : * avoid syncing under i_rwsem.
3986 : *
3987 : * Return:
3988 : * * number of bytes written, even for truncated writes
3989 : * * negative error code if no data has been written at all
3990 : */
3991 0 : ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3992 : {
3993 0 : struct file *file = iocb->ki_filp;
3994 0 : struct address_space *mapping = file->f_mapping;
3995 0 : struct inode *inode = mapping->host;
3996 0 : ssize_t written = 0;
3997 : ssize_t err;
3998 : ssize_t status;
3999 :
4000 : /* We can write back this queue in page reclaim */
4001 0 : current->backing_dev_info = inode_to_bdi(inode);
4002 0 : err = file_remove_privs(file);
4003 0 : if (err)
4004 : goto out;
4005 :
4006 0 : err = file_update_time(file);
4007 0 : if (err)
4008 : goto out;
4009 :
4010 0 : if (iocb->ki_flags & IOCB_DIRECT) {
4011 : loff_t pos, endbyte;
4012 :
4013 0 : written = generic_file_direct_write(iocb, from);
4014 : /*
4015 : * If the write stopped short of completing, fall back to
4016 : * buffered writes. Some filesystems do this for writes to
4017 : * holes, for example. For DAX files, a buffered write will
4018 : * not succeed (even if it did, DAX does not handle dirty
4019 : * page-cache pages correctly).
4020 : */
4021 0 : if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
4022 : goto out;
4023 :
4024 0 : pos = iocb->ki_pos;
4025 0 : status = generic_perform_write(iocb, from);
4026 : /*
4027 : * If generic_perform_write() returned a synchronous error
4028 : * then we want to return the number of bytes which were
4029 : * direct-written, or the error code if that was zero. Note
4030 : * that this differs from normal direct-io semantics, which
4031 : * will return -EFOO even if some bytes were written.
4032 : */
4033 0 : if (unlikely(status < 0)) {
4034 : err = status;
4035 : goto out;
4036 : }
4037 : /*
4038 : * We need to ensure that the page cache pages are written to
4039 : * disk and invalidated to preserve the expected O_DIRECT
4040 : * semantics.
4041 : */
4042 0 : endbyte = pos + status - 1;
4043 0 : err = filemap_write_and_wait_range(mapping, pos, endbyte);
4044 0 : if (err == 0) {
4045 0 : iocb->ki_pos = endbyte + 1;
4046 0 : written += status;
4047 0 : invalidate_mapping_pages(mapping,
4048 0 : pos >> PAGE_SHIFT,
4049 0 : endbyte >> PAGE_SHIFT);
4050 : } else {
4051 : /*
4052 : * We don't know how much we wrote, so just return
4053 : * the number of bytes which were direct-written
4054 : */
4055 : }
4056 : } else {
4057 0 : written = generic_perform_write(iocb, from);
4058 0 : if (likely(written > 0))
4059 0 : iocb->ki_pos += written;
4060 : }
4061 : out:
4062 0 : current->backing_dev_info = NULL;
4063 0 : return written ? written : err;
4064 : }
4065 : EXPORT_SYMBOL(__generic_file_write_iter);
4066 :
4067 : /**
4068 : * generic_file_write_iter - write data to a file
4069 : * @iocb: IO state structure
4070 : * @from: iov_iter with data to write
4071 : *
4072 : * This is a wrapper around __generic_file_write_iter() to be used by most
4073 : * filesystems. It takes care of syncing the file in case of O_SYNC file
4074 : * and acquires i_rwsem as needed.
4075 : * Return:
4076 : * * negative error code if no data has been written at all of
4077 : * vfs_fsync_range() failed for a synchronous write
4078 : * * number of bytes written, even for truncated writes
4079 : */
4080 0 : ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4081 : {
4082 0 : struct file *file = iocb->ki_filp;
4083 0 : struct inode *inode = file->f_mapping->host;
4084 : ssize_t ret;
4085 :
4086 0 : inode_lock(inode);
4087 0 : ret = generic_write_checks(iocb, from);
4088 0 : if (ret > 0)
4089 0 : ret = __generic_file_write_iter(iocb, from);
4090 0 : inode_unlock(inode);
4091 :
4092 0 : if (ret > 0)
4093 0 : ret = generic_write_sync(iocb, ret);
4094 0 : return ret;
4095 : }
4096 : EXPORT_SYMBOL(generic_file_write_iter);
4097 :
4098 : /**
4099 : * filemap_release_folio() - Release fs-specific metadata on a folio.
4100 : * @folio: The folio which the kernel is trying to free.
4101 : * @gfp: Memory allocation flags (and I/O mode).
4102 : *
4103 : * The address_space is trying to release any data attached to a folio
4104 : * (presumably at folio->private).
4105 : *
4106 : * This will also be called if the private_2 flag is set on a page,
4107 : * indicating that the folio has other metadata associated with it.
4108 : *
4109 : * The @gfp argument specifies whether I/O may be performed to release
4110 : * this page (__GFP_IO), and whether the call may block
4111 : * (__GFP_RECLAIM & __GFP_FS).
4112 : *
4113 : * Return: %true if the release was successful, otherwise %false.
4114 : */
4115 0 : bool filemap_release_folio(struct folio *folio, gfp_t gfp)
4116 : {
4117 0 : struct address_space * const mapping = folio->mapping;
4118 :
4119 0 : BUG_ON(!folio_test_locked(folio));
4120 0 : if (folio_test_writeback(folio))
4121 : return false;
4122 :
4123 0 : if (mapping && mapping->a_ops->release_folio)
4124 0 : return mapping->a_ops->release_folio(folio, gfp);
4125 0 : return try_to_free_buffers(folio);
4126 : }
4127 : EXPORT_SYMBOL(filemap_release_folio);
|
